Browse the corpus

f women, physicians, and facilities who are the subjects of this research. The Kaiser Permanente Washington breast imaging registry has received approval from the institutional review board for active or passive consenting processes or a waiver of consent to enroll participants, link data, and perform analytic studies. Women entered the cohort 6 months after completing their first risk factor questionnaire in the registry. To restrict to a cohort with negative screening findings, we excluded women diagnosed with in situ or invasive breast cancer within 6 months of their initial screening mammogram. All women aged 40 to 73 years at entry who attended at least 1 screening visit (baseline) with BI-RADS density recorded and did not have a prior diagnosis of invasive breast cancer or ductal carcinoma in situ were included; women with a lobular carcinoma in situ diagnosis before or at baseline were excluded. Some data from this cohort contributed to the BCSC model.13,14

Introduction Breast cancer is the most common cancer in women, with at least 1.7 million cases diagnosed and 0.5 million deaths per annum worldwide.1 Early diagnosis from mammography screening reduces breast cancer mortality by 20% to 40% in the general population.2,3 In women at an elevated risk of breast cancer, selective estrogen receptor modulator therapy for 5 years reduces the risk of breast cancer by about 40%, and the effect persists for at least 20 years.4 Risk-based breast cancer screening is not commonly adopted in the United States or elsewhere, but it has the potential to increase the benefits and decrease the harms of screening and increase the number of women eligible for preventive therapy. A prerequisite for the implementation of risk-adapted screening intervals and use of preventive therapies in precision medicine is accurate risk assessment.

elsewhere, but it has the potential to increase the benefits and decrease the harms of screening and increase the number of women eligible for preventive therapy. A prerequisite for the implementation of risk-adapted screening intervals and use of preventive therapies in precision medicine is accurate risk assessment. Breast cancer risk models have been used to guide entry criteria in prevention trials and to determine the eligibility of women for preventive therapy and supplemental screening by magnetic resonance imaging.5,6,7,8,9 The Tyrer-Cuzick model incorporates classic breast cancer risk factors, including information on affected second- and third-degree relatives, body mass index, menopause, and hormone therapy. However, the model identifies few women in the general population to be at high risk (taken to be an absolute 10-year risk of ≥8%).10,11 Accumulating evidence suggests that risk assessment may identify more high- and low-risk women when mammographic density is also taken into account,11,12,13,14,15 but the performance of the Tyrer-Cuzick model has not been directly assessed in any cohort study from a screening population in the United States.16,17

10,11 Accumulating evidence suggests that risk assessment may identify more high- and low-risk women when mammographic density is also taken into account,11,12,13,14,15 but the performance of the Tyrer-Cuzick model has not been directly assessed in any cohort study from a screening population in the United States.16,17 An important question for risk assessment is the follow-up time over which a model is accurate. Short-term predictions are useful for decisions such as additional screening modalities at the time of mammography, whereas longer-term risk predictions are important for deciding a risk-adapted screening regimen and eligibility for preventive therapy. Although several risk models provide the residual lifetime risk for a woman by year, studies to validate their performance have mostly considered cases within 5 years of risk assessment.18 The main aim of this study was to evaluate the Tyrer-Cuzick model to 19 years after risk assessment or 75 years of age in a screening cohort. We assessed the performance at different follow-up times and determined how much the accuracy of the model improves by adding a Breast Imaging and Reporting Data System (BI-RADS) measure of breast density.19

s to evaluate the Tyrer-Cuzick model to 19 years after risk assessment or 75 years of age in a screening cohort. We assessed the performance at different follow-up times and determined how much the accuracy of the model improves by adding a Breast Imaging and Reporting Data System (BI-RADS) measure of breast density.19 Methods Study Population The study included women in the Kaiser Permanente Washington Breast Cancer Surveillance Consortium (BCSC) registry who attended mammography screening from January 1, 1996, through December 31, 2013, with follow-up to December 31, 2014.20,21,22 This registry was used because of the more detailed family history of breast and ovarian cancer collected than for other BCSC registries.23 All procedures are compliant with the Health Insurance Portability and Accountability Act, and the registry has received a Federal Certificate of Confidentiality and other protection for the identities of women, physicians, and facilities who are the subjects of this research. The Kaiser Permanente Washington breast imaging registry has received approval from the institutional review board for active or passive consenting processes or a waiver of consent to enroll participants, link data, and perform analytic studies.

at least 1 screening visit (baseline) with BI-RADS density recorded and did not have a prior diagnosis of invasive breast cancer or ductal carcinoma in situ were included; women with a lobular carcinoma in situ diagnosis before or at baseline were excluded. Some data from this cohort contributed to the BCSC model.13,14 End Points The primary outcome was the time from 6 months after the entry questionnaire to diagnosis of invasive breast cancer or censoring. Women were censored at the earliest of death, health plan disenrollment, diagnosis of ductal carcinoma in situ, 75 years of age (the recommended end of screening), or the end of calendar time follow-up. Outcomes were obtained through linkage with the regional population-based Surveillance, Epidemiology, and End Results tumor registry24 and pathology databases. All benign and malignant breast pathologic findings at Kaiser Permanente Washington were collected from electronic medical records, given manual, standardized codes,25,26 and used to supplement tumor registry data.

al population-based Surveillance, Epidemiology, and End Results tumor registry24 and pathology databases. All benign and malignant breast pathologic findings at Kaiser Permanente Washington were collected from electronic medical records, given manual, standardized codes,25,26 and used to supplement tumor registry data. Exposure Variables Risk factors used in the Tyrer-Cuzick model (version 7.02) were investigated, and mammographic density risk was integrated using risk estimates from a different case-control study (A.R.B., Wendy F. Cohn, PhD, William A. Knaus, PhD, Martin J. Yaffe, PhD, J.C., and Jennifer A. Harvey, MD; unpublished data; December 2017) (eMethods in the Supplement). Risk factors were collected prospectively using a self-report form taken at the same time as the mammogram. Only screening mammograms using the radiologist’s indication for the examination were used. Risk factors included (1) first- and second-degree family history of breast cancer, including male relatives, and age affected (<50 or ≥50 years or unknown); (2) first-degree family history of ovarian cancer and age affected (<45 or ≥45 years or unknown); (3) age; (4) weight; (5) height; (6) parity, age at first child, or unknown; (7) premenopausal, perimenopausal, or postmenopausal status, age at menopause (<30, 30-39, 40-49 [or 40-44 or 45-49, with age categories collected changing over time], 50-54 or ≥55 years or unknown); (8) age at menarche (<11, 12, 13, 14, or ≥15 years or unknown); (9) benign breast disease, including number of biopsies, prior hyperplasia of the usual type (yes/no), or atypical hyperplasia (yes/no); (10) ovarian cancer, age at diagnosis (<45, 45-49, 50-54, or ≥55 years or unknown); and (11) BI-RADS breast density (almost entirely fat, scattered fibroglandular, heterogeneously dense, or extremely dense) reported by the interpreting radiologist. Body mass index was calculated using self-reported weight in kilograms divided the height in meters squared. Age categories were converted to single year for input into the Tyrer-Cuzick model by taking the midpoint or using the rules in eTable 1 in the Supplement for affected relatives. Questionnaire data were briefly reviewed by the mammogram technologist at the time of the examination and were checked for invalid values when they were scanned for research.

single year for input into the Tyrer-Cuzick model by taking the midpoint or using the rules in eTable 1 in the Supplement for affected relatives. Questionnaire data were briefly reviewed by the mammogram technologist at the time of the examination and were checked for invalid values when they were scanned for research. Approximately 5% of women undergoing screening opted out of having their questionnaire data used for research.25 Demographic factors included urban environment (metropolitan, micropolitan, small town, rural, or unknown) and a geocoded measure of median income determined by linking a woman’s address at the time of each questionnaire to income and urban environmental data from her 2010 US census tract.27

e data used for research.25 Demographic factors included urban environment (metropolitan, micropolitan, small town, rural, or unknown) and a geocoded measure of median income determined by linking a woman’s address at the time of each questionnaire to income and urban environmental data from her 2010 US census tract.27 Statistical Methods Data were analyzed from March 2, 2016, through November 13, 2017. Categories for demographic and risk factor categories were chosen based on questionnaire fields or established cut points, and their hazard ratios (HRs) were estimated with 95% Wald CIs. Breast cancer incidence was predicted to the end of each woman’s follow-up, converted to a cumulative hazard using a natural logarithm, and summated to provide the expected number of breast cancer diagnoses. Exact 95% CIs for the observed divided by the expected (O/E) numbers of cancer diagnoses assumed that the observed number was generated by a Poisson distribution, with a rate equal to the expected number if well calibrated. Population risks were shown using annual incidence rates (IRs) per 1000 women for the complete cohort and across 10-year risk subgroups using categories defined at baseline to be below average (<2%), average (2% to <3%), above average (3% to <5%), moderately increased (5% to <8%) and high (≥8%), following clinical guidelines in the United Kingdom.11 The top and bottom deciles of 10-year risk, relative to the middle 80%, were compared using Kaplan-Meier estimation and HRs and in a sensitivity analysis of the high-risk quantile. A proportional hazards model with a time-dependent covariate equal to the yearly predicted hazard rate with adjustment by 5-year age group was used to determine the calibration of relative risks overall and for each year of follow-up and visualized by a spline fitted to weighted Schoenfeld residuals.28,29 To quantify information conferred by risk models beyond age, we calculated the difference in likelihood ratio (LR) statistics between a proportional hazards model that included only age and one that additionally incorporated the yearly predicted hazard. All analysis was undertaken using statistical software R (version 3.4.1) and the survival, survminer, amd mgcv packages.29,30,31,32

alculated the difference in likelihood ratio (LR) statistics between a proportional hazards model that included only age and one that additionally incorporated the yearly predicted hazard. All analysis was undertaken using statistical software R (version 3.4.1) and the survival, survminer, amd mgcv packages.29,30,31,32 Results Cohort We included 132 139 women with a median follow-up of 5.2 years (interquartile range [IQR], 2.4-11.1 years). Follow-up was greater for younger women who entered the cohort earlier (eg, median of 10.8 years [IQR, 3.8-17.2 years] for 46 436 women younger than 60 years with entry before 2000). Most of the women were white (80.4%) and lived in a metropolitan area (95.4%) (Table 1). Median body mass index at baseline was 26.6 (IQR, 23.1-31.5).

greater for younger women who entered the cohort earlier (eg, median of 10.8 years [IQR, 3.8-17.2 years] for 46 436 women younger than 60 years with entry before 2000). Most of the women were white (80.4%) and lived in a metropolitan area (95.4%) (Table 1). Median body mass index at baseline was 26.6 (IQR, 23.1-31.5). Table 1. Invasive Breast Cancer Rate by Demographic and Other Factors Characteristic No. (%) of Womena Follow-up, 1000 Women-years No. of Invasive Cancer Cases IR per 1000 Women/y Age-Adjusted HR (95% CI) LR-χ2 Test P Value Overall 132 139 (100) 939 2699 2.9 Race White 106 191 (80.4) 778 2340 3.0 1 [Reference] 9.2b .06 Asian 11 690 (8.8) 70 152 2.2 0.80 (0.68-0.94) Black 5133 (3.9) 34 81 2.4 0.87 (0.70-1.08) >1 Race 3622 (2.7) 24 67 2.8 1.03 (0.81-1.32) Other 3470 (2.6) 22 59 2.7 1.01 (0.78-1.31) Unknown 2033 (1.5) 12 0 NA NA Ethnicity Non-Hispanic 123 750 (93.7) 882 2544 2.9 1 [Reference] 5.2c .02 Hispanic 6546 (5.0) 46 153 3.4 1.22 (1.03-1.43) Unknown Hispanic 1843 (1.4) 11 2 0.2 NA Urban environment Metropolitan 126 121 (95.4) 902 2632 2.9 1 [Reference] 18.9d <.001 Micropolitan 4073 (3.1) 27 48 1.8 0.58 (0.44-0.77) Small town 771 (0.6) 5 10 2.1 0.67 (0.36-1.25) Rural 519 (0.4) 3 5 1.5 0.46 (0.19-1.12) Unknown 655 (0.5) 2 4 1.8 0.79 (0.30-2.11) Income quartile (upper limit, $)e 1 ($68 005) 29 977 (22.7) 202 521 2.6 1 [Reference] 4.7d .03 2 ($79 932) 34 527 (26.1) 239 692 2.9 1.14 (1.02-1.28) 3 ($100 313) 32 667 (24.7) 234 657 2.8 1.11 (0.99-1.24) 4 (>$100 313) 32 571 (24.6) 248 771 3.1 1.18 (1.06-1.32) Unknown 2397 (1.8) 16 58 3.7 1.50 (1.14-1.97) Time to next screen (range), y >0.5 to 1.5 27 361 (20.7) 192 833 4.3 1.61 (1.48-1.75) 137.7d <.001 >1.5 to 2.5 54 811 (41.5) 498 1437 2.9 1 [Reference] >2.5 to 3.5 10 283 (7.8) 89 218 2.4 0.93 (0.81-1.07) >3.5 10 430 (7.9) 91 179 2.0 0.76 (0.65-0.89) Baseline only 29 254 (22.1) 69 32 0.5 0.24 (0.17-0.35) Abbreviations: HR, hazard ratio; IR, incidence rate; LR-χ2, age-adjusted likelihood ratio χ2 statistics, excluding unknown groups.

