Browse the corpus

Introduction Clinical documentation is among the most time-consuming and costly aspects of using an electronic health record (EHR) system.1,2 Speech recognition (SR), the automatic translation of voice into text, has been a promising technology for clinical documentation since the 1980s. A recent study reported that 90% of hospitals plan to expand their use of SR technology.3 There are 2 primary ways that SR can assist the clinical documentation process. In this study, we evaluated back-end SR (Figure, A), in which physicians’ dictations are captured and converted to text by an SR engine. The SR-generated text is edited by a professional medical transcriptionist (MT), then sent back to the physician for review. The other type is commonly called front-end SR (Figure, B). Here, physicians dictate directly into free-text fields of the EHR and edit the transcription before saving the document. Figure. Stages of Back-End and Front-End Dictation There are two 2 primary ways that speech recognition (SR) can assist the clinical documentation process. In back-end SR, clinicians’ dictations, the audio original (AO), are captured and converted to text by an SR engine. The SR-generated text is edited by a professional medical transcriptionist (MT), then sent back to the clinician for review and a signed note (SN). In front-end SR, clinicians dictate directly into free-text fields of the electronic health record and edit the transcription.

ptured and converted to text by an SR engine. The SR-generated text is edited by a professional medical transcriptionist (MT), then sent back to the clinician for review and a signed note (SN). In front-end SR, clinicians dictate directly into free-text fields of the electronic health record and edit the transcription. A recent study in Australia evaluated the type and prevalence of errors in documents created using a front-end SR system and those created using a keyboard and mouse.4 A higher prevalence of errors was found in notes created with SR, both overall and across most error types included in the analysis. While front-end SR is becoming more widely used, back-end SR systems remain in use in many health care institutions in the United States, and significant productivity enhancements associated with these systems have been demonstrated.5,6,7 However, to our knowledge, the quality and accuracy of clinical documents created using back-end SR have not been thoroughly studied.

k-end SR systems remain in use in many health care institutions in the United States, and significant productivity enhancements associated with these systems have been demonstrated.5,6,7 However, to our knowledge, the quality and accuracy of clinical documents created using back-end SR have not been thoroughly studied. Medical errors largely result from failed communication.8 Clinical documentation is essential for communication of a patient’s diagnosis and treatment and for care coordination between clinicians. Documentation errors can put patients at significant risk of harm.9 An analysis of medical malpractice cases found that incorrect information (eg, faulty data entry) was the top EHR-related contributing factor, contributing to 20% of reviewed cases.10,11 It is therefore in the best interest of both patients and clinicians that medical documents be accurate, complete, legible, and readily accessible for the purposes of patient safety, health care delivery, billing, audit, and possible litigation proceedings.12,13,14,15 As more medical institutions adopt SR software, we need to better understand how it can be used safely and efficiently.

that medical documents be accurate, complete, legible, and readily accessible for the purposes of patient safety, health care delivery, billing, audit, and possible litigation proceedings.12,13,14,15 As more medical institutions adopt SR software, we need to better understand how it can be used safely and efficiently. In this study, we analyzed errors at different processing stages of clinical documents collected from 2 institutions using the same back-end SR system. We hypothesized that error rates would be highest in original SR transcriptions, lower in notes edited by transcriptionists, and lower still in physicians’ signed notes (SN). We also expected significant differences in mean error rates between note types and between notes created by physicians of different specialties. We expected no significant difference in mean error rates among physicians of different sexes or from different institutions. Methods This cross-sectional study was conducted and reported following the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.16 Approval for this study was obtained from the Partners Human Research Committee and the Colorado Multiple Institutional Review Board. The study was determined by both institutional review boards to meet the criteria for a waiver of informed consent. Analysis was conducted from June 15, 2016, to November 17, 2017.

ne.16 Approval for this study was obtained from the Partners Human Research Committee and the Colorado Multiple Institutional Review Board. The study was determined by both institutional review boards to meet the criteria for a waiver of informed consent. Analysis was conducted from June 15, 2016, to November 17, 2017. Clinical Setting and Data Collection This study used 217 documents dictated between January 1 and December 31, 2016, from hospitals at 2 health care organizations: Partners HealthCare System in Boston, Massachusetts, and University of Colorado Health System (UCHealth) in Aurora, Colorado. Both organizations use Dragon Medical 360 | eScription (Nuance). Because hospitals use dictation for different note types, we collected a stratified random sample based on the different note types dictated at each hospital. The sample includes 44 operative notes, 83 office notes, and 40 discharge summaries from Partners HealthCare and 15 operative notes and 35 discharge summaries from UCHealth. We collected data for dictating physicians’ age, sex, and specialty. We reviewed each note at the 4 main processing stages of dictation. This included (1) our own transcription of the original audio recording (used as the criterion standard, described in the following section), (2) the note generated by the SR engine of the vendor transcription service (SR note), (3) the note following revision by a professional MT (MT note), and (4) the note after having been reviewed and signed by the physician (SN).

ginal audio recording (used as the criterion standard, described in the following section), (2) the note generated by the SR engine of the vendor transcription service (SR note), (3) the note following revision by a professional MT (MT note), and (4) the note after having been reviewed and signed by the physician (SN). Criterion Standard, Annotation Schema, and Annotation Process To create the criterion standard for each note, a PharmD candidate or medical student, under the supervision of 2 practicing physicians, created a transcription of the note while listening to the original audio and using the MT note as a reference. The audio was played repeatedly, at different speeds, to ensure the transcription’s accuracy. Medical record review was conducted to validate notes’ content, such as by referring to a patient’s structured medication list to verify a medication order that was partially inaudible in the original audio recording.

ference. The audio was played repeatedly, at different speeds, to ensure the transcription’s accuracy. Medical record review was conducted to validate notes’ content, such as by referring to a patient’s structured medication list to verify a medication order that was partially inaudible in the original audio recording. A team of clinical informaticians, computational linguists, and clinicians developed a schema for identifying and classifying errors iteratively over multiple annotation rounds. The schema includes 12 general types (eg, insertion), 14 semantic types (eg, medication), and a binary classification of clinical significance.17 An error was considered clinically significant if it could plausibly change a note’s interpretation, thereby potentially affecting a patient’s future care either directly (eg, by influencing clinical decisions or treatment options) or indirectly (eg, by causing billing errors or affecting litigation proceedings). The complete annotation schema is shown in Table 1. The schema includes brief descriptions of each type of error. Each error type is also accompanied by 1 or more examples found during the course of our annotation.

or treatment options) or indirectly (eg, by causing billing errors or affecting litigation proceedings). The complete annotation schema is shown in Table 1. The schema includes brief descriptions of each type of error. Each error type is also accompanied by 1 or more examples found during the course of our annotation. Table 1. Annotation Schema Error Type Description Examples General type Insertion One or more words was added to the transcription AO: There is distal biliary obstruction observed SR: There is no distal biliary obstruction observed Deletion One or more words was deleted from the transcription AO: CHADS2 VASC score 4 SR: score Substitution Enunciation An error due to a mispronunciation or failure to enunciate by the speaker AO: to find a homeopathic provider SR: defined homeopathic provider Suffix The root word is correct, but there is an incorrect, added, or omitted suffix AO: mental status worsened SR: mental status worsens Prefix The root word is correct, but there is an incorrect, added, or omitted prefix AO: Inadequate evaluation to exclude neoplasia SR: Adequate evaluation to exclude neoplasia Spelling The transcriptionist made a spelling error when editing the output of the SR system AO: we counseled him on risk of infection MT: we counseled hom on risk of infection Homophone One word has been substituted for another identically pronounced word AO: serial high resolution anoscopy SR: cereal high resolution anoscopy Dictionary An error likely due to the target word not being present in the SR system’s dictionary AO: driving a Camry and hit another car SR: driving an Academy and hit another car Nonsense A substitution that is so far off that it is unclear which category (if any) it falls under AO: follow up in 3 to 5 d SR: neck veins are evaluated No.a Any error involving a number, whether it is written as a digit (2) or as a word (two) AO: the patient is a 17-year-old female SR: the patient is a 70-year-old female Punctuationb A period, comma, or other punctuation mark was present where it should not have been AO: at discharge she had no flank tenderness SR: at discharge.

ror involving a number, whether it is written as a digit (2) or as a word (two) AO: the patient is a 17-year-old female SR: the patient is a 70-year-old female Punctuationb A period, comma, or other punctuation mark was present where it should not have been AO: at discharge she had no flank tenderness SR: at discharge. She had no flank tenderness Semantic type General English Any English words that do not fit into the categories below AO: which she would otherwise forget SR: which she would otherwise for gas that Stop word Common English words17 AO: intermittent pain under the right breast SR: intermittent pain in the right breast Medication Medication names and dose information AO: initiated on lamotrigine therapy SR: initiated on layman will try therapy Diagnosis Any words that are part of a specific medical diagnosis AO: Dengue SR: DKA [diabetic ketoacidosis] Laboratory test Laboratory test names and results AO: TSH of 26.7 SR: TSH of 22nd 6.7 Imaging test Imaging examination names and types and examination results AO: nonobstructive on CT imaging SR: nonobstructive on imaging Procedure Procedure names and descriptions AO: ligament was released on the leading edge SR: ligament was released operating edge Physical examination Any information directly related to the physical examination and any associated values AO: T 36.7 degrees SR: T3-T7 disease Patient or physician identifier Any words involving patient or physician metadata (names, MRN, etc) AO: SURGEON: [surgeon’s real name] SR: SURGEON: Stathis stairs Date Any dates, written with words (January 1, 2017) or with numbers (01/01/2017) AO: 10/10/2016 SR: 10/10/2000 Symptom Any symptom or description of symptoms AO: very mild arthralgias SR: very also arthralgias ??? When ??__?? (or similar) is left in the note, or when something is complete nonsense AO: no foreign material was identified MT: no foreign ??__??

ary 1, 2017) or with numbers (01/01/2017) AO: 10/10/2016 SR: 10/10/2000 Symptom Any symptom or description of symptoms AO: very mild arthralgias SR: very also arthralgias ??? When ??__?? (or similar) is left in the note, or when something is complete nonsense AO: no foreign material was identified MT: no foreign ??__?? was identified Clinical significance Any error that could plausibly change a note’s interpretation, thereby giving it the potential to affect a patient’s future care 1 AO: Long-term anticoagulation should be discussed with her PCP SR: Long-term she should be discussed with her PCP 2 AO: Rapamune 4 mg po daily SR: Verapamil 4 mg po daily 3 AO: The patient disagreed with this recommendation SR: The patient did agree with this recommendation 4 AO: Inadequate evaluation to exclude neoplasia SR: Adequate evaluation to exclude neoplasia 5 AO: Allergies: Furosemide, gabapentin, oxcarbazepine, yellow dye SR: Allergies: Yellow dye 6 AO: There is distal biliary obstruction observed SR: There is no distal biliary obstruction observed 7 AO: Oxycodone 5 mg oral q4h PRN for pain SR, MT, SN: Oxycodone 5 mg oral q4h for pain Abbreviations: AO, audio original (ie, criterion standard); MRN, medical record number; MT, medical transcriptionist-edited notes; SN, signed notes; SR, speech recognition. a A number error is a more specific type of insertion, deletion, or substitution. b A punctuation error is a more specific type of insertion or substitution.

was identified Clinical significance Any error that could plausibly change a note’s interpretation, thereby giving it the potential to affect a patient’s future care 1 AO: Long-term anticoagulation should be discussed with her PCP SR: Long-term she should be discussed with her PCP 2 AO: Rapamune 4 mg po daily SR: Verapamil 4 mg po daily 3 AO: The patient disagreed with this recommendation SR: The patient did agree with this recommendation 4 AO: Inadequate evaluation to exclude neoplasia SR: Adequate evaluation to exclude neoplasia 5 AO: Allergies: Furosemide, gabapentin, oxcarbazepine, yellow dye SR: Allergies: Yellow dye 6 AO: There is distal biliary obstruction observed SR: There is no distal biliary obstruction observed 7 AO: Oxycodone 5 mg oral q4h PRN for pain SR, MT, SN: Oxycodone 5 mg oral q4h for pain Abbreviations: AO, audio original (ie, criterion standard); MRN, medical record number; MT, medical transcriptionist-edited notes; SN, signed notes; SR, speech recognition. a A number error is a more specific type of insertion, deletion, or substitution. b A punctuation error is a more specific type of insertion or substitution. We used Knowtator,18 an open-source annotation tool, to annotate notes at each stage. Two annotators (1 computational linguist and 1 medical student) independently annotated the SR-transcribed, transcriptionist-edited, and signed versions of each note for errors. Each document was further annotated for the presence or absence the following changes: automatic abbreviation expansion by the SR system, disfluencies or misspoken words on the part of the dictating physician, stylistic changes (eg, rewording a grammatically incorrect sentence) by the transcriptionist and the signing physician, rearranging of the note’s content by the transcriptionist and the physician, and the addition and removal of content by the physician prior to signing.

en words on the part of the dictating physician, stylistic changes (eg, rewording a grammatically incorrect sentence) by the transcriptionist and the signing physician, rearranging of the note’s content by the transcriptionist and the physician, and the addition and removal of content by the physician prior to signing. Two practicing physicians independently evaluated errors for clinical significance, and disagreements were reconciled through discussion. Measures We determined the time required to dictate a note, along with each note’s turnaround time and clinician review time. We defined turnaround time as the length of time between the original dictation’s completion and when the transcriptionist-revised document was sent back to the EHR. Physician review time was the length of time between when the transcription was returned to the physician and when the physician signed the note.

ew time. We defined turnaround time as the length of time between the original dictation’s completion and when the transcriptionist-revised document was sent back to the EHR. Physician review time was the length of time between when the transcription was returned to the physician and when the physician signed the note. For each version of each note, we analyzed the differences between that note and the corresponding criterion standard note. We determined the error rate (ie, the number of errors per 100 words), the median error rate with interquartile ranges, the mean number of errors per note, the frequency of each error type (the number of errors of a specific type divided by the total number of errors), and the percentage of notes containing at least 1 error. We conducted these analyses for all errors and for just those errors that were found to be clinically significant. Throughout our analyses, a document’s error rate is defined as the total number of errors it contains (or, equivalently, the total number of insertions, deletions, and substitutions) divided by the number of words in the corresponding criterion standard.

r just those errors that were found to be clinically significant. Throughout our analyses, a document’s error rate is defined as the total number of errors it contains (or, equivalently, the total number of insertions, deletions, and substitutions) divided by the number of words in the corresponding criterion standard. We calculated interannotator agreement using a randomly selected subset of 33 notes, which included 7 SR-transcribed notes and 26 transcriptionist-edited notes, considering these stages’ variations in error complexity (eg, transcriptionists’ edits often involve subtle rewordings, which must be distinguished from true errors). Agreement was defined as the percentage of errors for which both annotators selected the same general and semantic type. For each error, we required only that the spans selected by each annotator overlap with one another to some degree, rather than requiring exact span matches. For clinical significance, agreement was defined as the percentage overlap between the 2 physicians’ classifications.

selected the same general and semantic type. For each error, we required only that the spans selected by each annotator overlap with one another to some degree, rather than requiring exact span matches. For clinical significance, agreement was defined as the percentage overlap between the 2 physicians’ classifications. Statistical Analysis Analyses were conducted in R statistical software (R Project for Statistical Computing)19 with t tests used to identify significant differences in mean error rates at each stage by sex, specialty, and note type. For comparisons involving more than 2 groups (eg, specialty), each group’s mean error rate was compared with that of all other groups combined. We calculated the Pearson correlation coefficient (r) to measure the strength of associations between error rate and physician age and between error rate and document length. We considered 2-sided P values of less than .05 to be statistically significant. Results Among the 217 notes, there were 144 unique dictating physicians: 44 female (30.6%) and 10 unknown sex (6.9%). Mean (SD) physician age was 52 (12.5) years (median [range] age, 54 [28-80] years). Among 121 physicians for whom specialty information was available (84.0%), 35 specialties were represented, including 45 surgeons (37.2%), 30 internists (24.8%), and 46 others (38.0%).

cians: 44 female (30.6%) and 10 unknown sex (6.9%). Mean (SD) physician age was 52 (12.5) years (median [range] age, 54 [28-80] years). Among 121 physicians for whom specialty information was available (84.0%), 35 specialties were represented, including 45 surgeons (37.2%), 30 internists (24.8%), and 46 others (38.0%). Original audio recordings contained a mean (SD) of 507 (296.9) words (median [range], 446 [59-1911] words) per document. The average dictation duration was 5 minutes, 46 seconds (median [range], 4 minutes, 45 seconds [21 seconds to 31 minutes, 35 seconds]). The average turnaround time was 3 hours, 37 minutes (median [range], 1 hour, 1 minute [2 minutes to 38 hours, 45 minutes]). The average physician review time was 4 days, 13 hours, 16 minutes (median [range], 23 hours, 25 minutes [0 minutes to 146 days, 4 hours, 54 minutes]). There were 329 errors in the 33-note subset. For the 171 errors that were identified by both annotators, interannotator agreement was 71.9%. Each of the annotators failed to identify a mean (SD) of 21.7% (1.5%) of the errors that were annotated by the other. Of the errors identified by only 1 annotator, 32 errors (20.3%) pertained to clinical information; the remaining 126 (79.7%) involved minor changes to general English words. Agreement for clinical significance was 85.7%. Examples of errors of each type that were identified in this data set can be found in Table 1.

her. Of the errors identified by only 1 annotator, 32 errors (20.3%) pertained to clinical information; the remaining 126 (79.7%) involved minor changes to general English words. Agreement for clinical significance was 85.7%. Examples of errors of each type that were identified in this data set can be found in Table 1. Detailed results of our error analysis are shown in Table 2 and Table 3. Errors were prevalent in original SR transcriptions, with an overall mean (SD) error rate of 7.4% (4.8%). The rate of errors decreased substantially following revision by MTs, to 0.4%. Errors were further reduced in SNs, which had an overall error rate of 0.3%. The number of notes containing at least 1 error also decreased with each processing stage. Of the 217 original SR transcriptions, 209 (96.3%) had errors. Following transcriptionist revision, this number decreased to 129 (58.1%), and by the time notes were signed, only 92 (42.4%) contained errors.

ll error rate of 0.3%. The number of notes containing at least 1 error also decreased with each processing stage. Of the 217 original SR transcriptions, 209 (96.3%) had errors. Following transcriptionist revision, this number decreased to 129 (58.1%), and by the time notes were signed, only 92 (42.4%) contained errors. Table 2. Summary of Error Rates by Note Type and Processing Stage Note Type (No.) Note Stage Errors, All Types Clinically Significant Errors No. of Total Errors/Total No. of Words in Criterion Standard (%) No. of Notes With Errors/Total No. of Notes (%) Mean Errors per Note, No. Error Rate by Note, Median (IQR) No. of Total Errors/Total No. of Words in Criterion Standard (%) No. of Notes With Errors/Total No. of Notes (%) Mean Errors per Note, No. Error Rate by Note, Median (IQR) Discharge summaries (75) SR 3892/42 873 (9.1) 75/75 (100.0) 51.9 8.2 (5.2-11.7) 248/42 873 (0.6) 69/75 (90.7) 3.3 0.2 (0-0.6) MT 195/42 873 (0.5) 46/75 (61.3) 2.6 0.3 (0.0-0.6) 13/42 873 (0.03) 12/75 (16.0) 0.2 0 SN 163/42 873 (0.4) 41/75 (54.7) 2.2 0.2 (0.0-0.5) 6/42 873 (0.01) 6/75 (8.0) 0.1 0 Office notes (83) SR 2513/35 841 (7.0) 76/83 (91.6) 30.3 5.8 (3.4-8.9) 136/35 841 (0.4) 39/83 (47.0) 1.6 0 (0-0.3) MT 239/35 841 (0.7) 43/83 (51.8) 2.9 0.2 (0.0-0.4) 17/35 841 (0.05) 12/83 (14.5) 0.2 0 SN 53/35 841 (0.1) 22/83 (26.5) 0.6 0.0 (0.0-0.2) 9/35 841 (0.03) 7/83 (8.4) 0.1 0 Operative notes (59) SR 1756/31 466 (5.6) 58/59 (98.3) 29.8 4.6 (3.2-8.2) 84/31 466 (0.3) 31/59 (52.5) 1.4 0 (0-0.2) MT 138/31 466 (0.4) 37/59 (62.7) 2.3 0.2 (0.0-0.6) 11/31 466 (0.03) 8/59 (13.6) 0.2 0 SN 112/31 466 (0.4) 29/59 (49.2) 1.9 0.0 (0.0-0.5) 6/31 466 (0.02) 4/59 (6.8) 0.2 0 All notes (217) SR 8161/110 180 (7.4) 209/217 (96.3) 37.6 6.1 (3.9-10.1) 468/110 180 (0.4) 138/217 (63.6) 2.2 0 (0-0.4) MT 461/110 180 (0.4) 129/217 (58.1) 2.1 0.2 (0.0-0.5) 41/110 180 (0.04) 32/217 (14.7) 0.2 0 SN 328/110 180 (0.3) 92/217 (42.4) 1.5 0.0 (0.0-0.4) 21/110 180 (0.02) 17/217 (7.8) 0.1 0 Abbreviations: IQR, interquartile range; MT, medical transcriptionist–edited notes; SN, signed notes; SR, automatic transcriptions by the speech recognition system.

17 (58.1) 2.1 0.2 (0.0-0.5) 41/110 180 (0.04) 32/217 (14.7) 0.2 0 SN 328/110 180 (0.3) 92/217 (42.4) 1.5 0.0 (0.0-0.4) 21/110 180 (0.02) 17/217 (7.8) 0.1 0 Abbreviations: IQR, interquartile range; MT, medical transcriptionist–edited notes; SN, signed notes; SR, automatic transcriptions by the speech recognition system. Table 3. Error Types in Dictated Notes by Note Type and Processing Stage Note Type (No.) Note Stage Total Errors, No. No. (%) Errors, General Typea Errors, Semantic Typea Deletion Insertion Substitution No.

17 (58.1) 2.1 0.2 (0.0-0.5) 41/110 180 (0.04) 32/217 (14.7) 0.2 0 SN 328/110 180 (0.3) 92/217 (42.4) 1.5 0.0 (0.0-0.4) 21/110 180 (0.02) 17/217 (7.8) 0.1 0 Abbreviations: IQR, interquartile range; MT, medical transcriptionist–edited notes; SN, signed notes; SR, automatic transcriptions by the speech recognition system. Table 3. Error Types in Dictated Notes by Note Type and Processing Stage Note Type (No.) Note Stage Total Errors, No. No. (%) Errors, General Typea Errors, Semantic Typea Deletion Insertion Substitution No. Punctuation General Englishc Clinical Information Enunciation Homonym Nonsense Otherb Medication Diagnosis Procedure Symptom Laboratory Test Physical Examination Imaging Test Otherd Discharge summaries (75) SR 3892 1395 (35.8) 1031 (26.5) 655 (16.8) 5 (0.1) 272 (7.0) 186 (4.8) 133 (3.4) 215 (5.5) 3314 (85.1) 127 (3.3) 76 (2.0) 13 (0.3) 60 (1.5) 50 (1.3) 30 (0.8) 25 (0.6) 197 (5.1) MT 195 87 (44.7) 36 (18.5) 35 (17.9) 0 5 (2.6) 20 (10.3) 9 (4.6) 3 (1.5) 133 (68.2) 14 (7.2) 9 (4.6) 2 (1.0) 7 (3.6) 2 (1.0) 14 (7.2) 2 (1.0) 12 (6.2) SN 163 74 (45.4) 29 (17.8) 29 (17.8) 0 1 (0.6) 19 (11.7) 8 (4.9) 3 (1.8) 114 (69.9) 6 (3.7) 9 (5.5) 2 (1.2) 7 (4.3) 2 (1.2) 14 (8.6) 2 (1.2) 7 (4.3) Office notes (83) SR 2513 875 (34.8) 549 (21.8) 608 (24.2) 10 (0.4) 178 (7.1) 128 (5.1) 41 (1.6) 124 (4.9) 2166 (86.2) 55 (2.2) 60 (2.4) 27 (1.1) 40 (1.6) 15 (0.6) 10 (0.4) 8 (0.3) 132 (5.3) MT 128 39 (30.5) 26 (20.3) 39 (30.5) 1 (0.8) 11 (8.6) 10 (7.8) 2 (1.6) 0 108 (84.4) 1 (0.8) 7 (5.5) 2 (1.6) 2 (1.6) 1 (0.8) 0 0 7 (5.5) SN 53 16 (30.2) 17 (32.1) 10 (18.9) 0 6 (11.3) 3 (5.7) 1 (1.9) 0 44 (83.0) 0 5 (9.4) 0 1 (1.9) 1 (1.9) 0 0 2 (3.8) Operative notes (59) SR 1756 559 (31.8) 620 (35.3) 226 (12.9) 13 (0.7) 124 (7.1) 77 (4.4) 40 (2.3) 97 (5.5) 1393 (79.3) 4 (0.2) 39 (2.2) 140 (8.0) 8 (0.5) 1 (0.1) 5 (0.3) 1 (0.1) 165 (9.4) MT 138 48 (34.8) 37 (26.8) 25 (18.1) 1 (0.7) 15 (10.9) 9 (6.5) 1 (0.7) 2 (1.4) 96 (69.6) 0 5 (3.6) 15 (10.9) 4 (2.9) 0 1 (0.7) 1 (0.7) 16 (11.6) SN 112 43 (38.4) 35 (31.3) 19 (17.0) 1 (0.9) 5 (4.5) 6 (5.4) 1 (0.9) 2 (1.8) 85 (75.9) 0 4 (3.6) 10 (8.9) 2 (1.8) 0 1 (0.9) 1 (0.9) 9 (8.0) All notes (217) SR 8161 2829 (34.7) 2200 (27.0) 1489 (18.2) 28 (0.3) 574 (7.0) 391 (4.8) 214 (2.6) 436 (5.3) 6873 (84.2) 186 (2.3) 175 (2.1) 180 (2.2) 108 (1.3) 66 (0.8) 45 (0.6) 34 (0.4) 494 (6.1) MT 461 174 (37.7) 99 (21.5) 99 (21.5) 2 (0.4) 31 (6.7) 39 (8.5) 12 (2.6) 5 (1.1) 337 (73.1) 15 (3.3) 21 (4.6) 19 (4.1) 13 (2.8) 3 (0.7) 15 (3.3) 3 (0.7) 35 (7.6) SN 328 133 (40.5) 81 (24.7) 58 (17.7) 1 (0.3) 12 (3.7) 28 (8.5) 10 (3.0) 5 (1.5) 243 (74.1) 6 (1.8) 18 (5.5) 12 (3.7) 10 (3.0) 3 (0.9) 15 (4.6) 3 (0.9) 18 (5.5) Abbreviations: MT, medical transcriptionist–edited notes; SN, signed notes; SR,

37 (73.1) 15 (3.3) 21 (4.6) 19 (4.1) 13 (2.8) 3 (0.7) 15 (3.3) 3 (0.7) 35 (7.6) SN 328 133 (40.5) 81 (24.7) 58 (17.7) 1 (0.3) 12 (3.7) 28 (8.5) 10 (3.0) 5 (1.5) 243 (74.1) 6 (1.8) 18 (5.5) 12 (3.7) 10 (3.0) 3 (0.9) 15 (4.6) 3 (0.9) 18 (5.5) Abbreviations: MT, medical transcriptionist–edited notes; SN, signed notes; SR, automatic transcriptions by the speech recognition system. a Percentages are equal to the number of errors of each type divided by the total number of errors; percentages may not sum to 100 because of rounding. b Includes suffix, prefix, dictionary, and spelling errors. c Includes general English, stop word, and date errors. d Includes patient or physician identifier, interpretation, psychological test, and ??? (unintelligible or otherwise unclassifiable) errors. The effect of human review on note accuracy becomes more pronounced when considering just those errors that are clinically significant, rather than treating all errors as equally meaningful. Prior to human revision, 138 of 217 notes (63.6%) had at least 1 clinically significant error, with a mean (SD) of 2.2 (2.7) errors per note. After being edited by an MT, 32 notes (14.7%) had clinically significant errors, and only 17 SNs (7.8%) contained such errors. However, the proportion of errors involving clinical information increased from 15.8% to 26.9% after transcriptionist revision, although it decreased slightly to 25.9% in SNs. Similarly, the proportion of errors that were clinically significant increased from 5.7% in the original SR transcriptions to 8.9% after being edited by an MT, then decreased to 6.4% in SNs.

inical information increased from 15.8% to 26.9% after transcriptionist revision, although it decreased slightly to 25.9% in SNs. Similarly, the proportion of errors that were clinically significant increased from 5.7% in the original SR transcriptions to 8.9% after being edited by an MT, then decreased to 6.4% in SNs. Table 3 shows the number and proportion of each error type across the 3 processing stages for each note type and for all notes combined. At all stages, deletion was the most prevalent general type (34.7%), followed by insertion (27.0%). The most frequent semantic type was general English. Medication was the most common clinical semantic type in the original SR transcriptions, while diagnosis was most common in the transcriptionist-edited and signed versions. Transcriptionists made stylistic changes to 180 (82.9%) of the 217 notes and rearranged the contents of 37 notes (17.1%), usually at the request of the dictating physician. Physicians made stylistic changes to 71 notes (32.7%) and rearranged the contents of 8 notes (3.7%). Finally, there were 59 notes (27.2%) to which the signing physician added information, and 37 (17.1%) from which the physician deleted information.

of 37 notes (17.1%), usually at the request of the dictating physician. Physicians made stylistic changes to 71 notes (32.7%) and rearranged the contents of 8 notes (3.7%). Finally, there were 59 notes (27.2%) to which the signing physician added information, and 37 (17.1%) from which the physician deleted information. The original SR transcriptions of notes created at institution A had a higher mean rate of errors compared with those created at institution B (Table 4), although not significantly, with mean error rates of 7.6% and 6.6%, respectively (difference, 1.0%; 95% CI, −0.2% to 2.8%; P = .10). Following human revision, however, institution A’s notes had lower error rates compared with notes from institution B, with mean error rates of 0.3% (institution A) and 0.7% (institution B) (difference, −0.3%; 95% CI, −0.63% to −0.04%; P = .03) after revision by an MT, and of 0.2% (institution A) and 0.6% (institution B) (difference, −0.4%; 95% CI, −0.7% to −0.2%; P = .003) after author review. Errors in original SR transcriptions occurred at similar frequencies for male and female physicians, with 7.5 and 7.7 mean errors per 100 words, respectively (difference, 0.2%; 95% CI, −1.2% to 1.6%; P = .78). A modest negative correlation was observed between age and error rate in original SR transcriptions (r = −0.20; 95% CI, −0.35 to −0.04; P = .01), with average error rates decreasing as physician age increased.

, with 7.5 and 7.7 mean errors per 100 words, respectively (difference, 0.2%; 95% CI, −1.2% to 1.6%; P = .78). A modest negative correlation was observed between age and error rate in original SR transcriptions (r = −0.20; 95% CI, −0.35 to −0.04; P = .01), with average error rates decreasing as physician age increased. Table 4. Mean Error Rates Compared by Institution, Note Type, Physician Sex, and Specialty Group 1 (No.) vs Group 2 (No.) SR MT SN Error Rate, Mean (SD) P Value Error Rate, Mean (SD) P Value Error Rate, Mean (SD) P Value Group 1 Group 2 Group 1 Group 2 Group 1 Group 2 Institutiona Institution A vs institution B 7.6 (5.1) 6.6 (3.4) .10 0.3 (0.5) 0.7 (1.0) .03 0.2 (0.4) 0.6 (1.0) .003 Note type Discharge summaries (75) vs others (142) 8.9 (4.6) 6.6 (4.7) <.001 0.4 (0.7) 0.4 (0.6) .51 0.4 (0.7) 0.2 (0.5) .08 Office notes (83) vs others (134) 7.0 (5.0) 7.6 (4.7) .33 0.3 (0.5) 0.4 (0.7) .28 0.1 (0.3) 0.4 (0.7) <.001 Operative notes (59) vs others (158) 6.1 (4.3) 7.9 (4.9) .01 0.4 (0.7) 0.4 (0.9) .72 0.4 (0.7) 0.3 (0.5) .34 Sexb Female (69) vs male (138) 7.7 (4.8) 7.5 (4.9) .78 0.5 (6.4) 0.4 (6.7) .65 0.3 (0.6) 0.3 (0.6) .96 Specialtyc General, internal, or family (53) vs others (138) 8.7 (4.6) 6.9 (4.7) .02 0.5 (0.8) 0.4 (0.6) .60 0.4 (0.7) 0.3 (0.6) .38 Surgery (63) vs others (128) 6.0 (4.3) 8.1 (4.7) .002 0.4 (0.5) 0.4 (0.7) .49 0.3 (0.5) 0.3 (0.7) .55 Subspecialties (75) vs others (116) 7.7 (4.8) 7.2 (4.6) .48 0.4 (0.7) 0.4 (0.6) .93 0.3 (0.6) 0.3 (0.6) .70 Abbreviations: MT, medical transcriptionist–edited notes; SN, signed notes; SR, automatic transcriptions by the speech recognition system.

6.0 (4.3) 8.1 (4.7) .002 0.4 (0.5) 0.4 (0.7) .49 0.3 (0.5) 0.3 (0.7) .55 Subspecialties (75) vs others (116) 7.7 (4.8) 7.2 (4.6) .48 0.4 (0.7) 0.4 (0.6) .93 0.3 (0.6) 0.3 (0.6) .70 Abbreviations: MT, medical transcriptionist–edited notes; SN, signed notes; SR, automatic transcriptions by the speech recognition system. a Number omitted to preserve deidentification. b Physician sex was missing for 10 notes. c Physician specialty was missing for 26 notes. Across all of the original SR transcriptions, discharge summaries had higher error rates than other note types (8.9% vs 6.6%; difference, 2.3%; 95% CI, 1.0%-3.6%; P < .001), and operative notes had lower error rates (6.1% vs 7.9%; difference, 1.8%; 95% CI, 0.4%-3.2%; P = .01). Likewise, notes dictated by surgeons had lower error rates than physicians of other specialties (6.0% vs 8.1%; difference, 2.2%; 95% CI, 0.8%-3.5%; P = .002). There was no significant difference in word counts between notes dictated by surgeons and notes dictated by physicians of other specialties; nor was there a significant difference in word counts between operative notes and other note types. No correlation was observed between word count and error rate at any stage (SR: r = 0.006; 95% CI, −0.13 to 0.14; P = .93 vs MT: r = 0.03; 95% CI, −0.1 to 0.2; P = .67 vs SN: r = 0.01; 95% CI, −0.1 to 0.1; P = .85).

ies; nor was there a significant difference in word counts between operative notes and other note types. No correlation was observed between word count and error rate at any stage (SR: r = 0.006; 95% CI, −0.13 to 0.14; P = .93 vs MT: r = 0.03; 95% CI, −0.1 to 0.2; P = .67 vs SN: r = 0.01; 95% CI, −0.1 to 0.1; P = .85). Discussion This study is among the first, to our knowledge, to analyze errors at the different processing stages of documents created with a back-end SR system. We defined a comprehensive schema to systematically classify and analyze errors across multiple note types. The comparatively large sample and the variety of clinicians and hospitals represented increase the robustness of our findings vs those of previous studies. Of 33 studies included in 2 recent systematic reviews of SR use in health care,5,7 most evaluated the productivity of SR-assisted dictation compared with traditional transcription and typing. Only 16, of which 7 were conducted in the United States, reported error or accuracy rates.20,21,22,23,24,25,26 Many had small samples (14 included <10 clinicians) and only reviewed notes from 1 medical specialty,27,28,29,30 usually radiology.20,31,32,33 In contrast, our findings are based on dictations from 144 physicians across a wide range of clinical settings and at 2 geographically distinct institutions.

0,21,22,23,24,25,26 Many had small samples (14 included <10 clinicians) and only reviewed notes from 1 medical specialty,27,28,29,30 usually radiology.20,31,32,33 In contrast, our findings are based on dictations from 144 physicians across a wide range of clinical settings and at 2 geographically distinct institutions. Speech recognition technology is being adopted at increasing rates at health care institutions across the country owing to its many advantages. Documentation is one of the most time-consuming parts of using EHR technology, and SR technology promises to improve documentation efficiency and save clinicians time. In back-end systems, SR software automatically converts clinicians’ dictations to text that MTs can quickly review and edit, reducing turnaround time and increasing productivity; however, it should be noted that turnaround times are typically stipulated in the contract with the transcriptionist vendor and may vary widely for this reason. Additionally, some notes remained unsigned for weeks or months, although they can still be viewed by other EHR users during this time. Many hospitals are adopting front-end dictation systems, where clinicians must review and edit their notes themselves, either as they dictate or at a later time. Clinicians face pressure to decrease documentation time and often only superficially review their notes before signing them.9 Fully shifting the editing responsibility from transcriptionists to clinicians may lead to increased documentation errors if clinicians are unable to adequately review their notes.

a later time. Clinicians face pressure to decrease documentation time and often only superficially review their notes before signing them.9 Fully shifting the editing responsibility from transcriptionists to clinicians may lead to increased documentation errors if clinicians are unable to adequately review their notes. Basma et al32 reported that SR-generated breast imaging reports were 8 times more likely than conventionally dictated reports (23% vs 4% before adjusting confounders) to contain major errors that could affect the understanding of a report or alter patient care. Our study also identified errors involving clinical information that could have such unintended impacts. For example, we found an SN that incorrectly listed a patient as having a “grown mass” instead of a “groin mass” because of an uncorrected error in the original SR transcription. We also found evidence suggesting some clinicians may not review their notes thoroughly, if they do so at all. Transcriptionists typically mark portions of the transcription that are unintelligible in the original audio recording with blank spaces (eg, ??__??), which the physician is then expected to fill in. However, we found 16 SNs (7.4%) that retained these marks, and in 3 instances, the missing word was discovered to be clinically significant. While additional medical record review found no evidence for the persistence of these omissions in subsequent documentation, such a risk still exists.

is then expected to fill in. However, we found 16 SNs (7.4%) that retained these marks, and in 3 instances, the missing word was discovered to be clinically significant. While additional medical record review found no evidence for the persistence of these omissions in subsequent documentation, such a risk still exists. Although adoption of SR technology is intended to ease some of the burden of documentation, that even readily apparent pieces of information at times remain uncorrected raises concerns about whether physicians have sufficient time and resources to review their dictated notes, even to a superficial degree. As previously mentioned, a recent study in Australia reported that emergency department clinicians needed 18% more time for documentation when using SR than when using a keyboard and mouse.4 The authors also observed 4.3 times as many errors in SR-generated documents compared with those created with a keyboard and mouse. We observed a similar trend; the SR transcriptions we reviewed had a mean (SD) of 7.4 (4.8) errors per 100 words, while in an earlier study we found errors in typed notes at a rate of 0.45 errors per 100 words. However, SR technology is continually improving, while clinicians’ skills with and attention to keyboard and mouse documentation may not be improving at a similar rate.

wed had a mean (SD) of 7.4 (4.8) errors per 100 words, while in an earlier study we found errors in typed notes at a rate of 0.45 errors per 100 words. However, SR technology is continually improving, while clinicians’ skills with and attention to keyboard and mouse documentation may not be improving at a similar rate. In general, health information technology and the EHR have introduced a number of potential sources for error. A recent study found higher rates of errors in the EHR than in paper records, possibly attributable to EHR-specific functionality such as templates and the ability to copy and paste text.34 Taken together, these findings demonstrate the necessity of further studies investigating clinicians’ use of and satisfaction with SR technology, its ability to integrate with clinicians’ existing workflows, and its effect on documentation quality and efficiency compared with other documentation methods. In addition, these findings indicate a need not only for clinical quality assurance and auditing programs, but also for clinician training and education to raise awareness of these errors and strategies to reduce them.

ws, and its effect on documentation quality and efficiency compared with other documentation methods. In addition, these findings indicate a need not only for clinical quality assurance and auditing programs, but also for clinician training and education to raise awareness of these errors and strategies to reduce them. Limitations The notes in our analysis were all created using the same back-end SR service. Furthermore, while larger in scale than many previous studies, our analysis was still conducted on a relatively small set of notes created in a limited number of clinical settings. As such, our findings may not be generalizable to SR-assisted documentation as a whole. Additionally, sex and specialty information was unavailable in the data sources to which we had access for 10 and 26 physicians, respectively. These missing data may limit our ability to draw conclusions about the effect these characteristics may have on error rates.

neralizable to SR-assisted documentation as a whole. Additionally, sex and specialty information was unavailable in the data sources to which we had access for 10 and 26 physicians, respectively. These missing data may limit our ability to draw conclusions about the effect these characteristics may have on error rates. Despite the iterative testing and revision that preceded the annotation schema’s finalization, there are some additional error types we may wish to include in subsequent work. For example, the lack of a body location semantic type resulted in some confusion over how errors involving these words should be annotated, potentially leading to inconsistent annotations. In some cases, it may have been useful to divide an existing type into more granular subtypes. In particular, the stop word semantic type, which was included to distinguish short, frequently used words from other general English terms, may have inadvertently masked the true prevalence of highly specific but still commonly observed errors, such as those involving pronouns (eg, he or she) or negations. Because of the time-intensive nature of the annotation task, we calculated interannotator agreement using only a small subset (33 of 651 [5.0%]), rather than requiring both individuals to annotate the full set of notes. This subset also included primarily notes that had been edited by MTs (26 of 33 [78.7%]), owing to the fact that errors in these notes are often more difficult to identify and may generate more disagreement.

nly a small subset (33 of 651 [5.0%]), rather than requiring both individuals to annotate the full set of notes. This subset also included primarily notes that had been edited by MTs (26 of 33 [78.7%]), owing to the fact that errors in these notes are often more difficult to identify and may generate more disagreement. Future Directions These findings lay the groundwork for many subsequent research activities. First, the developed schema can be used to annotate more notes, obtained from a wider variety of clinical domains, to create a robust corpus of errors in clinical documents created with SR technology. The benefits of such a corpus are considerable. Not only will it allow for more reliable error prevalence estimates, but it can also serve as training data for the development of an automatic error detection system. With the rapid adoption of SR in clinical settings, there is a need for automated methods based on natural language processing for identifying and correcting errors in SR-generated text. Such methods are vital to ensuring the effective use of clinicians’ time and to improving and maintaining documentation quality, all of which can, in turn, increase patient safety. Conclusions Seven in 100 words in unedited clinical documents created with SR technology involve errors and 1 in 250 words contains clinically significant errors. The comparatively low error rate in signed notes highlights the crucial role of manual editing and review in the SR-assisted documentation process.

Introduction Type 1 diabetes (T1D) treatment is centered around the improvement and maintenance of tight glycemic control, as assessed by levels of hemoglobin A1c (HbA1c), to prevent acute and chronic diabetes-related complications.1,2,3 Glycemic control can vary considerably from diabetes onset through adolescence,4,5,6 where fluctuations are known to occur during puberty3,4,7,8,9,10,11,12 and during early adulthood. Poorer glycemic control during early adulthood or from childhood to young adulthood has been attributed to a lack of continuity in diabetes-related clinical care4,11,12 as well as changes in self-care as children and adolescents with T1D grow into adulthood.9,10,13 However, glycemic control in youth and young adults with T1D is critical, as a higher average HbA1c level in this period of development is associated with impaired growth as well as diabetic complications.14,15,16,17

,11,12 as well as changes in self-care as children and adolescents with T1D grow into adulthood.9,10,13 However, glycemic control in youth and young adults with T1D is critical, as a higher average HbA1c level in this period of development is associated with impaired growth as well as diabetic complications.14,15,16,17 In cross-sectional studies of adolescents and young adults, glycemic control differs by racial and ethnic subgroups.18 African American, American Indian, Hispanic, and Asian or Pacific Islander youth with T1D are more likely to have higher HbA1c levels compared with non-Hispanic white youth.19 In longitudinal studies, nonwhite youth with T1D have increased markers of poor prognosis at diagnosis and 3 years following diagnosis, including higher HbA1c levels, more frequent diabetic ketoacidosis, and severe hypoglycemia.20 A constellation of sociodemographic factors related to race/ethnicity and glycemic control have been proposed, ranging from family dynamics, depressive symptoms, and quality of life13,21,22,23,24,25 to diabetes regimen.26,27,28 The role of socioeconomic position as a mediator of racial/ethnic associations remains controversial.28,29,30,31 Additionally, health care–specific factors such as disparities in health literacy, diabetes-related knowledge, or access to health care are known to contribute to pediatric health disparity but have not been well explored in T1D.32,33

osition as a mediator of racial/ethnic associations remains controversial.28,29,30,31 Additionally, health care–specific factors such as disparities in health literacy, diabetes-related knowledge, or access to health care are known to contribute to pediatric health disparity but have not been well explored in T1D.32,33 Latent class trajectory modeling has been used to identify subgroups who share a similar trajectory of HbA1c over time.34 Few studies have examined whether racial/ethnic disparities in glycemic control persist over time from childhood into young adulthood among individuals with T1D. Our objective was to first visualize major trajectories of glycemic control from childhood into young adulthood using all data from youth of all racial and ethnic groups and to then characterize specific associations between race/ethnicity and distinct longitudinal patterns of glycemic control. Our hypothesis was that non-Hispanic black and Hispanic youth would be more likely than non-Hispanic white youth to have unfavorable trajectory patterns representing poor glycemic control and that this association may be mediated by clinical factors such as diabetes regimen26,27,28 and by socioeconomic position.29,30,31

ur hypothesis was that non-Hispanic black and Hispanic youth would be more likely than non-Hispanic white youth to have unfavorable trajectory patterns representing poor glycemic control and that this association may be mediated by clinical factors such as diabetes regimen26,27,28 and by socioeconomic position.29,30,31 Methods Study Population The SEARCH for Diabetes in Youth study began in 2000 with an overarching objective to describe the incidence and prevalence of childhood diabetes among the 5 major racial and ethnic groups in the United States.35 Individuals with diabetes diagnosed before age 20 years were identified from a population-based incidence registry network at 5 US sites (South Carolina; Cincinnati, Ohio, and surrounding counties; Colorado with southwestern Native American sites; Seattle, Washington, and surrounding counties; and Kaiser Permanente, southern California).36 Patients were newly diagnosed with T1D in 2002 through 2005. Patients who could be contacted were asked to complete a short survey and recruited for a baseline visit. If they completed the first visit, they were asked to return for visits at 12, 24, and 60 months to measure risk factors for diabetes complications (Figure 1A). A subset of participants who were aged 10 years and older and had at least 5 years of diabetes duration were recruited for a follow-up cohort visit between 2012 and 2015. The subset of youth who were included in the SEARCH cohort visit were not significantly different from all other youth diagnosed between the years of 2002 and 2008 in terms of average age at diabetes onset, distribution of sex or race and ethnicity, or clinical measures.14

follow-up cohort visit between 2012 and 2015. The subset of youth who were included in the SEARCH cohort visit were not significantly different from all other youth diagnosed between the years of 2002 and 2008 in terms of average age at diabetes onset, distribution of sex or race and ethnicity, or clinical measures.14 Figure 1. Study Design and Sample Recruitment A, Study design of the SEARCH cohort study. B, Flowchart depicting participants in this report, including reasons for exclusion. The final sample included 1313 youths with type 1 diabetes. BV indicates baseline visit.

follow-up cohort visit between 2012 and 2015. The subset of youth who were included in the SEARCH cohort visit were not significantly different from all other youth diagnosed between the years of 2002 and 2008 in terms of average age at diabetes onset, distribution of sex or race and ethnicity, or clinical measures.14 Figure 1. Study Design and Sample Recruitment A, Study design of the SEARCH cohort study. B, Flowchart depicting participants in this report, including reasons for exclusion. The final sample included 1313 youths with type 1 diabetes. BV indicates baseline visit. Inclusion criteria for these analyses consisted of youth diagnosed with T1D between 2002 and 2005. Type 1 diabetes was based on the clinical diagnosis made by a physician or other health care professional at onset and was collected from these health care professionals or abstracted from medical records. Youth with a clinical diagnosis of type 1a, type 1b, or type 1 diabetes were included. Youth who had fewer than 3 measures of HbA1c from research visits during 6.1 to 13.3 years of follow-up were excluded (n = 618). Excluded individuals were not different with regard to HbA1c measures using available data from the study baseline and the cohort visit. The final study sample included 1313 youths with T1D (Figure 1B). The study was approved by institutional review boards with jurisdiction; the parent, the participant, or both provided written consent or assent for all participants (consent of ≥1 parent or legal guardian was required for participants aged <18 years). The study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.

with jurisdiction; the parent, the participant, or both provided written consent or assent for all participants (consent of ≥1 parent or legal guardian was required for participants aged <18 years). The study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline. Research Visits Trained personnel administered questionnaires; measured height, weight, and blood pressure; and obtained blood samples. Body mass index was defined as weight (kilograms) divided by height (meters squared) and converted to a z score.37 A blood draw occurred after an 8-hour overnight fast, and medications, including short-acting insulin, were withheld the morning of the visit. Laboratory Measures Blood samples were obtained under conditions of metabolic stability, defined as no episodes of diabetic ketoacidosis in the preceding month and the absence of fever and acute infections. They were processed locally and shipped within 24 hours to the central laboratory (Northwest Lipid Metabolism and Diabetes Research Laboratories). Hemoglobin A1c was measured by a dedicated ion exchange high-performance liquid chromatography instrument (TOSOH Bioscience).

and the absence of fever and acute infections. They were processed locally and shipped within 24 hours to the central laboratory (Northwest Lipid Metabolism and Diabetes Research Laboratories). Hemoglobin A1c was measured by a dedicated ion exchange high-performance liquid chromatography instrument (TOSOH Bioscience). Other Measures Self-reported race and ethnicity were collected based on questions modeled after the 2000 US Census38 and categorized as non-Hispanic white, non-Hispanic black, Hispanic, and other (Asian, Native American, Pacific Islander, other, and unknown). Although the US Census accommodates reporting of multiple races, the SEARCH study did not have sufficient participant numbers to allow evaluation of separate categories of reported multiple-race groups39 and used the National Center for Health Statistics plurality approach, in which data from a study designed to address multiple-race reporting was used to determine which single-race category should be assigned for specific combinations of multiple races reported.38

separate categories of reported multiple-race groups39 and used the National Center for Health Statistics plurality approach, in which data from a study designed to address multiple-race reporting was used to determine which single-race category should be assigned for specific combinations of multiple races reported.38 Insulin regimen was based on mode of insulin delivery, classified as pumps, long-acting with rapid-acting insulin injections with 3 or more injections per day, and any other form of multiple daily injections. Insulin dose was reported as units per kilogram of body weight. Frequency of self-monitoring of blood glucose was self-reported and categorized as less than 1 time per day, 1 to 3 times per day, and 4 or more times per day. Health insurance type was classified as none, private, Medicaid, or other. Parental education was based on the highest educational level attained by either parent and classified as less than high school degree, high school graduate, some college through associate’s degree, and bachelor’s degree or more. Household structure was classified as 2 parent, single parent, or other. Receipt of diabetes care was based on reported number of visits with prespecified diabetes health care professionals, including pediatric endocrinologists, adult diabetologists, and nurse diabetes educators, in the previous 6 months and classified based on the distribution: 0 to 1 visit, 2 to 3 visits, 4 to 5 visits, and 6 or more visits. Receipt of nondiabetes care was based on reported number of visits with prespecified nondiabetes health care professionals (pediatrician, family practice physician, general practice physician, internist, nurse practitioner or physician assistant, traditional healer, dietician, optometrist or ophthalmologist, and psychiatrist, psychologist, or mental health counselor) in the previous 6 months and classified as 0 to 1 visit, 2 to 3 visits, 4 to 6 visits, and 7 or more visits. Satisfaction with diabetes care was based on the response to the question, “How would you rate your diabetes care overall?” (possible responses were excellent, good, fair, poor, and not applicable).

h counselor) in the previous 6 months and classified as 0 to 1 visit, 2 to 3 visits, 4 to 6 visits, and 7 or more visits. Satisfaction with diabetes care was based on the response to the question, “How would you rate your diabetes care overall?” (possible responses were excellent, good, fair, poor, and not applicable). Statistical Analysis We used group-based trajectory modeling to identify trajectories of HbA1c among youth with T1D using duration of diabetes (months) as the time scale via the PROC TRAJ macro of SAS statistical software version 9.4 (SAS Institute Inc), which fits a semiparametric (discrete mixture) model for longitudinal data using the maximum-likelihood method.40,41,42,43,44 Trajectory analysis uses all available data for a participant and is robust to data that are missing at random. Details about trajectory analysis have been described elsewhere.43,44 The optimal number of groups was determined based on Bayesian information criterion and having at least 5% of the sample in the smallest trajectory group. We named the trajectories based on the baseline HbA1c value (from the initial research visit) and shape of the trajectory over the follow-up visits. We then calculated the posterior predicted probability for each participant of being a member of each trajectory group given his or her observed HbA1c pattern. Participants were assigned to the trajectory group for which they had the greatest posterior probability for group membership. Multinomial regression was used to assess the association of race/ethnicity (non-Hispanic white vs non-Hispanic black vs Hispanic) with HbA1c trajectory group membership. Youths who reported Asian or Pacific Islander, Native American, other, and unknown race/ethnicity (n = 34) were excluded from multinomial modeling. Non-Hispanic white was designated as the referent group.

on of race/ethnicity (non-Hispanic white vs non-Hispanic black vs Hispanic) with HbA1c trajectory group membership. Youths who reported Asian or Pacific Islander, Native American, other, and unknown race/ethnicity (n = 34) were excluded from multinomial modeling. Non-Hispanic white was designated as the referent group. All covariates were measured at baseline. Model 1 was unadjusted. Model 2 was adjusted for demographic factors (sex, age at diagnosis, and clinic site). Model 3 was additionally adjusted for clinical variables (body mass index z score, insulin regimen, insulin dose, and frequency of self-monitoring of blood glucose). Model 4 was further adjusted for socioeconomic position (highest parental education, household structure, and health insurance type). Given previous findings of health inequity,45 we tested for sex- and age-related subgroups who may be particularly vulnerable to the effects of heath inequity. Modification of race/ethnicity effects by age and sex was tested by adding an interaction term (race/ethnicity × sex and race/ethnicity × age at diagnosis, respectively) to model 4. The nature of the modification was explored in models stratified by sex and the median age of diagnosis (9 years old). Because of limited sample size, for stratified analyses, race/ethnicity was categorized into non-Hispanic white and other (defined as non-Hispanic black, Hispanic, Asian or Pacific Islander, Native American, other, and unknown).

ion was explored in models stratified by sex and the median age of diagnosis (9 years old). Because of limited sample size, for stratified analyses, race/ethnicity was categorized into non-Hispanic white and other (defined as non-Hispanic black, Hispanic, Asian or Pacific Islander, Native American, other, and unknown). All analyses were completed in SAS software in 2017. Statistical significance was based on a 2-sided P value of .05. Descriptive analyses used the mean and standard deviation or median and interquartile range (IQR) for nonnormal distributions and for continuous variables and frequencies to describe categorical variables. The means and frequencies of demographic and clinical characteristics were compared using χ2 test for categorical variables and analysis of variance or Kruskal-Wallis test for continuous variables. Results The sample of 1313 youths with T1D was 49.3% female (647 patients); 77.0% were non-Hispanic white (1011 patients); 10.7%, Hispanic (140 patients); 9.8%, non-Hispanic black (128 patients); and 2.6%, other race/ethnicity (34 patients) (Table 1). At the baseline visit, the mean (SD) age was 9.7 (4.3) years and the mean (SD) diabetes duration was 9.2 (6.3) months. Group-based trajectory modeling identified 3 distinct HbA1c trajectories over a mean (SD) follow-up of 108 (16) months (9.0 [1.4] years) of diabetes duration: group 1, low baseline and mild increases (50.7% [666 patients]); group 2, moderate baseline and moderate increases (41.7% [548 patients]); and group 3, moderate baseline and major increases (7.5% [99 patients]) (Figure 2).

rajectories over a mean (SD) follow-up of 108 (16) months (9.0 [1.4] years) of diabetes duration: group 1, low baseline and mild increases (50.7% [666 patients]); group 2, moderate baseline and moderate increases (41.7% [548 patients]); and group 3, moderate baseline and major increases (7.5% [99 patients]) (Figure 2). Table 1. Baseline Characteristics of 1313 Participants With Type 1 Diabetes by Hemoglobin A1c Trajectory Group Characteristic No.

rajectories over a mean (SD) follow-up of 108 (16) months (9.0 [1.4] years) of diabetes duration: group 1, low baseline and mild increases (50.7% [666 patients]); group 2, moderate baseline and moderate increases (41.7% [548 patients]); and group 3, moderate baseline and major increases (7.5% [99 patients]) (Figure 2). Table 1. Baseline Characteristics of 1313 Participants With Type 1 Diabetes by Hemoglobin A1c Trajectory Group Characteristic No. (%) P Valuea Total Participants (N = 1313) Group 1: Low Baseline and Mild Increases (n = 666) Group 2: Moderate Baseline and Moderate Increases (n = 548) Group 3: Moderate Baseline and Major Increases (n = 99) Age at diagnosis, mean (SD), y 8.9 (4.2) 8.8 (4.5) 8.5 (3.9) 11.3 (3.5) <.001 Age at baseline, mean (SD), y 9.7 (4.3) 9.6 (4.5) 9.3 (3.9) 12.2 (3.5) <.001 Diabetes duration, mean (SD), mo 9.2 (6.3) 9.0 (6.4) 9.3 (6.1) 10.4 (6.4) .13 Female 647 (49.3) 316 (47.5) 280 (51.1) 51 (51.5) .40 Nonwhite race/ethnicityb 302 (23.0) 102 (15.3) 153 (27.9) 47 (47.5) <.001 Race/ethnicityb Non-Hispanic white 1011 (77.0) 564 (84.7) 395 (72.1) 52 (52.5) <.001 Non-Hispanic black 128 (9.8) 34 (5.1) 69 (12.6) 25 (25.3) Hispanic 140 (10.7) 56 (8.4) 67 (12.2) 17 (17.2) Other 34 (2.6) 12 (1.8) 17 (3.1) 5 (5.1) Parental education Less than high school 48 (3.7) 20 (3.0) 20 (3.7) 8 (8.1) <.001 High school graduate 180 (13.8) 61 (9.2) 93 (17.1) 26 (26.3) Some college (through associate’s degree) 441 (33.8) 184 (27.8) 219 (40.3) 38 (38.4) Bachelor’s degree or more 636 (48.7) 397 (60.0) 212 (39.0) 27 (27.3) Insurance None 19 (1.5) 8 (1.2) 8 (1.5) 3 (3.0) <.001 Private 1052 (80.7) 586 (88.4) 402 (74.3) 64 (64.7) Medicaid 211 (16.2) 61 (9.2) 119 (22.0) 31 (31.3) Other 21 (1.6) 8 (1.2) 12 (2.2) 1 (1.1) Family structure Two-parent household 961 (73.6) 543 (81.9) 366 (67.4) 52 (52.5) <.001 Single-parent household 311 (23.8) 109 (16.4) 161 (29.7) 41 (41.4) Other structure 33 (2.53) 11 (1.7) 16 (3.0) 6 (6.1) Insulin regimen Pump 106 (8.15) 67 (10.1) 36 (6.6) 3 (3.1) .01 Long with short or rapid insulin, ≥3 times/d 418 (32.1) 225 (33.9) 164 (30.3) 29 (29.6) Long with other combinationc 779 (59.8) 371 (56.0) 342 (63.1) 66 (67.4) Insulin dose, mean (SD), units/kg 0.63 (0.42) 0.59 (0.46) 0.66 (0.38) 0.73 (0.38) .001 Blood glucose monitoring, times/d <1 10 (0.8) 14 (2.1) 11 (2.0) 4 (4.0) <.001 1-3 148 (11.5) 64 (9.6) 58 (10.6) 26 (26.5) ≥4 1134 (88.8) 588 (88.3) 478 (87.4) 68 (70.4) Body mass index z score, mean (SD) 0.58 (0.97) 0.40 (0.92) 0.66 (1.00) 0.68 (1.10) .02 Diabetes care visits in past 6 mo, No.d Mean (SD)d 3.9 (2.9) 3.9 (2.9) 4.1 (3.0) 3.6 (2.6) .31 0-1 173 (13.2) 95 (14.3) 65 (11.9) 1

4.0) <.001 1-3 148 (11.5) 64 (9.6) 58 (10.6) 26 (26.5) ≥4 1134 (88.8) 588 (88.3) 478 (87.4) 68 (70.4) Body mass index z score, mean (SD) 0.58 (0.97) 0.40 (0.92) 0.66 (1.00) 0.68 (1.10) .02 Diabetes care visits in past 6 mo, No.d Mean (SD)d 3.9 (2.9) 3.9 (2.9) 4.1 (3.0) 3.6 (2.6) .31 0-1 173 (13.2) 95 (14.3) 65 (11.9) 1 3 (13.1) .12 2-3 479 (36.5) 228 (34.2) 211 (38.5) 40 (40.4) 4-5 383 (29.2) 211 (31.7) 142 (25.9) 30 (30.30) ≥6 278 (21.2) 132 (19.8) 130 (23.7) 16 (16.2) Other care visits in past 6 mo, No.d Mean (SD)d 5.0 (4.1) 4.8 (3.9) 5.1 (4.3) 5.0 (4.5) .44 0-1 175 (13.3) 83 (12.5) 73 (13.3) 19 (19.2) .23 2-3 384 (29.3) 207 (31.1) 153 (27.9) 24 (24.2) 4-6 424 (33.1) 225 (33.8) 182 (33.2) 27 (27.3) ≥7 320 (24.4) 151 (22.7 140 (25.6) 29 (29.3) Satisfaction with diabetes caree Excellent 938 (72.4) 505 (77.2) 382 (70.6) 51 (54.8) <.001 Good 288 (22.4) 127 (19.4) 133 (24.6) 28 (30.1) Fair 49 (3.8) 16 (2.5) 21 (3.9) 12 (12.9) Poor 5 (0.4) 1 (0.2) 3 (0.6) 1 (1.1) a P values based on use of χ2 test and analysis of variance or Kruskal-Wallis test, as appropriate based on model assumptions. b Self-reported race and ethnicity were collected using 2000 US Census questions. White was defined as non-Hispanic white. Nonwhite was defined as non-Hispanic black, Hispanic, or other. Other was defined as Asian or Pacific Islander, Native American, other, or unknown. c Includes 2 or more times per day or any insulin combination (excluding long), 3 or more times per day or any insulin(s) taken once per day, or any insulin combination (excluding long) 2 or more times per day.

b Self-reported race and ethnicity were collected using 2000 US Census questions. White was defined as non-Hispanic white. Nonwhite was defined as non-Hispanic black, Hispanic, or other. Other was defined as Asian or Pacific Islander, Native American, other, or unknown. c Includes 2 or more times per day or any insulin combination (excluding long), 3 or more times per day or any insulin(s) taken once per day, or any insulin combination (excluding long) 2 or more times per day. d Diabetes care measured by frequency of visits with pediatric endocrinology, adult diabetologist, or nurse diabetes educator in the previous 6 months. Other care measured by frequency of visits with nondiabetes caregivers. Data are self-reported. e Based on response to the question, “How would you rate your diabetes care overall?” Possible answers were excellent, good, fair, poor, and not applicable. Figure 2. Trajectories of Hemoglobin A1c in 1313 Patients With Type 1 Diabetes in the SEARCH for Diabetes in Youth Study Group-based trajectory modeling identified 3 distinct hemoglobin A1c trajectories over a mean type 1 diabetes duration of 108 months. To convert hemoglobin A1c to proportion of total hemoglobin, multiply by 0.01.

jectories of Hemoglobin A1c in 1313 Patients With Type 1 Diabetes in the SEARCH for Diabetes in Youth Study Group-based trajectory modeling identified 3 distinct hemoglobin A1c trajectories over a mean type 1 diabetes duration of 108 months. To convert hemoglobin A1c to proportion of total hemoglobin, multiply by 0.01. The prevalence of black and Hispanic youth was the highest in group 3 and the lowest in group 1 (non-Hispanic black patients made up 5.1% of group 1, 12.6% of group 2, and 25.3% of group 3; Hispanic patients made up 8.4% of group 1, 12.2% of group 2, and 17.2% of group 3). For non-Hispanic black patients, the difference between group 1 and group 2 was 7.5% (95% CI, 4.2%-10.7%; P < .001); between group 1 and group 3, 20.2% (95% CI, 11.4%-28.9%; P < .001); and between group 2 and group 3, 12.6% (95% CI, 3.7%-21.7%; P = .001). For Hispanic patients, the difference between group 1 and group 2 was 3.8% (95% CI, 0.4%-7.3%; P = .03); between group 1 and group 3, 8.8% (95% CI, 1.0%-16.5%; P = .006); and between group 2 and group 3, 5.0% (95% CI, 3.0%-12.9%; P = .18). Group 3 was composed of 47.5% nonwhite youths (47 patients) (Table 1). Table 2 depicts the odds ratios (ORs) for non-Hispanic black and Hispanic vs non-Hispanic white race/ethnicity and HbA1c trajectory group in a series of sequentially adjusted models. Non-Hispanic black youth had 7.98 higher odds than non-Hispanic white youth of being in the highest HbA1c trajectory group relative to the lowest HbA1c trajectory group (unadjusted OR of non-Hispanic black race in group 3 vs group 1, 7.98; 95% CI, 4.42-14.38). After adjustment for baseline demographic characteristics, clinical factors, and socioeconomic position, non-Hispanic black youth had 4.54 times higher odds than non-Hispanic white youth of being in the highest HbA1c trajectory group relative to the lowest HbA1c trajectory group (adjusted OR [aOR] of non-Hispanic black race in group 3 vs group 1, 4.54; 95% CI, 2.08-9.89). Hispanic youth had 3.29 higher unadjusted odds than non-Hispanic white youth of being in the highest HbA1c trajectory group relative to the lowest HbA1c trajectory group (unadjusted OR of Hispanic ethnicity in group 3 vs group 1, 3.29; 95% CI, 1.78-6.08). Adjustment for baseline demographic characteristics, clinical factors, and socioeconomic position did not fully attenuate the association (aOR of Hispanic ethnicity in group 3 vs group 1, 2.24; 95% CI, 1.02-4.92).

A1c trajectory group (unadjusted OR of Hispanic ethnicity in group 3 vs group 1, 3.29; 95% CI, 1.78-6.08). Adjustment for baseline demographic characteristics, clinical factors, and socioeconomic position did not fully attenuate the association (aOR of Hispanic ethnicity in group 3 vs group 1, 2.24; 95% CI, 1.02-4.92). Adjustment for clinical variables diminished statistical significance associated with the moderate HbA1c trajectory (aOR of Hispanic ethnicity in group 2 vs group 1, 1.43; 95% CI, 0.90-2.27 vs unadjusted OR, 1.71; 95% CI, 1.17-2.49).

A1c trajectory group (unadjusted OR of Hispanic ethnicity in group 3 vs group 1, 3.29; 95% CI, 1.78-6.08). Adjustment for baseline demographic characteristics, clinical factors, and socioeconomic position did not fully attenuate the association (aOR of Hispanic ethnicity in group 3 vs group 1, 2.24; 95% CI, 1.02-4.92). Adjustment for clinical variables diminished statistical significance associated with the moderate HbA1c trajectory (aOR of Hispanic ethnicity in group 2 vs group 1, 1.43; 95% CI, 0.90-2.27 vs unadjusted OR, 1.71; 95% CI, 1.17-2.49). Table 2. Association of Black and Hispanic Race/Ethnicity, Compared With Non-Hispanic White Race/Ethnicity, With Hemoglobin A1c Trajectory Groups in 1011 Patients Modela Odds Ratio (95% CI) Black Race (n = 128)b Hispanic Ethnicity (n = 140)b Group 1: Low Baseline and Mild Increases Group 2: Moderate Baseline and Moderate Increases Group 3: Moderate Baseline and Major Increases Group 1: Low Baseline and Mild Increases Group 2: Moderate Baseline and Moderate Increases Group 3: Moderate Baseline and Major Increases Model 1 1 [Reference] 2.90 (1.88-4.46) 7.98 (4.42-14.38) 1 [Reference] 1.71 (1.17-2.49) 3.29 (1.78-6.08) Model 2 1 [Reference] 3.00 (1.92-4.67) 9.94 (5.15-19.20) 1 [Reference] 1.67 (1.08-2.58) 3.56 (1.75-7.21) Model 3 1 [Reference] 2.50 (1.54-4.05) 7.50 (3.68-15.26) 1 [Reference] 1.43 (0.90-2.27) 3.32 (1.60-6.91) Model 4 1 [Reference] 1.73 (1.04-2.90) 4.54 (2.08-9.89) 1 [Reference] 1.16 (0.71-1.89) 2.24 (1.02-4.92) a Model 1 was unadjusted. Model 2 was adjusted for demographic characteristics (age at diagnosis and clinic site). Model 3 further adjusted for body mass index (calculated as weight in kilograms divided by height in meters squared) z score, insulin regimen, insulin dose, and frequency of blood glucose monitoring. Model 4 further adjusted for socioeconomic position (maximum parental education, household structure, and health insurance type).

urther adjusted for body mass index (calculated as weight in kilograms divided by height in meters squared) z score, insulin regimen, insulin dose, and frequency of blood glucose monitoring. Model 4 further adjusted for socioeconomic position (maximum parental education, household structure, and health insurance type). b Self-reported race and ethnicity were collected using 2000 US Census questions and categorized as non-Hispanic white, non-Hispanic black, Hispanic, and other (Asian, Native American, Pacific Islander, other, and unknown). Respondents who self-reported as other were excluded from these analyses due to small sample size (n = 34).

rted race and ethnicity were collected using 2000 US Census questions and categorized as non-Hispanic white, non-Hispanic black, Hispanic, and other (Asian, Native American, Pacific Islander, other, and unknown). Respondents who self-reported as other were excluded from these analyses due to small sample size (n = 34). The association of race/ethnicity and HbA1c trajectory was modified by sex (P for interaction = .04) (Table 3). Nonwhite male patients had significantly elevated odds of membership in the highest HbA1c trajectory group (OR of group 3 vs group 1, 5.34; 95% CI, 2.16-13.2) and moderate HbA1c trajectory group (OR of group 2 vs group 1, 2.06; 95% CI, 1.18-3.57) relative to non-Hispanic white male patients. The associations were not significant in female patients (aOR of group 3 vs group 1, 1.48; 95% CI, 0.65-3.39 and aOR of group 2 vs group 1, 1.00; 95% CI, 0.61-1.64). The association of race/ethnicity and HbA1c trajectory was also modified by age at diagnosis (P for interaction = .02) (Table 3). Nonwhite youths diagnosed at or younger than 9 years had significantly elevated odds of membership in the highest HbA1c trajectory group (aOR of group 3 vs group 1, 5.37; 95% CI, 1.91-15.1) and the moderate HbA1c trajectory group (aOR of group 2 vs group 1, 2.04; 95% CI, 1.23-3.37). The association was not significant in youth who were diagnosed when they were older than 9 years (aOR of group 3 vs group 1, 1.65; 95% CI, 0.77-3.51 and aOR of group 2 vs group 1, 0.96; 95% CI, 0.55-1.65).

95% CI, 1.91-15.1) and the moderate HbA1c trajectory group (aOR of group 2 vs group 1, 2.04; 95% CI, 1.23-3.37). The association was not significant in youth who were diagnosed when they were older than 9 years (aOR of group 3 vs group 1, 1.65; 95% CI, 0.77-3.51 and aOR of group 2 vs group 1, 0.96; 95% CI, 0.55-1.65). Table 3. Association of Nonwhite Race/Ethnicity, Compared With Non-Hispanic White Race/Ethnicity, With Hemoglobin A1c Trajectory Group, Stratified by Sex and Age at Diagnosisa Modelb Odds Ratio (95% CI) P Value for Interaction Group 1: Low Baseline and Mild Increases Group 2: Moderate Baseline and Moderate Increases Group 3: Moderate Baseline and Major Increases Sex Female (n = 581) 1 [Reference] 1.00 (0.61-1.64) 1.48 (0.65-3.39) .04 Male (n = 593) 1 [Reference] 2.06 (1.18-3.57) 5.34 (2.16-13.2) Age at diagnosis, y 1 [Reference] ≤9 (n = 611) 1 [Reference] 2.04 (1.23-3.37) 5.37 (1.91-15.1) .02 >9 (n = 564) 1 [Reference] 0.96 (0.55-1.65) 1.65 (0.77-3.51) a Self-reported race and ethnicity were collected using 2000 US Census questions. White was defined as non-Hispanic white. Nonwhite was defined as non-Hispanic black, Hispanic, Asian or Pacific Islander, Native American, other, or unknown. b Fully adjusted for age at diagnosis, clinic site, maximum parental education, household structure, health insurance type, body mass index (calculated as weight in kilograms divided by height in meters squared) z score, insulin regimen, insulin dose, and frequency of blood glucose monitoring.

Table 3. Association of Nonwhite Race/Ethnicity, Compared With Non-Hispanic White Race/Ethnicity, With Hemoglobin A1c Trajectory Group, Stratified by Sex and Age at Diagnosisa Modelb Odds Ratio (95% CI) P Value for Interaction Group 1: Low Baseline and Mild Increases Group 2: Moderate Baseline and Moderate Increases Group 3: Moderate Baseline and Major Increases Sex Female (n = 581) 1 [Reference] 1.00 (0.61-1.64) 1.48 (0.65-3.39) .04 Male (n = 593) 1 [Reference] 2.06 (1.18-3.57) 5.34 (2.16-13.2) Age at diagnosis, y 1 [Reference] ≤9 (n = 611) 1 [Reference] 2.04 (1.23-3.37) 5.37 (1.91-15.1) .02 >9 (n = 564) 1 [Reference] 0.96 (0.55-1.65) 1.65 (0.77-3.51) a Self-reported race and ethnicity were collected using 2000 US Census questions. White was defined as non-Hispanic white. Nonwhite was defined as non-Hispanic black, Hispanic, Asian or Pacific Islander, Native American, other, or unknown. b Fully adjusted for age at diagnosis, clinic site, maximum parental education, household structure, health insurance type, body mass index (calculated as weight in kilograms divided by height in meters squared) z score, insulin regimen, insulin dose, and frequency of blood glucose monitoring. Discussion In a large, population-based multiethnic cohort of youth with T1D, we found 3 distinct HbA1c trajectories that deteriorated over a mean (SD) follow-up of 9.0 (1.4) years (range, 6.1-13.3 years) following diabetes diagnosis, reinforcing that early youth and the transition to adulthood are high-risk periods for worsening glycemic control.3,7,8 Black race and Hispanic ethnicity were associated with membership in the highest and most rapidly increasing (worsening) HbA1c trajectory group.

s (range, 6.1-13.3 years) following diabetes diagnosis, reinforcing that early youth and the transition to adulthood are high-risk periods for worsening glycemic control.3,7,8 Black race and Hispanic ethnicity were associated with membership in the highest and most rapidly increasing (worsening) HbA1c trajectory group. We tested the association of race/ethnicity with HbA1c trajectory by adjusting for other variables, including clinical factors and socioeconomic position. For example, prescribing practices may vary based on race/ethnicity27 and insulin pump use is known to be higher in white youth than non-Hispanic black or Hispanic youth.28 Lower socioeconomic position has been proposed as a major mediator of the association of race/ethnicity with health outcomes,29,30,31 including T1D complications, due to poorer self-management among persons whose socioeconomic conditions are less favorable.20,46 Despite adjustment for these known risk factors, black race remained significantly associated with HbA1c trajectory. Similarly, adjustment for demographic characteristics, clinical variables, and socioeconomic position did not fully attenuate the association of Hispanic ethnicity with the highest HbA1c trajectory, where the OR remained significantly elevated, suggesting remaining impact of inequity in this group. Evidence of disparity in glycemic control trajectory that exists particularly among nonwhite male patients and nonwhite youth with diabetes diagnosis at an early age (≤9 years) is consistent with previously reported patterns in acute glycemic complications that are more common among the youngest patients and male patients of all ages.47

ity in glycemic control trajectory that exists particularly among nonwhite male patients and nonwhite youth with diabetes diagnosis at an early age (≤9 years) is consistent with previously reported patterns in acute glycemic complications that are more common among the youngest patients and male patients of all ages.47 An important finding of the trajectory analysis was that the highest HbA1c trajectory subgroup also showed the highest mean HbA1c level at baseline, which occurred at a mean (SD) of 9.8 (6.3) months following diagnosis. This suggests that glycemic control obtained in the first year following diagnosis may confer information about longitudinal trends over time. Furthermore, the magnitude of racial/ethnic inequity over the longitudinal data are striking. Group 3 diverged over the follow-up period to give vastly different mean HbA1c measures at the cohort visit that may translate to significant increases in the risk for complications of diabetes based on evidence from the Diabetes Control and Complications Trial (DCCT) and the Epidemiology of Diabetes Interventions and Complications Study (EDIC).1,2,3,48 Disparity in glycemic control across trajectory groups in the present analyses even exceeds differences reported across groups of the DCCT/EDIC trial (which compared a median HbA1c of 7% of total hemoglobin in the intensive insulin treatment group with a median HbA1c of 9% of total hemoglobin in the conventional group), suggesting that those risk estimates may be conservative for youth who additionally face a longer period of disease-related exposures.49

/EDIC trial (which compared a median HbA1c of 7% of total hemoglobin in the intensive insulin treatment group with a median HbA1c of 9% of total hemoglobin in the conventional group), suggesting that those risk estimates may be conservative for youth who additionally face a longer period of disease-related exposures.49 Previous studies have shown that the migration status of parents is associated with glycemic control among youth with T1D.50 To address potential differences, we examined a subset of the sample with data on parental nativity (ie, US born vs foreign born) and found no significant differences across HbA1c trajectory groups. Adjustment for parental nativity did not attenuate the association of black race or Hispanic ethnicity with the moderate or highest HbA1c trajectory group, although the analysis is limited by small sample size (data not shown). Differences in youth and parental nativity status likely warrant future study in adequately powered samples.

arental nativity did not attenuate the association of black race or Hispanic ethnicity with the moderate or highest HbA1c trajectory group, although the analysis is limited by small sample size (data not shown). Differences in youth and parental nativity status likely warrant future study in adequately powered samples. Given the complexity of the study of race and health outcomes in the United States, in which health risks associated with race/ethnicity are not inherent but instead may signal underlying inequalities,51 we posit that our results may reflect health inequity in T1D operating at multiple levels. The social determinants of health operating outside of the health care system, including aspects of the physical environment, food security, social integration, barriers to health care,52 and complex patterns in health care utilization,53,54 may create race-based groups of individuals for whom glycemic control is challenged by inconsistencies in the availability of resources or support for T1D management. In general, adverse childhood experiences among nonwhite youth have been shown to result in a myriad of psychological and medical sequelae later in life.55

y create race-based groups of individuals for whom glycemic control is challenged by inconsistencies in the availability of resources or support for T1D management. In general, adverse childhood experiences among nonwhite youth have been shown to result in a myriad of psychological and medical sequelae later in life.55 There may also be modifiable aspects within the health care system, including racial/ethnic differences in the interpersonal dynamics of interactions between patients or parents and health care professionals that occur in pediatric clinical settings, extending from implicit bias and microaggressions to stereotyping, prejudice, and macroaggressions.56 Nonwhite youth and families report overtly weakened patient–health care professional communication and decreased participatory decision making.32,33 Implicit bias, the unconscious attitudes that unintentionally influence behavior, may affect health care professionals’ medical management decisions56 and perceptions about black, Hispanic, and young people of color in terms of disease experience57 and patient compliance.58 Higher levels of perceived bias or discrimination have been linked to worse diabetes care.59,60 The direct effect of implicit bias on HbA1c has not been well studied in pediatric diabetes. Finally, while social stigma associated with T1D is known,61,62,63 it may be more pronounced in specific communities where health literacy and resources are lacking or where T1D is significantly less common than type 2 diabetes. Nonwhite youth may struggle with misunderstanding and stigma that act as chronic stressors that indirectly affect glycemic control via psychosocial or behavioral effects,64,65 resulting in impaired self-care strategies or maladaptive coping behaviors that damage health.66

significantly less common than type 2 diabetes. Nonwhite youth may struggle with misunderstanding and stigma that act as chronic stressors that indirectly affect glycemic control via psychosocial or behavioral effects,64,65 resulting in impaired self-care strategies or maladaptive coping behaviors that damage health.66 Limitations A limitation of the study is that the observed inequity after adjustment for other factors may reflect racial and ethnic differences in the validity of HbA1c as a measure of average glycemia owing to racial differences in the glycation of hemoglobin or other factors affecting red blood cell turnover.67,68,69 However, the between-race differences that have been reported are small (0.4 percentage point in HbA1c69) relative to the differences in the present study, where the mean (SD) HbA1c of group 3 was 12.2% (1.5%) of total hemoglobin at the last visit, roughly 2.2% higher than group 2 and 4.4% higher than group 1 at that time. Combining individuals of many races, ethnicities, and cultures into single categories for analysis may result in residual confounding and underemphasize within-group heterogeneity. We are careful to avoid implying that all nonwhite youth have poor control; in our data, nearly a quarter of nonwhite youth had an HbA1c at or below 7.4% of total hemoglobin at the cohort visit (data not shown). Several of the variables measured at baseline may change over time, including health insurance status. Adjustment variables may provide information for future work that will delve into what drives the inequities. For example, measures of socioeconomic position may be improved by including other measures such as the ability to pay for medication, heath literacy, housing security, or food security. We did not control for diet and physical activity in these analyses. A larger sample may identify additional trajectories that capture the experience of smaller subpopulations, such as individuals who initially have low HbA1c that deteriorates later in the course of T1D. The outcome of trajectory group necessitated the use of logistic regression modeling, which may overestimate effect estimates, particularly when the outcome is common.70,71 Finally, there were relatively small numbers of participants across groups in the analyses stratified by sex and age at diagnosis. Larger studies are needed to further explore interactions and identify nonwhite youth who are at the highest risk for poor glycemic control over time.

larly when the outcome is common.70,71 Finally, there were relatively small numbers of participants across groups in the analyses stratified by sex and age at diagnosis. Larger studies are needed to further explore interactions and identify nonwhite youth who are at the highest risk for poor glycemic control over time. Finally, associations of data-driven trajectory models should be confirmed with future analyses that quantify and compare differences in longitudinal HbA1c across racial/ethnic groups. However, the study has several strengths, including the large, well-characterized, multiethnic cohort;72 the extended follow-up period; and the use of an analytic approach to characterize multiple common HbA1c trajectories and understand associated individual characteristics from an extensive collection of covariates. Conclusions Compared with non-Hispanic white youth with T1D, non-Hispanic black youth, Hispanic youth, and youth with other racial/ethnic backgrounds who are male and diagnosed earlier in life are more likely to show rapid deterioration in glycemic control within 9 years of T1D diagnosis. The findings of this study can be used to inform future research on the identification of factors that contribute to and reinforce racial and ethnic disparity among youth with T1D, particularly nonwhite male patients and nonwhite youth diagnosed earlier in life.

Introduction Guidelines for cancer surveillance routinely recommend aligning care with patients’ underlying cancer risk. Such risk-aligned surveillance entails more frequent surveillance for patients at high vs low risk of recurrence. For example, recommendations for computed tomography surveillance after treatment for kidney cancer range from every 3 to 6 months for stage II or III disease to every year for stage I disease.1 Similarly, after treatment for lung and prostate cancer, more frequent surveillance is recommended for patients at higher risk for recurrence.2,3 After treatment of colorectal adenoma, risk-aligned surveillance4 has already become a Medicare quality measure.5 Risk-aligned surveillance is likely to become even more relevant in the future, as the advent of personalized medicine increases the possibilities to predict each patient’s cancer risk.6

for recurrence.2,3 After treatment of colorectal adenoma, risk-aligned surveillance4 has already become a Medicare quality measure.5 Risk-aligned surveillance is likely to become even more relevant in the future, as the advent of personalized medicine increases the possibilities to predict each patient’s cancer risk.6 Surveillance after treatment of early-stage bladder cancer is a prime example of risk-aligned surveillance. Bladder cancer is the fourth most prevalent noncutaneous cancer in the United States.7 When cancer is identified, patients typically undergo transurethral resection.8 Based on pathologic findings, three-quarters of patients are diagnosed with early-stage disease and then undergo periodic cystoscopic surveillance with close inspection of the bladder mucosa.9,10 Broad consensus holds that the frequency of cystoscopic surveillance should align with each patient’s cancer risk,11 with risk-aligned surveillance recommended by 8 national and international panels.8,12,13,14,15,16,17,18 Specifically, low-risk patients are recommended to receive no more than 3 cystoscopy procedures during the first 2 years after diagnosis; high-risk patients should receive 6 to 8.8,11 Previous studies, including those using Surveillance Epidemiology and End Results Medicare data,19 have demonstrated that bladder cancer surveillance is often not aligned with underlying cancer risk. However, studies were limited to assessing care delivered in a fee-for-service environment and by lack of the longitudinal pathology data needed to accurately assign risk.

e Epidemiology and End Results Medicare data,19 have demonstrated that bladder cancer surveillance is often not aligned with underlying cancer risk. However, studies were limited to assessing care delivered in a fee-for-service environment and by lack of the longitudinal pathology data needed to accurately assign risk. We assessed the extent to which risk-aligned surveillance is practiced across and within Department of Veterans Affairs (VA) facilities by classifying surveillance patterns for low- vs high-risk patients with early-stage bladder cancer. Taking advantage of national data—including longitudinal pathology data—from the largest integrated health care system in the United States,20 we sought to study national patterns of care and simultaneously identify local models of best-practice risk-aligned surveillance.

-risk patients with early-stage bladder cancer. Taking advantage of national data—including longitudinal pathology data—from the largest integrated health care system in the United States,20 we sought to study national patterns of care and simultaneously identify local models of best-practice risk-aligned surveillance. Methods Our overall goal was to compare the frequency of cystoscopic surveillance among patients with low-risk vs high-risk bladder cancer across and within all facilities that perform bladder cancer surveillance in the VA. We proceeded along the following 4 steps: (1) identification of a cohort of patients with newly diagnosed bladder cancer who underwent surveillance in the VA, (2) assessment of low- or high-risk cancer status, (3) measurement of cystoscopic surveillance, and (4) analyses focused on comparing surveillance frequency among low- and high-risk patients across and within facilities. The study was approved by the Veteran’s Institutional Review Board of Northern New England and the University of Utah Institutional Review Board and follows the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline. Patient informed consent was waived because use of the data involved no more than minimum risk to the privacy of patients and the research could not be practically conducted without waiving informed consent.

d follows the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline. Patient informed consent was waived because use of the data involved no more than minimum risk to the privacy of patients and the research could not be practically conducted without waiving informed consent. Cohort Identification We used VA and Medicare administrative data as well as full-text pathology data supplemented by data abstracted by VA tumor registrars, as previously described and validated,21 to identify all patients older than 65 years who were newly diagnosed with bladder cancer between January 1, 2005, and December 31, 2011. The diagnosis date was assigned using a previously validated claims algorithm.21 Next, we excluded those who underwent no cystoscopic surveillance in the VA, those not eligible for cystoscopic surveillance (based on VA or Medicare evidence of cystectomy or radiotherapy within 6 months after diagnosis), and those without pathology data around the diagnosis date (Figure 1). Figure 1. Flow Diagram of Cohort Formation The final cohort comprised 1278 patients with low-risk bladder cancer and 2115 patients with high-risk bladder cancer. VA indicates Department of Veterans Affairs.

Cohort Identification We used VA and Medicare administrative data as well as full-text pathology data supplemented by data abstracted by VA tumor registrars, as previously described and validated,21 to identify all patients older than 65 years who were newly diagnosed with bladder cancer between January 1, 2005, and December 31, 2011. The diagnosis date was assigned using a previously validated claims algorithm.21 Next, we excluded those who underwent no cystoscopic surveillance in the VA, those not eligible for cystoscopic surveillance (based on VA or Medicare evidence of cystectomy or radiotherapy within 6 months after diagnosis), and those without pathology data around the diagnosis date (Figure 1). Figure 1. Flow Diagram of Cohort Formation The final cohort comprised 1278 patients with low-risk bladder cancer and 2115 patients with high-risk bladder cancer. VA indicates Department of Veterans Affairs. Because our focus was on facility-level patterns of care, we excluded patients who were treated in facilities with a substantial proportion (>80%) of missing pathology reports (256 patients from 39 facilities). After excluding patients with incomplete covariate data (n = 16), 6708 patients with complete clinical information remained. Of these, 2417 had ineligible pathology (no urothelial carcinoma or muscle invasion indicating non–early-stage disease) and thus were not eligible for cystoscopic surveillance. Eight hundred sixty patients had uncertain pathology (grade 2 or missing grade or stage, making classification into low-risk or high-risk impossible), leaving 3431 patients with early-stage bladder cancer from 91 facilities (Figure 1).

ng non–early-stage disease) and thus were not eligible for cystoscopic surveillance. Eight hundred sixty patients had uncertain pathology (grade 2 or missing grade or stage, making classification into low-risk or high-risk impossible), leaving 3431 patients with early-stage bladder cancer from 91 facilities (Figure 1). Assessment of Risk Patients with early-stage bladder cancer can be stratified into low- or high-risk categories based on the pathology at the time of diagnosis. Using the European Association of Urology risk-stratification guidelines,22 we defined low-risk cancer status as a primary low-grade noninvasive urothelial carcinoma and high-risk cancer status as a urothelial carcinoma that was either high-grade noninvasive, invasive into the lamina propria (T1), or associated with carcinoma in situ. We operationalized these definitions using data extracted from full-text pathology reports via validated natural language processing algorithms.23 We used pathology reports dated 90 days before to 90 days after the diagnosis date. Because of our focus on comparing cystoscopic surveillance for low- vs high-risk patients across and within facilities, we excluded 38 patients from 6 facilities where either only low- or high-risk patients were treated. Thus, 1278 low-risk and 2115 high-risk patients from 85 facilities remained for analyses (Figure 1).

ecause of our focus on comparing cystoscopic surveillance for low- vs high-risk patients across and within facilities, we excluded 38 patients from 6 facilities where either only low- or high-risk patients were treated. Thus, 1278 low-risk and 2115 high-risk patients from 85 facilities remained for analyses (Figure 1). Measuring Cystoscopic Surveillance We used Common Procedural Terminology and International Classification of Diseases, Ninth Revision, procedure codes to identify cystoscopy procedures during the follow-up period.21,24 The follow-up period started with the bladder cancer diagnosis date and ended with cancer recurrence, death, date of cystectomy or radiotherapy, date of last VA contact, or end of study (December 31, 2014), whichever occurred first. We followed patients only until they had a cancer recurrence, because a recurrence increases risk for further recurrences and thus changes a patient’s cancer risk status.22 We ascertained cancer recurrences using data extracted from full-text pathology reports via the validated natural language processing algorithms.23 Next, we enumerated the number of cystoscopy procedures each patient underwent during the follow-up period. We only counted those cystoscopy procedures that occurred at least 30 days following a previous procedure,25 because procedure codes occurring in close proximity to each other are unlikely to indicate routine surveillance (eg, cystoscopy with biopsy following shortly after a surveillance cystoscopy) and may sometimes even refer to the same procedure. The majority of patients (86.6% [2937 of 3393]) received cystoscopy procedures at only 1 facility; the remaining patients were assigned to the facility where they received the plurality of their cystoscopy procedures.

following shortly after a surveillance cystoscopy) and may sometimes even refer to the same procedure. The majority of patients (86.6% [2937 of 3393]) received cystoscopy procedures at only 1 facility; the remaining patients were assigned to the facility where they received the plurality of their cystoscopy procedures. Statistical Analysis Data management was performed from December 2015 to February 2017. Data analyses were performed from March 2017 to April 2018. Descriptive statistics were used to describe both patient and facility characteristics. Patient characteristics included age at the time of bladder cancer diagnosis, sex, race, year of diagnosis, and comorbidity using the enhanced Elixhauser index.26 Facility characteristics were obtained from the Veterans Health Administration’s Support Service Center and included number of hospital beds, unique patients per year, urology outpatient visits per year, urologist full-time equivalents, rural location, and academic affiliation.27

using the enhanced Elixhauser index.26 Facility characteristics were obtained from the Veterans Health Administration’s Support Service Center and included number of hospital beds, unique patients per year, urology outpatient visits per year, urologist full-time equivalents, rural location, and academic affiliation.27 Negative binomial multilevel models were used to calculate the adjusted cystoscopy frequency for each facility. The outcome was the patient-level number of cystoscopy procedures during the follow-up period. Models contained a random intercept for facility. They were adjusted for age (≥80 years) and comorbidity (>3 comorbidities) to account for differences across facilities in age and comorbidity, which might independently inform surveillance frequency. To account for time trends, we adjusted for year of diagnosis. We also included an indicator variable to denote whether patients were followed for longer than 2 years, because guideline recommendations specify that frequency of follow-up can be decreased at that point.18,28 The logarithm of length of follow-up was included as an offset.

e trends, we adjusted for year of diagnosis. We also included an indicator variable to denote whether patients were followed for longer than 2 years, because guideline recommendations specify that frequency of follow-up can be decreased at that point.18,28 The logarithm of length of follow-up was included as an offset. Separate models were fit for low- and high-risk patients. From these models, we then calculated the frequency of cystoscopic surveillance for each facility with corresponding 95% confidence intervals for an average low- or high-risk patient diagnosed in 2011 using empirical Bayes estimation.29 This approach accounted for differences in the reliability of estimated facility-level cystoscopy frequencies due to differences in sample size by shrinking the estimates of facilities with a small number of patients closer to the overall mean.29 To assess the strength of correlation between cystoscopy frequencies for low- and high-risk patients across facilities, we calculated the Pearson correlation coefficient. We determined which facilities, on average, performed at least 1 cystoscopy more over 2 years for high-risk than for low-risk patients. We also assessed whether surveillance frequency was statistically significantly different within each facility for high- vs low-risk patients. For this, a separate negative binomial model for each facility was fit, with the main exposure being high- vs low-risk cancer status while adjusting for the same covariates.

ts. We also assessed whether surveillance frequency was statistically significantly different within each facility for high- vs low-risk patients. For this, a separate negative binomial model for each facility was fit, with the main exposure being high- vs low-risk cancer status while adjusting for the same covariates. Sensitivity Analysis We performed 2 sensitivity analyses. First, we performed a simpler analysis to address the potential concern that the multivariable multilevel modeling, adjustment, and shrinkage may have affected our findings.29 In this analysis, we calculated the number of cystoscopies performed per 2 years for each patient. We only included patients who had at least 1 year of follow-up and followed them for up to 2 years after diagnosis. This was done to limit the effect of very short or very long follow-up on the calculated number of cystoscopies per 2 years. Next, we limited this sensitivity analysis to facilities that had at least 10 low-risk and 10 high-risk patients in the data set (618 low-risk and 709 high-risk patients across 33 facilities). For each facility, we calculated the unadjusted mean frequency of cystoscopy for low- and high-risk patients. Strength of correlation was assessed between these frequencies using the Pearson correlation coefficient.

nd 10 high-risk patients in the data set (618 low-risk and 709 high-risk patients across 33 facilities). For each facility, we calculated the unadjusted mean frequency of cystoscopy for low- and high-risk patients. Strength of correlation was assessed between these frequencies using the Pearson correlation coefficient. Second, to determine whether reliance on pathology data extracted via natural language processing affected our results, the main analyses were repeated using data abstracted by tumor registrars to characterize cancer risk rather than the natural language processing results. This sensitivity analysis included 1290 low-risk and 3987 high-risk patients. All analyses were performed using SAS statistical software version 9.4 (SAS Institute Inc) and Stata MP statistical software version 15 (StataCorp LLC). Statistical significance was set at 2-sided α < .05 for all analyses. Results We included 1278 low-risk and 2115 high-risk patients with a median (interquartile range [IQR]) age of 77 (71-82) years; 99% (3368 of 3393) were male. Patients received care across 85 VA facilities from 45 states, the District of Columbia, and Puerto Rico. Each facility contributed a mean (range) of 40 (3-150) patients. 1237 of 3393 patients (36.5%) were aged 80 years or older, and 670 of 3393 (19.8%) had more than 3 comorbidities (Table 1). Facilities had a median (IQR) of 135 (81-218) hospital beds, few were in rural locations, and most had an academic affiliation (Table 2). The median (IQR) proportion of low-risk patients among the facilities was 37% (27%-50%).

were aged 80 years or older, and 670 of 3393 (19.8%) had more than 3 comorbidities (Table 1). Facilities had a median (IQR) of 135 (81-218) hospital beds, few were in rural locations, and most had an academic affiliation (Table 2). The median (IQR) proportion of low-risk patients among the facilities was 37% (27%-50%). Table 1. Demographic Characteristics of Patientsa Characteristic No. (%) Low Risk (n = 1278) High Risk (n = 2115) Age, median (IQR), y 76 (71-81) 77 (72-83) ≥80 y 412 (32.2) 825 (39.0) Male 1265 (99.4) 2103 (99.0) Race White 1079 (84.4) 1745 (82.5) Black 88 (6.9) 157 (7.4) Asian 9 (0.7) 19 (0.9) Hispanic 18 (1.4) 39 (1.8) Native American 11 (0.9) 9 (0.4) Unknown 73 (5.7) 146 (6.9) Comorbidities, No. 0-3 1018 (79.7) 1705 (80.6) >3 260 (20.3) 410 (19.4) Year of diagnosis 2005 119 (9.3) 198 (9.4) 2006 131 (10.3) 243 (11.5) 2007 160 (12.5) 282 (13.3) 2008 202 (15.8) 331 (15.7) 2009 217 (17.0) 347 (16.4) 2010 228 (17.8) 390 (18.4) 2011 221 (17.3) 324 (15.3) Bladder cancer stageb Ta (noninvasive) 1278 (100.0) 754 (35.7) T1 (invasive or suspected invasion and superficial invasion depth) 0 1267 (59.9) Carcinoma in situ only 0 94 (4.4) Carcinoma in situb 0 417 (19.7) Bladder cancer gradeb Low 1278 (100.0) 235 (11.1) High 0 1880 (88.9) Abbreviation: IQR, interquartile range. a Based on full-text pathology risk assignment. b Derived via validated natural language processing algorithms.23

Characteristic No. (%) Low Risk (n = 1278) High Risk (n = 2115) Age, median (IQR), y 76 (71-81) 77 (72-83) ≥80 y 412 (32.2) 825 (39.0) Male 1265 (99.4) 2103 (99.0) Race White 1079 (84.4) 1745 (82.5) Black 88 (6.9) 157 (7.4) Asian 9 (0.7) 19 (0.9) Hispanic 18 (1.4) 39 (1.8) Native American 11 (0.9) 9 (0.4) Unknown 73 (5.7) 146 (6.9) Comorbidities, No. 0-3 1018 (79.7) 1705 (80.6) >3 260 (20.3) 410 (19.4) Year of diagnosis 2005 119 (9.3) 198 (9.4) 2006 131 (10.3) 243 (11.5) 2007 160 (12.5) 282 (13.3) 2008 202 (15.8) 331 (15.7) 2009 217 (17.0) 347 (16.4) 2010 228 (17.8) 390 (18.4) 2011 221 (17.3) 324 (15.3) Bladder cancer stageb Ta (noninvasive) 1278 (100.0) 754 (35.7) T1 (invasive or suspected invasion and superficial invasion depth) 0 1267 (59.9) Carcinoma in situ only 0 94 (4.4) Carcinoma in situb 0 417 (19.7) Bladder cancer gradeb Low 1278 (100.0) 235 (11.1) High 0 1880 (88.9) Abbreviation: IQR, interquartile range. a Based on full-text pathology risk assignment. b Derived via validated natural language processing algorithms.23 Table 2. Characteristics of the 85 Facilities Included Characteristica Value Hospital beds, median (IQR), No. 135 (81-218) Unique patients per year, median (IQR), No. 44 220 (32 915-57 776) Urology outpatient visits per year, median (IQR), No. 2722 (1881-3717) Urologist full-time equivalents, median (IQR), No. 1.7 (1.1-2.6) Rural facility, No. (%) 8 (9.4) Academic affiliation, No. (%) 72 (92.9) Abbreviation: IQR, interquartile range. a Characteristics were obtained from the Veterans Health Administration's Support Service Center.

Table 2. Characteristics of the 85 Facilities Included Characteristica Value Hospital beds, median (IQR), No. 135 (81-218) Unique patients per year, median (IQR), No. 44 220 (32 915-57 776) Urology outpatient visits per year, median (IQR), No. 2722 (1881-3717) Urologist full-time equivalents, median (IQR), No. 1.7 (1.1-2.6) Rural facility, No. (%) 8 (9.4) Academic affiliation, No. (%) 72 (92.9) Abbreviation: IQR, interquartile range. a Characteristics were obtained from the Veterans Health Administration's Support Service Center. Patients with low-risk cancer underwent a mean (SD) of 5.3 (3.4) cystoscopy procedures during a median (IQR) follow-up of 2.6 (0.9-4.7) years. As expected, patients with high-risk cancer had shorter follow-up periods due to a higher recurrence rate (median [IQR], 1.2 [0.6-3.7] years) during which they underwent a mean (SD) of 4.7 (3.7) cystoscopy procedures. Across all facilities, the adjusted frequency of surveillance cystoscopy ranged from 3.7 to 6.2 (mean, 4.8) procedures over 2 years for low-risk patients (Figure 2A) and from 4.6 to 6.0 (mean, 5.4) procedures over 2 years for high-risk patients (Figure 2B). For both low-risk and high-risk patients, overlap of 95% confidence intervals around adjusted cystoscopy frequencies suggested that a common mean frequency could be representative of most facilities (Figure 2).

ients (Figure 2A) and from 4.6 to 6.0 (mean, 5.4) procedures over 2 years for high-risk patients (Figure 2B). For both low-risk and high-risk patients, overlap of 95% confidence intervals around adjusted cystoscopy frequencies suggested that a common mean frequency could be representative of most facilities (Figure 2). Figure 2. Facility-Level Variation in Adjusted Frequency of Cystoscopy Procedures for Low-Risk and High-Risk Patients Facilities are ranked from lowest frequency to highest frequency of cystoscopy for patients with low-risk (A) and high-risk (B) bladder cancer. The mean frequency across all facilities is indicated on the y-axis. Error bars indicate 95% confidence intervals. Frequency of cystoscopy was adjusted for age, comorbidity, year of diagnosis, and length of follow-up.

cy to highest frequency of cystoscopy for patients with low-risk (A) and high-risk (B) bladder cancer. The mean frequency across all facilities is indicated on the y-axis. Error bars indicate 95% confidence intervals. Frequency of cystoscopy was adjusted for age, comorbidity, year of diagnosis, and length of follow-up. Within facilities, differences in cystoscopy frequency for high- vs low-risk patients were small (mean [range], 0.6 more over 2 years [0.15 fewer to 1.3 more]) (Figure 3). At most facilities (70 of 85), cystoscopy was performed at a similar frequency for high- and low-risk patients, with a difference of less than 1 cystoscopy over 2 years. At the remaining 15 facilities, there appeared to be more of a distinction between high- and low-risk patients, with a difference surpassing 1 cystoscopy over 2 years (range, 1.0-1.3 cystoscopies more over 2 years for high- vs low-risk patients). However, we found a statistically significantly higher cystoscopy frequency for high- vs low-risk patients only within 4 of the 85 facilities. Across all of the 85 facilities, these findings were reflected in a moderately strong correlation of cystoscopy frequencies for high-risk and low-risk patients (r = 0.52; P < .001) (Figure 3).

nd a statistically significantly higher cystoscopy frequency for high- vs low-risk patients only within 4 of the 85 facilities. Across all of the 85 facilities, these findings were reflected in a moderately strong correlation of cystoscopy frequencies for high-risk and low-risk patients (r = 0.52; P < .001) (Figure 3). Figure 3. Facility-Level Correlation of Cystoscopy Frequency Between Low-Risk and High-Risk Patients Each dot represents 1 facility. The line represents the same cystoscopy frequency for low-risk patients (x-axis) and high-risk patients (y-axis). The shaded area represents facilities where low-risk and high-risk patients undergo cystoscopy at comparable rates (ie, absolute difference of less than 1 cystoscopy over 2 years). Orange dots represent facilities with a statistically significantly higher frequency for high-risk vs low-risk patients. We performed sensitivity analyses using unadjusted data to characterize cystoscopy frequency for high- and low-risk patients at each facility. Again, cystoscopy frequency was moderately strongly correlated (r = 0.45; P = .008) and cystoscopy was performed at a similar frequency for high- and low-risk patients at most facilities (28 of the 33 included) (eFigure in Supplement 1). In a second sensitivity analysis, we used data abstracted by tumor registrars instead of full-text pathology data to characterize risk. Results from these analyses were not substantially different from those of the main analyses; thus, only the latter are presented.

es (28 of the 33 included) (eFigure in Supplement 1). In a second sensitivity analysis, we used data abstracted by tumor registrars instead of full-text pathology data to characterize risk. Results from these analyses were not substantially different from those of the main analyses; thus, only the latter are presented. Discussion We found that across the national VA integrated health care network, patients with high-risk bladder cancer undergo cystoscopic surveillance at comparable frequency to those with low-risk cancer, with few exceptions. At only 4 of 85 facilities was cystoscopic surveillance frequency significantly higher for high-risk patients than for low-risk patients, and even at these facilities high-risk patients underwent on average only approximately 1 cystoscopy more over the course of 2 years. This difference is much smaller than would be expected based on guideline recommendations, which recommend 6 to 8 cystoscopies over 2 years for high-risk patients and no more than 3 cystoscopies over 2 years for low-risk patients.11 Thus, we were not able to identify facilities that clearly exhibit local models of best practice risk-aligned care, although it may be possible to learn from facilities that appear to distinguish between low- and high-risk patients in their surveillance practices.

an 3 cystoscopies over 2 years for low-risk patients.11 Thus, we were not able to identify facilities that clearly exhibit local models of best practice risk-aligned care, although it may be possible to learn from facilities that appear to distinguish between low- and high-risk patients in their surveillance practices. This is the first study, to our knowledge, to examine whether risk-aligned cancer surveillance is performed in a national health system. As 1 of many cancers for which ongoing surveillance is routinely recommended, early-stage bladder cancer serves as a useful paradigm for assessing surveillance practices because of its high prevalence7 and because surveillance with cystoscopy is identifiable using administrative data. Our findings demonstrate that facilities with appropriately high surveillance rates for high-risk cancers also have inappropriately high surveillance rates for low-risk cancers and vice versa. They highlight how challenging it can be to routinely incorporate underlying cancer risk into cancer surveillance practice and suggest non–cancer-related factors may be driving surveillance rates, regardless of underlying risk of recurrence and progression.

y high surveillance rates for low-risk cancers and vice versa. They highlight how challenging it can be to routinely incorporate underlying cancer risk into cancer surveillance practice and suggest non–cancer-related factors may be driving surveillance rates, regardless of underlying risk of recurrence and progression. Prior work has examined the use of surveillance colonoscopy among patients diagnosed with colorectal adenoma. Similar to our findings, the frequency of surveillance colonoscopy was not aligned with the risk for progression to advanced lesions.30 Another example of patients with cancer not receiving risk-aligned care is imaging among low- and high-risk patients recently diagnosed with prostate or breast cancer.31,32 Diagnostic imaging is appropriate among high-risk patients and inappropriate among low-risk patients. Similar to our findings regarding risk-aligned cancer surveillance, appropriate and inappropriate imaging rates were closely correlated.31,32 Furthermore, efforts to decrease inappropriate imaging for patients with low-risk prostate cancer have produced the unintended consequence of decreasing appropriate imaging for high-risk patients.33 Taken together, these findings highlight the need to develop strategies that make it easier for physicians to deliver cancer care that is aligned with underlying cancer risk.

g for patients with low-risk prostate cancer have produced the unintended consequence of decreasing appropriate imaging for high-risk patients.33 Taken together, these findings highlight the need to develop strategies that make it easier for physicians to deliver cancer care that is aligned with underlying cancer risk. Limitations Although the present study is informed by national data and our findings reflect care received across a large number of distinct facilities, it is not without limitations. First, our data are from the VA and thus generalizability to other settings may be limited. However, previous studies of surveillance of early-stage bladder cancer using Surveillance Epidemiology and End Results Medicare data have also found that risk of cancer recurrence had little impact on the care patients receive.19 Second, the number of patients within each facility affected the power to detect any statistically significant differences in surveillance frequency between low-risk and high-risk patients. Thus, we calculated the shrunken surveillance frequency for each facility in our main analyses—an approach that adjusted for the reduced reliability of estimates in smaller facilities.29 Third, we acknowledge certain limitations of ascertaining cancer risk. While we used a validated natural language processing engine to extract information from pathology reports,23 certain characteristics that may contribute to an increased risk of cancer recurrence among low-risk patients (such as multifocality and large tumor size, classified as intermediate risk in the European Association of Urology risk-stratification guidelines22) could not be assessed. Thus, among low-risk patients, there are likely subgroups of patients who are at lower and somewhat higher risk of cancer recurrence. However, as all low-risk patients had low-grade disease, they are all still at substantially lower risk than the high-risk patients. Finally, the analysis examining differences in surveillance frequency between low- and high-risk patients is predicated on recommendations for risk-aligned surveillance that are based on lower level evidence.11 Nevertheless, 8 panels reviewing current evidence have recommended risk-aligned surveillance,8,12,13,14,15,16,17,18 which allows for timely detection of progression to muscle-invasive cancer among high-risk patients while sparing low-risk patients unnecessary procedures.

llance that are based on lower level evidence.11 Nevertheless, 8 panels reviewing current evidence have recommended risk-aligned surveillance,8,12,13,14,15,16,17,18 which allows for timely detection of progression to muscle-invasive cancer among high-risk patients while sparing low-risk patients unnecessary procedures. Timely detection is important because delays in diagnosis of muscle-invasive cancer are associated with increased mortality.34 Avoiding unnecessary procedures is relevant for patients as they lead to more anxiety, discomfort, and costs.35,36 Strengths These limitations notwithstanding, our study has important strengths. Because we had access to full-text pathology reports, this is the first population-based study, to our knowledge, on early-stage bladder cancer surveillance in which cancer recurrence could be ascertained. Because recurrence elevates patients’ future risk of further recurrence and thus influences further surveillance, we censored follow-up at the time of cancer recurrence, which has not been possible using other population-based data sets. Second, we used multilevel modeling to account for differences in reliability of facility-level estimates due to differences in the number of patients per facility. Third, to address potential concerns that the use of complex modeling or data obtained via natural language processing affected the validity of our findings, we performed sensitivity analyses using unadjusted data and data abstracted by tumor registrars, which supported our main findings.

mber of patients per facility. Third, to address potential concerns that the use of complex modeling or data obtained via natural language processing affected the validity of our findings, we performed sensitivity analyses using unadjusted data and data abstracted by tumor registrars, which supported our main findings. Our study is the first to assess risk-aligned bladder cancer surveillance in the VA, which has important implications for health policy and future research. Some may argue that financial incentives in a fee-for-service environment are an important factor contributing to more cystoscopy procedures than recommended. However, the VA is a capitated health care system in which such financial incentives do not exist, and we still find substantial overuse of cystoscopy among low-risk patients. In fact, we recently performed patient-level analyses to examine the extent of overuse and found that the number of cystoscopy procedures among low-risk patients is about double what it should be.37 Thus, other factors are likely important contributors to the lack of risk-aligned surveillance and may include limited clinician knowledge of the guidelines, clinicians’ habit to continue what they have always done, and complexity and variability of risk stratification across different guidelines.11 Future work should focus on understanding the barriers to risk-aligned surveillance and on developing strategies to improve surveillance. These strategies should facilitate providers’ ability to deliver risk-aligned surveillance, thus reducing adverse consequences of overuse of surveillance among low-risk patients and of underuse of surveillance among high-risk patients.

s to risk-aligned surveillance and on developing strategies to improve surveillance. These strategies should facilitate providers’ ability to deliver risk-aligned surveillance, thus reducing adverse consequences of overuse of surveillance among low-risk patients and of underuse of surveillance among high-risk patients. Conclusions In conclusion, our study highlights that risk-aligned surveillance for early-stage bladder cancer is not widely practiced. On the contrary, patients with high- and low-risk cancer undergo surveillance at comparable frequency, despite recommendations that high-risk patients warrant surveillance at least twice as often.11 Our findings should alert those who care for patients with bladder cancer and those who care for patients with other neoplasms for which risk-aligned surveillance is recommended. While risk factors, natural history, and tumor-specific characteristics differ across neoplasms, the challenges clinicians face to align surveillance with underlying cancer risk are likely similar. Supplement 1. eFigure. Sensitivity Analysis Examining Facility-Level Correlation of Cystoscopy Frequency Between Low-Risk (x-axis) and High-Risk (y-axis) Patients Click here for additional data file. Supplement 2. Data Sharing Statement Click here for additional data file.

a overexpression 19 (51.4) 34 (32.4) .04 E6 and E7 mRNA positivity 18 (48.7) 6 (5.7) <.001 Low p53 expression 23 (62.2) 40 (38.1) .01 Abbreviations: EAC, esophageal adenocarcinoma; EMR, endoscopic mucosal resection; HGD, high-grade dysplasia; HPV, human papillomavirus; mRNA, messenger RNA; RFA, radiofrequency ablation. a Denominators are listed when sample does not equal full sample size. b Differences between HPV-positive vs HPV-negative cases in regard to baseline characteristics were assessed using 2-sample t test for all numerical data and χ2 analysis for binary measurements. c Calculated as weight in kilograms divided by height in meters squared. d A pack-year indicates smoking 1 pack of cigarettes per day for a year. e Excess alcohol intake was defined as more than 21 units/wk for men and more than 14 units/wk for women. f TNM classification as per the American Joint Committee on Cancer’s Cancer Staging Manual, 7th Edition.

of cases interpreted by each pathologist, and np is the number of pathologists. Logistic regression models were used to test for a difference in accuracy between AJCC 7– and AJCC 8–based mappings. Models used robust estimators of the variance to account for correlation of case interpretations from the same pathologist. The reproducibility of participating pathologists’ interpretations were assessed as both intraobserver and interobserver concordance. Interobserver concordance considered all pairs of interpretations of the same invasive disease case by 2 different pathologists, and the proportion of those pairs for which interpretations were in the same diagnostic class was calculated. Although cases were restricted to those with invasive melanoma by consensus reference diagnosis, participating pathologist interpretations could include diagnoses in other noninvasive MPATH-Dx classes. Confidence intervals for interobserver concordance rates were bootstrap percentile intervals, and tests for differences between AJCC 7– and AJCC 8–based mappings used a Wald statistic based on the bootstrap standard error of the difference. A total of 3000 bootstrap samples were obtained by participant-level sampling with replacement and generation of all possible pairs of distinct sample participants for each sample.

Introduction Approximately 13% to 20% of adolescents experience minor depressive episodes (mDE) or major depressive episodes (MDE) annually.1 These adolescents have a higher incidence of medical illness2 than those without mDE and MDE, and are at higher risk for suicide and recurrent depression.3,4,5 Effective depression prevention programs are essential.6 Promising findings for depression prevention programs are available. A Cochrane meta-analysis of prevention trials favored the intervention group over the control group with an overall risk difference for depressive disorders of −0.03, and for depression symptoms a standard mean difference of −0.21.7 A review noted a 22% risk reduction of depressive episodes for adolescents.7,8 Another meta-analysis involving 19 randomized preventive trials demonstrated significant reduction in depressive symptoms over 2 years among adolescents with higher symptom levels.9 Another review of traditional therapies augmented with computerized communications demonstrated small-to-moderate effect sizes for depressive symptoms.10 A systematic review of primary care-based preventative interventions targeting depression identified 14 randomized clinical trials; only 1 included adolescents, and average effect sizes were small.11 Targeted interventions that show success during trials may not be scalable owing to practical issues such as cost, or prove ineffective in the broader community.12

reventative interventions targeting depression identified 14 randomized clinical trials; only 1 included adolescents, and average effect sizes were small.11 Targeted interventions that show success during trials may not be scalable owing to practical issues such as cost, or prove ineffective in the broader community.12 The primary care internet-based intervention, competent adulthood transition with cognitive behavioral humanistic and interpersonal training (CATCH-IT) addresses the need for a scalable intervention.13,14,15,16 Internet-based interventions are accessible, cost-effective, private, and acceptable because they reduce stigma.12 The CATCH-IT intervention is simple, consumer friendly, and more easily scaled up than more intensive, face-to-face interventions. A randomized clinical trial in China found that CATCH-IT lowered depressive symptoms in adolescents over 12 months.17

, cost-effective, private, and acceptable because they reduce stigma.12 The CATCH-IT intervention is simple, consumer friendly, and more easily scaled up than more intensive, face-to-face interventions. A randomized clinical trial in China found that CATCH-IT lowered depressive symptoms in adolescents over 12 months.17 We present a multisite randomized clinical trial testing the efficacy of CATCH-IT (version 3) vs an internet-based health education (HE) attention control in primary care. We aimed to prevent the onset of depressive episodes and lower symptoms in adolescents at intermediate-to-high risk for depression. Our primary hypothesis was that adolescents assigned to CATCH-IT relative to HE would have a lower hazard ratio (HR) for mDE or MDE at 6 months. We chose to evaluate group differences at 6 months to examine the potential of CATCH-IT as an immediate, medium-term response to depressive symptoms, given that follow-up intervals for such interventions in adolescents generally range from less than 6 to 12 months.12 We also hypothesize that adolescents in CATCH-IT would show improvement in depressed mood and functional status relative to HE.18

f CATCH-IT as an immediate, medium-term response to depressive symptoms, given that follow-up intervals for such interventions in adolescents generally range from less than 6 to 12 months.12 We also hypothesize that adolescents in CATCH-IT would show improvement in depressed mood and functional status relative to HE.18 Methods Study Design and Setting We conducted a hybrid type 1 effectiveness-implementation trial to test the efficacy of CATCH-IT in a scalable setting and collected information regarding implementation.19 This 2-site (Chicago, Illinois, and Boston, Massachusetts) randomized trial compared CATCH-IT vs HE for preventing depressive episode onset in an intermediate- to high-risk sample of adolescents in primary care. We defined risk status as teens’ current elevated symptoms of depression, history of depressive episode, or both. Depressive episodes are defined as a Depression Severity Rating (DSR) of 3 or more (exhibiting symptoms of subthreshold MDE). At baseline, the participants’ average Center for Epidemiologic Studies Depression scale (CES-D20) score was 16.9. Twelve percent of the sample enrolled with a past MDE only, 60% had current elevated symptoms only, and 28% had both a past MDE and elevated symptoms of depression. Participants were assessed at baseline and at 2 and 6 months postenrollment. Dates of depressive episodes were recorded. Depressive episodes were diagnosed through the use of Kiddie Schedule for Affective Disorders scale (K-SADS) interviews.

ymptoms only, and 28% had both a past MDE and elevated symptoms of depression. Participants were assessed at baseline and at 2 and 6 months postenrollment. Dates of depressive episodes were recorded. Depressive episodes were diagnosed through the use of Kiddie Schedule for Affective Disorders scale (K-SADS) interviews. Institutional review board approval was received from the central site, University of Illinois at Chicago, and local institutional review boards (IRB of Record, Wellesley College, Advocate Healthcare, Franciscan Alliance, Northshore University Health Systems, Northwestern, and Access Healthcare).20 Participants were recruited from 2012 to 2016 through a description of the study during doctor visits, recruitment letters, and posted flyers. Adolescents were screened for risk in-person or by phone. After parental consent, adolescents participated in an eligibility assessment by phone. The parent and adolescent attended an enrollment assessment at their primary care office, when written informed consent from parents and assent from adolescents were obtained, and assessments were administered to confirm eligibility.21 This study followed the Consolidated Standards of Reporting Trials (CONSORT) reporting guideline. The protocol, implementation process, and methods have been described in Supplement 1 and elsewhere.18 The study was conducted in clinics in Chicago, northern Indiana, and Boston.

ere administered to confirm eligibility.21 This study followed the Consolidated Standards of Reporting Trials (CONSORT) reporting guideline. The protocol, implementation process, and methods have been described in Supplement 1 and elsewhere.18 The study was conducted in clinics in Chicago, northern Indiana, and Boston. Inclusion and Exclusion Criteria Adolescents aged 13 to 18 years with elevated levels of depressive symptoms on the CES-D22 (scores 8-17 on the CES-D10 or scores ≥16 on the CES-D20) at screening or at baseline, and/or a history of depression or dysthymia,22,23,24 were eligible. Exclusion criteria included the following: current MDE diagnosis or treatment; past cognitive behavioral therapy; CES-D10 scores of more than 1722; schizophrenia, psychosis, or bipolar disorder; serious medical condition (ie, causing serious disability or dysfunction); significant reading impairment or developmental disability; imminent suicidal risk; and current drug or alcohol abuse.25,26 Criteria were selected to avoid confounding factors in depression etiology or treatment, consistent with the use of CATCH-IT as a preventive intervention. Randomization Participants were assigned randomly to CATCH-IT or HE (1:1 allocation) using a computer generated sequence blocked by site and time of entry (random blocks of size 4 and 6), stratified by risk severity (based on CES-D score, prior MDE, or dysthymia), sex and age (13-14 years or 15-18 years).26,27

Inclusion and Exclusion Criteria Adolescents aged 13 to 18 years with elevated levels of depressive symptoms on the CES-D22 (scores 8-17 on the CES-D10 or scores ≥16 on the CES-D20) at screening or at baseline, and/or a history of depression or dysthymia,22,23,24 were eligible. Exclusion criteria included the following: current MDE diagnosis or treatment; past cognitive behavioral therapy; CES-D10 scores of more than 1722; schizophrenia, psychosis, or bipolar disorder; serious medical condition (ie, causing serious disability or dysfunction); significant reading impairment or developmental disability; imminent suicidal risk; and current drug or alcohol abuse.25,26 Criteria were selected to avoid confounding factors in depression etiology or treatment, consistent with the use of CATCH-IT as a preventive intervention. Randomization Participants were assigned randomly to CATCH-IT or HE (1:1 allocation) using a computer generated sequence blocked by site and time of entry (random blocks of size 4 and 6), stratified by risk severity (based on CES-D score, prior MDE, or dysthymia), sex and age (13-14 years or 15-18 years).26,27 Blinding Randomization was concealed from investigators, clinicians, patients, and families until the baseline consent, enrollment, data collection, and assessment were completed. Study participants could not be blinded to their arm assignment. The health care professional was also not blinded, as he or she was expected to conduct 3 motivational interviews (MIs) for CATCH-IT participants. Assessors remained blinded throughout the study. Principal investigators (B.W.V.V. and T.R.G.G.) were blinded to between-group comparisons and group descriptive data until all 6-month follow-up data were collected.

blinded, as he or she was expected to conduct 3 motivational interviews (MIs) for CATCH-IT participants. Assessors remained blinded throughout the study. Principal investigators (B.W.V.V. and T.R.G.G.) were blinded to between-group comparisons and group descriptive data until all 6-month follow-up data were collected. Retention Challenges related to ongoing study participation were addressed by research staff. Approaches used to maintain contact were birthday cards and regular contact updates. Outcomes Occurrence of first depressive episode was determined by the DSR. A score indicating at least subthreshold major depression (a DSR of ≥3) was considered to be a depressive episode. To test for robustness of findings, we also examined data using a DSR cutoff of 4 or more, indicating probable MDE, and a DSR of 5, indicating the presence of MDE.28 Symptom outcomes include the CES-D1022 and Global Assessment Scale (GAS) scores.

major depression (a DSR of ≥3) was considered to be a depressive episode. To test for robustness of findings, we also examined data using a DSR cutoff of 4 or more, indicating probable MDE, and a DSR of 5, indicating the presence of MDE.28 Symptom outcomes include the CES-D1022 and Global Assessment Scale (GAS) scores. CATCH-IT Intervention The CATCH-IT intervention includes an internet component (15 adolescent modules, based primarily on the Coping with Depression Adolescent Course,29 behavioral activation,30 and interpersonal psychotherapy),31 a brief motivational component (3 physician MIs at 0, 2, and 12 months), and 1 to 3 staff coaching phone calls either at 1 month (Chicago) or at 2 and 4 weeks (Boston), and 18 months. There were also up to 3 check-in calls during weeks 1 through 3 to facilitate website use. The parent internet intervention component (5 modules) is based on an adaptation of the Preventive Intervention Project.32 A description of the intervention has been published.18,33 HE Intervention The HE intervention is an attention control internet site (14 modules) providing instruction on general health topics. The 14th module discusses mood and mental health treatment, and also addresses mental disorder stigma.14,34 Up to 3 check-in calls (weeks 1-3) were offered to ensure website access. The caregiver internet program (4 modules) is similar.

control internet site (14 modules) providing instruction on general health topics. The 14th module discusses mood and mental health treatment, and also addresses mental disorder stigma.14,34 Up to 3 check-in calls (weeks 1-3) were offered to ensure website access. The caregiver internet program (4 modules) is similar. Intervention Shared Elements Both interventions were consistent with guidelines for adolescent depression in primary care including the following: training clinicians in depression identification, diagnosis, and treatment; establishing referral relationships; screening; using a formal tool to determine depression risk; assessing depression; interviewing caregivers and adolescents; educating caregivers and adolescents on treatment; establishing treatment plans; and establishing safety plans.35 These steps are closely related to the Chronic Care Model.36 Rates of episodes were extremely low for this high-risk sample. When episodes were identified, adolescents were referred for treatment, and caregivers and pediatricians were notified.

n treatment; establishing treatment plans; and establishing safety plans.35 These steps are closely related to the Chronic Care Model.36 Rates of episodes were extremely low for this high-risk sample. When episodes were identified, adolescents were referred for treatment, and caregivers and pediatricians were notified. Instruments Instruments have been described previously.18 The 2-question screener was based on the Patient Health Questionnaire for adolescents.37,38 The K-SADS39,40 is a semistructured interview assessing current and lifetime psychiatric diagnoses in participants aged younger than 18 years, administered to parents and adolescents.39,41 The DSRs are obtained from the Kiddie Longitudinal Interval Follow-up Evaluation41 for each week of the follow-up interval, and GAS ratings were assigned at each assessment. For both the K-SADS and the Kiddie Longitudinal Interval Follow-up Evaluation, precipitating events were reviewed, and episodes secondary to medical concerns were indicated, if they occurred. The CES-D10 measures the frequency of 10 depressive symptoms over the past week, using a 4-point scale; it was completed at baseline, 2, and 6 months.22 Demographic information was collected at baseline, including race and ethnicity, using categories defined by the study team. Fidelity and exposure to the intervention were based on module completion, and completion and rating of the MIs, with 2 trained raters using the MI Treatment Integrity coding manual (version 4.2.1),42 and number of characters typed into the CATCH-IT website.

ce and ethnicity, using categories defined by the study team. Fidelity and exposure to the intervention were based on module completion, and completion and rating of the MIs, with 2 trained raters using the MI Treatment Integrity coding manual (version 4.2.1),42 and number of characters typed into the CATCH-IT website. Sample Size We required 200 participants per intervention condition to achieve 80% power based on a conservative application of our pilot study findings.40 These calculations assumed that in the control group 72% are free from depression after 1-year follow-up, and the second year continues to follow the same exponential rate for controls; for intervention, the hazard is a constant ratio of 0.62, and an attrition rate of 7% for each of the first 4 quarters, and 2% for each of the second 4 quarters (Trial Protocol in Supplement 1).

free from depression after 1-year follow-up, and the second year continues to follow the same exponential rate for controls; for intervention, the hazard is a constant ratio of 0.62, and an attrition rate of 7% for each of the first 4 quarters, and 2% for each of the second 4 quarters (Trial Protocol in Supplement 1). Statistical Analysis The trial tested for differences between group medians in website engagement using Wilcoxon rank sum tests. We estimated incidence rates by calculating the number of depressive episodes per 10 000 person-weeks of follow-up. Kaplan-Meier curves were used to estimate the time to first episode distribution for each intervention under 6 different treatment allocations (eTable 1 in Supplement 2). Treatment allocations were developed based on existent literature with regard to threshold effects of adherence as an a priori analytic strategy. The thresholds were applied in the same manner with both interventions with similar numbers of persons identified in each arm. Cox proportional hazard regression was used to estimate the HR comparing CATCH-IT with HE. We present adjusted (sex, ethnicity [Hispanic/nonHispanic], race [white/nonwhite], baseline age, site, and baseline CES-D10 score) and unadjusted HRs. The assumption of proportional hazards was checked by testing the independence between the Schoenfeld residuals and time.43 The trial examined moderating effects of baseline adolescent CES-D10 score as a continuous variable, exhibited across a range of possible CES-D10 values, by including interaction terms in the Cox models. We used linear mixed-effect growth models with random intercepts and slopes to examine differences between group change over time in CES-D10 and GAS. Analyses were adjusted for the covariates listed above. We used propensity scores to account for differences between treatment groups in the per protocol analysis (≥2 modules completed) that could otherwise confound treatment effect estimates controlling for: site, age, sex, ethnicity, race, mother’s education, parents’ marital status, number of siblings, firstborn child, times moved, current GAS score, most severe past GAS score, highest past GAS score, and baseline CES-D10. Analyses were conducted using R statistical software, version 3.3.1 (R Foundation Inc), SAS, version 9.4 (SAS Institute), and Mplus, version 8 (Muthén & Muthén).

marital status, number of siblings, firstborn child, times moved, current GAS score, most severe past GAS score, highest past GAS score, and baseline CES-D10. Analyses were conducted using R statistical software, version 3.3.1 (R Foundation Inc), SAS, version 9.4 (SAS Institute), and Mplus, version 8 (Muthén & Muthén). Missing Data The percentages of participants missing each K-SADS or CES-D10 assessment were calculated. We used a logistic regression model to determine whether those missing from follow-up differed from those who were not. We also used multiple imputation to assess the potential for differential follow-up by intervention condition. We constructed 50 data sets for each site and intervention condition with fully saturated specification by condition interacting with the following variables: all CES-D10 and GAS values (0, 2, and 6 months), screening CES-D10, baseline age, sex, ethnicity, race, and maternal education. These were combined into 50 complete imputed data sets and analyzed separately using the growth models described above; results were pooled. Results Implementation We implemented the study in 8 health systems from 31 practices in a defined population of more than 41 000 adolescents. There were 8499 adolescents screened, 2250 phone assessments, 446 enrolled, and 369 randomized. The 2 groups consisted of CATCH-IT (n = 193) and HE (n = 176) (Figure 1). Among these participants, 28% had both a past episode and subsyndromal depression; 12% had a past episode only, 59% had subsyndromal depression only, and 1% had borderline subsyndromal depression

one assessments, 446 enrolled, and 369 randomized. The 2 groups consisted of CATCH-IT (n = 193) and HE (n = 176) (Figure 1). Among these participants, 28% had both a past episode and subsyndromal depression; 12% had a past episode only, 59% had subsyndromal depression only, and 1% had borderline subsyndromal depression Figure 1. Consort Diagram CATCH-IT indicates competent adulthood transition with cognitive behavioral humanistic and interpersonal training; CES-D10, Center for Epidemiologic Studies Depression scale; GAS, Global Assessment Scale; MI, motivational interview. Sample Participants were aged 13 to 18 years (mean [SD] age, 15.4 [1.5] years; 251 women [68%]) with history of depression and/or current subsyndromal depressive symptoms. Participants were diverse in self-reported race and ethnicity: 21% Hispanic, 26% non-Hispanic black, 43% non-Hispanic white, 4% Asian, and 6% multiracial or other. Sixty-one percent had married parents, and 53% of the fathers were college graduates. Adolescents were moderately depressed (CES-D10: mean [SD], 9.4 [4.6]), moderately impaired (GAS: mean [SD], 78.1 [9.4%]), had a prior DSR of 3 or more (226 [62%]), and had a prior DSR of 4 or more (144 [40%]) (Table 1).

cial or other. Sixty-one percent had married parents, and 53% of the fathers were college graduates. Adolescents were moderately depressed (CES-D10: mean [SD], 9.4 [4.6]), moderately impaired (GAS: mean [SD], 78.1 [9.4%]), had a prior DSR of 3 or more (226 [62%]), and had a prior DSR of 4 or more (144 [40%]) (Table 1). Table 1. Participant Characteristics at Baseline Characteristic All (N = 369) CATCH-IT (n = 193) Health Education (n = 176) No. No. (%) No. No. (%) No. No. (%) Age, mean (SD), y 369 15.4 (1.5) 193 15.4 (1.5) 176 15.5 (1.5) Sex 369 193 176 Male 118 (32) 59 (31) 59 (34) Female 251 (68) 134 (69) 117 (66) Ethnicity 369 193 176 Hispanic 77 (21) 41 (21) 36 (20) Non-Hispanica 292 (79) 152 (79) 140 (80) Race 369 193 176 White 201 (54) 107 (55) 94 (53) Nonwhiteb 168 (46) 86 (45) 82 (47) Mother’s education 359 188 171 Some high school 12 (3) 5 (3) 7 (4) High school graduate/GED 45 (13) 20 (11) 25 (15) Some college 87 (24) 44 (23) 43 (25) College graduate 215 (60) 119 (63) 96 (56) Father’s education 336 177 159 Some high school 26 (8) 12 (7) 14 (9) High school graduate/GED 76 (23) 36 (20) 40 (25) Some college 55 (16) 37 (21) 18 (11) College graduate 179 (53) 92 (52) 87 (55) K-SADS GAS, mean (SD)c Current 367 78.1 (9.4) 193 78.3 (9.3) 174 78.0 (9.6) Most severe past 359 67.5 (10.9) 189 68.1 (10.3) 170 67.0 (11.5) Highest past 360 82.2 (8.5) 190 82.3 (8.4) 170 82.1 (8.5) DSR, mean (SD)d Most severe 364 3.1 (1.4) 189 3.1 (1.4) 175 3.2 (1.4) ≥3 226 (62) 113 (60) 113 (65) ≥4 144 (40) 75 (40) 69 (39) Current 365 1.8 (0.9) 190 1.7 (0.9) 175 1.8 (0.9) CES-D20, mean (SD)e 362 16.9 (8.7) 190 17.3 (8.7) 172 16.5 (8.8) CES-D10, mean (SD)f 362 9.4 (4.6) 190 9.5 (4.5) 172 9.4 (4.6) SCARED total score, mean (SD)g 312 25.3 (12.3) 171 25.5 (12.7) 141 25.2 (11.9) Abbreviations: CATCH-IT, Competent Adulthood Transition with Cognitive Behavioral Humanistic and Interpersonal Training; CES-D, Center for Epidemiologic Studies Depression scale; DSR, Depression Severity Rating; GAS, Global Assessment Scale; GED, general equivalency diploma; K-SADS, Kiddie Schedule for Affective Disorders scale; SCARED, screen for child and anxiety related disorders.

tive Behavioral Humanistic and Interpersonal Training; CES-D, Center for Epidemiologic Studies Depression scale; DSR, Depression Severity Rating; GAS, Global Assessment Scale; GED, general equivalency diploma; K-SADS, Kiddie Schedule for Affective Disorders scale; SCARED, screen for child and anxiety related disorders. a Participants with missing ethnicity data were coded as non-Hispanic (n = 6). b Participants with missing race data were coded as nonwhite (n = 20; most identified as Hispanic). c Possible range: 1 to 100; a higher score indicates higher functioning. d Possible range: 1 to 6; a higher score indicates more severe depression. e Possible range: 0 to 60; a higher score indicates more severe depression. f Possible range: 0 to 30; a higher score indicates more severe depression. g Possible range: 0 to 82; a higher score indicates greater anxiety.

c Possible range: 1 to 100; a higher score indicates higher functioning. d Possible range: 1 to 6; a higher score indicates more severe depression. e Possible range: 0 to 60; a higher score indicates more severe depression. f Possible range: 0 to 30; a higher score indicates more severe depression. g Possible range: 0 to 82; a higher score indicates greater anxiety. Fidelity and Intervention Exposure Intervention use was monitored and recorded. The number of MIs completed was recorded for CATCH-IT participants. Table 2 shows that CATCH-IT adolescents and parents spent more time using the intervention, but CATCH-IT adolescents completed fewer modules than HE adolescents (modules completed: median [interquartile range], 1.0 [4.0] vs 4.0 [14.0], respectively; P = .003). Both study arms received a sizable dose of the interventions, with the combined (parent + adolescent) module completion greater for HE (modules completed: median [interquartile range], 4.0 [8.0] vs 8.0 [17.0], respectively; P < .001). Adolescents and parents included in CATCH-IT typed a mean (SD) of 3071 (4572) and 716 (977) characters, respectively (Table 2). Over 73% of MIs and phone calls were completed. Mean (SD) interview length was 7.7 (4.0) minutes, mean (SD) technical global rating was 3.0 (0.5) on a 1 to 5 scale, and mean (SD) relational global rating was 2.9 (0.6) (eTable 2 in Supplement 2).

mean (SD) of 3071 (4572) and 716 (977) characters, respectively (Table 2). Over 73% of MIs and phone calls were completed. Mean (SD) interview length was 7.7 (4.0) minutes, mean (SD) technical global rating was 3.0 (0.5) on a 1 to 5 scale, and mean (SD) relational global rating was 2.9 (0.6) (eTable 2 in Supplement 2). Table 2. Fidelity Assessment of Internet Component Website Use Mean (SD) Difference CATCH-IT and Health Education, Mean (95% CI) Median (IQR) P Valuea CATCH-IT Health Education CATCH-IT Health Education Adolescents No. 193 176 Modules completed, No. 3.4 (4.7) 6.8 (6.5) −3.4 (−4.5 to −2.2) 1.0 (4.0) 4.0 (14.0) .003 Total time on site, min 100.2 (143.1) 22.8 (31.0) 77.4 (55.7 to 99.0) 39.6 (149.2) 8.4 (35.1) <.001 Days visited site 3.7 (4.5) 1.4 (1.6) 2.3 (1.6 to 3.0) 2.0 (4.0) 1.0 (2.0) <.001 Total characters typed, No. 3071 (4572) NA NA 923 (4469) NA NA Adolescents and parents Modules completed, No. 5.3 (5.8) 8.8 (7.3) −3.5 (−4.8 to −2.1) 4.0 (8.0) 8.0 (17.0) <.001 Total time on site, min 130.6 (157.9) 30.6 (35.6) 100.0 (76.1 to 124.0) 75.8 (192.2) 18.9 (40.8) <.001 Days visited site 5.2 (5.2) 2.2 (2.2) 2.9 (2.1 to 3.8) 4.0 (6.0) 2.0 (2.0) <.001 Total characters typed, No. 3713 (4932) NA NA 1899 (5792) NA NA Parents No. 165 157 Modules completed, No. 2.1 (2.0) 2.2 (1.9) −0.1 (−0.6 to 0.3) 2.0 (4.0) 4.0 (4.0) .80 Total time on site, min 32.6 (37.3) 8.6 (10.0) 24.0 (17.9 to 30.0) 22.4 (51.9) 5.6 (14.9) <.001 Days visited site 1.6 (1.6) 0.9 (1.1) 0.6 (0.3 to 0.9) 1.0 (2.0) 1.0 (1.0) <.001 Total characters typed, No. 716 (977) NA NA 101 (1205) NA NA Abbrevations: CATCH-IT, Competent Adulthood Transition with Cognitive Behavioral Humanistic and Interpersonal Training; IQR, interquartile range; NA, not available.

(14.9) <.001 Days visited site 1.6 (1.6) 0.9 (1.1) 0.6 (0.3 to 0.9) 1.0 (2.0) 1.0 (1.0) <.001 Total characters typed, No. 716 (977) NA NA 101 (1205) NA NA Abbrevations: CATCH-IT, Competent Adulthood Transition with Cognitive Behavioral Humanistic and Interpersonal Training; IQR, interquartile range; NA, not available. a Medians compared using Wilcoxon rank-sum test.

(14.9) <.001 Days visited site 1.6 (1.6) 0.9 (1.1) 0.6 (0.3 to 0.9) 1.0 (2.0) 1.0 (1.0) <.001 Total characters typed, No. 716 (977) NA NA 101 (1205) NA NA Abbrevations: CATCH-IT, Competent Adulthood Transition with Cognitive Behavioral Humanistic and Interpersonal Training; IQR, interquartile range; NA, not available. a Medians compared using Wilcoxon rank-sum test. Outcomes For the primary outcome of time-to-depressive episode (DSR ≥3) using intention-to-treat analyses (N = 369), unadjusted HR was 0.59 (95% CI, 0.27-1.29; P = .18), and adjusted HR was 0.53 (95% CI 0.23-1.23, P = .14). Proportional hazards assumption was met (P = .89). For per protocol analysis (≥2 modules completed on either arm, n = 245) (Figure 2), unadjusted HR was 0.41 (95% CI, 0.17-0.99; P = .05), and adjusted HR was 0.44 (95% CI, 0.18-1.08; P = .07). After adjusting for potential confounders using propensity scores, HR was 0.52 (95% CI, 0.19, 1.42; P = .20). Additional analyses are shown in eTable 3 in Supplement 2, and incidence rates in eTable 4 in Supplement 2. The trial calculated the number needed to treat to indicate the number of adolescents who would need to receive the intervention to prevent 1 additional onset of depressive disorder and found a number needed to treat of 36 for the main effect of CATCH-IT. Adolescents with higher baseline CES-D10 scores showed a significantly stronger effect of CATCH-IT on time to event relative to those with lower baseline scores (HR, 0.82; 95% CI, 0.67-0.99; P = .04) (Figure 3; eTable 5 in Supplement 2). For example, the hazard ratio for a CES-D10 score of 15 was 0.20 (95% CI, 0.05-0.77), compared with a hazard ratio of 1.44 (95% CI, 0.41-5.03) for a CES-D10 score of 5. Sex, ethnicity, race, and age did not predict outcome or interact significantly with the interventions and outcomes. Both CATCH-IT and HE demonstrated reduced depressed mood and improved functional status, with no statistically significant differences at 6 months (eTables 6 and 7 in Supplement 2).

a CES-D10 score of 5. Sex, ethnicity, race, and age did not predict outcome or interact significantly with the interventions and outcomes. Both CATCH-IT and HE demonstrated reduced depressed mood and improved functional status, with no statistically significant differences at 6 months (eTables 6 and 7 in Supplement 2). Figure 2. Time to First Depressive Episode for Those Completing 2 or More Modules (Per-Protocol 2 Analysis) CATCH-IT indicates competent adulthood transition with cognitive behavioral humanistic and interpersonal training. Figure 3. First Depressive Episode Survival Analysis by Adolescent Baseline CES-D Scores CATCH-IT indicates competent adulthood transition with cognitive behavioral humanistic and interpersonal training; and CES-D, Center for Epidemiologic Studies Depression scale.

Figure 2. Time to First Depressive Episode for Those Completing 2 or More Modules (Per-Protocol 2 Analysis) CATCH-IT indicates competent adulthood transition with cognitive behavioral humanistic and interpersonal training. Figure 3. First Depressive Episode Survival Analysis by Adolescent Baseline CES-D Scores CATCH-IT indicates competent adulthood transition with cognitive behavioral humanistic and interpersonal training; and CES-D, Center for Epidemiologic Studies Depression scale. Missing Data At least 1 follow-up K-SADS was completed for 85% of the sample, at least 1 follow-up CES-D10 was completed for 80% of the sample, and at least 1 follow-up GAS assessment was completed for 81% of the sample. Dropout between CATCH-IT and HE was different at 2 and 6 months for both K-SADS and CES-D10 (eTable 8A in Supplement 2). At 6 months, K-SADS data were missing for 48 participants (25%) of CATCH-IT and 23 participants (13%) of HE (P = .004), and CES-D10 was missing for 77 participants (40%) of CATCH-IT and 43 participants (24%) of HE (P = .002). Significant predictors of missing K-SADS at 6 months were randomization to CATCH-IT (CATCH-IT vs HE: odds ratio [OR], 2.62; 95% CI, 1.43-4.79; P = .002), living in Chicago (Boston vs Chicago: OR, 0.20; 95% CI, 0.09-0.46; P < .001), age at baseline (OR, 1.23; 95% CI, 1.02-1.49; P = .03), and maternal education (high school graduate or less vs college graduate: OR, 2.99; 95% CI, 1.37-6.53; P = .01) (eTable 8C in Supplement 2). Having a past episode or high CES-D10 at baseline was not associated with missing follow-up.

95% CI, 0.09-0.46; P < .001), age at baseline (OR, 1.23; 95% CI, 1.02-1.49; P = .03), and maternal education (high school graduate or less vs college graduate: OR, 2.99; 95% CI, 1.37-6.53; P = .01) (eTable 8C in Supplement 2). Having a past episode or high CES-D10 at baseline was not associated with missing follow-up. Discussion Overall, we observed a nonsignificant decrease in depressive disorders at 6 months in CATCH-IT as compared with HE. Adolescents and parents devoted substantial time to both interventions, and both conditions experienced decreased depressive symptoms and improved functional status. However, higher-risk adolescents demonstrated greater benefit from CATCH-IT, achieving as much as 80% risk reduction with a CES-D10 score of more than 15, but those without symptoms showed no such benefit. While regression to the mean is a possible explanation for the moderating effect of high CES-D on CATCH-IT, other studies have found that preventive effects for depression interventions are stronger for indicated vs universal samples.44 Moreover, the same effect did not emerge for the HE condition—higher CES-D scores did not moderate the effect of the HE condition—suggesting regression to the mean may not explain the group difference found. For the 66% of adolescent and parent pairs who completed at least 2 modules (63% for CATCH-IT and 70% for HE), the unadjusted analysis showed CATCH-IT reduced the risk of mDE and MDE by 59%, but this was not significant after adjustment for demographic factors or after analyses incorporating propensity scoring.

found. For the 66% of adolescent and parent pairs who completed at least 2 modules (63% for CATCH-IT and 70% for HE), the unadjusted analysis showed CATCH-IT reduced the risk of mDE and MDE by 59%, but this was not significant after adjustment for demographic factors or after analyses incorporating propensity scoring. To our knowledge, this is the first clinical trial in adolescents to evaluate whether depressive episodes can be prevented in primary care settings.11 Our finding that the risk of depressive episodes may be reduced for adolescents with subsyndromal depression is consistent with our earlier phase 2 clinical trial, which only included adolescents with subsyndromal depression.45 Results were not significant in the intention-to-treat main effect analysis, but this may be the result of heterogeneity of treatment effect whereby CATCH-IT is favored for those with subsyndromal depression, but not for those with prior depressive episode alone. Perhaps CATCH-IT bored or frustrated adolescents without current symptoms, or conversely, elicited increased surveillance of symptoms or stimulated memories of prior episodes.46,47 Alternatively, adolescents who are not symptomatic may be less motivated to complete CATCH-IT, the more self-directed intervention, and may actually prefer HE, which did not require substantial effort, perhaps even gaining a sense of self-efficacy.34,46,47,48,49 Also, despite spending substantial time engaged with this intervention, the low number of modules completed may have attenuated impact. Additionally, the borderline significant findings favoring CATCH-IT with the completion of 2 modules (eTable 3 in Supplement 2) suggest that there may also be a threshold effect whereby sufficient numbers of modules may need to be completed for CATCH-IT to be more efficacious than HE.

pleted may have attenuated impact. Additionally, the borderline significant findings favoring CATCH-IT with the completion of 2 modules (eTable 3 in Supplement 2) suggest that there may also be a threshold effect whereby sufficient numbers of modules may need to be completed for CATCH-IT to be more efficacious than HE. Depression prevention programs have shown mixed results.12 The only other primary care trial with adolescents demonstrated improvements in explanatory style but not depressed mood.50 Our findings showing that increased participation may predict better outcomes are consistent with prior reports.50,51 The observed risk reduction across multiple outcomes (DSR ≥3, ≥4, and 5; eTable 3 in Supplement 2), even if not statistically significant, is comparable with other trials.7,8,28 While most internet interventions demonstrate favorable changes in depressed mood, this study did not demonstrate between group differences for mood or functional status.10 However, this is similar to the phase 2 clinical trial of CATCH-IT, which demonstrated lower cumulative prevalence of depressive episodes, but not between group differences in depressed mood.38 It is possible the extensive human contact within this trial had an ameliorating effect on mood and strengthened functional status, effectively blurring between group results. A clinical trial of CATCH-IT in Hong Kong that only used self-report instruments demonstrated a significant between group effect (effect size = 0.36) for depressed mood at 12 months.17

ct within this trial had an ameliorating effect on mood and strengthened functional status, effectively blurring between group results. A clinical trial of CATCH-IT in Hong Kong that only used self-report instruments demonstrated a significant between group effect (effect size = 0.36) for depressed mood at 12 months.17 This study has a robust prevention design implemented in a population-based model in primary care.52,53 The implementation of our study at 2 sites and 8 health systems has rarely been accomplished in studies of child psychiatric conditions.20 This study fits the model by Curran et al19 of hybrid efficacy and implementation studies and substantially enhances generalizability. The attention control condition, which included guidelines for adolescent depression in primary care and chronic care model elements, no doubt reduced between-group differences.24,36 However, given the need for ethical care of adolescents at risk for depressive episodes, a “no intervention” or “wait list control” condition is not possible.53,54

ition, which included guidelines for adolescent depression in primary care and chronic care model elements, no doubt reduced between-group differences.24,36 However, given the need for ethical care of adolescents at risk for depressive episodes, a “no intervention” or “wait list control” condition is not possible.53,54 Limitations This study had limitations, including the relatively low adherence rate of teens and parents. Module completion for CATCH-IT was consistent with pilot findings. A review of internet-based mental health interventions for youths revealed completion ranged from 24% to 85%, and it was not necessary to complete the entire intervention for positive benefits to emerge.12 In addition, module completion does not correlate with time spent, as the HE modules are significantly shorter than CATCH-IT; overall, CATCH-IT participants spent more time using the intervention. Nevertheless, future research should examine why adolescents did not complete the interventions, and explore strategies for boosting adherence. Also, our incidence rate for depressive disorders was low, thus, increasing the number of participants needed to have adequate power to detect group differences. We do not know for certain whether intervention effects can be attributed to the internet-based modules or to the MIs, although results from our pilot study suggest that adolescents who did not get MIs still evidenced reduced symptoms of depression at follow-up. Other limitations include the findings of differential attrition, which were adjusted analytically, and the fact that researchers enrolled only 92% of the target sample.

c Calculated as weight in kilograms divided by height in meters squared. d A pack-year indicates smoking 1 pack of cigarettes per day for a year. e Excess alcohol intake was defined as more than 21 units/wk for men and more than 14 units/wk for women. f TNM classification as per the American Joint Committee on Cancer’s Cancer Staging Manual, 7th Edition. HPV Status and Survival Table 2 depicts the mean follow-up, DFS, OS, and recurrence or progression rate for the HPV-positive patients with HGD and EAC and the viral-negative groups. Mean (SD) DFS was 40.3 (33.8) months in the HPV-positive esophageal lesion group and 24.1 (25.5) months in the HPV-negative cohort (difference, 16.2 months; 95% CI, 5.7-26.8 months; P = .003). Similarly, mean (SD) OS was 43.7 (32.9) months in the viral-positive group compared with 29.8 (25.3) months in the HPV-negative patients (difference, 13.9 months; 95% CI, 3.6 to 24.3 months; P = .009). As expected, recurrence or progression was much reduced in the HPV-positive cohort compared with the HPV-negative cohort (9 of 37 [24.3%] vs 61 of 105 [58.1%]; difference, −33.8%; 95% CI, −50.5% to −17.0%; P < .001). Recurrence per se was almost a third less in the HPV-positive patients in comparison with HPV-negative patients (6 of 37 [16.2%] vs 46 of 105 [43.8%]; difference, −27.6%; 95% CI, −42.8% to −12.4%; P = .003). There appeared to be better (although nonsignificant) local and regional control for the HPV-positive patients compared with viral-negative individuals (6 of 37 [16.2%] vs 32 of 105 [30.5%]; difference, −14.3%; 95% CI, −29.0% to 0.5%; P = .09). Significantly, reduced distant metastasis was observed in the HPV-positive group (3 of 37 [8.1%] vs 29 of 105 [27.6%]; difference, −19.5%; 95% CI, −31.8% to −7.2%; P = .02) as were deaths from EAC (5 of 37 [13.5%] vs 38 of 105 [36.2%]; difference, −22.7%; 95% CI, −37.0% to −8.3%; P = .01).

e MIs, although results from our pilot study suggest that adolescents who did not get MIs still evidenced reduced symptoms of depression at follow-up. Other limitations include the findings of differential attrition, which were adjusted analytically, and the fact that researchers enrolled only 92% of the target sample. Conclusions Our long-term goal for the CATCH-IT intervention is to provide a first-line program for primary care physicians to offer as part of the guidelines for adolescent depression in primary care, to support adolescents while the need for further intervention can be evaluated. We continue to examine moderators that may explain who responds best to this approach. Future directions include the development of versions for personal devices (eg, tablets and mobile phones), and a version individualized for sexual- and gender-minority teens. A scalable, population-based approach to preventing depression in adolescents in primary care may be efficacious for adolescents with subsyndromal depression, but not for those with a prior episode alone. Supplement 1. Trial Protocol Click here for additional data file. Supplement 2. eTable 1. Definition for Criteria for Main Outcomes Analyses eTable 2. MI Fidelity: Mean MITI Summary Scores eTable 3. Hazard Ratio Estimate and 95% CI for First Depressive Episode Comparing CATCH-IT to Health Education eTable 4. Incidence of First Depressive Episode by Treatment Group eTable 5. Adjusted Cox Proportional Hazard Model Results, Including Adolescent Baseline CES-D10 as a Moderator eTable 6. Depressed Mood eTable 7. Functional Status

eTable 3. Hazard Ratio Estimate and 95% CI for First Depressive Episode Comparing CATCH-IT to Health Education eTable 4. Incidence of First Depressive Episode by Treatment Group eTable 5. Adjusted Cox Proportional Hazard Model Results, Including Adolescent Baseline CES-D10 as a Moderator eTable 6. Depressed Mood eTable 7. Functional Status eTable 8. A, Participants With Missing Data: KSADS and CES-D10. B, Predictors of Missing 2-Month Follow-up: Time to Last Assessment <2 Months. C, Predictors of Missing 6-Month Follow-up: Time to Last Assessment <6 Months. D, Predictors of Missing 2-Month CES-D10. E, Predictors of Missing 6-Month CES-D10 Click here for additional data file. Supplement 3. Data Sharing Statement Click here for additional data file.

Introduction Disease subclassification according to the AJCC Cancer Staging Manual by the American Joint Committee on Cancer (AJCC) is the customary and prevalent mode for stratifying patients with melanoma to estimate prognosis, determine appropriate surgical intervention, and assess eligibility for adjuvant therapies and clinical trials. The process presupposes that pathologists’ application of the AJCC histopathological criteria to individual cases of melanoma is accurate and reproducible. However, in the field of melanoma, there are only limited analyses quantifying the degree of reproducibility of AJCC microstaging between pathology observers.1 Extensive variability has been noted among pathologists in the diagnosis of invasive melanoma.2,3,4,5,6,7 One of the largest studies,2 our previously published Melanoma Pathology Study (M-Path) of 187 US pathologists, found less than 50% agreement between pathologists and a consensus-derived reference diagnosis of T1a invasive melanoma, with improvement to 72% concordance for invasive melanoma T1b or greater. Similarly, M-Path findings revealed only 46% interobserver agreement for T1a invasive melanoma, and 77% agreement for T1b or greater melanomas.2

agreement between pathologists and a consensus-derived reference diagnosis of T1a invasive melanoma, with improvement to 72% concordance for invasive melanoma T1b or greater. Similarly, M-Path findings revealed only 46% interobserver agreement for T1a invasive melanoma, and 77% agreement for T1b or greater melanomas.2 The previous study2 was conceived and executed in the context of the AJCC Cancer Staging Manual, 7th edition (AJCC 7) staging system. Across interpretations at 2 points, pathologists’ intraobserver reproducibility reached 63% for T1a melanomas and 83% for T1b or greater melanomas. Given the updated classification in the AJCC Cancer Staging Manual, 8th edition (AJCC 8), particularly with changes in definitions of T1a vs T1b or greater, the M-Path database enables a new comparison of pathologist concordance with a reference standard and reproducibility in the microstaging of melanoma according to both the existing AJCC 7 and the current AJCC 8.8,9 Briefly, in AJCC 8, the depth for stage T1a is established at 0.8 mm, rather than 1.0 mm, and the presence of ulceration continues to contribute to stage modification, but mitoses do not. In addition, the reporting of Breslow thickness is limited to intervals of tenths of a millimeter rather than hundredths. We assess whether changes in criteria in the newer AJCC 8 are associated with changes in concordance and reliability, and whether observer interpretations of histological alterations within melanocytic lesions are reliable in the context of the demands of microstaging and its consequences per the AJCC schema.

ndredths. We assess whether changes in criteria in the newer AJCC 8 are associated with changes in concordance and reliability, and whether observer interpretations of histological alterations within melanocytic lesions are reliable in the context of the demands of microstaging and its consequences per the AJCC schema. Methods Study Design The data used in this diagnostic study are derived from the M-Path study,2 which was described previously. Practicing pathologists from 10 US states who actively interpreted melanocytic skin biopsy lesions as part of their usual clinical practice and planned to continue practicing for a minimum of 2 subsequent years were invited to participate. This study was approved by the institutional review boards of Dartmouth College, the Fred Hutchinson Cancer Research Center, Oregon Health and Science University, and the University of Washington. Informed consent was obtained from every participating pathologist using an online platform. Each pathologist was randomized to interpret the same set of melanocytic skin biopsy cases on 2 occasions, at least 8 months apart. The study cases (n = 240) were assembled into 5 sets of 48 cases, each represented by a single glass slide. Each set included the full spectrum of melanocytic skin lesions (eg, from benign to invasive melanoma).

ndomized to interpret the same set of melanocytic skin biopsy cases on 2 occasions, at least 8 months apart. The study cases (n = 240) were assembled into 5 sets of 48 cases, each represented by a single glass slide. Each set included the full spectrum of melanocytic skin lesions (eg, from benign to invasive melanoma). Participating pathologists independently reviewed the same cases using the same glass slides. Participants entered diagnostic interpretations into an online Melanocytic Pathology Assessment Tool and Hierarchy for Diagnosis (MPATH-Dx) histology form for each case, choosing from a diverse and comprehensive list of more than 50 diagnostic terms. We asked participants to assume that the single glass slide for each case was representative of the entire lesion and that the margin was involved (irrespective of whether it involved the biopsy margin). Research analysts subsequently mapped diagnostic interpretations into 1 of 5 diagnostic classes according to the MPATH-Dx mapping scheme.10 Examples of diagnostic terms for each class and suggested treatment recommendations, provided under the assumption that specimen margins are positive, are depicted in Table 1. Because the AJCC 8 criteria changes only affect MPATH-Dx classes IV (T1a) and V (≥T1b), this article focuses on the distinction between invasive melanomas exclusively.

terms for each class and suggested treatment recommendations, provided under the assumption that specimen margins are positive, are depicted in Table 1. Because the AJCC 8 criteria changes only affect MPATH-Dx classes IV (T1a) and V (≥T1b), this article focuses on the distinction between invasive melanomas exclusively. Table 1. The MPATH-Dx Reporting Schema for Melanocytic Skin Lesion Classification Into 5 Diagnostic Classes, as Used in This Studya MPATH-Dx Class Perceived Risk for Progression Suggested Interventionb Examples 0 Incomplete study due to sampling or technical limitations Repeat biopsy or short-term follow-up NA I Very low risk No further treatment Common melanocytic nevus; blue nevus; mildly dysplastic nevus II Low risk Narrow but complete excision (<5 mm) Moderately dysplastic nevus; Spitz nevus III Slightly higher risk, greater need for intervention Complete excision with ≥5-mm but <1-cm margins Severely dysplastic nevus; melanoma in situ; atypical Spitz tumor IV Substantial risk for local or regional progression Wide local excision with ≥1-cm margins Thin invasive melanomas (eg, T1a) V Greatest risk for regional and/or distant metastases Wide local excision with ≥1-cm margins; consideration of staging sentinel lymph node biopsy; adjuvant therapy Thicker invasive melanoma (eg, T1b, stage ≥2) Abbreviations: MPATH-Dx, Melanocytic Pathology Assessment Tool and Hierarchy for Diagnosis; NA, not applicable.

Greatest risk for regional and/or distant metastases Wide local excision with ≥1-cm margins; consideration of staging sentinel lymph node biopsy; adjuvant therapy Thicker invasive melanoma (eg, T1b, stage ≥2) Abbreviations: MPATH-Dx, Melanocytic Pathology Assessment Tool and Hierarchy for Diagnosis; NA, not applicable. a Adapted from Piepkorn et al.10 These examples of suggested interventions were developed at the beginning of the study, are presented for consideration only, and may be out of date or controversial in some instances. Additional consensus development should proceed before these guidelines are adopted for general use, and they should be adapted according to individual national circumstances. In particular, the suggestions for melanoma should follow published national guidelines as most recently updated. b Assuming representative sampling of the lesion. Before data collection, a panel of 3 experienced dermatopathologists independently reviewed the hematoxylin-eosin–stained glass slides for each case followed by consensus review using a modified Delphi approach.11,12 This process was used to develop a consensus diagnosis for each of the M-Path study cases. Only 116 cases of invasive melanoma, as defined by the consensus diagnosis, were considered in this analysis. Three cases included in the original M-Path study as class IV were excluded here because classification was based on a treatment recommendation of wide excision but these cases were assessed as melanocytic lesions of uncertain malignant potential.

as defined by the consensus diagnosis, were considered in this analysis. Three cases included in the original M-Path study as class IV were excluded here because classification was based on a treatment recommendation of wide excision but these cases were assessed as melanocytic lesions of uncertain malignant potential. Statistical Analysis For each case, the consensus reference diagnosis and the participating pathologists’ interpretations were classified into the MPATH-Dx class IV (T1a) or class V (≥T1b) using both the AJCC 7 and AJCC 8 criteria.8,9 Accuracy outcome measures included overinterpretation, underinterpretation, and concordance of participant interpretations with the relevant (AJCC 7 or AJCC 8) reference diagnosis. We defined overinterpretation as diagnosing cases at a higher diagnostic class than the reference diagnosis, and underinterpretation as diagnosing cases at a lower diagnostic class than the reference diagnosis. Interpretations in agreement with the reference diagnosis were concordant. Confidence intervals accounted for both within-participant and across-participant variability by using variance estimates of the following form: {var(ratep) + [ave(ratep) × (1−ave(ratep))]/nc}/np, where ave(ratep) is the average rate among pathologists, var(ratep) is the sample variance of rates among pathologists, nc is the number of cases interpreted by each pathologist, and np is the number of pathologists. Logistic regression models were used to test for a difference in accuracy between AJCC 7– and AJCC 8–based mappings. Models used robust estimators of the variance to account for correlation of case interpretations from the same pathologist.

ts for differences between AJCC 7– and AJCC 8–based mappings used a Wald statistic based on the bootstrap standard error of the difference. A total of 3000 bootstrap samples were obtained by participant-level sampling with replacement and generation of all possible pairs of distinct sample participants for each sample. For intraobserver concordance among the 118 participants who interpreted the same glass slides on 2 occasions, we calculated the proportion of cases with both interpretations in the same diagnostic class. Confidence intervals for intraobserver concordance rates used a logit transformation and robust standard error that accounted for clustering at pathologist level. Logistic regression models were used to test for a difference in intraobserver concordance between AJCC 7– and AJCC 8–based mappings. All P values correspond to 2-tailed tests and differences with P < .05 were considered to be statistically significant. Analyses were performed using Stata statistical software (StataCorp), version 14. Results The 116 skin biopsy cases defined as invasive melanoma per the consensus reference diagnosis included 55 cases (47%) of T1a invasive melanoma and 61 cases (53%) of T1b or greater using AJCC 7. When AJCC 8 staging criteria were applied, the consensus reference diagnosis was upgraded from T1a to T1b or greater for 4 of 55 cases (7%) and downgraded from T1b or greater to T1a for 19 of 61 cases (31%). The reclassification of invasive cases by consensus reference diagnosis under AJCC 8 resulted in 70 T1a cases (60%) and 46 cases (40%) of T1b or greater.