1 (41.5) 498 1437 2.9 1 [Reference] >2.5 to 3.5 10 283 (7.8) 89 218 2.4 0.93 (0.81-1.07) >3.5 10 430 (7.9) 91 179 2.0 0.76 (0.65-0.89) Baseline only 29 254 (22.1) 69 32 0.5 0.24 (0.17-0.35) Abbreviations: HR, hazard ratio; IR, incidence rate; LR-χ2, age-adjusted likelihood ratio χ2 statistics, excluding unknown groups. a Percentages have been rounded and may not total 100. b Calculated as test of heterogeneity (df, 4). c Calculated as test of heterogeneity (df, 1). d Calculated as test for trend (df, 1). e Indicates the upper limit median family income from census data. Median age at entry was 50 years (IQR, 44-58 years). Two peaks in the entry distribution occurred at 40 and 50 years of age. These peaks reflect the cohort’s risk-based screening program21 in which high-risk women were recommended to start annual breast imaging at 40 years of age and low-risk women were recommended to start at 50 years of age (eFigure 1 in the Supplement). Most women had a second screen within 2 years of entry (82 172 [62.2%]) (Table 1), and 29 254 (22.1%) had a single baseline screening examination.

-risk women were recommended to start annual breast imaging at 40 years of age and low-risk women were recommended to start at 50 years of age (eFigure 1 in the Supplement). Most women had a second screen within 2 years of entry (82 172 [62.2%]) (Table 1), and 29 254 (22.1%) had a single baseline screening examination. In total, 2699 invasive breast cancers were diagnosed. Women were censored owing to disenrollment (62 331 [47.2%]), end of follow-up (48 317 [36.6%]), being 75 years of age (15 827 [12.0%]), death (2328 [1.8%]), or a diagnosis of ductal carcinoma in situ (637 [0.5%]). Of the 2699 invasive cancers, 412 were larger than 2 cm and had lymph node involvement (178 were of unknown size and/or nodal status). Invasive cancer rates increased from approximately 1.3 per 1000 women/y at 42 years of age to 5.1 per 1000 women/y at 70 years of age and were similar to recent rates in Washington State but departed from the marginal rate assumption in the Tyrer-Cuzick model (eFigure 2 in the Supplement).

wn size and/or nodal status). Invasive cancer rates increased from approximately 1.3 per 1000 women/y at 42 years of age to 5.1 per 1000 women/y at 70 years of age and were similar to recent rates in Washington State but departed from the marginal rate assumption in the Tyrer-Cuzick model (eFigure 2 in the Supplement). Risk factor HRs were in the expected direction (Table 2). Breast density was the strongest factor after age and had an approximate 4-fold difference between the most and least dense BI-RADS categories after adjustment for age and body mass index (2.21 [95% CI, 1.95-2.50] vs 0.55 [95% CI, 0.45-0.68]). In a multivariable analysis using risk factors included in the Gail model,5 most information was in age (LR-χ21 = 308.5; HR per 5 years, 1.24; 95% CI, 1.21-1.27), affected first-degree relatives (LR-χ22 = 125.3; HR for 1 vs none, 1.68; 95% CI, 1.53-1.85; HR for 2 vs none, 2.04; 95% CI, 1.54-2.61), and previous atypical hyperplasia diagnosis (LR-χ21 = 78.4; HR, 3.14; 95% CI, 2.34-4.23) (eTable 2 in the Supplement).

in age (LR-χ21 = 308.5; HR per 5 years, 1.24; 95% CI, 1.21-1.27), affected first-degree relatives (LR-χ22 = 125.3; HR for 1 vs none, 1.68; 95% CI, 1.53-1.85; HR for 2 vs none, 2.04; 95% CI, 1.54-2.61), and previous atypical hyperplasia diagnosis (LR-χ21 = 78.4; HR, 3.14; 95% CI, 2.34-4.23) (eTable 2 in the Supplement). Table 2. Invasive Breast Cancer Rate and HRs by Risk Factor at Baseline Risk Factor No. (%) of Womena Follow-up, 1000 Women-years No. of Invasive Cancer Cases IR per 1000 Women/y Age-Adjusted HR (95% CI) Trend Test LR-χ21 P Value Age at birth of first child, y Nulliparous 26 334 (19.9) 193 534 2.8 1 [Reference] 12.4 <.001 <20 20 014 (15.3) 134 362 2.7 0.83 (0.72-0.95) 20-24 37 718 (28.5) 265 811 3.1 0.93 (0.83-1.04) 25-29 24 336 (18.4) 177 514 2.9 0.97 (0.86-1.09) 30-34 12 846 (9.7) 98 278 2.8 1.07 (0.92-1.23) 35-39 5156 (3.9) 37 102 2.7 1.08 (0.88-1.34) ≥40 967 (0.7) 7 23 3.4 1.27 (0.83-1.92) Unknown 4768 (3.6) 27 75 2.7 0.89 (0.70-1.13) Age at menarche, y <11 3090 (2.3) 14 23 1.6 0.70 (0.45-1.09) 3.7 .055 11 7560 (5.7) 34 76 2.2 0.96 (0.73-1.27) 12 14 936 (11.3) 69 156 2.3 1 [Reference] 13 14 948 (11.3) 69 141 2.1 0.90 (0.71-1.13) 14 7146 (5.4) 32 55 1.7 0.75 (0.55-1.02) ≥15 7349 (5.8) 32 50 1.6 0.68 (0.49-0.93) Not asked 72 115 (54.6) 663 2154 3.3 1.09 (0.93-1.29) Asked but unknown 4995 (3.8) 25 44 1.8 0.73 (0.52-1.02) No. of affected first-degree relatives 0 113 685 (86.0) 810 2104 2.6 1 [Reference] 128.2 <.001 1 16 761 (12.7) 118 532 4.5 1.71 (1.55-1.88) ≥2 1693 (1.3) 11 63 5.8 2.04 (1.58-2.62) Age at menopause, y <30 3274 (2.5) 22 36 1.7 0.59 (0.42-0.83) 30.5 <.001 30-39 10 791 (8.2) 76 177 2.3 0.77 (0.65-0.91) 40-49 26 110 (19.8) 181 583 3.2 1 [Reference] 50-54 19 640 (14.9) 128 508 4.0 1.12 (0.99-1.26) ≥55 4776 (3.6) 28 134 4.8 1.24 (1.02-1.50) Premenopausal 51 891 (39.3) 389 905 2.3 1.03 (0.91-1.18) Unknown 15 657 (11.8) 115 356 3.1 1.05 (0.92-1.21) No.

(0.42-0.83) 30.5 <.001 30-39 10 791 (8.2) 76 177 2.3 0.77 (0.65-0.91) 40-49 26 110 (19.8) 181 583 3.2 1 [Reference] 50-54 19 640 (14.9) 128 508 4.0 1.12 (0.99-1.26) ≥55 4776 (3.6) 28 134 4.8 1.24 (1.02-1.50) Premenopausal 51 891 (39.3) 389 905 2.3 1.03 (0.91-1.18) Unknown 15 657 (11.8) 115 356 3.1 1.05 (0.92-1.21) No. of previous breast biopsiesb 0 123 370 (93.4) 856 2337 2.7 1 [Reference] 48.8 <.001 1 7213 (5.5) 67 284 4.3 1.56 (1.38-1.76) 2 1250 (0.9) 12 66 5.4 1.97 (1.54-2.52) ≥3 306 (0.2) 3 12 3.7 1.37 (0.78-2.41) Benign disease (highest grade) No biopsy 123 370 (93.4) 856 2337 2.7 1 [Reference] 104.5 <.001 Biopsy 6093 (4.6) 58 204 3.5 1.32 (1.14-1.52) Hyperplasia of usual type 2189 (1.7) 20 101 5.0 1.77 (1.45-2.17) Atypical hyperplasia 487 (0.4) 4 57 13.1 4.50 (3.46-5.85) Premenopausal BMI <20 2992 (5.8) 22 60 2.7 1.20 (0.92-1.58) 6.5 .01 20 to <25 19 756 (38.1) 153 355 2.3 1 [Reference] 25 to <30 13 487 (26.0) 100 254 2.5 1.08 (0.92-1.27) 30 to <35 7187 (13.9) 53 112 2.1 0.90 (0.73-1.11) ≥35 6728 (13.0) 48 89 1.9 0.81 (0.64-1.02) Unknown 1741 (3.4) 13 35 2.6 1.06 (0.75-1.50) Postmenopausal BMI <20 2688 (4.2) 18 42 2.3 0.79 (0.58-1.09) 6.0 .01 20 to <25 19 559 (30.3) 133 400 3.0 1 [Reference] 25 to <30 19 245 (29.8) 129 476 3.7 1.22 (1.07-1.39) 30 to <35 11 217 (17.4) 75 254 3.4 1.14 (0.98-1.34) ≥35 9457 (14.6) 64 214 3.3 1.19 (1.00-1.40) Unknown 2425 (3.8) 15 52 3.5 1.11 (0.83-1.49) Height, m <1.57 17 807 (13.5) 119 337 2.8 1.02 (0.91-1.15) 17.4 <.001 1.57-1.67 67 033 (50.7) 477 1305 2.7 1 [Reference] ≥1.67 43 685 (33.1) 321 989 3.1 1.18 (1.09-1.29) Unknown 3614 (2.7) 22 68 3.0 1.07 (0.84-1.37) BI-RADS densityc Fatty 10 138 (7.7) 65 100 1.5 0.55 (0.45-0.68) 191.4 <.001 Scattered 47 125 (35.7) 339 814 2.4 1 [Reference] Heterogeneous 55 943 (42.3) 396 1295 3.3 1.69 (1.54-1.85) Dense 18 933 (14.3) 139 490 3.5 2.21 (1.95-2.50) Abbreviations: BI-RADS, Breast Imaging and Reporting Data System; BMI, body mass index (calculated as weight in kilograms divided by the height in meters squared); heterogeneity test; HR, hazard ratio; IR, incidence rate; LR-χ2, age-adjusted likelihood ratio χ2 statistics, excluding unknown groups.

3.5 2.21 (1.95-2.50) Abbreviations: BI-RADS, Breast Imaging and Reporting Data System; BMI, body mass index (calculated as weight in kilograms divided by the height in meters squared); heterogeneity test; HR, hazard ratio; IR, incidence rate; LR-χ2, age-adjusted likelihood ratio χ2 statistics, excluding unknown groups. a Percentages have been rounded and may not total 100. b No unknown category was used; if none reported, number is 0. c Also adjusted for BMI owing to strong negative association. Evaluation of Risk Models We found good calibration of absolute risk during the entire follow-up period (O/E for the Tyrer-Cuzick model, 1.02 [95% CI, 0.98-1.06]; O/E for Tyrer-Cuzick model with density, 0.98 [95% CI, 0.94-1.02]) (Table 3). Absolute risk calibration varied by age at entry (eTable 3 in the Supplement), by which the general tendency was to predict relatively more cancers than observed in younger women but fewer in older women (O/E for Tyrer-Cuzick model in women aged 40-49 years, 0.81 [95% CI, 0.76-0.86]; in women aged 60-73 years, 1.18 [95% CI, 1.09-1.27]).

bration varied by age at entry (eTable 3 in the Supplement), by which the general tendency was to predict relatively more cancers than observed in younger women but fewer in older women (O/E for Tyrer-Cuzick model in women aged 40-49 years, 0.81 [95% CI, 0.76-0.86]; in women aged 60-73 years, 1.18 [95% CI, 1.09-1.27]). Table 3. Absolute Risk Calibration by Model and 10-Year Risk Subgroup Model by 10-y Risk No. (%) of Womena Follow-up, 1000 Women-years No. of Invasive Breast Cancer Cases O/E (95% CI) IR per 1000 Women/y IRR (95% CI) Observed Expected Observed Expected Tyrer-Cuzick All 132 139 (100) 939 2699 2645 1.02 (0.98-1.06) 2.9 2.8 NA <2% 47 975 (36.3) 347 648 533 1.22 (1.12-1.31) 1.9 1.5 0.73 (0.66-0.81) 2% to <3% 42 700 (32.3) 311 792 782 1.01 (0.94-1.09) 2.5 2.5 1 [Reference] 3% to <5% 29 523 (22.3) 202 779 763 1.02 (0.95-1.10) 3.9 3.8 1.52 (1.37-1.67) 5% to <8% 9387 (7.1) 62 333 382 0.87 (0.78-0.97) 5.4 6.2 2.12 (1.86-2.40) ≥8% 2554 (1.9) 17 147 185 0.79 (0.67-0.93) 8.7 11.0 3.43 (2.87-4.08) Tyrer-Cuzick with density All 132 139 (100) 939 2699 2757 0.98 (0.94-1.02) 2.9 2.9 NA <2% 53 436 (40.4) 390 641 548 1.17 (1.08-1.26) 1.6 1.4 0.63 (0.56-0.70) 2% to <3% 33 269 (25.2) 240 627 603 1.04 (0.96-1.12) 2.6 2.5 1 [Reference] 3% to <5% 29 477 (22.3) 203 779 784 0.99 (0.93-1.07) 3.8 3.9 1.47 (1.32-1.63) 5% to <8% 11 312 (8.6) 767 379 473 0.80 (0.72-0.89) 5.0 6.2 1.92 (1.69-2.18) ≥8% 4645 (3.5) 30 273 349 0.78 (0.69-0.88) 9.2 11.7 3.52 (3.05-4.05) Abbreviations: IR, incidence rate; IRR, IR ratio; NA, not applicable; O/E, observed divided by expected cases.