, the consensus reference diagnosis was upgraded from T1a to T1b or greater for 4 of 55 cases (7%) and downgraded from T1b or greater to T1a for 19 of 61 cases (31%). The reclassification of invasive cases by consensus reference diagnosis under AJCC 8 resulted in 70 T1a cases (60%) and 46 cases (40%) of T1b or greater. Of 301 eligible pathologists, 187 (62%) enrolled and completed independent interpretations. In the first round of interpretations, the pathologists completed 4342 independent case interpretations of the invasive melanoma cases. Similar to the aforementioned movement in consensus reference diagnoses, participant diagnoses were upgraded from T1a to T1b or greater for 136 of 1229 T1a assessments (11%) and downgraded from T1b or greater to T1a for 467 of 1841 assessments (25%). As shown in Table 2, concordance and reproducibility were improved when using the AJCC 8 criteria vs the earlier AJCC 7 criteria. With regard to T1a diagnoses, participating pathologists’ concordance with the consensus reference diagnosis increased from 44% (95% CI, 41%-48%), using AJCC 7 criteria, to 54% (95% CI, 51%-57%), using AJCC 8 criteria. The concordance for T1b or greater cases increased from 72% (95% CI, 69%-75%) to 78% (95% CI, 75%-80%). The increased concordance associated with using the AJCC 8 reduced both underinterpretation and overinterpretation.

from 44% (95% CI, 41%-48%), using AJCC 7 criteria, to 54% (95% CI, 51%-57%), using AJCC 8 criteria. The concordance for T1b or greater cases increased from 72% (95% CI, 69%-75%) to 78% (95% CI, 75%-80%). The increased concordance associated with using the AJCC 8 reduced both underinterpretation and overinterpretation. Table 2. Changes in Concordance, Interobserver Agreement, and Intraobserver Reproducibility When Comparing AJCC 7 With AJCC 8 AJCC Cancer Staging Manual Edition Total Invasive Melanoma Cases for Consensus, No. % (95% CI) Concordance With Consensus Reference Diagnosis Interobserver Agreement Intraobserver Reproducibility for Same Case at 2 Time Points Underinterpretation Concordance P Valuea Overinterpretation Concordance P Valuea Reproducibility P Valuea AJCC 7 T1a (MPATH-Dx class IV) 55 46 (43-50) 44 (41-48) 9 (8-12) 41 (39-44) 59 (56-63) T1b or greater (MPATH-Dx class V) 61 28 (25-31) 72 (69-75) NA 67 (64-69) 74 (71-76) AJCC 8 T1a 70 39 (36-42) 54 (51-57) <.001 7 (6-8) 51 (48-53) <.001 64 (62-67) .006 T1b or greater 46 22 (20-25) 78 (75-80) <.001 NA 69 (66-73) .02 77 (74-79) .11 Abbreviations: AJCC 7, AJCC Cancer Staging Manual, 7th edition; AJCC 8, AJCC Cancer Staging Manual, 8th edition; MPATH-Dx, Melanocytic Pathology Assessment Tool and Hierarchy for Diagnosis; NA, not applicable. a P values for test of concordance, interobserver agreement, and intraobserver reproducibility rate differences between AJCC 7– and AJCC 8–based mappings.

Table 2. Changes in Concordance, Interobserver Agreement, and Intraobserver Reproducibility When Comparing AJCC 7 With AJCC 8 AJCC Cancer Staging Manual Edition Total Invasive Melanoma Cases for Consensus, No. % (95% CI) Concordance With Consensus Reference Diagnosis Interobserver Agreement Intraobserver Reproducibility for Same Case at 2 Time Points Underinterpretation Concordance P Valuea Overinterpretation Concordance P Valuea Reproducibility P Valuea AJCC 7 T1a (MPATH-Dx class IV) 55 46 (43-50) 44 (41-48) 9 (8-12) 41 (39-44) 59 (56-63) T1b or greater (MPATH-Dx class V) 61 28 (25-31) 72 (69-75) NA 67 (64-69) 74 (71-76) AJCC 8 T1a 70 39 (36-42) 54 (51-57) <.001 7 (6-8) 51 (48-53) <.001 64 (62-67) .006 T1b or greater 46 22 (20-25) 78 (75-80) <.001 NA 69 (66-73) .02 77 (74-79) .11 Abbreviations: AJCC 7, AJCC Cancer Staging Manual, 7th edition; AJCC 8, AJCC Cancer Staging Manual, 8th edition; MPATH-Dx, Melanocytic Pathology Assessment Tool and Hierarchy for Diagnosis; NA, not applicable. a P values for test of concordance, interobserver agreement, and intraobserver reproducibility rate differences between AJCC 7– and AJCC 8–based mappings. The intraobserver reproducibility of diagnoses also improved when using the AJCC 8 criteria, increasing from 59% (95% CI, 56%-63%) to 64% (95% CI, 62%-67%) for T1a invasive melanoma, and from 74% (95% CI, 71%-76%) to 77% (95% CI, 74%-79%) for T1b or greater invasive melanoma cases. Average pairwise-interobserver agreement increased from 41% (95% CI, 39%-44%) to 51% (95% CI, 48%-53%) for T1a cases, and from 67% (95% CI, 64%-69%) to 69% (95% CI, 66%-73%) for T1b or greater cases.

a invasive melanoma, and from 74% (95% CI, 71%-76%) to 77% (95% CI, 74%-79%) for T1b or greater invasive melanoma cases. Average pairwise-interobserver agreement increased from 41% (95% CI, 39%-44%) to 51% (95% CI, 48%-53%) for T1a cases, and from 67% (95% CI, 64%-69%) to 69% (95% CI, 66%-73%) for T1b or greater cases. Discussion This analysis provides data that the new AJCC 8 criteria may lead to improved concordance and reproducibility among pathologists in the classification of invasive melanoma, although the size of this effect is modest. One explanation of the improvement in concordance of pathological staging of T1a and T1b melanoma in AJCC 8 is the change in stage T1 subgroups and criteria from AJCC 7. In AJCC 7, the criteria for T1b were presence of dermal mitotic activity, Breslow thickness, or epidermal ulceration,8 whereas in AJCC 8, the primary determinants for T1a vs T1b were Breslow thickness and ulceration, with the elimination of mitotic activity.9

s the change in stage T1 subgroups and criteria from AJCC 7. In AJCC 7, the criteria for T1b were presence of dermal mitotic activity, Breslow thickness, or epidermal ulceration,8 whereas in AJCC 8, the primary determinants for T1a vs T1b were Breslow thickness and ulceration, with the elimination of mitotic activity.9 In AJCC 8, T1b is now defined by Breslow thickness 0.8 mm or greater or ulceration in melanomas smaller than 0.8 mm. Because recognition of mitoses in thin melanomas is considered potentially unreliable13 and the recording of Breslow thickness more reliable,14 one would expect to find greater reliability of both T1a and T1b classification in the AJCC 8 staging. In fact, our results correspond exactly to this presupposed increase in reliability of classification of T1a and T1b in AJCC 8. A retrospective restaging of the Netherlands Cancer Registry database also reported a modest improvement in stratification of pT1 melanoma associated with the implementation of AJCC 8 criteria.15

fact, our results correspond exactly to this presupposed increase in reliability of classification of T1a and T1b in AJCC 8. A retrospective restaging of the Netherlands Cancer Registry database also reported a modest improvement in stratification of pT1 melanoma associated with the implementation of AJCC 8 criteria.15 Limitations Limitations of the study include interpretation of a single slide (although participants were asked to assume the slide was representative), use of a testing environment rather than a practice setting, and inability to obtain second opinions and clinical histories. Also, there is no established method to define a gold-standard diagnosis; therefore, improvement in concordance with an expert-defined reference should not necessarily be interpreted as improvement in accuracy. We chose to use the consensus of 3 experienced pathologists because this approach could be replicated in clinical practice. Finally, the relative proportions of cases used for this study are not representative of the population.16 Strengths include a large number of participating pathologists reviewing the same glass slides on 2 occasions and the ability to assess both concordance with a reference and reproducibility.

in clinical practice. Finally, the relative proportions of cases used for this study are not representative of the population.16 Strengths include a large number of participating pathologists reviewing the same glass slides on 2 occasions and the ability to assess both concordance with a reference and reproducibility. Conclusions Our results suggest that the changes in the AJCC staging will likely have a positive effect on patients. The consequences of melanoma staging to patients are substantial. Among these are patients’ perceptions of long-term implications to their health as determined by the particular stage assigned at diagnosis, economic consequences of health care services, and the magnitude of surgical interventions indicated by the staging classification (eg, size of wide local resection, eligibility for sentinel lymphatic mapping, and implications for other therapies). In view of these clinical ramifications, even modest improvements of 6% to 10% in diagnostic concordance resulting from changes from AJCC 7 to AJCC 8 are important. However, despite improvement, concordance and reproducibility remain low and suggest that conventional histopathology has been parsed to a degree that falls below the limits of reliability for the demands and consequences of the staging schemata that have evolved over time.

Introduction Attention-deficit/hyperactivity disorder (ADHD) is a common neurodevelopmental disorder that typically starts in childhood.1 The symptoms of ADHD, such as inattention, impulsivity, disorganization, and, to a lesser extent, hyperactivity, can carry into adulthood.1,2 More than two-thirds of children who receive a diagnosis of ADHD continue to experience at least 1 ADHD symptom during their adult lives.3 Current estimates show that approximately 4% of the US adult population lives with ADHD.4 Persistent ADHD symptoms can adversely affect many dimensions of an individual’s life, including physical, social, occupational, and behavioral functioning and can lower overall quality of life.5,6 Stimulants and the nonstimulant atomoxetine are recommended as first-line treatments of ADHD.7,8 Although long-term use of stimulants at appropriate therapeutic doses is considered safe, combination therapy of stimulants with other drugs with euphoric effects, such as opioids, may increase the risk of drug dependence.9 Both stimulants and opioids can increase dopamine release in the brain—the former directly enhances the effect of dopamine, whereas the latter indirectly promotes dopamine levels by activating opioid receptors.10,11,12 Thus, stimulants and opioids, when used concurrently, may synergistically reinforce dopamine signals and prolong the action of dopamine neuronal activities, leading to euphoric effects.10,11,12,13

nces the effect of dopamine, whereas the latter indirectly promotes dopamine levels by activating opioid receptors.10,11,12 Thus, stimulants and opioids, when used concurrently, may synergistically reinforce dopamine signals and prolong the action of dopamine neuronal activities, leading to euphoric effects.10,11,12,13 Despite the potential enhanced abuse risk of coadministered stimulant and opioid drugs, the United States does not restrict central nervous system stimulant prescriptions for individuals with substance use disorder or for those receiving opioid treatment, although such policy has been implemented elsewhere.14 Both opioids and stimulants (eg, methylphenidate, methamphetamine, or amphetamine) are regulated as controlled substances in the United States.15,16 Given the more than 8-fold increase in stimulant prescriptions and 4-fold increase in opioid prescriptions during the past 20 years,17 concerns have intensified regarding increased risk for misuse or abuse of these medications, alone or together, and the sequelae.18,19

s controlled substances in the United States.15,16 Given the more than 8-fold increase in stimulant prescriptions and 4-fold increase in opioid prescriptions during the past 20 years,17 concerns have intensified regarding increased risk for misuse or abuse of these medications, alone or together, and the sequelae.18,19 The prevalence of ADHD diagnosis and treatment has dramatically increased in the past decade among adults, especially among those insured by Medicaid.20 With this increase and with the potential risk of prescription drug abuse among adults with ADHD, it is crucial to understand whether long-term concurrent use of stimulants and opioids is common, whether the likelihood of such a drug combination changes over time, and whether there are certain factors associated with adults with ADHD that may cause them to be more likely to use both drugs long term. The present study provides evidence of the prevalence and secular trends of and the factors associated with long-term concurrent stimulant-opioid use among adult Medicaid enrollees with ADHD.

er there are certain factors associated with adults with ADHD that may cause them to be more likely to use both drugs long term. The present study provides evidence of the prevalence and secular trends of and the factors associated with long-term concurrent stimulant-opioid use among adult Medicaid enrollees with ADHD. Methods Study Design and Source This multiyear, cross-sectional study used Medicaid Analytic eXtract (MAX) files of 29 states from 1999 to 2010. The MAX files contain beneficiary billing records for inpatient and outpatient encounters and pharmacy-filled prescriptions. The inpatient and outpatient records include details on procedures and diagnoses coded using the International Classification of Disease, Ninth Revision, Clinical Modification (ICD-9-CM) billing codes. Beneficiary demographics, enrollment status, and mortality information are provided in the MAX personal summary file. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for cross-sectional studies. The University of Florida, Gainesville, and the Centers for Medicare & Medicaid Services (CMS) institutional review and privacy boards approved this study with a waiver of informed consent and a waiver of Health Insurance Portability and Accountability Act authorization.

reporting guideline for cross-sectional studies. The University of Florida, Gainesville, and the Centers for Medicare & Medicaid Services (CMS) institutional review and privacy boards approved this study with a waiver of informed consent and a waiver of Health Insurance Portability and Accountability Act authorization. The majority of the data were obtained before the CMS suppressed all substance use disorder–related encounter claims for its beneficiaries under the Confidentiality of Alcohol and Drug Abuse Patient Records Regulations. The regulations underwent a change in early 2017, revoking redaction of substance use disorder claims. Although we are awaiting CMS approval for a new data user agreement for more recent data (up to 2013), we used MAX data from 1999 to 2010 for the present study to provide timely reports on long-term concurrent use of stimulants and opioids among adults with ADHD. Data analyses were conducted between January 1 and December 31, 2017.

re awaiting CMS approval for a new data user agreement for more recent data (up to 2013), we used MAX data from 1999 to 2010 for the present study to provide timely reports on long-term concurrent use of stimulants and opioids among adults with ADHD. Data analyses were conducted between January 1 and December 31, 2017. Study Sample To construct the study sample, we first identified adult patients enrolled in Medicaid fee-for-service plans who were aged 20 to 64 years and had at least 1 inpatient or 2 outpatient visits that were coded with an ADHD diagnosis (ICD-9-CM codes 314.xx) during 1999 and 2010.21 To address enrollment gap issues (ie, disruptions in health plan enrollment that lead to no insurance coverage or temporary employer-based coverage, which is commonly seen in Medicaid beneficiaries),22 we randomly selected for each patient one 12-month continuous enrollment period following receipt of an ADHD diagnosis. We used the first half of each 12-month observation period to determine baseline demographic and clinical characteristics and used the second half (ie, 6-month follow-up) to ascertain long-term concurrent use of stimulants and opioids (outcome). This approach that randomly selects an observation period for each patient has been used in previous studies to generate population-based estimates of medication exposure.23,24

cal characteristics and used the second half (ie, 6-month follow-up) to ascertain long-term concurrent use of stimulants and opioids (outcome). This approach that randomly selects an observation period for each patient has been used in previous studies to generate population-based estimates of medication exposure.23,24 Medications of Interest The medications of interest captured through the MAX pharmacy files included drug treatment of ADHD and opioids. The ADHD medications for adults included atomoxetine and stimulants (methylphenidate, dexmethylphenidate, mixed amphetamine salts, dextroamphetamine, methamphetamine, and pemoline). We included the following opioid agents approved for use in the United States by 2010: buprenorphine, butorphanol, codeine, dihydrocodeine, fentanyl, hydrocodone, hydromorphone, levorphanol, nalbuphine, meperidine, methadone, morphine, opium, oxycodone, oxymorphone, pentazocine, propoxyphene, tapentadol, and tramadol.

We included the following opioid agents approved for use in the United States by 2010: buprenorphine, butorphanol, codeine, dihydrocodeine, fentanyl, hydrocodone, hydromorphone, levorphanol, nalbuphine, meperidine, methadone, morphine, opium, oxycodone, oxymorphone, pentazocine, propoxyphene, tapentadol, and tramadol. Measure of Long-term Concurrent Stimulant-Opioid Use We examined daily stimulant or opioid exposure based on the number of days’ supply of prescription claims during the 6-month follow-up of patients who received a diagnosis of ADHD. A grace period of 7 days was used to account for delayed prescription fills.25 Among adults with ADHD, we assessed the proportion of patients having concurrently used stimulants and opioids for at least 30 consecutive days. The minimum of 30 days was used to capture scenarios in which both stimulants and opioids appeared to be used long term, thus increasing the risk for substance use disorder and prescription drug abuse.19 To assess the robustness of our findings, we conducted a sensitivity analysis using a different cutoff period to define concurrent use of stimulants and opioids (ie, concurrent use for ≥15 days).

nd opioids appeared to be used long term, thus increasing the risk for substance use disorder and prescription drug abuse.19 To assess the robustness of our findings, we conducted a sensitivity analysis using a different cutoff period to define concurrent use of stimulants and opioids (ie, concurrent use for ≥15 days). Demographic and Clinical Characteristics Candidate factors associated with long-term concurrent stimulant-opioid use measured at baseline included sociodemographic characteristics and mental and physical comorbidities. Sociodemographic variables included age at baseline, sex, race/ethnicity, rural vs urban residence, reasons for Medicaid eligibility, and state of residence. Age was categorized into 4 groups (20-30, 31-40, 41-50, and 51-64 years). The race/ethnicity classification was based on the information available from the MAX data. Because of the small sample size, Hispanic, Asian, Pacific Islander, and Native American individuals were classified as “other.” The 29 states of residence were categorized into 4 census regions (South, Midwest, Northeast, and West; see the classification of states in the footnote to Table 1) to ensure sufficient sample size for analysis. Mental health comorbidities included schizophrenia, bipolar disorder, depression, anxiety disorder, and substance use disorder. Physical health comorbidities included obesity, diabetes, cardiovascular disease, chronic obstructive pulmonary disease (COPD), and chronic pain, which are prevalent among patients with ADHD26 and represent potential factors associated with prescription opioid use.27 Both mental and physical comorbidities were ascertained based on the presence of at least 1 relevant diagnosis code in any position on inpatient and outpatient encounter claims (eTable 1 in the Supplement).

e prevalent among patients with ADHD26 and represent potential factors associated with prescription opioid use.27 Both mental and physical comorbidities were ascertained based on the presence of at least 1 relevant diagnosis code in any position on inpatient and outpatient encounter claims (eTable 1 in the Supplement). Table 1. Baseline Characteristics of Eligible Medicaid-Enrolled Adults With ADHD Between 1999 and 2010, Overall and Stratified by Stimulant Usea Characteristic Adults With ADHD, No.

e prevalent among patients with ADHD26 and represent potential factors associated with prescription opioid use.27 Both mental and physical comorbidities were ascertained based on the presence of at least 1 relevant diagnosis code in any position on inpatient and outpatient encounter claims (eTable 1 in the Supplement). Table 1. Baseline Characteristics of Eligible Medicaid-Enrolled Adults With ADHD Between 1999 and 2010, Overall and Stratified by Stimulant Usea Characteristic Adults With ADHD, No. (%) Overall Sample No Stimulant Use (n = 44 683) Stimulant Use (n = 21 723) No Opioid Use Short-term Opioid Use Long-term Opioid Use No Opioid Use Short-term Stimulant-Opioid Use Long-term Stimulant-Opioid Use Sample size 66 406 (100) 27 476 (100) 11 381 (100) 5826 (100) 13 129 (100) 5004 (100) 3590 (100) Age, mean (SD), y 31.8 (10.2) 30.1 (9.7) 31.3 (9.5) 37.6 (10.4) 30.6 (9.8) 33.1 (9.9) 38.7 (10.3) 20-30 35 670 (53.7) 16 971 (61.8) 6267 (55.1) 1645 (28.2) 7667 (58.4) 2268 (45.3) 852 (23.7) 31-40 16 479 (24.8) 5964 (21.7) 2970 (26.1) 1848 (31.7) 3008 (22.9) 1542 (30.8) 1147 (32.0) 41-50 9817 (14.8) 3095 (11.3) 1550 (13.6) 1530 (26.3) 1760 (13.4) 855 (17.1) 1027 (28.6) 51-64 4440 (6.7) 1446 (5.3) 594 (5.2) 803 (13.8) 694 (5.3) 339 (6.8) 564 (15.7) Female 37 155 (56.0) 12 650 (46.0) 7688 (67.6) 3835 (65.8) 7122 (54.3) 3400 (68.0) 2460 (68.5) Race/ethnicity Non-Hispanic white 52 551 (79.1) 19 299 (70.2) 9368 (82.3) 5047 (86.6) 11 084 (84.4) 4486 (89.7) 3267 (91.0) Non-Hispanic black 7168 (10.8) 4517 (16.4) 1127 (9.9) 347 (6.0) 897 (6.8) 198 (4.0) 82 (2.3) Otherb 6687 (10.1) 3660 (13.3) 886 (7.8) 432 (7.4) 1148 (8.7) 320 (6.4) 241 (6.7) Rural residency 19 400 (29.2) 7264 (26.4) 3748 (32.9) 2023 (34.7) 3712 (28.3) 1557 (31.1) 1096 (30.5) Medicaid eligibilityc Cash assistance 47 814 (72.0) 20 896 (76.1) 8134 (71.5) 4342 (74.5) 8757 (66.7) 3229 (64.5) 2456 (68.4) Disability 38 205 (57.5) 18 056 (65.7) 5577 (49.0) 3385 (58.1) 7094 (54.0) 2222 (44.4) 1871 (52.1) Region at baselined South 22 621 (34.1) 8693 (31.6) 3943 (34.7) 2073 (35.6) 4589 (35.0) 1931 (38.6) 1392 (38.8) Midwest 27 058 (40.8) 9941 (36.2) 4883 (42.9) 2448 (42.0) 5888 (44.9) 2282 (45.6) 1616 (45.0) Northeast 8201 (12.4) 4996 (18.2) 888 (7.8) 443 (7.6) 1402 (10.7) 268 (5.4) 204 (5.7) West 8526 (12.8) 3846 (14.0) 1667 (14.7) 862 (14.8) 1250 (9.5) 523 (10.5) 378 (10.5) Year of index ADHD diagnosis 1999 2329 (3.5) 1234 (4.5) 467 (4.1) 148 (2.5) 329 (2.5) 101 (2.0) 50 (1.4) 2000 4027 (6.1) 2163 (7.9) 761 (6.7) 334 (5.7) 492 (3.8) 173 (3.5) 104 (2.9) 2001 4220 (6.4) 2165 (7.9) 870 (7.6) 344 (5.9) 557 (4.2) 182 (3.6) 102 (2.8) 2002 4491 (6.8) 2276 (8.3) 875 (7.7) 410 (7.0) 620 (4.7) 202 (4.0) 108 (3.0) 2003 61

nosis 1999 2329 (3.5) 1234 (4.5) 467 (4.1) 148 (2.5) 329 (2.5) 101 (2.0) 50 (1.4) 2000 4027 (6.1) 2163 (7.9) 761 (6.7) 334 (5.7) 492 (3.8) 173 (3.5) 104 (2.9) 2001 4220 (6.4) 2165 (7.9) 870 (7.6) 344 (5.9) 557 (4.2) 182 (3.6) 102 (2.8) 2002 4491 (6.8) 2276 (8.3) 875 (7.7) 410 (7.0) 620 (4.7) 202 (4.0) 108 (3.0) 2003 61 62 (9.3) 2662 (9.7) 1190 (10.5) 565 (9.7) 1107 (8.4) 415 (8.3) 223 (6.2) 2004 7925 (11.9) 2911 (10.6) 1461 (12.8) 681 (11.7) 1744 (13.3) 729 (14.6) 399 (11.1) 2005 7393 (11.1) 2823 (10.3) 1209 (10.6) 595 (10.2) 1636 (12.5) 672 (13.4) 458 (12.8) 2006 6370 (9.6) 2554 (9.3) 1055 (9.3) 562 (9.7) 1285 (9.8) 530 (10.6) 384 (10.7) 2007 5525 (8.3) 2189 (8.0) 866 (7.6) 541 (9.3) 1161 (8.8) 409 (8.2) 359 (10.0) 2008 6002 (9.0) 2379 (8.7) 927 (8.2) 592 (10.2) 1223 (9.3) 461 (9.2) 420 (11.7) 2009 6952 (10.5) 2488 (9.1) 1083 (9.5) 628 (10.8) 1577 (12.0) 644 (12.9) 532 (14.8) 2010 5010 (7.5) 1632 (5.9) 617 (5.4) 426 (7.3) 1398 (10.7) 486 (9.7) 451 (12.6) Mental comorbidity Depression 16 988 (25.6) 5578 (20.3) 3080 (27.1) 2067(35.5) 3286 (25.0) 1629 (32.6) 1348 (37.6) Bipolar disorder 10 638 (16.0) 3889 (14.2) 1838 (16.2) 1053 (18.1) 2023 (15.4) 1059 (21.2) 776 (21.6) Anxiety disorder 11 281 (17.0) 3394 (12.4) 2028 (17.8) 1606 (27.6) 2057 (15.7) 1086 (21.7) 1110 (30.9) Substance use disorder 8285 (12.5) 2997 (10.9) 1606 (14.1) 1177 (20.2) 1102 (8.4) 730 (14.6) 673 (18.8) Schizophrenia 5354 (8.1) 3066 (11.2) 809 (7.1) 406 (7.0) 690 (5.3) 228 (4.6) 155 (4.3) Pain diagnosis Any pain condition 1487 (2.2) 296 (1.1) 258 (2.3) 419 (7.2) 147 (1.1) 139 (2.8) 228 (6.4) Musculoskeletal pain 1402 (2.1) 282 (1.0) 241 (2.1) 391 (6.7) 142 (1.1) 135 (2.7) 211 (5.9) Select physical comorbidities Obesity 2519 (3.8) 883 (3.2) 508 (4.5) 353 (6.1) 365 (2.8) 221 (4.4) 189 (5.3) Diabetes 3564 (5.4) 1269 (4.6) 707 (6.2) 603 (10.4) 438 (3.3) 262 (5.2) 285 (7.9) Cardiovascular disease 7683 (11.6) 2494 (9.1) 1335 (11.7) 1390 (23.9) 1080 (8.2) 654 (13.1) 730 (20.3) Chronic obstructive pulmonary disease 3838 (5.8) 1050 (3.8) 840 (7.4) 765 (13.1) 423 (3.2) 331 (6.6) 429 (12.0) Abbreviation: ADHD, attention-deficit/hyperactivity disorder.

2) 603 (10.4) 438 (3.3) 262 (5.2) 285 (7.9) Cardiovascular disease 7683 (11.6) 2494 (9.1) 1335 (11.7) 1390 (23.9) 1080 (8.2) 654 (13.1) 730 (20.3) Chronic obstructive pulmonary disease 3838 (5.8) 1050 (3.8) 840 (7.4) 765 (13.1) 423 (3.2) 331 (6.6) 429 (12.0) Abbreviation: ADHD, attention-deficit/hyperactivity disorder. a Defined as the first 6 months of a randomly selected 12-month observation period of each patient. b Included Hispanic, Asian, Pacific Islander, and Native American individuals. c A patient may qualify for Medicaid for more than 1 reason (ie, both low income and disability). d South includes Florida, Georgia, North Carolina, South Carolina, Virginia, West Virginia, Alabama, Kentucky, Mississippi, Tennessee, Arkansas, Louisiana, and Texas; Midwest includes Illinois, Indiana, Ohio, Wisconsin, Iowa, Kansas, Minnesota, Missouri, and Nebraska; Northeast includes Massachusetts, New Jersey, and New York; and West includes Idaho, New Mexico, California, and Washington.

ia, West Virginia, Alabama, Kentucky, Mississippi, Tennessee, Arkansas, Louisiana, and Texas; Midwest includes Illinois, Indiana, Ohio, Wisconsin, Iowa, Kansas, Minnesota, Missouri, and Nebraska; Northeast includes Massachusetts, New Jersey, and New York; and West includes Idaho, New Mexico, California, and Washington. Statistical Analysis We reported the baseline characteristics of the overall study sample of adults with ADHD and of the groups with or without stimulant use during the 6-month follow-up between 1999 and 2010. None of the reported variables had missing data. Among the adults with ADHD who did or did not use stimulants, we calculated the proportions having (1) no opioid use, (2) concurrent short-term opioid use (1-29 days), and (3) concurrent long-term opioid use (≥30 days, denoted as long-term concurrent use) during the 6-month follow-up. Sensitivity analysis was performed using a shorter overlap of 15 days or more to define concurrent stimulant-opioid use. Among those who were long-term concurrent users, we further analyzed types of stimulant and opioid combinations, with opioids grouped as only short-acting opioids, only long-acting opioids, and both short- and long-acting opioids. Patients who had multiple episodes of concurrent stimulant-opioid use could contribute to more than 1 type of drug combination.

ent users, we further analyzed types of stimulant and opioid combinations, with opioids grouped as only short-acting opioids, only long-acting opioids, and both short- and long-acting opioids. Patients who had multiple episodes of concurrent stimulant-opioid use could contribute to more than 1 type of drug combination. We used multivariable modified Poisson regression models to analyze risk factors (ie, independent variables) associated with long-term concurrent use of prescription stimulants and opioids (ie, dependent variable) among adults with ADHD. The independent variables included age, sex, race/ethnicity, rural residency, region, mental comorbidity (depression, bipolar disorder, anxiety disorder, substance use disorder, or schizophrenia), and physical comorbidity (pain, obesity, diabetes, cardiovascular disease, and COPD). To test secular changes in the prevalence of long-term concurrent stimulant-opioid use, we included each calendar year as a dummy variable in the models. The coefficients of these annual dummy variables represented changes in the prevalence of long-term concurrent use for a given year compared with the reference year 1999. We expressed associations as prevalence relative ratios (PRRs) and reported associated 95% CIs. All analyses were performed using SAS, version 9.4 (SAS Institute Inc). A 2-sided P < .05 was considered statistically significant.

evalence of long-term concurrent use for a given year compared with the reference year 1999. We expressed associations as prevalence relative ratios (PRRs) and reported associated 95% CIs. All analyses were performed using SAS, version 9.4 (SAS Institute Inc). A 2-sided P < .05 was considered statistically significant. Results We identified 66 406 eligible adult patients with at least 1 inpatient or 2 outpatient visits coded for a diagnosis of ADHD between 1999 and 2010. The final study sample and the number of individuals excluded are shown in the study flowchart (Figure). In the final sample, 35 670 individuals (53.7%) were 20 to 30 years of age, 37 155 (56.0%) were female, 52 551 (79.1%) were non-Hispanic white, 47 006 (70.8%) were nonrural residents, and 38 205 (57.5%) qualified for Medicaid based on disability. During the 6-month baseline period, 25.6% of the individuals in the sample received a diagnosis of depression, and 2.2% received a diagnosis of chronic pain conditions. Of the patients who had received a code for a diagnosis of pain, most (1402 of 1487 [94.3%]) had musculoskeletal pain (Table 1). Figure. Flowchart of Patients With Attention-Deficit/Hyperactivity Disorder (ADHD) Included in the Study FFS indicates fee-for-service; ICD-9, International Classification of Diseases, Ninth Revision.

Results We identified 66 406 eligible adult patients with at least 1 inpatient or 2 outpatient visits coded for a diagnosis of ADHD between 1999 and 2010. The final study sample and the number of individuals excluded are shown in the study flowchart (Figure). In the final sample, 35 670 individuals (53.7%) were 20 to 30 years of age, 37 155 (56.0%) were female, 52 551 (79.1%) were non-Hispanic white, 47 006 (70.8%) were nonrural residents, and 38 205 (57.5%) qualified for Medicaid based on disability. During the 6-month baseline period, 25.6% of the individuals in the sample received a diagnosis of depression, and 2.2% received a diagnosis of chronic pain conditions. Of the patients who had received a code for a diagnosis of pain, most (1402 of 1487 [94.3%]) had musculoskeletal pain (Table 1). Figure. Flowchart of Patients With Attention-Deficit/Hyperactivity Disorder (ADHD) Included in the Study FFS indicates fee-for-service; ICD-9, International Classification of Diseases, Ninth Revision. Of the 21 723 adults (32.7%) with ADHD who used stimulants during follow-up, 13 129 (60.4%) received no opioids, 5004 (23.0%) received prescription stimulants and opioids for a short term (1-29 days), and 3590 (16.5%) used these 2 classes of medications concurrently for at least 30 days. Of the 44 683 patients with ADHD who did not use stimulants, 27 476 (61.5%) received no opioids, 11 381 (25.5%) used prescription opioids for a short term, and 5826 (13.0%) used prescription opioids for a long term. Differences were noted between individuals who used opioids and individuals who did not with regard to all characteristics studied among adults with ADHD who used stimulants, as well as among those who did not use stimulants. Overall, long-term opioid use was more common among adults who used stimulants (16.5%) than among those who did not use stimulants (13.0%) (Table 1).

oids and individuals who did not with regard to all characteristics studied among adults with ADHD who used stimulants, as well as among those who did not use stimulants. Overall, long-term opioid use was more common among adults who used stimulants (16.5%) than among those who did not use stimulants (13.0%) (Table 1). Among the 3590 adults with ADHD who used stimulants and opioids concurrently for a long term, 4 of every 5 adults (81.8%) used short-acting opioids, whereas only 1 of every 5 adults (20.6%) used long-acting opioids along with the stimulants. Nearly a quarter of the adults who used stimulants and opioids concurrently for a long term (23.2% [n = 832]) had prescriptions for both short- and long-acting opioids (Table 2).

very 5 adults (81.8%) used short-acting opioids, whereas only 1 of every 5 adults (20.6%) used long-acting opioids along with the stimulants. Nearly a quarter of the adults who used stimulants and opioids concurrently for a long term (23.2% [n = 832]) had prescriptions for both short- and long-acting opioids (Table 2). Table 2. Combinations of Long-term Concurrent Stimulant and Opioid Use Among Adults With Attention-Deficit/Hyperactivity Disorder, 1999-2010 Drug Combinationa Patients, No. (%) (n = 3590) Stimulant and short-acting opioid 2937 (81.8) Stimulant and long-acting opioid 738 (20.6) Stimulant, short-acting opioid, and long-acting opioid 832 (23.2) a A patient may have more than 1 type of drug combination during the 6-month follow-up. Short-acting opioids include hydrocodone, hydromorphone, morphine, oxymorphone, oxycodone, tapentadol, and tramadol in immediate-release forms; codeine and fentanyl in transmucosal and nontransdermal forms; and buprenorphine in nonpatch form. Long-acting opioids include hydromorphone, morphine, oxycodone, oxymorphone, tapentadol, and tramadol extended release, as well as buprenorphine patch, fentanyl transdermal system, levorphanol, and methadone.

lease forms; codeine and fentanyl in transmucosal and nontransdermal forms; and buprenorphine in nonpatch form. Long-acting opioids include hydromorphone, morphine, oxycodone, oxymorphone, tapentadol, and tramadol extended release, as well as buprenorphine patch, fentanyl transdermal system, levorphanol, and methadone. We observed a significant increasing prevalence of long-term concurrent stimulant-opioid use over time (for 2010 vs 1999, adjusted PRR, 1.12; 95% CI, 1.10-1.14) (Table 3). Compared with adults in their 20s, those in their 30s showed a significantly higher prevalence of long-term concurrent stimulant-opioid use (31-40 years vs 20-30 years, adjusted PRR, 1.07; 95% CI, 1.07-1.08). Prevalence further increased as patients advanced in age (41-50 years: adjusted PRR, 1.14; 95% CI, 1.12-1.15; 51-64 years: adjusted PRR, 1.17; 95% CI, 1.16-1.19). As expected, having pain was significantly associated with long-term concurrent stimulant-opioid use among adults with ADHD (vs no pain condition, adjusted PRR, 1.10; 95% CI, 1.07-1.13). Being of non-Hispanic white race/ethnicity (black PRR, 0.93; 95% CI, 0.92-0.93; other PRR, 0.97; 95% CI, 0.97-0.98; vs white), living in the South (Midwest PRR, 0.98; 95% CI, 0.97-0.98; Northeast PRR, 0.94; 95% CI, 0.93-0.94; West PRR, 0.95; 95% CI, 0.94-0.96; vs South), and having depression (PRR, 1.02; 95% CI, 1.01-1.03), anxiety disorder (PRR, 1.05; 95% CI, 1.04-1.07), substance use disorder (PRR, 1.04; 95% CI, 1.03-1.05), COPD (PRR, 1.05; 95% CI, 1.04-1.07), or cardiovascular disease (PRR, 1.02; 95% CI, 1.01-1.03) were also factors significantly associated with long-term concurrent stimulant-opioid use, whereas receiving a diagnosis of schizophrenia appeared to be associated with protection from long-term concurrent stimulant-opioid use (adjusted PRR, 0.95; 95% CI, 0.94-0.95) (Table 3). Similar associations were found in the sensitivity analysis in which long-term concurrent stimulant-opioid use was defined as concurrent use for 15 days or more. Using the shorter cutoff to define long-term concurrent use, we found that 5541 patients with ADHD (8.3%) used stimulants and opioids concurrently (eTable 2 in the Supplement).

ns were found in the sensitivity analysis in which long-term concurrent stimulant-opioid use was defined as concurrent use for 15 days or more. Using the shorter cutoff to define long-term concurrent use, we found that 5541 patients with ADHD (8.3%) used stimulants and opioids concurrently (eTable 2 in the Supplement). Table 3. Multivariable Modified Poisson Regression Analyses of Factors Associated With Long-term Concurrent Stimulant and Opioid Use Among Medicaid-Enrolled Adults With Attention-Deficit/Hyperactivity Disorder, 1999-2010 Variable Long-term Concurrent Stimulant and Opioid Use (Yes vs No), PRR (95% CI) Unadjusted Adjusteda Year 1999 1 [Reference] 1 [Reference] 2000 1.01 (0.99-1.02) 1.01 (0.99-1.02) 2001 1.01 (0.99-1.02) 1.00 (0.98-1.01) 2002 1.00 (0.99-1.02) 1.00 (0.98-1.01) 2003 1.03 (1.01-1.04) 1.02 (1.00-1.03) 2004 1.06 (1.04-1.07) 1.03 (1.02-1.05) 2005 1.08 (1.06-1.09) 1.06 (1.04-1.07) 2006 1.07 (1.06-1.09) 1.06 (1.05-1.08) 2007 1.08 (1.07-1.10) 1.08 (1.06-1.10) 2008 1.09 (1.08-1.11) 1.09 (1.07-1.11) 2009 1.11 (1.09-1.12) 1.10 (1.08-1.11) 2010 1.13 (1.11-1.15) 1.12 (1.10-1.14) Age, y 20-30 1 [Reference] 1 [Reference] 31-40 1.09 (1.08-1.10) 1.07 (1.07-1.08) 41-50 1.15 (1.14-1.17) 1.14 (1.12-1.15) 51-64 1.20 (1.18-1.22) 1.17 (1.16-1.19) Male (vs female as reference) 0.95 (0.95-0.96) 0.98 (0.98-0.99) Race/ethnicity Non-Hispanic white 1 [Reference] 1 [Reference] Non-Hispanic black 0.91 (0.90-0.92) 0.93 (0.92-0.93) Other 0.95 (0.94-0.96) 0.97 (0.97-0.98) Rural residency (yes vs no) 1.01 (1.00-1.01) 1.00 (0.99-1.01) US Region South 1 [Reference] 1 [Reference] Midwest 1.00 (0.99-1.00) 0.98 (0.97-0.98) Northeast 0.93 (0.93-0.94) 0.94 (0.93-0.94) West 0.97 (0.96-0.98) 0.95 (0.94-0.96) Mental comorbidity Depression (yes vs no) 1.06 (1.05-1.07) 1.02 (1.01-1.03) Bipolar didorder (yes vs no) 1.04 (1.03-1.05) 1.01 (1.00-1.02) Anxiety disorder (yes vs no) 1.10 (1.09-1.11) 1.05 (1.04-1.07) Substance use disorder (yes vs no) 1.06 (1.05-1.07) 1.04 (1.03-1.05) Schizophrenia (yes vs no) 0.95 (0.94-0.96) 0.95 (0.94-0.95) Physical comorbidity Pain (yes vs no) 1.18 (1.15-1.22) 1.10 (1.07-1.13) Obesity (yes vs no) 1.04 (1.02-1.06) 1.01 (0.99-1.03) Diabetes (yes vs no) 1.05 (1.03-1.07) 0.99 (0.97-1.00) Cardiovascular disease (yes vs no) 1.08 (1.07-1.10) 1.02 (1.01-1.03) COPD (yes vs no) 1.11 (1.09-1.13) 1.05 (1.04-1.07) Abbreviations: COPD, chronic obstructive pulmonary disease; PRR, prevalence relative ratio.

Obesity (yes vs no) 1.04 (1.02-1.06) 1.01 (0.99-1.03) Diabetes (yes vs no) 1.05 (1.03-1.07) 0.99 (0.97-1.00) Cardiovascular disease (yes vs no) 1.08 (1.07-1.10) 1.02 (1.01-1.03) COPD (yes vs no) 1.11 (1.09-1.13) 1.05 (1.04-1.07) Abbreviations: COPD, chronic obstructive pulmonary disease; PRR, prevalence relative ratio. a The model was simultaneously adjusted for the covariates listed. Discussion To our knowledge, the present study is among the first to provide population-based data on the long-term concurrent use of stimulants and opioids among adults with ADHD using Medicaid MAX files from 29 states. Our results comprising drug use data from the last decade indicated substantial and increasing long-term use of these 2 types of controlled prescription medications. Overall, 5% of adults with ADHD had concurrently used both prescription stimulants and opioids for at least 30 days during the 6-month follow-up period. The proportion was even higher among adults with ADHD who used stimulants, with 16.5% of these adults using both types of medications concurrently. The probability of long-term stimulant-opioid use compared with non–long-term use (ie, use of stimulants without opioids or short-term concurrent use of stimulants with opioids) increased by 12% between 1999 and 2010. Our findings suggest that long-term concurrent use of stimulants and opioids has become an increasingly common practice among adult patients with ADHD.

compared with non–long-term use (ie, use of stimulants without opioids or short-term concurrent use of stimulants with opioids) increased by 12% between 1999 and 2010. Our findings suggest that long-term concurrent use of stimulants and opioids has become an increasingly common practice among adult patients with ADHD. Our data also suggest that an increasing number of adults with ADHD may have developed moderate to severe chronic pain that required aggressive pain management. However, we observed only 2.2% of adults with ADHD who had received a diagnosis of any pain condition during the 6-month baseline, with the percentage higher among those with long-term opioid use (6.4% of adults who used stimulants and 7.2% of those who did not use stimulants). Our estimates are lower than self-reported data (from 18.8% to 76.7%) in surveys involving adults with ADHD,4,28,29 and the discrepancies may be due to differences in pain measurement, population, and time frame studied. The mechanism underlying ADHD symptoms and muscle pain is not entirely clear, but hypotheses have been proposed. Because of attention-deficit issues and increased risk-taking behavior, patients with ADHD are more prone to accidents causing physical injuries30 and, thereby, may be more likely to develop musculoskeletal chronic pain. Alternatively, some researchers have postulated that muscle pain is the long-term consequence of ADHD-associated motor problems (eg, difficulty in balancing, trouble with reciprocal coordination, and poor movement reflection and resistance).31 Given that the hypothesis involving increased pain prevalence is currently supported only by self-report survey data and small-scale clinical findings,4,28,29,31 formal observational studies are needed to further examine the association between ADHD and pain. An alternative hypothesis is inherent in reported associations between stimulant use and increased risk for substance use disorder,19 which may in turn result in increased prescription opioid use.

nical findings,4,28,29,31 formal observational studies are needed to further examine the association between ADHD and pain. An alternative hypothesis is inherent in reported associations between stimulant use and increased risk for substance use disorder,19 which may in turn result in increased prescription opioid use. Although the concurrent use of stimulants and opioids may initially have been prompted by ADHD symptoms and comorbid chronic pain, continued use of opioids alone or combined with central nervous system stimulants may result in drug dependence and other adverse effects (eg, overdose) because of the high potential for abuse and misuse.9,19,32,33 Both opioids and stimulants are prescription drugs commonly misused by adults.34,35,36 National data have shown an increase in the use of these medications, and the increasing use has contributed to an increasing number of emergency department visits and overdose deaths associated with both drug classes.9,19,32 Of particular concern is our finding that a high proportion (81.8%) of adults who use both drugs concurrently for a long term are prescribed stimulants and short-acting opioids. Short-acting opioids are recommended for acute pain, but their long-term use has been discouraged by clinical guidelines because of the risk of abuse, opioid tolerance, and dose escalation.37,38,39 The common long-term use of stimulants and opioids, especially short-acting agents, observed among adults with ADHD deserves further investigation to understand the association of the use of this drug combination with patient health outcomes.

because of the risk of abuse, opioid tolerance, and dose escalation.37,38,39 The common long-term use of stimulants and opioids, especially short-acting agents, observed among adults with ADHD deserves further investigation to understand the association of the use of this drug combination with patient health outcomes. Our study also showed several important social and clinical factors associated with long-term concurrent stimulant-opioid use among adults with ADHD. We noted a significant increase in stimulant-opioid use with increasing age, even after adjusting for aging-related chronic health conditions. Our finding is consistent with a recent study of any prescription opioid use, with a reported prevalence of 8.1% among adults older than 40 years compared with a prevalence of 4.7% among adults aged 20 to 39 years.40 These prevalence estimations from the general adult population, which did not require a minimum 30 days’ supply, were much lower than those of long-term opioid use observed among adults with ADHD (16.5% of those who used stimulants and 13.0% of those who did not use stimulants) in the present study. Furthermore, our findings of higher long-term concurrent stimulant-opioid use among non-Hispanic white adults compared with non-Hispanic black adults and among residents from the South compared with all other US regions are consistent with previous studies on prescription opioid use.40

stimulants) in the present study. Furthermore, our findings of higher long-term concurrent stimulant-opioid use among non-Hispanic white adults compared with non-Hispanic black adults and among residents from the South compared with all other US regions are consistent with previous studies on prescription opioid use.40 We also observed that adults with ADHD and comorbid depression, anxiety disorder, substance use disorder, cardiovascular disease, or COPD tended to use stimulants and opioids concurrently. Depression, anxiety disorder, and substance abuse disorder are strongly associated with long-term chronic opioid use,41,42,43,44 which may explain the increase in long-term stimulant-opioid use among adults with ADHD who have these comorbid conditions. Patients with cardiovascular disease (vs without) are more likely to experience pain,45 and patients with COPD symptoms are at high risk for pain and for nicotine dependence, the latter of which is an independent risk factor for substance use.46 Our study contributes to the understanding of the potential risk factors associated with long-term concurrent stimulant-opioid use among adults with ADHD. Identifying these high-risk patients allows for early intervention and may reduce the number of adverse events associated with the long-term use of these medications.

ur study contributes to the understanding of the potential risk factors associated with long-term concurrent stimulant-opioid use among adults with ADHD. Identifying these high-risk patients allows for early intervention and may reduce the number of adverse events associated with the long-term use of these medications. Strengths and Limitations Several strengths of this study are noteworthy. Our study provided important information on the treatment of adults with ADHD, a population that is increasing but has received limited research attention. Using more than a decade of administrative data that include detailed prescription information from 29 states allowed us to examine secular changes in concurrent stimulant-opioid use. We also explored a broad range of potentially associated factors, such as physical comorbidities, which have rarely been considered in the literature for stimulant-opioid polypharmacy.

that include detailed prescription information from 29 states allowed us to examine secular changes in concurrent stimulant-opioid use. We also explored a broad range of potentially associated factors, such as physical comorbidities, which have rarely been considered in the literature for stimulant-opioid polypharmacy. The present study shares limitations common to observational studies based on administrative data. First, MAX data captured only prescription medications filled and reimbursed by Medicaid, without information on medications obtained through other channels, such as paying for prescription fills with cash or obtaining medications through illicit sources. Considering that opioid prescription fills are commonly paid out of pocket,47 our reported prevalence of concurrent stimulant-opioid use may be too low. In addition, the data provided no information on whether prescribed opioids were intended for regular use or as needed. Using strict definitions of overlapping treatment periods that were defined based on the dispensed days’ supply (assuming daily use) might have underestimated long-term concurrent use. Furthermore, our data did not allow us to determine the intended clinical indications for each prescribed medication, which limited our ability to assess the appropriateness of the observed concurrent use of stimulants and opioids. We identified the study sample of patients with ADHD based on the disease diagnosis codes for inpatient or outpatient encounters, which may have excluded patients with ADHD who did not seek medical care. Our findings from the 1999 to 2010 MAX data may not reflect recent changes in clinical opioid prescribing practice because several initiatives were implemented after 2010 to educate physicians to reduce unsafe opioid prescribing. Our study results are only generalizable to the Medicaid fee-for-service plans of 29 states. We selected these states based on absolute Medicaid enrollment numbers, resulting in the capture of more than 80% of all Medicaid beneficiaries. However, restrictions to fee-for-service plans resulted in varying representation of individual states and study years (because of growing managed care penetration over time). Future studies that extend to other states’ populations of adults with ADHD in more recent time periods will contribute to our understanding of the differences in rates and patterns of concurrent stimulant-opioid use as well as to the association of concurrent use with adverse outcomes.

ged care penetration over time). Future studies that extend to other states’ populations of adults with ADHD in more recent time periods will contribute to our understanding of the differences in rates and patterns of concurrent stimulant-opioid use as well as to the association of concurrent use with adverse outcomes. Conclusions Among Medicaid-enrolled adults with ADHD, long-term concurrent use of stimulants and opioids was common and increased over time. Prevalent long-term stimulant-opioid use was associated with older age, non-Hispanic white race/ethnicity, southern US residency, and a diagnosis of substance abuse disorder, depression, anxiety disorder, chronic pain, COPD, or cardiovascular disease. Clinical and research priorities should be made toward understanding the benefits and risks of the long-term coadministration of stimulants and opioids in the management of ADHD and co-occurring pain conditions. Supplement. eTable 1. ICD-9 Codes for Studied Physical and Mental Comorbidities eTable 2. Multivariable Modified Poisson Regression Analyses of Factors Associated With Concurrent Stimulant-Opioid Use for ≥15 Days in Medicaid Adult Patients With ADHD, 29 States, 1999-2010 Click here for additional data file.

CI, −29.0% to 0.5%; P = .09). Significantly, reduced distant metastasis was observed in the HPV-positive group (3 of 37 [8.1%] vs 29 of 105 [27.6%]; difference, −19.5%; 95% CI, −31.8% to −7.2%; P = .02) as were deaths from EAC (5 of 37 [13.5%] vs 38 of 105 [36.2%]; difference, −22.7%; 95% CI, −37.0% to −8.3%; P = .01). Table 2. Comparison of Survival, Disease Relapse and Progression, and Site of Failure in HPV-Positive and HPV-Negative Patients Characteristic No. (%) P Valuea Patients With HPV-Positive HGD or EAC (n = 37) Patients With HPV-Negative HGD or EAC (n = 105) Disease-free survival, mean (SD), mo 40.3 (33.8) 24.1 (25.5) .003 Overall survival, mean, mo 43.7 (32.9) 29.8 (25.3) .009 Survival status (alive at last follow-up) 26 (70.3) 58 (55.2) .11 Recurrence or progression 9 (24.3) 61 (58.1) <.001 Recurrence 6 (16.2) 46 (43.8) .003 Local-regional failure 6 (16.2) 32 (30.5) .09 Distant metastases 3 (8.1) 29 (27.6) .02 Death due to EAC 5 (13.5) 38 (36.2) .01 Abbreviations: EAC, esophageal adenocarcinoma; HGD, high-grade dysplasia; HPV, human papillomavirus. a Differences between HPV-positive vs HPV-negative cases in regard to characteristics were assessed using 2-sample t test for all numerical data and χ2 analysis for binary measurements.

Introduction Esophageal adenocarcinoma (EAC) is one of the fastest-growing and deadliest cancers in the Western world,1,2 although recently the rate of increase has diminished and possibly plateaued in the United States and Sweden.3,4 Currently, Barrett esophagus (BE) is the only recognized visible precursor lesion for EAC. Intriguingly, these high rates of EAC have occurred against a backdrop of progressive reduction in the risk estimate of malignancy associated with BE.5 In this regard, our discovery of a strong association of transcriptionally active high-risk human papillomavirus (HPV) with a subset of Barrett dysplasia (BD) and EAC5 may be relevant.

ese high rates of EAC have occurred against a backdrop of progressive reduction in the risk estimate of malignancy associated with BE.5 In this regard, our discovery of a strong association of transcriptionally active high-risk human papillomavirus (HPV) with a subset of Barrett dysplasia (BD) and EAC5 may be relevant. Increasing high-risk HPV viral load and integration status has been linked with more severe disease along the Barrett metaplasia-dysplasia-adenocarcinoma sequence.6 Moreover, treatment failure after endoscopic ablation of BD or EAC is predicted by persistent high-risk HPV infection and overexpression of the p53 gene (TP53).7 We also discovered that HPV-positive EAC is molecularly distinct from HPV-negative EAC, indicating different biological mechanisms of tumor genesis.8 Hybrid sequences containing HPV-16 and the human genome have been identified, indicating a host-viral interaction.8 Aberrations of the retinoblastoma protein (pRb) pathway, ie, upregulation of p16INK4A and downregulation of pRb, as well as wild-type TP53 are hallmarks of active HPV involvement in BD and EAC. Additionally, the transformation-specific combination of high p16 expression and low pRb expression, which is a feature of HPV-driven lesions, has been identified in a significant proportion of virally positive BD and EAC.9

ulation of pRb, as well as wild-type TP53 are hallmarks of active HPV involvement in BD and EAC. Additionally, the transformation-specific combination of high p16 expression and low pRb expression, which is a feature of HPV-driven lesions, has been identified in a significant proportion of virally positive BD and EAC.9 Ability to detect HPV in earlier negative studies and more recent investigations may have been hampered by poor tissue classification and suboptimal testing methods. This is further exacerbated by the low HPV viral load in esophageal tissue. Some of these misclassified negative studies only used nondysplastic BE. Metaplastic tissue is not associated with HPV. Racial or geographic variations can also account for this anomaly.10,11,12,13,14 Interestingly, a systematic review has reported HPV prevalence rates of 35% in 174 patients with EAC, similar to our findings.15 Another systematic review that included 19 studies concluded that the pooled prevalence of HPV in EAC was 13%, suggesting the low prevalence rate may have been caused by small sample sizes and compromised detection methods.16

has reported HPV prevalence rates of 35% in 174 patients with EAC, similar to our findings.15 Another systematic review that included 19 studies concluded that the pooled prevalence of HPV in EAC was 13%, suggesting the low prevalence rate may have been caused by small sample sizes and compromised detection methods.16 It is well documented that patients with HPV-positive head and neck squamous cell carcinoma (HNSCC) have an improved rate of overall survival (OS) (hazard ratio [HR], 0.7; 95% CI, 0.5-1.0) and a reduced risk of recurrence (HR, 0.5; 95% CI, 0.4-0.7) compared with viral-negative tumors.17 A meta-analysis reported a 74% improved disease-free survival (DFS) and 53% better OS in HPV-associated HNSCC vs HPV-negative HNSCC.18 We therefore hypothesized that HPV-associated EAC and high-grade dysplasia (HGD) would show a similar favorable prognosis compared with HPV-negative esophageal lesions. Methods

It is well documented that patients with HPV-positive head and neck squamous cell carcinoma (HNSCC) have an improved rate of overall survival (OS) (hazard ratio [HR], 0.7; 95% CI, 0.5-1.0) and a reduced risk of recurrence (HR, 0.5; 95% CI, 0.4-0.7) compared with viral-negative tumors.17 A meta-analysis reported a 74% improved disease-free survival (DFS) and 53% better OS in HPV-associated HNSCC vs HPV-negative HNSCC.18 We therefore hypothesized that HPV-associated EAC and high-grade dysplasia (HGD) would show a similar favorable prognosis compared with HPV-negative esophageal lesions. Methods Study Design and Population This retrospective case-control study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline. Eligible patients were those with HGD or EAC (Siewert classification type I or II) deemed suitable for treatment, ie, for endotherapy (endoscopic mucosal resection [EMR] and/or radiofrequency ablation [RFA]) or esophagectomy with or without neoadjuvant chemotherapy and/or radiotherapy. The study period was from December 1, 2002, to November 28, 2017, and patients were enrolled from a tertiary referral center, Bankstown-Lidcombe Hospital, Sydney, Australia (n = 139), and a regional health care center, Launceston General Hospital, Launceston, Tasmania, Australia (n = 3). Demographic and clinical data were obtained from a prospectively maintained database. Inclusion and exclusion criteria have been previously documented.9 Pretreatment tissue was prospectively collected in 94 patients (both fresh frozen and formalin-fixed paraffin-embedded [FFPE]) and retrospectively retrieved (FFPE only) from the remaining 48 patients. Oral and written informed consent were obtained from participants prior to the investigation. This study was approved by the Human Research Ethics Committee, Tasmania, and South Western Sydney Local Health Network.

fixed paraffin-embedded [FFPE]) and retrospectively retrieved (FFPE only) from the remaining 48 patients. Oral and written informed consent were obtained from participants prior to the investigation. This study was approved by the Human Research Ethics Committee, Tasmania, and South Western Sydney Local Health Network. Endotherapy and/or Esophagectomy Staging endoscopic ultrasound examination was performed on all patients. Patients with nodular or ulcerating lesions without lymph node involvement underwent EMR. Apart from endoscopic ultrasound, positron emission tomography, computed tomography, and laparoscopy were performed as staging investigations in patients considered candidates for esophagectomy. Circumferential and focal RFA were used to ablate flat lesions to achieve complete eradication. Patients were scheduled to return for 3-, 6-, 9-, and 12-month follow-up endoscopy and biopsy. Residual lesions were ablated and nodules subjected to EMR. Patients with adenocarcinoma invading beyond the muscularis mucosa into the submucosa (T1b lesions) were excluded from endotherapy and underwent esophagectomy with D2 lymph node dissection. Siewert type I tumors were subjected to Ivor Lewis esophagogastrectomy and Siewert type II cancers to either radical total gastrectomy including lower esophagus or Ivor Lewis esophagectomy depending on the patient’s build.

esions) were excluded from endotherapy and underwent esophagectomy with D2 lymph node dissection. Siewert type I tumors were subjected to Ivor Lewis esophagogastrectomy and Siewert type II cancers to either radical total gastrectomy including lower esophagus or Ivor Lewis esophagectomy depending on the patient’s build. Staging was performed according to the seventh edition of the Cancer Staging Manual by the American Joint Committee on Cancer.19 Patients with locally advanced disease (T3 or T4 N0 or any T stage with N1) were potential candidates for neoadjuvant or adjuvant treatment. After treatment, repeat staging was done to exclude metastatic disease prior to proceeding to esophagectomy. As such, 14 patients did not undergo esophagectomy due to tumor-related factors or patient refusal. Nevertheless, they were included in the analysis as they had undergone radiotherapy and/or chemotherapy.

tment. After treatment, repeat staging was done to exclude metastatic disease prior to proceeding to esophagectomy. As such, 14 patients did not undergo esophagectomy due to tumor-related factors or patient refusal. Nevertheless, they were included in the analysis as they had undergone radiotherapy and/or chemotherapy. Laboratory Studies Detection of HPV in genomic DNA extracted from fresh frozen or formalin-fixed biopsy tissue was performed by nested polymerase chain reaction (PCR) amplification of a conserved viral L1 gene using MY09 and MY11 and GP5+ and GP6+ primers for both high-risk and low-risk HPV as previously published.5 To minimize contamination, separate rooms were used for reaction preparation, template handling, performing nested reactions, and post-PCR analysis. Routine decontamination by UV irradiation was performed in the DNA-free PCR hood before each run. To guard against systematic contamination of PCR reagent, appropriate positive (HPV-16–positive cervical cancer) and negative (deionized water and PCR master mix without template) controls were included in each step of the PCR process. Genotypes of HPV were determined by sequencing.5 Real-time PCR assays measuring HPV E6 and E7 copy numbers were performed to determine viral load using genotype-specific HPV-16 and HPV-18 primers.6

and negative (deionized water and PCR master mix without template) controls were included in each step of the PCR process. Genotypes of HPV were determined by sequencing.5 Real-time PCR assays measuring HPV E6 and E7 copy numbers were performed to determine viral load using genotype-specific HPV-16 and HPV-18 primers.6 In situ hybridization (ISH) of RNA for high-risk HPV-16 and HPV-18 E6 and E7 messenger RNA (mRNA) was performed manually using the RNAscope 2.5 High-Definition Assay (Advanced Cell Diagnostics Inc) with a cocktail of probes targeting 18 high-risk HPV types (HPV types 16, 18, 26, 31, 33, 35, 39, 45, 51, 52, 53, 56, 58, 59, 66, 68, 73, and 82) according to the manufacturer’s instructions.9 Negative and positive controls were probes recognizing the bacterial gene dapB and the endogenous ubiquitin C mRNA, respectively. In addition, cervical cancer tissue, HNSCC samples, BE and BD, EAC, and esophageal squamous cell carcinoma (ESCC), which all had detectable transcriptionally active HPV (DNA positive by PCR and the presence of ≥1 of 2 markers of biological activity, ie, E6 and E7 mRNA and/or p16INK4A), served as positive controls. We used HNSCC, BE and BD, EAC, and ESCC devoid of virus as negative controls. Positivity was defined as the presence of punctuate cytoplasmic and/or nuclear staining that exceeded the dapB (negative control) signal. Expression of the p16INK4A and p53 proteins was assessed by immunohistochemistry (IHC) on FFPE tissue using EnVision FLEX Mini Kits and CINtec Histology Kits (monoclonal mouse anti–human p16INK4A antibody, clone E6H4) (mtm laboratories), respectively, using appropriate positive and negative controls.9 All IHC scoring of slides was independently performed by 2 experienced gastrointestinal pathologists blinded to the virological status and the clinical outcome of patients. For p16INK4A, at least moderate staining of both nucleus and cytoplasm in more than 25% of esophageal dysplastic or tumor tissue was considered as p16 overexpression. Staining of 25% or less was deemed low expression.

gastrointestinal pathologists blinded to the virological status and the clinical outcome of patients. For p16INK4A, at least moderate staining of both nucleus and cytoplasm in more than 25% of esophageal dysplastic or tumor tissue was considered as p16 overexpression. Staining of 25% or less was deemed low expression. Using this criterion, we have demonstrated that high p16 expression is associated with HPV-associated BD and EAC with reasonable sensitivity and specificity as others have in HNSCC.9,20 In the case of p53, intense nuclear staining of more than 10% of esophageal cells was considered overexpression.7,9 Mutations of TP53 were confirmed by sequencing of the gene using the semiconductor-based Ion Torrent sequencing platform (ThermoFisher) according to the manufacturer’s instructions as previously published.9 Study End Points and Statistical Analysis The primary end points were DFS from the time of diagnosis to the date of first local, regional, or distant failure and OS, defined as time between diagnosis to the date of death or last follow-up.

Using this criterion, we have demonstrated that high p16 expression is associated with HPV-associated BD and EAC with reasonable sensitivity and specificity as others have in HNSCC.9,20 In the case of p53, intense nuclear staining of more than 10% of esophageal cells was considered overexpression.7,9 Mutations of TP53 were confirmed by sequencing of the gene using the semiconductor-based Ion Torrent sequencing platform (ThermoFisher) according to the manufacturer’s instructions as previously published.9 Study End Points and Statistical Analysis The primary end points were DFS from the time of diagnosis to the date of first local, regional, or distant failure and OS, defined as time between diagnosis to the date of death or last follow-up. Differences between HPV-positive and HPV-negative cases in regard to baseline characteristics were assessed using the 2-sample t test for comparing the mean values between the 2 groups in regard to all numerical data. A χ2 analysis was used for evaluating the association between the binary measurements in the viral-positive and viral-negative groups. Survival analysis was conducted using the Kaplan-Meier method to estimate the DFS and OS of the 5 HPV variables, ie, viral DNA status, transcriptionally active HPV, E6 and E7 mRNA, p16, and p53. The log-rank test was used to analyze the association between the HPV variables with DFS and OS. Cox proportional hazards regression models were used to estimate the importance of these biomarkers for DFS and OS after adjusting for age, sex, body mass index (calculated as weight in kilograms divided by height in meters squared), history of smoking, excess alcohol use, medications (proton-pump inhibitors, nonsteroidal anti-inflammatory drugs, or statins), T stage, N stage, surgical or endoscopic mucosal resection margin status, degree of tumor differentiation, and neoadjuvant or adjuvant treatment (model 2). Because viral status itself was associated with disease severity, adjusting for stage and treatment (chemotherapy and/or radiotherapy) may have prevented identification of prognostic effects derived exclusively from the 5 HPV variables. Therefore, additional Cox regression models were applied with all of the above covariates but excluding T and N stages as well as chemotherapy and radiotherapy (model 3). All statistical tests were performed using SAS statistical software version 9.4 (SAS Institute Inc), and the level of significance was set at .05 (2-sided).

re, additional Cox regression models were applied with all of the above covariates but excluding T and N stages as well as chemotherapy and radiotherapy (model 3). All statistical tests were performed using SAS statistical software version 9.4 (SAS Institute Inc), and the level of significance was set at .05 (2-sided). Results Patient Characteristics A total of 142 patients were tested for HPV status, p16INK4A IHC, and E6 and E7 mRNA ISH (eFigure 1 in the Supplement). Of 142 patients (126 [88.7%] male; mean [SD] age, 66.0 [12.1] years; 142 [100%] white), 37 were HPV positive and 105 were HPV negative. Mean (SD) follow-up time was 33.4 (28.0) months (range, 2-159 months) for the whole study population and 43.8 (29.4) months (range, 3-159 months) for survivors.

e 1 in the Supplement). Of 142 patients (126 [88.7%] male; mean [SD] age, 66.0 [12.1] years; 142 [100%] white), 37 were HPV positive and 105 were HPV negative. Mean (SD) follow-up time was 33.4 (28.0) months (range, 2-159 months) for the whole study population and 43.8 (29.4) months (range, 3-159 months) for survivors. HPV DNA, E6 and E7 mRNA, p16INK4A, and p53 Status Polymerase chain reaction analysis of HPV DNA was performed on all 142 patients with esophageal lesions (38 HGD and 104 EAC). Thirty-seven patients (11 with HGD and 26 with EAC) had positive results for HPV DNA; 33 had HPV-16, 1 had HPV-18, and the remaining 3 had low-risk types 6 and 11. No multiple genotypes were detected in the same patient. Amplifiable β-globin gene was present in all specimens, and median (range) viral load was 0.1 copy per 10 cell genomic DNA (0-1.12 copies per 10 cell genome). All specimens with measurable viral load revealed coherence between genotypes found on MY09 and MY11 and GP5+ and GP6+ PCR and those found in E6 and E7 analysis. In the 142 HGDs and EACs assessed, 33 of 34 high-risk HPV (types 16 and 18) (97.1%) had detectable viral load. Among the 37 DNA-positive lesions, 18 (48.7%) had E6 and E7 mRNA detection by ISH and 19 (51.4%) overexpressed p16INK4A. Eleven samples (29.7%) were positive for all 3 markers.

ose found in E6 and E7 analysis. In the 142 HGDs and EACs assessed, 33 of 34 high-risk HPV (types 16 and 18) (97.1%) had detectable viral load. Among the 37 DNA-positive lesions, 18 (48.7%) had E6 and E7 mRNA detection by ISH and 19 (51.4%) overexpressed p16INK4A. Eleven samples (29.7%) were positive for all 3 markers. Demographic, clinical, and pathological data are compared between HPV-positive and HPV-negative individuals with esophageal lesions in Table 1. Overexpression of p16INK4A and E6 and E7 mRNA presence were significantly greater in the HPV-positive group compared with the viral-negative cohort. Of note, HPV-positive patients had more early-stage (Tis, T1, and T2) esophageal lesions compared with viral-negative patients (75.7% vs 54.3%; difference, 21.4%; 95% CI, 4.6%-38.2%; P = .02), but nodal stage was similar between the groups. Most were stage N0 or N1 in both the HPV-positive (89.2%) and viral-negative (89.5%) patients with HGD and EAC. No significant differences were detected in any of the other clinical or pathological baseline characteristics. In the transcriptionally active HPV-positive group (HPV positive with p16INK4A, HPV positive with E6 and E7 mRNA, or HPV positive with both p16INK4A and E6 and E7 mRNA), the patients were significantly younger than the biologically inactive virus group (HPV positive without p16INK4A and/or E6 and E7 mRNA) (61.4 vs 67.2 years; difference, −5.8 years; 95% CI, −10.6 to −1.0 years; P = .02) (eTable in the Supplement).

E6 and E7 mRNA, or HPV positive with both p16INK4A and E6 and E7 mRNA), the patients were significantly younger than the biologically inactive virus group (HPV positive without p16INK4A and/or E6 and E7 mRNA) (61.4 vs 67.2 years; difference, −5.8 years; 95% CI, −10.6 to −1.0 years; P = .02) (eTable in the Supplement). Table 1. Demographic and Clinical Characteristics of the Study Population and Associated Tumors According to Patient Group Characteristic No. (%)a P Valueb Patients With HPV-Positive HGD or EAC (n = 37) Patients With HPV-Negative HGD or EAC (n = 105) Sex Male 33 (89.2) 93 (88.6) .92 Female 4 (10.8) 12 (11.4) Age, mean (SD) (range), y 65.2 (12.4) (33.0-89.0) 66.2 (12.0) (32.0-90.0) .65 Body mass index, mean (SD)c 27.7 (5.7) 27.0 (5.1) .53 Ever smoked 24 (64.9) 77 (73.3) .33 Smoked >10 pack-yearsd 22/24 (91.7) 69/76 (90.8) .90 Smoked >20 pack-yearsd 15/24 (62.5) 48/76 (63.2) .95 Alcohol intake 26/36 (72.2) 75 (71.4) .93 Excess alcohole 8/35 (22.9) 18 (17.1) .45 Proton-pump inhibitors 28 (75.7) 75 (71.4) .62 Nonsteroidal anti-inflammatory drugs 11 (29.7) 23 (21.9) .34 Statins 11 (29.7) 31 (29.5) .98 Esophagitis 4/36 (11.1) 14/104 (13.5) .72 Hiatal hernia 21/36 (58.3) 49/101 (48.5) .31 Resection margin R0 26/29 (89.7) 61/75 (81.3) .30 R1 or R2 3/29 (10.3) 14/75 (18.7) Histology, degree of differentiation Well 3 (8.1) 5 (4.8) .22 Moderate 14 (37.8) 31 (29.5) Poor 5 (13.5) 32 (30.5) Not documented 15 (40.5) 37 (35.2) Treatment endotherapy, EMR and RFA 17 (46.0) 41 (39.1) .46 Esophagectomy 21 (56.8) 54 (51.4) .58 Pathologic T stagef Tis, T1, or T2 28 (75.7) 57 (54.3) .02 T3 or T4 9 (24.3) 48 (45.7) Pathologic N stagef N0-N1 33 (89.2) 94 (89.5) .95 N2-N3 4 (10.8) 11 (10.5) Pathologic M stagef M0 36 (97.3) 99 (94.3) .47 M1 1 (2.7) 6 (5.7) Radiotherapy 8 (21.6) 28 (26.7) .54 Chemotherapy 13 (35.1) 39 (37.1) .83 p16INK4a overexpression 19 (51.4) 34 (32.4) .04 E6 and E7 mRNA positivity 18 (48.7) 6 (5.7) <.001 Low p53 expression 23 (62.2) 40 (38.1) .01 Abbreviations: EAC, esophageal adenocarcinoma; EMR, endoscopic mucosal resection; HGD, high-grade dysplasia; HPV, human papillomavirus; mRNA, messenger RNA; RFA, radiofrequency ablation.

Table 2. Comparison of Survival, Disease Relapse and Progression, and Site of Failure in HPV-Positive and HPV-Negative Patients Characteristic No. (%) P Valuea Patients With HPV-Positive HGD or EAC (n = 37) Patients With HPV-Negative HGD or EAC (n = 105) Disease-free survival, mean (SD), mo 40.3 (33.8) 24.1 (25.5) .003 Overall survival, mean, mo 43.7 (32.9) 29.8 (25.3) .009 Survival status (alive at last follow-up) 26 (70.3) 58 (55.2) .11 Recurrence or progression 9 (24.3) 61 (58.1) <.001 Recurrence 6 (16.2) 46 (43.8) .003 Local-regional failure 6 (16.2) 32 (30.5) .09 Distant metastases 3 (8.1) 29 (27.6) .02 Death due to EAC 5 (13.5) 38 (36.2) .01 Abbreviations: EAC, esophageal adenocarcinoma; HGD, high-grade dysplasia; HPV, human papillomavirus. a Differences between HPV-positive vs HPV-negative cases in regard to characteristics were assessed using 2-sample t test for all numerical data and χ2 analysis for binary measurements. Kaplan-Meier graphs for DFS and OS of patients categorized by HPV, transcriptionally active virus, E6 and E7 mRNA, and high expression of p16 and low expression of p53 are shown in the Figure and in eFigure 1, eFigure 2, and eFigure 3 in the Supplement. The log-rank test revealed that HPV status and transcriptionally active viral presence were individually associated with a superior DFS of 67% (HR, 0.33; 95% CI, 0.16-0.67; P = .002) and 56% (HR, 0.44; 95% CI, 0.22-0.88; P = .02), respectively (model 1 in Table 3). Conversely, positive status for E6 and E7 mRNA, high expression of p16INK4A, and low expression of p53 were not associated with improved DFS. Subsequently, Cox proportional hazards models were used to estimate the importance of these biomarkers for DFS and OS after adjusting for all the 14 variables mentioned under Study End Points and Statistical Analysis. This revealed statistically superior DFS for HPV (HR, 0.39; 95% CI, 0.18-0.85; P = .02), biologically active virus (HR, 0.36; 95% CI, 0.15-0.86; P = .02), E6 and E7 mRNA (HR, 0.36; 95% CI, 0.14-0.96; P = .04), and high p16 expression (HR, 0.49; 95% CI, 0.27-0.89; P = .02) (model 2 in Table 3).

and Statistical Analysis. This revealed statistically superior DFS for HPV (HR, 0.39; 95% CI, 0.18-0.85; P = .02), biologically active virus (HR, 0.36; 95% CI, 0.15-0.86; P = .02), E6 and E7 mRNA (HR, 0.36; 95% CI, 0.14-0.96; P = .04), and high p16 expression (HR, 0.49; 95% CI, 0.27-0.89; P = .02) (model 2 in Table 3). Figure. Survival Among Patients With High-Grade Dysplasia or Esophageal Adenocarcinoma as a Function of Human Papillomavirus (HPV) and Transcriptionally Active HPV Of 142 patients with either high-grade dysplasia or esophageal adenocarcinoma, 37 were HPV positive and 105 were HPV negative (A and B). Thirty patients had transcriptionally active HPV and 112 patients were HPV negative or had HPV that was not transcriptionally active (C and D). Table 3. Log-Rank and Multivariate Disease-Free Survival Analysis (Cox Regression) Characteristic Model 1a Model 2b Model 3c Disease-Free Survival, HR (95% CI) Unadjusted P Value Disease-Free Survival, HR (95% CI) Adjusted P Value Disease-Free Survival, HR (95% CI) Adjusted P Value HPV positive 0.33 (0.16-0.67) .002 0.39 (0.18-0.85) .02 0.36 (0.17-0.77) .009 Transcriptionally active HPV positive 0.44 (0.22-0.88) .02 0.36 (0.15-0.86) .02 0.31 (0.13-0.72) .006 E6 and E7 mRNA positive 0.50 (0.24-1.05) .07 0.36 (0.14-0.96) .04 0.32 (0.12-0.83) .02 High p16 expression 0.76 (0.46-1.25) .28 0.49 (0.27-0.89) .02 0.59 (0.34-1.02) .06 Low p53 expression 0.85 (0.53-1.37) .51 0.89 (0.53-1.51) .66 0.84 (0.51-1.39) .50 Abbreviations: HPV, human papillomavirus, HR, hazard ratio; mRNA, messenger RNA.

NA positive 0.50 (0.24-1.05) .07 0.36 (0.14-0.96) .04 0.32 (0.12-0.83) .02 High p16 expression 0.76 (0.46-1.25) .28 0.49 (0.27-0.89) .02 0.59 (0.34-1.02) .06 Low p53 expression 0.85 (0.53-1.37) .51 0.89 (0.53-1.51) .66 0.84 (0.51-1.39) .50 Abbreviations: HPV, human papillomavirus, HR, hazard ratio; mRNA, messenger RNA. a Model 1 was a univariate analysis on each characteristic with disease-free survival. b In model 2, each characteristic was analyzed in a multivariate Cox regression separately, adjusted by the following covariates: age, sex, body mass index (calculated as weight in kilograms divided by height in meters squared), ever smoked, excess alcohol, proton-pump inhibitor use, nonsteroidal anti-inflammatory ingestion, statin use, T stage, N stage, R0 resection margin, chemotherapy and radiotherapy, and tumor differentiation. c Model 3 was the same as model 2 but excluded the following covariates in the adjustment: T stage, N stage, chemotherapy, and radiotherapy. In regard to OS, HPV status was not statistically associated with improved prognosis (HR, 0.54; 95% CI, 0.28-1.04; log-rank P = .06) (model 1 in Table 4).

b In model 2, each characteristic was analyzed in a multivariate Cox regression separately, adjusted by the following covariates: age, sex, body mass index (calculated as weight in kilograms divided by height in meters squared), ever smoked, excess alcohol, proton-pump inhibitor use, nonsteroidal anti-inflammatory ingestion, statin use, T stage, N stage, R0 resection margin, chemotherapy and radiotherapy, and tumor differentiation. c Model 3 was the same as model 2 but excluded the following covariates in the adjustment: T stage, N stage, chemotherapy, and radiotherapy. In regard to OS, HPV status was not statistically associated with improved prognosis (HR, 0.54; 95% CI, 0.28-1.04; log-rank P = .06) (model 1 in Table 4). Table 4. Log-Rank and Multivariate Overall Survival Analysis (Cox Regression) Characteristic Model 1a Model 2b Model 3c Overall Survival, HR (95% CI) Unadjusted P Value Overall Survival, HR (95% CI) Adjusted P Value Overall Survival, HR (95% CI) Adjusted P Value HPV positive 0.54 (0.28-1.04) .06 1.06 (0.46-2.46) .89 0.64 (0.31-1.31) .22 Transcriptionally active HPV positive 1.09 (0.61-1.98) .76 1.80 (0.80-4.05) .16 1.01 (0.52-1.98) .98 E6 and E7 mRNA positive 1.23 (0.66-2.28) .52 1.09 (0.47-2.51) .84 1.02 (0.49-2.14) .96 High p16 expression 1.32 (0.79-2.22) .29 1.26 (0.66-2.40) .48 1.06 (0.60-1.89) .84 Low p53 expression 1.27 (0.76-2.12) .37 1.62 (0.90-2.93) .11 1.32 (0.76-2.32) .33 Abbreviations: HPV, human papillomavirus; HR, hazard ratio; mRNA, messenger RNA. a Model 1 was a univariate analysis on each characteristic with overall survival.

Table 4. Log-Rank and Multivariate Overall Survival Analysis (Cox Regression) Characteristic Model 1a Model 2b Model 3c Overall Survival, HR (95% CI) Unadjusted P Value Overall Survival, HR (95% CI) Adjusted P Value Overall Survival, HR (95% CI) Adjusted P Value HPV positive 0.54 (0.28-1.04) .06 1.06 (0.46-2.46) .89 0.64 (0.31-1.31) .22 Transcriptionally active HPV positive 1.09 (0.61-1.98) .76 1.80 (0.80-4.05) .16 1.01 (0.52-1.98) .98 E6 and E7 mRNA positive 1.23 (0.66-2.28) .52 1.09 (0.47-2.51) .84 1.02 (0.49-2.14) .96 High p16 expression 1.32 (0.79-2.22) .29 1.26 (0.66-2.40) .48 1.06 (0.60-1.89) .84 Low p53 expression 1.27 (0.76-2.12) .37 1.62 (0.90-2.93) .11 1.32 (0.76-2.32) .33 Abbreviations: HPV, human papillomavirus; HR, hazard ratio; mRNA, messenger RNA. a Model 1 was a univariate analysis on each characteristic with overall survival. b In model 2, each characteristic was analyzed in a multivariate Cox regression separately, adjusted by the following covariates: age, sex, body mass index (calculated as weight in kilograms divided by height in meters squared), ever smoked, excess alcohol, proton-pump inhibitor use, nonsteroidal anti-inflammatory ingestion, statin use, T stage, N stage, R0 resection margin, chemotherapy and radiotherapy, and tumor differentiation. c Model 3 was the same as model 2 but excluded the following covariates in the adjustment: T stage, N stage, chemotherapy, and radiotherapy.

b In model 2, each characteristic was analyzed in a multivariate Cox regression separately, adjusted by the following covariates: age, sex, body mass index (calculated as weight in kilograms divided by height in meters squared), ever smoked, excess alcohol, proton-pump inhibitor use, nonsteroidal anti-inflammatory ingestion, statin use, T stage, N stage, R0 resection margin, chemotherapy and radiotherapy, and tumor differentiation. c Model 3 was the same as model 2 but excluded the following covariates in the adjustment: T stage, N stage, chemotherapy, and radiotherapy. As viral status itself was associated with disease severity and the likelihood of too many covariates diluting the multivariate models, further analysis of these HPV-related factors was undertaken using Cox models without adjustment for stage or treatment options (model 3 in Table 3 and Table 4). This resulted in a significantly enhanced DFS for HPV DNA positivity (HR, 0.36; 95% CI, 0.17-0.77; P = .009), presence of transcriptionally active virus (HR, 0.31; 95% CI, 0.13-0.72; P = .006), and E6 and E7 mRNA detection (HR, 0.32; 95% CI, 0.12-0.83; P = .02) (model 3 in Table 3). Although the HR was lower for all 5 of these variables in relation to OS, none were statistically significant (model 3 in Table 4).

77; P = .009), presence of transcriptionally active virus (HR, 0.31; 95% CI, 0.13-0.72; P = .006), and E6 and E7 mRNA detection (HR, 0.32; 95% CI, 0.12-0.83; P = .02) (model 3 in Table 3). Although the HR was lower for all 5 of these variables in relation to OS, none were statistically significant (model 3 in Table 4). Next-Generation Sequencing of TP53 Of 142 HGD and EAC specimens, 132 (93.0%) were successfully sequenced, and in 104 of 132 (78.8%) p53 IHC and sequencing data matched. Seventy-three (55.3%) harbored mutated TP53 and 59 (44.7%) had wild type. In the 73 specimens with TP53 mutations, 50 (68.5%) were missense mutations present in the DNA binding domain (exons 5-8; aa102-292), 2 (2.7%) were missense mutations in the oligomerization domain (exons 9-10), and 2 contained frameshift mutations in exons 5 and 8 that induced a loss of both DNA binding and oligomerization domains. In 35 of 37 HPV-positive patients with HGD and EAC with successful sequencing, TP53 mutations were detected in only 9 patients (25.7%) compared with 64 of 97 (66.0%) in the viral-negative group (difference, −40.3%; 95% CI, −57.5% to −23.0%; P < .001 [Fisher exact test]).

of both DNA binding and oligomerization domains. In 35 of 37 HPV-positive patients with HGD and EAC with successful sequencing, TP53 mutations were detected in only 9 patients (25.7%) compared with 64 of 97 (66.0%) in the viral-negative group (difference, −40.3%; 95% CI, −57.5% to −23.0%; P < .001 [Fisher exact test]). Discussion This is the first study, to our knowledge, to show improved survival associated with HPV-positive HGD and EAC. Esophageal lesional HPV status and associated viral transcriptional markers, ie, E6 and E7 mRNA (gold standard) and p16INK4A (surrogate marker) are associated with improved DFS in patients with HGD and EAC. It was mainly due to a reduction in distant metastasis and possibly better local and regional control in HPV-positive patients compared with HPV-negative patients. This resulted in a lower mortality from EAC. Mean duration of OS was again significantly improved in the HPV-positive group compared with the HPV-negative group. Nevertheless, the association of HPV status with OS failed to reach significance by the log-rank test. Although 26 of 37 HPV-positive individuals (70.3%) were alive at the end of the follow-up period compared with 58 of 105 HPV-negative individuals (55.2%), this was not statistically significant, possibly because of the modest sample size and associated comorbidities. These findings are somewhat similar to the data in head and neck cancers. Human papillomavirus–positive HNSCCs have improved survival and lower rate of local and regional recurrence compared with HPV-negative head and neck cancers. No significant differences were detected in distant metastases, possibly because the studies were insufficiently powered.21,22,23 In ESCC, it has been reported that patients with p16-positive cancers had superior 5-year OS and PFS rates compared with patients with p16-negative cancers.24 Similarly, Kumar and colleagues25 found that ESCC patients with p16-positive tumors subjected to neoadjuvant chemotherapy had better complete remission rates than the p16-negative group. Conversely, in a recent publication, HPV, p16, and p53 were not found to be prognostic factors in ESCC.26

ith p16-negative cancers.24 Similarly, Kumar and colleagues25 found that ESCC patients with p16-positive tumors subjected to neoadjuvant chemotherapy had better complete remission rates than the p16-negative group. Conversely, in a recent publication, HPV, p16, and p53 were not found to be prognostic factors in ESCC.26 The results of this study are consistent with our previous work that demonstrated HPV-positive and HPV-negative EAC are distinct diseases just as in HNSCC. A previous study found that HPV-positive EAC harbored approximately 50% fewer nonsilent somatic mutations than HPV-negative EAC.8 Moreover, TP53 aberrations are less frequent in HPV-positive BD and EAC compared with viral-negative esophageal lesions.7,8,9 Prior studies from our group have also demonstrated that HPV-positive BD and EAC are mostly wild-type TP53.7,9 In this study, we similarly found that only a quarter of HPV-positive HGD and EAC lesions harbored TP53 mutations vs two-thirds of HPV-negative HGD and EAC having the same molecular aberration. Although patients with EAC who have p53 mutations have been shown to have reduced OS in a recent meta-analysis, we could not confirm the same in our study.27 Possible reasons include the small sample size and other nonviral interactions involving p53 function, including smoking (which in this study was high in both HPV-positive and HPV-negative patients) and alcohol use.28 Not surprisingly, comorbidity, especially smoking related, also influences survival negatively.29 Response to chemotherapy in the presence of other antiapoptotic proteins not analyzed in this study (eg, Bcl-2 and Bcl-xL, which are associated with both chemotherapy and radiation resistance) could be another reason for this discrepancy.30,31,32

idity, especially smoking related, also influences survival negatively.29 Response to chemotherapy in the presence of other antiapoptotic proteins not analyzed in this study (eg, Bcl-2 and Bcl-xL, which are associated with both chemotherapy and radiation resistance) could be another reason for this discrepancy.30,31,32 Patients whose HGD and EAC harbored transcriptionally active HPV (n = 30) were significantly younger (mean [SD] age, 61.4 [11.9] years) compared with patients with esophageal lesions devoid of biologically active virus (n = 112) (mean [SD] age, 67.2 [11.9] years). This is in keeping with findings from our previous small study on exome sequencing of HPV-positive and HPV-negative patients.8 Interestingly, three-quarters of HPV-positive esophageal lesions were early T stage (Tis, T1, and T2) compared with slightly more than half of the HPV-negative group. Nevertheless, the vast majority of both cohorts consisted of nodal stage N0 or N1 (89.2% of HPV-positive patients and 89.5% of viral negative patients), which is probably indicative of patient selection for endotherapy or esophagectomy with curative intent. Our results concur somewhat with data from HNSCC, another malignant neoplasm in which a subset are HPV driven. Patients with HPV-positive HNSCC are generally 10 years younger33 and have more early T-stage but advanced N-stage disease, although they respond better to treatment and have a better outcome than patients with HPV-negative HNSCC.34,35

t with data from HNSCC, another malignant neoplasm in which a subset are HPV driven. Patients with HPV-positive HNSCC are generally 10 years younger33 and have more early T-stage but advanced N-stage disease, although they respond better to treatment and have a better outcome than patients with HPV-negative HNSCC.34,35 The determination of HPV presence in BD or EAC can be difficult using FFPE specimens. While PCR is more sensitive than ISH for HPV DNA detection, it risks false-positivity. Thus, in addition to PCR for viral DNA, we also included E6 and E7 mRNA transcript analysis by ISH, which is a reliable marker of HPV involvement, and IHC for p16 overexpression, a surrogate marker for HPV infection. Sequencing and IHC for p53 were also undertaken given their importance in BD and EAC progression.36 Central reporting of biopsy specimens by experienced gastrointestinal pathologists was an added strength.

ich is a reliable marker of HPV involvement, and IHC for p16 overexpression, a surrogate marker for HPV infection. Sequencing and IHC for p53 were also undertaken given their importance in BD and EAC progression.36 Central reporting of biopsy specimens by experienced gastrointestinal pathologists was an added strength. Limitations This investigation was retrospective in nature and the study sample was small. The case-control nature of the study introduces biases pertaining to selection, information, and observation as well as confounding. As HPV status was not known at the time of enrollment and treatment decision, it minimized both selection and observer bias. Measurement bias was addressed by blinding the scientist and pathologists to the clinical and virological status of the patients and to treatment outcome. Confounding was mitigated with adjustment for potential confounders in the multivariate statistical analysis. The independence of prognostic effects of the 5 dichotomous variables, HPV DNA positivity, transcriptionally active HPV, E6 and E7 mRNA detection, and p16 and p53 overexpression, is important. Nevertheless, we could not include these variables in a single multivariate model as they were intercorrelated. Thus, we were unable to identify the independent effects of these factors. Some of the specimens analyzed were more than 10 years old, which increases the risk of DNA and RNA invalidity. Moreover, IHC analysis is subjective and lacks uniform scoring systems, which can hamper reproducibility between studies.

correlated. Thus, we were unable to identify the independent effects of these factors. Some of the specimens analyzed were more than 10 years old, which increases the risk of DNA and RNA invalidity. Moreover, IHC analysis is subjective and lacks uniform scoring systems, which can hamper reproducibility between studies. Conclusions If these findings of a favorable prognosis of HPV-positive HGD and EAC are confirmed in larger cohorts with more advanced disease, it presents an opportunity for treatment de-escalation in the hope of reducing toxic effects without deleteriously affecting survival. Supplement. eFigure 1. Flow Diagram Showing HPV DNA and Transcriptional Marker Status of Enrolled Study Patients eFigure 2. E6 and E7 mRNA Status and p16 Overexpression eFigure 3. p53 Expression eTable. Demographic and Clinical Characteristics of the Study Population and Associated Tumors According to Patient Group Click here for additional data file.

Introduction Reducing guideline-discordant prostate cancer staging imaging is an important national priority. Minimizing guideline-discordant imaging for men with low-risk prostate cancer is listed as a primary focus for reducing inefficient health care utilization in the Choosing Wisely campaign.1,2,3 Within the context of the Veterans Health Administration (VA) system, the Veterans Access, Choice, and Accountability Act, also known as the Choice Act, was passed in 2014 for the goal of reducing wait times for veterans seeking access to specialized health care services by providing funds for patients to seek care outside the VA.4 Changes in policy, such as the Choice Act, have unclear implications for the quality and cost-effectiveness of care that patients receive.

was passed in 2014 for the goal of reducing wait times for veterans seeking access to specialized health care services by providing funds for patients to seek care outside the VA.4 Changes in policy, such as the Choice Act, have unclear implications for the quality and cost-effectiveness of care that patients receive. Prostate cancer imaging rates appear to vary among veterans depending on patients’ ability to seek care outside of the VA health system. Earlier research found higher rates of guideline-discordant prostate cancer imaging among VA patients with low-risk prostate cancer who used Medicare services than among those with no Medicare utilization.5 Within the VA, physicians typically receive a set salary that does not include financial incentives to provide more health care services.6 Outside the VA, the fee-for-service model used in Medicare and in most health care systems in the United States may encourage provision of more health care services7 because of direct physician reimbursement8,9 and patient self-referral. Earlier qualitative work found that physicians practicing at the VA are cognizant of this incentive difference between VA and non-VA practice and acknowledge that their own ordering behavior varies by setting.10 Many of these additional services that are associated with potential financial incentives may have limited efficacy or even be quantifiably unnecessary.11 Specifically, it is unclear whether a difference in care patterns among veterans seeking care only through the VA vs through the VA and Medicare would also apply to Medicare patients with no access to the VA.

associated with potential financial incentives may have limited efficacy or even be quantifiably unnecessary.11 Specifically, it is unclear whether a difference in care patterns among veterans seeking care only through the VA vs through the VA and Medicare would also apply to Medicare patients with no access to the VA. The aim of this study was to directly assess the association between quality of health care within the VA health system vs a fee-for-service system (Medicare) by comparing rates of guideline-discordant and guideline-concordant imaging among patients with prostate cancer. To do this, we categorized patients into 1 of 3 groups: those who received health care through a fee-for-service system (the Medicare-only group), those who received health care through an integrated health system (the VA-only group), and those who received health care through a combination of the 2 systems (VA with some Medicare use; the VA and Medicare group).

f 3 groups: those who received health care through a fee-for-service system (the Medicare-only group), those who received health care through an integrated health system (the VA-only group), and those who received health care through a combination of the 2 systems (VA with some Medicare use; the VA and Medicare group). Outside the VA health system, there may be higher rates of guideline-discordant prostate cancer imaging, suggesting a trade-off of resources for quality of care in a fee-for-service setting. If the rates of guideline-discordant prostate cancer imaging are actually lower outside the VA health system, it is possible that there is a problem in terms of quality of care in an integrated health system. We hypothesized that men with prostate cancer who used Medicare only would have the highest rate of guideline-discordant imaging, that those who used both health systems would have the next highest rate, and that those who used the VA only would have the lowest rate. The results of this study may help policy makers understand the implications of particular health care policies, such as the Choice Act, and highlight areas for improvement in the cost-effectiveness and quality of health care for veterans.

xt highest rate, and that those who used the VA only would have the lowest rate. The results of this study may help policy makers understand the implications of particular health care policies, such as the Choice Act, and highlight areas for improvement in the cost-effectiveness and quality of health care for veterans. Methods We conducted a retrospective cohort study to compare rates of prostate cancer staging imaging among distinct health care settings. This study was approved by the institutional review boards of VA New York Harbor Healthcare and VA Puget Sound Healthcare Systems with a waiver of informed consent to include review of patient records. We used the VA Central Cancer Registry (VACCR), linked to administrate claims and Medicare utilization records, and the Surveillance, Epidemiology, and End Results Program (SEER) database to compose our cohorts according to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.12,13

VA Central Cancer Registry (VACCR), linked to administrate claims and Medicare utilization records, and the Surveillance, Epidemiology, and End Results Program (SEER) database to compose our cohorts according to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.12,13 We created 3 nationally representative cohorts of men who received a diagnosis of prostate cancer between 2004 and 2008: veterans who received a diagnosis at the VA, men whose diagnoses were found in SEER Medicare claims, and veterans who received a diagnosis at the VA and chose to receive subsequent care at private facilities using Medicare insurance.5 The VA patients were identified through the VA Information Resource Center Corporate Data Warehouse using inpatient, outpatient, surgery, and vital measures tables. Medicare patients were identified from the SEER Medicare database, which includes inpatient (MedPAR), outpatient, and physician claims. To identify dual users, we searched for any Medicare claims for veterans in our VA cohort and included these claims in our cohort data.14,15 The dual user group comprised VA patients who were eligible for Medicare benefits; there is no mechanism in place for Medicare patients to be referred to the VA for care. Of note, the VA sometimes outsources patient care to community health care professionals when the equipment or staff required to provide a specialized service is not available in a local VA medical center and the patient is unable to travel.16,17 Veterans who receive health care services at private facilities in this way, known as fee basis services in the VA, may have images obtained in the same facilities as Medicare users. Although we included these privately obtained services in our analysis, we classified these patients as part of the VA-only cohort because their imaging was directed solely by VA health care professionals. For all 3 groups, we identified prostate cancer diagnoses and associated care using International Classification of Diseases, Ninth Revision (ICD-9) diagnosis and procedure codes (92.14, 92.18, 88.01, and 88.95) and Current Procedural Terminology/Healthcare Common Procedure Coding System codes (78306, 78315, 78102, 78103, 78104, 72191, 72192, 72193, 72194, 74150, 74160, 74170, 74175, 72198, 74185, 72195, 72196, 72197, 74181, 74182, and 74183).

inth Revision (ICD-9) diagnosis and procedure codes (92.14, 92.18, 88.01, and 88.95) and Current Procedural Terminology/Healthcare Common Procedure Coding System codes (78306, 78315, 78102, 78103, 78104, 72191, 72192, 72193, 72194, 74150, 74160, 74170, 74175, 72198, 74185, 72195, 72196, 72197, 74181, 74182, and 74183). The study population consisted of men who received a diagnosis of prostate cancer between January 1, 2004, and March 31, 2008. We included patients younger than 85 years at diagnosis with a pathologically confirmed, registry-documented diagnosis, thus eliminating men with a diagnosis at autopsy or on the death certificate. We excluded men who died within 3 months after diagnosis and men without utilization of any health benefits in months 2 to 6 after diagnosis, a situation suggesting complete departure from the health care system.5 In addition, men who received a diagnosis at facilities with a prostate cancer diagnosis volume less than 25 cases per year were eliminated because of the unreliability of clinical data from such institutions.18 We also excluded men without high-risk features but missing any tumor risk characteristic (prostate-specific antigen [PSA], Gleason grade, or clinical stage), because we could not confirm the risk classification of these men.

r year were eliminated because of the unreliability of clinical data from such institutions.18 We also excluded men without high-risk features but missing any tumor risk characteristic (prostate-specific antigen [PSA], Gleason grade, or clinical stage), because we could not confirm the risk classification of these men. Our primary dependent variable of interest was receipt of any of the following imaging studies: radionuclide bone scan, computed tomography (CT), or magnetic resonance imaging (MRI). Guideline concordance was determined on the basis of National Comprehensive Cancer Network (NCCN) recommendations for imaging to stage incident prostate cancer.19 Receipt of bone scan was considered to be guideline discordant unless a patient had any of the following high-risk characteristics: clinical stage T3 or higher, Gleason score 8 or higher, and PSA level of 20 ng/mL or higher. Receipt of CT or MRI was considered to be guideline discordant unless the patient had a 20% or higher risk of positive lymph nodes, estimated from the Partin tables.20 Men who met the requirements for bone scan, CT, or MRI were classified as having high-risk prostate cancer, whereas all other men were classified as having low-risk prostate cancer. Two exceptions were made to these rules regarding guideline concordance of imaging. Receipt of CT was considered to be guideline concordant for patients with low-risk prostate cancer if they were undergoing radiation therapy, because CT may have been used for treatment planning rather than for disease staging.21 In addition, receipt of bone scan by a patient with low-risk prostate cancer was considered to be guideline concordant if the patient had a diagnosis of spinal or pathologic fractures in the 3-month period before prostate cancer diagnosis or until 6 months after prostate cancer diagnosis.22 Patients with low-risk prostate cancer were classified as guideline discordant if they received an inappropriate bone scan, CT, or MRI. Patients with high-risk prostate cancer were classified as guideline concordant if they received a bone scan, CT, or MRI. It is possible that patients received multiple scans; we did not measure surplus guideline-discordant imaging among patients with low-risk prostate cancer or surplus guideline-concordant imaging among patients with high-risk prostate cancer for this study.

guideline concordant if they received a bone scan, CT, or MRI. It is possible that patients received multiple scans; we did not measure surplus guideline-discordant imaging among patients with low-risk prostate cancer or surplus guideline-concordant imaging among patients with high-risk prostate cancer for this study. The VA and Medicare claims were reviewed for the period beginning 3 months before diagnosis until cancer treatment initiation, death, or 6 months after diagnosis, whichever was earliest. Patient treatment for prostate cancer was identified using ICD-9 diagnosis codes and categorized on the basis of the earliest treatment date as radical prostatectomy, radiation, or other treatment. Other treatment included androgen deprivation, cryosurgery, and watchful waiting (ie, no treatment found). Data also included age at diagnosis, race, marital status, geographical region of the institution where the diagnosis was given, clinical stage, PSA level at diagnosis, Gleason score at diagnosis, and diagnosis year. In addition, we analyzed claims for the year before prostate cancer diagnosis to calculate an unweighted Elixhauser comorbidity score.23 Median household income and proportion of the population with a college degree were identified for each patient’s county of residence on the basis of 2010 US Census data.24 The institution where a diagnosis was given was categorized on the basis of the average number of prostate cancer cases diagnosed per year during the study period as high volume (>99 cases), medium volume (60-99 cases), or low volume (<60 cases).5

ach patient’s county of residence on the basis of 2010 US Census data.24 The institution where a diagnosis was given was categorized on the basis of the average number of prostate cancer cases diagnosed per year during the study period as high volume (>99 cases), medium volume (60-99 cases), or low volume (<60 cases).5 Statistical Analysis We created a multivariable logistic regression model to determine the association between all described covariates and the receipt of imaging, stratified by high-risk and low-risk cancer and clustered by region. All covariates were considered to be theoretically important and thus remained in the model regardless of their level of statistical significance. Because the VA cohort included men younger than 65 years and the Medicare cohort did not, we performed a sensitivity analysis that included only men 65 years and older. We report adjusted risk ratios (RRs) computed from the fitted logistic regression model. Statistical analyses were performed using Stata software (version 12.0; StataCorp). All P values were 2-sided with statistical significance at α = .05. Analysis of the data was completed between March 2016 and February 2018.

older. We report adjusted risk ratios (RRs) computed from the fitted logistic regression model. Statistical analyses were performed using Stata software (version 12.0; StataCorp). All P values were 2-sided with statistical significance at α = .05. Analysis of the data was completed between March 2016 and February 2018. Results The cohort included 98 867 men with incident prostate cancer (mean [SD] age, 70.26 [7.48] years). Table 1 shows that the majority of the sample (57.3%) were in the Medicare-only group, followed by 28.1% in the VA-only group and 14.5% in the VA and Medicare group. Reflecting current nationwide patterns,5,25 the majority of the cohort (69.8%) were categorized as having low-risk prostate cancer. The overall study sample was predominantly white (77.4%), followed by black (17.4%), other races (2.8%), and missing or unknown race (2.3%). The largest age group in the cohort was 75 to 85 years (28.9%), followed by 70 to 74 years (25.9%), 65 to 69 years (25.1%), and 64 years and younger (20.0%). Because we limited the Medicare group to men who were age qualified, this group had no members younger than 65 years. Most of the men in the cohort (62.9%) were married. For all variables across the 3 groups, χ2 tests revealed statistically significant differences. Results of the sensitivity analysis limited to men 65 years and older revealed no significant differences in the magnitude or direction of the odds ratios from the primary analysis. Table 1. Demographic and Clinical Characteristics of Men With Incident Prostate Cancer Characteristic No.

Results The cohort included 98 867 men with incident prostate cancer (mean [SD] age, 70.26 [7.48] years). Table 1 shows that the majority of the sample (57.3%) were in the Medicare-only group, followed by 28.1% in the VA-only group and 14.5% in the VA and Medicare group. Reflecting current nationwide patterns,5,25 the majority of the cohort (69.8%) were categorized as having low-risk prostate cancer. The overall study sample was predominantly white (77.4%), followed by black (17.4%), other races (2.8%), and missing or unknown race (2.3%). The largest age group in the cohort was 75 to 85 years (28.9%), followed by 70 to 74 years (25.9%), 65 to 69 years (25.1%), and 64 years and younger (20.0%). Because we limited the Medicare group to men who were age qualified, this group had no members younger than 65 years. Most of the men in the cohort (62.9%) were married. For all variables across the 3 groups, χ2 tests revealed statistically significant differences. Results of the sensitivity analysis limited to men 65 years and older revealed no significant differences in the magnitude or direction of the odds ratios from the primary analysis. Table 1. Demographic and Clinical Characteristics of Men With Incident Prostate Cancer Characteristic No. (%) of Men VA only (n = 27 811) VA and Medicare (n = 14 385) Medicare Only (n = 56 671) Total (N = 98 867) Prostate cancer imaging group Low-risk prostate cancer 20 066 (72.2) 9963 (69.3) 38 973 (68.8) 69 002 (69.8) High-risk prostate cancer 7745 (27.8) 4422 (30.7) 17 698 (31.2) 29 865 (30.2) Clinical stage T1 17 626 (63.4) 8817 (61.3) 31 203 (55.1) 57 646 (58.3) T2NOS 1237 (4.5) 747 (5.2) 13 806 (24.4) 15 790 (16.0) T2A 3046 (11.0) 1592 (11.1) 2722 (4.8) 7360 (7.4) T2B 1280 (4.6) 719 (5.0) 1159 (2.1) 3158 (3.2) T2C 3489 (12.6) 2012 (14.0) 4562 (8.1) 10 063 (10.2) T3 626 (2.3) 280 (2.0) 1658 (2.9) 2564 (2.6) T4 209 (0.8) 100 (0.7) 605 (1.1) 914 (0.9) Missing 298 (1.1) 118 (0.8) 956 (1.7) 1372 (1.4) Gleason grade <7 12 845 (46.2) 6163 (42.8) 23 698 (41.8) 42 706 (43.2) 3 + 4 7408 (26.6) 3684 (25.6) 14 229 (25.1) 25 321 (25.6) 4 + 3 2794 (10.1) 1613 (11.2) 6390 (11.3) 10 797 (10.9) ≥8 4566 (16.4) 2802 (19.5) 11 179 (19.7) 18 547 (18.8) Missing 198 (0.7) 123 (0.9) 1175 (2.1) 1496 (1.5) PSA level, ng/mL 0-4 3172 (11.4) 1543 (10.7) 6763 (11.9) 11 478 (11.6) >4 to 10 16 103 (57.9) 8077 (56.2) 30 894 (54.5) 55 074 (55.7) >10 to 20 4070 (14.6) 2387 (16.6) 9375 (16.5) 15 832 (16.0) >20 4315 (15.5) 2285 (15.9) 7702 (13.6) 14 302 (14.5) Missing 151 (0.5) 93 (0.7) 1937 (3.4) 2181 (2.2) Race Black 8014 (28.8) 2752 (19.1) 6429 (11.3) 17 195 (17.4) White 18 834 (67.7) 11 041 (76.8) 46 694 (82.4) 76 569 (77.4) Other 283 (1.0) 142 (1.0) 2372 (4.2) 2797 (2.8) Missing 680 (2.5) 450 (3.1) 1176 (2.1) 2306 (2.3) Age, y <65 17 792 (64.0) 1999 (13.9) 0 19 791 (20.0) 65-69 3817 (13.7) 3750 (26.1) 17 291 (30.5) 24 858 (25.1) 70-74 3198 (11.5) 4356 (30.3) 18 094 (31.9) 25 648 (25.9) ≥75 3004 (10.8) 4280 (29.8) 21 286 (37.6) 28 570 (28.9) Marital status Married 13 784 (49.6) 9025 (62.7) 39 402 (69.5) 62 211 (62.9) Single, divorced, or widowed 13 872 (49.9) 5311 (36.9) 11 780 (20.8) 30 963 (31.3) Missing 155 (0.6) 49 (0.3) 5489 (9.7) 5693 (5.8) Medical comorbidities 0 7344 (26.4) 3381 (23.5) 42 436 (74.9) 53 161 (53.8) 1-2 9217 (33.1) 4585 (31.9) 13 209 (23.3) 27 011 (27.3) ≥3 11 250 (40.5) 6419 (44.6) 1026 (1.8) 18 695 (18.9) Treatment Watchful waiting and/or hormone therapy 11 327 (40.7)

(20.8) 30 963 (31.3) Missing 155 (0.6) 49 (0.3) 5489 (9.7) 5693 (5.8) Medical comorbidities 0 7344 (26.4) 3381 (23.5) 42 436 (74.9) 53 161 (53.8) 1-2 9217 (33.1) 4585 (31.9) 13 209 (23.3) 27 011 (27.3) ≥3 11 250 (40.5) 6419 (44.6) 1026 (1.8) 18 695 (18.9) Treatment Watchful waiting and/or hormone therapy 11 327 (40.7) 5928 (41.2) 22 934 (40.5) 40 189 (40.6) Prostatectomy 7913 (28.5) 2646 (18.4) 13 784 (24.3) 24 343 (24.6) Radiation therapy 8571 (30.8) 5811 (40.4) 19 953 (35.2) 34 335 (34.7) Hospital volume category, cases per y <60 2739 (9.9) 2019 (14.0) 24 922 (44.0) 29 680 (30.0) 60-99 6656 (23.9) 3506 (24.4) 9830 (17.4) 19 992 (20.2) >99 18 416 (66.2) 8860 (61.6) 10 826 (19.1) 38 102 (38.5) Missing 0 0 11 093 (19.6) 11 093 (88.8) Census tract per capita income, $ <25 000 3261 (11.7) 1386 (9.6) 4250 (7.5) 8897 (9.0) 25 000-34 999 8716 (31.3) 4929 (34.3) 10 219 (18.0) 23 864 (24.1) 35 000-44 999 7953 (28.6) 4243 (29.5) 11 875 (21.0) 24 071 (24.3) 45 000-54 999 3692 (13.3) 1799 (12.5) 9776 (17.3) 15 267 (15.4) >55 000 3098 (11.1) 1505 (10.5) 20 210 (35.7) 24 813 (25.1) Missing 1091 (3.9) 523 (3.6) 341 (0.6) 1955 (2.0) Census tract population with ≥4 y of college, % <10 4792 (17.2) 2438(17.0) 9206 (16.2) 16 436 (16.6) 10 to <20 12 069 (43.4) 6612 (46.0) 16 278 (28.7) 34 959 (35.4) 20 to <30 5064 (18.2) 2395 (16.7) 9863 (17.4) 17 322 (17.5) ≥30 4468 (16.1) 2212 (15.4) 20 409 (36.0) 27 089 (27.4) Missing 1418 (5.1) 728 (5.1) 915 (1.6) 3061 (3.1) Year of diagnosis 2004 6575 (23.6) 2723 (18.9) 14 253 (25.2) 23 551 (23.8) 2005 6036 (21.7) 3171 (22.0) 13 657 (24.1) 22 864 (23.1) 2006 6365 (22.9) 3592 (25.0) 14 451 (25.5) 24 408 (24.7) 2007 7262 (26.1) 4030 (28.0) 14 310 (25.3) 25 602 (25.9) 2008 1573 (5.7) 869 (6.0) 0 2442 (2.5) Abbreviations: PSA, prostate-specific antigen; VA, Veterans Health Administration.

14 253 (25.2) 23 551 (23.8) 2005 6036 (21.7) 3171 (22.0) 13 657 (24.1) 22 864 (23.1) 2006 6365 (22.9) 3592 (25.0) 14 451 (25.5) 24 408 (24.7) 2007 7262 (26.1) 4030 (28.0) 14 310 (25.3) 25 602 (25.9) 2008 1573 (5.7) 869 (6.0) 0 2442 (2.5) Abbreviations: PSA, prostate-specific antigen; VA, Veterans Health Administration. Men With Low-Risk Prostate Cancer Among men with low-risk prostate cancer, more men in the Medicare-only group received at least 1 imaging test for staging (52.5%) compared with the VA and Medicare group (50.9%) and VA-only group (45.9%) (P < .001) (Table 2). The Medicare-only group was also least likely to receive guideline-concordant care (53.1%) vs the VA and Medicare group (56.4%) and VA-only group (60.6%) (P < .001) (Table 2). Results of the multivariable model showed that, compared with the Medicare-only reference group, being a VA and Medicare group patient (RR, 0.87; 95% CI, 0.76-0.98) or a VA-only group patient (RR, 0.79; 95% CI, 0.67-0.92) was associated with reduced guideline-discordant imaging (Table 3).26 For men with low-risk prostate cancer, all clinical markers were significantly associated with receipt of imaging. Clinical stage of T2B (RR, 1.25; 95% CI, 1.17-1.33) or T2C (RR, 1.21; 95% CI, 1.15-1.27), Gleason score of 7 or higher (Gleason score 3 + 4: RR, 1.23; 95% CI, 1.17-1.28; Gleason score 4 + 3: RR, 1.38; 95% CI, 1.30-1.46), and PSA level greater than 10 ng/mL (RR, 1.52; 95% CI, 1.39-1.64) were associated with increased receipt of guideline-discordant imaging. In addition, an increased level of comorbidity was associated with greater risk of guideline-discordant imaging, whereas living in a county with higher educational levels was associated with lower rates of imaging (Table 3).

.52; 95% CI, 1.39-1.64) were associated with increased receipt of guideline-discordant imaging. In addition, an increased level of comorbidity was associated with greater risk of guideline-discordant imaging, whereas living in a county with higher educational levels was associated with lower rates of imaging (Table 3). Table 2. Imaging Use Among Men Who Received a Diagnosis of Incident Prostate Cancera Imaging Modality No. (%) of Men VA Only (n = 27 811) VA and Medicare (n = 14 385) Medicare Only (n = 56 671) Men with low-risk disease characteristics Guideline-concordant care 12 161 (60.6) 5620 (56.4) 20 686 (53.1) Any imaging 9214 (45.9) 5068 (50.9) 20 455 (52.5) Bone scan 6474 (32.3) 3552 (35.7) 16 014 (41.1) CT 6769 (33.7) 3694 (37.1) 13 516 (34.7) MRI 363 (1.8) 369 (3.7) 2287 (5.9) Men with high-risk disease characteristics Guideline-concordant care 5322 (68.7) 3181 (71.2) 11 814 (66.8) Any imaging 5833 (75.3) 3494 (79.0) 13 583 (76.8) Bone scan 5166 (66.7) 3111 (70.4) 12 454 (70.4) CT 4488 (58.0) 2580 (58.3) 9645 (54.5) MRI 270 (3.5) 219 (5.0) 1363 (7.7) Abbreviations: CT, computed tomography; MRI, magnetic resonance imaging; VA, Veterans Health Administration. a Within 3 months before prostate cancer diagnosis and up to 6 months after diagnosis. P < .001 for all associations by χ2 test. Table 3.

Table 2. Imaging Use Among Men Who Received a Diagnosis of Incident Prostate Cancera Imaging Modality No. (%) of Men VA Only (n = 27 811) VA and Medicare (n = 14 385) Medicare Only (n = 56 671) Men with low-risk disease characteristics Guideline-concordant care 12 161 (60.6) 5620 (56.4) 20 686 (53.1) Any imaging 9214 (45.9) 5068 (50.9) 20 455 (52.5) Bone scan 6474 (32.3) 3552 (35.7) 16 014 (41.1) CT 6769 (33.7) 3694 (37.1) 13 516 (34.7) MRI 363 (1.8) 369 (3.7) 2287 (5.9) Men with high-risk disease characteristics Guideline-concordant care 5322 (68.7) 3181 (71.2) 11 814 (66.8) Any imaging 5833 (75.3) 3494 (79.0) 13 583 (76.8) Bone scan 5166 (66.7) 3111 (70.4) 12 454 (70.4) CT 4488 (58.0) 2580 (58.3) 9645 (54.5) MRI 270 (3.5) 219 (5.0) 1363 (7.7) Abbreviations: CT, computed tomography; MRI, magnetic resonance imaging; VA, Veterans Health Administration. a Within 3 months before prostate cancer diagnosis and up to 6 months after diagnosis. P < .001 for all associations by χ2 test. Table 3. Adjusted Risk Ratios of Receipt of Imaging Staging Test Associated With Clinical and Demographic Factors Among Men With Incident Prostate Cancer, Stratified by Imaging Indication and Clustered by VA Region Variable Risk Ratio (95% CI) Low-Risk Prostate Cancer High-Risk Prostate Cancer Insurance group Medicare only 1 [Reference] 1 [Reference] VA only 0.79 (0.67-0.92)a 1.00 (0.95-1.06) VA and Medicare 0.87 (0.76-0.98)a 1.04 (0.98-1.09) Clinical stage T1 1 [Reference] 1 [Reference] T2NOS 1.03 (0.96-1.09) 1.08 (1.05-1.11)a T2A 1.07 (1.00-1.15) 0.99 (0.92-1.05) T2B 1.25 (1.17-1.33)a 1.07 (1.02-1.13)a T2C 1.21 (1.15-1.27)a 1.08 (1.04-1.12)a T3 NA 1.06 T4 NA 0.98 Missing NA 1.03 Gleason grade <7 1 [Reference] 1 [Reference] 3 + 4 1.23 (1.17-1.28)a 1.24 (1.15-1.35)a 4 + 3 1.38 (1.30-1.46)a 1.32 (1.20-1.43)a ≥8 NA 1.45 (1.31-1.58) Missing NA 1.14 PSA level, ng/mL 0-4 1 [Reference] 1 [Reference] >4 to 10 0.95 (0.92-0.99)a 0.97 (0.93-1.01) >10 to 20 1.52 (1.39-1.64)a 1.06 (1.02-1.11)a >20 NA 1.1 Missing NA 0.91 Race Black 1 [Reference] 1 [Reference] White 0.94 (0.86-1.01) 0.98 (0.95-1.02) Other 0.97 (0.81-1.12) 1.02 (0.96-1.08) Missing 0.85 (0.73-096) 0.89 (0.82-0.97)a Age, y <65 1 [Reference] 1 [Reference] 65-69 0.98 (0.92-1.03) 1.01 (0.98-1.04) 70-74 1.00 (0.94-1.07) 1.04 (1.00-1.07)a ≥75 1.02 (0.95-1.09) 0.99 (0.95-1.02) Marital status Married 0.98 (0.95-1.01) 1.00 (0.98-1.03) Single, divorced, or widowed 1 [Reference] 1 [Reference] Missing 0.97 (0.92-1.03) 1.07 (1.02-1.11) Medical comorbidities 0 1 [Reference] 1 [Reference] 1-2 1.08 (1.05-1.11)a 1.03 (1.01-1.06)a ≥3 1.12 (1.07-1.16)a 1.01 (0.99-1.03) Hospital volume category, cases per y <60 1 [Reference] 1 [Reference] 60-99 0.90 (0.81-0.98)a 1.00 (0.95-1.05) >99 0.93 (0.84-1.01) 1.03 (0.98-1.08) Missing 0.78 (0.72-0.83) 0.86 (0.82-0.90) Census tract per capita income, $ <25 000 1 [Reference] 1 [Reference] 25 000-34 999 0.90 (0.79-1.02) 0.96 (0.91-1.02) 35 000-44 999 0.95 (0.95-0.96)a 1.01 (1.01-1.01)a 45 000-54 999 0.97 (0.96-0.97)a 1.02 (1.01-1.02)a >55 000 1.05 (1.04-1.05)a 1.03 (1.03-1.04)a Missing or unknown 1.11 (1.09-1.12) 1.02 (1.02-1.02) Census tract population with ≥4 y of college, % <10 1 [Reference] 1 [Refer

25 000-34 999 0.90 (0.79-1.02) 0.96 (0.91-1.02) 35 000-44 999 0.95 (0.95-0.96)a 1.01 (1.01-1.01)a 45 000-54 999 0.97 (0.96-0.97)a 1.02 (1.01-1.02)a >55 000 1.05 (1.04-1.05)a 1.03 (1.03-1.04)a Missing or unknown 1.11 (1.09-1.12) 1.02 (1.02-1.02) Census tract population with ≥4 y of college, % <10 1 [Reference] 1 [Refer ence] 10 to <20 0.99 (0.94-1.05) 1.00 (0.97-1.03) 20 to <30 0.94 (0.86-1.01) 1.00 (0.96-1.04) ≥30 0.91 (0.83-1.00)a 1.01 (0.97-1.05) Missing 0.89 (0.79-0.99) 0.97 (0.90-1.05) Year of diagnosis 2004 1 [Reference] 1 [Reference] 2005 1.01 (0.98-1.04) 1.05 (1.02-1.09)a 2006 1.03 (0.99-1.07) 1.09 (1.05-1.13)a 2007 1.03 (0.98-1.08) 1.13 (1.09-1.18)a 2008 0.96 (0.88-1.04) 1.18 (1.10-1.26)a Abbreviations: NA, not available; PSA, prostate-specific antigen; VA, Veterans Health Administration. a P < .05, determined using adjusted risk ratios from multivariate logistic regression models with delta method standard errors.26

ence] 10 to <20 0.99 (0.94-1.05) 1.00 (0.97-1.03) 20 to <30 0.94 (0.86-1.01) 1.00 (0.96-1.04) ≥30 0.91 (0.83-1.00)a 1.01 (0.97-1.05) Missing 0.89 (0.79-0.99) 0.97 (0.90-1.05) Year of diagnosis 2004 1 [Reference] 1 [Reference] 2005 1.01 (0.98-1.04) 1.05 (1.02-1.09)a 2006 1.03 (0.99-1.07) 1.09 (1.05-1.13)a 2007 1.03 (0.98-1.08) 1.13 (1.09-1.18)a 2008 0.96 (0.88-1.04) 1.18 (1.10-1.26)a Abbreviations: NA, not available; PSA, prostate-specific antigen; VA, Veterans Health Administration. a P < .05, determined using adjusted risk ratios from multivariate logistic regression models with delta method standard errors.26 Men With High-Risk Prostate Cancer The proportion of men with high-risk prostate cancer who received at least 1 imaging test was similar among the 3 health care delivery systems, ranging from 75.3% to 79.0% (P < .001) (Table 2). The rate of guideline-concordant imaging among patients with high-risk prostate cancer was also similar among the health care delivery systems, with the VA and Medicare group having a slightly higher percentage of guideline-concordant patients with high-risk prostate cancer (71.2%), followed by the VA-only group (68.7%) and the Medicare-only group (66.8%) (Table 2). The multivariate model confirmed that, for men who received a diagnosis of high-risk prostate cancer, where they sought care was not significantly associated with receipt of guideline-concordant imaging. Gleason score of 7 or higher (Gleason score 3 + 4: RR, 1.24 95% CI, 1.15-1.35; Gleason score 4 + 3: RR, 1.32; 95% CI, 1.20-1.43; and Gleason score ≥8: RR, 1.45; 95% CI, 1.31-1.58), greater median county income (>$55 000: RR, 1.03; 95% CI, 1.03-1.04), and later year of diagnosis (2005: RR, 1.05; 95% CI, 1.02-1.09; 2006: RR, 1.09; 95% CI, 1.05-1.13; 2007: RR, 1.13; 95% CI, 1.09-1.18; and 2008: RR, 1.18; 95% CI, 1.10-1.26) had a statistically significant association with receipt of guideline-concordant imaging among men with high-risk prostate cancer (Table 3).

1.04), and later year of diagnosis (2005: RR, 1.05; 95% CI, 1.02-1.09; 2006: RR, 1.09; 95% CI, 1.05-1.13; 2007: RR, 1.13; 95% CI, 1.09-1.18; and 2008: RR, 1.18; 95% CI, 1.10-1.26) had a statistically significant association with receipt of guideline-concordant imaging among men with high-risk prostate cancer (Table 3). Discussion In this study, we analyzed rates of prostate cancer staging imaging among men with a diagnosis of prostate cancer in 3 types of health care delivery settings. We found that men with low-risk disease were less likely to receive guideline-discordant staging imaging if they received a diagnosis and treatment in the VA exclusively compared with those who received care using Medicare or a combination of VA and Medicare services. Furthermore, dividing our study sample into 3 cohorts based on delivery system revealed a dose-response relationship, with men with low-risk prostate cancer in the VA-only group having the lowest likelihood of guideline-discordant imaging, those in the VA and Medicare group having the next highest likelihood of guideline-discordant imaging, and those in the Medicare-only group having the highest likelihood of guideline-discordant imaging. The VA and Medicare health systems differ in notable ways, including their patient demographics and institutional characteristics,27,28,29 but it is possible that the differing financial incentives for physicians between these 2 health care delivery systems contributed to significantly different risks of guideline-concordant imaging among men with low-risk cancer. Although differing financial incentives were associated with differences in the rates of guideline-discordant imaging between the 2 groups, the rate of guideline-discordant imaging among patients with low-risk prostate cancer in the VA was high (45.9%) compared with that in other health care delivery systems without financial incentives for overuse, such as in Sweden, where recent guideline-discordant imaging rates were as low as 3% after an intervention to bring imaging practices in line with guidelines.30 This finding suggests that factors in addition to financial incentives, such as physician education, culture, and habituation, could also contribute to guideline-discordant prostate cancer imaging rates in the United States.

s were as low as 3% after an intervention to bring imaging practices in line with guidelines.30 This finding suggests that factors in addition to financial incentives, such as physician education, culture, and habituation, could also contribute to guideline-discordant prostate cancer imaging rates in the United States. Disease characteristics were also associated with the receipt of imaging, whether guideline discordant for men with low-risk disease or guideline concordant for men with high-risk disease. Presentation with higher clinical stage, Gleason grade, or PSA level was associated with men with low-risk prostate cancer receiving presumably unwarranted imaging tests. For men with high-risk disease, only increased Gleason grade was associated with guideline-concordant imaging. These findings suggest that physicians find it difficult to adhere to evidence-based heuristics when faced with test results that approach high-risk levels, even when these test results are not sufficient to categorize the patient as having high-risk disease.19 There is thus an opportunity to reinforce physician knowledge and confidence in nomograms, such as the updated Partin tables,31 when making decisions regarding imaging of patients with prostate cancer.

-risk levels, even when these test results are not sufficient to categorize the patient as having high-risk disease.19 There is thus an opportunity to reinforce physician knowledge and confidence in nomograms, such as the updated Partin tables,31 when making decisions regarding imaging of patients with prostate cancer. Among men with high-risk disease, we found that the likelihood of receiving guideline-concordant imaging increased with each calendar year. This finding is encouraging and suggests that health care professionals became more aware across the study period of the importance of imaging in accordance with guidelines to stage more advanced cases of prostate cancer. Because this pattern did not occur among men with low-risk prostate cancer, findings suggest that physician education efforts may need to be targeted specifically with regard to the patients with low-risk prostate cancer.

portance of imaging in accordance with guidelines to stage more advanced cases of prostate cancer. Because this pattern did not occur among men with low-risk prostate cancer, findings suggest that physician education efforts may need to be targeted specifically with regard to the patients with low-risk prostate cancer. There are numerous indications that guideline-discordant imaging among men with low-risk prostate cancer is still a salient problem. A 2017 systematic review identified 14 articles on overuse of imaging among patients with low-risk prostate cancer, representing data from 2004 to 2012.32 In the previous 10 years, there were several published efforts to curb guideline-discordant imaging, some of which have been successful, like the Michigan Urological Surgery Improvement Collaborative,33 and others which have been less so.34 Guideline-discordant staging imaging for low-risk prostate cancer continues to be regarded as a Choosing Wisely priority. This study provides important baseline data, highlights a prevalent practice pattern targeted for deimplementation, and may serve as a benchmark for more updated analyses to understand where quality improvement efforts may be successful and where they might be failing.

nues to be regarded as a Choosing Wisely priority. This study provides important baseline data, highlights a prevalent practice pattern targeted for deimplementation, and may serve as a benchmark for more updated analyses to understand where quality improvement efforts may be successful and where they might be failing. Strengths and Limitations Our study is strengthened by the large size of our study sample and the meticulous formation of each of the 3 delivery system cohorts. Because our study included men who received diagnosis and treatment using the VA and Medicare, which are 2 of the largest health care systems in the United States,35,36 our results are both reliable and generalizable. By creating a distinct cohort of men who used both the VA and Medicare, we were able to uniquely investigate the relative effectiveness of this commonly used approach to health care utilization.

e, which are 2 of the largest health care systems in the United States,35,36 our results are both reliable and generalizable. By creating a distinct cohort of men who used both the VA and Medicare, we were able to uniquely investigate the relative effectiveness of this commonly used approach to health care utilization. We conducted our assessment of imaging utilization solely through analyses of claims data, and thus there was an inherent possibility of misclassification bias. Situations in which imaging was warranted for a patient with low-risk prostate cancer because of clinical information that was not noted specifically in claims may have been misclassified as guideline discordant. In addition, both Medicare and the VA have reported possible problems with data reliability regarding clinical stage and PSA level.37 Incorrect clinical stage or PSA level could have caused patients to be classified in the incorrect risk group, although this type of broad misclassification is rare. It is also possible that health care professionals have dual appointments and treat patients with prostate cancer both at the VA and at an affiliated medical center. Based on an earlier qualitative study, individual health care professionals have expressed that they alter their imaging practices on the basis of health care setting.10 Because we did not study the breakdown of health care professionals with dual appointments among the cohorts, the effect of this practice would be unclear in our results and should be further studied.

health care professionals have expressed that they alter their imaging practices on the basis of health care setting.10 Because we did not study the breakdown of health care professionals with dual appointments among the cohorts, the effect of this practice would be unclear in our results and should be further studied. Our analysis data ended in 2008 and would benefit from an update to include more recent years. However, we carefully constructed these cohorts from multiple data sources and data sets, which is a time-consuming process with lags in availability, to uncover these important results. Veterans Health Administration regulatory requirements and new data server requirements to obtain updated data would require more than $50 000 in additional funding support just to obtain the data, before consideration of any analyst time, which is why we have not pursued establishing this data set as a repository. In addition, the approval timeline would be approximately 6 to 9 months, further impeding feasibility. The assembly of multiple data sources, even within the VA, requires complex data use agreements, which is one of the reasons that the analytic data set that we have assembled, although somewhat historical, is uniquely valuable.

n, the approval timeline would be approximately 6 to 9 months, further impeding feasibility. The assembly of multiple data sources, even within the VA, requires complex data use agreements, which is one of the reasons that the analytic data set that we have assembled, although somewhat historical, is uniquely valuable. Conclusions In this analysis, which merged 2 large prostate cancer cohorts (VACCR and SEER-Medicare), we found that veterans treated in a VA-only setting received the least amount of guideline-discordant imaging among patients with low-risk prostate cancer, without any significant difference in imaging utilization for patients with high-risk prostate cancer, for whom the imaging was necessary. In addition, although veterans who sought care through both the VA and Medicare would seem to be the most likely to experience the most imaging overuse because of the potential for fragmented care in 2 systems, patients seen in the Medicare-only setting had the most guideline-discordant imaging among patients with low-risk prostate cancer. These results reveal important differences between integrated and fee-for-service health systems regarding guideline concordance and quality of care. The results also suggest that patients using the Choice Act are likely to experience more utilization of care without a guarantee of improved quality of care. Future research to improve guideline-concordant care for prostate cancer imaging should consider and explore varying contexts and the role of unique settings of different health care systems. In addition, future studies should consider the cost implications of guideline-discordant imaging and the potential savings from an effort to align practice with evidence.

ealth care top state employers published a CSR report, and 261 of 500 (52%) on the Fortune 500,21 385 of 500 (77%) on the S&P 500,4 and 31 of 36 (86%) of non–health care top state employers reported CO2 emissions (Figure). Of the 12 top state employer nonprofit state university systems, 8 (66%) mentioned CO2 emissions. Figure. Sustainability Reporting by Health Care Sector Compared With All Sectors Sources are the 2016 Fortune 500, S&P 500, and Forbes 100 Charities; June 2016 largest employer in every state was compiled by 24/7 Wall St. Corporate social responsibility (CSR) data for companies on the 2016 S&P 500 list were obtained from the Governance & Accountability Institute, Inc and from the Carbon Disclosure Project. Data for companies on the 2016 Fortune 500 were compiled by CSRHub. Data from the the Global Reporting Initiative (GRI) database and corporate websites of all health care corporations were from the authors’ analysis. CO2 indicates carbon dioxide; S&P, Standard & Poor.

reduction diminished with time; the hazard ratio was 0.71 (95% CI, 0.57-0.87) in the first interval and 1.06 (95% CI, 0.77-1.46) in the third interval. In the matched lumpectomy cohort, radiotherapy was also associated with a significant reduction in contralateral invasive breast cancers (HR, 0.91; 95% CI, 0.85-0.97). Table 4. Hazard Ratios for Mortality From Breast Cancer Associated With Time Period (Time From Ductal Carcinoma In Situ Diagnosis) in Matched Patients Treated With Lumpectomy and Radiation vs Lumpectomy Alone Time Period, ya Comparison Hazard Ratio (95% CI) P Value 0-5.0 Lumpectomy alone 1 [Reference] .001 Lumpectomy plus radiotherapy 0.71 (0.57-0.87) 5.1-10.0 Lumpectomy alone 1 [Reference] .005 Lumpectomy plus radiotherapy 0.72 (0.58-0.91) 10.1-15.0 Lumpectomy alone 1 [Reference] .74 Lumpectomy plus radiotherapy 1.06 (0.77-1.46) a Global test for interaction not statistically significant (P = .31).

titute, Inc and from the Carbon Disclosure Project. Data for companies on the 2016 Fortune 500 were compiled by CSRHub. Data from the the Global Reporting Initiative (GRI) database and corporate websites of all health care corporations were from the authors’ analysis. CO2 indicates carbon dioxide; S&P, Standard & Poor. If no CSR report was found and sustainability activities were not mentioned on the main corporate website, a Google search was performed to determine if other health care corporate material mentioned sustainability terms. Results were as follows: 5 of 8 (63%) on the Fortune 500; 2 of 3 (66%) on the S&P 500; 8 of 14 (64%) of the largest state employers; 4 of 11 (36%) of the top state for-profit HCOs; and 8 of 17 (47%) of the top state nonprofit HCOs. Of the 12 top state employer nonprofit state university systems, 11 (92%) mentioned sustainability via Google search. Mention of at least 1 sustainability metric was identified for health care corporations as follows: 5 of 8 (63%) on the Fortune 500; 2 of 3 (66%) on the S&P 500; 4 of 14 (14%) of the largest state employers; 4 of 11 (36%) of the largest for-profit HCOs; and 5 of 17 (29%) of the largest nonprofit HCOs by number of facilities owned and operated.

tainability metric was identified for health care corporations as follows: 5 of 8 (63%) on the Fortune 500; 2 of 3 (66%) on the S&P 500; 4 of 14 (14%) of the largest state employers; 4 of 11 (36%) of the largest for-profit HCOs; and 5 of 17 (29%) of the largest nonprofit HCOs by number of facilities owned and operated. Search of the CSRHub database for reporting by Forbes 100 Largest Charities revealed few published CSR reports. Nevertheless, we analyzed all 8 health care corporations on the Forbes 100 Largest Charities list and found the following: 1 of 8 (13%) published a CSR report found in the GRI database; 1 of 8 (13%) mentioned a sustainability on the main website; 5 of 8 (63%) mentioned sustainability elsewhere via Google search; 3 of 8 (38%) reported at least 1 sustainability metric; and 1 of 8 (13%) mentioned CO2 emissions.

Introduction Over the past 2 decades many large corporations have begun to acknowledge that they must hold themselves responsible and take measures to reduce the environmental degradation, pollution, climate change, and social disruption that results from their activities. This represents a major transformation in the business landscape that challenges the traditional model that corporations are responsible only to their shareholders without regard to the social and environmental consequences of their actions. At the same time, investors and other stakeholders, such as governments and consumers, are increasingly demanding transparency from corporate leadership about the social and environmental impacts of business operations.

to their shareholders without regard to the social and environmental consequences of their actions. At the same time, investors and other stakeholders, such as governments and consumers, are increasingly demanding transparency from corporate leadership about the social and environmental impacts of business operations. To meet these demands many companies now publish annual reports detailing their efforts to measure, manage, and mitigate their environmental and social impacts. Commonly referred to in the business world as corporate social responsibility (CSR) reporting, these efforts are also known as sustainability reporting, nonfinancial reporting, integrated reporting, corporate citizenship reporting, triple bottom line reporting, and environmental, social, and governance reporting. Historically, these reports began as a platform to highlight corporate philanthropy but have evolved to include corporate performance on environmental measures (waste, water, and pollution) and on social impacts such as workforce well-being, diversity and equality practices, labor and management relations, human rights, and effects on local communities throughout the supply chain. For this article we refer to sustainability reporting to reference activities primarily in the environmental sphere and CSR to reflect a formal published report and/or activities that broaden efforts to include some degree of social impacts. Corporate social responsibility reports, which are currently voluntary in the United States and distinct from regulatory reporting or required US Securities and Exchange Commission corporate financial filings, have become the conventional format for companies to communicate performance on environmental, social, and governance issues and are now published by a majority of the largest publicly traded companies as well as many private and nonprofit companies. In 2016, 82% of S&P (Standard & Poor) 500 publicly traded companies published a CSR report compared with just 20% reported in 2011.1 In 2015, 90% of the world’s 250 largest companies published a CSR report.2 The Centre for Sustainability and Excellence, a US-based sustainability management consulting firm, analyzed 550 North American companies and found 79% published a CSR report in 2015 and 2016.3 In addition to preparing CSR reports, many large corporations signal their commitment to environmental sustainability by disclosing their carbon emissions to third-party accounting groups.

inability management consulting firm, analyzed 550 North American companies and found 79% published a CSR report in 2015 and 2016.3 In addition to preparing CSR reports, many large corporations signal their commitment to environmental sustainability by disclosing their carbon emissions to third-party accounting groups. In 2017, 70% of S&P 500 companies voluntarily reported their carbon emissions to the Carbon Disclosure Project, a UK-based nongovernmental organization operating a global disclosure system to help companies, cities, and governments report on and manage their carbon emissions; to date, more than 6000 companies from around the world have disclosed their carbon emissions.4 More recently, pressure to report sustainability and corporate responsibility activities intensified in 2017 when Vanguard, the largest provider of mutual funds, with $4.5 trillion in assets, notified publicly traded companies seeking support that it wants full disclosure of all sustainability risks. In early 2018, BlackRock, the world’s largest investor with $6.3 trillion under management, notified chief executive officers that it will no longer seek investments in companies that do not demonstrate CSR activities.

publicly traded companies seeking support that it wants full disclosure of all sustainability risks. In early 2018, BlackRock, the world’s largest investor with $6.3 trillion under management, notified chief executive officers that it will no longer seek investments in companies that do not demonstrate CSR activities. To standardize reporting across all economic sectors, a number of reporting frameworks have been developed and are now used by a majority of companies. One of the most common frameworks is promulgated by the Global Reporting Initiative (GRI), an independent standards organization that maintains a searchable database of CSR reports of companies from around the world. Despite increasing stakeholder pressure and the availability of frameworks, the sophistication, depth, transparency, and quality of sustainability and CSR reporting is still evolving. As such, the act of sharing sustainability information via websites or publishing a CSR report does not necessarily verify that a corporation is maximizing corporate citizenship efforts.

ailability of frameworks, the sophistication, depth, transparency, and quality of sustainability and CSR reporting is still evolving. As such, the act of sharing sustainability information via websites or publishing a CSR report does not necessarily verify that a corporation is maximizing corporate citizenship efforts. The health care industry is one of the largest and fastest growing economic sectors in the United States, representing 17.7% of the GDP in 2015 with projections to reach 20% by 2025.5 In 2017, for the first time, health care surpassed retail and manufacturing to become the largest source of jobs in the United States.6 Spurred by the passage of the Affordable Care Act and the move to value-based reimbursement models, mergers and acquisitions among health care organizations (HCOs) increased by 55% between 2010 and 2016,7 resulting in larger corporate entities and highly concentrated hospital markets in 90% of metropolitan areas.8 Health care organizations are now among the largest corporations in the United States and generate large revenues. For example, Hospital Corporation of America Holdings, the largest for-profit health system in the United States, had more than $44 billion in revenue in 2016 and ranked 63rd on the Fortune 500.9 Kaiser Permanente, the nation’s largest nonprofit health system, generated more than $64 billion in revenue in 2016.10 If the organization was eligible for the Fortune 500 list, it would rank 39th, ahead of MetLife, Pepsi, and Disney.

es, had more than $44 billion in revenue in 2016 and ranked 63rd on the Fortune 500.9 Kaiser Permanente, the nation’s largest nonprofit health system, generated more than $64 billion in revenue in 2016.10 If the organization was eligible for the Fortune 500 list, it would rank 39th, ahead of MetLife, Pepsi, and Disney. The US health care delivery industry consumes vast resources, the majority of which become waste—nearly 7000 tons of hospital waste is created daily.11 The health care delivery sector is responsible for 10% of all greenhouse gas emissions, 12% of acid rain, 10% of smog formation, and 9% of criteria air pollutants (ground-level ozone, particulate matter, carbon monoxide, lead, sulfur dioxide, and nitrogen dioxide), which leads to indirect health burdens commensurate with the 44 000 to 98 000 hospital deaths each year from preventable medical errors.12

ns, 12% of acid rain, 10% of smog formation, and 9% of criteria air pollutants (ground-level ozone, particulate matter, carbon monoxide, lead, sulfur dioxide, and nitrogen dioxide), which leads to indirect health burdens commensurate with the 44 000 to 98 000 hospital deaths each year from preventable medical errors.12 Corporate social responsibility activities have been shown to provide a positive return on investment; enhance employee recruitment, retention, productivity, and well-being; and create positive consumer sentiment.13 There is natural synergy between the mission of health care delivery, sustainability, and CSR activities. All seek to improve human well-being, the health care enterprise directly through the provision of medical care, sustainability by improving the environment, and CSR by including efforts to improve the social welfare of employees, consumers, and communities. The degree to which large health care corporations participate in the business trends to report on sustainability and/or CSR activities by publishing reports or by providing information via corporate websites is evaluated in this article.

g efforts to improve the social welfare of employees, consumers, and communities. The degree to which large health care corporations participate in the business trends to report on sustainability and/or CSR activities by publishing reports or by providing information via corporate websites is evaluated in this article. Methods Health care delivery companies were defined as companies that own and operate facilities providing direct patient care. Large health care delivery companies were defined by their inclusion on one of the following lists: 2016 Fortune 500,9 S&P 500,14 Forbes 100 Largest Charities,15 2015 largest for-profit16 and nonprofit17 health care systems by number of facilities complied by Becker’s Hospital Review, and June 2016 largest employer in every state compiled by 24/7 Wall St.18 This study follows the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.

2015 largest for-profit16 and nonprofit17 health care systems by number of facilities complied by Becker’s Hospital Review, and June 2016 largest employer in every state compiled by 24/7 Wall St.18 This study follows the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline. The GRI database19 was searched for a published CSR report for each of the large health care delivery corporations. Many large corporations with robust sustainability or CSR programs prominently feature their efforts on their main corporate website. Therefore, sustainability and/or CSR information included on the main webpage was considered a proxy for importance given these activities by corporate leadership. Each large health care corporation was evaluated through a search of the main corporate website landing page or drop-down menu on the landing page and through Google search with corporate name and at least 1 of the following terms: sustainability, corporate responsibility, social responsibility, citizenship, or environment. If no main website information was found and/or no CSR report was identified via the main website or GRI database, a Google search was performed to determine if corporate materials on basic sustainability activities (eg, waste, energy, recycling, reprocessing, and carbon dioxide [CO2] emissions) were available elsewhere. This was used as a proxy to signal some level of corporate awareness of basic sustainability activities but reflecting less of a corporate commitment than main website mention or the publication of a CSR report. A sustainability metric was considered to be mentioned if it included a quantity (pounds, gallons, British thermal units, etc), whether or not it was benchmarked or tracked over time. Mention of CO2 was considered to be “yes” if emissions were noted as total emissions for the institution or goals in reduction were described numerically (without total emissions noted or report to the Carbon Disclosure Project). Mention of at least 1 sustainability metric was considered to be a proxy to help distinguish between a narrative in corporate materials that conveys support for sustainability concepts vs some evidence of actual or potential measurement and management of waste generation and energy use.

the Carbon Disclosure Project). Mention of at least 1 sustainability metric was considered to be a proxy to help distinguish between a narrative in corporate materials that conveys support for sustainability concepts vs some evidence of actual or potential measurement and management of waste generation and energy use. We individually analyzed in detail only the health care corporations on each list of large US corporations, since data for total CSR reporting was readily available via published reports or through search of publicly available databases. The total percentage of CSR reporting by all corporations on the 2016 S&P 500 was compiled and publicly reported by the Governance & Accountability Institute, Inc.1 Reporting of CO2 emissions by all corporations on the S&P 500 was compiled and published by the Carbon Disclosure Project.20 The total percentage of sustainability reporting and CO2 emissions disclosure by corporations on the 2016 Fortune 500 and 2016 Forbes 100 Largest Charities was compiled by CSRHub, the largest database of corporate social responsibility, environmental, and governance ratings and information worldwide.21 Reporting of CSR in these databases included GRI as well as other reporting formats. For each list, percentages of health care corporations reporting sustainability activities ascertained by our analysis were compared with the total percentage of sustainability reporting for all corporations on each list as compiled by the mentioned firms.

hese databases included GRI as well as other reporting formats. For each list, percentages of health care corporations reporting sustainability activities ascertained by our analysis were compared with the total percentage of sustainability reporting for all corporations on each list as compiled by the mentioned firms. Results A total of 49 health care corporations were analyzed (10 appeared on >1 list but were analyzed only once). In 2016, 8 health care corporations appeared on the Fortune 500, 3 on the S&P 500, and 8 on the Forbes 100 Largest Charities. In 2016, HCOs were the top employers in 14 states and in 2015 a total of 28 for-profit and nonprofit health systems were listed as the largest by number of facilities owned and operated. Because of ties in the number of facilities owned and operated, 11 top 10 largest for-profit HCOs and 17 top 10 largest nonprofit HCOs were included (Table). Table. Demographics of HCOs Analyzed List No. of HCOs Total For-Profit Nonprofit Northeast Southeast Midwest Southwest West Fortune 500 8 8 0 1 4 0 1 2 S&P 500 3 3 0 1 1 0 0 1 Forbes 100 Largest Charities 8 0 8 4 2 2 0 0 10 Largest for-profit HCOs 11 11 0 2 6 1 0 2 10 Largest nonprofit HCOs 17 0 17 2 3 2 3 7 Largest state employers 14 7 7 7 0 4 0 3 Abbreviations: HCO, health care organization; S&P, Standard & Poor.

st Midwest Southwest West Fortune 500 8 8 0 1 4 0 1 2 S&P 500 3 3 0 1 1 0 0 1 Forbes 100 Largest Charities 8 0 8 4 2 2 0 0 10 Largest for-profit HCOs 11 11 0 2 6 1 0 2 10 Largest nonprofit HCOs 17 0 17 2 3 2 3 7 Largest state employers 14 7 7 7 0 4 0 3 Abbreviations: HCO, health care organization; S&P, Standard & Poor. Of the 49 HCOs analyzed, 6 (12%) published a CSR report as determined by a search of the GRI database: 4 of 8 (50%) on the Fortune 500; 1 of 3 (33%) on the S&P 500; 1 of 8 (13%) on the Forbes 100 Largest Charities; 2 of 11 (18%) of the largest for-profit HCOs; and 1 of 17 (6%) of the largest nonprofit HCOs by number of facilities owned and operated (some companies appeared on >1 list). No health care corporation on the largest state employer list published a CSR report. Mention of sustainability terms were found on the main corporate website for health care corporations as follows: 2 of 8 (25%) on the Fortune 500; 1 of 3 (33%) on the S&P 500; 5 of 14 (36%) of the largest state employers; 1 of 11 (9%) of the largest for-profit HCOs; and 5 of 17 (29%) of the largest nonprofit HCOs by number of facilities owned and operated. Mention of CO2 emissions by health care corporations were as follows: 1 of 8 (13%) on the Fortune 500; 2 of 14 (14%) of the health care top state employers; 1 of the 11 (9%) of the largest for-profit HCOs; and 5 of 17 (29%) of the largest nonprofit HCOs by facilities owned and operated. No S&P 500 health care company mentioned CO2 emissions.

s by health care corporations were as follows: 1 of 8 (13%) on the Fortune 500; 2 of 14 (14%) of the health care top state employers; 1 of the 11 (9%) of the largest for-profit HCOs; and 5 of 17 (29%) of the largest nonprofit HCOs by facilities owned and operated. No S&P 500 health care company mentioned CO2 emissions. Our findings for health care corporations on the Fortune 500, S&P 500, and largest state employers were compared with the data for all corporations available from published reports and publicly available databases. Among all corporations, 389 of 500 (78%) on the Fortune 500,21 410 of 500 (82%) on the S&P 500,1 and 24 of 36 (67%) of non–health care top state employers published a CSR report, and 261 of 500 (52%) on the Fortune 500,21 385 of 500 (77%) on the S&P 500,4 and 31 of 36 (86%) of non–health care top state employers reported CO2 emissions (Figure). Of the 12 top state employer nonprofit state university systems, 8 (66%) mentioned CO2 emissions.

s list and found the following: 1 of 8 (13%) published a CSR report found in the GRI database; 1 of 8 (13%) mentioned a sustainability on the main website; 5 of 8 (63%) mentioned sustainability elsewhere via Google search; 3 of 8 (38%) reported at least 1 sustainability metric; and 1 of 8 (13%) mentioned CO2 emissions. Discussion To our knowledge, this is the first assessment of sustainability reporting by large corporations in the health care delivery sector. Our analysis finds the health care delivery sector lags far behind other economic sectors in communicating sustainability activities. In 2016, 389 of 500 (78%) of Fortune 500 companies and 410 of 500 (82%) of S&P 500 companies reported CSR activities compared with 4 of 8 (50%) Fortune 500, 1 of 3 (33%) S&P 500, and 6 of 49 (12%) of all health care corporations appearing on any list. If mention of sustainability terms on the main corporate website is a proxy for corporate commitment to these activities, then health care corporations lag; only 2 of 8 (25%) of Fortune 500 companies; 1 of 3 (33%) of S&P 500 companies, 1 of 8 (13%) of Forbes 100 Largest Charities, and 5 of 14 (36%) of top state employers mentioned sustainability or CSR terms on their main websites. Furthermore, 77% of all S&P 500 and 52% of all Fortune 500 companies disclosed carbon emission information to the Carbon Disclosure Project, but only 1 of 8 (13%) of Fortune 500 health care companies, 1 of 8 (13%) of Forbes 100 Largest Charities, 2 of 14 (14%) of the largest state employer health care systems, and no S&P 500 health care companies mentioned carbon emissions in corporate material.

carbon emission information to the Carbon Disclosure Project, but only 1 of 8 (13%) of Fortune 500 health care companies, 1 of 8 (13%) of Forbes 100 Largest Charities, 2 of 14 (14%) of the largest state employer health care systems, and no S&P 500 health care companies mentioned carbon emissions in corporate material. While overall performance in sustainability reporting is poor compared with other economic sectors, there are indications that these activities have some importance to health care corporations. A greater number of health care corporations were found to mention basic sustainability information when searched via Google: 2 of 3 (66%) of S&P 500, 5 of 8 (63%) of Fortune 500, and 8 of 14 (64%) of the largest state employer HCOs. Most also included at least 1 metric, which provides some evidence of commitment to sustainability; however, without publishing a CSR report or including mention of these efforts on the main website, it is difficult to determine to what degree HCOs truly value and pursue these activities.

e largest state employer HCOs. Most also included at least 1 metric, which provides some evidence of commitment to sustainability; however, without publishing a CSR report or including mention of these efforts on the main website, it is difficult to determine to what degree HCOs truly value and pursue these activities. Perhaps the starkest contrast in reporting is seen in the top state employers. In 2015 Walmart was the largest employer in 19 states; Boeing, Intel, MGM Grand, Hannaford Supermarkets, and Wakefern Foods were the largest in 1 state each. All of these large corporations have robust sustainability programs, include CSR on their websites, and publish CSR reports, and all but 1 reported CO2 emissions. By contrast, only 5 of 14 (36%) of top state employer HCOs included sustainability information on their main website, only 1 mentioned CO2 emissions, and none had CSR reports searchable via GRI.

robust sustainability programs, include CSR on their websites, and publish CSR reports, and all but 1 reported CO2 emissions. By contrast, only 5 of 14 (36%) of top state employer HCOs included sustainability information on their main website, only 1 mentioned CO2 emissions, and none had CSR reports searchable via GRI. It could be argued that private and charitable corporations are exempt from sustainability or more robust CSR activities because they do not face shareholder pressure. However, many large private companies do participate in some form of CSR. For example, of the 5 largest privately held corporations in the United States—Cargill, Koch, Mars, PricewaterhouseCoopers, and Bechtel—only Koch Industries does not publish any form of sustainability information. Furthermore, nonprofit state university systems—the top employers in 12 states—are more likely to mention sustainability via a Google search (11 of 12 [92%]) and mention CO2 emissions (8 of 12 [66%]) than top state employer HCOs. It is noteworthy that the largest nonprofit HCOs by number of facilities owned are in fact more likely than the largest for-profit HCOs to address sustainability and report CO2 emissions (5 of 17 [29%] vs 1 of 11 [9%]) for both measures.

12 [92%]) and mention CO2 emissions (8 of 12 [66%]) than top state employer HCOs. It is noteworthy that the largest nonprofit HCOs by number of facilities owned are in fact more likely than the largest for-profit HCOs to address sustainability and report CO2 emissions (5 of 17 [29%] vs 1 of 11 [9%]) for both measures. Efforts of CSR can yield a large positive return on investment. An examination of the potential cost savings that could be realized from application of best practices in energy and waste management for the US health care system estimated a return on investment of $5.4 billion over 5 years and $15 billion over 10 years.22 Reduction of pollution and greenhouse gas emissions would also provide a cobenefit for human health by reducing disease burden and medical costs, all of which directly support the mission of health care.

US health care system estimated a return on investment of $5.4 billion over 5 years and $15 billion over 10 years.22 Reduction of pollution and greenhouse gas emissions would also provide a cobenefit for human health by reducing disease burden and medical costs, all of which directly support the mission of health care. Limitations Limitations of this study include the possibility that large health care corporations are actively involved in sustainability or CSR activities but do not communicate them publicly via websites or published reports. Additionally, we searched only the GRI database for a published report and no other reporting frameworks, therefore possibly underestimating reporting by health care corporations. Additionally, we directly assessed only health care corporations and not all corporations, relying instead on published reports and databases for overall reporting. Therefore, we may have overestimated or underestimated reporting by all corporations on each list. In the business world, many companies with robust CSR programs feature their reports prominently on the main corporate website. In our analysis we made the assumption that if it is not on the main website, it is of lower priority to corporate leadership. It was our judgment that inclusion on the main corporate website is a better indicator of the importance given to sustainability or CSR activities by corporate leadership whether or not a formal CSR report was identified; however, we did not test this assumption directly. Furthermore, we did not directly survey any corporation to ascertain attitudes, actual sustainability, and CSR activities as opposed to what they are reporting. We may give a false impression of inactivity or lack of corporate commitment to sustainability and CSR on the part of the health care delivery sector. In addition, the rapid pace of corporate mergers in the health care delivery sector, spurred by the transition to population-based medicine, may mean corporate leadership has not had time to gather and publish relevant sustainability or CSR information. It is also important to note that we focused primarily on environmental metrics for our analysis and not on broader social impacts (eg, employee well-being and human rights) that are also key components of CSR. Therefore, it is possible that we have underestimated efforts on social measures by health care corporations.

It is also important to note that we focused primarily on environmental metrics for our analysis and not on broader social impacts (eg, employee well-being and human rights) that are also key components of CSR. Therefore, it is possible that we have underestimated efforts on social measures by health care corporations. Additionally, we looked only at large economic entities and not midsize or small corporations and our analysis may not accurately reflect the totality of efforts by the health care delivery sector since many smaller entities may be engaged in and reporting on sustainability and/or CSR activities. Finally, this report is limited by the fact it is a descriptive assessment with no statistical analysis. Conclusions This study did not explore why health care corporations have lagged behind corporations in other economic sectors in adopting sustainability reporting. Further studies should evaluate to what degree this is associated with challenges such as slim operating margins, regulatory burdens, political uncertainty, evolving payment models, lack of familiarity with CSR, misperceptions about cost-effectiveness, and a sentiment that charity status and the already substantial societal benefit of providing medical care exempt HCOs from CSR. Our analysis provides no information about the actual quality, depth, and impact of sustainability and/or CSR activities in any economic sector.

CSR, misperceptions about cost-effectiveness, and a sentiment that charity status and the already substantial societal benefit of providing medical care exempt HCOs from CSR. Our analysis provides no information about the actual quality, depth, and impact of sustainability and/or CSR activities in any economic sector. Despite the overall paucity of sustainability reporting in the health care sector, efforts to improve the footprint of the health care industry have been under way for years championed by programs such as the American Hospital Association’s Sustainability Roadmap for Hospitals, Healthcare Without Harm, and Practice Greenhealth. Numerous case studies have been published of health facilities that have reduced costs while improving their environmental footprint. Many health systems are innovating ways to build sustainable, LEED (Leadership in Energy and Environmental Design)–certified facilities (a rating system for environmentally responsible construction practices)23 and to incorporate wellness design aimed at improving recovery time through natural light, views, and ventilation. New York City hospitals have signed on to an ambitious pledge from the mayor to reduce carbon emissions by at least 40% by 2030. A few early CSR adopters such as Kaiser Permanente and Cleveland Clinic have kept pace with the business world, and Cleveland Clinic publishes a formal CSR report annually.

, and ventilation. New York City hospitals have signed on to an ambitious pledge from the mayor to reduce carbon emissions by at least 40% by 2030. A few early CSR adopters such as Kaiser Permanente and Cleveland Clinic have kept pace with the business world, and Cleveland Clinic publishes a formal CSR report annually. Despite these exemplars, the business world has substantially surpassed the health care delivery industry in responding to the demand for greater accountability for the environmental and social impacts of the enterprise by creating frameworks and metrics for measuring and reporting on sustainability and social responsibility. It is a business maxim that “you can’t manage what you don’t measure” and these readily available frameworks would help align the business of health care with its mission and create uniformity in reporting, providing greater accountability and better information for all stakeholders. The health care delivery sector has the opportunity to lead in sustainability and CSR—no other industry has a mission that approximates so closely the ultimate goal of all CSR activities—the protection of the environmental and social systems that protect and enhance human health and well-being. Furthermore, greater participation by the health care sector in sustainability efforts is consistent with the intent of the United Nations (UN) Global Compact and will advance attainment of the UN Sustainable Development Goals. The UN Global Compact, also called the world’s largest corporate sustainability initiative, is an effort to enlist corporations worldwide to align their business strategies and operations with universal principles of human rights, labor, environment, and anticorruption. To date, more than 9500 companies from 160 countries in the developed and developing world are signatories to the UN Global Compact. A strong argument could be made that health care delivery corporations should align their operations with these goals. Corporate social responsibility continues to evolve but it is not likely to be a passing fad. It is safe to say that no economic sector will be exempted from the expectation for good corporate citizenship; in this regard, health care corporations are part of the problem but critical to the solution.

Introduction Clinical trials require time to generate and disseminate new knowledge. The time lag is subject to fixed constraints, such as the follow-up period for the primary end point, and modifiable factors, such as participant enrollment time and time to publication after completion of data collection. Furthermore, time to publication could be affected by the time needed for data entry, adjudication, cleaning, analysis and interpretation, manuscript preparation, peer reviews, and actual publication by the journal after acceptance. Some of these tasks could be conducted, in large part, in parallel with the conduct of the trial. As medical knowledge and clinical practice rapidly evolve,1 the faster the variable aspects of a trial are accomplished, the more relevant the results are to current practice.2,3 There is an imperative to publish clinical trial results in a timely way and reduce unnecessary delays. Previous studies of published clinical trials showed that it took, on average, 2 years for these trials to be published after completion (ie, time to publication).4,5 Little is known about the overall age of data at publication, the contribution of the time to publication to the data age, or the time spent enrolling participants. Such information might identify opportunities to accelerate the timeliness of the clinical trial process and reporting. Accordingly, we sought to characterize data age, enrollment time, and publication time of all clinical trials published in 2015 by the medical journals with the highest impact factors.

g participants. Such information might identify opportunities to accelerate the timeliness of the clinical trial process and reporting. Accordingly, we sought to characterize data age, enrollment time, and publication time of all clinical trials published in 2015 by the medical journals with the highest impact factors. Methods Data Collection We screened all original research articles published (either online or in print) from January 1 through December 31, 2015, in 6 general and internal medicine journals with high impact factors6: the New England Journal of Medicine (NEJM), Lancet, JAMA, BMJ, Annals of Internal Medicine, and JAMA Internal Medicine. We identified only trials with a randomized comparison of an intervention with a control and excluded those that did not represent a primary analysis or that did not report the time of starting enrollment or ending data collection in a specific month (eFigure 1 in the Supplement). Approval of this study was waived by the Yale University institutional review board as it did not meet the definition of human participants research. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines. All data elements were independently abstracted by 1 of us (J.W., L.B., C.O.Z., or L.M.) and were then checked for accuracy by a different investigator from our group. For 335 of the 341 data elements (98.2%), the results of both abstractions were in agreement. All discrepancies were resolved through discussion with a third investigator (J.W., J.S.R., or H.M.K.).

ed by 1 of us (J.W., L.B., C.O.Z., or L.M.) and were then checked for accuracy by a different investigator from our group. For 335 of the 341 data elements (98.2%), the results of both abstractions were in agreement. All discrepancies were resolved through discussion with a third investigator (J.W., J.S.R., or H.M.K.). Outcome Measures The outcome measures of interest were the midpoint of data collection until publication (data age), the time from first participant enrollment to last participant enrollment (enrollment time), and the time from final data collection to publication (publication time). We defined the data collection period as the start of enrollment to the end of follow-up. The definition of data age was determined as such to convey the mean age of the entire sample. The first patient enrolled in the trial could be viewed as the age of the data, as could the last patient follow-up collected. However, using these markers would not capture the sample as well as the midpoint of these 2 time points (Figure 1). We extracted the start and end dates of enrollment, as well as the final data collection date from the main text of each article. When those dates were not available, we checked the trial protocol or appendix and, if necessary, its registration on ClinicalTrials.gov. For trials that provided only a month as a start or end date, we assumed that the start date was the first day of that month and that the end date was the last day of that month. Only 27 of 341 trials (7.9%) were missing the start or end date, and the maximum that a trial duration could have been overestimated would be by 2 months. We extracted the publication date based on the e-publication ahead of print date when it was listed; otherwise, it was based on the print publication date.

at month. Only 27 of 341 trials (7.9%) were missing the start or end date, and the maximum that a trial duration could have been overestimated would be by 2 months. We extracted the publication date based on the e-publication ahead of print date when it was listed; otherwise, it was based on the print publication date. Figure 1. A Mock Trial With Different Time Periods Labeled Progression of a mock trial defining data age, data collection period, and publication time. The black lines indicate an individual patient’s own timeline throughout the study.

at month. Only 27 of 341 trials (7.9%) were missing the start or end date, and the maximum that a trial duration could have been overestimated would be by 2 months. We extracted the publication date based on the e-publication ahead of print date when it was listed; otherwise, it was based on the print publication date. Figure 1. A Mock Trial With Different Time Periods Labeled Progression of a mock trial defining data age, data collection period, and publication time. The black lines indicate an individual patient’s own timeline throughout the study. Independent Variables Our independent variables included important trial features to characterize our sample and features that may be associated with data age at publication. Specifically, we collected data on the type of intervention (drug, device, or other), early study termination for any reason, number of patients enrolled, trial location (United States only, United States and outside of United States, or outside of United States only), number of trial centers, number of manuscript authors, author affiliation (government or private industry), funding source (government, nonprofit, private industry, or combination), whether the trial was registered on ClinicalTrials.gov or registered on another website, and whether results were posted on ClinicalTrials.gov. We also determined the favorability of the findings by designating trials with primary end point results that yielded statistically significant better outcomes for the treatment population compared with the control population as “favorable,” statistically significant worse outcome for the treatment population compared with the control population as “unfavorable,” and all others as “inconclusive.”

primary end point results that yielded statistically significant better outcomes for the treatment population compared with the control population as “favorable,” statistically significant worse outcome for the treatment population compared with the control population as “unfavorable,” and all others as “inconclusive.” Statistical Analysis We calculated the median values and interquartile ranges (IQRs) for continuous variables because we presumed the distributions of these variables were not normal; we calculated the total counts and percentages for categorical variables. We reported the data age, enrollment time, and publication time, overall and by follow-up duration (<1 month, 1 month to 1 year, and >1 year). We conducted bivariate analysis to test whether data age, enrollment time, and publication time were associated with each of the following variables: type of intervention, early study termination for any reason, number of patients enrolled, trial location, number of trial centers, number of manuscript authors, author affiliation, funding source, favorability of the findings, trial registration on ClinicalTrials.gov, and results posted on ClinicalTrials.gov. To identify trial characteristics associated with data age, enrollment time, and publication time, we further developed multivariable linear regression models. We adjusted for variables that reached statistical significance at 2-sided P < .05 in bivariate analyses. Because longer follow-up duration was associated with older data age of the trials, we also adjusted for follow-up time in the model to account for the different follow-up times in these studies. In a post hoc analysis, we characterized the 10 studies with the longest data age and the 10 studies with the shortest data age. All data were analyzed with R, version 3.3.2 (The R Foundation for Statistical Computing).

usted for follow-up time in the model to account for the different follow-up times in these studies. In a post hoc analysis, we characterized the 10 studies with the longest data age and the 10 studies with the shortest data age. All data were analyzed with R, version 3.3.2 (The R Foundation for Statistical Computing). Results Our search identified 979 original research articles, of which 566 were excluded because they were not randomized clinical trials, 23 because they did not represent a primary analysis, 46 because they had missing data on outcomes of interest, and 3 for other reasons; thus, our final analysis included 341 trials (eFigure 1 in the Supplement). The 341 trials assessed drugs (206 [60.4%]), devices (21 [6.2%]), and other interventions (114 [33.4%]) (Table 1). Among these trials, 37 (10.9%) had a follow-up period of less than 1 month, 172 (50.4%) had a follow-up period between 1 month and 1 year, and 132 (38.7%) had a follow-up period of more than 1 year. The median number of enrollees was 467 (IQR, 212-1260), the median number of trial centers was 23 (IQR, 6-62), and the median number of authors was 16 (IQR, 11-22).