77 (22.3) 203 779 784 0.99 (0.93-1.07) 3.8 3.9 1.47 (1.32-1.63) 5% to <8% 11 312 (8.6) 767 379 473 0.80 (0.72-0.89) 5.0 6.2 1.92 (1.69-2.18) ≥8% 4645 (3.5) 30 273 349 0.78 (0.69-0.88) 9.2 11.7 3.52 (3.05-4.05) Abbreviations: IR, incidence rate; IRR, IR ratio; NA, not applicable; O/E, observed divided by expected cases. a Percentages have been rounded and may not total 100. Figure 1 shows continued separation for baseline risk groups in estimated cumulative risk curves through 19 years after risk assessment, where the end of the curves represent proportionally more younger women at entry owing to censoring at 75 years of age. The Tyrer-Cuzick model identified 10-year risk in 2554 women (1.9%) to be 8% or greater, in whom 147 cancers (5.4%; IR per 1000 women, 8.7) were subsequently diagnosed as invasive breast cancer. The Tyrer-Cuzick model with density identified more women (4645 [3.5%]; 273 cancers [10.1%]; IR per 1000 women, 9.2). However, risk was overestimated in this group (O/E for the Tyrer-Cuzick model, 0.78 [95% CI, 0.69-0.88]; O/E for the Tyrer-Cuzick model with density, 0.79 [95% CI, 0.67-0.93]), and risk was underestimated in the group with a 10-year risk of less than 2% (O/E for the Tyrer-Cuzick model, 1.22 [95% CI, 1.12-1.31]; O/E for the Tyrer-Cuzick model with density, 1.17 [95% CI, 1.08-1.26]).

-Cuzick model, 0.78 [95% CI, 0.69-0.88]; O/E for the Tyrer-Cuzick model with density, 0.79 [95% CI, 0.67-0.93]), and risk was underestimated in the group with a 10-year risk of less than 2% (O/E for the Tyrer-Cuzick model, 1.22 [95% CI, 1.12-1.31]; O/E for the Tyrer-Cuzick model with density, 1.17 [95% CI, 1.08-1.26]). Figure 1. Observed Cumulative Invasive Breast Cancer Risk by 10-Year Risk Group The risk groups are from the 10-year risk assessment. The width of the fan represents a pointwise 95% CI. At 10 years the observed risk for the Tyrer-Cuzick model and the Tyrer-Cuzick model with density was 1.8% and 1.6%, respectively, for the group with predicted risk of less than 2%; 2.6% and 2.6%, respectively, for predicted risk of 2% to less than 3%; 4.1% and 3.8%, respectively, for predicted risk of 3% to less than 5%; 5.5% and 5.4%, respectively, for predicted risk of 5% to less than 8%; and 8.2% and 9.0%, respectively, for predicted risk of 8% or greater.

with predicted risk of less than 2%; 2.6% and 2.6%, respectively, for predicted risk of 2% to less than 3%; 4.1% and 3.8%, respectively, for predicted risk of 3% to less than 5%; 5.5% and 5.4%, respectively, for predicted risk of 5% to less than 8%; and 8.2% and 9.0%, respectively, for predicted risk of 8% or greater. Overestimation of the highest decile relative to the middle 80% was also apparent (Figure 2). The hazard ratio for the top decile was 2.22 (95% CI, 2.02-2.45) for the Tyrer-Cuzick model compared with 2.55 (95% CI, 2.33-2.80) for the Tyrer-Cuzick model with density, and the results were robust to choice of upper quantile (eFigure 4 in the Supplement). The hazard ratio for the bottom decile was 0.50 (95% CI, 0.42-0.61) for the Tyrer-Cuzick model but 0.36 (95% CI, 0.29-0.45) for the Tyrer-Cuzick model with density. Incorporating density in the model provided a greater range of observed risk between the top and bottom deciles (Figure 2). The Tyrer-Cuzick model also overestimated relative risks after allowing for age (0.67; 95% CI, 0.60-0.75) (eTable 4 in the Supplement) but showed little evidence of a change in relative risk calibration during follow-up for the Tyrer-Cuzick model (age-adjusted intercept, 0.69 [95% CI, 0.58-0.81]; age-adjusted slope, −0.003 [95% CI, −0.018 to 0.012]) and the Tyrer-Cuzick model with density (age-adjusted intercept, 0.78 [95% CI, 0.68-0.88]; age-adjusted slope, −0.008 [95% CI, −0.020 to 0.004]) (eTable 4 and eFigure 5 in the Supplement).

for the Tyrer-Cuzick model (age-adjusted intercept, 0.69 [95% CI, 0.58-0.81]; age-adjusted slope, −0.003 [95% CI, −0.018 to 0.012]) and the Tyrer-Cuzick model with density (age-adjusted intercept, 0.78 [95% CI, 0.68-0.88]; age-adjusted slope, −0.008 [95% CI, −0.020 to 0.004]) (eTable 4 and eFigure 5 in the Supplement). Figure 2. Observed and Expected Cumulative Invasive Breast Cancer Risk by Quantile The risk groups are from the predicted 10-year risk assessment (lowest decile, middle 80% and top decile). Solid lines indicate observed risk; broken lines, expected risk. At 10 years the observed risk for the Tyrer-Cuzick model and the Tyrer-Cuzick model with density was 1.4% and 1.0%, respectively, for the bottom decile of risk; 2.7% and 2.6%, respectively, for the middle 80% of risk; and 5.9% and 7.0%, respectively, for the top decile of risk. Other analyses showed more predictive information than age in the models (Tyrer-Cuzick model, ΔLR-χ2 = 290.5; Tyrer-Cuzick model with density, ΔLR-χ2 = 541.4); density added approximately 86% to all factors in the Tyrer-Cuzick model other than age. A reclassification matrix demonstrated that this information translated into improved risk stratification for individual women (eTable 5 in the Supplement).

k model, ΔLR-χ2 = 290.5; Tyrer-Cuzick model with density, ΔLR-χ2 = 541.4); density added approximately 86% to all factors in the Tyrer-Cuzick model other than age. A reclassification matrix demonstrated that this information translated into improved risk stratification for individual women (eTable 5 in the Supplement). Discussion In this article, we evaluated the accuracy of long-term breast cancer risk assessment in a US screening cohort and found that breast cancer risk models based on classic risk factors and mammographic density remain accurate during a longer period than considered to date. We found continued differences in observed risk during a 19-year period between predicted risk strata formed at baseline (Figure 1).

nt in a US screening cohort and found that breast cancer risk models based on classic risk factors and mammographic density remain accurate during a longer period than considered to date. We found continued differences in observed risk during a 19-year period between predicted risk strata formed at baseline (Figure 1). The long-term calibration of breast cancer risk models has important clinical implications. Arguably the main role of breast cancer risk assessment to date has been to triage women for genetic counseling and thereby guide their eligibility for genetic testing, preventive therapy, and screening modalities in addition to mammography. Our results lend support to extending such triage to more general high-risk clinics based on risk to 19 years using a combined risk assessment, not just familial risk associated with BRCA1/2 mutations or other inherited genetic factors. Combining mammographic density with classic risk factors appears to be particularly important for this aim because the strategy almost doubled the number identified in a high-risk group. Genetic or high-risk clinics may only have a moderate effect on breast cancer in the general population because most breast cancers are attributable to nongenetic factors. Most of these reproductive, hormonal, lifestyle, or other factors are common but relatively weak factors for individual women (Table 2), and few women at a very high long-term risk are identified by them. However, accurate risk assessment can also play a role for women not included in a high-risk group by helping to personalize risk-adapted screening strategies.

yle, or other factors are common but relatively weak factors for individual women (Table 2), and few women at a very high long-term risk are identified by them. However, accurate risk assessment can also play a role for women not included in a high-risk group by helping to personalize risk-adapted screening strategies. Implementing risk assessment and prevention strategies that are effective during a longer period could be easier than more frequent risk assessment, but updated risk assessments are likely to also play a role because they will be more precise for individual women. For example, risk would increase after a first diagnosis of proliferative benign disease, and taking into account a sequence of mammographic density measurements will be more precise than only using the most recent measurement.33

nts are likely to also play a role because they will be more precise for individual women. For example, risk would increase after a first diagnosis of proliferative benign disease, and taking into account a sequence of mammographic density measurements will be more precise than only using the most recent measurement.33 The overall rates predicted by the models were broadly consistent with the observed rates, but some evidence suggested overestimation for the women at highest risk, and the models also predicted relatively more cancers than observed in younger women and fewer in older women. Although changing risk thresholds or applying a recalibration of the absolute risks may be considered, this is unnecessary because the aim is to use the risk model to form broad risk strata. For example, the observed risks in the chosen groups were consistent with the predicted 10-year risk (Figure 1). Another issue is that although assessment of absolute risk calibration is important, it is not straightforward to evaluate absolute risk in screening cohorts, in part because the evaluation is affected by the process of screening. Risk models are calibrated to breast cancer rates in the population, not just participants who attend screening or who have had a negative finding. Thus, one might expect observed rates to be higher than predicted by the risk models. However, the analytic approach of removing cancers detected at the first screening initially makes incidence lower than that in the population owing to the removal of a pool of cancers and the time taken for new cancers to develop. This aspect is reflected in the age-specific rates for women aged 41 years in eFigure 2 in the Supplement. In line with both these points, absolute risk calibration varied by age group, with relatively more cancers predicted than observed for women in their 40s. Half of the cohort entered when in the 40-year age group, and a reduction in the rates relative to the general population conferred by a negative finding on initial screening would lead to fewer than expected cancers. Women in their 70s had relatively fewer cancers predicted than observed, which is likely owing to screening being recommended to 75 years of age in this cohort. The Tyrer-Cuzick model background rates are based on a UK sample in which population screening ended at 70 years of age (eFigure 2B in the Supplement).

ancers. Women in their 70s had relatively fewer cancers predicted than observed, which is likely owing to screening being recommended to 75 years of age in this cohort. The Tyrer-Cuzick model background rates are based on a UK sample in which population screening ended at 70 years of age (eFigure 2B in the Supplement). Validated and freely available models for invasive breast cancer have merit for guiding personalized breast cancer screening and prevention strategies,34,35,36 but models for subtypes could also play a role in decision making. For instance, it has long been considered likely that mammographic screening in women younger than 50 years should be more frequent than in older women despite their lower risk because on average tumor progression is more rapid in the younger group and the breast tissue is denser; one might seek to use models that assess risk of aggressive or lethal types of cancer and a false-negative mammography screening result.

50 years should be more frequent than in older women despite their lower risk because on average tumor progression is more rapid in the younger group and the breast tissue is denser; one might seek to use models that assess risk of aggressive or lethal types of cancer and a false-negative mammography screening result. Limitations This study has several limitations. Results are derived from a single registry in 1 area of the United States with one of the highest IRs for breast cancer in the nation37 and an active risk-based screening program, which will provide information on younger women who are at higher risk. All women had health insurance, the median census family income was relatively high, and the cohort mainly consisted of women who regularly attend screening, which might represent more healthy individuals. The relative homogeneity of the sample has the potential to limit the factors that influence the model and mask the influence of socioeconomic or other risk factors. Some data were missing (Table 2), for which the Tyrer-Cuzick model in general assumes the population risk (relative risk of 1.00). Missing risk factor data could reduce the predictive ability but were uncommon (Table 2). Finally, follow-up was only during enrollment in the health plan. Although specific information about reasons for health plan disenrollment were not collected, in general they were owing to an employer no longer offering the health plan, choosing a different option during annual open enrollment, or a new job.

Table 2). Finally, follow-up was only during enrollment in the health plan. Although specific information about reasons for health plan disenrollment were not collected, in general they were owing to an employer no longer offering the health plan, choosing a different option during annual open enrollment, or a new job. Conclusions Risk models combining classic risk factors with mammographic density were informative to 19 years after risk assessment. Mammographic density helped to identify a greater number of women at the extremes of the risk distribution where preventive measures or different screening intervals might be considered to minimize intervention-associated harms and the public health burden of breast cancer. Supplement. eMethods. Tyrer Cuzick Models eTable 1. Age of Affected Relative Used for Input to the Tyrer-Cuzick Model for Breast Cancer– or Ovarian Cancer–Affected Relatives eTable 2. Estimated Multivariate Hazard Ratios for Gail Model Factors From This Cohort and Analysis of Deviance Results eTable 3. Calibration of Models in 3 Age Groups by 10-Year Risk Group at Entry eTable 4. Calibration of the Relative Risks From the Models After Accounting for Age-Specific Baseline Hazard Functions in 5-Year Groups eTable 5. Reclassification Matrix for 10-Year Risk Groups in the Tyrer-Cuzick Model and the Tyrer-Cuzick Model With Mammographic Density eFigure 1. Some Characteristics of the Cohort eFigure 2. Age-Specific Rates eFigure 3. Observed Cumulative Risk by Quantile Risk Group and Age Group eFigure 4. Further Comparison of Observed Risks by Decile

eTable 5. Reclassification Matrix for 10-Year Risk Groups in the Tyrer-Cuzick Model and the Tyrer-Cuzick Model With Mammographic Density eFigure 1. Some Characteristics of the Cohort eFigure 2. Age-Specific Rates eFigure 3. Observed Cumulative Risk by Quantile Risk Group and Age Group eFigure 4. Further Comparison of Observed Risks by Decile eFigure 5. Calibration of Relative Risks After Allowing For Recalibration of Age-Specific Rates Click here for additional data file.