-up period of less than 1 month, 172 (50.4%) had a follow-up period between 1 month and 1 year, and 132 (38.7%) had a follow-up period of more than 1 year. The median number of enrollees was 467 (IQR, 212-1260), the median number of trial centers was 23 (IQR, 6-62), and the median number of authors was 16 (IQR, 11-22). Table 1. Characteristics for Randomized Trials Published in 2015 in 6 Journals With High Impact Factors Characteristic Trials, No. (%) (N = 341) Included trial articles by journal Annals of Internal Medicine 17 (5.0) BMJ 24 (7.0) JAMA 58 (17.0) JAMA Internal Medicine 17 (5.0) New England Journal of Medicine 124 (36.4) Lancet 101 (29.6) Enrolled patients, median (IQR), No. 467 (212-1260) Trial type Drug 206 (60.4) Device 21 (6.2) Other 114 (33.4) No. of trial centers, median (IQR) 23 (6-62) Trial location United States only 70 (20.5) United States and outside United States 104 (30.5) Outside United States only 167 (49.0) No. of authors, median (IQR) 16 (11-22) Manuscripts with ≥1 author primarily employed by private industry, No./total No. (%) ≥1 124/340 (36.5) 0 216/340 (63.5) Early stoppage of trial Yes 21 (6.2) No 320 (93.8) Trial registration and results reporting on ClinicalTrials.gov Yes 100 (29.3) No 179 (52.5) Registered on a site other than ClinicalTrials.gov 60 (17.6) Unregistered 2 (0.6) Favorability of findings for the treatment population relative to the control population Favorable 231 (67.7) Unfavorable 12 (3.5) Inconclusive 96 (28.2) NAa 2 (0.6) Funding source Government 112 (32.8) Nonprofit 35 (10.3) Private industry 108 (31.7) Government, nonprofit 36 (10.6) Government, private industry 17 (5.0) Nonprofit, private industry 10 (2.9) Government, nonprofit, private industry 20 (5.9) Not disclosed 2 (0.6) None 1 (0.3) Abbreviations: IQR, interquartile range; NA, not applicable.

overnment 112 (32.8) Nonprofit 35 (10.3) Private industry 108 (31.7) Government, nonprofit 36 (10.6) Government, private industry 17 (5.0) Nonprofit, private industry 10 (2.9) Government, nonprofit, private industry 20 (5.9) Not disclosed 2 (0.6) None 1 (0.3) Abbreviations: IQR, interquartile range; NA, not applicable. a Two trials had more than 1 primary outcome and had mixed findings. Data Age, Enrollment Time, and Publication Time The median data age (midpoint of data collection to publication) was 33.9 months (IQR, 23.5-46.3 months; range [minimum to maximum], 2.2-131.8 months). A total of 88 trials (25.8%) reported data age of less than 2 years; 31 trials (9.1%) reported data age of 5 years or more (Figure 2 and eFigure 2 in the Supplement). The median data age was 30.6 months (IQR, 18.6-39.0 months) for trials with a follow-up period of 1 month, 31.8 months (IQR, 21.0-41.7 months) for trials with a follow-up period of 1 month to 1 year, and 40.1 months (IQR, 30.3-51.9 months) for trials with a follow-up period of more than 1 year. Figure 2. Distributions of Data Age, Enrollment Time, and Publication Time The ends of the boxes indicate the upper and lower quartiles, so the box spans the interquartile range. The middle line indicates the median, the whiskers are the 2 lines outside the box that extend to the highest and lowest observations, and the circles indicate the extreme values of the observations.

Publication Time The ends of the boxes indicate the upper and lower quartiles, so the box spans the interquartile range. The middle line indicates the median, the whiskers are the 2 lines outside the box that extend to the highest and lowest observations, and the circles indicate the extreme values of the observations. The median enrollment time was 26.2 months (IQR, 14.2-42.3 months; range, 0.3-141.1 months), and the median mean enrollment time per person across trials was 1.4 days (IQR, 0.5-3.8 days; range 0.0002-69.0 days); 64 of 313 trials (20.4%) completed enrollment within 1 year; 251 of 313 trials (80.2%) completed enrollment within 4 years; and 68 trials (19.9%) required more than 4 years to complete enrollment. A total of 257 of 313 trials (82.1%) required fewer than 5 days per enrolled participant; 32 of 313 trials (10.2%) required 9 days or more (Figure 2 and eFigure 3 in the Supplement). The median time to publication was 14.8 months (IQR, 7.4-22.2 months; range, 0.5-90.3 months) (Figure 2). A total of 138 trials (40.5%) were published within 1 year after completing the final data collection, and 63 trials (18.5%) were published more than 2 years after completing the final data collection (eFigure 4 in the Supplement). Overall, 43.2% (IQR, 26.8%-61.6%) of the data age was the time to publication.

igure 2). A total of 138 trials (40.5%) were published within 1 year after completing the final data collection, and 63 trials (18.5%) were published more than 2 years after completing the final data collection (eFigure 4 in the Supplement). Overall, 43.2% (IQR, 26.8%-61.6%) of the data age was the time to publication. Factors Associated With Data Age, Enrollment Time, and Publication Time In multivariable analyses, some factors were associated with older data age, adjusted for follow-up time (Table 2). Compared with favorable trials, inconclusive or unfavorable trials had a median data age that was 235 days longer (95% CI, 108-362 days). Each additional day of follow-up duration was also associated with an additional 0.6 days (95% CI, 0.5-0.8) of data age.

d with older data age, adjusted for follow-up time (Table 2). Compared with favorable trials, inconclusive or unfavorable trials had a median data age that was 235 days longer (95% CI, 108-362 days). Each additional day of follow-up duration was also associated with an additional 0.6 days (95% CI, 0.5-0.8) of data age. Table 2. Factors Associated With Data Age (in Days) in Bivariate and Multivariable Analyses Factor Coefficient (95% CI) Bivariate Analysis Multivariable Analysisa Trial type Drug 1 [Reference] NA Device −22 (−238 to 282) NA Other 91 (−42 to 223) NA No. of enrolled patients (per 1000) 0.1 (−1.3 to 1.5) NA No. of trial centers 0.1 (−0.5 to 0.6) NA No. of authors −3.1 (−9.3 to 3.1) NA Trial location United States only 1 [Reference] NA Outside of United States only 86 (−75 to 246) NA United States and outside of United States −116 (−290 to 58) NA Manuscripts with ≥1 author primarily employed by private industry No 1 [Reference] NA Yes −276 (−401 to −151)b −182 (−376 to 13) Early stoppage of trial No 1 [Reference] NA Yes −47 (−303 to 210) NA Trial registration and results reported at ClinicalTrials.gov Yes 1 [Reference] NA No or unregistered on any site −30 (−173 to 112) NA Registered on a site other than ClinicalTrials.gov 15 (−171 to 200) NA Favorability of findings for the treatment population relative to the control population Favorable 1 [Reference] NA Unfavorable or inconclusiveb 307 (179 to 436) 235 (108 to 362) Funding source Government 1 [Reference] NA Nonprofit 15 (−199 to 229) 76 (−119 to 271) Private industry −266 (−415 to −117)b −74 (−297 to 149) Government and nonprofit −0.9 (−213 to 211) 46 (−155 to 246) Private industry and others (government or nonprofit) 267 (30 to 504)b 191 (−43 to 425) Government, nonprofit, and private industry −17 (−251 to 286) 219 (−49 to 487) Follow-up duration (per day)b 0.6 (0.4 to 0.7) 0.6 (0.5 to 0.8) Abbreviation: NA, not applicable.

nment and nonprofit −0.9 (−213 to 211) 46 (−155 to 246) Private industry and others (government or nonprofit) 267 (30 to 504)b 191 (−43 to 425) Government, nonprofit, and private industry −17 (−251 to 286) 219 (−49 to 487) Follow-up duration (per day)b 0.6 (0.4 to 0.7) 0.6 (0.5 to 0.8) Abbreviation: NA, not applicable. a We selected variables that were significant in the bivariate analysis and included them in the multivariable analysis. We also adjusted for follow-up time in the multivariable analysis to account for the different follow-up time in these studies. b Estimates were statistically significant at P < .05. We also found several characteristics associated with significantly longer enrollment time (Table 3). Specifically, trials that had no authors affiliated with private industry were associated with a longer enrollment time (by 566 days; 95% CI, 306-827 days) than those with at least 1 author affiliated with industry. Compared with trials that had only government funding, trials that had funding from both private industry and government or from both private industry and nonprofit agencies (by 460 days; 95% CI, 152-768 days) or all 3 sources together (by 676 days; 95% CI, 322-1031 days) were also associated with a longer enrollment time. Compared with trials that were funded only by private industry, trials that were only government funded were associated with an additional 180 days (95% CI, 18-343 days) to publish (Table 4).

152-768 days) or all 3 sources together (by 676 days; 95% CI, 322-1031 days) were also associated with a longer enrollment time. Compared with trials that were funded only by private industry, trials that were only government funded were associated with an additional 180 days (95% CI, 18-343 days) to publish (Table 4). Table 3. Factors Associated With Enrollment Time (in Days) in Bivariate and Multivariable Analyses Factor Coefficient (95% CI) Bivariate Analysis Multivariable Analysisa Trial type Drug 1 [Reference] NA Device 170 (−155 to 496) NA Other 63 (−108 to 235) NA No. of enrolled patients (per 1000) −1.1 (−1.1 to 0.9) NA No. of trial centers 0.4 (−0.3 to 1.0) NA No. of authors 2.1 (−5.9 to 10) NA Trial location United States only 1 [Reference] NA Outside of United States only 148 (−62 to 358) NA United States and outside of United States −91 (−320 to 138) NA Manuscripts with ≥1 author primarily employed by private industry No 1 [Reference] NA Yesb −457 (−618 to −297) −566 (−827 to −306) Early stoppage of trial No 1 [Reference] NA Yes 35 (−285 to 355) NA Trial registration and results reported at ClinicalTrials.gov Yes 1 [Reference] NA No or unregistered on any site 71 (−113 to 255) NA Registered on a site other than ClinicalTrials.gov 3.1 (−236 to 242) NA Funding source Government 1 [Reference] NA Nonprofit 36 (−99 to 171) −23 (−283 to 236) Private industry −157 (−251 to −63)b 156 (−136 to 448) Government and nonprofit −67 (−200 to 67) 129 (−138 to 395) Private industry and others (government or nonprofit) −17 (−166 to 132) 460 (152 to 768)b Government, nonprofit, and private industry −175 (−344 to −6) 676 (322 to 1031)b Abbreviation: NA, not applicable.

Private industry −157 (−251 to −63)b 156 (−136 to 448) Government and nonprofit −67 (−200 to 67) 129 (−138 to 395) Private industry and others (government or nonprofit) −17 (−166 to 132) 460 (152 to 768)b Government, nonprofit, and private industry −175 (−344 to −6) 676 (322 to 1031)b Abbreviation: NA, not applicable. a Only variables significant in the bivariate analysis were included in the multivariable analysis. b Estimates were statistically significant at P < .05.

Private industry −157 (−251 to −63)b 156 (−136 to 448) Government and nonprofit −67 (−200 to 67) 129 (−138 to 395) Private industry and others (government or nonprofit) −17 (−166 to 132) 460 (152 to 768)b Government, nonprofit, and private industry −175 (−344 to −6) 676 (322 to 1031)b Abbreviation: NA, not applicable. a Only variables significant in the bivariate analysis were included in the multivariable analysis. b Estimates were statistically significant at P < .05. Table 4. Factors Associated With Publication Time (in Days) in Bivariate and Multivariable Analyses Factor Coefficient (95% CI) Bivariate Analysis Multivariate Analysisa Trial type Drug 1 [Reference] NA Device −22 (−183 to 139) NA Other 76 (−5.9 to 158) NA No. of enrolled patients (per 1000) 0.6 (−0.3 to 1.4) NA No. of trial centersb −0.5 (−0.8 to −0.2) −0.4 (−0.7 to −0.1) No. of authorsb −6.6 (−10.4 to −2.8) −5.9 (−9.9 to −2.0) Trial location United States only 1 [Reference] NA Outside of United States only 11 (−88 to 111) 62.3 (−42 to 167) United States and outside of United States −123 (−231 to −16)b 30 (−104 to 164) Manuscripts with ≥1 author primarily employed by private industry No 1 [Reference] NA Yes −88 (−167 to −9)b 98 (−35 to 231) Early stoppage of trial No 1 [Reference] NA Yes −115 (−274 to 44) NA Trial registration and results reported at ClinicalTrials.gov Yes 1 [Reference] NA No or unregistered on any site −19 (−107 to 69) NA Registered on a site other than ClinicalTrials.gov 70 (−45 to 185) NA Favorability of findings for the treatment population relative to the control population Favorable 1 [Reference] NA Unfavorable or inconclusive 93 (11 to 175)b 38 (−51 to 128) Funding source Government 1 [Reference] NA Nonprofit 36 (−98 to 171) −4.5 (−145 to 136) Private industryb −157 (−251 to −63) −180 (−343 to −18) Government and nonprofit −67 (−200 to 67) −63 (−201 to 75) Private industry and others (government or nonprofit) −17 (−166 to 133) −27 (−184 to 131) Government, nonprofit, and private industry −175 (−344 to −6.3)b −167 (−351 to 17) Abbreviation: NA, not applicable.

rivate industryb −157 (−251 to −63) −180 (−343 to −18) Government and nonprofit −67 (−200 to 67) −63 (−201 to 75) Private industry and others (government or nonprofit) −17 (−166 to 133) −27 (−184 to 131) Government, nonprofit, and private industry −175 (−344 to −6.3)b −167 (−351 to 17) Abbreviation: NA, not applicable. a Only variables significant in the bivariate analysis were included in the multivariable analysis. b Estimates were statistically significant at P < .05. In a post hoc analysis, we characterized 10 studies with the longest data age. They were mostly drug trials (7), were mostly non-US based (6), and had an end of follow-up to publication time range of 7.8 to 91.5 months. In contrast, the 10 studies with the shortest data age shared the following characteristics: 9 were drug trials, all had no follow-up or relatively short follow-up, the end of follow-up to publication time range was 0.5 to 6 months, they involved higher numbers of trial centers, and the area of study was predominately of the hepatitis C virus or Ebola.

he shortest data age shared the following characteristics: 9 were drug trials, all had no follow-up or relatively short follow-up, the end of follow-up to publication time range was 0.5 to 6 months, they involved higher numbers of trial centers, and the area of study was predominately of the hepatitis C virus or Ebola. Discussion In our review of clinical trials published in 2015 in 6 journals with high impact factors, we found that by the time of publication, the median data age was nearly 3 years and the median publication time was more than 1.2 years, with 63 trials (18.5%) taking 2 years or more to complete. For some trials, enrollment periods required as many as 9 days per participant. In multivariable analyses, inconclusive or unfavorable trial results (vs favorable results) were significantly associated with older data age after adjusting for follow-up time. Government-funded trials took 6 months longer in time to publication. Collectively, these findings suggest opportunities to adjust various processes related to clinical trials to improve the timeliness for dissemination of the final results.

gnificantly associated with older data age after adjusting for follow-up time. Government-funded trials took 6 months longer in time to publication. Collectively, these findings suggest opportunities to adjust various processes related to clinical trials to improve the timeliness for dissemination of the final results. Our study extends the current literature in 2 important ways. First, previous studies have shown delays in publication, which we confirm with a comprehensive assessment, in addition to elucidating data age and enrollment time as important time markers of clinical trials.4,5 Our descriptive analysis of the data age, enrollment time, and time to publication of randomized trials in medical journals with the highest impact factors provides benchmarks and indicates leverage points to improve the timeliness for research dissemination. As medical knowledge rapidly evolves, an old data age and a long delay in publication time can result in the knowledge generated from trials being less relevant to contemporary clinical practice.7,8,9 In addition, there could be implications for research—researchers trying to apply the trial findings and advance the science will be delayed in adapting the new knowledge.

e and a long delay in publication time can result in the knowledge generated from trials being less relevant to contemporary clinical practice.7,8,9 In addition, there could be implications for research—researchers trying to apply the trial findings and advance the science will be delayed in adapting the new knowledge. Our study also adds to the literature on which trial characteristics are associated with the time needed for each phase of a trial from the start of enrollment to the dissemination of results to other investigators, clinicians, policy makers, and patients. We found that trials with more centers and more authors were significantly associated with shorter times to publication, but the effect sizes were small. Of greater importance, government funding was associated with substantially longer times to publication. Researchers funded by private companies may have shorter publication times because of greater incentives to produce and distribute findings compared with researchers funded by government grants or nonprofit foundations. Private funders may impose greater accountability on the clinical trial process to match the performance of industry; they may also provide more resources, better staffing, larger infrastructure, and share knowledge of patent drugs, devices, or other interventions to improve timeliness. There are many other factors that may be responsible for the differences between trials funded by private industry and those funded by the government, including resources, the use of contract research organizations, and motivation. The ultimate goal is to identify best practices and spread them. For some trials, particularly in the area of prevention, which depend on the accumulation of the hard end points over time, a longer follow-up time is required, and it is inevitable that these trials will have older data age at the time of publication. However, our study shows that, aside from the follow-up time, multiple areas contribute to older data age, including enrollment and publication times. Our findings reveal many opportunities in these areas where the clinical trial process can be accelerated and the time from data collection to publication (ie, the data age) can be shortened.

ws that, aside from the follow-up time, multiple areas contribute to older data age, including enrollment and publication times. Our findings reveal many opportunities in these areas where the clinical trial process can be accelerated and the time from data collection to publication (ie, the data age) can be shortened. First, because there is variability in enrollment rates, the trials with relatively slow enrollment (>9 days per participant) may need to consider innovative strategies. Enrollment time might be shortened by integrating the randomization process into clinical practice, such as by the use of already existing clinical registries.10 Examples include the SAFE-PCI for Women (Study of Access Site for Enhancement of PCI for Women) trial11 using the National Cardiovascular Research Infrastructure as the platform for randomization and data collection, and the ADAPTABLE (Aspirin Dosing: A Patient-Centric Trial Assessing Benefits and Long-term Effectiveness) trial12 using the National Patient-Centered Clinical Research Network to support rapid and efficient randomization of patients. In this way, participants could be enrolled (and data generated) more quickly. Another approach might be preregistration of participants (ie, creating a pool of people who are amenable to enrollment in trials), so that such individuals are easier to identify and invite to participate. It may also be useful to make enrollment less reliant on clinicians and pursue more direct-to-participant strategies. The idea of participant-partnered research is growing and could provide opportunities to disrupt the current approaches.13,14

, so that such individuals are easier to identify and invite to participate. It may also be useful to make enrollment less reliant on clinicians and pursue more direct-to-participant strategies. The idea of participant-partnered research is growing and could provide opportunities to disrupt the current approaches.13,14 Another opportunity for improvement is publication time, which might be shortened by accelerating the aggregation and analysis of data, the time to write manuscripts, and the submission and revision processes before publication. Of the 6 journals that published the work examined in this study, only BMJ publicly provides this information, and it shows that the peer review process contributes to a substantial proportion of the publication time. It is possible that some manuscripts may be reviewed and rejected at other journals before they are accepted by a different journal, which could lead to some delays. An implication of this study is that there may be more a priori preparation for completion of a manuscript, even before the final data are analyzed. Authors can be preparing the Introduction and Methods of the manuscript even before final results are known, and discussions with journal editors may proceed before the calculation of results. In the end, there needs to be an imperative to report the results of a completed trial quickly and comprehensively. Many recent trials, including the recently published CANTOS (Canakinumab Anti-inflammatory Thrombosis Outcomes Study) trial, demonstrated that a complex trial could have a very short time from completion to publication (58 days).15

be an imperative to report the results of a completed trial quickly and comprehensively. Many recent trials, including the recently published CANTOS (Canakinumab Anti-inflammatory Thrombosis Outcomes Study) trial, demonstrated that a complex trial could have a very short time from completion to publication (58 days).15 Limitations Our study has several limitations. First, this study, by design, focuses on trials from 6 general medicine journals with the highest impact factors in 1 year, not a complete sample of trials published across all general medicine journals. We selected these journals because of their prominence and because they publish the trials that are likely to inform clinical practice, and thus they should be the best examples for how quickly trials are conducted and published. Because the general medical journals with high impact factors analyzed in this study are more likely to publish more quickly compared with other general medical journals, we may expect longer delays if we include all general medicine journals. Second, we used the midpoint of data collection until publication as the definition of data age, but data might not be collected evenly over time. Therefore, our results may not precisely reflect the true data age, but we expect this difference to be small. Third, we do not have information about duration of the peer review process except for 1 journal; therefore, we are unable to determine whether longer times to publication were caused by submission delays or by the time required for peer review, acceptance, and publication. Fourth, this study cannot determine the effect of older data age in some trials. However, practice is changing rapidly, and it could be that the clinical care patterns at the end of the trial were different than at the beginning. Such interactions of effect are rarely tested. Also, all things being equal, more recent data and a more quickly completed study are preferable. Therefore, data age does seem to be a relevant metric worthy of more attention and study. This study also did not evaluate delays in the translation of the trial findings into practice. Often, even well-done trials experience delays after publication. This issue—which was beyond the scope of our present article—also deserves attention, along with more timely knowledge generation in the course of trials.

dy. This study also did not evaluate delays in the translation of the trial findings into practice. Often, even well-done trials experience delays after publication. This issue—which was beyond the scope of our present article—also deserves attention, along with more timely knowledge generation in the course of trials. Conclusions Clinical trials in 6 journals with high impact factors were published with a median data age of nearly 3 years. For a substantial proportion of these trials, there were extended times for enrollment and publication that led to markedly older data at the time of publication. There are seemingly many opportunities for improvement in the clinical trial process and in the work of trialists with journal editors. Supplement. eFigure 1. PRISMA Flowchart eFigure 2. Distribution of Data Age for Randomized Trials Published in 6 High-Impact Journals in 2015 eFigure 3. Distribution of Enrollment Time for Randomized Trials Published in 6 High-Impact Journals in 2015 eFigure 4. Distribution of Publication Time for Randomized Trials Published in 6 High-Impact Journals in 2015 Click here for additional data file.

Introduction Q fever (or Query fever) is a globally widespread zoonosis that was first described in 1937. The previous dichotomy between acute and chronic Q fever, which was based on serologic criteria, has led to the current confusion between chronic infection and post–Q fever fatigue syndrome.1 The diagnosis of infection requires the demonstration of an organic lesion (identified as the infectious focus) and evidence of a microbial infection (as proven by serologic findings, polymerase chain reaction analysis, culture, and/or immunohistochemical analysis using anti–Coxiella burnetii antibodies).2,3,4,5 This concept of the disease and paradigm shift was made possible by revolutionary improvements in imaging during the 21st century. The systematic and early use of transthoracic echocardiography (TTE) and positron emission tomographic (PET) scanning has allowed the identification of infectious foci that were previously undetected and that are now crucial and decisive for the diagnosis of C burnetii infection and to guide the choice of therapeutic approach. Therefore, new definition criteria have recently been published for C burnetii endocarditis, vascular infection, osteoarticular infection, lymphadenitis, and interstitial lung disease.2,3,4,5 Consequently, the term chronic Q fever should no longer be used.2

tii infection and to guide the choice of therapeutic approach. Therefore, new definition criteria have recently been published for C burnetii endocarditis, vascular infection, osteoarticular infection, lymphadenitis, and interstitial lung disease.2,3,4,5 Consequently, the term chronic Q fever should no longer be used.2 This change is particularly important because the serologic response is strain dependent. As an example, IgG antibody titers to phase I C burnetii are higher in French Guiana, where a unique strain is endemic, than in metropolitan France.6,7 For this reason, Q fever postinfectious syndrome is defined by an association of elevated IgG titers to phase I C burnetii with subjective symptoms only. This definition is different from that of the Netherlands team, who consider these cases to be chronic infection.1 The 26-year experience of the French National Reference Center for Q fever, which contains data collected prospectively from patients worldwide, gave us the opportunity to reanalyze C burnetii infection using the clarified 21st century definition and to highlight new, rare, and unusual infectious foci of the disease.

1 The 26-year experience of the French National Reference Center for Q fever, which contains data collected prospectively from patients worldwide, gave us the opportunity to reanalyze C burnetii infection using the clarified 21st century definition and to highlight new, rare, and unusual infectious foci of the disease. Methods Study Design and Setting The French National Reference Center for Q fever is designed by the French government to collect data from patients with Q fever as part of epidemiologic surveillance and receives serum samples from France and abroad (eFigure 1 in the Supplement). Epidemiologic, clinical, and biological data from positive cases are collected as described later. In this prospective cohort study, we report the epidemiologic, clinical, and biological data collected in the French National Reference Center from January 1, 1991, to December 31, 2016. We followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines.8 According to the procedures of the French Commission for Data Protection (Commission Nationale de l’Informatique et des Libertés), collected data were anonymized. The study was approved by the local ethics committee of IHU (Institut Hospitalo-Universitaire)–Méditerranée Infection and by the French National Drugs and Health Products Agency. An oral consent was obtained from study participants by the referral physician.

nformatique et des Libertés), collected data were anonymized. The study was approved by the local ethics committee of IHU (Institut Hospitalo-Universitaire)–Méditerranée Infection and by the French National Drugs and Health Products Agency. An oral consent was obtained from study participants by the referral physician. Participants and Follow-up For each positive C burnetii test finding, clinical data were collected by telephone from the referral physician using a standardized questionnaire (eFigure 2 in the Supplement). Patients with positive serologic test results who presented with clinical manifestations consistent with an active C burnetii infection were included. We asked the referral physician for serologic tests 3 and 6 months after diagnosis to monitor clinical and serologic markers of improvement. The duration of monitoring was set at 5 years in case of persistent focalized infections. Patients with unavailable clinical data or unidentified infectious focus were excluded from the study analysis. Diagnosis of C burnetii Infection Serologic testing and molecular detection were performed as previously described.9,10,11 Culture and immunohistochemistry using specific anti–C burnetii antibodies and fluorescence in situ hybridization were performed as previously described.9,12,13,14,15

Participants and Follow-up For each positive C burnetii test finding, clinical data were collected by telephone from the referral physician using a standardized questionnaire (eFigure 2 in the Supplement). Patients with positive serologic test results who presented with clinical manifestations consistent with an active C burnetii infection were included. We asked the referral physician for serologic tests 3 and 6 months after diagnosis to monitor clinical and serologic markers of improvement. The duration of monitoring was set at 5 years in case of persistent focalized infections. Patients with unavailable clinical data or unidentified infectious focus were excluded from the study analysis. Diagnosis of C burnetii Infection Serologic testing and molecular detection were performed as previously described.9,10,11 Culture and immunohistochemistry using specific anti–C burnetii antibodies and fluorescence in situ hybridization were performed as previously described.9,12,13,14,15 Case Definition Primary (acute) C burnetii infection was defined by the association of acute clinical symptoms with the following serologic criteria: IgG titers representing phase II (≥200) and IgM titers representing phase II (≥50) or seroconversion within 3 months of the primary symptoms.16 Persistent C burnetii focal infection was diagnosed using the recently updated criteria as persistence of clinical symptoms for more than 3 months in addition to the identification of an infectious focus (eTables 1-3 in the Supplement).2,3,4 Immunosuppression was defined in patients with known organ deficiency (those undergoing hemodialysis or before transplant), patients who were receiving an immunosuppressive drug or who underwent splenectomy, and patients with polymetastatic cancer.

tification of an infectious focus (eTables 1-3 in the Supplement).2,3,4 Immunosuppression was defined in patients with known organ deficiency (those undergoing hemodialysis or before transplant), patients who were receiving an immunosuppressive drug or who underwent splenectomy, and patients with polymetastatic cancer. Biological Variables Findings for G-isotype anticardiolipin (IgG aCL) antibodies were defined as positive at greater than 22 IgG anti–phospholipid-binding units (GPLU). Since 2012, tests have been systematically performed when an active C burnetii infection has been identified.17,18 Imaging For all patients with a positive C burnetii serologic test result since 2001, we recommend a cardiac TTE to detect known or unknown valvular defects or new valvular lesions compatible with endocarditis, because these conditions require prophylactic or curative treatment.17,18 Since 2009, the use of a PET scan is systematically recommended, when accessible, to detect deep infectious foci when focalized persistent infection is suspected.2

or unknown valvular defects or new valvular lesions compatible with endocarditis, because these conditions require prophylactic or curative treatment.17,18 Since 2009, the use of a PET scan is systematically recommended, when accessible, to detect deep infectious foci when focalized persistent infection is suspected.2 Statistical Analysis To compare the distribution of continuous or dichotomous variables between 2 groups, we used the 2-sided t test or the 2-sided Fisher test, respectively. Two-sided P < .05 was considered to indicate a significant difference between 2 groups. The mortality rate was computed as the number of deaths occurring in the cohort divided by the number of person-years during the study period. The Cox proportional hazards regression model was used to determine the factors associated with mortality risk. The proportional hazards assumption was tested in the Cox regression models by examining the rescaled Schoenfeld residuals. The indicative factors for complications were determined using the logistic regression model (for evolution to a persistent focalized C burnetii infection), the Poisson regression model (in cases of rare manifestations of acute Q fever), or the Cox proportional hazards regression model (for lymphoma). All multivariate models were adjusted for sex, age, and year category at baseline (before 2009, 2009-2012, and after 2012), and interactions were also tested. We used Stata/SE software (version 14.2; StataCorp LP) for all the analyses.

of acute Q fever), or the Cox proportional hazards regression model (for lymphoma). All multivariate models were adjusted for sex, age, and year category at baseline (before 2009, 2009-2012, and after 2012), and interactions were also tested. We used Stata/SE software (version 14.2; StataCorp LP) for all the analyses. Results Clinical Presentation of the Whole Cohort From 1991 to 2016, 277 666 serum specimens were tested for antibodies to C burnetii (eFigure 1 in the Supplement). Of 180 483 patients undergoing testing, 2918 had positive findings for C burnetti. Of these, 2434 had a positive serologic result and an identified infectious focus (1674 [68.8%] men and 760 [31.2%] women) (Figure 1). A total of 2105 patients (86.5%) lived in metropolitan France, and 222 (9.1%) lived in Latin America, mostly in French Guiana (eTable 4 in the Supplement). The mean (SD) age of patients was 51.8 (17.4) years (range, 0-98 years); 58 (2.4% of the patients) were 16 years or younger. The ratio of men to women was 2.2 in adults and 0.9 in children (27 males [46%] and 31 females [54%]) (eFigure 3 in the Supplement). Mean (SD) follow-up was 16 (29) months. The medical records and cardiac TTEs at the time of diagnosis revealed that 640 patients (26.3%) presented with a valvulopathy, 91 (3.7%) were immunosuppressed (eTable 5 in the Supplement), and 36 (1.5%) were pregnant women (eTable 6 in the Supplement). Positron emission tomographic scanning was performed for 291 patients (12.0%) (eFigure 4 in the Supplement).

e time of diagnosis revealed that 640 patients (26.3%) presented with a valvulopathy, 91 (3.7%) were immunosuppressed (eTable 5 in the Supplement), and 36 (1.5%) were pregnant women (eTable 6 in the Supplement). Positron emission tomographic scanning was performed for 291 patients (12.0%) (eFigure 4 in the Supplement). Figure 1. Study Flowchart Among the 2434 patients included in the study analysis, 1668 had only acute Q fever, 628 had only persistent focalized C burnetii infection, and 138 had an acute Q fever that evolved to a persistent C burnetii infection. NRC indicates National Reference Center. Among the 2434 patients included, 602 (24.7%) had a single-day follow-up. Nine hundred twenty-four of 1806 patients with acute Q fever (51.2%) were followed up for more than 3 months, and 149 of 766 patients with persistent C burnetii infection (19.5%) were followed up for more than 5 years. C burnetii Infection In the overall cohort, 1806 patients (74.2%) had acute Q fever, and 31.5% (766 of 2434) presented with a persistent focalized infection. Among the 2434 patients, hepatitis was the most frequent clinical form of Q fever (933 [38.3%]), followed by endocarditis (533 [21.9%]), and pneumonia (618 [25.4%]) (Figure 2). Hepatitis (836 [46.3%]), pneumonia (480 [26.6%]), and flulike syndrome (350 [19.4%]) were the main clinical presentations of acute Q fever, followed by lymphadenitis (97 [5.4%]) (eTables 7 and 8 in the Supplement).

of Q fever (933 [38.3%]), followed by endocarditis (533 [21.9%]), and pneumonia (618 [25.4%]) (Figure 2). Hepatitis (836 [46.3%]), pneumonia (480 [26.6%]), and flulike syndrome (350 [19.4%]) were the main clinical presentations of acute Q fever, followed by lymphadenitis (97 [5.4%]) (eTables 7 and 8 in the Supplement). Figure 2. Clinical Presentations of Coxiella burnetii Infection Includes a total of 2434 patients with positive C burnetti serologic findings consistent with C burnetti infection. Among the 766 patients diagnosed as having persistent focalized C burnetii infection, 581 (75.8%) presented with endocarditis, 145 (18.9%) had a vascular infection, and 56 (7.3%) had an osteoarticular infection (eTables 9-11 in the Supplement). The mean (SD) age at diagnosis was 60 (17) years for these 766 patients, and the median duration of follow-up was 15.1 months (interquartile range, 1.7-45.3 months). In 91 patients (15.7% of patients with a final diagnosis of endocarditis), the use of transesophageal echography was critical to identify a valvular lesion, which was not detected with TTE. We witnessed the evolution of acute Q fever to a persistent C burnetii infection in 138 patients (7.6%) (Table 1). Table 1. Evolution to Persistent Coxiella burnetii Infection in 1806 Patients With Acute Q Fever Patient Characteristic No.

Among the 766 patients diagnosed as having persistent focalized C burnetii infection, 581 (75.8%) presented with endocarditis, 145 (18.9%) had a vascular infection, and 56 (7.3%) had an osteoarticular infection (eTables 9-11 in the Supplement). The mean (SD) age at diagnosis was 60 (17) years for these 766 patients, and the median duration of follow-up was 15.1 months (interquartile range, 1.7-45.3 months). In 91 patients (15.7% of patients with a final diagnosis of endocarditis), the use of transesophageal echography was critical to identify a valvular lesion, which was not detected with TTE. We witnessed the evolution of acute Q fever to a persistent C burnetii infection in 138 patients (7.6%) (Table 1). Table 1. Evolution to Persistent Coxiella burnetii Infection in 1806 Patients With Acute Q Fever Patient Characteristic No. (%) of Patients Univariate Analysis P Valuea Logistic Regression Acute Q Fever Without Persistent C burnetii Infection (n = 1668) Acute Q Fever Progressing to Persistent C burnetii Infection (n = 138) Univariate Multivariate OR (95% CI) P Value OR (95% CI) P Value Immunosuppression No 1608 (96.4) 132 (95.7) NA 1 [Reference] NA NA NA Yes 60 (3.6) 6 (4.3) .63 1.2 (0.5-2.9) .65 NR NA Valvulopathy No 1498 (89.8) 53 (38.4) NA 1 [Reference] NA 1 [Reference] NA Yes 170 (10.2) 85 (61.6) <.001 14.1 (9.7-20.6) <.001 9.8 (6.1-15.8) <.001 Sex Male 1115 (66.8) 110 (79.7) .002 1.9 (1.3-3.0) .002 1.9 (1.1-3.1) .01 Female 553 (33.2) 28 (20.3) NA 1 [Reference] NA 1 [Reference] NA Age at baseline, median (IQR), y 48 (37-59) 55.5 (46-68) <.001 1.03 (1.02-1.04) <.001 1.01 (1.00-1.03) .03 Year category at baseline Before 2009 186 (11.2) 55 (39.9) NA 1 [Reference] NA 1 [Reference] NA 2009-2012 680 (40.8) 32 (23.2) NA 4.6 (3.1-7.0) <.001 3.2 (1.9-5.3) <.001 After 2012 802 (48.1) 51 (37.0) .03 0.7 (0.5-1.2) .19 0.8 (0.5-1.4) .42 Pneumonia No 1211 (72.6) 115 (83.3) NA 1 [Reference] NA 1 [Reference] NA Yes 457 (27.4) 23 (16.7) .005 0.5 (0.3-0.8) .007 0.6 (0.3-1.0) .07 Lymphadenitis No 1616 (96.9) 124 (89.9) NA 1 [Reference] NA 1 [Reference] NA Yes 52 (3.1) 14 (10.1) <.001 3.5 (1.9-6.5) <.001 3.3 (1.6-7.1) .002 Thrombosis No 1657 (99.3) 133 (96.4) NA 1 [Reference] NA 1 [Reference] NA Yes 11 (0.7) 5 (3.6) .005 5.7 (1.9-16.5) .002 6.8 (1.9-24.8) .004 Acute endocarditis No 1636 (98.1) 120 (87.0) NA 1 [Reference] NA 1 [Reference] NA Yes 32 (1.9) 18 (13.0) <.001 7.7 (4.2-14.1) .001 3.8 (1.5-9.8) .006 IgG titer to phase I on first serologic analysis ≤800 1476 (88.5) 84 (60.9) NA 1 [Reference] NA NA NA >800 192 (11.5) 54 (39.1) <.001 4.9 (3.4-7.2) <.001 NR NA Maximum IgG titer to phase I ≤800 1313 (78.8) 48 (34.8) NA 1 [Reference] NA 1 [Reference] NA >800 354 (21.2) 90 (65.2) <.001 7.0 (4.8-10.1) <.001 5.2 (3.3-8.1) <.001 IgG aCL antibody titer ≤90 GPLU 722 (77.4) 59 (67.8) NA 1 [Reference] NA NA NA >90 GPLU 211 (22.6) 28 (32.2) .048 1.6 (1.0-2.6) .046 NR NA Positive C burnetii PCR No 1481 (91.3) 105 (77.8) NA 1 [Reference] NA 1 [Reference] NA Yes 142 (8.7) 30 (22.2) <.001 3.0 (1.9-4.6) <.001 1.

7.0 (4.8-10.1) <.001 5.2 (3.3-8.1) <.001 IgG aCL antibody titer ≤90 GPLU 722 (77.4) 59 (67.8) NA 1 [Reference] NA NA NA >90 GPLU 211 (22.6) 28 (32.2) .048 1.6 (1.0-2.6) .046 NR NA Positive C burnetii PCR No 1481 (91.3) 105 (77.8) NA 1 [Reference] NA 1 [Reference] NA Yes 142 (8.7) 30 (22.2) <.001 3.0 (1.9-4.6) <.001 1. 9 (1.0-3.4) .03 Abbreviations: GPLU, IgG anti–phospholipid-binding units; IgG aCL, G-isotype anticardiolipin; IQR, interquartile range; NA, not applicable; NR, not retained in the model; OR, odds ratio; PCR, polymerase chain reaction. a Calculated using the 2-sided Fisher exact test or 2-sided t test. New Clinical Presentations Lymphadenitis and Lymphoma Lymphadenitis was identified in 97 patients (4.0%). Lymphadenitis was concomitant with persistent focal C burnetii infection in 36 of 97 cases and was the unique infective focus in 23 cases (eTable 8 and eFigure 5 in the Supplement). Positron emission tomographic scanning enabled the identification of deep lymphadenitis in 18 of 41 cases (43.9%). In patients diagnosed with Q fever, 16 were diagnosed as having a lymphoma. Fourteen (87.5%) were men, 14 had B-cell non–Hodgkin lymphoma, and 2 had T-cell lymphoma (6 diffuse large B-cell lymphomas, 2 follicular lymphomas, 1 gastric lymphoma, 1 mucosa-associated lymphoid tissue lymphoma, 2 marginal zone lymphomas, 1 mantle cell lymphoma, and 1 lymphoplasmacytic lymphoma) (eFigure 6 in the Supplement). Fourteen patients with lymphoma presented lymphadenitis. One patient had concomitant hemophagocytic syndrome and no lymphadenitis.

phomas, 1 gastric lymphoma, 1 mucosa-associated lymphoid tissue lymphoma, 2 marginal zone lymphomas, 1 mantle cell lymphoma, and 1 lymphoplasmacytic lymphoma) (eFigure 6 in the Supplement). Fourteen patients with lymphoma presented lymphadenitis. One patient had concomitant hemophagocytic syndrome and no lymphadenitis. Coagulation Disorder, Elevated aCL Titers, and Acute Q Fever Endocarditis Acute Q fever endocarditis (50 patients [2.1%]) was a risk factor for evolution to persistent focal C burnetii infection. Acute Q fever endocarditis occurred with hepatitis in 25 cases (50.0%) and with pneumonia in 13 (26.0%). Thrombosis Thrombosis was diagnosed in 23 patients of our cohort. Thrombosis was concomitant with persistent focalized C burnetii infection in 12 cases, including 8 with endocarditis, 5 with vascular infections, and 1 with osteoarticular infection (3 patients presented with endocarditis and vascular infection). Atypical Forms of Q Fever Among the 2434 patients included in the analysis, neurologic involvement was diagnosed in 32 (1.3%). Fifteen patients presented with meningoencephalitis; 11, with meningitis; and 6, with encephalitis. In 25 of these patients, the neurologic infection occurred as a part of acute Q fever; in 7, it occurred as part of persistent endocarditis. Three patients who developed encephalitis as a complication of septic emboli died.

teen patients presented with meningoencephalitis; 11, with meningitis; and 6, with encephalitis. In 25 of these patients, the neurologic infection occurred as a part of acute Q fever; in 7, it occurred as part of persistent endocarditis. Three patients who developed encephalitis as a complication of septic emboli died. Thirty-one patients of the cohort (1.3%) had pericarditis. Twenty-three had acute Q fever, 11 had a persistent C burnetii infection, and 3 had both. Among the 7 patients with myocarditis, 2 developed severe complications, including 1 conduction defect and 1 case of endomyocardial fibrosis. Eye involvement (n = 10) was the unique possible persistent focus of C burnetii infection in 7 cases and was concomitant with persistent lymphadenitis, pneumonia, and endocarditis in 1 patient each. Uveitis was the main observed manifestation (n = 7), followed by papillitis (n = 2), chorioretinitis (n = 1), and optic neuritis (n = 1). One patient presented with uveitis and papillitis. Alithiasic cholecystitis was diagnosed in 11 patients (0.4%). In 4 of the 5 patients for whom IgG aCL was measured, these antibody titers were elevated (Table 2). For 2 patients, histologic analysis of the gallbladder showed inflammatory infiltrates, but immunohistochemical analysis yielded negative findings in both cases.

hiasic cholecystitis was diagnosed in 11 patients (0.4%). In 4 of the 5 patients for whom IgG aCL was measured, these antibody titers were elevated (Table 2). For 2 patients, histologic analysis of the gallbladder showed inflammatory infiltrates, but immunohistochemical analysis yielded negative findings in both cases. Table 2. Positive aCL Antibodies Associated With Clinical Complications of Coxiella burnetii Infection in 1328 Patients With Available IgG aCL Titers Acute Q Fever Manifestation No. (%) of Patients Univariate Analysis P Valuea Multivariate Logistic Regressionb IgG aCL ≤22 GPLU (n = 830) IgG aCL >22 GPLU (n = 498) OR or IRR (95% CI) P Value Pneumonia (n = 319) 236 (28.4) 83 (16.7) <.001 0.5 (0.4-0.6)c <.001 Hepatitis (n = 503) 217 (26.1) 286 (57.4) <.001 3.7 (2.9-4.7)c <.001 Cholecystitis (n = 5) 1 (0.1) 4 (0.8) .07 6.9 (0.7-62.8)d .09 Hemophagocytic syndrome (n = 9) 0 9 (1.8) <.001 NR NR Acute endocarditis (n = 42) 13 (1.6) 28 (5.6) <.001 3.9 (2.0-7.5)d <.001 Thrombosis (n = 21) 10 (1.2) 11 (2.2) .18 2.1 (0.9-5.2)d .09 Abbreviations: aCL, anticardiolipin; GPLU, IgG anti–phospholipid-binding units; IRR, incidence rate ratio; NR, not retained in this model; OR, odds ratio. a Calculated using the 2-sided Fisher exact test or χ2 test. b All multivariate models are adjusted for sex, age, and year category at baseline (before 2009, 2009-2012, and after 2012). c Odds ratio calculated using multivariate logistic regression. d Incidence rate ratio calculated using multivariate Poisson regression.

Table 2. Positive aCL Antibodies Associated With Clinical Complications of Coxiella burnetii Infection in 1328 Patients With Available IgG aCL Titers Acute Q Fever Manifestation No. (%) of Patients Univariate Analysis P Valuea Multivariate Logistic Regressionb IgG aCL ≤22 GPLU (n = 830) IgG aCL >22 GPLU (n = 498) OR or IRR (95% CI) P Value Pneumonia (n = 319) 236 (28.4) 83 (16.7) <.001 0.5 (0.4-0.6)c <.001 Hepatitis (n = 503) 217 (26.1) 286 (57.4) <.001 3.7 (2.9-4.7)c <.001 Cholecystitis (n = 5) 1 (0.1) 4 (0.8) .07 6.9 (0.7-62.8)d .09 Hemophagocytic syndrome (n = 9) 0 9 (1.8) <.001 NR NR Acute endocarditis (n = 42) 13 (1.6) 28 (5.6) <.001 3.9 (2.0-7.5)d <.001 Thrombosis (n = 21) 10 (1.2) 11 (2.2) .18 2.1 (0.9-5.2)d .09 Abbreviations: aCL, anticardiolipin; GPLU, IgG anti–phospholipid-binding units; IRR, incidence rate ratio; NR, not retained in this model; OR, odds ratio. a Calculated using the 2-sided Fisher exact test or χ2 test. b All multivariate models are adjusted for sex, age, and year category at baseline (before 2009, 2009-2012, and after 2012). c Odds ratio calculated using multivariate logistic regression. d Incidence rate ratio calculated using multivariate Poisson regression. Hemophagocytic syndrome was diagnosed in 9 patients (0.4%), all with acute Q fever. In 1 patient, evolution to endocarditis occurred; 1 had evolution to vascular C burnetii infection; and 1 presented with a marginal B-cell lymphoma diagnosed based on a splenic biopsy. All 9 patients had elevated IgG aCL titers (>22 GPLU) (Table 2).

c syndrome was diagnosed in 9 patients (0.4%), all with acute Q fever. In 1 patient, evolution to endocarditis occurred; 1 had evolution to vascular C burnetii infection; and 1 presented with a marginal B-cell lymphoma diagnosed based on a splenic biopsy. All 9 patients had elevated IgG aCL titers (>22 GPLU) (Table 2). Seven patients with persistent focal C burnetii infection presented with interstitial lung disease. All patients had severe and advanced fibrotic lung lesions.19 A pseudotumor of the lung was detected in 3 patients. One had persistent endocarditis, 1 had persistent lymphadenitis, and 1 had acute Q fever.18 Seven patients with positive findings for C burnetii infection presented with giant cell arteritis, 5 had acute Q fever, and 1 had vascular infection. Half of the patients had positive IgG aCL titers. Peculiarities of Q Fever in French Guiana In French Guiana, the ratio of men to women was 1.5, and acute pneumonia represented clinical presentation in 154 of 220 (70.0%). Only 13 patients (5.9%) presented with elevated aCL antibodies in the acute phase of the disease, which is much lower than that observed in metropolitan France (1115 of 2105 [53.0%]) (P < .001). Host Factors Immunocompromised Patients Ninety-one patients were immunocompromised. Among these, 52 were receiving immunosuppressive therapy. Sixty-six patients (72.5%) had acute Q fever, and 31 (34.1%) presented with persistent focal infection (eTable 5 in the Supplement).

Peculiarities of Q Fever in French Guiana In French Guiana, the ratio of men to women was 1.5, and acute pneumonia represented clinical presentation in 154 of 220 (70.0%). Only 13 patients (5.9%) presented with elevated aCL antibodies in the acute phase of the disease, which is much lower than that observed in metropolitan France (1115 of 2105 [53.0%]) (P < .001). Host Factors Immunocompromised Patients Ninety-one patients were immunocompromised. Among these, 52 were receiving immunosuppressive therapy. Sixty-six patients (72.5%) had acute Q fever, and 31 (34.1%) presented with persistent focal infection (eTable 5 in the Supplement). Children Among 58 children, 4 were neonates. Fourteen children presented with persistent endocarditis, whereas lymphadenitis was the unique clinical presentation in 2 children (eTable 12 in the Supplement). Pregnant Women Thirty-six included patients were pregnant women (mean [SD] age, 30 [6] years). Infection occurred mostly during the 6 first months of pregnancy. Pregnancy complications were identified in 22 of these patients (61.1%) (eTable 6 in the Supplement). Sex Being male was associated with an increased risk of vascular infection independent of age (odds ratio [OR], 3.4; 95% CI, 2.0-5.7; P < .001). Men presented with higher IgG aCL titers than women in the primary phase of the disease (OR, 1.6; 95% CI, 1.3-2.1; P < .001), independent of age. Twenty-seven of 58 children (46.6%) were boys; 1654 of 2376 adults (69.6%) were men (P < .001).

nfection independent of age (odds ratio [OR], 3.4; 95% CI, 2.0-5.7; P < .001). Men presented with higher IgG aCL titers than women in the primary phase of the disease (OR, 1.6; 95% CI, 1.3-2.1; P < .001), independent of age. Twenty-seven of 58 children (46.6%) were boys; 1654 of 2376 adults (69.6%) were men (P < .001). Mortality and Complications Fifty-eight (2.4%) of the 2434 patients diagnosed with Q fever died. Among these, 46 were men (79.3%), the mean (SD) age at the time of Q fever diagnosis was 65.8 (12.9) years, and the mean (SD) age at the time of death was 69 (12) years. Among the 1806 patients with acute Q fever, 3 died (2 of complications of a solid tumor, and 1 of fulminant Q fever hepatitis). Among 766 patients with persistent focalized infection, 55 died, 43 had endocarditis (4 with concomitant vascular infection), and 16 had a vascular infection (of whom 4 had concomitant spondylodiscitis) (Figure 3). Figure 3. Kaplan-Meier Survival Analysis Includes patients with Coxiella burnetii infection. PEI indicates persistent endocarditis; PVI, persistent vascular infection.

Mortality and Complications Fifty-eight (2.4%) of the 2434 patients diagnosed with Q fever died. Among these, 46 were men (79.3%), the mean (SD) age at the time of Q fever diagnosis was 65.8 (12.9) years, and the mean (SD) age at the time of death was 69 (12) years. Among the 1806 patients with acute Q fever, 3 died (2 of complications of a solid tumor, and 1 of fulminant Q fever hepatitis). Among 766 patients with persistent focalized infection, 55 died, 43 had endocarditis (4 with concomitant vascular infection), and 16 had a vascular infection (of whom 4 had concomitant spondylodiscitis) (Figure 3). Figure 3. Kaplan-Meier Survival Analysis Includes patients with Coxiella burnetii infection. PEI indicates persistent endocarditis; PVI, persistent vascular infection. Factors Associated With Mortality Among the 2434 patients with Q fever (accounting for a total duration of follow-up of 3276 person-years, with a median follow-up duration of 4.6 months [interquartile range, 0.1-18.1 months]), the mortality rate was 1.8 (95% CI, 1.4-2.3) per 100 person-years. The mortality rate was 2.4 (95% CI, 1.8-3.3) per 100 person-years among patients with C burnetii persistent endocarditis, and 6.2 (95% CI, 3.8-10.1) per 100 person-years among patients with C burnetii vascular infection.

artile range, 0.1-18.1 months]), the mortality rate was 1.8 (95% CI, 1.4-2.3) per 100 person-years. The mortality rate was 2.4 (95% CI, 1.8-3.3) per 100 person-years among patients with C burnetii persistent endocarditis, and 6.2 (95% CI, 3.8-10.1) per 100 person-years among patients with C burnetii vascular infection. All Cox proportional hazards regression risk models used to test the association between each condition and mortality were adjusted for sex, age, and calendar period. Persistent focal C burnetii infections were associated with an increased risk of death (hazard ratio [HR], 10.9; 95% CI, 3.2-37.1; P < .001). Persistent endocarditis (HR, 2.4; 95% CI, 1.1-5.1; P = .02) and vascular infection (HR, 3.1; 95% CI, 1.7-5.7; P < .001) were associated with an increased risk of death. Meningitis (HR, 4.0; 95% CI, 1.4-11.6; P = .009) and spondylodiscitis (HR, 8.3; 95% CI, 3.3-20.9; P < .001), when associated with cardiovascular infection, were also associated with a higher risk of death.

.02) and vascular infection (HR, 3.1; 95% CI, 1.7-5.7; P < .001) were associated with an increased risk of death. Meningitis (HR, 4.0; 95% CI, 1.4-11.6; P = .009) and spondylodiscitis (HR, 8.3; 95% CI, 3.3-20.9; P < .001), when associated with cardiovascular infection, were also associated with a higher risk of death. Factors Associated With Complications According to the Cox proportional hazards regression model with lymphoma as the dependent variable, after adjustment for sex, age, and year category at baseline, we found that the presence of lymphadenitis was associated with a higher risk of lymphoma (HR, 77.4; 95% CI, 21.2-281.8; P < .001) as was the presence of hemophagocytic syndrome (HR, 19.1; 95% CI, 3.4-108.6; P < .001). Valvulopathy, thrombosis, lymphadenitis, maximum high IgG titer (>800), acute Q fever endocarditis, and male sex were identified as factors associated with evolution to persistent focal C burnetii infection (Table 1).

1.8; P < .001) as was the presence of hemophagocytic syndrome (HR, 19.1; 95% CI, 3.4-108.6; P < .001). Valvulopathy, thrombosis, lymphadenitis, maximum high IgG titer (>800), acute Q fever endocarditis, and male sex were identified as factors associated with evolution to persistent focal C burnetii infection (Table 1). IgG anticardiolipin titers were available for 1328 patients (54.6%). Positive aCL antibody findings were indicative of acute Q fever endocarditis (incidence rate ratio, 3.9; 95% CI, 2.0-7.5; P < .001) and hemophagocytic syndrome (all patients in this group were positive for aCL antibodies) (Table 2). Immunosuppression was not indicative of any complication. In case of acute Q fever, the receiver operating characteristics analysis showed that the presence of aCL antibodies were significantly associated with acute Q fever complications such as acute Q fever endocarditis (area under the curve [AUC], 0.67; 95% CI, 0.58-0.76; P < .001), thrombosis (AUC, 0.72; 95% CI, 0.60-0.85; P = .002), hemophagocytic syndrome (AUC, 0.78; 95% CI, 0.67-0.89; P = .003), meningitis (AUC, 0.68; 95% CI, 0.56-0.79; P = .01), and alithiasic cholecystitis (AUC, 0.75; 95% CI, 0.60-0.90; P = .05) (eTable 13 in the Supplement). Discussion We present a comprehensive description of a 26-year cohort of patients with Q fever from the French National Reference Center for Q fever. Endocarditis was the second infectious C burnetii focus identified.

IgG anticardiolipin titers were available for 1328 patients (54.6%). Positive aCL antibody findings were indicative of acute Q fever endocarditis (incidence rate ratio, 3.9; 95% CI, 2.0-7.5; P < .001) and hemophagocytic syndrome (all patients in this group were positive for aCL antibodies) (Table 2). Immunosuppression was not indicative of any complication. In case of acute Q fever, the receiver operating characteristics analysis showed that the presence of aCL antibodies were significantly associated with acute Q fever complications such as acute Q fever endocarditis (area under the curve [AUC], 0.67; 95% CI, 0.58-0.76; P < .001), thrombosis (AUC, 0.72; 95% CI, 0.60-0.85; P = .002), hemophagocytic syndrome (AUC, 0.78; 95% CI, 0.67-0.89; P = .003), meningitis (AUC, 0.68; 95% CI, 0.56-0.79; P = .01), and alithiasic cholecystitis (AUC, 0.75; 95% CI, 0.60-0.90; P = .05) (eTable 13 in the Supplement). Discussion We present a comprehensive description of a 26-year cohort of patients with Q fever from the French National Reference Center for Q fever. Endocarditis was the second infectious C burnetii focus identified. New complications identified included acute Q fever endocarditis, hemophagocytic syndrome, thrombosis, lymphadenitis, and lymphoma. Anticardiolipin antibodies (IgG aCL) during acute Q fever were indicative of hepatitis, cholecystitis, endocarditis, thrombosis, hemophagocytic syndrome, meningitis, and progression to persistent endocarditis. Meningitis and spondylodiscitis complications were associated with an increased risk of death.20

and lymphoma. Anticardiolipin antibodies (IgG aCL) during acute Q fever were indicative of hepatitis, cholecystitis, endocarditis, thrombosis, hemophagocytic syndrome, meningitis, and progression to persistent endocarditis. Meningitis and spondylodiscitis complications were associated with an increased risk of death.20 In the Netherlands, where Q fever has been responsible of 4000 cases in 4 years, physicians used a definitive criterion based on a serologic cutoff rather than the clinical features. Consequently, Q fever complication and unusual presentation of the diseases are probably underestimated.21,22,23 In the Netherlands, the mortality rate varies from 1% in cases of acute Q fever to 13% in cases of persistent focalized C burnetii infection (9% for endocarditis and 21% for vascular infection).22,24 To our knowledge, no study on aCL associated with Q fever has been published from the Netherlands. In addition, TTE is not performed in case of acute Q fever.25 Therefore, comparison of the clinical features observed herein with those observed in the Netherlands remains difficult. By contrast, in French Guiana, clinical manifestations of the disease presented some peculiarities.26 Strikingly, the disease affects men and women almost equally. Although the Guiana strain has been described as a highly virulent strain in vivo (unpublished data; C.M., Aurelia Caputo, PhD, Yassina Bechah, PhD, et al; June 2018), no elevation of aCL antibody levels was observed in this region. Regarding lymphoma, a prospective study needs to be performed with the definition criteria used herein.

ana strain has been described as a highly virulent strain in vivo (unpublished data; C.M., Aurelia Caputo, PhD, Yassina Bechah, PhD, et al; June 2018), no elevation of aCL antibody levels was observed in this region. Regarding lymphoma, a prospective study needs to be performed with the definition criteria used herein. Host Factors In pregnant women, the highest proportion of complications (61.1%) corroborates previous dramatic reports on Q fever during pregnancy.27 Placentitis and microthrombi have been described.23 In children, Q fever has been marked by an age-related increase in incidence.28 The imbalance in the sex ratio distribution of the disease occurred after puberty, and males were most affected by vascular infections and presented with a higher secretion of aCL antibodies during acute Q fever. Thus, sex hormones likely influence the host’s response during Q fever, and further investigation is warranted to determine the mechanism involved.29

n of the disease occurred after puberty, and males were most affected by vascular infections and presented with a higher secretion of aCL antibodies during acute Q fever. Thus, sex hormones likely influence the host’s response during Q fever, and further investigation is warranted to determine the mechanism involved.29 New Clinical Manifestations Lymphadenitis and Lymphoma In 2015, Melenotte et al13 described an association between C burnetii infection and non–Hodgkin lymphoma. In that study, C burnetii lymphadenitis and hemophagocytic syndrome were identified as risk factors of lymphoma. To determine how bone marrow and lymph nodes could influence lymphomagenesis, further investigations are warranted. In any event, because lymphadenitis was the unique infective focus of C burnetii in 23 patients (0.9%) in our cohort, and because lymphadenitis was identified in 43.9% of the cases with PET scan imaging, the latter is justified to identify C burnetii lymphadenitis as a prelymphomatous stage.3 Coagulation Disorder and Elevated aCL Levels Acute Q Fever Endocarditis First described as a subacute disease and later considered a fatal chronic disease, heart valvular injury is now an acute Q fever clinical entity.30,31 Acute Q fever endocarditis is associated with evolution to persistent C burnetii endocarditis and must be seriously considered with immediate and systematic TTE.17

itis First described as a subacute disease and later considered a fatal chronic disease, heart valvular injury is now an acute Q fever clinical entity.30,31 Acute Q fever endocarditis is associated with evolution to persistent C burnetii endocarditis and must be seriously considered with immediate and systematic TTE.17 Atypical Presentation of Q Fever The proportion of neurologic involvement (1.3%) in our cohort is consistent with that in 14 previous international publications.2,32,33,34 Capillary thrombi and small perivascular hemorrhages have been described, and C burnetii has been identified by immunofluorescence in the brain.35,36,37 Regarding pericarditis and myocarditis, systematic TTE and the systematic prescription of serologic findings for C burnetii have improved the diagnosis of Q fever pericarditis.38 Among 22 cases of myocarditis described in the literature, myocardial necrosis has been anecdotally reported.9,39,40 Cases of optic neuritis (n = 6) and uveitis (n = 21) reported in the literature have been considered to be inflammatory phenomena triggered by the bacterium without evidence of C burnetii in the intraocular specimen.41,42,43,44,45 In the literature, 17 cases of C burnetii alithiasic cholecystitis were reported. One case was confirmed by positive polymerase chain reaction results for C burnetii in the gallbladder, and 2 were associated with antiphospholipid antibodies.46,47,48,49,50,51,52,53,54,55,56 Thirteen cases of Q fever hemophagocytic syndrome associated with acute Q fever have been reported in the literature, and an increase in aCL antibodies was reported in only 1 case.57,58,59,60,61 Coxiella burnetii interstitial lung disease, which was first described after outbreaks in the United States and Russia, is a rare and severe persistent focal C burnetii infection with advanced fibrotic lesions and poor clinical outcome.62,63,64 First described in 1983 by Janigan and Marrie,65 pseudotumors of the lung are rare manifestations of C burnetii.66 After resection to exclude a tumoral process, histologic findings showed macrophage infiltration with C burnetii.65,66 Finally, large vessel vasculitis in association with C burnetii infection has been described in 5 case reports.67,68,69,70 For one of these cases, a high IgG aCL titer was observed.68

of C burnetii.66 After resection to exclude a tumoral process, histologic findings showed macrophage infiltration with C burnetii.65,66 Finally, large vessel vasculitis in association with C burnetii infection has been described in 5 case reports.67,68,69,70 For one of these cases, a high IgG aCL titer was observed.68 Limitations Some limitations of the study need to be acknowledged. Six hundred fifteen patients with acute Q fever (25.3%) were lost to follow-up because in most cases their clinical course was favorable, and they no longer consulted their referring physician. In addition, cardiovascular C burnetii infections were probably overrepresented in this cohort because, as a reference center, we are solicited for severe C burnetii infections. Conversely, the mortality rate might be underestimated because of potential loss to follow-up.

no longer consulted their referring physician. In addition, cardiovascular C burnetii infections were probably overrepresented in this cohort because, as a reference center, we are solicited for severe C burnetii infections. Conversely, the mortality rate might be underestimated because of potential loss to follow-up. Conclusions Based on the new definition criteria, hitherto neglected foci of infection include the lymphatic system (ie, bone marrow, lymphadenitis) with a risk of lymphoma. Cardiovascular infections were the main fatal complications, highlighting the importance of routine screening of valvular heart disease and vascular anomalies during acute Q fever. Routine screening for aCL antibodies during acute Q fever can help prevent complications. Further investigations are necessary to evaluate the addition of hydroxychloroquine sulfate to doxycycline in cases of elevated aCL titers. A PET scan could be performed for all patients with suspected persistent focalized infection for early diagnosis of vascular and lymphatic infections associated with death and lymphoma, respectively. Supplement. eFigure 1. Serological Test Performed Each Year in the French National Reference Center of Coxiella burnetii Infection eFigure 2. Standardized Questionnaire for Q Fever Cases in the French National Reference Center eFigure 3. Age and Sex Distribution at Diagnosis of C burnetii Infection eFigure 4. Number of Cases of Persistent Focalized Infection Over Time Regarding the Use of Systematic TTE (2001) and PET Scanning (2009)

Supplement. eFigure 1. Serological Test Performed Each Year in the French National Reference Center of Coxiella burnetii Infection eFigure 2. Standardized Questionnaire for Q Fever Cases in the French National Reference Center eFigure 3. Age and Sex Distribution at Diagnosis of C burnetii Infection eFigure 4. Number of Cases of Persistent Focalized Infection Over Time Regarding the Use of Systematic TTE (2001) and PET Scanning (2009) eFigure 5. Focalized Persistent C burnetii Lymphadenitis as the Unique Focus of C burnetii Persistent Infection eFigure 6. Lymphoma and Q Fever eTable 1. Diagnostic Criteria of C burnetii Persistent Focalized Infection eTable 2. Definition Criteria for Patients With C burnetii Persistent Infection and Interstitial Lung Diseases (ILD) eTable 3. Diagnostic Criteria of C burnetii Acute Endocarditis eTable 4. Geographic Origin of Serum Sample eTable 5. Immunosuppression Characteristics of Patients (n = 91) eTable 6. Clinical Manifestation of Q Fever During Pregnancy (n = 36) eTable 7. Clinical Presentation of Acute Q Fever in 1806 Patients eTable 8. Patients With Q Fever Lymphadenitis (n = 97) eTable 9. Clinical Presentation of Persistent C burnetii Infections in 766 Patients eTable 10. Osteoarticular Infection in Q Fever (n = 56) eTable 11. Diagnosis of C burnetii Osteoarticular Infection eTable 12. C burnetii Infection in Children (n = 58) eTable 13. ROC Analysis of IgG Anticardiolipin Antibodies and Acute Q Fever Complications Click here for additional data file.

Introduction Ductal carcinoma in situ (DCIS) refers to the histologic appearance of cancer cells within the breast ductule and/or lobule without evidence of cancer present beyond the basement membrane.1 This condition is generally identified in asymptomatic women in the context of screening mammography, and the incidence of DCIS in a population closely mirrors the extent of mammographic screening.2 In about 15% of cases of DCIS treated with breast-conserving surgery, the woman will experience an in-breast invasive recurrence in the same breast within 15 years.3 In about 6% of cases, women with DCIS will develop a contralateral invasive breast cancer within 15 years.3 In about 3% of cases, women with DCIS will die of breast cancer within 15 years.4 The risk of death from breast cancer increases greatly after an in-breast invasive recurrence; however, about 50% of women who die of breast cancer after DCIS have no record of an invasive recurrence.4

sive breast cancer within 15 years.3 In about 3% of cases, women with DCIS will die of breast cancer within 15 years.4 The risk of death from breast cancer increases greatly after an in-breast invasive recurrence; however, about 50% of women who die of breast cancer after DCIS have no record of an invasive recurrence.4 The dual goals of treatment are to prevent invasive local recurrence and to reduce death from breast cancer. The risk of death from breast cancer for patients with DCIS is approximately the same for women treated with mastectomy as it is for those treated with lumpectomy without radiotherapy, despite the fact that women in the latter group experience many more local recurrences.3,4,5,6,7 There is emerging evidence that, after a diagnosis of DCIS, the addition of radiotherapy to lumpectomy reduces the risk of death from breast cancer (as well as reducing the risk of local recurrence).8 Because of the low mortality associated with DCIS, it is difficult to study deaths from DCIS using small cohort studies or randomized trials. As a result, most clinical trials have been designed to study local recurrence. It is challenging to study mortality because the effect sizes are small and it is necessary to compare groups of women with similar risk profiles, ie, hazard ratios must be adjusted for variations in both pathologic features and treatments. We conducted a historical cohort study of women with pure DCIS (ie, without microinvasion) using the Surveillance, Epidemiology, and End Results (SEER) database. We extracted data on age and year of diagnosis, tumor size, tumor grade, treatments (surgery and radiation), and outcomes (local invasive recurrence, contralateral invasive breast cancer, and death from breast cancer). We sought to measure the extent to which radiotherapy is associated with a reduced risk of breast cancer death in this cohort of women and to identify subgroups of women who might benefit from radiotherapy the most.

utcomes (local invasive recurrence, contralateral invasive breast cancer, and death from breast cancer). We sought to measure the extent to which radiotherapy is associated with a reduced risk of breast cancer death in this cohort of women and to identify subgroups of women who might benefit from radiotherapy the most. Methods We used SEER*Stat statistical software version 8.3.4 to conduct a case-listing session and retrieved all cases of first primary DCIS (stage 0) diagnosed between 1998 and 2014 in the SEER 18 registries research database (November 2016 submission). We selected all cases with the American Joint Committee on Cancer primary tumor classification Tis (carcinoma in situ; no evidence of an invasive component). Among the cases classified as Tis, we excluded those associated with lobular carcinoma in situ, nonepithelial histologies, Paget disease of the nipple, or diffuse DCIS. We also excluded cases with unknown laterality, unknown or no surgical intervention on the primary tumor, and unknown radiation treatment status. Information on exclusions is provided in eTable 1 in the Supplement. Because patients cannot be identified, the research ethics board of the Women’s College Hospital exempted this study from review, and patient informed consent was not required. This article follows the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for cohort studies.

Because patients cannot be identified, the research ethics board of the Women’s College Hospital exempted this study from review, and patient informed consent was not required. This article follows the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for cohort studies. For each case, we retrieved information on the year of breast cancer diagnosis, age at diagnosis, ethnicity, household income, tumor laterality, tumor size, tumor grade, estrogen receptor (ER) status, progesterone receptor status, use of radiotherapy, use of chemotherapy, type of surgery, and cause of death. We assessed the vital status at the time of last follow-up. We extracted the information on survival time from the variable survival time months. The SEER*Stat program estimates survival time by subtracting the date of diagnosis from the date of last contact (the study cutoff). For each case we linked all additional cancer events that followed the DCIS diagnosis. Ipsilateral invasive recurrence was defined as the earliest new primary record that was an invasive breast cancer (stage I to IV) that occurred in the same breast as the DCIS. We retrieved all tumor characteristics and treatments for the ipsilateral invasive recurrence. We defined 3 time intervals: time from DCIS to end of follow-up, time from DCIS to ipsilateral invasive recurrence, and time from DCIS to contralateral invasive breast cancer. Outcome events were breast cancer–specific mortality, ipsilateral invasive recurrence, and contralateral invasive breast cancer, respectively.

. We defined 3 time intervals: time from DCIS to end of follow-up, time from DCIS to ipsilateral invasive recurrence, and time from DCIS to contralateral invasive breast cancer. Outcome events were breast cancer–specific mortality, ipsilateral invasive recurrence, and contralateral invasive breast cancer, respectively. Study participants were categorized into 3 groups: mastectomy, lumpectomy without radiation, and lumpectomy with radiation. The groups were compared for a range of demographic, pathologic, and treatment variables and differences were evaluated using standardized differences. Matching We conducted 3 separate cohort comparisons using 1:1 matching: lumpectomy with radiation vs lumpectomy without radiation, lumpectomy without radiation vs mastectomy, and lumpectomy with radiation vs mastectomy. In each analysis, patients were matched on year of diagnosis (same year), age at diagnosis (within 2 years), tumor grade (I, II, III, or IV), ER status (positive, negative, or unknown), and propensity score. The propensity score took into account ethnicity, household income, tumor size, and progesterone receptor status. Caliper matching was done by matching participants who were within 0.2 times the standard deviation of their propensity score.9 A standardized difference of greater than 0.1 was considered a meaningful imbalance between comparison groups.10 Variable distributions for the matched cohorts are available in eTables 2, 3, and 4 in the Supplement.

ne by matching participants who were within 0.2 times the standard deviation of their propensity score.9 A standardized difference of greater than 0.1 was considered a meaningful imbalance between comparison groups.10 Variable distributions for the matched cohorts are available in eTables 2, 3, and 4 in the Supplement. Statistical Analysis We estimated the crude cumulative breast cancer–specific mortality rates for the 3 treatment-matched subgroups using the Kaplan-Meier method. We then estimated the crude rates for invasive local recurrence (from the date of diagnosis of DCIS to the date of ipsilateral invasive recurrence for the 3 treatment groups). Hazard ratios (HRs) were calculated using the Cox proportional hazards model in SAS statistical software, version 9.4 (SAS Institute Inc). Patients were followed up from the date of DCIS until the outcome of interest, the end of follow-up, death from another cause, or loss to follow-up. Adjusted HRs were generated using a Cox proportional hazards model on the matched subgroups. Among all participants treated with lumpectomy, we conducted subgroup comparisons by age, ethnicity, ER status, tumor grade, and tumor size (using inverse probability of treatment weighting) to determine the extent to which radiation was associated with decreased risk of death in these various subgroups. Stabilized inverse probability of treatment–weighted estimates were truncated at the 1st and 99th percentile.10,11

ER status, tumor grade, and tumor size (using inverse probability of treatment weighting) to determine the extent to which radiation was associated with decreased risk of death in these various subgroups. Stabilized inverse probability of treatment–weighted estimates were truncated at the 1st and 99th percentile.10,11 Breast cancer–specific mortality hazard rates were calculated for each year following DCIS diagnosis. We compared hazard rates for the entire 15-year interval assuming a proportional hazard and then for three 5-year subintervals (0-5 years, 5-10 years, and 10-15 years after diagnosis). In this analysis, the hazard rate was permitted to vary between intervals but was proportional within a given interval. A log-rank test was used to compare differences across groups with the Kaplan-Meier method. We generated 95% confidence limits for all HRs in the analysis. All P values were 2-tailed and statistically significant at a level of .05 or less. Results Among the 140 366 patients with DCIS in the cohort (109 712 [78.2%] white; mean [SD] age, 58.8 [12.3] years), 100 371 patients (71.5%) were treated with lumpectomy (35 070 [25.0%] with lumpectomy alone and 65 301 [46.5%] with lumpectomy and radiotherapy) and 39 995 patients (28.5%) were treated with mastectomy (Table 1). The patients treated with mastectomy were slightly younger on average than those treated with lumpectomy (mean [SD] age, 56.5 [12.6] years vs 59.8 [12.0] years). The likelihood of having a mastectomy increased with tumor size and with tumor grade.

apy) and 39 995 patients (28.5%) were treated with mastectomy (Table 1). The patients treated with mastectomy were slightly younger on average than those treated with lumpectomy (mean [SD] age, 56.5 [12.6] years vs 59.8 [12.0] years). The likelihood of having a mastectomy increased with tumor size and with tumor grade. Table 1. Baseline Characteristics of All Patients With Ductal Carcinoma In Situ, According to Treatment Group Value No.

apy) and 39 995 patients (28.5%) were treated with mastectomy (Table 1). The patients treated with mastectomy were slightly younger on average than those treated with lumpectomy (mean [SD] age, 56.5 [12.6] years vs 59.8 [12.0] years). The likelihood of having a mastectomy increased with tumor size and with tumor grade. Table 1. Baseline Characteristics of All Patients With Ductal Carcinoma In Situ, According to Treatment Group Value No. (%) P Valuea Overall Lumpectomy Alone Lumpectomy Plus Radiotherapy Mastectomy Patients 140 366 (100) 35 070 (25.0) 65 301 (46.5) 39 995 (28.5) Year of diagnosis 1998-2004 47 675 (34.0) 13 619 (38.8) 20 343 (31.2) 13 713 (34.3) <.001 2005-2009 45 502 (32.4) 10 923 (31.1) 21 957 (33.6) 12 622 (31.6) 2010-2014 47 189 (33.6) 10 528 (30.0) 23 001 (35.2) 13 660 (34.2) Age at diagnosis, y Mean (SD) 58.8 (12.3) 62.1 (13.2) 58.5 (11.1) 56.5 (12.6) <.001 Median (IQR) 58.0 (49.0-68.0) 61.0 (52.0-72.0) 58.0 (50.0-67.0) 55.0 (47.0-66.0) <.001 <40 4657 (3.3) 780 (2.2) 1414 (2.2) 2463 (6.2) <.001 40-49 31 047 (22.1) 6114 (17.4) 14 014 (21.5) 10 919 (27.3) 50-59 40 338 (28.7) 8947 (25.5) 20 277 (31.1) 11 114 (27.8) 60-69 34 504 (24.6) 8151 (23.2) 17 856 (27.3) 8497 (21.2) 70-79 22 116 (15.8) 7135 (20.3) 9733 (14.9) 5248 (13.1) ≥80 7704 (5.5) 3943 (11.2) 2007 (3.1) 1754 (4.4) Ethnicity White 109 712 (78.2) 27 765 (79.2) 51 261 (78.5) 30 686 (76.7) <.001 Black 14 904 (10.6) 3542 (10.1) 6910 (10.6) 4452 (11.1) East Asian 5983 (4.3) 1336 (3.8) 2915 (4.5) 1732 (4.3) Southeast Asian 5364 (3.8) 1183 (3.4) 2412 (3.7) 1769 (4.4) Other or unknown 4403 (3.1) 1244 (3.5) 1803 (2.8) 1356 (3.4) Annual household income, $ <30 000 38 844 (27.7) 8282 (23.6) 18 426 (28.2) 12 136 (30.3) <.001 30 000-34 999 35 561 (25.3) 11 165 (31.8) 14 559 (22.3) 9837 (24.6) 35 000-39 999 27 795 (19.8) 6210 (17.7) 13 752 (21.1) 7833 (19.6) ≥40 000 38 153 (27.2) 9408 (26.8) 18 561 (28.4) 10 184 (25.5) Unknown 13 (0.0) 5 (0.0) 3 (0.0) 5 (0.0) Tumor grade I 16 620 (11.8) 6198 (17.7) 7166 (11.0) 3256 (8.1) <.001 II 48 404 (34.5) 13 259 (37.8) 22 859 (35.0) 12 286 (30.7) III or IV 53 597 (38.2) 8696 (24.8) 26 276 (40.2) 18 625 (46.6) Unknown 21 745 (15.5) 6917 (19.7) 9000 (13.8) 5828 (14.6) Tumor size, cm Mean (SD) 1.7 (2.1) 1.3 (2.0) 1.4 (1.5) 2.6 (2.7) <.001 Median (IQR) 1.1 (0.6-2.0) 0.8 (0.5-1.5) 1.0 (0.5-1.7) 1.8 (1.0-3.5) <.001 <1.0 42 267 (30.1) 12 861 (36.7) 22 381 (34.3) 7025 (17.6) <.001 1.0-1.9 28 500 (20.3) 5814 (16.6) 15 208 (23.3) 7478 (18.7) 2.0-2.9 12 434 (8.9) 2094 (6.0) 5700 (8.7) 4640 (11.6) 3.0-4.9 9263 (6.6) 1385 (3.9) 3450 (5.3) 4428 (11.1) ≥5.0 6823 (4.9) 874 (2.5) 1421 (2.2) 4528 (11.3) Unknown

(0.5-1.7) 1.8 (1.0-3.5) <.001 <1.0 42 267 (30.1) 12 861 (36.7) 22 381 (34.3) 7025 (17.6) <.001 1.0-1.9 28 500 (20.3) 5814 (16.6) 15 208 (23.3) 7478 (18.7) 2.0-2.9 12 434 (8.9) 2094 (6.0) 5700 (8.7) 4640 (11.6) 3.0-4.9 9263 (6.6) 1385 (3.9) 3450 (5.3) 4428 (11.1) ≥5.0 6823 (4.9) 874 (2.5) 1421 (2.2) 4528 (11.3) Unknown 41 079 (29.3) 12 042 (34.3) 17 141 (26.2) 11 896 (29.7) Estrogen receptor status Negative 13 823 (9.8) 2021 (5.8) 6576 (10.1) 5226 (13.1) <.001 Positive 77 023 (54.9) 17 050 (48.6) 39 242 (60.1) 20 731 (51.8) Unknown 49 520 (35.3) 15 999 (45.6) 19 483 (29.8) 14 038 (35.1) Progesterone receptor status Negative 21 482 (15.3) 3399 (9.7) 10 497 (16.1) 7586 (19.0) <.001 Positive 63 877 (45.5) 14 364 (41.0) 32 690 (50.1) 16 823 (42.1) Unknown 55 007 (39.2) 17 307 (49.3) 22 114 (33.9) 15 586 (39.0) Abbreviation: IQR, interquartile range. a Variables statistically different across all treatment combinations. We used χ2 tests for categorical variables and t tests and Mann-Whitney tests for continuous variables. Among the patients treated with lumpectomy, 65 301 (65%) received radiotherapy and 35 070 (35%) did not. Among these patients, those who received radiotherapy were on average 3.6 years younger than those who did not (mean [SD] age, 58.5 [11.1] years vs 62.1 [13.2] years) (Table 1). The use of radiotherapy also increased with increasing tumor grade. Radiotherapy was less commonly used for women with cancers of less than 1 cm (64%) than for women with larger cancers (72%).

apy were on average 3.6 years younger than those who did not (mean [SD] age, 58.5 [11.1] years vs 62.1 [13.2] years) (Table 1). The use of radiotherapy also increased with increasing tumor grade. Radiotherapy was less commonly used for women with cancers of less than 1 cm (64%) than for women with larger cancers (72%). For all participants combined, the cumulative mortality from breast cancer at 15 years was 2.03% (annual rates provided in eTable 5 in the Supplement). The risk was 2.26% for participants treated with mastectomy and 1.94% for participants treated with lumpectomy. The actuarial 15-year mortality rate for women who had a mastectomy (2.26%) was similar to the rate for women who had lumpectomy without radiotherapy (2.33%). The adjusted HR for death for mastectomy vs lumpectomy alone (based on 20 832 propensity-matched pairs) was 0.91 (95% CI, 0.78-1.05) (Table 2; eFigure 1 in the Supplement).

al 15-year mortality rate for women who had a mastectomy (2.26%) was similar to the rate for women who had lumpectomy without radiotherapy (2.33%). The adjusted HR for death for mastectomy vs lumpectomy alone (based on 20 832 propensity-matched pairs) was 0.91 (95% CI, 0.78-1.05) (Table 2; eFigure 1 in the Supplement). Table 2. Hazard Ratios Associated With Radiation and Extent of Surgery in 1:1 Propensity-Matched Subgroups Comparison Hazard Ratio (95% CI) P Value Lumpectomy plus radiotherapy vs lumpectomy alone 0.77 (0.67-0.88) <.001 Mastectomy vs lumpectomy alone 0.91 (0.78-1.05) .20 Lumpectomy plus radiotherapy vs mastectomy 0.75 (0.65-0.87) <.001 Among patients treated with lumpectomy, the actuarial 15-year mortality rate was 25% less for those who received radiotherapy than for those who did not (1.74% vs 2.33%). The adjusted HR associated with radiotherapy (based on 29 465 propensity-matched pairs) was 0.77 (95% CI, 0.67-0.88; P < .001) (Table 2 and the Figure). The adjusted HR for death associated with lumpectomy and radiotherapy vs mastectomy (based on 29 865 propensity-matched pairs) was 0.75 (95% CI, 0.65-0.87; P < .001). The results of the adjusted analysis did not change substantially when competing risks of death were considered in the model (model 2 in eTable 6 in the Supplement) or when inverse probability of treatment weighting was used to compare treatment groups (model 3 in eTable 6 in the Supplement). In the matched lumpectomy cohort, radiotherapy was associated with an absolute reduction in local recurrences of 2.82% (eTable 7 and eFigure 2 in the Supplement) and a reduction in deaths from breast cancer of 0.27% (eTable 7 in the Supplement; Figure). In the matched comparison of patients treated with lumpectomy and radiation vs mastectomy, mastectomy was associated with an absolute reduction in local recurrences of 4.31% (eTable 8 and eFigure 3 in the Supplement) and an absolute increase in breast cancer deaths of 0.28% (eTable 8 and eFigure 4 in the Supplement).