Introduction Multiple myeloma is a clonal neoplasm of differentiated B cells (plasma cells). It is one of the most common hematologic malignant neoplasms among adults, affecting approximately 100 000 persons currently living with the disease in the United States, and has an age-adjusted incidence rate of 6.53 per 100 000 per year.1,2,3 Multiple myeloma is most frequently diagnosed among people aged 65 to 74 years; only approximately 5% of cases are diagnosed before 50 years.2 In the general population, approximately 80% of patients with multiple myeloma have expression of IgH (referred to as “monoclonal-(M)-protein”) and 20% have abnormal light-chain proteins detectable in peripheral blood.4 Evidence from a large, prospective, population-based cancer screening trial shows that IgH-secreting and light-chain–secreting multiple myeloma are consistently preceded by their respective precursor states, monoclonal gammopathy of undetermined significance (MGUS) and light-chain MGUS, which can be detected in peripheral blood.5

om a large, prospective, population-based cancer screening trial shows that IgH-secreting and light-chain–secreting multiple myeloma are consistently preceded by their respective precursor states, monoclonal gammopathy of undetermined significance (MGUS) and light-chain MGUS, which can be detected in peripheral blood.5 Although the cause of multiple myeloma and its precursor conditions (MGUS and light-chain MGUS) remains largely unclear, previous studies have reported an increased risk among individuals exposed to known and suspected carcinogens including polychlorinated biphenyl (PCB), dioxins, polycyclic aromatic hydrocarbons (PAHs), and asbestos.6,7,8 The attacks on the World Trade Center (WTC) on September 11, 2001 (9/11), created an unprecedented environmental exposure to aerosolized dust and gases that contained known carcinogens including PCBs and PAHs.9 These substances were produced by the collapsed and burning buildings and by the diesel smoke emitted from heavy equipment used during the 10-month rescue and recovery effort. Recent cohort studies of first responders, construction workers, and volunteers at the WTC site provided evidence linking exposure to the WTC aerosolized dust and gases with multiple myeloma and other malignant neoplasms. In 2009, a case series (N = 8) suggested an excess of early-onset of multiple myeloma among WTC-exposed first responders in the General Responder Cohort; 4 of the individuals were 45 years or younger at diagnosis.10 Since 2011, studies have examined the post-9/11 incidence of multiple myeloma, and other cancers, in 3 WTC-exposed cohorts compared with the general population. A study of 55 778 New York state residents enrolled in the WTC Health Registry (including rescue workers, recovery workers, and those who lived or worked near the WTC) reported a nearly 3-fold (standardized incidence ratio [SIR], 2.85; 95% CI, 1.15-5.88) higher risk of multiple myeloma, based on 7 cases.11 However, a follow-up study focusing on 10-year cancer incidence patterns in the same population observed a nonsignificant 1.4-fold (SIR, 1.35; 95% CI, 0.70-2.36) increased risk of multiple myeloma.12 Similar studies among 8927 Fire Department of the City of New York (FDNY) WTC-exposed firefighters reported an SIR of 1.49 (95% CI, 0.56-3.97)13 while another among 20 984 non-FDNY responders from the General Responder Cohort reported an SIR of 1.41 (95% CI, 0.64-2.67).14

.70-2.36) increased risk of multiple myeloma.12 Similar studies among 8927 Fire Department of the City of New York (FDNY) WTC-exposed firefighters reported an SIR of 1.49 (95% CI, 0.56-3.97)13 while another among 20 984 non-FDNY responders from the General Responder Cohort reported an SIR of 1.41 (95% CI, 0.64-2.67).14 To improve our understanding of this association, we identified and characterized all WTC-exposed white, male firefighters from FDNY who received a diagnosis of multiple myeloma from September 12, 2001, to July 1, 2017. Second, we conducted a screening study for myeloma precursor disease among the FDNY subset of 781 white male WTC-exposed firefighters older than 50 years. The aim of our screening study was to define the age-specific prevalence of MGUS and light-chain MGUS in WTC-exposed New York City male firefighters and to compare the FDNY prevalence with that in the male Olmsted County, Minnesota, population. We also assessed patterns of myeloma precursor disease in relation to our exposure metric (time of initial arrival at the WTC site) to test for a possible exposure-response association. Methods The case series and screening studies were approved by the Institutional Review Boards of Montefiore Medical Center and Albert Einstein College of Medicine. All participants provided written consent to research.

To improve our understanding of this association, we identified and characterized all WTC-exposed white, male firefighters from FDNY who received a diagnosis of multiple myeloma from September 12, 2001, to July 1, 2017. Second, we conducted a screening study for myeloma precursor disease among the FDNY subset of 781 white male WTC-exposed firefighters older than 50 years. The aim of our screening study was to define the age-specific prevalence of MGUS and light-chain MGUS in WTC-exposed New York City male firefighters and to compare the FDNY prevalence with that in the male Olmsted County, Minnesota, population. We also assessed patterns of myeloma precursor disease in relation to our exposure metric (time of initial arrival at the WTC site) to test for a possible exposure-response association. Methods The case series and screening studies were approved by the Institutional Review Boards of Montefiore Medical Center and Albert Einstein College of Medicine. All participants provided written consent to research. Multiple Myeloma Case Series Population and Case Information Through the WTC Health Program, FDNY WTC-exposed firefighters (N = 12 942) receive comprehensive physical and mental health services. All white, male firefighters with a post-9/11 diagnosis of multiple myeloma (n = 16) in the FDNY WTC Health Program as of July 1, 2017, were included in the case series population (Figure 1A). The cases were confirmed in 2 ways: (1) via state tumor registry matches with New York, New Jersey, Connecticut, Pennsylvania, Florida, North Carolina, South Carolina, Arizona, and Virginia. A total of 8622 of 8830 (98%) retired FDNY firefighters in this cohort currently reside in these states. All active FDNY firefighters are required to live in New York City or neighboring New York state counties of Westchester, Rockland, Orange, Nassau, or Suffolk; (2) via FDNY WTC Health Program medical assessments/records reviewed by an experienced clinician (N.J.).13 Using medical records from the time of diagnosis, we extracted information on age at diagnosis, bone marrow aspirate and biopsy reports, and serum and urine protein testing. We performed complete case analysis when outcome data were missing.

ram medical assessments/records reviewed by an experienced clinician (N.J.).13 Using medical records from the time of diagnosis, we extracted information on age at diagnosis, bone marrow aspirate and biopsy reports, and serum and urine protein testing. We performed complete case analysis when outcome data were missing. Figure 1. Exclusion Criteria FDNY indicates Fire Department of the City of New York; MGUS, monoclonal gammopathy of undetermined significance; WTC, World Trade Center. FDNY MGUS Screening Study Population The FDNY WTC Health Program provides regular monitoring examinations approximately every 12 to 18 months. From December 2013 through October 2015, serum samples were collected from 1173 WTC-exposed firefighters during routine monitoring examinations (Figure 1b). If a firefighter had more than 1 monitoring examination during the study period, serum was collected from the first examination. To facilitate comparison with our multiple myeloma results and our external comparison group, we restricted the analysis to white men. We excluded specimens from 5 who had received a diagnosis of myeloma or non-Hodgkin lymphoma prior to blood sampling. We also excluded participants younger than 50 or older than 79 years at the time of blood sampling. The final study cohort included 781 white, male, WTC-exposed firefighters (Table). All serum samples were analyzed for MGUS or light-chain MGUS in 2016.

agnosis of myeloma or non-Hodgkin lymphoma prior to blood sampling. We also excluded participants younger than 50 or older than 79 years at the time of blood sampling. The final study cohort included 781 white, male, WTC-exposed firefighters (Table). All serum samples were analyzed for MGUS or light-chain MGUS in 2016. Table. Characteristics of World Trade Center (WTC)-Exposed Fire Department of the City of New York (FDNY) White Male Firefighters in the Screening Study and Olmsted County, Minnesota, White Male Cohort Characteristic No. (%) WTC-Exposed FDNY Firefighters (N = 781) Olmsted County (N = 7612) Age, y 50-59 482 (61.7) 3450 (45.3) 60-69 225 (28.8) 2554 (33.6) 70-79 74 (9.5) 1608 (21.1) WTC arrival date, 2001 Morning of Sep 11 116 (14.9) 0 Afternoon of Sep 11 419 (53.6) 0 Sep 12 125 (16.0) 0 Sep 13-24 112 (14.3) 0 Later than Sep 24 9 (1.2) 0 Time spent at WTC site, mean (SD), mo 3.17 (2.68) 0 Demographic data including age at blood sample collection, race, and sex were obtained from the FDNY employee database. Additionally, data from the first self-administered health questionnaire, which began in October 2001, were used to categorize level of WTC exposure based on initial arrival time (arriving the morning of 9/11 [most highly exposed]; arriving the afternoon of 9/11; arriving September 12, 2001; arriving between September 13 and 24, 2001; and arriving between September 25, 2001, and July 24, 2002 [least exposed], when the WTC site closed) and duration at the WTC site (months in which a participant worked at least 1 day at the WTC site).

hly exposed]; arriving the afternoon of 9/11; arriving September 12, 2001; arriving between September 13 and 24, 2001; and arriving between September 25, 2001, and July 24, 2002 [least exposed], when the WTC site closed) and duration at the WTC site (months in which a participant worked at least 1 day at the WTC site). Serum Specimen and Laboratory Methods We obtained a 0.5-mL aliquot for each study participant who consented to the FDNY research protocol. Each aliquot tube was labeled only with the participant’s coded identification number. All specimens were shipped on dry ice to the Protein Laboratory at Memorial Sloan Kettering Cancer Center, where protein assays were performed. The samples were tested concurrently and results were assessed by 2 of us (O.L. and K.M.) in a blinded fashion.5,15,16,17 Comparison Population: Olmsted County, Minnesota We used published data from the population-based Olmsted County, Minnesota, study, the only available screening study including both MGUS and light-chain MGUS assays,18 as our comparison population. The racial distribution of the population in Olmsted County is predominantly white.19 We focused on the Olmsted County MGUS and light-chain MGUS rates in men 50 to 79 years old (N = 7612). Among the 7612 men, the prevalence of overall MGUS (ie, MGUS and light-chain MGUS), MGUS, and light-chain MGUS was 4.4% (n = 333), 3.4% (n = 258), and 1.0% (n = 75), respectively.18

ed County is predominantly white.19 We focused on the Olmsted County MGUS and light-chain MGUS rates in men 50 to 79 years old (N = 7612). Among the 7612 men, the prevalence of overall MGUS (ie, MGUS and light-chain MGUS), MGUS, and light-chain MGUS was 4.4% (n = 333), 3.4% (n = 258), and 1.0% (n = 75), respectively.18 Statistical Analysis The crude age-specific prevalence rates were calculated for white men as the total number of cases within each age stratum divided by the total number of individuals within that age stratum. Prevalence rates for overall MGUS, MGUS, and light-chain MGUS were calculated for the FDNY study population. Additionally, to enable external comparison, prevalence rates were age standardized to the US 2000 male population, ages 50 to 79 years. Age-adjusted 95% confidence intervals were calculated for directly standardized relative rates (RRs) using the modified γ approximation method, which assumes a Poisson distribution. Standard errors for 95% Mantel-Haenszel confidence limits of standardized relative risks were calculated using the Greenland and Robins20 variance formula. Participants older than 79 years were excluded from this analysis due to small numbers in the firefighting cohort. Exposure to the WTC, using time of arrival at the WTC site as a proxy for intensity, was evaluated separately in a stratified analysis. All analyses were performed using SAS, version 9.4, and R v.3.2.0.

a. Participants older than 79 years were excluded from this analysis due to small numbers in the firefighting cohort. Exposure to the WTC, using time of arrival at the WTC site as a proxy for intensity, was evaluated separately in a stratified analysis. All analyses were performed using SAS, version 9.4, and R v.3.2.0. Results Multiple Myeloma Case Series We identified 16 white male responders from the FDNY firefighter cohort with a post-9/11 diagnosis of multiple myeloma. The median age at diagnosis was 57 years (range, 38-76 years). The median time between 9/11 and diagnosis was 12.0 years (range, 1.0-15.7 years). The myeloma cell infiltration of the bone marrow ranged between less than 10% and 90% across individuals; immunophenotypic characterization of the bone marrow sample revealed CD20-positive plasma cells in 5 of 7 (71%; 95% CI, 36%-92%) tested cases. Results on serum and/or urine monoclonal proteins isotype and free light chains were available for 14 cases; 7 (50%; 95% CI, 27%-73%) had light-chain multiple myeloma. Individual-level data for the case series, including plasma-cell percentage, serum and/or urine monoclonal protein isotype and free light chains, and plasma-cell CD20 expression are provided in eTable 1 in the Supplement.