. In the matched comparison of patients treated with lumpectomy and radiation vs mastectomy, mastectomy was associated with an absolute reduction in local recurrences of 4.31% (eTable 8 and eFigure 3 in the Supplement) and an absolute increase in breast cancer deaths of 0.28% (eTable 8 and eFigure 4 in the Supplement). Figure. Breast Cancer–Specific Mortality After Ductal Carcinoma In Situ in Propensity-Matched Patients Treated With Lumpectomy Alone vs Lumpectomy and Radiotherapy The protective effect of radiotherapy on mortality was measured for different subgroups of patients who underwent lumpectomy using inverse probability of treatment weighting (Table 3). The HR was 0.59 (95% CI, 0.43-0.80) for patients younger than 50 years and 0.86 (95% CI, 0.73-1.01) for patients aged 50 years and older. The HR was 0.67 (95% CI, 0.51-0.87) for patients with ER-positive cancers, 0.50 (95% CI, 0.32-0.78) for patients with ER-negative cancers, and 0.93 (95% CI, 0.77-1.13) for patients with unknown ER status. The HR was 0.69 (95% CI, 0.50-0.96) for black women and 0.83 (95% CI, 0.71-0.98) for white women. The HR was 1.00 (95% CI, 0.79-1.27) for patients with low- or intermediate-grade tumors (grade I or II) and 0.59 (95% CI, 0.47-0.75) for patients with high-grade tumors (grade III or IV).

r patients with unknown ER status. The HR was 0.69 (95% CI, 0.50-0.96) for black women and 0.83 (95% CI, 0.71-0.98) for white women. The HR was 1.00 (95% CI, 0.79-1.27) for patients with low- or intermediate-grade tumors (grade I or II) and 0.59 (95% CI, 0.47-0.75) for patients with high-grade tumors (grade III or IV). Table 3. Hazard Ratios Associated With Lumpectomy and Radiotherapy vs Lumpectomy Alone for Various Patient Subgroups (Adjusted Using Inverse Probability of Treatment Weighting) Subgroup Comparison Hazard Ratio (95% CI) P Value Estrogen receptor statusa Positive Lumpectomy alone 1 [Reference] .003 Lumpectomy plus radiotherapy 0.67 (0.51-0.87) Negative Lumpectomy alone 1 [Reference] .002 Lumpectomy plus radiotherapy 0.50 (0.32-0.78) Unknown Lumpectomy alone 1 [Reference] .46 Lumpectomy plus radiotherapy 0.93 (0.77-1.13) Age at diagnosis, y <40 Lumpectomy alone 1 [Reference] .09 Lumpectomy plus radiotherapy 0.54 (0.26-1.09) 40-49 Lumpectomy alone 1 [Reference] .004 Lumpectomy plus radiotherapy 0.59 (0.42-0.84) 50-59 Lumpectomy alone 1 [Reference] .01 Lumpectomy plus radiotherapy 0.68 (0.50-0.92) ≥60 Lumpectomy alone 1 [Reference] .29 Lumpectomy plus radiotherapy 0.90 (0.74-1.09) Ethnicity White Lumpectomy alone 1 [Reference] .03 Lumpectomy plus radiotherapy 0.83 (0.71-0.98) Black Lumpectomy alone 1 [Reference] .03 Lumpectomy plus radiotherapy 0.69 (0.50-0.96) Tumor gradea I Lumpectomy alone 1 [Reference] .09 Lumpectomy plus radiotherapy 1.54 (0.94-2.53) II Lumpectomy alone 1 [Reference] .31 Lumpectomy plus radiotherapy 0.87 (0.67-1.14) III or IV Lumpectomy alone 1 [Reference] <.001 Lumpectomy plus radiotherapy 0.59 (0.47-0.75) Tumor size, cma <1.0 Lumpectomy alone 1 [Reference] .58 Lumpectomy plus radiotherapy 0.92 (0.68-1.24) 1.0-1.9 Lumpectomy alone 1 [Reference] .01 Lumpectomy plus radiotherapy 0.68 (0.50-0.92) 2.0-2.9 Lumpectomy alone 1 [Reference] .24 Lumpectomy plus radiotherapy 0.75 (0.47-1.21) 3.0-4.9 Lumpectomy alone 1 [Reference] .07 Lumpectomy plus radiotherapy 0.54 (0.27-1.06) ≥5.0 Lumpectomy alone 1 [Reference] <.001 Lumpectomy plus radiotherapy 0.20 (0.09-0.49) a Global test for interaction statistically significant (P < .05).

ectomy alone 1 [Reference] .24 Lumpectomy plus radiotherapy 0.75 (0.47-1.21) 3.0-4.9 Lumpectomy alone 1 [Reference] .07 Lumpectomy plus radiotherapy 0.54 (0.27-1.06) ≥5.0 Lumpectomy alone 1 [Reference] <.001 Lumpectomy plus radiotherapy 0.20 (0.09-0.49) a Global test for interaction statistically significant (P < .05). In the matched cohort of patients who underwent lumpectomy, actuarial breast cancer mortality at 15 years was reduced by 0.27% with radiotherapy (from 2.05% to 1.78%). The difference was greater than this for women younger than 50 years (1.59%; from 3.06% to 1.47%), black women (0.87%; from 4.28% to 3.41%), and women with ER-negative cancers (0.57%; from 2.99% to 2.42%). On average, 370 women would need to be treated with radiotherapy to save 1 life. This count was fewer for black women (115 treated) and for women younger than 50 years (63 treated).

s (1.59%; from 3.06% to 1.47%), black women (0.87%; from 4.28% to 3.41%), and women with ER-negative cancers (0.57%; from 2.99% to 2.42%). On average, 370 women would need to be treated with radiotherapy to save 1 life. This count was fewer for black women (115 treated) and for women younger than 50 years (63 treated). We sought to better characterize the time-dependent effect of the association between radiotherapy and mortality. To do this, we divided the follow-up period into three 5-year intervals and constructed interval-specific hazard rates and HRs for the matched lumpectomy cohort (Table 4). The risk of dying of breast cancer increased with time since DCIS diagnosis, from 76.4 per 100 000 person-years in the first interval to 179.1 per 100 000 person-years in the third interval. In contrast, the benefit of radiotherapy in terms of mortality reduction diminished with time; the hazard ratio was 0.71 (95% CI, 0.57-0.87) in the first interval and 1.06 (95% CI, 0.77-1.46) in the third interval. In the matched lumpectomy cohort, radiotherapy was also associated with a significant reduction in contralateral invasive breast cancers (HR, 0.91; 95% CI, 0.85-0.97).

ne 1 [Reference] .001 Lumpectomy plus radiotherapy 0.71 (0.57-0.87) 5.1-10.0 Lumpectomy alone 1 [Reference] .005 Lumpectomy plus radiotherapy 0.72 (0.58-0.91) 10.1-15.0 Lumpectomy alone 1 [Reference] .74 Lumpectomy plus radiotherapy 1.06 (0.77-1.46) a Global test for interaction not statistically significant (P = .31). Discussion Among patients with DCIS treated with lumpectomy, adjuvant radiation was associated with a 23% reduced risk of dying of breast cancer; the cumulative mortality at 15 years was 2.33% for patients with DCIS treated with lumpectomy alone and 1.74% for women treated with lumpectomy and radiotherapy (adjusted HR, 0.77; 95% CI, 0.67-0.88; P < .001). The relative risk reduction in mortality of 23% is substantial, but the absolute risk reduction was only 0.27%, and it is doubtful whether a benefit of this size is large enough to warrant radiotherapy. It would be necessary to treat 370 women to save 1 life. The mortality benefit for black women was larger (1 death prevented for every 115 women treated), but the small size of this difference makes it difficult to personalize treatment. We believe that the mortality benefit is attributable to radiotherapy and not to a baseline imbalance in pathologic features or treatments; we used matching and propensity scoring to generate comparable groups (eTables 2-4 in the Supplement). Women who received radiation were younger, on average, and were more likely to have high-grade cancers than the women who did not receive radiation (Table 1), but these factors were accounted for in the matched analysis.

used matching and propensity scoring to generate comparable groups (eTables 2-4 in the Supplement). Women who received radiation were younger, on average, and were more likely to have high-grade cancers than the women who did not receive radiation (Table 1), but these factors were accounted for in the matched analysis. In the 2010 Early Breast Cancer Trialists’ Collaborative Group (EBCTCG) overview of randomized trials evaluating radiotherapy after lumpectomy in women with DCIS,14 radiotherapy decreased ipsilateral breast events by one-half (HR, 0.46; P < .001), but had no effect on breast cancer mortality (HR, 1.22; P > .1). Many population-based studies examining the various treatments in patients with DCIS have confirmed a reduction in local recurrences with local therapies (mastectomy vs lumpectomy and lumpectomy plus radiotherapy vs lumpectomy alone)4,5,6,15; however, most have reported no significant difference in breast cancer mortality.4,5,6,7,8,15,16

ed studies examining the various treatments in patients with DCIS have confirmed a reduction in local recurrences with local therapies (mastectomy vs lumpectomy and lumpectomy plus radiotherapy vs lumpectomy alone)4,5,6,15; however, most have reported no significant difference in breast cancer mortality.4,5,6,7,8,15,16 In our previous analysis of the SEER DCIS cohort,4 we observed a nonsignificant decrease in breast cancer mortality associated with radiotherapy after lumpectomy (adjusted HR, 0.81; 95% CI, 0.63-1.04) and a nonsignificant increase in breast cancer mortality associated with mastectomy compared with lumpectomy (adjusted HR, 1.20; 95% CI, 0.96-1.50). The current analysis examines a larger cohort of patients, and we used a propensity score–based 1:1 matching approach to compare the treatment groups. This approach eliminates the potential influence of outliers in the data set. We report HRs similar in size to those of the previous study, but which now reach statistical significance (Table 2). In 2016, Sagara et al8 studied 32 144 lumpectomy-treated patients with DCIS diagnosed between 1998 and 2007 in the SEER database. In a multivariable analysis adjusted by patient age, year, patient race, tumor size, and tumor grade, the HR for death associated with radiotherapy was 0.73 (95% CI, 0.62-0.88). However, this study did not include patients treated with mastectomy; we show, to our knowledge for the first time, a survival benefit of lumpectomy plus radiotherapy compared with mastectomy (HR, 0.75; 95% CI, 0.65-0.87; P < .001) (Table 2).

, the HR for death associated with radiotherapy was 0.73 (95% CI, 0.62-0.88). However, this study did not include patients treated with mastectomy; we show, to our knowledge for the first time, a survival benefit of lumpectomy plus radiotherapy compared with mastectomy (HR, 0.75; 95% CI, 0.65-0.87; P < .001) (Table 2). In theory, there are various mechanisms whereby radiation might reduce mortality in patients with DCIS. If radiation exerts its effect through local control, ie, if radiation prevents local recurrences, and if local recurrences are the source of metastases, then radiation should prevent some deaths. Elsewhere we have argued against this model.17 It is often stated, based on results of the EBCTCG study of invasive breast cancer,18,19 that for every 4 local recurrences prevented, 1 death is prevented (radiation-prevented local recurrences and deaths in a ratio of 4 to 1). The association is insufficient to infer causality. In the present study, radiation after lumpectomy was associated with reductions in local recurrences by 2.82% and of deaths by 0.27%, ie, the ratio of local recurrences prevented to deaths prevented was approximately 10 to 1 (Figure; eTable 7 and eFigure 2 in the Supplement). However, we cannot infer that the decline in deaths was a consequence of avoiding recurrences because there is no direct evidence that the women who survived were those who avoided local recurrence. Moreover, in comparing the lumpectomy plus radiation cohort with the mastectomy cohort, we observed a marked decrease in local recurrences with mastectomy (4.31%), but an increase in deaths of 0.28% (eTable 8, eFigures 3 and 4 in the Supplement). If the salutary effect of radiation on mortality were effected through local control, we would expect to see the same effect (or a greater effect) with mastectomy.

observed a marked decrease in local recurrences with mastectomy (4.31%), but an increase in deaths of 0.28% (eTable 8, eFigures 3 and 4 in the Supplement). If the salutary effect of radiation on mortality were effected through local control, we would expect to see the same effect (or a greater effect) with mastectomy. Similar results have been reported for patients with invasive cancer. In the 7 trials comparing mastectomy alone with lumpectomy and radiotherapy among women with node-negative invasive breast cancer,19 the rate ratio for local recurrence was 0.54 (P < .001) and the rate ratio for breast cancer mortality was 0.98 (P = .80). Several studies in patients with early invasive breast cancer have shown that lumpectomy and radiotherapy combined are superior to mastectomy in terms of survival, despite being less effective in terms of local control.20,21,22,23

ce was 0.54 (P < .001) and the rate ratio for breast cancer mortality was 0.98 (P = .80). Several studies in patients with early invasive breast cancer have shown that lumpectomy and radiotherapy combined are superior to mastectomy in terms of survival, despite being less effective in terms of local control.20,21,22,23 These results support our conclusion that the survival benefits of radiotherapy seen in both patients with DCIS and patients with invasive breast cancer cannot be explained by improving local control. We must seek an alternative explanation, namely that radiation to the breast acts as a systemic therapy to eradicate subclinical latent metastases. If a patient dies of breast cancer following DCIS, it is reasonable to conclude that undetected metastatic deposits were present at the time of diagnosis, and that may lead to generalized metastatic clinical disease and death. Perhaps radiation induces an immune response or activates another defense mechanism, thereby preventing the emergence or expansion of subclinical metastatic clones.24 Possible considerations include radiation to the blood as it circulates through the breast, radiation to stromal elements in the breast matrix, and radiation scatter to tissues beyond the breast. These areas are deserving of future study.

thereby preventing the emergence or expansion of subclinical metastatic clones.24 Possible considerations include radiation to the blood as it circulates through the breast, radiation to stromal elements in the breast matrix, and radiation scatter to tissues beyond the breast. These areas are deserving of future study. Support for the notion that local radiation induces systemic antitumor effects is the observation of a significant reduction in contralateral invasive breast cancers in the matched comparison of lumpectomy and radiotherapy vs lumpectomy alone (HR, 0.91; 95% CI, 0.85 to 0.97) (eFigure 5 in the Supplement). A 2017 meta-analysis of all observational and randomized studies in patients with DCIS reported an HR for radiotherapy on contralateral breast cancer of 0.95 (95% CI, 0.44-1.82).25 Future studies are required to more closely examine this association. This study of patients with DCIS is ideal, as fewer patients will receive chemotherapy or other systemic therapies that could affect risk.

atients with DCIS reported an HR for radiotherapy on contralateral breast cancer of 0.95 (95% CI, 0.44-1.82).25 Future studies are required to more closely examine this association. This study of patients with DCIS is ideal, as fewer patients will receive chemotherapy or other systemic therapies that could affect risk. Limitations Our study has several inherent limitations. It has been acknowledged that the rates of local recurrence among patients with DCIS in SEER are lower than expected, but this should not affect the mortality results. We might have misclassified some of the cases of DCIS with microinvasion as pure DCIS. In the SEER database there are currently 13 cases of pure DCIS recorded for every case of DCIS with microinvasion.12 Including patients with DCIS with microinvasion should not affect the protective association with radiotherapy unless women with microinvasion were less likely to receive radiotherapy than those without microinvasion. Data were missing for many individuals for key variables, including tumor size, grade, and ER status. We did not have information on tamoxifen use. It has been reported that radiotherapy is underreported in the SEER database13; however, we do not think that there are false-positive reports of radiotherapy and we accept that the women who reported having radiotherapy were likely to have had it. Therefore, the effect of misclassification should be small. The treatments in the study population were not assigned at random, and there is always the possibility that the decision to undergo radiotherapy was associated with other favorable prognostic factors (latent confounding) related to the tumor, demographic factors, or the treatment itself. The matching process requires the exclusion of a significant proportion of the cohort; thus, the results may not be generalizable to all patients with DCIS.

ndergo radiotherapy was associated with other favorable prognostic factors (latent confounding) related to the tumor, demographic factors, or the treatment itself. The matching process requires the exclusion of a significant proportion of the cohort; thus, the results may not be generalizable to all patients with DCIS. Conclusions Among patients with DCIS, treatment with lumpectomy and radiotherapy is associated with a significant reduction in breast cancer mortality compared with either lumpectomy alone or mastectomy. Although the clinical benefit is small, it is intriguing that radiotherapy has this effect, which appears to be attributable to systemic activity rather than local control. How exactly radiotherapy affects survival is an important question that should be explored in future studies. Supplement. eTable 1. Excluded Cases of Stage 0 Breast Cancer Identified in SEER From 1998 to 2014 eTable 2. Matched DCIS Patients Treated With Lumpectomy Alone Versus Lumpectomy and Radiation eTable 3. Matched DCIS Patients Treated With Lumpectomy Alone Versus Mastectomy eTable 4. Matched DCIS Patients Treated With Lumpectomy and Radiation Versus Mastectomy eTable 5. Breast Cancer-Specific Mortality Rates From DCIS Diagnosis for the Entire Cohort and According to Treatment Group (Mastectomy, Lumpectomy Alone, Lumpectomy and Radiotherapy) eTable 6. Hazard Ratios Associated With Radiation/Extent of Surgery Using Multivariate Cox Regression, Inverse Probability Treatment Weighting and 1:1 Propensity Score-Based Matching, With and Without Accounting for Competing Risks of Death Among Matched Subgroups

eTable 5. Breast Cancer-Specific Mortality Rates From DCIS Diagnosis for the Entire Cohort and According to Treatment Group (Mastectomy, Lumpectomy Alone, Lumpectomy and Radiotherapy) eTable 6. Hazard Ratios Associated With Radiation/Extent of Surgery Using Multivariate Cox Regression, Inverse Probability Treatment Weighting and 1:1 Propensity Score-Based Matching, With and Without Accounting for Competing Risks of Death Among Matched Subgroups eTable 7. Breast Cancer-Specific Mortality And Ipsilateral Invasive Recurrence Rates From DCIS Diagnosis Among Matched DCIS Patients Treated With Lumpectomy Alone Versus Lumpectomy and Radiation eTable 8. Breast Cancer-Specific Mortality and Ipsilateral Invasive Recurrence Rates From DCIS Diagnosis Among Matched DCIS Patients Treated With Mastectomy Versus Lumpectomy and Radiation eFigure 1. Breast Cancer-Specific Survival Post-DCIS in Propensity-Matched Patients Treated With Lumpectomy Alone vs. With Mastectomy eFigure 2. Ipsilateral Invasive Recurrence-Free Survival Post-DCIS in Propensity-Matched Patients Treated With Lumpectomy and Radiation vs. With Lumpectomy Alone eFigure 3. Breast Cancer-Specific Survival Post-DCIS in Propensity-Matched Patients Treated With Lumpectomy and Radiation vs. With Mastectomy eFigure 4. Ipsilateral Invasive Recurrence-Free Survival Post-DCIS in Propensity-Matched Patients Treated With Lumpectomy and Radiation vs. With Mastectomy eFigure 5. Contralateral Invasive Breast Cancer-Free Survival Post-DCIS in Propensity-Matched Patients Treated With Lumpectomy and Radiation vs. With Lumpectomy Alone

eFigure 3. Breast Cancer-Specific Survival Post-DCIS in Propensity-Matched Patients Treated With Lumpectomy and Radiation vs. With Mastectomy eFigure 4. Ipsilateral Invasive Recurrence-Free Survival Post-DCIS in Propensity-Matched Patients Treated With Lumpectomy and Radiation vs. With Mastectomy eFigure 5. Contralateral Invasive Breast Cancer-Free Survival Post-DCIS in Propensity-Matched Patients Treated With Lumpectomy and Radiation vs. With Lumpectomy Alone Click here for additional data file.

Introduction Breast cancer is the most frequent cancer among women, with psychological, professional, and social effects.1,2 Diagnosis and treatment, as well the consequences on womanhood and everyday life, are often harsh. In this “hostile” environment, most women undergo breast surgery, which can lead to even more stress.3 Studies4,5 have reported that within the first postoperative days patients experience moderate or severe pain (10%-20%), nausea/vomiting (10%-15%), drowsiness (10%), sore throat (15%), and hoarseness, causing anxiety and discomfort. Therefore, reduction of these postoperative complications remains a medical challenge, with economic issues that include delayed discharge from the postanesthesia care unit (PACU) and center, medical and pharmacological rescue, and readmission.

s (10%), sore throat (15%), and hoarseness, causing anxiety and discomfort. Therefore, reduction of these postoperative complications remains a medical challenge, with economic issues that include delayed discharge from the postanesthesia care unit (PACU) and center, medical and pharmacological rescue, and readmission. Several strategies, such as premedication or pharmacological multimodal approaches, have been proposed to reduce these effects.6,7 Nonpharmacological alternatives, including music, relaxation therapy, or medical hypnosis, have been shown to decrease perioperative anxiety, pain, medication requirement, and nausea/vomiting.8,9,10,11 Among them, hypnosis seems to be a convenient method, with advantages that include an inexpensive and simple technique with no specific adverse effects. Observational studies and meta-analyses have reported the benefit of hypnosis on postoperative pain and other postoperative adverse effects, anesthetic intake, and duration of stay in the PACU.12,13,14 In 2007, Montgomery et al12 showed that a brief hypnosis session (15 minutes) performed within 1 hour before breast surgery reduced both adverse effects and cost. Hypnosis has become popular in numerous hospitals, leading to enhanced empathy and verbal and nonverbal communication in the preoperative and intraoperative settings.10,12,13,14 However, these hypnosis techniques and attitudes have not been validated by rigorous randomized studies to date.

oth adverse effects and cost. Hypnosis has become popular in numerous hospitals, leading to enhanced empathy and verbal and nonverbal communication in the preoperative and intraoperative settings.10,12,13,14 However, these hypnosis techniques and attitudes have not been validated by rigorous randomized studies to date. The HYPNOSEIN prospective single-blind randomized clinical trial evaluated the effect of hypnosis performed immediately before general anesthesia on main adverse effects in patients scheduled for day-case breast cancer surgery. The primary objective of this multicenter study in France was to evaluate the efficacy of a preoperative hypnosis session for reducing postoperative breast pain in patients who underwent minor breast cancer surgery, assessed using a visual analog scale (VAS) in the PACU. We also investigated the effect of hypnosis on nausea/vomiting, fatigue, comfort/well-being, and anxiety.

was to evaluate the efficacy of a preoperative hypnosis session for reducing postoperative breast pain in patients who underwent minor breast cancer surgery, assessed using a visual analog scale (VAS) in the PACU. We also investigated the effect of hypnosis on nausea/vomiting, fatigue, comfort/well-being, and anxiety. Methods Study Design and Setting This multicenter, prospective, randomized, single-blind, phase 3 clinical trial was conducted in the following centers: Montpellier Cancer Institute (ICM) and Montpellier University Hospital, Montpellier, France, and Paoli-Calmettes Institute, Marseille, France. The study was approved by the local institutional and ethics committee (Comité de Protection des Personnes [CPP], Sud Méditerrannée III, Nîmes, France). It was conducted in accord with the ethical standards of the Declaration of Helsinki15 and the Good Clinical Practice requirements.16 All participants provided written informed consent before the study began. This study followed the Consolidated Standards of Reporting Trials (CONSORT) reporting guidelines. The complete trial protocol is available in the Supplement.

ndards of the Declaration of Helsinki15 and the Good Clinical Practice requirements.16 All participants provided written informed consent before the study began. This study followed the Consolidated Standards of Reporting Trials (CONSORT) reporting guidelines. The complete trial protocol is available in the Supplement. Study Population Patients were included in the study if they were women older than 18 years with an indication for minor breast cancer surgery (cancer tumorectomy or tumorectomy with limited axillary node dissection). They had to be scheduled for day-case breast surgery (ambulatory, discharge on the same day, or discharge on the following day), with general anesthesia. Patients were excluded if they had an American Society of Anesthesiologists score of 4 or higher, body mass index (calculated as weight in kilograms divided by height in meters squared) less than 15 or greater than 45, or an indication for major surgery (mastectomy, bilateral surgery, full axillary dissection, major breast reconstruction, lumpectomy, or previous duration of surgery exceeding 2 hours). Also excluded were patients who refused hypnosis or had undergone previous surgery with hypnosis and patients with psychiatric disorders, chronic pain, or the use of therapeutic opioids for more than 3 months.

llary dissection, major breast reconstruction, lumpectomy, or previous duration of surgery exceeding 2 hours). Also excluded were patients who refused hypnosis or had undergone previous surgery with hypnosis and patients with psychiatric disorders, chronic pain, or the use of therapeutic opioids for more than 3 months. Randomization and Blinding On the day of surgery, eligible patients were randomly assigned (1:1) to the control arm or the hypnosis arm. Patients (but not the care teams) were blinded to the arm to which they were assigned. To reduce bias (excessive empathy in the control arm), 2 different anesthesiologist teams were involved in patient care. Patients in the control arm were managed by caregivers without formal hypnosis training, and patients in the hypnosis arm were cared for by a dedicated team of staff trained in hypnosis.

re assigned. To reduce bias (excessive empathy in the control arm), 2 different anesthesiologist teams were involved in patient care. Patients in the control arm were managed by caregivers without formal hypnosis training, and patients in the hypnosis arm were cared for by a dedicated team of staff trained in hypnosis. Intervention Preoperative Preparation and Hypnosis No premedication was given, and music therapy was not allowed. Patients in the control arm were prepared for surgery by the control team, with no specific recommendation on wording or nonverbal communication, and standard general anesthesia was used. In the hypnosis arm, patients were prepared for surgery by the hypnosis team, who started verbal and nonverbal communication immediately. A short individual hypnosis session (≤15 minutes) that was personalized to each patient was performed in all centers by a trained anesthesiologist who had been practicing the technique for more than 1 year. It was recommended that the anesthesiologist use sensorial language and paraverbal and rewording techniques to promote patient comfort/well-being according to her choice of a safe place or leisure activity. Family concerns and negative topics were avoided. During the hypnosis session, only the anesthesiologist talked with the patient. When the physician thought that the patient was ready for anesthesia, pharmacological anesthesia induction was performed in a manner similar to that for the control arm.

or leisure activity. Family concerns and negative topics were avoided. During the hypnosis session, only the anesthesiologist talked with the patient. When the physician thought that the patient was ready for anesthesia, pharmacological anesthesia induction was performed in a manner similar to that for the control arm. Anesthetic Management, Analgesia, and Postoperative Care Preoperative preparation, analgesia, surgery, and postoperative care were similar in the 2 arms in accord with the standards of the participating centers. Anesthesia induction was started intravenously with lidocaine hydrochloride (1 mg/kg), propofol (2-3 mg/kg), sufentanil citrate (0.1-0.2 μg/kg), and cis-atracurium (0.3-0.5 mg/kg), if necessary. The airway was secured with an endotracheal or supraglottic tube, and the lungs were ventilated with a mix of oxygen and air (50%-50%). The tidal volume was set to 6 to 8 mL/kg of ideal body weight. Anesthesia was maintained using sevoflurane (1%-2%), and additional intravenous sufentanil citrate (5 μg) was administered during surgery if the heartbeat or systolic blood pressure increased more than 20%. During surgery, analgesics were administered, including acetaminophen (1 g), ketoprofen (50 mg), and ketamine hydrochloride (0.1-0.2 mg/kg). No local anesthetic solution was administered.

enous sufentanil citrate (5 μg) was administered during surgery if the heartbeat or systolic blood pressure increased more than 20%. During surgery, analgesics were administered, including acetaminophen (1 g), ketoprofen (50 mg), and ketamine hydrochloride (0.1-0.2 mg/kg). No local anesthetic solution was administered. In the PACU, rescue analgesia (intravenous tramadol hydrochloride [50 mg]) was given to patients whose pain exceeded 3 on a 10-point numeric scale. If a score above 3 remained after 30 minutes, morphine sulfate titration (2 mg for body mass <60 kg and 3 mg for body mass ≥60 kg) at 5-minute intervals was administered until a score of 3 or lower was reached. In the first postoperative 48 hours, all patients received acetaminophen (1 g) and ketoprofen (50 mg) at 6-hour intervals by mouth. As nausea/vomiting prophylaxis, all patients with an Apfel score exceeding 2 were given intravenous dexamethasone sodium phosphate (4 mg) and droperidol (1.25 mg) after anesthesia induction.17,18 To treat nausea/vomiting after extubation, patients received intravenous ondansetron hydrochloride (4 mg). When necessary, oral ondansetron was administered during the first 48 hours after surgery (4 mg every 6 hours).

enous dexamethasone sodium phosphate (4 mg) and droperidol (1.25 mg) after anesthesia induction.17,18 To treat nausea/vomiting after extubation, patients received intravenous ondansetron hydrochloride (4 mg). When necessary, oral ondansetron was administered during the first 48 hours after surgery (4 mg every 6 hours). Clinical Assessment of Outcomes The primary end point was breast pain reduction, assessed immediately before discharge from the PACU by 2 on the VAS. Secondary end points were evaluation on the VAS of the following: postsurgical nausea/vomiting, fatigue, comfort/well-being, anxiety, PACU length of stay, operative time, use and dose of antiemetics, analgesic consumption, and number of failed day-case surgical procedures. Outcomes were recorded using the VAS (range, 0-10) by an independent clinical research associate before surgery, immediately before PACU discharge, at patient discharge on the evening of surgery, and at days 1, 7, and 30 after surgery. Assessments of outcomes in the PACU were performed by a blinded nurse. Patient satisfaction with care was evaluated the day after surgery on a scale of 0 to 10. Statistical Analysis Randomization was stratified according to center. The sample size calculation was based on a difference of at least 2 on the VAS in the PACU between the 2 arms in terms of pain severity. To detect such a difference with σ of 3.5, 2-sided α risk of 5%, and power of 90% (β = .10), 66 patients per arm were required. Considering 10% of nonevaluable patients, a recruitment total of 150 patients (75 per arm) was planned.19

nce of at least 2 on the VAS in the PACU between the 2 arms in terms of pain severity. To detect such a difference with σ of 3.5, 2-sided α risk of 5%, and power of 90% (β = .10), 66 patients per arm were required. Considering 10% of nonevaluable patients, a recruitment total of 150 patients (75 per arm) was planned.19 All analyses were performed on an intent-to-treat basis according to the statistical analysis plan (Supplement). No imputation method was used in the case of missing data. Data were recorded by treatment group. Continuous variables were described using means (SDs) and medians (ranges). For categorical variables, frequencies and percentages were computed. For both quantitative and qualitative variables, missing data were reported. To compare the distribution of continuous variables, t test or Kruskal-Wallis test was used. Categorical variables were compared using χ2 test or Fisher exact test. An exploratory subgroup analysis was conducted to assess the effect of treatment group perception on end points. During their PACU stay, patients were asked whether they thought that they had received hypnosis (“Do you think you received hypnosis before anesthesia?”), and 2 subgroups (perceived hypnosis and no perceived hypnosis) were considered based on their perception. Accordingly, the pain, fatigue, comfort/well-being, and anxiety in each subgroup were described. All tests were 2-sided, and P < .05 was considered statistically significant. Statistical analyses were performed using a software program (Stata, version 13.0; StataCorp LP).

An exploratory subgroup analysis was conducted to assess the effect of treatment group perception on end points. During their PACU stay, patients were asked whether they thought that they had received hypnosis (“Do you think you received hypnosis before anesthesia?”), and 2 subgroups (perceived hypnosis and no perceived hypnosis) were considered based on their perception. Accordingly, the pain, fatigue, comfort/well-being, and anxiety in each subgroup were described. All tests were 2-sided, and P < .05 was considered statistically significant. Statistical analyses were performed using a software program (Stata, version 13.0; StataCorp LP). Results Patients Between October 7, 2014, and April 5, 2016, a total of 150 patients were randomized, and 73 and 77 were allocated to the control and hypnosis arms, respectively (Figure). Two patients were excluded from the safety and efficacy analysis (1 patient was not treated according to the protocol, and 1 patient did not meet an eligibility criterion). Patient characteristics were well balanced between the 2 arms (Table 1). Briefly, the mean patient age was 57 years (range, 33-79 years) in the control arm and 53 years (range, 20-84 years) in the hypnosis arm. Most patients had undergone previous surgery (91.5% [65 of 71] in the control arm and 83.1% [64 of 77] in the hypnosis arm). Pain, nausea/vomiting, fatigue, comfort/well-being, and anxiety assessed at baseline immediately before entering the operating room were similar for patients in the 2 arms.

) in the hypnosis arm. Most patients had undergone previous surgery (91.5% [65 of 71] in the control arm and 83.1% [64 of 77] in the hypnosis arm). Pain, nausea/vomiting, fatigue, comfort/well-being, and anxiety assessed at baseline immediately before entering the operating room were similar for patients in the 2 arms. Figure. CONSORT Diagram of the Study CONSORT indicates Consolidated Standards of Reporting Trials. aPatients screened were reported in 2 participating centers and not in the third (center 3).

) in the hypnosis arm. Most patients had undergone previous surgery (91.5% [65 of 71] in the control arm and 83.1% [64 of 77] in the hypnosis arm). Pain, nausea/vomiting, fatigue, comfort/well-being, and anxiety assessed at baseline immediately before entering the operating room were similar for patients in the 2 arms. Figure. CONSORT Diagram of the Study CONSORT indicates Consolidated Standards of Reporting Trials. aPatients screened were reported in 2 participating centers and not in the third (center 3). Table 1. Patient Characteristics at Baselinea Variable Control Arm (n = 71) Hypnosis Arm (n = 77) Age, median (range), y 57 (33-79) 53 (20-84) BMI, median (range) 23 (15-32) 23 (16-38) Apfel score, median (range) 2 (1-4) 2 (1-4) Missing (n = 6) (n = 7) APAIS score, median (range) 14 (6-26) 13 (6-25) Missing (n = 3) (n = 1) ASA score, median (range) 1 (1-3) 1 (1-3) Education, No./total No. (%) <Bachelor degree 18/68 (26.5) 23/72 (31.9) Bachelor degree 17/68 (25.0) 12/72 (16.7) Postgraduate 33/68 (48.5) 37/72 (51.4) Missing (n = 3) (n = 5) Occupation, No./total No. (%) Active 31/71 (43.7) 41/75 (54.7) Inactive, unemployed, retired 40/71 (56.3) 34/75 (45.3) Missing (n = 0) (n = 2) Medical history, No. (%) 66 (93.0) 68 (88.3) Previous surgery, No. (%) 65 (91.5) 64 (83.1) Premedication, No./total No. (%) 18/67 (26.9) 11/74 (14.9) Missing (n = 4) (n = 3) VAS score, mean (SD)b Breast pain 0.39 (0.72) 0.51 (1.23) General pain 1.44 (1.78) 1.45 (2.05) Nausea/vomiting 0.24 (0.98) 0.22 (0.80) Fatigue 2.11 (2.05) 2.47 (2.33) Comfort/well-being 7.65 (2.04) 7.42 (2.13) Anxiety 4.48 (2.73) 3.84 (2.70) Abbreviations: APAIS, Amsterdam Preoperative Anxiety and Information Scale; ASA, American Society of Anesthesiologists; BMI, body mass index (calculated as weight in kilograms divided by height in meters squared); VAS, visual analog scale.

2.33) Comfort/well-being 7.65 (2.04) 7.42 (2.13) Anxiety 4.48 (2.73) 3.84 (2.70) Abbreviations: APAIS, Amsterdam Preoperative Anxiety and Information Scale; ASA, American Society of Anesthesiologists; BMI, body mass index (calculated as weight in kilograms divided by height in meters squared); VAS, visual analog scale. a All baseline characteristics were well balanced between the 2 arms. b Measured immediately before entering the operating room using a VAS graded from 0 to 10. Hypnosis Session In the hypnosis arm, the median duration of the hypnosis session was 6 minutes (range, 2-15 minutes). The use of intraoperative opioids and hypnotics was lower in the hypnosis arm. After surgery, 25.0% (14 of 56) of patients in the control arm thought that they had received hypnosis, and 21.7% (15 of 69) of patients in the hypnosis arm thought that they had not received hypnosis.

on was 6 minutes (range, 2-15 minutes). The use of intraoperative opioids and hypnotics was lower in the hypnosis arm. After surgery, 25.0% (14 of 56) of patients in the control arm thought that they had received hypnosis, and 21.7% (15 of 69) of patients in the hypnosis arm thought that they had not received hypnosis. Treatments Almost all patients (100% [71 of 71] in the control arm and 98.7% [76 of 77] in the hypnosis arm) underwent tumorectomy or quadrantectomy (Table 2). The median operative time was the same in the control arm (50 minutes [range, 15-193 minutes]) and the hypnosis arm (50 minutes [range, 15-220 minutes]). Drug consumption was similar overall except for doses of propofol and sufentanil, which were both lower in the hypnosis arm: the doses in the control vs hypnosis arms were 240 mg (range, 120-450 mg) vs 200 mg (range, 100-450 mg) (P = .01) for propofol and 19 μg (range, 10-30 μg) vs 15 μg (range, 10-25 μg) (P = .05) for sufentanil citrate. The proportion of patients who received lidocaine was lower in the control arm than in the hypnosis arm (46.5% [33 of 71] vs 68.8% [53 of 77], P = .008). The choice of airway management technique was also significantly different, favoring a noninvasive laryngeal mask in the hypnosis arm (70.1% [54 of 77]) vs in the control arm (53.5% [38 of 71]) (P = .04). The median PACU length of stay was 60 minutes (range, 20-290 minutes) in the control arm vs 46 minutes (range, 5-100 minutes) in the hypnosis arm (P = .002).

was also significantly different, favoring a noninvasive laryngeal mask in the hypnosis arm (70.1% [54 of 77]) vs in the control arm (53.5% [38 of 71]) (P = .04). The median PACU length of stay was 60 minutes (range, 20-290 minutes) in the control arm vs 46 minutes (range, 5-100 minutes) in the hypnosis arm (P = .002). Table 2. Surgical, Anesthesia, and Postoperative Data Variable Control Arm (n = 71) Hypnosis Arm (n = 77) P Value Surgical Data Surgery type, No. (%) Tumorectomy or quadrantectomy 71 (100) 76 (98.7) >.99 Oncoplasty 0 1 (1.3) Operative time, median (range), min 50 (15-193) 50 (15-220) .77 Missing (n = 8) (n = 6) Anesthesia Data Anesthesia duration, median (range), min 95 (25-232) 87 (41-210) .27 Missing (n = 15) (n = 9) Intubation type, No. (%) Orotracheal 33 (46.5) 23 (29.9) .04 Laryngeal mask 38 (53.5) 54 (70.1) Midazolam hydrochloride, No. (%) 4 (5.6) 6 (7.8) .75 Median (range), mg 1 (1-2) 1 (1-2) .83 Propofol, No. (%) 71 (100) 77 (100) >.99 Median (range), mg 240 (120-450) 200 (100-450) .01 Missing (n = 2) (n = 0) Sufentanil citrate, No. (%) 60 (84.5) 64 (83.1) >.99 Median (range), μg 19 (10-30) 15 (10-25) .05 Morphine sulfate reinjection, No. (%) 22 (31.0) 16 (20.8) .19 Median (range), μg 7.5 (0.1-460.0) 5.0 (0.1-15.0) .11 Ketamine hydrochloride, No. (%) 37 (52.1) 49 (63.6) .18 Median (range), mg 15 (10-30) 15 (10-40) .91 Lidocaine hydrochloride, No. (%) 33 (46.5) 53 (68.8) .008 Median (range), mg 50 (10-80) 50 (30-150) .18 Curare, No. (%) 12 (16.9) 9 (11.7) .48 Median (range), mg 6 (5-10) 6 (4-20) .90 Missing (n = 0) (n = 1) PONV prophylaxis, No. (%) 34 (47.9) 19 (24.7) .004 Postoperative Data PACU length of stay, median (range), min 60 (20-290) 46 (5-100) .002 Missing (n = 4) (n = 3) Antiemetic treatment, No./total No. (%) 4/69 (5.8) 8/77 (10.4) .38 Missing (n = 2) (n = 0) Including ondansetron hydrochloride, No./total No. (%) 3/4 (75.0) 7/8 (87.5) >.99 Analgesic consumption, No./total No. (%) 32/69 (46.4) 34/77 (44.2) .87 Missing (n = 2) (n = 0) Including morphine sulfate, No./total No. (%) 2/32 (6.3) 6/33 (18.2) .26 Missing (n = 0) (n = 1) Patient discharge, No./total No. (%) Day 0 52/70 (74.3) 61/77 (79.2) .56 Day 1a 13/70 (18.6) 12/77 (15.6) .91 Missing (n = 2) (n = 0) Patient satisfaction regarding anesthesia care and management, mean (SD)b 8.9 (1.5) 9.5 (1.1) .02 Missing (n = 7) (n = 2) Abbreviations: PACU, postanesthesia care unit; PONV, postoperative nausea/vomiting.

No./total No. (%) Day 0 52/70 (74.3) 61/77 (79.2) .56 Day 1a 13/70 (18.6) 12/77 (15.6) .91 Missing (n = 2) (n = 0) Patient satisfaction regarding anesthesia care and management, mean (SD)b 8.9 (1.5) 9.5 (1.1) .02 Missing (n = 7) (n = 2) Abbreviations: PACU, postanesthesia care unit; PONV, postoperative nausea/vomiting. a Patients discharged on day 1 were not discharged on day 0 only for logistic or familial reasons. No complications were reported. b Patient satisfaction was evaluated on the day after surgery. The postoperative nausea/vomiting (PONV) prophylaxis was different in the 2 arms: 47.9% (34 of 71) of patients received prophylaxis in the control arm vs 24.7% (19 of 77) in the hypnosis arm (P = .004), which prohibited interpretation of these results and analysis of the effect of hypnosis on PONV. One month after surgery, no serious adverse event or medical or surgical complication had been reported in either of the 2 arms. Postoperative Pain Primary End Point The mean (SD) breast pain score (range, 0-10), assessed immediately before PACU discharge, was 1.75 (1.59) in the control arm vs 2.63 (1.62) in the hypnosis arm (P = .004), favoring the control arm (difference, −0.88; 95% CI, −1.45 to −0.29) (Table 3). At discharge on the evening of surgery and with longer follow-up (postoperative days 1, 7, and 30), no difference in the breast and general pain scores was observed between the 2 arms.

trol arm vs 2.63 (1.62) in the hypnosis arm (P = .004), favoring the control arm (difference, −0.88; 95% CI, −1.45 to −0.29) (Table 3). At discharge on the evening of surgery and with longer follow-up (postoperative days 1, 7, and 30), no difference in the breast and general pain scores was observed between the 2 arms. Table 3. Patient Outcomes Assessed on the Day of Surgery (in the PACU and at Patient Discharge on the Evening of Surgery) and on Day 1 in the Control and Hypnosis Armsa Variable Control Arm (n = 71) Hypnosis Arm (n = 77) P Value Difference (95% CI)b Median (Range) Mean (SD) Median (Range) Mean (SD) Breast Pain PACU 2 (0 to 6) 1.75 (1.59) 3 (0 to 7) 2.63 (1.62) .004 −0.88 (−1.45 to −0.29) Missing (n = 18) (n = 7) Evening of surgery 2 (0 to 6) 2.14 (1.68) 2 (0 to 6) 2.27 (1.61) .62 −0.13 (−0.67 to 0.41) Missing (n = 2) (n = 2) Day 1 1 (0 to 8) 1.45 (1.58) 2 (0 to 7) 1.84 (1.77) .21 −0.39 (−0.95 to 0.27) Missing (n = 5) (n = 1) General Pain PACU 0 (0 to 4) 0.75 (1.19) 0 (0 to 6) 1.00 (1.57) .77 −0.25 (−0.76 to 0.27) Missing (n = 18) (n = 7) Evening of surgery 0 (0 to 6) 0.88 (1.57) 0 (0 to 7) 0.88 (1.45) .87 0.00 (−0.49 to 0.50) Missing (n = 2) (n = 2) Day 1 0 (0 to 5) 0.64 (1.13) 0 (0 to 7) 0.87 (1.72) .88 −0.24 (−0.72 to 0.26) Missing (n = 5) (n = 1) Fatigue PACU 4 (0 to 10) 3.81 (2.56) 4 (0 to 10) 3.66 (2.40) .76 0.15 (−0.74 to 1.04) Missing (n = 18) (n = 6) Evening of surgery 4 (0 to 9) 3.81 (2.15) 3 (0 to 9) 2.99 (2.56) .03 0.82 (0.03 to 1.61) Missing (n = 3) (n = 3) Day 1 3 (0 to 7) 2.79 (2.03) 2 (0 to 9) 2.45 (2.37) .24 0.34 (−0.40 to 1.08) Missing (n = 5) (n = 1) Comfort/Well-being PACU 8 (0 to 10) 7.43 (2.26) 8 (0 to 10) 7.15 (2.36) .57 0.28 (−0.55 to 1.11) Missing (n = 18) (n = 6) Evening of surgery 7 (0 to 10) 6.91 (2.26) 8 (1 to 10) 7.31 (2.16 .31 −0.40 (−1.12 to 0.33) Missing (n = 2) (n = 2) Day 1 8 (0 to 10) 7.18 (2.53) 8 (1 to 10) 7.71 (1.96) .39 −0.53 (−1.27 to 0.22) Missing (n = 5) (n = 1) Anxiety PACU 0 (0 to 9) 1.68 (2.23) 0 (0 to 7) 1.14 (1.78) .35 0.54 (−0.18 to 1.25) Missing (n = 18) (n = 6) Evening of surgery 0 (0 to 9) 1.41 (2.07) 0 (0 to 10) 0.91 (1.85) .11 0.50 (−0.15 to 1.14) Missing (n = 2) (n = 2) Day 1 0 (0 to 10) 1.17 (2.13) 0 (0 to 9) 1.08 (2.14) .56 0.09 (−0.62 to 0.81) Missing (n = 5) (n = 1) Abbreviation: PACU, postanesthesia care unit.

(1.78) .35 0.54 (−0.18 to 1.25) Missing (n = 18) (n = 6) Evening of surgery 0 (0 to 9) 1.41 (2.07) 0 (0 to 10) 0.91 (1.85) .11 0.50 (−0.15 to 1.14) Missing (n = 2) (n = 2) Day 1 0 (0 to 10) 1.17 (2.13) 0 (0 to 9) 1.08 (2.14) .56 0.09 (−0.62 to 0.81) Missing (n = 5) (n = 1) Abbreviation: PACU, postanesthesia care unit. a All variables were assessed using a visual analog scale scored from 0 to 10 (0 indicates not at all, and 10 indicates unbearable pain, extreme nausea/vomiting, extreme fatigue requiring bed rest, and maximal possible anxiety). For comfort/well-being, 0 indicates very uncomfortable, and 10 indicates maximal comfort desired. b Difference is the visual analog scale score in the control arm minus the visual analog scale score in the hypnosis arm.

a All variables were assessed using a visual analog scale scored from 0 to 10 (0 indicates not at all, and 10 indicates unbearable pain, extreme nausea/vomiting, extreme fatigue requiring bed rest, and maximal possible anxiety). For comfort/well-being, 0 indicates very uncomfortable, and 10 indicates maximal comfort desired. b Difference is the visual analog scale score in the control arm minus the visual analog scale score in the hypnosis arm. Secondary End Points The mean (SD) VAS scores for fatigue in the PACU were 3.81 (2.56) in the control arm vs 3.66 (2.40) in the hypnosis arm, which was not significantly different (difference, 0.15; 95% CI, −0.74 to 1.04) (Table 3). The mean (SD) VAS score for fatigue on the evening of surgery was significantly lower in the hypnosis arm (3.81 [2.15] in the control arm vs 2.99 [2.56] in the hypnosis arm) (difference, 0.82, 95% CI, 0.03-1.61; P = .03). Levels of comfort/well-being and anxiety immediately after surgery were similar in the 2 arms. At days 1, 7, and 30 after surgery, there was a tendency toward lower levels of fatigue, comfort/well-being, and anxiety in the hypnosis arm compared with the control arm, but the differences were not statistically significant. The mean (SD) perioperative care patient satisfaction evaluated on the day after surgery was 8.9 (1.5) in the control arm vs 9.5 (1.1) in the hypnosis arm (P = .02) (Table 2).

igue, comfort/well-being, and anxiety in the hypnosis arm compared with the control arm, but the differences were not statistically significant. The mean (SD) perioperative care patient satisfaction evaluated on the day after surgery was 8.9 (1.5) in the control arm vs 9.5 (1.1) in the hypnosis arm (P = .02) (Table 2). Exploratory analyses were conducted according to patient perception of whether she received hypnosis. Patient characteristics were well balanced for the 2 subgroups of perceived hypnosis (68 of 125 patients [54.4%]) and no perceived hypnosis (57 of 125 patients [45.6%]). Significantly lower fatigue scores were reported in the perceived hypnosis subgroup on the evening of surgery (mean [SD], 4.13 [2.26] for no perceived hypnosis vs 2.97 [2.42] for perceived hypnosis) (difference, 1.16; 95% CI, 0.31-2.00; P = .01) and on day 1 after surgery (mean [SD], 3.09 [2.15] for no perceived hypnosis vs 2.33 [2.37] for perceived hypnosis; difference, 0.76; 95% CI, −0.06 to 1.58; P = .048) (Table 4). The level of anxiety was also significantly lower on the evening of surgery in the perceived hypnosis subgroup (mean [SD], 1.67 [2.29] for no perceived hypnosis vs 0.75 [1.64] for perceived hypnosis (difference, 0.92; 95% CI, 0.22-1.62; P = .03). At days 1, 7, and 30 after surgery, levels of fatigue, comfort/well-being, and anxiety were not statistically different in the 2 treatment perception subgroups.

eived hypnosis subgroup (mean [SD], 1.67 [2.29] for no perceived hypnosis vs 0.75 [1.64] for perceived hypnosis (difference, 0.92; 95% CI, 0.22-1.62; P = .03). At days 1, 7, and 30 after surgery, levels of fatigue, comfort/well-being, and anxiety were not statistically different in the 2 treatment perception subgroups. Table 4. Patient Outcomes Assessed on the Day of Surgery (in the PACU and at Patient Discharge on the Evening of Surgery) and on Day 1 in the Control and Hypnosis Arms, Stratified by Treatment Group Perceptiona Variable Treatment Group Perception P Value Difference (95% CI)b No Perceived Hypnosis Subgroup (n = 57) Perceived Hypnosis Subgroup (n = 68) Median (Range) Mean (SD) Median (Range) Mean (SD) Breast Pain PACU 2 (0 to 5) 1.80 (1.64) 3 (0 to 7) 2.78 (1.67) .005 −0.98 (−1.61 to −0.34) Missing (n = 7) (n = 10) Evening of surgery 2 (0 to 6) 2.07 (1.73) 2 (0 to 6) 2.52 (1.54) .10 −0.45 (−1.03 to 0.13) Missing (n = 0) (n = 0) Day 1 2 (0 to 8) 1.75 (1.62) 1 (0 to 7) 1.67 (1.70) .65 0.08 (−0.52 to 0.68) Missing (n = 2) (n = 2) General Pain PACU 0 (0 to 4) 0.71 (1.20) 0 (0 to 6) 1.09 (1.66) .48 −0.38 (−0.94 to 0.18) Missing (n = 6) (n = 11) Evening of surgery 0 (0 to 6) 0.58 (1.07) 0 (0 to 7) 1.15 (1.70) .19 −0.57 (−1.08 to −0.05) Missing (n = 0) (n = 0) Day 1 0 (0 to 7) 0.78 (1.47) 0 (0 to 6) 0.74 (1.41) .88 0.04 (−0.48 to 0.56) Missing (n = 2) (n = 2) Fatigue PACU 4 (0 to 10) 4.20 (2.59) 3 (0 to 10) 3.26 (2.26) .07 0.94 (0.02 to 1.86) Missing (n = 6) (n = 10) Evening of surgery 4 (0 to 9) 4.13 (2.26) 3 (0 to 8) 2.97 (2.42) .01 1.16 (0.31 to 2.00) Missing (n = 1) (n = 1) Day 1 3 (0 to 8) 3.09 (2.15) 2 (0 to 9) 2.33 (2.37) .048 0.76 (−0.06 to 1.58) Missing (n = 2) (n = 2) Comfort/Well-being PACU 8 (0 to 10) 7.35 (2.08) 7 (0 to 10) 7.10 (2.48) .77 0.25 (−0.63 to 1.13) Missing (n = 6) (n = 10) Evening of surgery 7 (0 to 10) 6.63 (2.01) 8 (1 to 10) 7.13 (2.30) .14 −0.50 (−1.27 to 0.27) Missing (n = 0) (n = 0) Day 1 7 (0 to 10) 7.00 (2.20) 8 (1 to 10) 7.68 (2.18) .05 −0.68 (−1.47 to 0.11) Missing 2 2 Anxiety PACU 0 (0 to 9) 1.69 (2.34) 0 (0 to 5) 1.05 (1.58) .37 0.64 (−0.12 to 1.38) Missing (n = 6) (n = 10) Evening of surgery 0 (0 to 9) 1.67 (2.29) 0 (0 to 10) 0.75 (1.64) .03 0.92 (0.22 to 1.62) Missing (n = 0) (n = 0) Day 1 0 (0 to 10) 1.53 (2.42) 0 (0 to 8) 0.82 (1.79) .16 0.71 (−0.05 to 1.47) Missing (n = 2) (n = 2) Abbreviation: PACU, postanesthesia care unit.

5 (1.58) .37 0.64 (−0.12 to 1.38) Missing (n = 6) (n = 10) Evening of surgery 0 (0 to 9) 1.67 (2.29) 0 (0 to 10) 0.75 (1.64) .03 0.92 (0.22 to 1.62) Missing (n = 0) (n = 0) Day 1 0 (0 to 10) 1.53 (2.42) 0 (0 to 8) 0.82 (1.79) .16 0.71 (−0.05 to 1.47) Missing (n = 2) (n = 2) Abbreviation: PACU, postanesthesia care unit. a All variables were assessed using a visual analog scale scored from 0 to 10 (0 indicates not at all, and 10 indicates unbearable pain, extreme nausea/vomiting, extreme fatigue requiring bed rest, and maximal possible anxiety). For comfort/well-being, 0 indicates very uncomfortable, and 10 indicates maximal comfort desired. b Difference is the visual analog scale score in the no perceived hypnosis subgroup minus the visual analog scale score in the perceived hypnosis subgroup. Discussion We report the results of a multicenter, prospective, randomized, single-blind phase 3 clinical trial evaluating the benefit of preoperative hypnosis sessions in day-case minor breast cancer surgery. There was no benefit of a hypnosis session performed immediately before general anesthesia induction on postoperative breast pain in these patients. However, patients who thought that they had received hypnosis had significantly lower postoperative fatigue and anxiety.

sessions in day-case minor breast cancer surgery. There was no benefit of a hypnosis session performed immediately before general anesthesia induction on postoperative breast pain in these patients. However, patients who thought that they had received hypnosis had significantly lower postoperative fatigue and anxiety. Although breast pain was scored significantly higher by the patients in the hypnosis arm, this difference in pain level did not result at the clinical level in a higher consumption of ketamine and morphine in the PACU for these patients. Indeed, patients had similar need for rescue analgesia and pain relief in the PACU in the 2 arms. A 2014 meta-analysis by Kekecs et al14 showed that suggestive or hypnosis techniques reduced anxiety but did not significantly affect postoperative analgesic consumption. Furthermore, in the present study, an observed difference in breast pain was significant in the PACU in favor of the control arm but was not found for later time points.

by Kekecs et al14 showed that suggestive or hypnosis techniques reduced anxiety but did not significantly affect postoperative analgesic consumption. Furthermore, in the present study, an observed difference in breast pain was significant in the PACU in favor of the control arm but was not found for later time points. There may be 2 reasons for our results herein. First, patients in the hypnosis arm received lower doses of the intraoperative anesthetic drugs propofol and sufentanil, which may explain the lower pain levels in the control arm. This may be linked to the statistically significant reduced PACU length of stay for patients in the hypnosis arm (median, 60 minutes [range, 20-290 minutes] in the control arm vs 46 minutes [range, 5-100 minutes] in the hypnosis arm; P = .002). Also, fewer patients received lidocaine in the control arm than in the hypnosis arm. This may have affected postoperative pain perception, together with the fact that more patients received PONV prophylaxis in the control arm, which may have resulted in less pain compared with the hypnosis arm. Second, the choice of surgery may also explain our results. Patients underwent minor breast cancer surgery, an intervention that induced limited postoperative pain because of a minimally invasive surgical technique (tumorectomy). We used a multimodal analgesic strategy (ie, a combination of several analgesic drugs injected before the end of anesthesia), which is the standard of care for pain control in these patients. In both study arms, patients received intravenous acetaminophen, ketoprofen, and ketamine. This strategy enabled effective pain relief by our teams for all patients. Indeed, patients in the control arm reported a low median pain intensity in the PACU (2 [range, 0-6] for breast pain and 0 [range, 0-4] for general pain). Reducing pain even further with hypnosis may have been an unreasonable goal because of these low pain levels.

ed effective pain relief by our teams for all patients. Indeed, patients in the control arm reported a low median pain intensity in the PACU (2 [range, 0-6] for breast pain and 0 [range, 0-4] for general pain). Reducing pain even further with hypnosis may have been an unreasonable goal because of these low pain levels. The randomized clinical trial by Montgomery et al12 included 200 patients scheduled for breast surgery. Patients in the hypnosis arm underwent a 15-minute hypnosis session conducted by a psychologist. In that study, patients were not blinded to their study arm assignment, and the effectiveness of blinding of the research and clinical staff was not formally assessed. Blinding regarding treatment assignment usually gives stronger evidence of treatment efficacy than studies of unblinded design. When the treatment is hypnosis, blinding may decrease the potential for hypnotic suggestion and thus a positive effect of hypnosis. Montgomery et al12 reported a lower pain intensity in the PACU in the hypnosis arm compared with the control arm (mean VAS score [range, 0-100], 22.43 vs 47.83; P < .001). In that study, data were not available on pain reduction in the days after surgery. Regarding air management herein, we found a significant difference between the 2 study arms. A noninvasive laryngeal mask was used more frequently for patients in the hypnosis arm (53.5% [38 of 71] of control patients vs 70.1% [54 of 77] of hypnosis patients, P = .04). This suggests that hypnosis may improve the quality of anesthesia induction and may allow better drug titration.

ence between the 2 study arms. A noninvasive laryngeal mask was used more frequently for patients in the hypnosis arm (53.5% [38 of 71] of control patients vs 70.1% [54 of 77] of hypnosis patients, P = .04). This suggests that hypnosis may improve the quality of anesthesia induction and may allow better drug titration. In our study, we assessed fatigue, comfort/well-being, and anxiety immediately before PACU discharge, at patient discharge (on the evening of surgery), and on the day after surgery. Only fatigue was found to be significantly lower among patients in the hypnosis arm on the evening of surgery compared with patients in the control arm. We performed a subgroup exploratory analysis that took into account perception of hypnosis by the patients. Indeed, 25.0% (14 of 56) of patients in the control arm thought that they had received hypnosis, while only standard welcoming and patient care had been used by their staff members, who were not trained in hypnosis techniques. This emphasizes the difficulty in analyzing the specific and intrinsic effect of hypnosis vs a putative placebo effect. The subgroup analysis showed significantly lower postoperative fatigue and anxiety levels and better comfort/well-being in the perceived hypnosis subgroup compared with the no perceived hypnosis subgroup. To our knowledge, this is the first study to report such an effect of hypnosis on postoperative fatigue. In oncology, cancer-related fatigue is a frequently present and disabling symptom, which often influences patient quality of life20,21 and for which no treatment is available. Therefore, hypnosis may mitigate postoperative fatigue in patients already experiencing cancer-related fatigue. However, no such effect was found 7 days after surgery in our study. An intermediate hypnosis session 3 days after surgery may extend the duration of the benefit of hypnosis on postoperative fatigue. Further studies could assess the effect of hypnosis on fatigue as a primary end point in patients undergoing oncologic surgery or surgery for other diseases. Studies assessing anxiety will also be relevant, with a previous study22 showing that nonpharmacological techniques, such as auricular acupuncture, can reduce dental anxiety to the same extent as intranasal midazolam hydrochloride.

rimary end point in patients undergoing oncologic surgery or surgery for other diseases. Studies assessing anxiety will also be relevant, with a previous study22 showing that nonpharmacological techniques, such as auricular acupuncture, can reduce dental anxiety to the same extent as intranasal midazolam hydrochloride. Patient satisfaction regarding anesthesia care and global management (range, 0-10) was higher in the hypnosis arm (mean [SD], 8.9 [1.5] in the control arm vs 9.5 [1.1] in the hypnosis arm; P = .02]). A previous study7 showed that premedication with benzodiazepines did not improve the perioperative experience of patients when it was correlated with adverse effects, such as late extubation or a low early cognitive recovery rate, effects that have not been shown after hypnosis. Further studies are needed to objectivize and explain this higher global satisfaction and evaluate its causes, including simple attention and care linked to hypnosis or hypnosis-induced hormonal mediation influencing patient well-being. Although not analyzed in our study, it may be possible that hypnosis has an effect on the operating staff members. In the hypnosis arm, patient preparation was performed with extra courtesy and in a low-noise and relaxed environment, which if studied may reveal a benefit on the whole team. Regarding the quality of life at work, it may be relevant to study this aspect and the effect of hypnosis on surgeons, nurses, and anesthesiologists in further studies.23

nt preparation was performed with extra courtesy and in a low-noise and relaxed environment, which if studied may reveal a benefit on the whole team. Regarding the quality of life at work, it may be relevant to study this aspect and the effect of hypnosis on surgeons, nurses, and anesthesiologists in further studies.23 Limitations Our study has some limitations. As already discussed, the choice of surgery that induced limited postoperative pain is a major limitation and contributed to the negative primary end point of our study. Blinding may have decreased the effect of hypnosis because of hypnotic suggestion. Also, participation in the study may have caused a behavioral change among the medical and paramedical teams in the global care and management of patients, especially for patients in the control arm, who may have been cared for with more empathy than usual care. This may also have introduced a bias and decreased the difference in treatment group perception between the 2 arms and thus the effect of hypnosis in the hypnosis arm. Although patients in the control arm were prepared for surgery by staff not trained in hypnosis, a positive effect in addressing and caring for the patients makes it a limitation of the study.

decreased the difference in treatment group perception between the 2 arms and thus the effect of hypnosis in the hypnosis arm. Although patients in the control arm were prepared for surgery by staff not trained in hypnosis, a positive effect in addressing and caring for the patients makes it a limitation of the study. Conclusions Our study shows no benefit of a short perioperative hypnosis session on postoperative pain in women eligible for minor breast cancer surgery. However, hypnosis seems to have other benefits regarding fatigue and anxiety, especially in patients who thought that they received hypnosis. Patient satisfaction is also improved with hypnosis. Further studies are needed to objectivize the benefit of hypnosis in this population. Supplement. Trial Protocol Click here for additional data file.

Introduction Readmission reduction is linked to improved quality of care and cost savings across health care systems in many countries, including the United States.1 Reduction of 30-day hospital readmission among patients with acute ischemic stroke (AIS) is a quality metric established by the Centers for Medicare & Medicaid Services (CMS) in the United States. However, despite the emphasis on tracking and reporting 30-day readmission for CMS beneficiaries with AIS, little is known about the nationwide proportion of patients with ischemic and hemorrhagic stroke readmitted within 30 days. Prior published reports of stroke-related readmission are exclusive to the CMS patient population 65 years or older and do not represent more recent data.2,3,4 Establishing contemporary, nationwide readmission metrics for various stroke subtypes among patients of all ages insured by all payer types (including the uninsured) is necessary to investigate the effectiveness of readmission reduction measures and is of interest to multiple stakeholders. In addition, quantifying the trend of stroke-related readmission is important to comprehensively evaluate the association of future readmission reduction strategies over time. It has also been reported that the overall burden of readmission is disproportionately borne by high-volume, academic hospitals.5 These health care facilities serve as safety-net hospitals in their communities and provide access to a wide case, payer, and population mix of patients. The population-level disproportionate burden across various facility types for stroke-related readmission has not been quantified to date. Finally, there is a dearth of information about the population-wide association of stroke-related readmission in terms of hospital outcomes and resource use relative to an early readmission. Understanding this association with resources and reasons for readmission will be helpful for constructing future preventive frameworks.

, there is a dearth of information about the population-wide association of stroke-related readmission in terms of hospital outcomes and resource use relative to an early readmission. Understanding this association with resources and reasons for readmission will be helpful for constructing future preventive frameworks. Our objective was to examine population-based, nationwide 30-day stroke-related readmission metrics over a recent 6-year period (between January 1, 2010, and September 30, 2015) stratified by stroke subtype. We hypothesized that there would be a significant reduction of 30-day stroke-related readmission during this period. We further aimed to investigate the association between hospitals’ stroke discharge volume (SDV) and their teaching status and readmission rate. We hypothesized that large-volume teaching hospitals would bear a disproportionate burden of stroke-related readmission. Finally, we explored whether readmitted patients have higher in-hospital mortality, longer length of stay (LOS), and higher costs compared with the mean of these metrics at index admission among patients with various stroke subtypes.

teaching hospitals would bear a disproportionate burden of stroke-related readmission. Finally, we explored whether readmitted patients have higher in-hospital mortality, longer length of stay (LOS), and higher costs compared with the mean of these metrics at index admission among patients with various stroke subtypes. Methods Design and Database This was a population-based cohort study in which eligible patients with stroke were identified and followed up for 30-day hospital readmission during each individual year of the US Nationwide Readmissions Database between 2010 and 2015. The Nationwide Readmissions Database is part of the Health Care Utilization Project (HCUP) of the Agency for Healthcare Research and Quality (AHRQ).6 It comprises a nationally representative, weighted probability sample of approximately 36 million discharges per year from short-term hospitals of 22 geographically dispersed states in the United States (27 states were included for 2015). The nonweighted sample represents 50% of all US hospitalizations. Investigators received requisite training and conformed to the data use agreements with the HCUP. The HCUP databases conform to the definition of a limited data set; as per the data use agreement with the HCUP, review by an institutional review board is not required for use of limited data sets. The analysis and reporting followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.7

conform to the definition of a limited data set; as per the data use agreement with the HCUP, review by an institutional review board is not required for use of limited data sets. The analysis and reporting followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.7 Target Patient Population and Case Definitions We used International Classification of Diseases, Ninth Revision (ICD-9) codes to identify adults (≥18 years) with a primary diagnosis of ischemic stroke (codes 433.x1, 434.x1, and 436), primary intracerebral hemorrhage (ICH) (code 431), and subarachnoid hemorrhage (SAH) (code 430). Procedure codes were used to identify patients with stroke who received intravenous thrombolytics or endovascular treatment and underwent procedures, such as extraventricular drain placement, surgical decompression or hematoma evaluation, mechanical ventilation, and endotracheal or gastric tube placement. To avoid coding inconsistencies between ICD-9 and International Statistical Classification of Diseases and Related Health Problems, Tenth Revision, data from the last quarter of 2015 were excluded. We also excluded same-day events, discharges in the last month of analysis (by each year), and in-hospital deaths (for index admission). For analysis of patients with hemorrhagic stroke (ICH or SAH), we excluded all patients who had a concurrent diagnosis of head trauma. Eligible patients had their first (index) hospital admission tagged for further analyses. Readmission was defined as any admission within 30 days of index hospitalization discharge. Using CMS-defined algorithms, events were classified as planned or unplanned, and potentially preventable readmissions attributable to ambulatory care–sensitive conditions were identified (eAppendix in the Supplement).8 For subanalyses, we also identified patients with the same primary discharge diagnosis on readmission as that of their index admission and evaluated reasons for readmission using the AHRQ’s Clinical Classification Software–based diagnostic categories.9

ensitive conditions were identified (eAppendix in the Supplement).8 For subanalyses, we also identified patients with the same primary discharge diagnosis on readmission as that of their index admission and evaluated reasons for readmission using the AHRQ’s Clinical Classification Software–based diagnostic categories.9 Covariates, Subgroups, and Cost We categorized hospitals based on their annual SDV as low (11-50), medium (51-175), high (176-350), or very high (>350). Hospitals with an SDV of 10 or less were excluded. Hospitals were classified as teaching hospitals if they had an American Medical Association–approved residency program or had a ratio of full-time equivalent interns and residents to beds of 0.25 or higher. For some analyses, we dichotomized the period of investigation as preimplementation and postimplementation of excess readmission penalties under the CMS’ Hospital Readmissions Reduction Program (HRRP) (discharges before vs on or after October 1, 2012).10 Costs were obtained by using the AHRQ’s ratios of cost to charge and were inflation adjusted for 2014 using the Chained Consumer Price Index for all urban consumers and medical care services from the US Bureau of Labor Statistics.11

missions Reduction Program (HRRP) (discharges before vs on or after October 1, 2012).10 Costs were obtained by using the AHRQ’s ratios of cost to charge and were inflation adjusted for 2014 using the Chained Consumer Price Index for all urban consumers and medical care services from the US Bureau of Labor Statistics.11 Statistical Analysis We used survey design methods by taking into account sampling weights and clustering of discharges within hospitals to report national estimates of proportions and 95% CIs for overall, planned, and potentially preventable 30-day readmission during each individual year of analysis. Similar estimates are provided for readmission among patients discharged from teaching vs nonteaching hospitals and from hospitals with varying SDV. We fit survey design multivariable logistic regression models to assess the trend in annual proportion of 30-day readmission over the period of investigation and to investigate the association between CMS’ HRRP implementation and stroke-related readmission. Odds ratios (ORs) and 95% CIs are reported as estimates of likelihood of 30-day readmission across years and periods. Analyses controlled for potential changes in patients’ demographic, social, and comorbidity profiles. We also fit similar models to evaluate the association between stroke-related readmission and hospitals’ SDV and their teaching status and explored the interaction between SDV and teaching status for their association with readmission. Goodness-of-fit tests and area under the curve statistics were used to assess model fit and discrimination. With 252 406 events (30-day readmissions) in the combined data set, we had greater than 95% power to satisfy all of our analytical aims, including evaluation of interaction terms. Analyses were performed using statistical software (Stata, version 14; StataCorp LP).

tics were used to assess model fit and discrimination. With 252 406 events (30-day readmissions) in the combined data set, we had greater than 95% power to satisfy all of our analytical aims, including evaluation of interaction terms. Analyses were performed using statistical software (Stata, version 14; StataCorp LP). Results Analysis Population and Readmission Proportions The Nationwide Readmissions Database contains information on 208 363 328 hospital discharges between January 1, 2010, and September 30, 2015. Of these, 3 123 362 records were identified as having a primary discharge diagnosis of a stroke subtype (ICH, AIS, or SAH). Based on our criteria, 2 200 688 eligible stroke hospital discharges were included in our analyses. Of these, 87.6% were AIS, 8.7% were ICH, and 3.6% were SAH. Among the 2 078 854 index events (first event during each year) for all stroke subtypes, the mean (SE) patient age was 70.02 (0.07) years, and 51.9% were female. Summary data on demographics, hospital factors, comorbidity, disease severity, and treatment variables for the overall, readmitted, and nonreadmitted stroke discharges are listed in eTable 1 in the Supplement. Based on analysis of index events, 30-day readmission was highest for patients with ICH (13.70%; 95% CI, 13.40%-13.99%), followed by patients with AIS (12.44%; 95% CI, 12.33%-12.55%) and patients with SAH (11.48%; 95% CI, 11.01%-11.96%). More than 90% of all 30-day stroke-related readmissions were unplanned; depending on stroke subtype, up to 13.6% were deemed potentially preventable. Details on selection of eligible stroke discharges, index events, and stroke subtype–specific readmission metrics are shown in Figure 1.

(11.48%; 95% CI, 11.01%-11.96%). More than 90% of all 30-day stroke-related readmissions were unplanned; depending on stroke subtype, up to 13.6% were deemed potentially preventable. Details on selection of eligible stroke discharges, index events, and stroke subtype–specific readmission metrics are shown in Figure 1. Figure 1. Eligible Stroke Discharges, Reasons for Exclusion, and Proportion of Overall, Unplanned, and Potentially Preventable Stroke-Related Nationwide 30-Day Readmissions Between January 1, 2010, and September 30, 2015 The proportion (95% CI) of 30-day overall readmissions among index events for acute ischemic stroke was 12.44% (12.33%-12.55%); intracerebral hemorrhage, 13.70% (13.40%-13.99%); and subarachnoid hemorrhage, 11.48% (11.01%-11.96%). The proportion (95% CI) of unplanned readmissions among 30-day readmissions for acute ischemic stroke was 90.00% (89.74%-90.26%); intracerebral hemorrhage, 94.15% (93.56%-94.70%); and subarachnoid hemorrhage, 90.63% (89.40%-91.73%). The proportion (95% CI) of potentially preventable readmissions among 30-day readmissions for acute ischemic stroke was 13.28% (13.03%-13.53%); intracerebral hemorrhage, 13.57% (12.82%-14.35%); and subarachnoid hemorrhage, 9.27% (8.26%-10.39%). aInclusive of data between January 1, 2010, and September 30, 2015. bReasons not mutually exclusive. cUnplanned and potentially preventable readmissions not mutually exclusive.

Figure 1. Eligible Stroke Discharges, Reasons for Exclusion, and Proportion of Overall, Unplanned, and Potentially Preventable Stroke-Related Nationwide 30-Day Readmissions Between January 1, 2010, and September 30, 2015 The proportion (95% CI) of 30-day overall readmissions among index events for acute ischemic stroke was 12.44% (12.33%-12.55%); intracerebral hemorrhage, 13.70% (13.40%-13.99%); and subarachnoid hemorrhage, 11.48% (11.01%-11.96%). The proportion (95% CI) of unplanned readmissions among 30-day readmissions for acute ischemic stroke was 90.00% (89.74%-90.26%); intracerebral hemorrhage, 94.15% (93.56%-94.70%); and subarachnoid hemorrhage, 90.63% (89.40%-91.73%). The proportion (95% CI) of potentially preventable readmissions among 30-day readmissions for acute ischemic stroke was 13.28% (13.03%-13.53%); intracerebral hemorrhage, 13.57% (12.82%-14.35%); and subarachnoid hemorrhage, 9.27% (8.26%-10.39%). aInclusive of data between January 1, 2010, and September 30, 2015. bReasons not mutually exclusive. cUnplanned and potentially preventable readmissions not mutually exclusive. Trends in Nationwide Stroke-Related 30-Day Readmissions The overall proportion of 30-day stroke-related readmission was highest in 2010 (13.47%; 95% CI, 13.19%-13.76%). On average, there was a 3.3% annual decline in 30-day readmission between 2010 and 2014. There was a statistically significant annual decline in likelihood of 30-day readmission by 4.0% over the period of investigation (OR, 0.96; 95% CI, 0.95-0.97) after controlling for potential changes in demographic, social, and comorbidity case mix across years. Our adjusted multivariable analyses further indicated that the probability of a nationwide 30-day stroke-related readmission was estimated to be 17.2% and 16.5% less likely during 2014 and 2015, respectively, compared with 2010 and 12.1% less likely after implementation of the CMS’ HRRP penalties compared with the preimplementation period. A similar decline in stroke-related readmission was observed across all stroke subtypes, with ICH readmissions remaining the highest across the analysis period. Figure 2 shows stroke subtype–specific proportion of 30-day readmission across the years of investigation, and the Table lists proportions, ORs, and 95% CIs for annual likelihood of 30-day readmission for all stroke subtypes, specific stroke subtypes, and before and after CMS’ HRRP implementation.

the analysis period. Figure 2 shows stroke subtype–specific proportion of 30-day readmission across the years of investigation, and the Table lists proportions, ORs, and 95% CIs for annual likelihood of 30-day readmission for all stroke subtypes, specific stroke subtypes, and before and after CMS’ HRRP implementation. Figure 2. Proportion of Patients With Intracerebral Hemorrhage, Acute Ischemic Stroke, and Subarachnoid Hemorrhage Readmitted Within 30 Days of Index Hospital Discharge by Year of Investigation Adjusted odds ratios of 30-day readmissions by year were 0.96 (95% CI, 0.95-0.97) for intracerebral hemorrhage, 0.97 (95% CI, 0.95-0.98) for acute ischemic stroke, and 0.96 (95% CI, 0.94-0.99) for subarachnoid hemorrhage. Odd ratios were adjusted for the following: age, sex, insurance, patient location, median household income for patient’s zip code, Charlson Comorbidity Index, number of chronic conditions, atrial fibrillation, alcohol abuse, deficiency anemias, chronic blood loss anemia, congestive heart failure, coagulopathy, uncomplicated diabetes, diabetes with chronic complications, hypertension (combined uncomplicated and complicated), liver disease, fluid and electrolyte disorders, other neurological disorders, obesity, pulmonary circulation disorders, renal failure, solid tumor without metastasis, psychoses, depression, chronic pulmonary disease, drug abuse, peripheral vascular disorders, peptic ulcer disease (excluding bleeding), and valvular disease. Results shown for 2015 include data between January 1 and September 30, 2015.

ity, pulmonary circulation disorders, renal failure, solid tumor without metastasis, psychoses, depression, chronic pulmonary disease, drug abuse, peripheral vascular disorders, peptic ulcer disease (excluding bleeding), and valvular disease. Results shown for 2015 include data between January 1 and September 30, 2015. Table. Proportions, Odds Ratios, and 95% CIs for Year-Wise 30-Day Readmission Among Patients With All Stroke Subtypes, Intracerebral Hemorrhage, Acute Ischemic Stroke, and Subarachnoid Hemorrhagea Variable All Stroke Subtypes Intracerebral Hemorrhage Acute Ischemic Stroke Subarachnoid Hemorrhage Readmission, % (95% CI) OR (95% CI) Readmission, % (95% CI) OR (95% CI) Readmission, % (95% CI) OR (95% CI) Readmission, % (95% CI) OR (95% CI) Year of Discharge 2010 13.47 (13.19-13.76) 1 [Reference] 13.44 (13.15-13.74) 1 [Reference] 14.46 (13.76-13.15) 1 [Reference] 11.79 (13.74-13.71) 1 [Reference] 2011 13.05 (12.72-13.39) 0.96 (0.92-0.99) 13.01 (12.68-13.36) 0.96 (0.88-1.05) 14.16 (13.39-12.68) 0.96 (0.92-1.00) 12.16 (13.36-13.41) 1.02 (0.86-1.22) 2012 12.52 (12.20-12.84) 0.91 (0.87-0.94) 12.45 (12.15-12.75) 0.95 (0.87-1.04) 13.97 (12.84-12.15) 0.91 (0.87-0.94) 11.49 (12.75-13.20) 0.96 (0.81-1.13) 2013 12.15 (11.93-12.37) 0.87 (0.84-0.90) 12.10 (11.87-12.32) 0.88 (0.81-0.96) 13.20 (12.37-11.87) 0.87 (0.84-0.90) 11.22 (12.32-12.55) 0.91 (0.77-1.06) 2014 11.79 (11.59-12.00) 0.83 (0.80-0.86) 11.71 (11.50-11.92) 0.87 (0.81-0.95) 13.25 (12.00-11.50) 0.83 (0.80-0.85) 10.96 (11.92-12.64) 0.87 (0.74-1.01) 2015 11.98 (11.79-12.18) 0.83 (0.81-0.86) 11.94 (11.73-12.14) 0.85 (0.78-0.93) 13.05 (12.18-11.73) 0.84 (0.81-0.86) 11.20 (12.14-12.38) 0.86 (0.73-1.00) Hospital Readmissions Reduction Program Status Before the CMS’ HRRPb 13.08 (12.90-13.27) 1 [Reference] 14.25 (13.80-14.72) 1 [Reference] 13.04 (12.85-13.22) 1 [Reference] 11.91 (11.15-12.71) 1 [Reference] After the CMS’ HRRPc 11.96 (11.85-12.08) 0.88 (0.86-0.90) 13.19 (12.83-13.57) 0.89 (0.85-0.94) 11.90 (11.78-12.02) 0.88 (0.86-0.90) 11.08 (10.57-11.60) 0.87 (0.80-0.96) Abbreviations: CMS, Centers for Medicare & Medicaid Services; HRRP, Hospital Readmissions Reduction Program; OR, odds ratio.

11.15-12.71) 1 [Reference] After the CMS’ HRRPc 11.96 (11.85-12.08) 0.88 (0.86-0.90) 13.19 (12.83-13.57) 0.89 (0.85-0.94) 11.90 (11.78-12.02) 0.88 (0.86-0.90) 11.08 (10.57-11.60) 0.87 (0.80-0.96) Abbreviations: CMS, Centers for Medicare & Medicaid Services; HRRP, Hospital Readmissions Reduction Program; OR, odds ratio. a Odds ratios were adjusted for the following: age, sex, insurance, patient location, median household income for patient’s zip code, Charlson Comorbidity Index, number of chronic conditions, atrial fibrillation, alcohol abuse, deficiency anemias, chronic blood loss anemia, congestive heart failure, coagulopathy, uncomplicated diabetes, diabetes with chronic complications, hypertension (combined uncomplicated and complicated), liver disease, fluid and electrolyte disorders, other neurological disorders, obesity, pulmonary circulation disorders, renal failure, solid tumor without metastasis, psychoses, depression, chronic pulmonary disease, drug abuse, peripheral vascular disorders, peptic ulcer disease (excluding bleeding), and valvular disease. b Defined as discharges before October 1, 2012, based on effective date of the CMS’ HRRP implementation. c Defined as discharges after October 1, 2012, based on effective date of the CMS’ HRRP implementation.

a Odds ratios were adjusted for the following: age, sex, insurance, patient location, median household income for patient’s zip code, Charlson Comorbidity Index, number of chronic conditions, atrial fibrillation, alcohol abuse, deficiency anemias, chronic blood loss anemia, congestive heart failure, coagulopathy, uncomplicated diabetes, diabetes with chronic complications, hypertension (combined uncomplicated and complicated), liver disease, fluid and electrolyte disorders, other neurological disorders, obesity, pulmonary circulation disorders, renal failure, solid tumor without metastasis, psychoses, depression, chronic pulmonary disease, drug abuse, peripheral vascular disorders, peptic ulcer disease (excluding bleeding), and valvular disease. b Defined as discharges before October 1, 2012, based on effective date of the CMS’ HRRP implementation. c Defined as discharges after October 1, 2012, based on effective date of the CMS’ HRRP implementation. Burden of Stroke-Related 30-Day Readmission Across Hospitals A mean of 1397 hospitals per year were included in our analyses. The median annual SDV among these hospitals was 99 (interquartile range, 37-230). Our adjusted multivariable model indicated that higher SDV and nonteaching status were independently associated with higher stroke-related readmission for hospitals with very high vs low SDV (OR, 1.08; 95% CI, 1.03-1.12) and for nonteaching vs teaching hospitals (OR, 1.05; 95% CI, 1.03-1.07). We found a significant effect modification of teaching status and readmission by hospitals’ SDV. A higher SDV was significantly associated with 30-day stroke-related readmission among nonteaching hospitals (OR, 1.06; 95% CI, 1.02-1.11 for high vs low SVD and OR, 1.13; 95% CI, 1.07-1.18 for very high vs low SDV); however, a similar association was not observed for the 2 comparisons across teaching hospitals (2-sided P for interaction = .02 and .01, respectively) (eTable 2 and eFigure in the Supplement). We further characterized hospitals’ SDV based on increments of 50; after adjusting for demographic and comorbidity variables, there was a significantly increasing likelihood of 30-day stroke-related readmission for nonteaching hospitals with an annual SDV of 300 or higher compared with teaching hospitals with similar SDV (Figure 3).

racterized hospitals’ SDV based on increments of 50; after adjusting for demographic and comorbidity variables, there was a significantly increasing likelihood of 30-day stroke-related readmission for nonteaching hospitals with an annual SDV of 300 or higher compared with teaching hospitals with similar SDV (Figure 3). Figure 3. Probability and 95% CI of 30-Day Stroke-Related Readmission Based on Hospitals’ Stroke Discharge Volume in Increments of 50 for Teaching and Nonteaching Hospitals in the 2010 to 2015 Nationwide Readmissions Database Values on the x-axis were obtained from multivariable survey design models adjusted for patient case mix of demographic and comorbidity variables. The y-axis represents hospitals’ annual stroke discharge volume. Hospitals with stroke discharge volume of 10 or less were excluded. The maximum stroke discharge volume for nonteaching hospitals is 698. Data for nonteaching hospitals beyond the maximum are modeled for comparison with teaching hospitals.

dity variables. The y-axis represents hospitals’ annual stroke discharge volume. Hospitals with stroke discharge volume of 10 or less were excluded. The maximum stroke discharge volume for nonteaching hospitals is 698. Data for nonteaching hospitals beyond the maximum are modeled for comparison with teaching hospitals. Reasons, Resources, and Outcomes for Stroke-Related Readmissions Among all 30-day stroke-related readmissions, 18.91% (95% CI, 18.64%-19.18%) had the same primary discharge diagnosis on readmission as that of their index hospitalization. This proportion was highest among patients with AIS (17.91%; 95% CI, 17.62%-18.19%), whereas it was 9.16% (95% CI, 8.52%-9.84%) for patients with ICH and 7.53% (95% CI, 6.68%-8.48%) for patients with SAH. Furthermore, this proportion increased over time, ranging from 17.66% (95% CI, 17.02%-18.32%) in 2010 to 19.94% (95% CI, 19.27%-20.62%) in 2015. The proportion of patients among the readmitted patients with the same primary diagnosis on readmission as that of their index admission for all stroke subtypes and for each stroke subtype is shown in Figure 4; eTable 3 in the Supplement lists proportions and 95% CIs for unplanned and potentially preventable readmissions by stroke subtype for each year among patients with the same readmission diagnosis as that of their index hospitalization. The analysis of readmission reasons yielded acute cerebrovascular disease and septicemia as the first and second leading reasons for readmission for all years and all stroke subtypes, accounting for 19.6% and 10.0%, respectively, of all readmissions. The top 25 reasons for readmission by year for each stroke subtype are listed in eTables 4, 5, 6, and 7 in the Supplement. On 30-day readmission, a higher proportion of patients with AIS experienced in-hospital mortality (6.54%; 95% CI, 6.36%-6.73%) compared with the proportion of patients who died during initial hospitalization (5.13%; 95% CI, 5.05%-5.22%). This was not observed for patients with ICH or SAH. The mean LOS on readmission was also longer for patients with AIS (6.50 days) compared with the mean index LOS (4.92 days). Likewise, the inflation-adjusted mean cost per stay was higher on readmission for patients with AIS ($10 881 vs $12 303 for index vs readmission). Comparative data between index hospitalization and readmission for mortality, LOS, and cost are listed in eTable 8 in the Supplement.

ed with the mean index LOS (4.92 days). Likewise, the inflation-adjusted mean cost per stay was higher on readmission for patients with AIS ($10 881 vs $12 303 for index vs readmission). Comparative data between index hospitalization and readmission for mortality, LOS, and cost are listed in eTable 8 in the Supplement. Figure 4. Year-Wise Proportion of 30-Day Hospital Readmission Among Patients With All Stroke Subtypes, Intracerebral Hemorrhage, Acute Ischemic Stroke, and Subarachnoid Hemorrhage With the Same Primary Diagnosis on Readmission as That on Index Hospitalization Results shown for 2015 include data between January 1 and September 30, 2015. Discussion We present nationally representative readmission metrics during a contemporary 6-year period for patients with ICH, AIS, and SAH of all ages and all payer types. These analyses serve to address knowledge gaps in trends of readmission metrics and highlight that patients with stroke discharged from higher-volume nonteaching hospitals may be targeted for readmission reduction. These metrics are helpful for defining policy and practice and provide an estimate to measure the effectiveness of various readmission reduction strategies at the local level. To our knowledge, such data have not been previously reported for patients with stroke.

ing hospitals may be targeted for readmission reduction. These metrics are helpful for defining policy and practice and provide an estimate to measure the effectiveness of various readmission reduction strategies at the local level. To our knowledge, such data have not been previously reported for patients with stroke. We report that a considerable proportion of patients with stroke are readmitted within 30 days of hospital discharge, with a significantly higher proportion among patients with ICH compared with patients with AIS (13.7% vs 12.4%). More than 90% of readmissions are unplanned, and up to 13% may be potentially preventable. Publicly reported readmission measures based on analyses of the CMS data are limited to patients with AIS who are 65 years or older. Our analyses of more than 2 million patients indicate that up to 35% of patients with stroke are younger than 65 years and that more than 25% were not insured by the CMS. Even among patients with AIS, the 30-day readmission was significantly lower for patients younger than 65 years compared with those 65 years or older (10.8% vs 13.3%). Furthermore, there were significant differences in readmission rates between payer types (OR, 0.70; 95% CI, 0.68-0.72 for private insurance vs Medicare) even after controlling for age. These data suggest that CMS estimates, although meaningful for CMS-insured elderly patients with AIS, may not be nationally representative and indicate that additional analyses and reporting of readmission metrics among patients of all stroke subtypes, all age groups, and all payer types are warranted to fill the knowledge gaps. Sole reliance on CMS measures to penalize hospitals for excess readmissions has also been previously questioned.12 Our analyses indicate that stroke-related readmission gradually declined between 2010 and 2014, with a mean annual reduction of 3.3% and a reduction of 12.5% between 2010 and 2014. A similar decline was not observed between 2014 and 2015, possibly because of the exclusion of data from the third quarter of 2015. However, CMS13 data for patients with AIS also suggest a steady readmission rate between 2013-2014 and 2014-2015. While there are no prior reports of nationwide trends among all stroke-related readmissions, our findings are comparable to gains in readmission reduction previously reported for other HRRP-targeted14 and HRRP-nontargeted conditions between 2007-2008 and 2014-2015.

eady readmission rate between 2013-2014 and 2014-2015. While there are no prior reports of nationwide trends among all stroke-related readmissions, our findings are comparable to gains in readmission reduction previously reported for other HRRP-targeted14 and HRRP-nontargeted conditions between 2007-2008 and 2014-2015. Furthermore, our data show a readmission reduction of 4.9% between 2012 and 2014 for patients with AIS. This estimate is comparable to a readmission reduction of 4.7% for patients with AIS in CMS data during a similar period.13 Also, the cumulative decline in readmission for 4 other high-volume conditions between 2009 and 2013 has been reported to be 5.4%.15 Therefore, we believe that an annual 3.3% decline in stroke-related readmission can be used to guide and monitor the effectiveness of stroke-related readmission reduction programs.

iod.13 Also, the cumulative decline in readmission for 4 other high-volume conditions between 2009 and 2013 has been reported to be 5.4%.15 Therefore, we believe that an annual 3.3% decline in stroke-related readmission can be used to guide and monitor the effectiveness of stroke-related readmission reduction programs. In addition to the absence of stroke-related readmission reduction programs at the national level, there could be multiple other reasons for the nationwide decline. First, patients with stroke are elderly with multiple comorbidities, particularly cardiovascular comorbidities. Because our analyses include all-cause readmissions, it is likely that readmission reduction among patients with stroke is corollary to the overall readmission reduction reported for patients without stroke. It is important to highlight that 30-day readmission attributed to the same primary diagnosis as that of the index admission did not show a decreasing trend. Second, it is also possible that an emphasis on improving postacute transition of care (TOC) has accounted for some of this decline.16 Multiple regional and local initiatives are under way to improve hospital discharge processes.17,18,19,20,21 However, there remains considerable lack of standardization in TOC models across the nation for patients with stroke, and the overall influence of TOC improvement on stroke-related readmission remains to be systematically demonstrated.22 Third, it is also likely that stroke-related readmission metrics have been influenced by well-documented effects of measuring and reporting health care quality.23

the nation for patients with stroke, and the overall influence of TOC improvement on stroke-related readmission remains to be systematically demonstrated.22 Third, it is also likely that stroke-related readmission metrics have been influenced by well-documented effects of measuring and reporting health care quality.23 Our adjusted analyses indicate that likelihood of readmission significantly increases with higher SDV for nonteaching hospitals, but a similar outcome is not observed for teaching institutions. Although seemingly small, these associations translate into a population-based increased risk of readmission among patients with stroke at nonteaching hospitals. There is a growing concern that various quality programs, including the HRRP, will disproportionately penalize safety-net hospitals.24 It has also been reported that large hospitals and teaching hospitals had a significantly higher likelihood of penalization under the HRRP.25 However, a prior analysis of CMS data from 2006 found no differences in readmission rates between patients with stroke treated at primary stroke centers and those discharged from noncertified centers.26 This is important because many high-volume teaching hospitals serve as safety-net hospitals for their communities, catering to the health care needs of the uninsured and underinsured. The effect modification by hospitals’ teaching status is likely explained by better adherence to quality-of-care metrics at these centers, including an emphasis on TOC variables. For example, many teaching hospitals have adjoining outpatient clinics or are staffed by faculty who also attend in an ambulatory clinic. Certain reports indicate better adherence to in-hospital quality of care and improved outcomes for patients discharged from hospitals contributing data to national stroke registries.27,28 Such improvements specifically include greater proportions of patients with AIS receiving thrombolytic therapies and shorter door-to-needle times. Significantly lower rates of 30-day readmission were previously demonstrated among patients with AIS who received thrombolytic therapies.29 Improved acute treatment metrics have also been strongly attributed to the use of telestroke technology and organization of care delivery.30,31 It is likely that a larger number of teaching hospitals are reporting their data to national registries and are regional telestroke centers and thus consistently provide better quality of care to patients with stroke even at higher volumes.

ted to the use of telestroke technology and organization of care delivery.30,31 It is likely that a larger number of teaching hospitals are reporting their data to national registries and are regional telestroke centers and thus consistently provide better quality of care to patients with stroke even at higher volumes. Identifying specific patient populations with stroke at higher risk of readmission are much needed for practice and policy decisions.

ted to the use of telestroke technology and organization of care delivery.30,31 It is likely that a larger number of teaching hospitals are reporting their data to national registries and are regional telestroke centers and thus consistently provide better quality of care to patients with stroke even at higher volumes. Identifying specific patient populations with stroke at higher risk of readmission are much needed for practice and policy decisions. We examined reasons for readmission using 2 different approaches and consistently found that approximately 1 in 5 readmissions were attributable to acute cerebrovascular disease or had the same principal diagnosis as that of the index hospitalization. The second most prevalent reason for readmission was septicemia. These findings were consistent across all years of analysis for all stroke subtypes. Our findings verify prior individual reports and systematic reviews regarding reasons for readmission among patients with stroke.2,3,29,32 These findings highlight the importance of secondary stroke prevention and warrant a deeper examination of various mechanistic factors that mediate early readmission in patients with stroke. We also report that patients with AIS who experience 30-day readmission have greater mortality, longer LOS, and higher cost per stay on readmission compared with the mean of these metrics on index admission. It has been previously shown that readmission is costlier compared with the index admission for a several other high-volume conditions as well.15 We believe that excess cost on readmission is mostly explained by longer LOS and is likely attributable to the need for higher intensity of care. Further analyses are warranted to elucidate factors that lead to excess resource use and mortality among patients with AIS on 30-day readmission.

volume conditions as well.15 We believe that excess cost on readmission is mostly explained by longer LOS and is likely attributable to the need for higher intensity of care. Further analyses are warranted to elucidate factors that lead to excess resource use and mortality among patients with AIS on 30-day readmission. Limitations Our analyses should be interpreted in light of the following limitations. First, the use of an administrative database did not allow us to control for stroke-specific severity measures. However, a large sample size rendered enough power to adjust for an extensive set of comorbidity and other sociodemographic variables and adequately test for effect modification. Second, the design of the Nationwide Readmissions Database did not enable us to track individual patients across multiple years. Therefore, our analyses represent cross-sectional estimates for each included year. Likewise, the readmission reduction reported across the period of CMS’ HRRP implementation does not convey a direct influence of the HRRP on stroke-related readmission. Third, potentially preventable readmissions were defined based on CMS methods for analyses of administrative data. Reasons for readmission are multifactorial (patient, quality of care, social, and access to care); therefore, classification of a readmission as potentially preventable requires deeper, prospective, contextual analyses. Until such time as these factors are clearly elucidated, using readmission reduction as a primary outcome in clinical trials will remain challenging. Fourth, the use of ICD-9 codes to identify cases may have been sensitive to variations in coding practices across various settings or over time. Such variations may introduce bias in risk-adjustment models or case ascertainment. However, sensitivity, specificity, and positive predictive value of ICD-9 codes for stroke have been reported to be high.33,34,35,36,37 Furthermore, we did not include data from the last quarter of 2015 to avoid the possible influence of changes in coding practices. While our work may address certain limitations of existing CMS readmission measures, it does not completely eliminate concerns raised in the literature regarding the broad use of hospital readmission as a quality measure for hospital rankings.12 Although nationally derived, our estimates are best suited for understanding and improving care at a local level, rather than establishing a national performance standard.

completely eliminate concerns raised in the literature regarding the broad use of hospital readmission as a quality measure for hospital rankings.12 Although nationally derived, our estimates are best suited for understanding and improving care at a local level, rather than establishing a national performance standard. Further work is warranted to establish the validity of national readmission measures, particularly in the context of emerging evidence of compromised survival associated with readmission reduction for other conditions.38 Conclusions We provide novel and contemporary evidence of declining 30-day readmission among patients with all stroke subtypes. This decline seems to be explained largely by the overall decline in 30-day readmission for other high-volume conditions. We further identify nonteaching hospitals with higher SDV as potential targets for improvement in stroke-related readmission, perhaps focused on secondary stroke prevention. We provide an estimate of a temporal trend as a metric for planning and evaluation of readmission reduction strategies at the level of individual hospitals. However, patient-level and health care environment–specific prospective research is needed to fully understand reasons for readmission and development of targeted readmission reduction strategies among patients with stroke. Supplement. eAppendix. Supplemental Methods eTable 1. Descriptive Univariable Analysis of the Overall, Readmitted, and Non-Readmitted Index Stroke Discharges in the National Readmission Database (January 2010-September 2015)

Conclusions We provide novel and contemporary evidence of declining 30-day readmission among patients with all stroke subtypes. This decline seems to be explained largely by the overall decline in 30-day readmission for other high-volume conditions. We further identify nonteaching hospitals with higher SDV as potential targets for improvement in stroke-related readmission, perhaps focused on secondary stroke prevention. We provide an estimate of a temporal trend as a metric for planning and evaluation of readmission reduction strategies at the level of individual hospitals. However, patient-level and health care environment–specific prospective research is needed to fully understand reasons for readmission and development of targeted readmission reduction strategies among patients with stroke. Supplement. eAppendix. Supplemental Methods eTable 1. Descriptive Univariable Analysis of the Overall, Readmitted, and Non-Readmitted Index Stroke Discharges in the National Readmission Database (January 2010-September 2015) eTable 2. Multivariate Analyses for Association Between 30-Day Readmission and Hospital Teaching Status and Stroke Discharge Volume eTable 3. Proportion and 95% CI for Same Primary Diagnosis, Unplanned, and Preventable Readmissions Among Readmitted Patients by Year for Stroke Subtypes eTable 4. Top 25 Causes of 30-Day Readmission for All Stroke Subtypes by Year eTable 5. Top 25 Causes of 30-Day Readmission for Ischemic Stroke by Year eTable 6. Top 25 Causes of 30-Day Readmission for Intracerebral Hemorrhage by Year

eTable 3. Proportion and 95% CI for Same Primary Diagnosis, Unplanned, and Preventable Readmissions Among Readmitted Patients by Year for Stroke Subtypes eTable 4. Top 25 Causes of 30-Day Readmission for All Stroke Subtypes by Year eTable 5. Top 25 Causes of 30-Day Readmission for Ischemic Stroke by Year eTable 6. Top 25 Causes of 30-Day Readmission for Intracerebral Hemorrhage by Year eTable 7. Top 25 Causes of 30-Day Readmission for Subarachnoid Hemorrhage by Year eTable 8. Comparison of Length of Stay, In-hospital Mortality, and Cost of Care for Ischemic and Hemorrhagic Stroke by Year eFigure. Probability and 95% CI (Y-Axis) for 30-Day Stroke Related Readmissions for Hospitals With Varying Stroke Discharge Volume (X-Axis) Based on Hospitals’ Teaching Status eReferences Click here for additional data file.

Introduction Historically, stereotactic ablative radiotherapy (SABR)—giving small numbers of high ablative doses of radiotherapy over a short period—has produced long-term rates of control of local and regional disease exceeding 80% when it is used as first-line treatment for appropriately chosen patients with early-stage non–small cell lung cancer (NCSLC).1 Stereotactic ablative radiotherapy has recently been shown to produce survival and cancer-specific outcomes comparable with those of patients who have undergone lobectomy, but with less morbidity, and today represents first-line therapy for patients whose disease is inoperable.2,3,4 Although the use of SABR for patients with operable disease remains under investigation, the elderly population is composing a greater proportion of all patients treated; as such, the number of patients with early-stage NSCLC that is inoperable, and thus deferred to definitive SABR treatment, continues to rise.5,6 Recurrence patterns after SABR have been reported, but, to date, outcomes have not.4 Until now, little was known about the 1 in 6 patients who develop isolated local recurrence (iLR) or isolated regional recurrence (iRR) after first-line SABR.1,2,3,4 Thus, for thoracic oncologists, clinical questions about the outcomes for such patients (whose disease is potentially curable) and how best to manage recurrences have remained largely unanswered.

n 6 patients who develop isolated local recurrence (iLR) or isolated regional recurrence (iRR) after first-line SABR.1,2,3,4 Thus, for thoracic oncologists, clinical questions about the outcomes for such patients (whose disease is potentially curable) and how best to manage recurrences have remained largely unanswered. Although options for treating recurrence (such as surgery and reirradiation) are offered in guidelines from the National Comprehensive Cancer Network7 and the European Society for Medical Oncology,8 they tend to not apply easily to the population of patients undergoing SABR, most of whom were not candidates for surgery and had already received radiotherapy. Thus, no evidence-based guidelines or large-scale studies specifying how to determine when a given salvage technique would be appropriate for these patients have been available. Moreover, since much of the evidence to support salvage treatment after SABR has been limited to studies of small, heterogeneous groups of patients,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23 little is known of outcomes after iLR or iRR after SABR for early-stage NSCLC. We sought to fill that void by reporting long-term outcomes for a large group of patients with iLR or iRR after SABR for early-stage NSCLC. Our findings on survival, disease control, and toxic effects after various salvage techniques serve to inform treatment decision making for these patients with potentially curable disease.

t to fill that void by reporting long-term outcomes for a large group of patients with iLR or iRR after SABR for early-stage NSCLC. Our findings on survival, disease control, and toxic effects after various salvage techniques serve to inform treatment decision making for these patients with potentially curable disease. Methods Patients We analyzed 912 patients with clinical stage T1 to T3 (satellite nodule) N0M0 NSCLC not involving the bronchial tree or other critical structures, who had received image-guided SABR on an institutional protocol at MD Anderson Cancer Center, Houston, Texas, from January 1, 2004, through December 31, 2014. All patients had been registered prospectively, and their records were reviewed retrospectively for this analysis, which took place from June 1 to August 30, 2017. Before SABR, disease was staged by chest computed tomography (CT) and positron emission tomography (PET) with CT, with brain CT or magnetic resonance imaging as needed. Images suggesting mediastinal disease were followed up with endobronchial ultrasound–guided biopsy to rule out nodal metastases. This study was approved by the MD Anderson Cancer Center institutional review board, and the requirement for informed consent was waived owing to deidentification of patient data. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines.24

al metastases. This study was approved by the MD Anderson Cancer Center institutional review board, and the requirement for informed consent was waived owing to deidentification of patient data. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines.24 SABR Protocol Four-dimensional CT images were obtained in all cases to account for tumor motion, and respiratory gating was used for patients whose tumor moved more than 1 cm. Most patients were treated with 50 Gy in 4 fractions (to convert gray to rad, multiply by 100), except for patients with large or central lesions (ie, within 2 cm of critical mediastinal structures or the brachial plexus), for whom dose-volume constraints for normal tissues could not be achieved. Such patients were treated with 70 Gy in 10 fractions or other regimens with a lower biologically effective dose.25,26 Doses (eg, 50 Gy in 4 fractions or 70 Gy in 10 fractions) were typically prescribed to the 70% to 90% isodose line covering the planning treatment volume (PTV).

or normal tissues could not be achieved. Such patients were treated with 70 Gy in 10 fractions or other regimens with a lower biologically effective dose.25,26 Doses (eg, 50 Gy in 4 fractions or 70 Gy in 10 fractions) were typically prescribed to the 70% to 90% isodose line covering the planning treatment volume (PTV). For plans for intensity-modulated radiotherapy or volumetric modulated arc therapy, an integrated boost to the gross tumor volume brought the total dose to 60 Gy in 4 fractions or 85 Gy in 10 fractions; this boost was done to mimic 3-dimensional conformal radiation–based SABR planning to generate a high-dose region inside the gross tumor volume. Treatment was delivered on consecutive weekdays with a break on intervening weekend days, if applicable. Other SABR treatment planning and delivery details have been previously described.27

done to mimic 3-dimensional conformal radiation–based SABR planning to generate a high-dose region inside the gross tumor volume. Treatment was delivered on consecutive weekdays with a break on intervening weekend days, if applicable. Other SABR treatment planning and delivery details have been previously described.27 Follow-up Evaluations and Definitions of Treatment Failure Follow-up evaluations after initial SABR included chest CT scans every 3 months for the first 2 years, every 6 months for the next 3 years, and annually thereafter. Scanning with PET and CT was commonly performed at 3 to 12 months after SABR to evaluate response and detect early recurrence. Local recurrence (LR) was defined as evidence on CT of progressive soft-tissue abnormalities in the same lobe as the primary tumor that then corresponded to areas avid on PET or positive biopsy findings.28 Regional recurrence (RR) was defined as similar CT, PET, or biopsy findings in the hila or mediastinum. Recurrence in previously uninvolved lobes or outside the thorax was defined as distant failure. Isolated local recurrence and iRR were defined as LR or RR with no other recurrence. In-field LRs were within 1 cm of the initial SABR PTV, marginal LRs overlapped with the PTV plus 1 cm, and out-of-field LRs appeared beyond the PTV plus 1 cm. Any patient with confirmed LR or RR also received PET, brain magnetic resonance imaging, and/or mediastinal endobronchial ultrasonography as indicated for restaging.

ield LRs were within 1 cm of the initial SABR PTV, marginal LRs overlapped with the PTV plus 1 cm, and out-of-field LRs appeared beyond the PTV plus 1 cm. Any patient with confirmed LR or RR also received PET, brain magnetic resonance imaging, and/or mediastinal endobronchial ultrasonography as indicated for restaging. Second primary lung carcinomas were defined by the modified Martini and Melamed criteria29 as a new tumor of different histologic or molecular subtype, or a new tumor of the same histologic characteristics in a different lobe appearing after a tumor-free interval of more than 2 years.3 All cases were reviewed before initial SABR and at recurrence by a multidisciplinary treatment team (D.R.G., Z.L., M.J., M.O., J.W.W., Q.-N. N., J.J.E., G.E., K.A., M.B.A., S.M.H., J.V.H., D.C.R., and J.Y.C.) consisting of thoracic surgeons, medical oncologists, radiation oncologists, interventional radiologists, pulmonologists, and radiologists. All available information was reviewed, including pathologic findings, clinical history, and imaging features. In all cases, this team determined which treatments were possible and reached consensus on a preferred treatment approach for each patient, as described below. Salvage Therapy In all cases, the choice of salvage therapy for iLR or iRR was made via consistent multidisciplinary evaluation. The process for salvage therapy selection and management approach is summarized in Figure 1.

Second primary lung carcinomas were defined by the modified Martini and Melamed criteria29 as a new tumor of different histologic or molecular subtype, or a new tumor of the same histologic characteristics in a different lobe appearing after a tumor-free interval of more than 2 years.3 All cases were reviewed before initial SABR and at recurrence by a multidisciplinary treatment team (D.R.G., Z.L., M.J., M.O., J.W.W., Q.-N. N., J.J.E., G.E., K.A., M.B.A., S.M.H., J.V.H., D.C.R., and J.Y.C.) consisting of thoracic surgeons, medical oncologists, radiation oncologists, interventional radiologists, pulmonologists, and radiologists. All available information was reviewed, including pathologic findings, clinical history, and imaging features. In all cases, this team determined which treatments were possible and reached consensus on a preferred treatment approach for each patient, as described below. Salvage Therapy In all cases, the choice of salvage therapy for iLR or iRR was made via consistent multidisciplinary evaluation. The process for salvage therapy selection and management approach is summarized in Figure 1. Figure 1. Management Guide for Treating Isolated Local Recurrence (iLR) and Isolated Regional Recurrence (iRR) After Stereotactic Ablative Radiotherapy (SABR) for Early-Stage Non–Small Cell Lung Cancer The workup involves positron emission tomography with computed tomography, magnetic resonance imaging of the brain, endobronchial ultrasound (if applicable based on findings of imaging), liver function tests, complete blood cell count, basic metabolic panel, and pulmonary function tests (if surgery is considered). Systemic therapy may be carefully considered in conjunction with locally directed therapy for iLR or iRR given the rates of distant metastases observed. SLR indicates sublobar resection.

ng), liver function tests, complete blood cell count, basic metabolic panel, and pulmonary function tests (if surgery is considered). Systemic therapy may be carefully considered in conjunction with locally directed therapy for iLR or iRR given the rates of distant metastases observed. SLR indicates sublobar resection. For iLR, repeated SABR was the preferred salvage therapy because of its low morbidity.23 Repeated SABR was possible when the iLR was sufficiently far from critical central chest structures25 or was outside the original SABR treatment volume (>1 cm from the initial SABR PTV). For iLRs encroaching on prior treatment fields (ie, marginal recurrences), an alternative to 50 Gy in 4 fractions (often 70 Gy in 10 fractions) was used for safety. All marginal recurrences for repeated SABR were peripheral and away from central chest structures and were discussed in a multidisciplinary setting to determine if the cumulative dose to the prior irradiated volume was safe. For patients who were not candidates for repeated SABR, or who were candidates for surgery, surgical resection was the next preferred option.19,20,21,22 All patients considering surgery had sufficient pulmonary function (predicted postoperative diffusing capacity for carbon monoxide and forced expiratory volume in 1 second >40%) and were deemed adequate risk candidates by a thoracic surgeon. For iLRs that could not be safely treated with reirradiation or with surgery, thermal ablation was preferred. Currently, thermal ablation can be done percutaneously, is suitable for lesions up to 3 cm in diameter, and can be used on tumors 1 cm or more from central chest structures.30,31,32,33 Patients undergoing thermal ablation must be able to safely tolerate a small pneumothorax.

f recurrence, median (range), mo 2.0 (0.0-25.4) 1.4 (0.1-30.1) NA Abbreviations: CT, computed tomography; ECOG, Eastern Cooperative Oncology Group; NA, not applicable; NSC NOS, non–small cell not otherwise specified; PET-CT, positron emission tomography and computed tomography; SABR, stereotactic ablative radiotherapy. a iRR vs no recurrence, P = .04. b ECOG score 0-1 vs 2-3 for iRR vs no recurrence, P = .02. Salvage Therapy Characteristics and Toxic Effects Several types of therapy were used for salvage treatment. Among patients with iLR, 15 had SABR as salvage treatment, 10 had surgery, 6 had thermal ablation, 5 had chemotherapy only, 2 had chemoradiotherapy, 1 had conventional radiotherapy, and 10 had no treatment; among patients with iRR, 26 had chemoradiotherapy, 12 had chemotherapy only, 8 had conventional radiotherapy, 1 had surgery, 1 had brachytherapy, and 5 had no treatment (eTables 1 and 2 in the Supplement). Among patients with iLR, grade 3 or greater toxic effects occurred in 1 of the 15 patients who had SABR (6.7%; pneumonitis), 4 of the 10 who had surgery (40.0%; postoperative renal, cardiac, and/or pulmonary sequelae; the 90-day mortality rate was 0% and symptoms resolved in all 4 patients), none of the 6 who had thermal ablation (0%), and 2 of the 5 patients who had systemic treatment (40.0%; hematologic). Among patients with iRR, grade 3 or greater toxic effects occurred in 10 of 26 patients who had chemoradiotherapy (38.5%; most common were esophagitis, fatigue, and hematologic effects), 1 of 8 who had conventional radiotherapy (12.5%; dyspnea), and 4 of 12 who had systemic therapy (33.3%; most common was fatigue). No patient experienced any salvage-related grade 5 event. Further details on the type and grade of toxic effects experienced for each salvage treatment can be found in eTables 1 and 2 in the Supplement.

ith surgery, thermal ablation was preferred. Currently, thermal ablation can be done percutaneously, is suitable for lesions up to 3 cm in diameter, and can be used on tumors 1 cm or more from central chest structures.30,31,32,33 Patients undergoing thermal ablation must be able to safely tolerate a small pneumothorax. For patients with iRR, bimodality treatment with nodal irradiation and systemic therapy was preferred. This approach is similar to that for patients with stage II or III NSCLC presenting with node-positive disease. The most common systemic therapy was platinum-paclitaxel doublet therapy, and local control was attempted with conventional radiotherapy to the involved nodes. Doses of 60 to 70 Gy in 2-Gy fractions were preferred, but in some circumstances, the dose was reduced to meet normal tissue (or patient) tolerance. Patients at high risk of complications from platinum-based doublet therapy were given mono-agent cytotoxic therapy. Patients who could not tolerate chemoradiotherapy were given either systemic therapy or radiotherapy alone. For patients unable to tolerate radiotherapy, the systemic agent was chosen based on toxic effects and appropriateness given the tumor’s molecular characteristics, and used until progression, eradication of disease, or death. If systemic therapy was not possible, definitive radiotherapy was given to a dose as close as possible to that used for stage III disease (60-70 Gy in 2-Gy fractions). Patients unable to undergo additional local or systemic therapy were given best supportive care.34

until progression, eradication of disease, or death. If systemic therapy was not possible, definitive radiotherapy was given to a dose as close as possible to that used for stage III disease (60-70 Gy in 2-Gy fractions). Patients unable to undergo additional local or systemic therapy were given best supportive care.34 Statistical Analysis The Kaplan-Meier method was used to estimate probabilities of overall survival (OS) and progression-free survival. Overall survival was calculated from completion of SABR to death from any cause and was also calculated from time of iLR or iRR to death from any cause. Time-varying covariate analysis using recurrence as the covariate was also used when examining OS between patients with iLR or iRR and no recurrence to account for survival bias. Progression-free survival was calculated from completion of SABR to the first failure at any site or death. Times to LR, RR, or distant recurrence were calculated from completion of SABR to the development of local, regional, or distant failure as both first events and cumulatively as concurrent or subsequent events over the course of the study.

as calculated from completion of SABR to the first failure at any site or death. Times to LR, RR, or distant recurrence were calculated from completion of SABR to the development of local, regional, or distant failure as both first events and cumulatively as concurrent or subsequent events over the course of the study. In addition to reporting crude recurrence rates, we also calculated the incidence of local, regional, and distant failure by using the Kaplan-Meier method with death as a competing risk.35 These criteria were also applied to reporting rates of second primary lung cancer. Treatment-related toxic effects were scored with the Common Terminology Criteria for Adverse Events, version 4.0.36 P < .05 (2-sided) was considered statistically significant. χ2 Analysis was used for categorical variables. Data were analyzed with SPSS, version 21.0 (IBM Corp), with a macro to calculate the cumulative incidence with competing risk.

s were scored with the Common Terminology Criteria for Adverse Events, version 4.0.36 P < .05 (2-sided) was considered statistically significant. χ2 Analysis was used for categorical variables. Data were analyzed with SPSS, version 21.0 (IBM Corp), with a macro to calculate the cumulative incidence with competing risk. Results The study population comprised 912 patients consecutively treated with SABR in 2004-2015 (Table 1 and eFigure 1 in the Supplement). The median patient age was 72 years (range, 46-91 years), 456 (50.0%) were men and 456 (50.0%) were women, 756 patients (82.9%) had clinical T1 disease, and 156 patients (17.1%) had clinical T2 or T3 disease (per the American Joint Committee on Cancer, 7th edition, guidelines).37 Among the 912 patients, 502 tumors (55.0%) were adenocarcinomas and 309 (33.9%) were squamous cell carcinomas. Nearly all cases (903 [99.0%]) had been confirmed by biopsy. The median follow-up time was 59.3 months (interquartile range [IQR], 37.7-87.9 months) from the initial SABR. About one-third of patients (318 [34.9%]) had staging mediastinal endobronchial ultrasonography for suspected lymphadenopathy on PET or CT (eg, node ≥1 cm). Most patients (773 [84.8%]) had been referred for initial SABR for inoperable disease or medical contraindications, and the other 139 patients (15.2%) had declined surgery or were randomized to SABR on the STARS (Stereotactic Ablative Radiotherapy [SABR] in Stage I Non-small Cell Lung Cancer Patients) protocol (ClinicalTrials.gov identifier NCT02357992). Most patients (754 [82.7%]) had good performance status, with Eastern Cooperative Oncology Group scores of 0 to 1 at diagnosis.

1-30.1 months) for those with iRR; this interval was used to exclude distant disease and to allow case review by the multidisciplinary team. Time to recurrence was numerically shorter for those with iRR than for those with iLR. Other characteristics between patients with iLR and those with iRR are presented in Table 2. Table 2. Characteristics of All Patients With Isolated Local Recurrence (iLR) or Isolated Regional Recurrence (iRR) and Patients Without Recurrence Characteristic Patients, No. (%) With iLR (n = 49) With iRR (n = 53) Without Recurrence (n = 658) Age at recurrence, median (range), y 74 (57-89) 70 (49-89) NA Sex Male 25 (51.0) 34 (64.2) 323 (49.1)a Female 24 (49.0) 19 (35.8) 335 (50.9) ECOG score at recurrence 0 1 (2.0) 1 (1.9) 63 (9.6) 1 35 (71.4) 36 (67.9) 481 (73.1) 2 10 (20.4) 12 (22.6) 104 (15.8)b 3 3 (6.1) 4 (7.5) 10 (1.5) Tumor status (initial stage) T1 41 (83.7) 44 (83.0) 553 (84.0) T2 (T2a: ≤5 cm, pleural invasion) 7 (14.3) 8 (15.1) 96 (14.6) T3 (with satellite nodule) 1 (2.0) 1 (1.9) 9 (1.4) Tumor histologic findings (initial) Adenocarcinoma 23 (46.9) 25 (47.2) 366 (55.6) Squamous cell carcinoma 22 (44.9) 21 (39.6) 217 (33.0) Other 1 (2.0) 2 (3.8) 14 (2.1) NSC NOS 2 (4.1) 5 (9.4) 54 (8.2) Unknown or no pathologic findings obtained 1 (2.0) 0 7 (1.1) Recurrence confirmed Biopsy 38 (77.6) 40 (75.5) NA PET-CT 8 (16.3) 13 (24.5) NA CT 3 (6.1) 0 NA Time to recurrence, median (range), mo 14.5 (1.5-60.8) 9.0 (1.9-70.7) NA Received salvage treatment, No. (%) 39 (79.6) 48 (90.6) NA SABR 15 (38.5) NA NA Surgery 10 (25.6) 1 (2.1) NA Thermal ablation 6 (15.4) NA NA Chemoradiotherapy 2 (5.1) 26 (54.2) NA Radiotherapy alone 1 (2.6) 12 (25) NA Systemic therapy alone 5 (12.8) 8 (16.7) NA Other NA 1 (2.1) NA Time to salvage from time of recurrence, median (range), mo 2.0 (0.0-25.4) 1.4 (0.1-30.1) NA Abbreviations: CT, computed tomography; ECOG, Eastern Cooperative Oncology Group; NA, not applicable; NSC NOS, non–small cell not otherwise specified; PET-CT, positron emission tomography and computed tomography; SABR, stereotactic ablative radiotherapy.

surgery or were randomized to SABR on the STARS (Stereotactic Ablative Radiotherapy [SABR] in Stage I Non-small Cell Lung Cancer Patients) protocol (ClinicalTrials.gov identifier NCT02357992). Most patients (754 [82.7%]) had good performance status, with Eastern Cooperative Oncology Group scores of 0 to 1 at diagnosis. Table 1. Characteristics and Outcomes for All Patients Initially Treated With Stereotactic Ablative Radiotherapy for Early-Stage Non–Small Cell Lung Cancer Characteristic Patients, No. (%) (N = 912) Age, median (range), y 72 (46-91) Sex Male 456 (50.0) Female 456 (50.0) Tumor status T1 756 (82.9) T2 (T2a: ≤5 cm, pleural invasion) 140 (15.4) T3 (with satellite nodule) 16 (1.8) Tumor histologic findings Adenocarcinoma 502 (55.0) Squamous cell carcinoma 309 (33.9) Other 23 (2.5) NSC NOS 69 (7.6) No pathologic findings obtained 9 (1.0) Tumor location Peripheral 760 (83.3) Central 152 (16.7) SABR dose/fraction (BED) 50 Gy/4 fractions (112.5 Gy) 720 (78.9) 70 Gy/10 fractions (119 Gy) 124 (13.6) Others (75-180 Gy) 68 (7.5) Endobronchial ultrasonography Yes 318 (34.9) No 594 (65.1) Reason for SABR Inoperable disease 773 (84.8) Declined surgery 139 (15.2) ECOG score 0 87 (9.5) 1 667 (73.1) 2 146 (16.0) 3 12 (1.3) First site of recurrence Isolated LR 49 (5.4) Isolated RR 46 (5.0) Isolated DM 96 (10.5) Concurrent LR and RR 7 (0.8) Concurrent LR and DM 19 (2.1) Concurrent RR and DM 29 (3.2) All sites failure 8 (0.9) Cumulative recurrence of events over the entire time course LR 91 (10.0) RR 105 (11.5) DM 183 (20.1) Time to any recurrence as first event, median (range), mo LR 14.9 (1.5-91.9) RR 10.5 (1.4-70.7) DM 11.6 (0.2-91.9) Follow-up time, median (IQR), mo 59.3 (37.7-87.9) Overall survival time, median (95% CI), mo 56.3 (51.4-61.2) 1-y rate, % 88.8 3-y rate, % 64.9 5-y rate, % 47.7 Progression-free survival time, median (95% CI), mo 39.7 (34.6-44.8) 1-y rate, % 78.1 3-y rate, % 52.7 5-y rate, % 39.1 Second primary lung cancer 68 (7.5) Time to second primary lung cancer, median (range), mo 23.6 (1.2-122.4) Abbreviations: BED, biological effective dose; DM, distant metastasis; ECOG, Eastern Cooperative Oncology Group; IQR, interquartile range; LR local recurrence; NSC NOS, non–small cell not otherwise specified; PFS, progression-free survival; RR, regional recurrence; SABR, stereotactic ablative radiotherapy.

(1.2-122.4) Abbreviations: BED, biological effective dose; DM, distant metastasis; ECOG, Eastern Cooperative Oncology Group; IQR, interquartile range; LR local recurrence; NSC NOS, non–small cell not otherwise specified; PFS, progression-free survival; RR, regional recurrence; SABR, stereotactic ablative radiotherapy. SI conversion factor: To convert gray to rad, multiply by 100. Recurrence Patterns and Survival After Initial SABR Recurrences as cumulative and first events for the 912 patients are presented in Table 1 and eFigure 2 in the Supplement. Most patients (658 [72.1%]) did not experience recurrence. First failures were iLR in 49 patients (5.4%), iRR in 46 (5.0%), and simultaneous iLR and iRR in 7 (0.8%). (For the purposes of this analysis, these 7 patients with simultaneous iLR and iRR were considered to have iRR, bringing the total number of patients with iLR or iRR to 102 [11.2%].) Distant failure as a first event, alone or in combination with other failure, was the predominant pattern of failure (152 patients [16.7%]). The median time to iLR was 14.9 months (IQR, 1.5-91.9 months), to iRR was 10.5 months (IQR, 1.4-70.7 months), and to distant failure was 11.6 months (IQR, 0.2-91.9 months). The cumulative rates of recurrence (calculated not with the Kaplan-Meier method, but rather considering subsequent events in addition to first events) were 10.0% for LR (91 of 912), 11.5% for RR (105 of 912), and 20.1% for distant failure (183 of 912). The cumulative crude rate of second primary lung cancer was 7.5% (68 of 912).

tive rates of recurrence (calculated not with the Kaplan-Meier method, but rather considering subsequent events in addition to first events) were 10.0% for LR (91 of 912), 11.5% for RR (105 of 912), and 20.1% for distant failure (183 of 912). The cumulative crude rate of second primary lung cancer was 7.5% (68 of 912). The cumulative incidence of LR, RR, distant metastasis, and second primary lung cancer calculated with the Kaplan-Meier method with death as a competing risk is presented in eFigure 2 in the Supplement. The cumulative rates for LR with death as a competing risk were 4% at 1 year, 9% at 3 years, and 11% at 5 years; corresponding rates for RR were 6% at 1 year, 11% at 3 years, and 12% at 5 years; for distant failure, 10% at 1 year, 18% at 3 years, and 21% at 5 years; and for second primary lung cancer, 5.9% at 1 year, 10.9% at 3 years, and 11.9% at 5 years. Rates of OS for all 912 patients were 88.8% at 1 year, 64.9% at 3 years, and 47.7% at 5 years; corresponding rates of progression-free survival were 78.1% at 1 year, 52.7% at 3 years, and 39.1% at 5 years (eFigure 3 in the Supplement).

second primary lung cancer, 5.9% at 1 year, 10.9% at 3 years, and 11.9% at 5 years. Rates of OS for all 912 patients were 88.8% at 1 year, 64.9% at 3 years, and 47.7% at 5 years; corresponding rates of progression-free survival were 78.1% at 1 year, 52.7% at 3 years, and 39.1% at 5 years (eFigure 3 in the Supplement). Characteristics of Patients Who Received Salvage Therapy for iLR or iRR Most patients with iLR (38 of 49 [77.6%]) or iRR (40 of 53 [75.5%]) had biopsy confirmation of recurrence; 39 patients with iLR (79.6%) and 48 patients with iRR (90.6%) received salvage therapy (Table 2). Median times to recurrence after SABR were 14.5 months (range, 1.5-60.8 months) for iLR and 9.0 months (range, 1.9-70.7 months) for iRR. The median follow-up time for patients with iLR or iRR was 57.2 months (IQR, 37.7-87.6 months) from the initial SABR and 38.5 months (IQR, 19.9-69.3 months) after the recurrence. The mean time from recurrence to initiation of salvage treatment was 2.0 months (IQR, 0.0-25.4 months) for those with iLR and 1.4 months (IQR, 0.1-30.1 months) for those with iRR; this interval was used to exclude distant disease and to allow case review by the multidisciplinary team. Time to recurrence was numerically shorter for those with iRR than for those with iLR. Other characteristics between patients with iLR and those with iRR are presented in Table 2.

f recurrence, median (range), mo 2.0 (0.0-25.4) 1.4 (0.1-30.1) NA Abbreviations: CT, computed tomography; ECOG, Eastern Cooperative Oncology Group; NA, not applicable; NSC NOS, non–small cell not otherwise specified; PET-CT, positron emission tomography and computed tomography; SABR, stereotactic ablative radiotherapy. a iRR vs no recurrence, P = .04. b ECOG score 0-1 vs 2-3 for iRR vs no recurrence, P = .02.

ad conventional radiotherapy (12.5%; dyspnea), and 4 of 12 who had systemic therapy (33.3%; most common was fatigue). No patient experienced any salvage-related grade 5 event. Further details on the type and grade of toxic effects experienced for each salvage treatment can be found in eTables 1 and 2 in the Supplement. Although systemic therapy alone was not considered definitive for local salvage, it was included as a form of salvage treatment in this study given its presumed role in reducing morbidity and mortality from recurrent disease. One patient with iRR had brachytherapy as salvage treatment for mediastinal recurrence invading the trachea, and another underwent surgery for a low disease burden. Two patients with iLR received chemoradiotherapy, one for aggressive management and the other as induction therapy to reduce the size of the radiotherapy field.

atient with iRR had brachytherapy as salvage treatment for mediastinal recurrence invading the trachea, and another underwent surgery for a low disease burden. Two patients with iLR received chemoradiotherapy, one for aggressive management and the other as induction therapy to reduce the size of the radiotherapy field. Survival After iLR and iRR Overall survival time was significantly longer for patients with iLR or iRR who received salvage treatment (n = 87) than for those with iLR or iRR who did not receive salvage treatment (n = 15) (37 vs 7 months from time of recurrence; hazard ratio [HR], 0.40; 95% CI, 0.09-0.66; P = .006; Figure 2A). Rates of OS after recurrence for patients with iLR plus salvage treatment were 92.0% at 1 year, 55.3% at 3 years, and 33.2% at 5 years (Figure 2B); for patients with iRR plus salvage treatment, rates of OS were 80.3% at 1 year, 40.4% at 3 years, and 20.7% at 5 years (Figure 2C). Rates of OS after recurrence for patients with untreated iLR were lower, at 64.8% at 1 year, 34.2% at 3 years, and 0% at 5 years; for patients with untreated iRR, rates of OS were 20.0% at 1 year, 0% at 3 years, and 0% at 5 years (Figure 2A).

OS were 80.3% at 1 year, 40.4% at 3 years, and 20.7% at 5 years (Figure 2C). Rates of OS after recurrence for patients with untreated iLR were lower, at 64.8% at 1 year, 34.2% at 3 years, and 0% at 5 years; for patients with untreated iRR, rates of OS were 20.0% at 1 year, 0% at 3 years, and 0% at 5 years (Figure 2A). Figure 2. Survival Outcomes After Salvage Therapy for Isolated Local Recurrence (iLR) or Isolated Regional Recurrence (iRR) A, Overall survival after recurrence for patients with iLR or iRR who did or did not undergo salvage therapy. B, Overall survival from the time of initial stereotactic ablative radiation therapy (SABR) for patients with iLR who received salvage treatment vs for patients with no recurrence. C, Overall survival from the time of initial SABR for patients with iRR who received salvage treatment vs for patients with no recurrence.

B, Overall survival from the time of initial stereotactic ablative radiation therapy (SABR) for patients with iLR who received salvage treatment vs for patients with no recurrence. C, Overall survival from the time of initial SABR for patients with iRR who received salvage treatment vs for patients with no recurrence. When evaluating whether salvageable recurrence adversely affects survival, we found that OS was no different for patients with iLR who received salvage treatment than for patients who had no recurrence after initial SABR (log-rank P = .65); rates of OS at 5 years from initial SABR were no different between patients with iLR and salvage treatment (57.9%) and patients with no recurrence (54.9%; HR, 0.89; 95% CI, 0.56-1.43; time-varying P = .10; HR, 1.51; 95% CI, 0.92-2.47; Figure 2B and Table 3). However, rates of OS at 5 years for patients with iRR who received salvage treatment (31.1%) were significantly lower than those for patients with no recurrence (log-rank P = .049; HR, 1.43; 95% CI, 1.00-2.34; time-varying P < .001; HR, 2.08; 95% CI, 1.45-3.01; Figure 2C and Table 3).

0.92-2.47; Figure 2B and Table 3). However, rates of OS at 5 years for patients with iRR who received salvage treatment (31.1%) were significantly lower than those for patients with no recurrence (log-rank P = .049; HR, 1.43; 95% CI, 1.00-2.34; time-varying P < .001; HR, 2.08; 95% CI, 1.45-3.01; Figure 2C and Table 3). Table 3. Outcomes of Patients After Salvage for Isolated Local Recurrence (iLR) or Isolated Regional Recurrence (iRR) Compared With Patients Without Recurrence Characteristic Patients With iLR (n = 39) Patients With iRR (n = 48) Patients Without Recurrence (n = 658) OS time from initial SABR, median (95% CI), mo 62.3 (38.2-86.4) 52.3 (35.3-69.2) 65.3 (60.3-70.3)a,b OS rate, % 1 y 97.4 95.8 88.8 3 y 77.3 58.1 69.5 5 y 57.9 31.1 54.9 OS time after recurrence, median (95% CI), mo 51.6 (0.0-110.6) 28.0 (14.6-41.5) OS rate, % 1 y 92.0 80.3 NA 3 y 55.3 40.4 NA 5 y 33.2 20.7 NA Cumulative subsequent recurrence for patients after salvage, No. (%) LR 7 (17.9) 1 (2.1) NA RR 9 (23.1) 2 (4.2) NA DM 10 (25.6) 14 (29.2) NA Time to subsequent event for patients after salvage, median (range), mo LR 11.8 (2.7-20.8) 9.6c NA RR 10.3 (1.2-16.2) 23.4 (12.1-34.6) NA DM 12.5 (0.3-78.6) 8.3 (1.2-34.6) NA Location of distant metastases, No./total No. (%) Intrathoracic 9/10 (90.0) 8/14 (57.1) NA Extrathoracic 1/10 (10.0) 6/14 (42.9) NA Abbreviations: DM, distant mestastasis; LR local recurrence; NA, not applicable; OS, overall survival; RR, regional recurrence; SABR, stereotactic ablative radiotherapy.

8.3 (1.2-34.6) NA Location of distant metastases, No./total No. (%) Intrathoracic 9/10 (90.0) 8/14 (57.1) NA Extrathoracic 1/10 (10.0) 6/14 (42.9) NA Abbreviations: DM, distant mestastasis; LR local recurrence; NA, not applicable; OS, overall survival; RR, regional recurrence; SABR, stereotactic ablative radiotherapy. a For iLR vs no recurrence, P = .65 by Kaplan-Meier analysis, and P = .10 by time-varying analysis. b For iRR vs no recurrence, P = .049 by Kaplan-Meier analysis, and P < .001 by time-varying analysis. c There is no range because there was only 1 patient. Disease Progression After Salvage Therapy Subsequent recurrence events after salvage are presented in Table 3. Nineteen of 39 patients with iLR (48.7%) and 33 of 48 patients with iRR (68.8%) had no further recurrence. Subsequent LR occurred in 7 patients with iLR (17.9%) and 1 patient with iRR (2.1%), subsequent RR occurred in 9 patients with iLR (23.1%) and in 2 patients with iRR (4.2%), and subsequent distant failure occurred in 10 patients with iLR (25.6%) and 14 patients with iRR (28.6%). Sites of distant failure differed for patients with iLR (9 of 10 [90.0%], lungs and 1 of 10 [10.0%], extrathoracic) vs iRR (8 of 14 [57.1%], lungs and 6 of 14 [42.9%], extrathoracic): extrathoracic sites included the bone, liver, adrenal glands, and brain. All patients with iLR or iRR who did not receive salvage therapy had progressive disease (n = 15).

or patients with iLR (9 of 10 [90.0%], lungs and 1 of 10 [10.0%], extrathoracic) vs iRR (8 of 14 [57.1%], lungs and 6 of 14 [42.9%], extrathoracic): extrathoracic sites included the bone, liver, adrenal glands, and brain. All patients with iLR or iRR who did not receive salvage therapy had progressive disease (n = 15). Discussion Our key findings from this large study of long-term outcomes after salvage treatment for locally or regionally recurrent disease after SABR for early-stage NSCLC are as follows. First, life expectancy for patients with iLR after SABR who subsequently received salvage treatment was virtually the same as that for patients without recurrence. Moreover, at 3 years after recurrence, 50% to 60% of patients with iLR or iRR who received salvage treatment never had another recurrence, showing that the potential cure rate with salvage treatment for such patients can be substantial. The OS for patients with iRR was poorer than that for patients with iLR or no recurrence, but was akin to that for patients with stage III disease. Thus, although salvage treatment offers better outcomes as a whole, iLR and iRR represent 2 distinct clinical paths, an important distinction for clinicians managing such cases.

S for patients with iRR was poorer than that for patients with iLR or no recurrence, but was akin to that for patients with stage III disease. Thus, although salvage treatment offers better outcomes as a whole, iLR and iRR represent 2 distinct clinical paths, an important distinction for clinicians managing such cases. We further found support for using salvage treatment, because patients receiving any salvage had better OS than patients who did not. Although one might expect survival in patients who did not receive salvage treatment to be poorer (perhaps because comorbidities precluded salvage), we found that all patients with recurrence who did not receive salvage treatment experienced progression and none were alive at 3 to 5 years after recurrence. For those who experienced progression after salvage treatment, that progression was mostly distant, and sites varied between the 2 recurrence groups. Specifically, 90% of recurrences after salvage treatment for iLR occurred in a different lung lobe, whereas distant failure for patients with iRR who received salvage treatment was more often extrathoracic and disseminated, which may have contributed to the poorer OS in the iRR subgroup.

etween the 2 recurrence groups. Specifically, 90% of recurrences after salvage treatment for iLR occurred in a different lung lobe, whereas distant failure for patients with iRR who received salvage treatment was more often extrathoracic and disseminated, which may have contributed to the poorer OS in the iRR subgroup. Patients with iLR who received salvage treatment had higher rates of subsequent LR and RR events than did patients with iRR who received salvage treatment. This result was not surprising because local lobar disease was apparently controlled in most patients with iRR but not in patients with iLR. Furthermore, unlike patients with iRR, most patients with iLR did not receive nodal or mediastinal sterilizing therapy (ie, chemoradiotherapy), which could make regional nodes the most logical location for recurrence, should recurrence take place. Subsequent LR and RR was managed with the same approach as that for initial salvage treatment, likely contributing to the favorable OS for patients with iLR.

r mediastinal sterilizing therapy (ie, chemoradiotherapy), which could make regional nodes the most logical location for recurrence, should recurrence take place. Subsequent LR and RR was managed with the same approach as that for initial salvage treatment, likely contributing to the favorable OS for patients with iLR. Although most patients achieved disease control, the 40% rate of recurrence after salvage treatment suggests the potential for systemic therapy upfront for patients with either iLR or iRR to eradicate distant or residual microscopic disease at the time of recurrence. To this end, the addition of immunotherapy to SABR for patients with newly diagnosed early-stage disease or iLR after SABR (I-SABR [ClinicalTrials.gov identifier NCT03110978]) is being tested.38 Finally, we showed that a variety of salvage techniques, including thermal ablation (not currently included in national guidelines), could be successful for patients who are unable to undergo other locally directed therapies.

ease or iLR after SABR (I-SABR [ClinicalTrials.gov identifier NCT03110978]) is being tested.38 Finally, we showed that a variety of salvage techniques, including thermal ablation (not currently included in national guidelines), could be successful for patients who are unable to undergo other locally directed therapies. Limitations This retrospective review provides data on outcomes for patients with recurrence after SABR. Our approach is similar to the National Comprehensive Cancer Network consensus algorithm. Our study has some limitations, chief among them its retrospective nature, with all the inherent biases. Any nonrandomized comparison of the effectiveness of various salvage techniques is limited by bias in assigning patients to a particular salvage therapy based on performance and disease status. Our results, based on a low rate (11%) of iLR or iRR in patients undergoing SABR, indicate that a prospective clinical trial would be a challenge. Finally, the single-institution nature of this study was both a limitation and a strength in that it allowed a relatively complete long-term analysis. Conclusions Life expectancy after salvage treatment for iLR was similar to that for patients without recurrence, but survival after salvage treatment for iRR was similar to that of patients with stage III NSCLC. Because salvage treatment for iLR or iRR was based on a consistent multidisciplinary approach, the results of this study may help clinicians and patients in treatment decision making. Supplement. eFigure 1. CONSORT-Style Diagram Showing Patterns of First Recurrence

Conclusions Life expectancy after salvage treatment for iLR was similar to that for patients without recurrence, but survival after salvage treatment for iRR was similar to that of patients with stage III NSCLC. Because salvage treatment for iLR or iRR was based on a consistent multidisciplinary approach, the results of this study may help clinicians and patients in treatment decision making. Supplement. eFigure 1. CONSORT-Style Diagram Showing Patterns of First Recurrence eFigure 2. Cumulative Incidence of Local Recurrence (LR), Regional Recurrence (RR), Distant Metastasis (DM), and Second Primary Lung Cancer (SPLC) After Definitive Stereotactic Ablative Radiation Therapy (SABR) Among 912 Patients With Early-stage Non-Small Cell Lung Cancer eFigure 3. Progression-Free Survival (PFS) (A) and Overall Survival (B) Outcomes for 912 Patients, Beginning at the Completion of Stereotactic Ablative Radiation Therapy for Early-stage Non-Small Cell Lung Cancer eTable 1. Salvage Therapy for Isolated Local Recurrences eTable 2. Salvage Therapy for Isolated Regional Recurrence Click here for additional data file.

Introduction Feasible, accurate, and reproducible assessment of left ventricular ejection fraction (LVEF) is an important objective of noninvasive cardiac imaging. Whether LVEF is preserved or reduced currently forms the basis for the classification of patients with heart failure. Additionally, LVEF is an important predictor of prognosis in patients with myocardial infarction,1,2,3 heart failure,4,5,6 and valve disease.7 Moreover, current practice guidelines use LVEF thresholds for decision making in different clinical scenarios, such as the recommendation regarding device implantation or pharmacologic therapy in patients with heart failure8,9 and the recommendation for valve replacement in patients with severe valvular heart disease.10 Left ventricular ejection fraction is also a common enrollment criterion and/or end point for clinical trials.11

h as the recommendation regarding device implantation or pharmacologic therapy in patients with heart failure8,9 and the recommendation for valve replacement in patients with severe valvular heart disease.10 Left ventricular ejection fraction is also a common enrollment criterion and/or end point for clinical trials.11 Left ventricular ejection fraction can be determined by using multiple noninvasive imaging modalities, including echocardiography, cardiac magnetic resonance (CMR) imaging, and gated single-photon emission computed tomography (SPECT) imaging. All of these methods are routinely used for clinical decision making as well as research study enrollment. However, few data exist regarding the agreement between LVEF determined by these different methods. Prior studies have been limited by small numbers of participants, sometimes including only healthy volunteers, with imaging performed by a single center, and have compared only 2 imaging modalities.12,13,14 As LVEF cut points are often the basis for clinical management decisions and trial eligibility, the implications of variability are substantial.

by small numbers of participants, sometimes including only healthy volunteers, with imaging performed by a single center, and have compared only 2 imaging modalities.12,13,14 As LVEF cut points are often the basis for clinical management decisions and trial eligibility, the implications of variability are substantial. The Surgical Treatment for Ischemic Heart Failure (STICH) was an international multicenter trial aimed to compare coronary artery bypass grafting (CABG) and medical therapy for patients with heart failure, coronary artery disease (CAD), and left ventricular (LV) systolic dysfunction defined as LVEF of 35% or less.15,16 In this trial, any of 3 diagnostic methods (echocardiography, gated SPECT imaging, or CMR) could be used by a local clinical site to measure LVEF in order to determine a patient’s trial eligibility. All patients enrolled in the STICH trial were required to have a baseline determination of LVEF, and a subset of patients underwent this determination by multiple modalities, including echocardiography, gated SPECT imaging, and/or CMR. All LVEF data obtained by echocardiography, CMR, and SPECT were measured by respective core laboratories. Therefore, the STICH trial provides a unique opportunity to correlate core laboratory assessment of LVEF data between different modalities.

e modalities, including echocardiography, gated SPECT imaging, and/or CMR. All LVEF data obtained by echocardiography, CMR, and SPECT were measured by respective core laboratories. Therefore, the STICH trial provides a unique opportunity to correlate core laboratory assessment of LVEF data between different modalities. We conducted this study to determine the variability among imaging modalities and among different echocardiographic methods for assessing LVEF in patients with reduced LV systolic function, and to compare the association between these measurements and subsequent mortality in patients with ischemic cardiomyopathy.

e modalities, including echocardiography, gated SPECT imaging, and/or CMR. All LVEF data obtained by echocardiography, CMR, and SPECT were measured by respective core laboratories. Therefore, the STICH trial provides a unique opportunity to correlate core laboratory assessment of LVEF data between different modalities. We conducted this study to determine the variability among imaging modalities and among different echocardiographic methods for assessing LVEF in patients with reduced LV systolic function, and to compare the association between these measurements and subsequent mortality in patients with ischemic cardiomyopathy. Methods Patients A total of 2136 patients with LVEF of 35% or less were enrolled in the STICH trial.16 All patients provided written informed consent, as approved by the local institutional review board. Patients were enrolled at 127 clinical sites in 26 countries from July 24, 2002, to May 5, 2007; all had CAD amenable to CABG. Each of the 127 enrolling sites had to obtain institutional review board approval for STICH. We followed the Standards for Reporting of Diagnostic Accuracy (STARD) reporting guideline. A manual of operation for each modality was produced by the core laboratory for that modality to standardize imaging technique. Each site was required to submit 1 to 3 studies that fulfilled imaging requirements before enrollment. When studies did not meet these requirements, additional studies were requested until requirements were met. Patients with baseline imaging data received by the core laboratories for 1 or more imaging modalities were considered for inclusion. Patients with LVEF measured 90 days or more from study randomization or with study quality deemed by the core laboratory as being unusable for measurement were excluded. Determination of LVEF was performed by a separate core laboratory for each modality, independent of clinical information, treatment assignment, and data from other modalities.17 Each core laboratory provided oversight of quality control and assessed the quality of each study as excellent, good, fair, borderline, or unusable. Left ventricular end-systolic volume for each modality was indexed for body surface area.

of clinical information, treatment assignment, and data from other modalities.17 Each core laboratory provided oversight of quality control and assessed the quality of each study as excellent, good, fair, borderline, or unusable. Left ventricular end-systolic volume for each modality was indexed for body surface area. Calculation of LVEF Left ventricular ejection fraction was determined from LV end-diastolic volume and end-systolic volume using the following standard formula: LVEF = [(end-diastolic volume) – (end-systolic volume)]/end-diastolic volume Imaging Echocardiography was performed at most sites for patient enrollment. Determination of LVEF was attempted according to the recommendations of the American Society of Echocardiography18 using the Simpson method.17 Measurements were averaged over 3 cardiac cycles for patients in sinus rhythm, and 3 to 5 cardiac cycles for those in atrial fibrillation. If 2 apical views were not available for LV volume measurement, only 1 apical view (single-plane Simpson measurement) was used for determination of LVEF. For Simpson measurements, the LV endocardial border was traced contiguously from one side of the mitral annulus to the other side excluding the papillary muscles and trabeculations. Left ventricular ejection fraction was also estimated visually in most patients and when the definition of the LV endocardial border was not satisfactory from any of the apical views, visual estimate was the only echocardiographic determination of LVEF.19

e other side excluding the papillary muscles and trabeculations. Left ventricular ejection fraction was also estimated visually in most patients and when the definition of the LV endocardial border was not satisfactory from any of the apical views, visual estimate was the only echocardiographic determination of LVEF.19 Gated myocardial perfusion SPECT imaging, predominantly using sestamibi, was performed at clinical sites using a standard protocol. The gated raw projections were reconstructed by the radionuclide core laboratory using automatic software (AutoSPECT). When appropriate, an algorithm was applied to correct for motion. Resting studies accounted for 82% of measurements of LVEF; the remainder was obtained from poststress studies. Gated short-axis images were reviewed by a core laboratory technologist to optimize the accuracy of automatically determined LV contours, which were measured in end-systole and end-diastole. Manual adjustment addressed incorrect valve plane placement or contour deviations because of extracardiac radioactivity. Gated SPECT images were analyzed for LVEF by the radionuclide core laboratory using quantitative gated SPECT (QGS) software.20

rmined LV contours, which were measured in end-systole and end-diastole. Manual adjustment addressed incorrect valve plane placement or contour deviations because of extracardiac radioactivity. Gated SPECT images were analyzed for LVEF by the radionuclide core laboratory using quantitative gated SPECT (QGS) software.20 Cardiac magnetic resonance imaging was performed by clinical sites that had the required CMR platform and software.17,21 A minimum of 2 data sets of short- and long-axis views were required to allow the core laboratory to select images of the highest quality. All gated data were displayed and reviewed by CMR core laboratory expert technologists to verify that LV boundaries were accurately demarcated. Short-axis images allowed determination of LVEF, using software developed by the CMR core laboratory at the University of Southern California (USC Cardio) and based on the Simpson method. Papillary muscles and trabeculations were considered to be part of the LV cavity. All short-axis data were reviewed by a technologist and adjusted manually, if necessary, to optimize accuracy of LV contour borders.

MR core laboratory at the University of Southern California (USC Cardio) and based on the Simpson method. Papillary muscles and trabeculations were considered to be part of the LV cavity. All short-axis data were reviewed by a technologist and adjusted manually, if necessary, to optimize accuracy of LV contour borders. Statistical Analysis In this secondary study of the STICH trial, agreement of the core laboratory determinations of LVEF between modalities were assessed using 5 indices of variability: mean signed difference, mean absolute difference, Pearson correlation coefficient, Bland-Altman plots in which the mean of 2 measurements was plotted against the difference,22 and coverage probability.23 For determining LVEF by echocardiography, 3 methods (biplane, single plane, and visual estimation) were included in this comparison. For the coverage probability analysis, which is an assessment of the proportion of participants where a prespecified level of agreement is present, agreement was assumed to be present if the LVEF measures being compared were within 5% of each other. Because LVEF measures with SPECT performed after stress could be influenced by ischemia or stunning, Bland-Altman plots were repeated after exclusion of the 18% with SPECT LVEF assessed following stress. To assess the impact of nonsimultaneous imaging, the number of days between performances of various modalities was considered. Finally, the prognostic effect of the different measures of LVEF for association with all-cause mortality was assessed using Cox regression models. In each case, the relationship of LVEF with mortality was modeled using restricted cubic spline functions.24 The relative risk of every 5% LVEF increment, expressed as hazard ratio and 95% confidence interval, was also calculated using the Cox model. The longest available follow-up information was used for each patient.16,25 This was a 2-sided test with a P value less than .05 required for significance.

ic spline functions.24 The relative risk of every 5% LVEF increment, expressed as hazard ratio and 95% confidence interval, was also calculated using the Cox model. The longest available follow-up information was used for each patient.16,25 This was a 2-sided test with a P value less than .05 required for significance. Results Clinical Characteristics The population included 2032 patients (95.1%) (mean [SD] age, 60.9 [9.6] years; 1759 [86.6%] male) of the 2136 patients in the STICH trial who had LVEF assessed by at least 1 imaging modality. A baseline echocardiographic assessment of LVEF was received by the core laboratory in 1978 patients. Of these, 30 (1.5%) were excluded (imaging was performed >90 days before or after randomization in 24 cases and images were unusable in 6 cases). The remaining 1948 patients constitute the population studied with echocardiography. Biplane Simpson method of determining LVEF was performed in 897 (46%) of the 1948 patients. Only 3 patients (<1%) received echo contrast. Gated SPECT images for assessment of LVEF were received by the core laboratory in 790 patients. Of these, 16 (2.0%) were excluded (imaging was performed >90 days before or after randomization in 14 cases and images were unusable in 2 cases). The remaining 774 patients constitute the population studied with SPECT. Cardiac magnetic resonance assessment of LVEF was performed and received by the core laboratory in 425 patients. Of these, 7 (1.6%) were excluded (imaging was performed >90 days before or after randomization in 6 cases and 1 patient with out-of-range LVEF measurement that could not be confirmed by the CMR core laboratory). The remaining 417 patients constitute the population studied with CMR. Although most patients had echocardiography, the population included 84 patients who had only gated SPECT or CMR. Thus, 1107 (57%) of the 1948 patients with echocardiographic determination of LVEF also had LVEF determined by a second modality. Characteristics of the patients who were evaluable are shown in Table 1. Patients with better-quality echocardiographic images were more likely to undergo imaging with SPECT or CMR. For those with excellent echocardiographic image quality, 42.3% had SPECT (42.5% were good, 35.6% were fair, and 29.7% were borderline; P < .001). For those with excellent echocardiographic image quality, 36.6% had CMR (21.1% were good, 18.2% were fair, and 15.6% were borderline; P < .001).

o imaging with SPECT or CMR. For those with excellent echocardiographic image quality, 42.3% had SPECT (42.5% were good, 35.6% were fair, and 29.7% were borderline; P < .001). For those with excellent echocardiographic image quality, 36.6% had CMR (21.1% were good, 18.2% were fair, and 15.6% were borderline; P < .001). Compared with those who had only imaging by echocardiography, patients who also had imaging by SPECT or CMR more often had prior myocardial infarction (79.9% vs 83.9%; P = .02), prior percutaneous coronary revascularization (9.9% vs 21.0%; P < .001), greater anterior akinesia or dyskinesia (45.3 [50.5%] vs 48.4 [26.7%]; P < .001), and lower New York Heart Association heart failure class (class I or II in 58% vs 61.6%; P < .001). Left ventricular ejection fraction was measured by all 3 modalities in 127 patients; these patients compared with those with LVEF measured by 1 or 2 modalities more often had prior myocardial infarction (93.7% vs 80.9%; P < .001), prior percutaneous coronary revascularization (34.6% vs 14.1%; P < .001), greater anterior akinesia or dyskinesia (57.5 [13.1%] vs 46.0 [40.3%]; P < .001), and lower New York Heart Association heart failure class (class I or II in 68.5% vs 56.7%; P = .02).

ten had prior myocardial infarction (93.7% vs 80.9%; P < .001), prior percutaneous coronary revascularization (34.6% vs 14.1%; P < .001), greater anterior akinesia or dyskinesia (57.5 [13.1%] vs 46.0 [40.3%]; P < .001), and lower New York Heart Association heart failure class (class I or II in 68.5% vs 56.7%; P = .02). Table 1. Baseline Characteristics of 2032 Patients With Left Ventricular Ejection Fraction Data From Core Laboratoriesa,b Characteristics No. (%) Patients With Echocardiographic EF (n = 1948) Patients With SPECT EF (n = 774) Patients With CMR EF (n = 417) Age, mean (SD), y 60.9 (9.5) 61.2 (9.4) 61.0 (9.6) Male 1687 (86.6) 667 (86.2) 359 (86.1) BMI, mean (SD) 27.4 (4.6) 27.6 (4.4) 27.4 (4.5) Hyperlipidemia 1271 (65.4) 515 (66.8) 300 (72.6) Hypertension 1165 (59.8) 443 (57.2) 241 (57.8) Current smoker 408 (21.0) 141 (18.2) 100 (24.0) Diabetes 712 (36.6) 284 (36.7) 151 (36.2) Peripheral vascular disease 292 (15.0) 122 (15.8) 51 (12.2) Chronic renal insufficiency 152 (7.8) 60 (7.8) 26 (6.2) Stroke 133 (6.8) 57 (7.4) 22 (5.3) Myocardial infarction 1595 (81.9) 642 (82.9) 364 (87.3) Previous CABG 56 (2.9) 18 (2.3) 10 (2.4) Previous PCI 301 (15.5) 160 (20.7) 100 (24.0) Atrial flutter or fibrillation 237 (12.2) 94 (12.1) 47 (11.3) Current NYHA heart failure class I 200 (10.3) 80 (10.3) 35 (8.4) II 915 (47.0) 429 (55.4) 201 (48.2) III 757 (38.9) 240 (31.0) 156 (37.4) IV 76 (3.9) 25 (3.2) 25 (6.0) Anterior akinesia or dyskinesia, mean (SD), % 47.0 (39.2) 46.3 (29.0) 56.0 (14.6) Study quality Excellent 71 (3.6) 250 (32.3) 334 (80.3) Good 791 (40.6) 380 (49.1) 66 (15.9) Fair 578 (29.7) 134 (17.3) 13 (3.1) Borderline 508 (26.1) 10 (1.3) 3 (0.7) Abbreviations: BMI, body mass index (calculated as weight in kilograms divided by height in meters squared); CABG, coronary artery bypass grafting; CMR, cardiovascular magnetic resonance; EF, ejection fraction; NYHA, New York Heart Association; PCI, percutaneous coronary intervention; SPECT, single-photon emission computed tomography.

MI, body mass index (calculated as weight in kilograms divided by height in meters squared); CABG, coronary artery bypass grafting; CMR, cardiovascular magnetic resonance; EF, ejection fraction; NYHA, New York Heart Association; PCI, percutaneous coronary intervention; SPECT, single-photon emission computed tomography. a Some patients have left ventricular ejection fraction from more than 1 core laboratory. b Continuous variables are presented as mean (SD) and categorical variables are presented as No. (%). Imaging The median time interval between echocardiography and SPECT was 3.0 days (interquartile range, 1.0-9.0 days); between echocardiography and CMR was 2.0 days (interquartile range, 1.0-6.0 days); and between SPECT and CMR was 1.0 days (interquartile range, 1.0-5.0 days). All patients qualified for participation in the study based on LVEF of 35% or less as assessed by the recruiting site. Thus, in some patients, an alternative modality of determination of LVEF yielded a result of 35% or greater. The mean (SD) for the LVEF by the core laboratories for each modality and for the various echocardiographic methods and the numbers and percentages with LVEF of 35% or less for each modality are shown in Table 2.

hus, in some patients, an alternative modality of determination of LVEF yielded a result of 35% or greater. The mean (SD) for the LVEF by the core laboratories for each modality and for the various echocardiographic methods and the numbers and percentages with LVEF of 35% or less for each modality are shown in Table 2. Table 2. Data on LVEF According to Echocardiographic Method and Imaging Modality Variables Patients With Data by Modality, No. (%) LVEF, Mean (SD), % Patients With LVEF ≤35%, No. (%) Echocardiographic EF 1948 (95.9) 29.0 (8.2) 1555 (79.8) Echocardiographic biplane EF 897 (44.1) 28.7 (8.2) 709 (79.0) Echocardiographic single-plane EF 725 (35.7) 29.2 (8.6) 552 (76.1) Echocardiographic visual EF 1941 (95.5) 28.5 (7.8) 1679 (86.5) SPECT EF 774 (38.1) 26.8 (8.3) 648 (83.7) CMR EF 417 (20.5) 27.2 (10.8) 328 (78.7) Abbreviations: CMR, cardiovascular magnetic resonance; EF, ejection fraction; LVEF, left ventricular ejection fraction; SPECT, single-photon emission computed tomography.

.1) Echocardiographic visual EF 1941 (95.5) 28.5 (7.8) 1679 (86.5) SPECT EF 774 (38.1) 26.8 (8.3) 648 (83.7) CMR EF 417 (20.5) 27.2 (10.8) 328 (78.7) Abbreviations: CMR, cardiovascular magnetic resonance; EF, ejection fraction; LVEF, left ventricular ejection fraction; SPECT, single-photon emission computed tomography. Comparison of Imaging Modalities Among 1948 patients who had LVEF assessed by echocardiography, 1437 (73.8%) of them had both visual estimate and quantitative LVEF. The mean absolute differences between LVEF as determined by quantitative vs visual echocardiographic methods were all less than 5% (mean absolute difference 2.7% for biplane and visual, 3.0% for single plane and visual, and 2.9% for biplane and single plane). The mean absolute differences of LVEF by echocardiography and by other modalities were all greater than 5% but were lowest when biplane echocardiography, rather than other echocardiographic methods, was used (data not shown). Mean absolute difference of LVEF between modalities was 5.9% for biplane echocardiography and SPECT (n = 385), 7.3% for CMR and biplane echocardiography (n = 204), and 5.9% for CMR and SPECT (n = 134). The variability measurements were similar in women and men (data not shown). When only data for studies performed within 3 days were considered, results were similar: mean absolute difference 5.8% for biplane echocardiography and SPECT (n = 213), 7.2% for CMR and biplane echocardiography (n = 126), and 5.9% for CMR and SPECT (n = 84). Even when only data for studies performed the same day were considered, results were similar: mean absolute difference 5.2% for biplane echocardiography and SPECT (n = 68 ), 8.6% for CMR and biplane echocardiography (n = 33), and 4.1% for CMR and SPECT (n = 10). When only data for studies rated as having excellent or good quality were included, mean absolute differences were again similar: 5.6% for biplane echocardiography and SPECT (n = 317), 7.3% for CMR and biplane echocardiography (n = 184), and 5.7% for CMR and SPECT (n = 116). For the 127 patients in whom LVEF was measured by all 3 modalities, mean absolute difference between modalities was 6.0% for echocardiography and SPECT, 6.8% for CMR and echocardiography, and 5.9% for CMR and SPECT.