ght chains were available for 14 cases; 7 (50%; 95% CI, 27%-73%) had light-chain multiple myeloma. Individual-level data for the case series, including plasma-cell percentage, serum and/or urine monoclonal protein isotype and free light chains, and plasma-cell CD20 expression are provided in eTable 1 in the Supplement. MGUS Screening Study Demographic Characteristics The Table provides selected demographic characteristics of the FDNY firefighter cohort and Olmsted County comparison population. The median age at time of FDNY specimen collection was 57 years (interquartile range, 54-62 years). The majority of firefighters arrived at the WTC site on the day of the attacks (535 [68.5%]) and spent a mean (SD) 3.17 (2.68) months working at the WTC site. The Olmsted County cohort was assumed to have no WTC exposure. The median age of firefighters with MGUS was nonsignificantly younger than that among the reference population: 62 years (FDNY firefighter cohort) vs 70 years (Olmsted County comparison population). Similarly, firefighters with light-chain MGUS were nonsignificantly younger than the reference population (median age, 61 vs 68 years). Specific characteristics of MGUS are found in eTable 2 in the Supplement.

e reference population: 62 years (FDNY firefighter cohort) vs 70 years (Olmsted County comparison population). Similarly, firefighters with light-chain MGUS were nonsignificantly younger than the reference population (median age, 61 vs 68 years). Specific characteristics of MGUS are found in eTable 2 in the Supplement. MGUS and Light-Chain MGUS Age-Standardized and Relative Rates Among white men aged 50 to 79 years, the age-standardized prevalence rate (ASR) of overall MGUS among WTC-exposed FDNY firefighters was 7.63 per 100 persons (95% CI, 5.45-9.81), which was 1.8-fold significantly higher compared with the rate among those from Olmsted County, Minnesota (ASR, 4.34 per 100 persons; 95% CI, 3.88-4.81 per 100 persons and RR, 1.76; 95% CI, 1.34-2.29) (Figure 2A). Figure 2B shows that, among the same population, the ASR of FDNY firefighters with only light-chain MGUS was 3.08 per 100 persons (95% CI, 1.66-4.50 per 100 persons), much greater than that of Olmsted County, which had an ASR of 0.98 per 100 persons (95% CI, 0.76-1.21 per 100 persons). The relative risk of having light-chain MGUS was 3.1-fold significantly higher (RR, 3.13; 95% CI, 1.99-4.93) when comparing WTC-exposed FDNY firefighters with the Olmsted County population. Last, Figure 2C shows that, among the same population, the ASR of FDNY firefighters with only MGUS was 4.55 per 100 persons (95% CI, 2.90-6.21 per 100 persons) and was slightly but not significantly higher than the Olmsted County cohort (ASR, 3.36; 95% CI, 2.95-3.77; RR, 1.35; 95% CI, 0.96-1.91) (Figure 2C). Overall MGUS was assessed by WTC arrival time and by duration at the WTC site; for all arrival times the ASRs were greater than in the reference population, although we did not find an exposure gradient (data not shown). Additionally, there was no significant difference in ASRs when duration of work at the WTC site was included in the analyses (data not shown).

me and by duration at the WTC site; for all arrival times the ASRs were greater than in the reference population, although we did not find an exposure gradient (data not shown). Additionally, there was no significant difference in ASRs when duration of work at the WTC site was included in the analyses (data not shown). Figure 2. Prevalence of Monoclonal Gammopathy of Undetermined Significance (MGUS) and Light-Chain MGUS in World Trade Center–Exposed Fire Department of the City of New York (FDNY) White Male Firefighters and Comparison Population aOverall MGUS includes both MGUS and light-chain MGUS cases. bPrevalence rates are age standardized to US 2000 male population age 50 to 79 years.

Figure 2. Prevalence of Monoclonal Gammopathy of Undetermined Significance (MGUS) and Light-Chain MGUS in World Trade Center–Exposed Fire Department of the City of New York (FDNY) White Male Firefighters and Comparison Population aOverall MGUS includes both MGUS and light-chain MGUS cases. bPrevalence rates are age standardized to US 2000 male population age 50 to 79 years. Discussion In this first comprehensive study focusing on characteristics of multiple myeloma and its precursor disease (MGUS and light-chain MGUS) among WTC-exposed first responders, we observe striking patterns. We found the median age of multiple myeloma diagnosis to be 57 years, which is roughly 12 years younger than what is seen nationally,2 and because symptoms usually develop shortly after clinical manifestation of the disease (approximately 1 year), this would argue against a lead-time bias. Furthermore, the proportion of participants with CD20-expressing plasma cells—characteristics associated with a poorer prognosis—was more than 3.5-fold higher than found in other populations (71% vs approximately 20%).2,5,21,22 Among WTC-exposed white, male firefighters, the proportion of light-chain multiple myeloma was more than double that of the general population (50% vs approximately 20%).4 Similarly, in our screening study, we found a 2-fold significantly higher risk of myeloma precursor disease, particularly light-chain MGUS, the precursor of light-chain multiple myeloma.5 As suggested previously based on smaller numbers,10,11,12,13,14 our study shows that WTC exposure may be a risk factor for the development of multiple myeloma and its precursor disease.

significantly higher risk of myeloma precursor disease, particularly light-chain MGUS, the precursor of light-chain multiple myeloma.5 As suggested previously based on smaller numbers,10,11,12,13,14 our study shows that WTC exposure may be a risk factor for the development of multiple myeloma and its precursor disease. Prior studies in cohorts without WTC exposure have found an increased risk of MGUS and light-chain MGUS among individuals exposed to known and suspected carcinogens including PCB, dioxins, PAHs, and asbestos,6,7,8 including exposure to Agent Orange (which contains the human carcinogen 2,3,7,8-tetrachlorodibenzo-p-dioxin).6 Interestingly, among veterans exposed to Agent Orange who were found to have myeloma precursor disease, there was a particular excess risk of light-chain MGUS (11 of 34 [32%] of the MGUS cases were light-chain MGUS).6 In our multiple myeloma case series, we found 7 of 14 (50%) cases to be of light-chain multiple myeloma subtype, which is more than double the rate in other populations (50% vs approximately 20%).4 In our screening study, we observed 18 of 47 (38%; 95% CI, 26%-53%) of the precursor cases to have light-chain MGUS, the precursor of light-chain multiple myeloma. Furthermore, we found that the risk of having light-chain MGUS was 3 times higher (RR, 3.13; 95% CI, 1.99-4.93) among our population compared with the Olmsted County population. These findings are of interest due to previously observed associations between light-chain multiple myeloma and light-chain MGUS and exposure to toxins6,17 and chronic immune stimulation.23

ight-chain MGUS was 3 times higher (RR, 3.13; 95% CI, 1.99-4.93) among our population compared with the Olmsted County population. These findings are of interest due to previously observed associations between light-chain multiple myeloma and light-chain MGUS and exposure to toxins6,17 and chronic immune stimulation.23 Many WTC first responders were initially exposed to aerosolized dust and toxic fumes from burning jet fuel and building materials. For the next 10 months of the rescue, recovery, and cleanup effort, responders were exposed to burning subterranean fires that released trapped gases and dust, and were not extinguished until the end of December. The final insult included diesel fuel combustion byproducts from heavy equipment used at the site. The WTC dust itself included pulverized cement, glass fibers, asbestos, lead, PAHs, PCBs, and polychlorinated furans and dioxins produced as combustion byproducts from the collapsed and burning buildings.9 Many of these substances are known carcinogens, providing biologic plausibility for an association between WTC exposure and cancer.9,24,25,26 Although some contaminants could directly cause cancer, WTC exposure has been shown to trigger chronic inflammation resulting in upper and lower respiratory diseases and autoimmune diseases,27,28,29 and, therefore, inflammation-induced oncogenesis should also be considered as a potential mechanism.23

cancer.9,24,25,26 Although some contaminants could directly cause cancer, WTC exposure has been shown to trigger chronic inflammation resulting in upper and lower respiratory diseases and autoimmune diseases,27,28,29 and, therefore, inflammation-induced oncogenesis should also be considered as a potential mechanism.23 Several surface antigens are used to characterize individual plasma cells as malignant or normal. For example, compared with normal plasma cells, abnormal plasma cells tend to be low in the expression of CD19 and CD27, have weaker expression of CD45, and increased expression of CD28, CD56, and CD117.30 Expression of CD20 is typically seen during the maturation process of B cells and absent from plasma cells; however, CD20 expression can be detected in 13% to 22% of patients with multiple myeloma diagnosed in the general population.22 Preliminary data suggest that subsets of CD20-positive multiple myeloma patients have a poorer prognosis.5,21 In the present study, albeit based on small numbers, immunophenotypic characterization revealed CD20-positive plasma cells in 5 of 7 (71%) multiple myeloma cases tested among WTC-exposed responders. Future work is needed to expand on these observations.

20-positive multiple myeloma patients have a poorer prognosis.5,21 In the present study, albeit based on small numbers, immunophenotypic characterization revealed CD20-positive plasma cells in 5 of 7 (71%) multiple myeloma cases tested among WTC-exposed responders. Future work is needed to expand on these observations. Limitations We acknowledge that our study has limitations. First, in our myeloma case series analysis, we were not able to rigorously compare FDNY and national age-adjusted incidence rates due to small or zero counts in certain age strata. This is often a challenge in the study of rare cancers.31 Additionally, when MGUS was assessed by WTC arrival time,32 for all arrival times the ASRs were greater than in the reference population. However, the study was underpowered to detect an exposure-response gradient association between WTC exposure and MGUS. Nonetheless, we did observe an elevated risk of overall MGUS and light-chain MGUS compared with the Olmsted County population. While a comparison group composed exclusively of firefighters with no exposure to the WTC disaster or a truly random sample of the US population would be most desirable, no such cohort data were available. Specifically, no other study meeting those criteria screened all participants for light-chain MGUS in the same manner that we did. The Olmsted County population matched our testing protocol and was demographically similar and thus provided a valuable comparison for this study. Similarly, due to small numbers in the firefighter cohort and lack of an adequate comparison population, races other than white and women were excluded. Future research investigating MGUS and light-chain MGUS may provide additional comparisons to our full population, as well as other WTC-exposed populations, important populations in need of follow-up.

n the firefighter cohort and lack of an adequate comparison population, races other than white and women were excluded. Future research investigating MGUS and light-chain MGUS may provide additional comparisons to our full population, as well as other WTC-exposed populations, important populations in need of follow-up. Furthermore, in our case series analysis it is possible that the onset of precursor disease states may have preceded WTC exposure; however, 75% of myeloma cases were diagnosed more than 5 years after 9/11, with half being at least 12 years following the attacks. This suggests that, for most, the precursor disease likely developed after 9/11.33 Last, while we controlled for the main risk factors for MGUS, we cannot rule out the possibility of uncontrolled confounding between the FDNY population and the Olmsted County population. This study had a number of strengths. First, we believe that our case ascertainment was excellent; this included matching to tumor registries where more than 98% of our cohort reside, as well as full access to the FDNY electronic medical record where approximately 87% have had a monitoring or treatment visit within the past 2 years. Second, to our knowledge, this is the largest study to characterize multiple myeloma in WTC-exposed responders. More importantly, this is the first study to establish the age-specific prevalence of MGUS/light-chain MGUS in a well-defined population of WTC-exposed responders.

or treatment visit within the past 2 years. Second, to our knowledge, this is the largest study to characterize multiple myeloma in WTC-exposed responders. More importantly, this is the first study to establish the age-specific prevalence of MGUS/light-chain MGUS in a well-defined population of WTC-exposed responders. Conclusions In summary, we identified and characterized all WTC-exposed white, male FDNY firefighters who received a diagnosis of multiple myeloma from September 12, 2001 to July 1, 2017, and 50% (7 of 14) of these cases were light-chain multiple myeloma. The median age at multiple myeloma diagnosis was 57 years, which is 12 years younger than what is observed in national data.2 A high proportion of patients with multiple myeloma had CD20-expressing plasma cells, which is a characteristic associated with a poorer prognosis.5,21 In the screening study including 781 white male WTC-exposed FDNY firefighters, we found the risk of overall MGUS to be 2-fold higher compared with the rate in the Olmsted County reference population; in particular, the risk of light-chain MGUS was higher, which may have important prognostic implications. Taken together, our results show that environmental exposure due to the WTC attacks is associated with myeloma precursor disease (MGUS and light-chain MGUS) and may be a risk factor for the development of multiple myeloma at an earlier age, particularly the light-chain subtype. Supplement. eTable 1. Post-9/11 multiple myeloma cases among WTC-exposed FDNY white male firefighters

Conclusions In summary, we identified and characterized all WTC-exposed white, male FDNY firefighters who received a diagnosis of multiple myeloma from September 12, 2001 to July 1, 2017, and 50% (7 of 14) of these cases were light-chain multiple myeloma. The median age at multiple myeloma diagnosis was 57 years, which is 12 years younger than what is observed in national data.2 A high proportion of patients with multiple myeloma had CD20-expressing plasma cells, which is a characteristic associated with a poorer prognosis.5,21 In the screening study including 781 white male WTC-exposed FDNY firefighters, we found the risk of overall MGUS to be 2-fold higher compared with the rate in the Olmsted County reference population; in particular, the risk of light-chain MGUS was higher, which may have important prognostic implications. Taken together, our results show that environmental exposure due to the WTC attacks is associated with myeloma precursor disease (MGUS and light-chain MGUS) and may be a risk factor for the development of multiple myeloma at an earlier age, particularly the light-chain subtype. Supplement. eTable 1. Post-9/11 multiple myeloma cases among WTC-exposed FDNY white male firefighters eTable 2. Myeloma precursor disease screening study: characteristics of MGUS and light-chain-MGUS in WTC-exposed FDNY white male firefighters Click here for additional data file.

Introduction In the United Kingdom, upper gastrointestinal symptoms account for at least 3% of consultations in primary care.1 The national esophagogastric cancer (OGC) audit reported 7044 cases of OGC diagnosed in 2016. Many patients present with advanced-stage disease and only 38% of cases can be treated with a curative intent.2 Current UK referral guidelines for suspected OGC focus on alarm symptoms such as dysphagia and odynophagia, despite these symptoms having poor sensitivity and specificity for OGC and often only occur in advanced disease translating into a poor outcome and overall survival.3 There is a wide range in the rate of oesophagogastro duodenoscopy (OGD) among general practice populations in England and OGC patients belonging to practices with the lowest rates of OGD referral are at greatest risk of poor overall survival owing to advanced tumor stage at diagnosis.4 Furthermore, OGD is an expensive invasive investigation, with poor uptake in specific ethic minority populations consequently affecting survival.5 This high prevalence of upper gastrointestinal symptoms coupled with the low incidence of OGC and the nonspecific nature of symptoms in early disease highlight the need for a triage test to direct patients to have OGD.

estigation, with poor uptake in specific ethic minority populations consequently affecting survival.5 This high prevalence of upper gastrointestinal symptoms coupled with the low incidence of OGC and the nonspecific nature of symptoms in early disease highlight the need for a triage test to direct patients to have OGD. Volatile organic compounds (VOCs) emitted from the human body have been of interest to researchers for several decades,6 with associations previously suggested between specific VOCs and breath and lung, bladder, and breast cancers.7,8,9 We analyzed exhaled breath samples using selected ion flow-tube mass spectrometry (SIFT-MS) from 210 patients, 81 with OGC and 129 control patients. A diagnostic model of 13 VOCs was able to diagnose OGC with a sensitivity of 89% and specificity of 94%.10 Several phase 1 biomarker studies linking noninvasively measured VOCs to the presence of cancer similar to our own have been published,7,8,9,11,12,13 with very few attempts to externally validate these findings in a further prospective cohort of patients from several centers. The objective of this multicenter validation study was to establish the diagnostic accuracy of a previously identified set of breath VOCs dysregulated with the presence of OGC in a multicenter setting.

Volatile organic compounds (VOCs) emitted from the human body have been of interest to researchers for several decades,6 with associations previously suggested between specific VOCs and breath and lung, bladder, and breast cancers.7,8,9 We analyzed exhaled breath samples using selected ion flow-tube mass spectrometry (SIFT-MS) from 210 patients, 81 with OGC and 129 control patients. A diagnostic model of 13 VOCs was able to diagnose OGC with a sensitivity of 89% and specificity of 94%.10 Several phase 1 biomarker studies linking noninvasively measured VOCs to the presence of cancer similar to our own have been published,7,8,9,11,12,13 with very few attempts to externally validate these findings in a further prospective cohort of patients from several centers. The objective of this multicenter validation study was to establish the diagnostic accuracy of a previously identified set of breath VOCs dysregulated with the presence of OGC in a multicenter setting. Methods Multivariable logistic regression model (stepwise regression) (eMethods 1 in the Supplement) was used to create a 5-VOCs model which were butyric acid, pentanoic acid, hexanoic acid, butanal, and decanal from our previously published data set.10 The predictive probabilities generated by this 5-VOC diagnostic model were then used to generate an receiver operating characteristic (ROC) curve, which showed a good diagnostic accuracy with an area under the curve of 0.90 (SD, 0.02).

acid, hexanoic acid, butanal, and decanal from our previously published data set.10 The predictive probabilities generated by this 5-VOC diagnostic model were then used to generate an receiver operating characteristic (ROC) curve, which showed a good diagnostic accuracy with an area under the curve of 0.90 (SD, 0.02). Based on 50% of patients in the study population having cancer (1 patient with a benign abnormality was recruited to 1 patient with cancer) and maintaining a sensitivity and specificity of 80% for the diagnostic model derived from our previous research, the sample size estimated for the multicenter external validation study was 325 patients; 162 patients with esophageal or gastric cancer and 163 patients with benign conditions or a normal upper gastrointestinal tract. Breath samples were taken from 3 hospitals (St Mary’s Hospital Imperial College London, University College London Hospital, and The Royal Marsden Hospital) and transported to St Mary’s VOC laboratory for SIFT-MS analysis. The National Health Service (NHS) Health Research Authority (NRES Committee London–Camden and Islington) approved the study, and written informed consent was obtained. The full protocol for this study was previously published.14 The study was reported according to STARD 2015 (Standards for Reporting of Diagnostic Accuracy Studies) guidelines (eMethods 2 in the Supplement).15 The trial was registered with the National Health Service, Health Research Authority

Breath samples were taken from 3 hospitals (St Mary’s Hospital Imperial College London, University College London Hospital, and The Royal Marsden Hospital) and transported to St Mary’s VOC laboratory for SIFT-MS analysis. The National Health Service (NHS) Health Research Authority (NRES Committee London–Camden and Islington) approved the study, and written informed consent was obtained. The full protocol for this study was previously published.14 The study was reported according to STARD 2015 (Standards for Reporting of Diagnostic Accuracy Studies) guidelines (eMethods 2 in the Supplement).15 The trial was registered with the National Health Service, Health Research Authority Patients 18 years or older with upper gastrointestinal symptoms attending for endoscopy or surgery were eligible. In the cancer cohort only patients with histologically confirmed nonmetastatic esophagogastric adenocarcinoma (stage I-III) were included. All patients in the cancer cohort were sampled when they were neoadjuvant naive. Patients who had a documented active infection, were unable to provide informed consent, or unable to provide a 500-mL breath sample were excluded. Patients with Barrett esophagus were excluded from the control group (this is a premalignant condition worthy of independent investigation).

Patients 18 years or older with upper gastrointestinal symptoms attending for endoscopy or surgery were eligible. In the cancer cohort only patients with histologically confirmed nonmetastatic esophagogastric adenocarcinoma (stage I-III) were included. All patients in the cancer cohort were sampled when they were neoadjuvant naive. Patients who had a documented active infection, were unable to provide informed consent, or unable to provide a 500-mL breath sample were excluded. Patients with Barrett esophagus were excluded from the control group (this is a premalignant condition worthy of independent investigation). Breath Sampling Methodology After informed consent was obtained from all patients, we followed the sampling protocol used in our previous clinical studies,10 which was informed by our investigations on the influence of breath maneuvers and hospital environment on VOC measurements.16,17 Patients fasted for a minimum of 4 hours prior to their breath sample collection. Patients rested in the same area for at least 20 minutes prior to exhaled breath collection and all samples were obtained immediately prior to endoscopy or surgery. Patients were asked to perform a single deep nasal inhalation followed by complete exhalation via their mouth into secure 500-mL steel breath bag (GastroCHECK) via a 1-mL Luer lok syringe (Terumo Europe, Leuven, Belgium). Patients in the cancer and control groups were recruited consecutively. The research team were aware of clinical diagnosis when breath sampling the patients, however the clinical team performing the OGD or surgery were blind to the results of the breath analysis.

1-mL Luer lok syringe (Terumo Europe, Leuven, Belgium). Patients in the cancer and control groups were recruited consecutively. The research team were aware of clinical diagnosis when breath sampling the patients, however the clinical team performing the OGD or surgery were blind to the results of the breath analysis. For each VOC measurement, the syringe plunger was removed from the 1-mL Luer lok syringe and the steel breath bag was directly connected via the syringe barrel to the sample inlet arm of the SIFT-MS instrument. For the multi-ion monitoring mode, selective VOCs from breath were analyzed for a total of 60 seconds and measured concentrations were averaged over this time for each VOC. SIFT-MS permits online, real-time VOC quantification.18,19 It has been used in the study of VOCs in breath and urine from patients with conditions including cystic fibrosis and bladder cancer.20,21 The SIFT-MS instrument allows real-time detection and quantification of VOCs in biological samples such as exhaled breath without sample preparation.22 We have previously confirmed the reproducibility of VOCs measurements using SIFT-MS.23 A panel of 30 VOCs including the 5 VOCs forming our diagnostic model were analyzed for each breath sample, as previously described.10

d quantification of VOCs in biological samples such as exhaled breath without sample preparation.22 We have previously confirmed the reproducibility of VOCs measurements using SIFT-MS.23 A panel of 30 VOCs including the 5 VOCs forming our diagnostic model were analyzed for each breath sample, as previously described.10 Clinical Data A detailed medical proforma was completed by the consenting clinician or research fellow using information provided by the patient as well as clinical investigations. These data included patient demographics, tumor characteristics, comorbidities, medications, and lifestyle measures. Diagnostic endoscopy and/or operative findings were recorded for each patient. Histopathologic examination of tissues obtained via endoscopy or from surgically resected specimens was carried out. The reference test was considered positive on OGC histopathologic diagnosis.

Clinical Data A detailed medical proforma was completed by the consenting clinician or research fellow using information provided by the patient as well as clinical investigations. These data included patient demographics, tumor characteristics, comorbidities, medications, and lifestyle measures. Diagnostic endoscopy and/or operative findings were recorded for each patient. Histopathologic examination of tissues obtained via endoscopy or from surgically resected specimens was carried out. The reference test was considered positive on OGC histopathologic diagnosis. SIFT-MS instrument was calibrated daily to 6% water in human exhaled breath. All breath samples were tested using SIFT-MS to ensure that percentage water from the exhaled breath sample in the bag was between 5% and 7%. If this was not the case the sample was discarded because it was likely to be unreliable and representative of bag malfunction. All samples were analyzed within 4 hours of collection. Our methodological studies demonstrated the stability of trace VOCs up to 48 hours from the time of patient sampling when using the GastroCHECK steel breath bag (eMethods 3 in the Supplement). Weekly samples were taken from the ambient room air at the participating hospitals where patients were being breath sampled and also from the laboratory air from where samples were analyzed. This was to ensure that there was no contamination from the ambient room air causing anomalous results, which could represent an important confounding factor (eMethods 4 in the Supplement). Breath sampling methodology was standardized. We performed human factors analysis, which demonstrated several potential sources of error in breath sampling that can affect the results of the analysis. Therefore, all clinicians and researchers participating in this study underwent a thorough credentialing process involving observation of consent, performing breath sampling, and storage of samples prior to participating in the study (eMethods 5 in the Supplement). Threshold of detection of SIFT-MS analysis was defined as 1 part per billion by volume (ppbv), based on previously performed statistical modeling (eMethods 6 in the Supplement).

observation of consent, performing breath sampling, and storage of samples prior to participating in the study (eMethods 5 in the Supplement). Threshold of detection of SIFT-MS analysis was defined as 1 part per billion by volume (ppbv), based on previously performed statistical modeling (eMethods 6 in the Supplement). To confirm the identified VOCs obtained in the exhaled breath using SIFT-MS; we conducted cross-platform validation with Gas chromatography mass spectrometry ([GC-MS] considered the gold standard for compound identification owing to the use of chromatographic separation). Exhaled breath was collected using the same method from 20 patients. The VOC content from each GastroCHECK bag was transferred using an air-sampling pocket pump (SKC 210-1002 series) at 50 mL/min onto inert coated stainless steel Tenax/Carbograph-5TD sorbent tubes (Markes International Ltd, Llantrisant) prior to GC-MS analysis (eMethods 7 in the Supplement).

ame method from 20 patients. The VOC content from each GastroCHECK bag was transferred using an air-sampling pocket pump (SKC 210-1002 series) at 50 mL/min onto inert coated stainless steel Tenax/Carbograph-5TD sorbent tubes (Markes International Ltd, Llantrisant) prior to GC-MS analysis (eMethods 7 in the Supplement). Statistical Analysis Comparison of predicted cancer risk and actual OGD findings or histology from endoscopic biopsies (reference standard test) was then made, and the overall diagnostic accuracy (sensitivity, specificity, and ROC analysis) for this noninvasive diagnostic investigation was determined. A similar ROC analysis was performed based on predicted cancer risk from clinical parameters defined by National Institute for Health and Care Excellence (NICE) criteria.3 Potential confounding factors across the study groups were evaluated by employing the Kruskal-Wallis test for continuous variables and χ2 test for discrete variables. Linear regression models were used to assess any influence of patient demographic factors, or medications, on VOC concentrations measured. P < .05 was used to assign statistical significance. All statistical analysis was performed using the statistical software SPSS (version 22, IBM). Results Only 1 invited patient declined to participate in the study with a patient acceptability rate of 99.7% to undertake and complete the test. No adverse events were observed during breath sampling.