CT (n = 317), 7.3% for CMR and biplane echocardiography (n = 184), and 5.7% for CMR and SPECT (n = 116). For the 127 patients in whom LVEF was measured by all 3 modalities, mean absolute difference between modalities was 6.0% for echocardiography and SPECT, 6.8% for CMR and echocardiography, and 5.9% for CMR and SPECT. Bland-Altman comparisons between modalities for determining LVEF are shown in the Figure. Limits of agreement between modalities were broad, ranging from 28.27% to 35.31%. Bland-Altman comparison with SPECT repeated after exclusion of LVEF assessed following stress with SPECT were similar (eFigure in the Supplement). The mean signed difference between LVEF measured by biplane and single-plane echocardiography was closest to 0, indicating no substantial overestimation or underestimation of LVEF by either echocardiographic method. The mean signed differences for LVEF between modalities were larger with wider 95% confidence intervals. Figure. Bland-Altman Plots for Left Ventricular Ejection Fraction (EF) Plots are compared for biplane Simpson method and visual estimation for echocardiography (A), biplane Simpson method by echocardiography and gated single-photon emission computed tomography (SPECT) (B), biplane Simpson method by echocardiography and cardiovascular magnetic resonance (CMR) (C), and gated SPECT and CMR (D).

ts are compared for biplane Simpson method and visual estimation for echocardiography (A), biplane Simpson method by echocardiography and gated single-photon emission computed tomography (SPECT) (B), biplane Simpson method by echocardiography and cardiovascular magnetic resonance (CMR) (C), and gated SPECT and CMR (D). Correlations between LVEF as determined by quantitative vs visual echocardiographic methods (r = 0.898 for biplane vs visual, r = 0.876 for single plane vs visual, r = 0.874 for biplane and single plane) were higher than the correlations of LVEF assessed between different modalities (r = 0.601 for biplane echocardiography and SPECT, r = 0.660 for CMR and SPECT, and r = 0.493 for CMR and biplane echocardiography). Left ventricular ejection fraction measurements were within 5% in 54.0% for biplane echocardiography and SPECT, 48.5% for SPECT and CMR, and 43.1% for biplane echocardiography and CMR. Using CMR as the standard and comparing it with SPECT and echocardiography as to whether there was intermodality agreement for LVEF greater than 35% is shown in Table 3. The results of the 4 variability indices comparing LVEF and end-systolic volume index for the 3 modalities are summarized in Table 4.

ography and CMR. Using CMR as the standard and comparing it with SPECT and echocardiography as to whether there was intermodality agreement for LVEF greater than 35% is shown in Table 3. The results of the 4 variability indices comparing LVEF and end-systolic volume index for the 3 modalities are summarized in Table 4. Table 3. Agreement and Disagreement for LVEF 35% or Greater According to Echocardiographic Method and Imaging Modality Using LVEF by CMR as the Standard Comparison LVEF No. of Patients With Both LVEFs No. (%) of Patients Both EF ≤35% Both EF >35% 2 EFs Agreed CMR EF ≤35% and Comparison EF >35% CMR EF >35% and Comparison EF ≤35% 2 EFs Disagreed Echocardiographic EF 377 243 (64.5) 37 (9.8) 280 (74.3) 54 (14.3) 43 (11.4) 97 (25.7) Echocardiographic biplane EF 204 134 (65.7) 18 (8.8) 152 (74.5) 34 (16.7) 18 (8.8) 52 (25.5) Echocardiographic single-plane EF 130 80 (61.5) 17 (13.1) 97 (74.6) 18 (13.9) 15 (11.5) 33 (25.4) Echocardiographic visual EF 375 271 (72.3) 25 (6.7) 296 (78.9) 24 (6.4) 55 (14.7) 79 (21.1) SPECT EF 134 90 (67.1) 19 (14.2) 109 (81.3) 10 (7.5) 15 (11.2) 25 (18.7) Abbreviations: CMR, cardiovascular magnetic resonance; EF, ejection fraction; LVEF, left ventricular ejection fraction; SPECT, single-photon emission computed tomography.

ic visual EF 375 271 (72.3) 25 (6.7) 296 (78.9) 24 (6.4) 55 (14.7) 79 (21.1) SPECT EF 134 90 (67.1) 19 (14.2) 109 (81.3) 10 (7.5) 15 (11.2) 25 (18.7) Abbreviations: CMR, cardiovascular magnetic resonance; EF, ejection fraction; LVEF, left ventricular ejection fraction; SPECT, single-photon emission computed tomography. Table 4. Variability Indices for LVEF and ESVI Measures Between Modalities Pairwise Comparison of Core Laboratory LVEF or ESVI Measures Mean Signed Difference Mean Absolute Difference Correlation Coefficient Bland-Altman Limit of Agreement Rangea Coverage Probabilityb Echocardiographic biplane EF vs SPECT EF, % (n = 385) 2.2 5.9 0.601 28.27 0.540 SPECT EF vs CMR EF, % (n = 134) 0.8 5.9 0.660 30.36 0.485 Echocardiographic biplane EF vs CMR EF, % (n = 204) 2.5 7.3 0.493 35.31 0.431 Echocardiographic biplane ESVI vs SPECT ESVI, mL/m2 (n = 386) −12.8 20.8 0.821 97.92 0.332 SPECT ESVI vs CMR ESVI, mL/m2 (n = 134) 11.0 17.9 0.821 83.04 0.358 Echocardiographic biplane ESVI vs CMR ESVI, mL/m2 (n = 204) −3.8 18.8 0.786 26.67 0.353 Abbreviations: CMR, cardiovascular magnetic resonance; EF, ejection fraction; ESVI, end-systolic volume index; LVEF, left ventricular ejection fraction; SPECT, single-photon emission computed tomography. a Bland-Altman limit of agreement range is between 1.96 standard deviation of the differences between the 2 imaging modalities. Approximately 95% of the differences are expected to fall within this range.

Table 4. Variability Indices for LVEF and ESVI Measures Between Modalities Pairwise Comparison of Core Laboratory LVEF or ESVI Measures Mean Signed Difference Mean Absolute Difference Correlation Coefficient Bland-Altman Limit of Agreement Rangea Coverage Probabilityb Echocardiographic biplane EF vs SPECT EF, % (n = 385) 2.2 5.9 0.601 28.27 0.540 SPECT EF vs CMR EF, % (n = 134) 0.8 5.9 0.660 30.36 0.485 Echocardiographic biplane EF vs CMR EF, % (n = 204) 2.5 7.3 0.493 35.31 0.431 Echocardiographic biplane ESVI vs SPECT ESVI, mL/m2 (n = 386) −12.8 20.8 0.821 97.92 0.332 SPECT ESVI vs CMR ESVI, mL/m2 (n = 134) 11.0 17.9 0.821 83.04 0.358 Echocardiographic biplane ESVI vs CMR ESVI, mL/m2 (n = 204) −3.8 18.8 0.786 26.67 0.353 Abbreviations: CMR, cardiovascular magnetic resonance; EF, ejection fraction; ESVI, end-systolic volume index; LVEF, left ventricular ejection fraction; SPECT, single-photon emission computed tomography. a Bland-Altman limit of agreement range is between 1.96 standard deviation of the differences between the 2 imaging modalities. Approximately 95% of the differences are expected to fall within this range. b Coverage probability is the proportion of participants who fall within the prespecified acceptable paired absolute difference. For LVEF, the prespecified acceptable paired absolute difference was 5% or less. For ESVI, the prespecified acceptable paired absolute difference was 10 mL/m2.

a Bland-Altman limit of agreement range is between 1.96 standard deviation of the differences between the 2 imaging modalities. Approximately 95% of the differences are expected to fall within this range. b Coverage probability is the proportion of participants who fall within the prespecified acceptable paired absolute difference. For LVEF, the prespecified acceptable paired absolute difference was 5% or less. For ESVI, the prespecified acceptable paired absolute difference was 10 mL/m2. Prognostic Impact of LVEF According to Modality Among the 2032 patients who had LVEF assessment, follow-up was available in all, although it was abbreviated in 32 cases (1.6%). The prognostic effect of LVEF by modality and outcome was assessed. During a mean (SD) follow-up of 5.0 (3.3) years, there were 966 deaths. Among 1948 patients who had echocardiographic LVEF assessment, the prognostic effect of LVEF by method (biplane, single plane, or visual estimation) was examined. Left ventricular ejection fractions measured by different methods were all statistically significant in terms of their prognostic value with respect to all-cause mortality. The HR (0.83-0.89) and the associated 95% confidence intervals for every 5% LVEF increment were all similar and below 1.00. This indicates increased LVEF was associated with decreased mortality risk, regardless which modality for determining LVEF was used (eTable in the Supplement).

with respect to all-cause mortality. The HR (0.83-0.89) and the associated 95% confidence intervals for every 5% LVEF increment were all similar and below 1.00. This indicates increased LVEF was associated with decreased mortality risk, regardless which modality for determining LVEF was used (eTable in the Supplement). Discussion Left ventricular ejection fraction refers to the fraction of LV end-diastolic volume ejected during systole. It is the most widely used measure of assessment for LV systolic function and is familiar to patients and clinicians. This is the first study to compare echocardiographic, CMR, and SPECT methods for determination of LVEF in a large, international, multicenter cohort of patients included in a clinical trial with extensive follow-up. In this population with LV dysfunction and CAD, we found that there was substantial variation among modalities for determination of LVEF even though these measures were made by specialized core laboratories, each of which followed specific plans for image analysis and measurements.19,26 Acquisition of data at 127 sites worldwide may have contributed to this variability. Moreover, variation was not predictable; there was no substantial overestimation or underestimation of LVEF by any modality relative to another.

oratories, each of which followed specific plans for image analysis and measurements.19,26 Acquisition of data at 127 sites worldwide may have contributed to this variability. Moreover, variation was not predictable; there was no substantial overestimation or underestimation of LVEF by any modality relative to another. Surprisingly, few studies have assessed the differences in LVEF as measured by different imaging modalities.12,13,14,27 Previous studies were predominantly single center, most with fewer than 100 participants, and most recent studies have focused on 3-dimensional or contrast echocardiography. These newer techniques were associated with improved reproducibility and agreement with CMR.28,29

ured by different imaging modalities.12,13,14,27 Previous studies were predominantly single center, most with fewer than 100 participants, and most recent studies have focused on 3-dimensional or contrast echocardiography. These newer techniques were associated with improved reproducibility and agreement with CMR.28,29 Correlations between various methods of determination of LVEF by a single modality, echocardiography, were similar and better (r = 0.898 for biplane and visual estimation; r = 0.874 for single plane and biplane; and r = 0.876 for single plane and visual) than were correlations between different modalities, which ranged from r = 0.493 (for biplane echocardiography and CMR) to r = 0.660 (for CMR and SPECT). The good correlation between visual estimation and measurement of echocardiographic images is of interest; however, it should be noted that visual estimates were made by echocardiographic core laboratory staff, who had advanced training and extensive experience. Results might be worse with less experienced reviewers. Although the 3 echocardiographic methods assessed were well correlated, LVEF by biplane method of disks should be used when feasible as it correlated best with LVEF by other modalities.

hic core laboratory staff, who had advanced training and extensive experience. Results might be worse with less experienced reviewers. Although the 3 echocardiographic methods assessed were well correlated, LVEF by biplane method of disks should be used when feasible as it correlated best with LVEF by other modalities. Bland-Altman analysis showed no substantial overestimation or underestimation of LVEF by different modalities. Biplane quantitation with echocardiography averaged only 2.5% higher than CMR and 2.2% higher than gated SPECT, and SPECT was 0.8% higher than CMR. Limits of agreement were broad. Variability between modalities for measures of LV end-systolic volume index were also broad. Variation of LVEF within 5% between modalities might be considered clinically acceptable. However, the percentage of observations that fell within a range of 5% ranged from 43% to 54% between different imaging modalities. Discordance between modalities as to whether LVEF was greater than 35% was present in about 20% to 25% of cases. Had core laboratory determination of LVEF been used for enrollment in the STICH trial, which required LVEF of 35% or less, many of these patients would have been ineligible.

een different imaging modalities. Discordance between modalities as to whether LVEF was greater than 35% was present in about 20% to 25% of cases. Had core laboratory determination of LVEF been used for enrollment in the STICH trial, which required LVEF of 35% or less, many of these patients would have been ineligible. The decision to perform SPECT or CMR was made by the clinical site. Patients with greater akinesia or dyskinesia were more likely to undergo CMR. Geometric abnormalities of the left ventricle may have contributed to the discordance between modalities for determining LVEF. In particular, echocardiographic biplane and single-plane methods rely on geometric assumptions, whereas SPECT and CMR methods assess the LV tomographically, without these assumptions. This may have contributed to the poor concordance of echocardiography and the other methods. Had patients with worse echocardiographic images been more likely to be referred for additional imaging, this might have accounted for a poorer correlation between echocardiography and other modalities. However, patients with better echocardiographic image quality more often underwent performance of additional imaging with SPECT or CMR.

worse echocardiographic images been more likely to be referred for additional imaging, this might have accounted for a poorer correlation between echocardiography and other modalities. However, patients with better echocardiographic image quality more often underwent performance of additional imaging with SPECT or CMR. The prognostic effect of LVEF by different echocardiographic methods and between modalities was also assessed. Although LVEF measurements from different methods and modalities have significant variations, analyses in this study indicate LVEF was strongly prognostic of all-cause mortality in the univariable Cox regression models no matter which method or modality was used. Because LVEF was determined by different modalities for different patients, direct comparison of the prognostic effect for LVEF by various modalities was not possible.

study indicate LVEF was strongly prognostic of all-cause mortality in the univariable Cox regression models no matter which method or modality was used. Because LVEF was determined by different modalities for different patients, direct comparison of the prognostic effect for LVEF by various modalities was not possible. Limitations No gold standard exists for determining LVEF; agreement between modalities was assessed by various statistical methods. All patients had reduced LVEF. Correlations may have been better had the population included patients with a full spectrum of values. A broader spectrum of values of LVEF would be expected to be encountered in most clinical situations. Despite recommendations that LVEF be determined using biplane Simpson method, this was possible in only 897 of patients with echocardiography (46%). Ultrasonography contrast agents, recommended for use when imaging is suboptimal,30 have been shown to improve correlation with CMR but were used in only 3 patients.14,31 Three-dimensional echocardiographic imaging is increasingly being used for determination of LVEF and may improve the agreement between echocardiography and other methods, but it is not available on all ultrasonography systems and was not used in this study.32 Similarly, technology for SPECT and CMR also continues to be refined; such improvements may reduce differences between modalities. The SPECT LVEF measurements obtained from poststress studies may have been lower than rest studies in patients with extensive ischemia. However, even after exclusion of data in which SPECT LVEF was assessed following stress, results were similar with wide limits of agreement. Only a subset of patients was referred for SPECT and CMR; decisions about referral were made at local sites. Only 57% of patients who underwent echocardiography had LVEF determined by a second modality. Only 38% of patients included in the analysis had LVEF determined by SPECT, and only 20.5% of patients included in the analysis had LVEF determined by CMR. Since not all patients included in the STICH trial had LVEF determined by 2 different modalities, there may have been selection bias in choosing patients who had LVEF determined by a second imaging modality.

is had LVEF determined by SPECT, and only 20.5% of patients included in the analysis had LVEF determined by CMR. Since not all patients included in the STICH trial had LVEF determined by 2 different modalities, there may have been selection bias in choosing patients who had LVEF determined by a second imaging modality. Assessments by different modalities were not simultaneous, and intervening medical therapy or ischemia may have affected LVEF. Left ventricular ejection fraction can be affected by changes in preload, afterload, and LV remodeling.33 Data regarding changes in pharmacologic therapy or blood pressure between tests were not available. The median time interval between studies was 3 days between echocardiography and SPECT studies, 2 days between echocardiography and CMR studies, and 1 day between SPECT and CMR studies. Correlations were not consistently improved when tests were performed within 3 days or on the same day. Reproducibility of LVEF determination may be affected by image quality,34 but correlations were not improved when only images of good or excellent quality were included.

dies, and 1 day between SPECT and CMR studies. Correlations were not consistently improved when tests were performed within 3 days or on the same day. Reproducibility of LVEF determination may be affected by image quality,34 but correlations were not improved when only images of good or excellent quality were included. Conclusions Although LVEF is a widely reported measurement and is the cornerstone of many treatment decisions, there is substantial variability in its measurement using different modalities, even when assessed by core laboratories. In this international study in which LVEF imaging was performed at 127 clinical sites using up to 3 widely used imaging modalities and LVEF was independently measured by core laboratories according to standard protocols, the variability in LVEF measurement exceeded 5% in about half of the patients. Variability was less for different methods of determining LVEF when a single imaging modality (echocardiography, in this case) was used. Longitudinal assessments of a given patient may best be accomplished using a single imaging modality. The diagnostic and prognostic importance of LVEF as well documented in numerous studies is not disputed. Left ventricular ejection fraction by each modality was associated with mortality. However, variability in LVEF assessment by different imaging modalities should be considered in trial design and clinical management. Considering this variability, cut points in LVEF should not be the sole basis for decision making.

ted. Left ventricular ejection fraction by each modality was associated with mortality. However, variability in LVEF assessment by different imaging modalities should be considered in trial design and clinical management. Considering this variability, cut points in LVEF should not be the sole basis for decision making. Supplement. eFigure. Bland-Altman Plots for LVEF After Exclusion of SPECT LVEF Data When Obtained After Stress eTable. Prognostic Effect of LVEF Measures in Cox Regression Model Click here for additional data file.

Introduction Delirium is a short-term change in attention and awareness that typically affects hospitalized, high-risk older adults.1 Many older adults who present to the emergency department and are subsequently admitted to the hospital have prevalent delirium. Han et al2 found that up to 15% of patients present to the emergency department with delirium, and delirium at acute care admission for older adults has a reported prevalence of 18% to 39%.3,4,5 This prevalence increases to 57% among older adults admitted with a diagnosis of preexisting dementia.6 Despite the high prevalence, delirium is poorly recognized in the acute care setting. Early research in delirium found that the diagnosis was missed in up to two-thirds of older adults,7 and despite advances in delirium science, delirium continues to go unrecognized at presentation to the emergency department or hospital admission.8,9,10 Although prevalent delirium cannot be prevented, failure to recognize delirium is associated with increased length of stay, health complications, discharge to a skilled nursing facility,4 and increased risk of death.11,12,13 In addition, delirium is associated with increased distress for individuals experiencing delirium and their families, as well as health care professionals.14,15

nize delirium is associated with increased length of stay, health complications, discharge to a skilled nursing facility,4 and increased risk of death.11,12,13 In addition, delirium is associated with increased distress for individuals experiencing delirium and their families, as well as health care professionals.14,15 Delirium prediction algorithms have been used to stratify those at highest risk for delirium so that increased resources and efforts can be allocated to those in greatest need.16,17,18 Identification of those at highest risk for delirium allows for improved clinical efficiency and more timely recognition and diagnosis of delirium. Older adults admitted to the hospital with delirium need diagnosis, identification of underlying risk factors and potential causes, and appropriate management of and intervention for delirium.19 The National Institute for Health and Clinical Excellence (NICE) in the United Kingdom performed a comprehensive systematic review and meta-analysis of delirium risk factors as part of a delirium clinical practice guideline.20,21 The factors identified in the NICE meta-analysis have been prospectively confirmed for delirium that develops after admission.16,17

nical Excellence (NICE) in the United Kingdom performed a comprehensive systematic review and meta-analysis of delirium risk factors as part of a delirium clinical practice guideline.20,21 The factors identified in the NICE meta-analysis have been prospectively confirmed for delirium that develops after admission.16,17 The primary purposes of this study were to use the framework of the NICE meta-analysis and delirium clinical practice guidelines21 to compare the performance of 3 prediction rules for prevalent delirium and to consolidate the prediction rule components to the minimum information necessary to maximize predictive ability. We hypothesized that a consolidated prediction rule for delirium at admission would perform better than existing NICE-based delirium prediction rules. Having a singular, consolidated delirium prediction rule could target clinical efforts of screening toward those who need immediate evaluation for and diagnosis of delirium.

ypothesized that a consolidated prediction rule for delirium at admission would perform better than existing NICE-based delirium prediction rules. Having a singular, consolidated delirium prediction rule could target clinical efforts of screening toward those who need immediate evaluation for and diagnosis of delirium. Methods Sample The sample for this analysis was drawn from the Veteran Affairs (VA) External Peer Review Program (EPRP) at 118 VA medical centers with inpatient facilities.22 Medical records of patients admitted for congestive heart failure, acute coronary syndrome, community-acquired pneumonia, and chronic obstructive pulmonary disease were randomly selected for electronic medical record (EMR) review by trained nurses for the presence of delirium and its risk factors. Interrater reliability assessments were built into the data collection process. From October 1, 2012, to September 30, 2013, a total of 27 625 VA hospital admissions were abstracted for patients 65 years or older; this group composed the derivation cohort. The confirmation cohort consisted of 11 752 patients from the EPRP sample from October 1, 2013, until March 31, 2014. This analysis follows the recommendations of the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) reporting guideline (eAppendix in the Supplement).23 Data were not identifiable; however, because of the linear age variable, some veterans older than 90 years were identifiable. The VA Providence Institutional Review Board approved this analysis. This was a quality improvement project, and a waiver of informed consent from study participants was granted by the VA Providence Institutional Review Board and from the Health Insurance Portability and Accountability Act of 1996.

s were identifiable. The VA Providence Institutional Review Board approved this analysis. This was a quality improvement project, and a waiver of informed consent from study participants was granted by the VA Providence Institutional Review Board and from the Health Insurance Portability and Accountability Act of 1996. NICE Delirium Prediction Rule A systematic review and meta-analysis of delirium risk factors conducted by the NICE delirium clinical practice guideline identified 6 risk factors for delirium: (1) age with cutoff points at 65 and 80 years, (2) cognitive impairment, (3) illness severity, (4) infection, (5) fracture, and (6) visual impairment. These factors have been independently confirmed for incident and any delirium.16

by the NICE delirium clinical practice guideline identified 6 risk factors for delirium: (1) age with cutoff points at 65 and 80 years, (2) cognitive impairment, (3) illness severity, (4) infection, (5) fracture, and (6) visual impairment. These factors have been independently confirmed for incident and any delirium.16 Electronic NICE Delirium Prediction Rule The electronic NICE (eNICE) delirium prediction rule used criteria defined by the initial meta-analysis presented by the NICE delirium clinical practice guideline. All criteria were pulled from the EMR to confirm retrospective and prospective cases of delirium.16 Patients were admitted for general medical and surgical reasons, such as cardiac issues and infection. Age points were defined as 65 years and older and 80 years and older. Cognitive impairment was defined as EMR diagnosis or medication used to treat dementia at admission. Severity of illness was based on laboratory and vital sign data. Admitting diagnoses were screened to assess for infection and/or fracture. Visual impairment was determined by the EMR data. Delirium was assessed daily by a trained physician using Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition, Text Revision criteria and was defined as any delirium at admission or during hospitalization.16

ened to assess for infection and/or fracture. Visual impairment was determined by the EMR data. Delirium was assessed daily by a trained physician using Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition, Text Revision criteria and was defined as any delirium at admission or during hospitalization.16 Pendlebury NICE Delirium Prediction Rule Pendlebury modified the NICE factors to confirm a rule for patients admitted to acute care hospitals.17 This was a prospective report, and delirium was screened for at admission and daily during hospitalization according to Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition criteria by a physician. Criteria included in the prediction rule included cognitive impairment, age of 80 years and older, infection, visual impairment, and systemic inflammatory response syndrome (SIRS).17 Patients were admitted to the same medical team. Cognitive impairment was assessed using an abbreviated mental test score or Mini-Mental State Examination score and/or dementia diagnosis in the EMR. Visual impairment was noted in the EMR or if signs were evident during the patient visit. The Pendlebury NICE delirium prediction rule used SIRS criteria of respiratory rate, pulse, and white blood cell count as the criteria for acute illness,17 and these were taken from the EMR at admission.

ntia diagnosis in the EMR. Visual impairment was noted in the EMR or if signs were evident during the patient visit. The Pendlebury NICE delirium prediction rule used SIRS criteria of respiratory rate, pulse, and white blood cell count as the criteria for acute illness,17 and these were taken from the EMR at admission. Proposed Consolidated NICE Delirium Prediction Rule For the proposed consolidated NICE delirium prediction rule, age was abstracted from the medical record as the age at hospital admission and categorized as 65 years and older and 80 years and older. Cognitive impairment was defined as prior diagnosis of dementia in the EMR or outpatient prescription of a medication for dementia at admission (eg, donepezil). Severity of illness was calculated using an acute physiologic score from laboratory (ie, sodium level, bilirubin level, creatinine concentration, hematocrit, albumin level, blood urea nitrogen level, glucose level, and white blood cell count) and vital sign (pulse, respiratory rate, and blood pressure) data collected from the EMR with cutoffs similar to those of prior severity rules.24 Laboratory data used for the analysis were those most proximal to the admission to allow for the capture of laboratory values that were obtained in the emergency department. Infection and fracture were considered to be present if they were listed in the admission diagnoses. Infection data were abstracted by trained nurse reviewers and included the top 10 infections at the VA (ie, pneumonia, influenza, urinary tract infection, septicemia or sepsis, cellulitis, diverticulitis, peritonitis, appendicitis, osteomyelitis, and meningitis). Fracture was limited to the short-term presence of a femoral, vertebral, humeral, tibial, fibular, radial, or ulnar fracture. Visual impairment was based on review of the problem list and nursing admission notes for evidence of prior visual deficit diagnosis or for inability to correct vision during acute care admission. Multiple imputation was used for missing data (eTable 1 in the Supplement).

al, tibial, fibular, radial, or ulnar fracture. Visual impairment was based on review of the problem list and nursing admission notes for evidence of prior visual deficit diagnosis or for inability to correct vision during acute care admission. Multiple imputation was used for missing data (eTable 1 in the Supplement). Outcomes The primary outcome for the consolidated NICE prediction rule was prevalent delirium, which was defined as the presence of 1 or more of the following terms or symptoms of delirium in the EMR within 24 hours of admission: (1) delirium, (2) change in mental status, (3) disoriented, (4) confused, (5) unarousable, (6) lethargic, and (7) obtunded. Interrater reliability, performed routinely within the EPRP, found 92% agreement among reviewers for prevalent delirium.16 Statistical Analysis The derivation and confirmation cohorts were compared using standardized differences. From the derivation cohort, we developed the eNICE and Pendlebury NICE prediction rules using the criteria outlined earlier from the initial studies.16,17

Outcomes The primary outcome for the consolidated NICE prediction rule was prevalent delirium, which was defined as the presence of 1 or more of the following terms or symptoms of delirium in the EMR within 24 hours of admission: (1) delirium, (2) change in mental status, (3) disoriented, (4) confused, (5) unarousable, (6) lethargic, and (7) obtunded. Interrater reliability, performed routinely within the EPRP, found 92% agreement among reviewers for prevalent delirium.16 Statistical Analysis The derivation and confirmation cohorts were compared using standardized differences. From the derivation cohort, we developed the eNICE and Pendlebury NICE prediction rules using the criteria outlined earlier from the initial studies.16,17 Random Forest Modeling The random forest algorithm was used to create the consolidated NICE score. Random forests are a classification tool that automatically constructs and classifies multiple decision trees and uses ensemble learning superimposed on regression to select independent variables. Random forest models reduce the model overfitting common with standard regression modeling25 by bootstrapping the decision trees consisting of the NICE predictive factors. The automated random forest algorithm provided an importance measure: mean decrease in accuracy (percentage increase in mean squared error; higher is better). On the basis of this measure, variable selection focused on net accuracy. More in-depth definitions of these importance measures can be found elsewhere.25,26,27 For our consolidated NICE model, we included age as a continuous variable to create an age cutoff using a separate random forest analysis. The same methods were applied to all other continuous variables, including laboratory measures, to find cutoffs. Because of the large size of our sample, we chose to eliminate the use of more complicated laboratory and vital scores (Acute Physiologic Assessment and Chronic Health Evaluation and SIRS criteria) and included each of the factors in the APACHE and SIRS criteria as individual features in our random forest model to determine the most important features when identifying prevalent delirium. All laboratory values were continuous in the data set to allow the random forest mechanism to capture the maximal possible information, after which we created an additive clinical diagnostic score based on the remaining important features. The feature weights were determined using random forest importance level. Clinical cutoff scores were created after the important features were selected using the modeling techniques. After producing a receiver operating characteristic (ROC) curve, we determined low-, intermediate-, and high-risk cut points for the consolidated NICE score, which were created for presentation.

portance level. Clinical cutoff scores were created after the important features were selected using the modeling techniques. After producing a receiver operating characteristic (ROC) curve, we determined low-, intermediate-, and high-risk cut points for the consolidated NICE score, which were created for presentation. We examined discriminatory performance of the 3 delirium prediction rules for delirium at admission in the derivation and confirmation cohorts using area under the ROC (AUROC) curve (C statistic) to test for model consistency and equality. Histograms and box plots were used to give a visual representation of the association between NICE score and delirium status in the derivation cohort. A density histogram provides a probability breakdown of each NICE score by delirium status. Comparison between the rules used a χ2 test. For comparisons, we set statistical significance at P < .05. Stata statistical software, version 14.2 (StataCorp) was used for data manipulation, eNICE score generation, and table creation. R, version 3.3.2 (R Foundation for Statistical Computing) was used to construct ROC curves. The randomForest program within R, version 3.3.2 was used to create the random forest algorithms.27

Stata statistical software, version 14.2 (StataCorp) was used for data manipulation, eNICE score generation, and table creation. R, version 3.3.2 (R Foundation for Statistical Computing) was used to construct ROC curves. The randomForest program within R, version 3.3.2 was used to create the random forest algorithms.27 Results A total of 27 625 patients were included in the derivation cohort (28 118 [92.2%] male; mean [SD] age, 75.95 [8.61] years) and 11 752 in the confirmation cohort (11 536 [98.2%] male; mean [SD] age, 75.43 [8.55] years). Delirium at admission was identified in 2343 patients (8.5%) in the derivation cohort and 882 patients (7.0%) in the confirmation cohort. The derivation and confirmation cohorts are compared in Table 1. Although statistically significant differences were found between the derivation and confirmation cohorts, the actual differences were small and not clinically significant. For example, the analysis identified a difference between hematocrit in the derivation vs confirmation cohort (35.3% vs 34.9% [to convert to a proportion of 1.0, multiply by 0.01]; P < .001), which is statistically but not clinically significant. Means and percentages of vital signs, medications, laboratory values, and NICE scores across both cohorts were clinically similar but often statistically different.

onfirmation cohort (35.3% vs 34.9% [to convert to a proportion of 1.0, multiply by 0.01]; P < .001), which is statistically but not clinically significant. Means and percentages of vital signs, medications, laboratory values, and NICE scores across both cohorts were clinically similar but often statistically different. Table 1. Characteristics of the Derivation and Confirmation Cohortsa Characteristic Derivation Cohort (n = 27 625) Confirmation Cohort (n = 11 752) Standardized Difference Age, mean (SD), y 75.95 (8.61) 75.43 (8.55) −0.007 Age ≥80 y 9953 (36.0) 3985 (33.9) −0.044 Male 27 118 (98.2) 11 536 (98.2) 0 Fracture 351 (1.3) 142 (1.2) −0.006 Infection 9180 (33.2) 3699 (31.5) −0.038 Dementia diagnosis 3356 (12.1) 1290 (11.0) −0.037 Dementia medications 950 (3.4) 338 (2.9) −0.032 Visual impairment 12 667 (45.9) 5051 (43.0) −0.0579 Pulse, mean (SD), beats/min 80.94 (18.12) 80.12 (16.64) −0.003 Respiratory rate, mean (SD), breaths/min 19.35 (3.51) 19.17 (3.21) −0.016 Systolic blood pressure, mean (SD), mm Hg 134.03 (23.97) 134.98 (22.81) 0.002 Diastolic blood pressure, mean (SD), mm Hg 73.74 (13.27) 74.47 (12.99) 0.004 Arterial pressure, mean (SD), mg/dL 93.61 (14.80) 94.41 (14.39) 0.004 Creatinine level, mean (SD), mg/dL 1.42 (1.24) 1.43 (1.51) 0.005 Blood urea nitrogen level, mean (SD), mg/dL 25.88 (16.53) 25.69 (16.56) −0.001 Glucose level, mean (SD), g/dL 136.28 (58.79) 135.00 (56.41) −0.022 Albumin level, mean (SD), mg/dL 3.31 (0.60) 3.29 (0.59) −0.057 Bilirubin level, mean (SD), mg/dL 0.87 (0.97) 0.84 (0.62) −0.043 White blood cell count, mean (SD), /μL 9690 (5700) 9650 (5510) −0.001 Hematocrit, mean (SD), % 35.27 (5.91) 34.90 (5.94) −0.011 eNICE score, mean (SD) 5.11 (2.79) 4.90 (2.71) −0.028 Pendlebury NICE score, mean (SD) 2.24 (1.73) 2.08 (1.67) −0.056 Consolidated NICE score, mean (SD) 1.45 (1.59) 1.33 (1.51) −0.031 Delirium at admission 2343 (8.5) 822 (7.0) −0.056 Length of stay, mean (SD), d 5.70 (6.69) 5.74 (6.97) 0.001 Abbreviations: eNICE, electronic National Institute for Health and Care Excellence; NA, not applicable; NICE, National Institute for Health and Care Excellence.

CE score, mean (SD) 1.45 (1.59) 1.33 (1.51) −0.031 Delirium at admission 2343 (8.5) 822 (7.0) −0.056 Length of stay, mean (SD), d 5.70 (6.69) 5.74 (6.97) 0.001 Abbreviations: eNICE, electronic National Institute for Health and Care Excellence; NA, not applicable; NICE, National Institute for Health and Care Excellence. SI conversion factors: to convert albumin to grams per liter, multiply by 10; bilirubin to micromoles per liter, multiply by 17.104; creatinine to micromoles per liter, multiply by 88.4; glucose to millimoles per liter, multiply by 0.0555; hematocrit to a proportion of 1, multiply by 0.01; urea nitrogen to millimoles per liter, multiply by 0.357; and white blood cell count to ×109/L, multiply by 0.001. a Data are presented as number (percentage) of patients unless otherwise indicated.

SI conversion factors: to convert albumin to grams per liter, multiply by 10; bilirubin to micromoles per liter, multiply by 17.104; creatinine to micromoles per liter, multiply by 88.4; glucose to millimoles per liter, multiply by 0.0555; hematocrit to a proportion of 1, multiply by 0.01; urea nitrogen to millimoles per liter, multiply by 0.357; and white blood cell count to ×109/L, multiply by 0.001. a Data are presented as number (percentage) of patients unless otherwise indicated. In terms of predictive power and accuracy, cognitive impairment was the most important factor, followed by infection, sodium level, and age. The mean decrease in accuracies for the NICE features included in our random forest model is depicted in Figure 1. A cut point for age was determined using the median age node cutoff (≥80 years of age). A cut point for sodium level was determined using the median node cutoff and knowledge of normal sodium levels in VA patients. When we examined the breakdown of the sodium level cut point of 137 mEq/L (to convert to millimoles per liter, multiply by 1) (the median cut point determined by the random forest algorithm), no difference was found in the incidence of delirium at admission. Because the random forest model cannot pick a double cutoff, it picked within the reference range as the singular cutoff. At the VA, the reference range for sodium level in veterans is 135 to 145 mEq/L. Using our knowledge of VA patients, we decided on a double cutoff of less than 135 mEq/L or greater than 145 mEq/L as 1 point in our consolidated NICE score. eTable 1 in the Supplement gives the power of our 4 chosen predictive factors. There was an increase in the proportion of patients with delirium vs those without (eTable 1 in the Supplement). For the identified risk factors, patients with delirium were statistically more likely to be 80 years or older (1432 [14.4%] vs 911 [5.2%], P < .001), have cognitive impairment (2069 [41.8%] vs 274 [1.2%], P < .001), have infection diagnoses (1528 [16.6%] vs 815 [4.4%], P < .001), and have abnormal (<135 or >145 mEq/L) sodium levels (592 [11.0%] vs 1720 [8.0%], P < .001). Weights for the 4 delirium risk factors of cognitive impairment (3 points), age of 80 years or older (1 point), sodium level (1 point), and infection (1 point) were chosen based on mean decrease in accuracy measures (Table 2). A key observation was the reduced overlap of populations in the consolidated NICE model and the large difference in median consolidated NICE scores among those with delirium (Figure 2).

or older (1 point), sodium level (1 point), and infection (1 point) were chosen based on mean decrease in accuracy measures (Table 2). A key observation was the reduced overlap of populations in the consolidated NICE model and the large difference in median consolidated NICE scores among those with delirium (Figure 2). The consolidated NICE scores discriminated between those with and without prevalent delirium (low risk, 252 [1.1%]; intermediate risk, 990 [29.4%]; high risk, 1101 [50.9%]; AUROC curve, 0.91; 95% CI, 0.90-0.92; P < .001) (eTable 2 in the Supplement). A comparison of the ROC curves of our consolidated NICE score vs those for the eNICE and Pendlebury NICE scores revealed the improvement in discrimination of the random forest prediction rule over existing approaches (eFigure in the Supplement). Figure 1. Importance Plot for the Random Forest Model Used to Generate the Consolidated National Institute for Health and Clinical Excellence (NICE) Score The mean decrease in accuracy shows the predictive power of each measure. Higher values indicate measures of higher importance. Cognitive impairment, infection, sodium level, and age were measures selected to be used in the consolidated NICE score. BUN indicates blood urea nitrogen; MAP, mean arterial pressure; WBC, white blood cell.

se in accuracy shows the predictive power of each measure. Higher values indicate measures of higher importance. Cognitive impairment, infection, sodium level, and age were measures selected to be used in the consolidated NICE score. BUN indicates blood urea nitrogen; MAP, mean arterial pressure; WBC, white blood cell. Table 2. Independent Risk Factors for Delirium From NICE-Based Prediction Rules Risk Factor for Delirium NICE Meta-analysis21 Odds Ratio (95% CI) Prediction Rule Weights eNICE Pendlebury NICE Consolidated NICE Cognitive impairment 6.3 (2.9-13.7) 4 2 3 Age ≥65 y 3.0 (1.2-7.7) 2 NA NA Age ≥80 y 5.2 (2.6-10.4) 3 2 1 Infection 3.0 (1.4-6.1) 2 1 1 Fracture 6.6 (2.2-19.3) 4 NA NA Visual impairment 1.7 (1.0-2.8) 1 1 NA Severe illness 3.5 (1.5-8.2) NA NA NA Acute physiology score NA 2 NA NA SIRS criteriaa NA NA 1 NA Serum sodium level NA NA NA 1 Abbreviations: eNICE, electronic National Institute for Health and Care Excellence; NA, not applicable; NICE, National Institute for Health and Care Excellence; SIRS, systematic inflammatory response syndrome. a SIRS was noted as positive if at least 2 of the following were present: heart rate greater than 90 beats/min, respiratory rate greater than 20 breaths/min, and white blood cell count less than 4000/μL or greater than 12 000/μL (to convert to ×109/L, multiply by 0.001). Temperature was not available in our data sets and therefore was excluded from SIRS calculations.

following were present: heart rate greater than 90 beats/min, respiratory rate greater than 20 breaths/min, and white blood cell count less than 4000/μL or greater than 12 000/μL (to convert to ×109/L, multiply by 0.001). Temperature was not available in our data sets and therefore was excluded from SIRS calculations. Figure 2. Comparison of the Consolidated National Institute for Health and Clinical Excellence (NICE) Score and Delirium Status A, The density histograms provide a probability breakdown of the consolidated NICE score by delirium status. B, The box plots show the association between the consolidated NICE score and delirium status in the derivation cohort. Center line in box indicates median; lower box border, 25th percentile or quartile 1; upper box border, 75th percentile or quartile 3; lower whisker, quartile 1 – 1.5 × interquartile range; upper whisker, quartile 3 + 1.5 × interquartile range.

n the consolidated NICE score and delirium status in the derivation cohort. Center line in box indicates median; lower box border, 25th percentile or quartile 1; upper box border, 75th percentile or quartile 3; lower whisker, quartile 1 – 1.5 × interquartile range; upper whisker, quartile 3 + 1.5 × interquartile range. For the confirmation cohort, each of the 3 NICE rules were associated with delirium at admission (Table 3). Table 3 highlights the discriminatory function of each of the models. In each model, increasing delirium risk points was associated with increased risk of delirium. eTable 3 in the Supplement presents additional statistics that demonstrate the consolidated NICE score’s improvement in the percentage of those classified correctly and positive predictive value. All scores had high negative predictive values: 0.96 (consolidated NICE score), 0.98 (eNICE score), and 0.96 (Pendlebury NICE score). The consolidated NICE model had significantly higher predictive capability compared with the other models in both cohorts (eTable 3 in the Supplement). In the derivation cohort, the consolidated NICE score correctly classified 25 332 patients (91.7%) with an AUROC curve of 0.91 (95% CI, 0.91-0.92; P < .001), eNICE correctly classified 18 896 patients (68.4%) with an AUROC curve of 0.81 (95% CI, 0.80-0.82; P < .001), and the Pendlebury NICE score correctly classified 24 697 patients (89.4%) with an AUROC curve of 0.87 (95% CI, 0.86-0.88; P < .001). In the confirmation cohort, the consolidated NICE score correctly classified 10 882 patients (92.6%) with an AUROC curve of 0.91 (95% CI, 0.90-0.92; P < .001), eNICE correctly classified 8332 patients (70.9%) with an AUROC curve of 0.83 (95% CI, 0.81-0.84; P < .001), and the Pendlebury NICE score correctly classified 10 647 patients (90.6%) with an AUROC curve of 0.87 (95% CI, 0.86-0.88; P < .001).

82 patients (92.6%) with an AUROC curve of 0.91 (95% CI, 0.90-0.92; P < .001), eNICE correctly classified 8332 patients (70.9%) with an AUROC curve of 0.83 (95% CI, 0.81-0.84; P < .001), and the Pendlebury NICE score correctly classified 10 647 patients (90.6%) with an AUROC curve of 0.87 (95% CI, 0.86-0.88; P < .001). Table 3. Comparison of 3 NICE Scores and Delirium Risk in the Derivation and Confirmation Cohorts Validated Prediction Rule and Delirium Risk (Points) Derivation Confirmation No. (%) With Delirium (n = 27 625) AUROC Curve (95% CI) No. (%) With Delirium (n = 11 752) AUROC Curve (95% CI) eNICE score Low (0-2) 75 (1.4) 0.81 (0.80-0.82) 27 (1.0) 0.83 (0.81-0.84) Intermediate (3-5) 415 (3.5) 136 (2.6) High (6-9) 917 (11.9) 344 (11.3) Very high (10-18) 936 (38.9) 315 (36.0) Pendlebury NICE score Low (0-1) 71 (0.6) 0.87 (0.86-0.88) 30 (0.6) 0.87 (0.86-0.88) Intermediate (2-4) 940 (7.0) 378 (6.8) High (5-7) 1332 (41.0) 414 (37.2) Consolidated NICE score Low (0-2) 252 (1.1) 0.91 (0.91-0.92) 103 (1.1) 0.91 (0.90-0.92) Intermediate (3-4) 990 (29.4) 371 (27.7) High (5-6) 1101 (50.9) 348 (46.5) Abbreviations: AUROC, area under the receiver operating characteristic; eNICE, electronic National Institute for Health and Care Excellence; NICE, National Institute for Health and Care Excellence.

.1) 0.91 (0.91-0.92) 103 (1.1) 0.91 (0.90-0.92) Intermediate (3-4) 990 (29.4) 371 (27.7) High (5-6) 1101 (50.9) 348 (46.5) Abbreviations: AUROC, area under the receiver operating characteristic; eNICE, electronic National Institute for Health and Care Excellence; NICE, National Institute for Health and Care Excellence. Discussion This analysis may provide an improvement over existing approaches to screening for delirium at admission by targeting 4 criteria and improving the accuracy compared with other algorithms. Furthermore, it revealed consistency in the NICE factors for delirium found in prior studies.16,17,21 By using the random forest method, we were able to identify cognitive impairment, age, sodium level, and infection as the variables that were primarily associated with delirium at admission. In addition, the use of separate but not prospective cohorts for derivation of the random forest algorithm and confirmation adds to the utility of this consolidated NICE delirium prediction rule for future research and implementation. This consolidated prediction rule may have clinical utility when used to identify patients in need of additional cognitive assessment and monitoring.

derivation of the random forest algorithm and confirmation adds to the utility of this consolidated NICE delirium prediction rule for future research and implementation. This consolidated prediction rule may have clinical utility when used to identify patients in need of additional cognitive assessment and monitoring. There may be some clinical advantages to recognizing delirium at admission. First, recent work has identified that a large percentage of patients at high risk for delirium may be missed with usual clinical care in the emergency and acute care settings.8,9,10 Of importance, patients with missed delirium have had negative outcomes in prior studies.11,12,13 Delirium at admission accounts for up to one-third of cases of delirium.28 Second, prior work with prevention strategies for delirium has excluded persons with delirium before enrollment.29,30,31 Third, the clinical guidelines lean toward routine screening for delirium specifically to identify and treat delirium early.20,32,33 Because of the low delirium recognition and poor outcomes associated with unrecognized delirium, an EMR-based delirium prediction tool may have some advantages after it is externally confirmed.

the clinical guidelines lean toward routine screening for delirium specifically to identify and treat delirium early.20,32,33 Because of the low delirium recognition and poor outcomes associated with unrecognized delirium, an EMR-based delirium prediction tool may have some advantages after it is externally confirmed. The value of any delirium prediction rule lies in its application to clinical practice. Because many cases of delirium are missed during routine clinical care, the use of an electronic tool to focus clinical assessment may efficiently direct clinical efforts. For example, if the consolidated NICE rule is embedded in the EMR and an electronic flag identifies the patient as high risk, a trained, frontline nurse can perform an ultrabrief assessment of cognitive function, such as the modified Richmond Agitation and Sedation Scale (15 seconds) or Months of the Year Backward (2 minutes),34,35 and initiate nonpharmacologic prevention strategies or ask for a more comprehensive assessment of delirium diagnosis and treatable causes. This brief, 2-step process may aid in quickly identifying patients at high risk for delirium and in need of more in-depth assessment and intervention for delirium. In addition, high delirium risk conveys prognostic information that is valuable to practitioners. For example, the eNICE confirmation study16 found that those at high and very high delirium risk at admission had increased length of stay, discharge to a rehabilitation facility, and readmissions compared with low-risk patients. As a result, systematic use of a delirium prediction rule, particularly an electronic measure, may efficiently identify patients who would benefit from additional clinical assessment.

isk at admission had increased length of stay, discharge to a rehabilitation facility, and readmissions compared with low-risk patients. As a result, systematic use of a delirium prediction rule, particularly an electronic measure, may efficiently identify patients who would benefit from additional clinical assessment. Strengths and Limitations The strengths of this analysis are the use of large derivation and confirmation cohorts from centers across the United States, the random forest method, and the association with delirium at admission. Data were systematically collected from the VA EPRP and have been demonstrated to be highly reliable. Use of the NICE meta-analysis as the core of the random forest method enhances the validity and generalizability.

ters across the United States, the random forest method, and the association with delirium at admission. Data were systematically collected from the VA EPRP and have been demonstrated to be highly reliable. Use of the NICE meta-analysis as the core of the random forest method enhances the validity and generalizability. Despite the strengths of this work, there are significant limitations, and the results must be interpreted within these limits. First, generalizability is affected by the use of VA medical centers, which have a high proportion of men. Second, the VA system comprises 150 medical centers, and there is inherent variability in the assessment, diagnosis, and treatment of delirium. In this study, there was no external confirmation; additional, prospective confirmation in other health care systems, particularly those with EMR systems, is needed. Next, the EMR-based abstraction could limit the selection of additional variables because of availability, coding, or delay to diagnosis. Because our analysis was restricted to variables collected, there could be variables associated with delirium (eg, medications, prior living arrangement) that were not included in our random forest method. This factor may have affected the features selected by building random forests because of unavailability of some patient characteristics, such as vision or hearing data. Another limitation is the use of retrospective cohort EMR terms for delirium, which underestimates the prevalence of delirium.36 These terms also favor hyperactive delirium compared with hypoactive delirium, which is an inherent bias. Another limitation is that admissions were limited to 4 admission groups, which did not include groups at high risk for delirium, such as those with fractures.

um, which underestimates the prevalence of delirium.36 These terms also favor hyperactive delirium compared with hypoactive delirium, which is an inherent bias. Another limitation is that admissions were limited to 4 admission groups, which did not include groups at high risk for delirium, such as those with fractures. The next step to move the consolidated NICE algorithm forward is confirmation in an external, prospective cohort with standardized assessment to improve validity for delirium at admission and delirium that develops after admission. Similar findings in such a prospective cohort with the reference standard delirium diagnosis may suggest that the internal validity of this retrospective study was not compromised. Conclusions Practitioners frequently miss delirium at admission. Using advanced random forest methods for variable selection, this analysis found good discriminatory function for delirium at admission with 4 elements: cognitive impairment, age, sodium level, and infection. Further prospective examination of the consolidated NICE screening algorithm is required. Building screening algorithms such as this one into an EMR system in the future may help alert practitioners to individuals who would benefit from standardized cognitive assessment and appropriate interventions. Supplement. eAppendix. Methods of Creating the Consolidated National Institute for Health and Care Excellence eTable 1. Random Forest Selected Variables and Association With Delirium at Admission eTable 2. Categories of Consolidated NICE Delirium Prediction Rule

Conclusions Practitioners frequently miss delirium at admission. Using advanced random forest methods for variable selection, this analysis found good discriminatory function for delirium at admission with 4 elements: cognitive impairment, age, sodium level, and infection. Further prospective examination of the consolidated NICE screening algorithm is required. Building screening algorithms such as this one into an EMR system in the future may help alert practitioners to individuals who would benefit from standardized cognitive assessment and appropriate interventions. Supplement. eAppendix. Methods of Creating the Consolidated National Institute for Health and Care Excellence eTable 1. Random Forest Selected Variables and Association With Delirium at Admission eTable 2. Categories of Consolidated NICE Delirium Prediction Rule eTable 3. Additional Performance Metrics of the 3 Prediction Rules eFigure. Receiver Operating Curves for the 3 NICE Rules for Delirium Upon Admission Click here for additional data file.

Introduction We have tracked rates of substance use among American Indian adolescents attending schools on or near reservations since 1974, and American Indian adolescents have consistently reported the highest levels of substance use compared with other US racial/ethnic groups.1,2,3,4 Other studies of mostly nonreservation American Indian youths report similar results.5,6,7 When compared with a national sample of students from Monitoring the Future (MTF), our study of data from 2009 to 2012 indicated that American Indian students reported lifetime rates of marijuana 3.4 and 1.6 times higher for 8th- and 12th-grade students, respectively.8 Differences for current marijuana use for these grades were larger, at 4.8 to 1.6 times higher, respectively. An important consideration for American Indian youths is that, in addition to high rates of substance use, risk for substance use begins early.9,10,11 As previously reported, we compared initiation rates of alcohol, marijuana, and inhalants between a population-based sample of reservation-based American Indian youths with white youths attending the same schools, and odds ratios for initiation comparing these 2 groups were 2.3 (95% CI, 1.5-3.4) for intoxication, 10.5 (95% CI, 6.0-18.5) for marijuana, and 1.8 (95% CI, 1.3-2.4) for inhalants, with American Indian youths more likely to initiate early use.12

based American Indian youths with white youths attending the same schools, and odds ratios for initiation comparing these 2 groups were 2.3 (95% CI, 1.5-3.4) for intoxication, 10.5 (95% CI, 6.0-18.5) for marijuana, and 1.8 (95% CI, 1.3-2.4) for inhalants, with American Indian youths more likely to initiate early use.12 The harm associated with high rates of use and early initiation for American Indian youths include increasing rates of use in early and later adulthood, higher risk of developing a substance use disorder, and more alcohol-related problems, including alcohol-attributable death.7,13,14 Furthermore, American Indian and Alaska Native youths are more likely to need treatment for a substance use problem than all other US racial/ethnic groups.15 These findings underscore the need for continuing surveillance of this at-risk group, particularly given changing trends in perceived harmfulness of illicit substances as new statutes alter access to medical and recreational use of cannabis.16,17 Our previous surveys corresponded closely to items from the MTF study, an ongoing population-based study of US youths. In coordination with the National Institute on Drug Abuse Epidemiology Research Branch and MTF staff, we revised our survey (Our Youth, Our Future [OYOF]) to use identical items from MTF, making direct comparisons possible. This allows for ongoing comparisons of substance use rates between population-based samples of reservation-based American Indian students and national US students.

ogy Research Branch and MTF staff, we revised our survey (Our Youth, Our Future [OYOF]) to use identical items from MTF, making direct comparisons possible. This allows for ongoing comparisons of substance use rates between population-based samples of reservation-based American Indian students and national US students. Methods Sample and Recruitment The data are part of an ongoing epidemiologic study of substance use by American Indian youths living on or near reservations, where students in grades 7 through 12 in sampled schools complete online surveys of substance use. The sampling frame was built from 3 primary sources: the National Center for Education Statistics Common Core of Data, the National Center for Education Statistics Private School Universe Survey, and the Bureau of Indian Education National Directory. We followed the American Association for Public Opinion Research (AAPOR) reporting guideline. The final sampling frame contained 353 schools that included students in seventh grade or higher, located on or within 25 miles of a reservation or tribal lands, with at least 20% of students enrolled being American Indian. Schools outside the continental United States were excluded, as were Oklahoma tribal statistical areas. Schools were stratified into 7 regions (Northeast, Northwest, Southeast, Southwest, Northern Plains, Southern Plains, and Upper Great Lakes) based on cultural and other similarities among American Indian groups.18

n. Schools outside the continental United States were excluded, as were Oklahoma tribal statistical areas. Schools were stratified into 7 regions (Northeast, Northwest, Southeast, Southwest, Northern Plains, Southern Plains, and Upper Great Lakes) based on cultural and other similarities among American Indian groups.18 Each year schools within each region are randomly drawn from the sampling frame to reflect the regional distribution of American Indian persons residing in each stratum based on the US 2010 Census data. Because of the small number of schools in the Northeast, Southern Plains, and Southeast, all schools and/or school districts meeting the requirements of the sample are invited to participate. For schools not including high school grades (eg, middle schools), the high school most likely to be attended by the American Indian middle school students is determined and added to the drawn sample. Participating schools were reasonably representative of sampled schools in several measurable demographic characteristics. Approximately 28% of sampled schools and 21% of surveyed schools were Bureau of Indian Education schools, while 67% of sampled schools and 79% of participating schools were public. Middle schools composed 27% of sampled schools and 21% of surveyed schools; K-12 schools were 11% of sampled schools and 10% of participating schools; and K-8 (or similar) schools were 27% of sampled schools and 21% of surveyed schools. Schools with greater than 80% American Indian students composed 50% of sampled schools and 41% of participating schools, and schools on reservations were 65% of the sample vs 55% of participating schools.

10% of participating schools; and K-8 (or similar) schools were 27% of sampled schools and 21% of surveyed schools. Schools with greater than 80% American Indian students composed 50% of sampled schools and 41% of participating schools, and schools on reservations were 65% of the sample vs 55% of participating schools. For each participating school, the appropriate tribal and school board approvals were obtained. Each school received a comprehensive report of their survey findings and compensation for resources used to complete the survey process, ranging from $750 to $5000 depending on enrollment, with median payment being $1500. All survey responses were collected anonymously, and all procedures were approved by the Colorado State University institutional review board. The board approved waiver of signed parental consent; however, assent was obtained from both parents and students. Participants This study used survey data from participating schools during fall 2016 to spring 2017 semesters of the 2016-2017 academic year. Specific identities of tribes and reservations were kept confidential. Across participating schools, the number of completed surveys (n = 6065) represented 87.7% of student enrollment in those schools. To make direct comparisons with MTF, only students in 8th, 10th, and 12th grade who self-identified as American Indian were included in the analysis.

nd reservations were kept confidential. Across participating schools, the number of completed surveys (n = 6065) represented 87.7% of student enrollment in those schools. To make direct comparisons with MTF, only students in 8th, 10th, and 12th grade who self-identified as American Indian were included in the analysis. Sample sizes for MTF ranged from 8450 to 16 900 for eighth graders, 7350 to 15 900 for 10th graders, and 3933 to 11 800 for 12th graders, depending on the substance. The number of cases varied by substance because multiple questionnaire forms are randomly distributed at each grade level, and not all questions are asked on all forms to reduce response burden.19 Procedure Approximately 3 weeks before the scheduled survey, letters were sent to parents of enrolled students in grades 7 and higher describing the survey and providing instructions for opting their child out of the survey. This information was also posted on other local media sites, where parents were likely to see it. Less than 1% of students refused to take the survey or were opted out by their parents. The OYOF survey was administered online to students using Qualtrics software. School staff read directions prior to survey administration indicating that students could decline to participate or leave blank any questions they did not wish to answer, and these instructions were repeated in the online survey. Staff were instructed to remain in an area of the room where they could not observe students’ responses.

ad directions prior to survey administration indicating that students could decline to participate or leave blank any questions they did not wish to answer, and these instructions were repeated in the online survey. Staff were instructed to remain in an area of the room where they could not observe students’ responses. Measures The OYOF survey contains a verbatim subset of the most recent substance use questions asked in the MTF survey. For each substance, the OYOF survey asks about lifetime and last-30-day use in addition to measures of demographic characteristics. Substance use measures for each question were coded as 1 for any use and 0 for no use. In addition to single substance use measures, a composite measure of illicit drug use was calculated, using the MTF definition, as any use of lysergic acid diethylamide (LSD), other hallucinogens, crack, cocaine other than crack, heroin, any use of narcotics other than heroin (for 12th grade only), amphetamines, sedatives (barbiturates), or tranquilizers not under a physician’s orders.

illicit drug use was calculated, using the MTF definition, as any use of lysergic acid diethylamide (LSD), other hallucinogens, crack, cocaine other than crack, heroin, any use of narcotics other than heroin (for 12th grade only), amphetamines, sedatives (barbiturates), or tranquilizers not under a physician’s orders. Statistical Analysis Demographic characteristics and outcome data are reported using descriptive statistics, proportions with 95% confidence intervals, and relative risk (RR) ratios with 95% confidence intervals. For each outcome measure at each grade, lifetime and last-30-day use prevalence rates and their 95% confidence intervals were computed, excluding missing data, using Stata version 15 (StataCorp) survey commands, with stratification by region and with school as the primary sampling unit. Missingness varied between 0.4% and 2.6% for individual substance measures except hallucinogens other than LSD, for which missing rates varied from 4.8% for grade 10 to 14.4% for grade 8. This was because 1 tribal group requested that this question not be asked. Observations were weighted to correct for overrepresentation or underrepresentation by region, with weights based on the US 2010 Census reservation and off-reservation trust land state populations for ages 10 to 19 years. Weights varied from 0.39 (Northeast) to 2.25 (Southeast), with the remaining weights between 0.70 and 1.30. Comparable MTF substance use rates were obtained from MTF’s National Survey Results for Drug Use 1975 to 2016.19

S 2010 Census reservation and off-reservation trust land state populations for ages 10 to 19 years. Weights varied from 0.39 (Northeast) to 2.25 (Southeast), with the remaining weights between 0.70 and 1.30. Comparable MTF substance use rates were obtained from MTF’s National Survey Results for Drug Use 1975 to 2016.19 The RRs comparing American Indian and MTF students’ lifetime and last-30-day use prevalence rates and their 95% confidence intervals with a test of significant difference from 1 (P < .05, 2-sided) were calculated using MedCalc statistical software version 16.4.3 (MedCalc Software). To account for nesting of students within schools, effective sample sizes for each substance within a sample were computed as actual sample size divided by design effect for that substance.20 We compared lifetime and last-30-day use RRs from our 2009 to 2012 data6 with the current RRs averaged across the 3 grades for alcohol, marijuana, and other drugs. A composite alcohol RR was calculated as the mean alcohol, intoxication, and binge drinking RR (for last 30 days), while the other-drug RR was the mean RR for all other drugs except marijuana.

use RRs from our 2009 to 2012 data6 with the current RRs averaged across the 3 grades for alcohol, marijuana, and other drugs. A composite alcohol RR was calculated as the mean alcohol, intoxication, and binge drinking RR (for last 30 days), while the other-drug RR was the mean RR for all other drugs except marijuana. Results Thirty-one schools participated in the OYOF survey with the following regional distribution of students: Northeast, 6.1%; Northwest, 9.7%; Northern Plains, 20.5%; Southeast, 9.9%; Southwest, 43.3%; and Upper Great Lakes, 10.6%. A total of 87.7% of enrolled students were surveyed. Participants included 570 students in eighth grade (49.6% girls; mean [SD] age, 13.5 [0.71] years), 582 in 10th grade (50.0% girls; mean [SD] age, 15.4 [0.67] years), and 508 in 12th grade (53.5% girls; mean [SD] age, 17.4 [0.70] years). Table 1 and Table 2 present lifetime and last-30-day substance use prevalence rates for 8th-, 10th-, and 12th-grade students in the American Indian and MTF samples. In addition to point estimates and confidence intervals for American Indian students, the RR ratios are also presented, along with 95% confidence intervals and a test of significant difference from 1 (P < .05, 2-sided).

se prevalence rates for 8th-, 10th-, and 12th-grade students in the American Indian and MTF samples. In addition to point estimates and confidence intervals for American Indian students, the RR ratios are also presented, along with 95% confidence intervals and a test of significant difference from 1 (P < .05, 2-sided). Table 1. Lifetime Prevalence of Alcohol and Drug Use Comparing Reservation-Based American Indian Students (2016-2017) With MTF Students (2016) Type of Substance Use Grade 8 Grade 10 Grade 12 American Indian, % (95% CI) MTF, %a RR (95% CI) American Indian, % (95% CI) MTF, %a RR (95% CI) American Indian, % (95% CI) MTF, %a RR (95% CI) Alcohol 39.7 (31.4-48.6) 22.8 1.7 (1.4-2.2)b 52.9 (46.5-59.2) 43.4 1.2 (1.1-1.4)b 72.5 (66.0-78.1) 61.2 1.2 (1.1-1.3)b Been drunk 22.9 (17.3-29.7) 8.6 2.7 (2.0-3.5)b 39.2 (31.6-47.4) 26.0 1.5 (1.2-1.9)b 56.5 (49.2-63.5) 46.3 1.2 (1.0-1.4)b Marijuana 43.7 (35.1-52.7) 12.8 3.4 (2.8-4.2)b 55.6 (46.7-64.2) 29.7 1.9 (1.6-2.2)b 66.4 (57.3-74.5) 44.5 1.5 (1.3-1.7)b Any illicit drug, not marijuanac 16.2 (12.7-20.5) 8.9 1.8 (1.5-2.3)b 19.1 (12.8-27.6) 14.0 1.4 (0.9-2.0) 24.4 (18.8-31.0) 20.7 1.2 (0.9-1.5) Inhalants 13.2 (9.7-17.7) 7.7 1.7 (1.3-2.3)b 10.7 (8.6-13.1) 6.6 1.6 (1.3-2.1)b 10.8 (8.0-14.3) 5.0 2.2 (1.3-3.5)b Tranquilizers 3.6 (2.2-5.7) 3.0 1.2 (0.8-2.0) 6.2 (3.6-10.7) 6.1 1.0 (0.6-1.7) 5.0 (3.5-7.1) 7.6 0.7 (0.5-0.9)b Narcotics other than heroin 3.0 (2.0-4.5) NA NA 8.1 (5.2-12.3) NA NA 10.9 (7.6-15.4) 7.8 1.4 (0.9-2.0) Amphetamines 4.0 (2.3-6.8) 5.7 0.7 (0.5-1.2) 5.8 (3.5-9.4) 8.8 0.7 (0.4-1.1) 10.0 (5.9-16.4) 10.0 1.0 (0.7-1.5) Cocaine 4.3 (2.8-6.5) 1.1 3.9 (2.3-6.5)b 6.4 (3.6-11.0) 1.9 3.4 (1.8-6.0)b 11.8 (8.3-16.4) 3.3 3.6 (2.3-5.4)b Crack 3.0 (1.7-5.2) 0.9 3.3 (1.8-5.7)b 3.8 (2.3-6.2) 0.8 4.8 (2.8-8.2)b 5.6 (3.4-9.3) 1.4 4.0 (2.2-6.9)b LSD 3.9 (2.6-5.9) 1.2 3.3(2.1-5.1)b 6.1 (3.6-10.3) 3.2 1.9 (1.2-3.3)b 9.8 (6.3-14.9) 4.9 2.0 (1.2-3.2)b Hallucinogens other than LSD 8.1 (5.6-11.6) 1.9 4.3 (2.8-6.4)b 9.9 (7.3-13.4) 3.1 3.2 (2.3-4.4)b 12.3 (8.6-17.4) 4.7 2.8 (1.9-4.1)b Heroin 2.8 (1.8-4.3) 0.5 5.6 (3.3-9.8)b 2.4 (1.4-3.9) 0.6 4.0 (2.3-7.4)b 3.2 (2.2-4.6) 0.7 4.6 (2.9-7.3)b Crystal methd 2.6 (1.4-4.8) 0.6 4.3 (2.2-7.9)b 5.3 (4.0-7.0) 0.7 7.6 (5.1-10.8)b 8.0 (4.2-14.7) 1.4 5.7 (3.0-12.6)b Cigarettes 29.7 (22.4-38.1) 9.8 3.0 (2.3-4.0)b 42.0 (36.2-47.9) 17.5 2.4 (2.0-2.8)b 49.7 (44.2-55.1) 28.3 1.8 (1.6-2.01)b Abbreviations: LSD, lysergic acid diethylamide; MTF, Monitoring the Future; NA, not available; RR, relative risk.

4.3 (2.2-7.9)b 5.3 (4.0-7.0) 0.7 7.6 (5.1-10.8)b 8.0 (4.2-14.7) 1.4 5.7 (3.0-12.6)b Cigarettes 29.7 (22.4-38.1) 9.8 3.0 (2.3-4.0)b 42.0 (36.2-47.9) 17.5 2.4 (2.0-2.8)b 49.7 (44.2-55.1) 28.3 1.8 (1.6-2.01)b Abbreviations: LSD, lysergic acid diethylamide; MTF, Monitoring the Future; NA, not available; RR, relative risk. a Confidence intervals for MTF data can be found in Miech et al,19 Tables 4-1a through d. b P < .05. c Use of any illicit drug includes any use of LSD, other hallucinogens, crack, cocaine other than crack, heroin, any use of narcotics other than heroin (grade 12 only), amphetamines, or tranquilizers not under a physician’s orders. d Numbers for MTF grades 8 and 10 are for methamphetamine (including crystal meth). Table 2.

b P < .05. c Use of any illicit drug includes any use of LSD, other hallucinogens, crack, cocaine other than crack, heroin, any use of narcotics other than heroin (grade 12 only), amphetamines, or tranquilizers not under a physician’s orders. d Numbers for MTF grades 8 and 10 are for methamphetamine (including crystal meth). Table 2. Last-30-Day Prevalence of Alcohol and Drug Use Comparing Reservation-Based American Indian Students (2016-2017) With MTF Students (2016) Type of Substance Use Grade 8 Grade 10 Grade 12 American Indian, % (95% CI) MTF, %a RR (95% CI) American Indian, % (95% CI) MTF, %a RR (95% CI) American Indian, % (95% CI) MTF, %a RR (95% CI) Alcohol 15.8 (10.7-22.7) 7.3 2.1 (1.4-3.0)b 24.1 (20.0-28.7) 19.9 1.2 (1.0-1.5) 30.7 (25.1-36.9) 33.2 0.9 (0.8-1.1) Been drunk 9.6 (5.8-15.4) 1.8 5.3 (3.3-8.9)b 16.5 (12.9-20.8) 9.0 1.8 (1.4-2.4)b 23.2 (17.7-29.8) 20.4 1.1 (0.8-1.5) Binge drinking 11.8 (6.4-20.6) 3.4 3.5 (2.0-6.0)b 16.6 (13.6-20.0) 9.7 1.7 (1.4-2.1)b 22.8 (18.3-28.1) 15.5 1.5 (1.2-1.9)b Marijuana 22.5 (16.1-30.5) 5.4 4.2 (3.1-5.8)b 35.1 (28.2-42.8) 14.0 2.5 (2.0-3.1)b 39.3 (32.1- 46.9) 22.5 1.7 (1.4-2.2)b Any illicit drug, not marijuanac 6.4 (4.6-8.9) 2.7 2.4 (1.7-3.3)b 6.7 (3.8-11.7) 4.4 1.5 (0.9-2.7) 9.7 (7.0-13.3) 6.9 1.4 (0.9-2.0) Inhalants 4.9 (3.4-7.2) 1.8 2.7 (1.8-4.1)b 2.2 (1.2-3.9) 1.0 2.2 (1.2-4.3)b 2.1 (1.2-3.6) 0.8 2.6 (1.2-6.1)b Tranquilizers 1.4 (0.7-2.9) 0.8 1.8 (0.9-3.5) 1.6 (0.9-3.0) 1.5 1.1 (0.5-1.9) 2.7 (1.8-4.2) 1.9 1.4 (0.9-2.3) Narcotics other than heroin 1.3 (0.7-2.3) NA NA 2.8 (1.6-4.9) NA NA 4.9 (2.8-8.2) 1.7 2.9 (1.6-5.2)b Amphetamines 1.6 (0.8-3.4) 1.7 0.9 (0.5-1.8) 2.5 (1.5-4.1) 2.7 0.9 (0.5-1.6) 4.5 (2.8-7.3) 3.0 1.5 (0.8-2.6) Cocaine 1.2 (0.6-2.5) 0.3 4.0 (1.9-9.4)b 2.4 (1.3-4.4) 0.3 8.0 (3.7-17.6)b 4.1 (2.8-6.1) 0.6 6.8 (3.5-13.1)b Crack 0.8 (0.3-2.1) 0.2 4.0 (1.5-10.1)b 1.3 (5.9-2.9) 0.2 6.5 (2.5-15.0)b 1.7 (0.6-4.8) 0.5 3.4 (1.6-12.3)b LSD 1.5 (0.8-2.8) 0.4 3.8 (2.3-9.1)b 1.5 (0.6-4.2) 0.7 2.1 (0.8-6.4) 2.5 (1.4-4.6) 1.0 2.5 (1.2-5.1)b Hallucinogens other than LSD 2.2 (1.3-3.8) 0.3 7.3 (4.1-13.1)b 3.7 (1.8-7.4) 0.5 7.4 (3.4-15.1)b 2.8 (1.7-4.6) 0.7 4.0 (2.2-7.3)b Heroin 0.8 (0.3-2.1) 0.2 4.0 (1.4-10.9)b 0.6 (0.2-1.3) 0.2 3.0 (1.2-8.2)b 0.5 (0.1-3.2) 0.2 2.5 (0.3-18.3) Crystal methd 1.0 (0.4-2.4) 0.3 3.3 (2.6-4.3)b 1.4 (0.8-2.7) 0.2 7.0 (3.1-14.9)b 3.3 (1.6-6.8) 0.4 8.3 (3.0-19.1)b Cigarettes 10.6 (7.8-14.2) 2.6 4.1 (2.9-5.8)b 15.1 (10.4-21.3) 4.9 3.1 (2.1-4.7)b 23.1 (17.9-29.4) 10.5 2.2 (1.6-2.9)b Abbreviations: LSD, lysergic acid diethylam

1.2-8.2)b 0.5 (0.1-3.2) 0.2 2.5 (0.3-18.3) Crystal methd 1.0 (0.4-2.4) 0.3 3.3 (2.6-4.3)b 1.4 (0.8-2.7) 0.2 7.0 (3.1-14.9)b 3.3 (1.6-6.8) 0.4 8.3 (3.0-19.1)b Cigarettes 10.6 (7.8-14.2) 2.6 4.1 (2.9-5.8)b 15.1 (10.4-21.3) 4.9 3.1 (2.1-4.7)b 23.1 (17.9-29.4) 10.5 2.2 (1.6-2.9)b Abbreviations: LSD, lysergic acid diethylam ide; MTF, Monitoring the Future; NA, not available; RR, relative risk. a Confidence intervals for MTF data can be found in Miech et al,19 Tables 4-1a through d. b P < .05. c Use of any illicit drug includes any use of LSD, other hallucinogens, crack, cocaine other than crack, heroin, any use of narcotics other than heroin (grade 12 only), amphetamines, or tranquilizers not under a physician’s order. d Numbers for MTF grades 8 and 10 are for methamphetamine (including crystal meth).

a Confidence intervals for MTF data can be found in Miech et al,19 Tables 4-1a through d. b P < .05. c Use of any illicit drug includes any use of LSD, other hallucinogens, crack, cocaine other than crack, heroin, any use of narcotics other than heroin (grade 12 only), amphetamines, or tranquilizers not under a physician’s order. d Numbers for MTF grades 8 and 10 are for methamphetamine (including crystal meth). Lifetime Prevalence Lifetime rates of American Indian students for all substance measures except tranquilizers and amphetamines for each grade were higher than MTF rates at P < .05 (Table 1). The highest lifetime rates for American Indian students in grade 8 were for marijuana (43.7% [95% CI, 35.1%-52.7%]), alcohol (39.7% [95% CI, 31.4%-48.6%]), cigarettes (29.7% [95% CI, 22.4%-38.1%]), and having been drunk (22.9% [95% CI, 17.3%-29.7%]), with respective RRs of 3.4 (95% CI, 2.8-4.2), 1.7 (95% CI, 1.4-2.2), 3.0 (95% CI, 2.3-4.0), and 2.7 (95% CI, 2.0-3.5). Results for students in grades 10 and 12 were similar, but with greater reported use as grades increased for both American Indian and MTF students and lower RRs, although they were still significantly different than 1. For example, the American Indian students’ rate of marijuana use increased to 55.6% (95% CI, 46.7%-64.2%) by 10th grade, but its corresponding RR decreased to 1.9 (95% CI, 1.6-2.2).

use as grades increased for both American Indian and MTF students and lower RRs, although they were still significantly different than 1. For example, the American Indian students’ rate of marijuana use increased to 55.6% (95% CI, 46.7%-64.2%) by 10th grade, but its corresponding RR decreased to 1.9 (95% CI, 1.6-2.2). Lifetime illicit drug rates excluding marijuana were 16.2% (95% CI, 12.7%-20.5%), 19.1% (95% CI, 12.8%-27.6%), and 24.4% (95% CI, 18.8%-31.0%) for American Indian students in grades 8, 10, and 12, respectively. The lifetime illicit drug RR for grade 8 was significantly different from 1 (1.8 [95% CI, 1.5-2.3]) but the RRs were not significantly different from 1 for grade 10 (1.4 [95% CI, 0.9-2.0]) or grade 12 (1.2 [95% CI, 0.9-1.5]).

4.4% (95% CI, 18.8%-31.0%) for American Indian students in grades 8, 10, and 12, respectively. The lifetime illicit drug RR for grade 8 was significantly different from 1 (1.8 [95% CI, 1.5-2.3]) but the RRs were not significantly different from 1 for grade 10 (1.4 [95% CI, 0.9-2.0]) or grade 12 (1.2 [95% CI, 0.9-1.5]). Last-30-Day Prevalence As with lifetime use, for American Indian students in grade 8, last-30-day use rates for all substance measures except tranquilizers and amphetamines were higher than MTF rates at P < .05 (Table 2). The highest rates were for marijuana (22.5% [95% CI, 16.1%-30.5%]), alcohol (15.8% [95% CI, 10.7%-22.7%]), binge drinking (11.8% [95% CI, 6.4%-20.6%]), and cigarettes (10.6% [95% CI, 7.8%-14.2%]), with respective RRs of 4.2 (95% CI, 3.1-5.8), 2.1 (95% CI, 1.4-3.0), 3.5 (95% CI, 2.0-6.0), and 4.1 (95% CI, 2.9-5.8), respectively. These RRs (excluding binge drinking, which has no lifetime measure) are greater than those for lifetime use. As grade increased, last-30-day use for these measures increased for both American Indian and MTF students, but RRs decreased. The RRs for 10th and 12th graders for alcohol were not significantly different from 1, nor was the RR for 12th graders for having been drunk.

e measure) are greater than those for lifetime use. As grade increased, last-30-day use for these measures increased for both American Indian and MTF students, but RRs decreased. The RRs for 10th and 12th graders for alcohol were not significantly different from 1, nor was the RR for 12th graders for having been drunk. Last-30-day illicit drug use rates excluding marijuana for American Indian students in grades 8, 10, and 12 were 6.4% (95% CI, 4.6%-8.9%), 6.7% (95% CI, 3.8%-11.7%), and 9.7% (95% CI, 7.0%-13.3%), compared with respective MTF rates of 2.7%, 4.4%, and 6.9%. The RR associated with these rates for eighth graders was 2.4 (95% CI, 1.7-3.3), while the rates for 10th and 12th graders were not significantly different from 1 at P < .05. In addition, RRs for amphetamines and tranquilizers were not significantly different from 1, nor was the RR for LSD for 10th graders or the RR for heroin for 10th and 12th graders. All other RRs were significantly different from 1, with use by American Indian students being greater than use by MTF students.

at P < .05. In addition, RRs for amphetamines and tranquilizers were not significantly different from 1, nor was the RR for LSD for 10th graders or the RR for heroin for 10th and 12th graders. All other RRs were significantly different from 1, with use by American Indian students being greater than use by MTF students. Change in RR From 2009-2012 and 2016-2017 The RR between American Indian and MTF students for lifetime alcohol use changed little from 2009-2012 (RR, 1.5 [95% CI, 1.4-1.6]) to 2016-2017 (RR, 1.3 [95% CI, 1.2-1.4]). This was also true for lifetime marijuana use (2009-2012: RR, 2.0 [95% CI, 1.8-2.1]; 2016-2017: RR, 2.1 [95% CI, 1.9-2.3]). However, lifetime RR for use of other drugs increased substantially across these years with an RR of 1.8 (95% CI, 1.7-1.9) for 2009 to 2012 and an RR of 3.0 (95% CI, 2.9-3.2) for 2016-2017. For last-30-day use, the RR increased for last-month alcohol use from 1.5 (95% CI, 1.3-1.7) to 2.1 (95% CI, 1.9-2.3), but no change was found for RR for marijuana use, with the RR at 2.8 for both points. The RR for other drug use increased from 2.5 (95% CI, 2.3-2.8) to 3.9 (95% CI, 3.6-4.2).

CI, 2.9-3.2) for 2016-2017. For last-30-day use, the RR increased for last-month alcohol use from 1.5 (95% CI, 1.3-1.7) to 2.1 (95% CI, 1.9-2.3), but no change was found for RR for marijuana use, with the RR at 2.8 for both points. The RR for other drug use increased from 2.5 (95% CI, 2.3-2.8) to 3.9 (95% CI, 3.6-4.2). Discussion Findings from this study demonstrate that American Indian adolescents who reside on or near reservations continue a trend of using nearly all substances at substantially higher rates than adolescents from a nationally representative sample (MTF). Lifetime exposure was higher for American Indian reservation-based youths, with significant RRs compared with national youths. The only exceptions to this pattern were for amphetamines and tranquilizers at all grades and lifetime use of any illicit drug other than marijuana at grades 10 and 12. Similarly, higher rates for American Indian youths were found for current (last 30 days) substance use, with the exception of amphetamines and tranquilizers at all grades, alcohol at grades 10 and 12, having been drunk at grade 12, any illicit drug other than marijuana at grades 10 and 12, LSD at grade 10, and heroin at grade 12. The RRs decreased with grade in school; by grade 12, MTF students were more similar to American Indian students, but American Indian students’ rates remained an average of at least twice that of MTF students’ rates.

12, any illicit drug other than marijuana at grades 10 and 12, LSD at grade 10, and heroin at grade 12. The RRs decreased with grade in school; by grade 12, MTF students were more similar to American Indian students, but American Indian students’ rates remained an average of at least twice that of MTF students’ rates. The higher rate of substance use among American Indian students compared with MTF students at grade 8 stresses the critical need for early prevention intervention efforts for American Indian youths living on or near reservations. Yet, few interventions have been developed and tested for this group.21 The distinct living environment of the reservation, coupled with high normative rates of use by peers and adults,22 creates prevention and treatment challenges unique to these youths. While American Indian youths are similar to other youths in many respects, with similar risk and protective factor profiles,22,23,24 these youths experience high rates of trauma and loss, such as suicide, accidents, violence, and substance abuse, in addition to other adverse childhood experiences such as child abuse and household dysfunction.25,26,27 Studies have established relationships between these experiences and higher rates of alcohol and drug use.28,29,30,31 Prevention efforts found to be effective in the general population may show less effectiveness in this population because of differences in childhood experiences and environment. Moreover, prevention efforts that do not attend carefully to cultural adaptation may not be acceptable, increasing the chance for failure.32 Cultural and value-based characteristics unique to American Indian populations, such as traditional Native American spirituality and the importance of expanded kinship networks, may provide beneficial targets for prevention programming, although there is still little etiologic evidence on how these work to prevent risky behaviors. These can be incorporated into existing evidence-based interventions or used in developing interventions from the ground up to create a portfolio of evidence-based, culturally grounded interventions.32 While the need for prevention and intervention is high and the research base is low compared with programs for other minority groups, scientific evidence is building.

-based interventions or used in developing interventions from the ground up to create a portfolio of evidence-based, culturally grounded interventions.32 While the need for prevention and intervention is high and the research base is low compared with programs for other minority groups, scientific evidence is building. For example, under the Interventions for Health Promotion and Disease Prevention in Native American Populations (PAR-14-260), the National Institutes of Health have funded a number of studies to adapt, develop, and test substance use prevention programs with traditional Native American practices and cultural traditions to establish evidence-based practices for this population.33

Disease Prevention in Native American Populations (PAR-14-260), the National Institutes of Health have funded a number of studies to adapt, develop, and test substance use prevention programs with traditional Native American practices and cultural traditions to establish evidence-based practices for this population.33 Concurrent to increasing prevention efforts among this group, it is important to closely monitor and screen American Indian youths for substance use. High rates of both lifetime exposure and current use by eighth grade place these youths at enhanced risk for development of substance use disorders as well as other substance-related problems.6,13,14 Of particular concern are alcohol and marijuana, as 4 in 10 students have used alcohol, nearly 1 in 4 students have gotten drunk, and more than 4 in 10 have used marijuana. These rates are 1.7 to 3.4 times higher than for MTF students. Given new national legal statutes regarding recreational use of marijuana, changing attitudes of youths toward perceived harmfulness of marijuana may be associated with these alarming rates of use. To date, legalization appears to be increasing marijuana use among young people already using, not among nonusers.34 There is also evidence that legalization is increasing use among heavy alcohol users.35 While the legal status of cannabis, both at the federal and state levels, remains in flux, some reservations are considering moves toward legalization.36

creasing marijuana use among young people already using, not among nonusers.34 There is also evidence that legalization is increasing use among heavy alcohol users.35 While the legal status of cannabis, both at the federal and state levels, remains in flux, some reservations are considering moves toward legalization.36 For American Indian youths, early initiation and the combined use of alcohol and marijuana were associated with increased risk for abuse or dependence from 2 to 5 times compared with those who use only 1 of these substances.9 Physicians and other medical staff treating American Indian youths (eg, Indian Health Service) need to be particularly alert to screen for emerging and established substance use, abuse, and dependence among both younger and older American Indian adolescents. In light of the general public’s lowering of perceived harm of marijuana use, a helpful resource for medical and other treating staff is the recently published volume from the National Academy of Sciences regarding health effects of cannabis and cannabinoids.37 This volume provides a comprehensive and fair presentation of both therapeutic and negative health effects of cannabis, along with the most currently available medical evidence.