Statistical Analysis Comparison of predicted cancer risk and actual OGD findings or histology from endoscopic biopsies (reference standard test) was then made, and the overall diagnostic accuracy (sensitivity, specificity, and ROC analysis) for this noninvasive diagnostic investigation was determined. A similar ROC analysis was performed based on predicted cancer risk from clinical parameters defined by National Institute for Health and Care Excellence (NICE) criteria.3 Potential confounding factors across the study groups were evaluated by employing the Kruskal-Wallis test for continuous variables and χ2 test for discrete variables. Linear regression models were used to assess any influence of patient demographic factors, or medications, on VOC concentrations measured. P < .05 was used to assign statistical significance. All statistical analysis was performed using the statistical software SPSS (version 22, IBM). Results Only 1 invited patient declined to participate in the study with a patient acceptability rate of 99.7% to undertake and complete the test. No adverse events were observed during breath sampling. Patient Demographics and Tumor Factors After necessary exclusions owing to sample quality, defined as inadequate percentage of water (n = 20), 335 patients in total were included; 172 patients in the control group and 163 patients with esophageal or gastric cancer. In the control group, 89 (51.7%) patients had a normal upper gastrointestinal tract on endoscopy or only the presence of a hiatal hernia. The most common diagnoses among the remaining participants in the control group were esophagitis, gastritis, or duodenitis with or without erosions in 59 (34.3%) patients, followed by the presence of benign gastric polyps in 12 (7.0%) patients and achalasia or esophageal stricture in 11 (6.4%) patients.

hiatal hernia. The most common diagnoses among the remaining participants in the control group were esophagitis, gastritis, or duodenitis with or without erosions in 59 (34.3%) patients, followed by the presence of benign gastric polyps in 12 (7.0%) patients and achalasia or esophageal stricture in 11 (6.4%) patients. In the cancer group there were significant increases in patient age, proportion of male and white patients, ex-smokers, ASA grade 3, and hypertensive patients, with a reduced proportion of patients with liver impairment (Table 1). There were also significant increases in the use of statin, β-blocker, and ACE-inhibitor medications in the cancer group (Table 1). Dysphagia, vomiting, and gastrointestinal bleeding were increased, and abdominal pain reduced, as presenting symptoms in the cancer group (eMethods 8 in the Supplement). Furthermore, the breakdown of the cancer-specific factors including stage and tumor location is provided in Table 2 with 72 (44.2%) of tumors being gastric in origin, 123 (69.3%) being T3 or T4, and 106 (65%) being nodal positive.

pain reduced, as presenting symptoms in the cancer group (eMethods 8 in the Supplement). Furthermore, the breakdown of the cancer-specific factors including stage and tumor location is provided in Table 2 with 72 (44.2%) of tumors being gastric in origin, 123 (69.3%) being T3 or T4, and 106 (65%) being nodal positive. Table 1. Comparison of Demographic Factors and Medication Between Cancer and Control Patients Variable Control Group, No. (%) Cancer Group, No. (%) P Value Age, median (IQR) 55 (41–69) 68 (60–75) <.001 Sex Female 91 (52.6) 29 (17.8) <.001 Male 81 (47.4) 134 (82.2) White 88 (51.5) 114 (69.9) .001 Smoking history Current 31 (18.1) 22 (13.5) <.001 Ex smoker 40 (23.4) 72 (44.2) Alcohol history Current 77 (45.0) 87 (53.4) .30 Ex alcohol user 19 (11.1) 17 (10.4) ASA grade 1 72 (42.1) 41 (25.2) .001 2 91 (53.2) 101 (62.0) 3 8 (4.7) 21 (12.9) Comorbidities Diabetes 28 (16.4) 26 (16.0) .92 Renal impairment 8 (4.7) 3 (1.8) .15 Chronic obstructive pulmonary disease 10 (5.8) 7 (4.3) .52 Ischaemic heart disease 19 (11.1) 20 (12.3) .74 Liver impairment 16 (9.4) 1 (0.6) <.001 Hypertension 45 (26.3) 62 (38.0) .02 Asthma 18 (10.5) 19 (11.7) .74 Medication Proton pump inhibitor 83 (48.5) 93 (57.1) .12 Statin 35 (20.5) 56 (34.4) .004 β-Blocker 12 (7.0) 27 (16.6) .007 ACE inhibitor 13 (7.6) 37 (22.7) <.001 Amlodipine 17 (9.9) 17 (10.4) .88 Aspirin 13 (7.6) 14 (8.6) .74 Clopidogrel 7 (4.1) 5 (3.1) .62 Metformin 20 (11.7) 17 (10.4) .71 Diuretic 4 (2.3) 7 (4.3) .31 Abbreviations: ACE, angiotensin converting enzyme; ASA, American Society of Anesthesiologists.

ker 12 (7.0) 27 (16.6) .007 ACE inhibitor 13 (7.6) 37 (22.7) <.001 Amlodipine 17 (9.9) 17 (10.4) .88 Aspirin 13 (7.6) 14 (8.6) .74 Clopidogrel 7 (4.1) 5 (3.1) .62 Metformin 20 (11.7) 17 (10.4) .71 Diuretic 4 (2.3) 7 (4.3) .31 Abbreviations: ACE, angiotensin converting enzyme; ASA, American Society of Anesthesiologists. Table 2. Description of Cancer-Specific Factors Tumor-Related Factor Patients, No. (%) Tumor location Gastric 72 (44.2) Gastroesophageal junction 36 (22.1) Oesophageal 55 (33.7) Clinical T stage 1 18 (11.0) 2 32 (19.6) 3 61 (37.4) 4 52 (31.9) Clinical N stage 0 57 (35.0) 1 58 (35.6) 2 22 (13.5) 3 26 (16.0) Cross-Platform GC-MS Validation In conjunction with SIFT-MS analysis, TD-GC-MS analysis was applied to breath from the same cohort of patients (n = 20) to cross-validate the identity of measured VOCs. eMethods 9 in the Supplement summarizes the VOC identification by GC-MS via mass spectrum (MS) matching of detected compounds to the commercial NIST library, as well as their calculated retention indices (RI) to those of authentic chemical standards undergone separation on the ZB-624 column. Overall, the presence of 30 VOCs was confirmed except for pentanol owing to its limited level of signal-to-noise ratio (SNR<5) among the analyzed samples.

o the commercial NIST library, as well as their calculated retention indices (RI) to those of authentic chemical standards undergone separation on the ZB-624 column. Overall, the presence of 30 VOCs was confirmed except for pentanol owing to its limited level of signal-to-noise ratio (SNR<5) among the analyzed samples. SIFT-MS Analysis and VOC-Based OGC Diagnosis The concentration of butyric acid, hexanoic acid, butanal, and decanal showed significant differences between the cancer and control groups (eMethods 10 in the Supplement). Five VOCs were then taken forward to form a risk-prediction model for the diagnosis of esophagogastric cancer, and included in a multivariable logistic regression analysis with cancer diagnosis as the dependent variable (eMethods 10 in the Supplement). To ensure these factors were not associated with a confounding demographic variable or presenting symptom that differed between the comparison groups, linear regression models were performed for each of the 5 VOCs (eMethods 10 in the Supplement). There were no significant differences in the concentration of these 5 VOCs between patients with esophageal or gastric cancer (eMethods 10 in the Supplement).

r presenting symptom that differed between the comparison groups, linear regression models were performed for each of the 5 VOCs (eMethods 10 in the Supplement). There were no significant differences in the concentration of these 5 VOCs between patients with esophageal or gastric cancer (eMethods 10 in the Supplement). The predictive probabilities generated by this 5-VOC diagnostic model were then used to generate an ROC curve, which showed a good diagnostic accuracy with an area under the curve of 0.85 (SD, 0.02) (Figure). This translated to a sensitivity of 80% and specificity of 81% for the diagnosis of esophagogastric cancer. This compared with the diagnostic accuracy generated by a clinical parameters test based on NICE guidelines for endoscopy referral,3 which had an area under the curve of 0.73 (SD, 0.03), sensitivity of 59%, and specificity of 81% (eMethods 11 in the Supplement). Figure. ROC Curve for the 5-VOC Breath Model in the Diagnosis of Esophagogastric Cancer in the Multicenter Clinical Triala Abbreviations: ROC, receiver operating characteristic curve; VOC, volatile organic compounds. aArea under the curve of 0.85 (SD, 0.02).

The predictive probabilities generated by this 5-VOC diagnostic model were then used to generate an ROC curve, which showed a good diagnostic accuracy with an area under the curve of 0.85 (SD, 0.02) (Figure). This translated to a sensitivity of 80% and specificity of 81% for the diagnosis of esophagogastric cancer. This compared with the diagnostic accuracy generated by a clinical parameters test based on NICE guidelines for endoscopy referral,3 which had an area under the curve of 0.73 (SD, 0.03), sensitivity of 59%, and specificity of 81% (eMethods 11 in the Supplement). Figure. ROC Curve for the 5-VOC Breath Model in the Diagnosis of Esophagogastric Cancer in the Multicenter Clinical Triala Abbreviations: ROC, receiver operating characteristic curve; VOC, volatile organic compounds. aArea under the curve of 0.85 (SD, 0.02). Discussion This multicenter study demonstrated a sensitivity of 80% and specificity of 81% of a single breath test in the diagnosis of esophagogastric cancer, thus validating the 5-VOC breath model. All patients with cancer included in the study were receiving a curative treatment pathway, highlighting the potential value of the test in detecting operable disease and the potential impact on survival. The sensitivity of 80% compares favorably to the existing technologies such as fecal occult blood test (sensitivity ranging from 30%-70%) for colorectal cancer,24 and more specifically to upper gastrointestinal disease, the cytosponge (sensitivity 73%) for Barrett esophagus.25 An important finding with both these technologies was an increase in sensitivity associated with multiple episodes of testing, which could be an important area for further research of breath testing.

er,24 and more specifically to upper gastrointestinal disease, the cytosponge (sensitivity 73%) for Barrett esophagus.25 An important finding with both these technologies was an increase in sensitivity associated with multiple episodes of testing, which could be an important area for further research of breath testing. At present the NICE guidance for endoscopy referral is based on age threshold and symptom criteria. Patients aged over 55 years with dyspepsia, or those of any age with alarm-type symptoms are considered eligible for direct referral for endoscopy and assessment of the upper gastrointestinal tract.3 Despite these guidelines a huge degree of variability remains in referral patterns for endoscopy.4 The breath test for esophagogastric cancer aims to provide clinicians with an objective assessment of need for endoscopic referral. Given the association of all 5 VOCs with esophagogastric cancer, this may in the future allow for calculation of stratified risk for individual patients, which would require an independent large-scale study to fully validate. The consensus of key stakeholders in a decision workshop was to locate breath testing in primary care to triage patients with nonspecific symptoms to have endoscopy based on risk of OGC (eMethods 12 in Supplement 2). This view has been endorsed by our recent finding that the diagnostic model for OGC is different from that for colorectal cancer,26 providing the concept for a single breath test for multiple gastrointestinal cancers. If a clinician is presented with a patient with gastrointestinal symptoms that do not prompt referral based on NICE criteria, he/she would not need to watch and wait to see if symptoms worsen but could offer the exhaled breath test immediately. The clinician would order a breath test in much the same way as routine blood tests. A nurse can perform the test and send breath samples to a regional laboratory for analysis. A positive result would warrant immediate referral for endoscopy. A negative test would permit the clinician to reassure the patient and offer retesting if symptoms persist. Because endoscopy is an expensive investigation,1 with a low diagnostic yield of 2% to 5%,2 a triage breath test prior to endoscopy could substantially reduce the number of negative endoscopies and increase the cancer yield making the diagnostic pathway more effective with improved patient experience. Avoiding unnecessary investigations would also free up resources in the NHS.

w diagnostic yield of 2% to 5%,2 a triage breath test prior to endoscopy could substantially reduce the number of negative endoscopies and increase the cancer yield making the diagnostic pathway more effective with improved patient experience. Avoiding unnecessary investigations would also free up resources in the NHS. Concerns regarding clinical application of breath sampling and transport have led to the development of thermal desorption tubes, which allow breath samples to be stored for up to 1.5 months and transported between sites.27,28 These tubes can be used multiple times after cleaning and potentially for multiple diseases using the same analytical platform, which may serve to further lower the cost of a breath test. The mechanism of production of these VOCs in the cancer state may involve changes at a genetic and cellular level causing metabolic alterations in enzymatic pathways. Aldehydes have generated much research interest given their link as a possible carcinogen and also their elevation in other types of cancers.28 Genetic dysregulation of aldehyde metabolism is present in patients with esophageal cancer.29,30 Lipid peroxidation flux may provide a link between inflammation, aldehydes, and cancer.12,31 Gastric microbiome associated with cancer may also be a contributing factor to the production of VOCs, yet to be defined.32,33

.28 Genetic dysregulation of aldehyde metabolism is present in patients with esophageal cancer.29,30 Lipid peroxidation flux may provide a link between inflammation, aldehydes, and cancer.12,31 Gastric microbiome associated with cancer may also be a contributing factor to the production of VOCs, yet to be defined.32,33 Limitations There are limitations associated with this study that must be considered in the interpretation of the findings. Although demographic data were collected and regressed for in the analysis, there may have been other unmeasured confounding variables that could have influenced the changes in VOCs observed. Also, the reference standard test was histopathologically proven tissue diagnosis through endoscopy or from surgically resected specimen, although gastric and esophageal cancers can be missed in up to 8% of diagnostic endoscopies,34,35 however endoscopy remains the best diagnostic test currently available. Furthermore, most patients presented in the current study have T3 esophagogastric cancer, in line with disease patterns in the UK. Therefore the diagnostic accuracy of the test to identify early stage (T1) cancer remains undetermined by the current study. It must be acknowledged that given a current sensitivity of 80% there is still potential for further refinement of exhaled breath testing and thereby improvements in cancer detection rates; a successful evolution observed in other triage investigations such as stool DNA testing for colorectal cancer.36,37 Further investigations are also needed to examine the sensitivity of breath analysis on multiple testing samples in patients who initially have a negative result.

improvements in cancer detection rates; a successful evolution observed in other triage investigations such as stool DNA testing for colorectal cancer.36,37 Further investigations are also needed to examine the sensitivity of breath analysis on multiple testing samples in patients who initially have a negative result. Conclusion This validation study showed a sensitivity of 80% and a specificity of 81% for the breath test to diagnose esophagogastric cancer. The next stage is a large-scale diagnostic accuracy study among the primary care population where the test is intended to be employed. Supplement. eMethods 1: VOC breath model refinement eMethods 2: STARD 2015 list eMethods 3: Optimisation of Bag materials eMethods 4: Effect of ambient room air upon analysis of trace VOCs eMethods 5: Human factor analysis of breath bag sampling eMethods 6: Detection limit of SIFT-MS identified as 1ppbv eMethods 7: GC-MS analysis eMethods 8: Comparison of presenting symptom between cancer and control patients eMethods 9: GC-MS cross platform validation eMethods 10: Comparison of VOCs concentrations between cancer and controls and regression for confounding variables eMethods 11: Diagnostic accuracy of test based upon clinical parameters for NICE guidelines for endoscopy referral eMethods 12: Decision conferencing – patient pathway Click here for additional data file.

Introduction Thyroid nodules are common, with as many as two-thirds of adults harboring nodules detectable by ultrasound and about 5% by palpation.1,2 Ultrasound has proven useful in estimating the likelihood of malignant tumors and selecting nodules for fine-needle aspiration (FNA) biopsy.3,4,5,6,7 More than 600 000 thyroid FNAs are performed every year in the United States alone, and the number has been increasing annually by 16%.8,9 Thyroid FNA cytology can accurately classify most nodules as benign and a minority as malignant, but the results remain indeterminate in about 20% (range, 10%-38%) of nodules when cytological features lack specific characteristics needed for a definitive diagnosis.10 Further, the proportion of indeterminate cytology results appears to be rising.11

rately classify most nodules as benign and a minority as malignant, but the results remain indeterminate in about 20% (range, 10%-38%) of nodules when cytological features lack specific characteristics needed for a definitive diagnosis.10 Further, the proportion of indeterminate cytology results appears to be rising.11 The Bethesda System for Reporting Thyroid Cytopathology12,13 includes 2 common categories of indeterminate cytology, atypical or follicular lesion of undetermined significance (Bethesda category III) and follicular neoplasm/suspicious for follicular (or Hürthle cell) neoplasm (Bethesda category IV), each accounting for approximately 10% of all thyroid FNA results.10 The observed rates of cancer in these categories vary widely by institution, ranging from 6% to 48% for Bethesda III and 14% to 34% for Bethesda IV,10 which poses diagnostic uncertainty that greatly confounds patient treatment, often resulting in repeat FNA and/or unnecessary diagnostic surgery.3 Another cytologic category generally considered as indeterminate is suspicious for malignancy (Bethesda category V), which comprises 2% to 3% of all FNAs.10,12,13 The probability of cancer in these nodules is much higher, 53% to 97%,10 with surgery indicated in most cases, although the extent of surgery (thyroid lobectomy or total thyroidectomy with possible elective central lymph node dissection) could be informed by a more precise cancer probability assessment.3

FNAs.10,12,13 The probability of cancer in these nodules is much higher, 53% to 97%,10 with surgery indicated in most cases, although the extent of surgery (thyroid lobectomy or total thyroidectomy with possible elective central lymph node dissection) could be informed by a more precise cancer probability assessment.3 Most thyroid cancers are well differentiated and have an indolent clinical course and low mortality. As a result, limited surgery and lower intensity postsurgical treatment and surveillance may be considered for cancers with low to intermediate risk for recurrence.3 In addition, the histologic assessment (the gold standard for cytology) of benign vs malignant thyroid nodules is in transition with both intraobserved and interobserved variability rates that can be high14 and a recent change in nomenclature for the noninvasive encapsulated follicular variant of papillary thyroid cancer.15 Now variably considered as nonmalignant, premalignant, or possibly carcinoma in situ, and redefined as a noninvasive follicular thyroid neoplasm with papillary-like nuclear features (NIFTP),15,16 these tumors require surgery for diagnosis and treatment, but can usually be adequately treated by lobectomy.17,18 Furthermore, thyroid cancers driven by distinct mutations (most commonly BRAF V600E or RAS) differ with respect to their pathologic and clinical properties,19,20 and accumulation of additional mutations such as TERT may identify thyroid cancers with the highest risk for tumor recurrence and disease-specific mortality.21,22

hermore, thyroid cancers driven by distinct mutations (most commonly BRAF V600E or RAS) differ with respect to their pathologic and clinical properties,19,20 and accumulation of additional mutations such as TERT may identify thyroid cancers with the highest risk for tumor recurrence and disease-specific mortality.21,22 Over the past decade, molecular testing of thyroid nodules was developed to improve diagnostic accuracy of FNA cytology.23,24 The initial small gene mutation panels offered high PPV for cancer detection but lacked sufficiently high NPV to reliably exclude malignant disease in test-negative samples.25,26 More advanced molecular tests were subsequently developed using gene expression profiling, broader panels of mutational markers, or combinations of different markers.27,28,29,30,31 Overall, they offered a significantly improved sensitivity and NPV. However, they suffer from either relatively low specificity and PPV, particularly for certain types of thyroid cancer, such as Hürthle cell tumors, limited clinical validation, and/or lack of reporting specific molecular information for more refined cancer risk assessment.

a significantly improved sensitivity and NPV. However, they suffer from either relatively low specificity and PPV, particularly for certain types of thyroid cancer, such as Hürthle cell tumors, limited clinical validation, and/or lack of reporting specific molecular information for more refined cancer risk assessment. Recently, a new 112-gene test was developed (ThyroSeq v3 Genomic Classifier [GC]) to include a broad range of thyroid cancer-related point mutations, gene fusions, copy number alterations and gene expression alterations with the goals of achieving both high sensitivity and specificity in detecting all types of thyroid cancer and providing detailed genomic information on the nodules sampled by FNA biopsy.32 This prospective, blinded, multicenter clinical validation study was undertaken to assess the diagnostic performance of this GC test in cytologically indeterminate thyroid nodules. Methods Study Population Patients eligible for this study were aged 18 years or older, had 1 or more thyroid nodules, underwent a routine FNA procedure to collect samples for cytological examination, and agreed to provide material for molecular analysis. After FNA cytology was reported, only those patients who had at least 1 nodule that yielded a cytologic diagnosis of Bethesda III, IV, or V and underwent thyroid surgery to remove 1 or more nodules were included in the study.

ct samples for cytological examination, and agreed to provide material for molecular analysis. After FNA cytology was reported, only those patients who had at least 1 nodule that yielded a cytologic diagnosis of Bethesda III, IV, or V and underwent thyroid surgery to remove 1 or more nodules were included in the study. Study Design and Sample Collection This prospective cohort study recruited 782 patients with 1013 thyroid nodules clinically evaluated at 10 sites, 9 in the United States and 1 in Singapore, between January 2015 and December 2016. All FNA were performed using a 22g, 25g, or 27g needle depending on institutional practice. Samples were collected for molecular analysis by either (1) rinsing the residual material in the aspiration needle from all passes or (2) collecting a dedicated pass into a preservative solution tube (ThyroSeqPreserve) and stored at −20°C. Samples from nodules diagnosed as Bethesda III, IV, or V with surgical follow-up were retained as eligible and shipped to the University of Pittsburgh Medical Center (UPMC) for GC testing. Application of the eligibility criteria resulted in 256 patients with thyroid nodules that yielded 286 FNA samples available for molecular analysis (Figure 1). Central pathology review was performed on 274 (96%) nodules by a panel of expert thyroid pathologists (eMethods 1 in the Supplement). Figure 1. Recruitment and Exclusion of Patients and Samples in the Study FNA indicates fine-needle aspiration; TNA, total nucleic acids; NIFTP, noninvasive follicular thyroid neoplasm with papillary-like nuclear features.

Study Design and Sample Collection This prospective cohort study recruited 782 patients with 1013 thyroid nodules clinically evaluated at 10 sites, 9 in the United States and 1 in Singapore, between January 2015 and December 2016. All FNA were performed using a 22g, 25g, or 27g needle depending on institutional practice. Samples were collected for molecular analysis by either (1) rinsing the residual material in the aspiration needle from all passes or (2) collecting a dedicated pass into a preservative solution tube (ThyroSeqPreserve) and stored at −20°C. Samples from nodules diagnosed as Bethesda III, IV, or V with surgical follow-up were retained as eligible and shipped to the University of Pittsburgh Medical Center (UPMC) for GC testing. Application of the eligibility criteria resulted in 256 patients with thyroid nodules that yielded 286 FNA samples available for molecular analysis (Figure 1). Central pathology review was performed on 274 (96%) nodules by a panel of expert thyroid pathologists (eMethods 1 in the Supplement). Figure 1. Recruitment and Exclusion of Patients and Samples in the Study FNA indicates fine-needle aspiration; TNA, total nucleic acids; NIFTP, noninvasive follicular thyroid neoplasm with papillary-like nuclear features. The study was double-blinded; neither cytologists nor pathologists were aware of molecular analysis results and none of the personnel involved in performing molecular analysis were aware of cytology and histopathology results. The study was approved by the institutional review boards or ethics committees of all participating study sites. Written informed consent was obtained and patients were not compensated for participation. The study protocol is available (ClinicalTrials.gov identifier: NCT02352766).

f cytology and histopathology results. The study was approved by the institutional review boards or ethics committees of all participating study sites. Written informed consent was obtained and patients were not compensated for participation. The study protocol is available (ClinicalTrials.gov identifier: NCT02352766). Molecular Analysis The ThyroSeq v3 GC is a targeted next-generation sequencing test that interrogates selected regions of 112 thyroid cancer-related genes for point mutations, insertions/deletions, gene fusions, copy number alterations, or gene expression alterations.32 The assay was performed at the UPMC Molecular and Genomic Pathology laboratory.32 The genomic classifier was applied to assign a value to each detected genetic alteration based on the strength of association with malignancy: 0 (no association with cancer), 1 (low cancer probability), or 2 (high cancer probability). A GC score calculated for each sample is a sum of individual values of all detected alterations, with GC scores 0 and 1 accepted as test negative (score 1 commercially reported as currently negative) and scores 2 and above as test positive.32

ion with cancer), 1 (low cancer probability), or 2 (high cancer probability). A GC score calculated for each sample is a sum of individual values of all detected alterations, with GC scores 0 and 1 accepted as test negative (score 1 commercially reported as currently negative) and scores 2 and above as test positive.32 Study Outcomes The primary outcome was the sensitivity, specificity, NPV, and PPV of the multigene GC to predict the histopathologic diagnosis of benign nodule vs cancer/NIFTP in indeterminate thyroid nodules with Bethesda III and IV cytology. In data analysis, NIFTP was grouped together with cancer because it also represents a tumor type that requires surgery based on current practice guidelines.17,18 The secondary outcome was the prediction of cancer/NIFTP by specific genetic alterations in Bethesda III, IV, and V cytology nodules.

sda III and IV cytology. In data analysis, NIFTP was grouped together with cancer because it also represents a tumor type that requires surgery based on current practice guidelines.17,18 The secondary outcome was the prediction of cancer/NIFTP by specific genetic alterations in Bethesda III, IV, and V cytology nodules. Statistical Analysis For the primary and secondary outcomes, the test sensitivity, specificity, PPV, and NPV with 95% Wilson confidence intervals were calculated33 for individual nodules using the consensus diagnosis of central pathology as the reference standard. Using observed sensitivity and specificity, hypothetical positive and negative predictive value curves were calculated over the entire range (0%-100%) of possible disease prevalence. Among patients with nodules yielding indeterminate cytology, baseline characteristics of the included and excluded patients were compared using the Wilcoxon test and Fisher exact test. Statistical analysis was conducted with the R software package (version 3.4.2, R Foundation).34 Sample size justification and the programming code used to generate results are described in eMethods 2 in the Supplement.

cs of the included and excluded patients were compared using the Wilcoxon test and Fisher exact test. Statistical analysis was conducted with the R software package (version 3.4.2, R Foundation).34 Sample size justification and the programming code used to generate results are described in eMethods 2 in the Supplement. Results Patients and Nodules Of the 256 eligible patients, 202 were female (79%), with a median age of 53 years (range, 18-90 years); biopsied nodules had a median size of 2.4 cm (range, 0.5-7 cm). Among the 286 eligible samples, FNA cytology diagnosis was Bethesda III in 172, Bethesda IV in 101, and Bethesda V in 13 cases. Based on the results of central pathology review, 206 nodules (72%) were classified as benign, 69 (24%) as malignant, and 11 (4%) as NIFTP. The prevalence of conditions requiring surgery, ie, cancer and NIFTP, was 28% in the entire cohort, ranging from 9% to 60% among study sites (eTable 2 in the Supplement).

13 cases. Based on the results of central pathology review, 206 nodules (72%) were classified as benign, 69 (24%) as malignant, and 11 (4%) as NIFTP. The prevalence of conditions requiring surgery, ie, cancer and NIFTP, was 28% in the entire cohort, ranging from 9% to 60% among study sites (eTable 2 in the Supplement). Molecular Analysis Of 286 samples subjected to molecular analysis, 20 (7%) failed a presequencing step owing to low total nucleic acid quantity reflecting low sample cellularity, and 9 (3%) were inadequate on postsequencing analysis because the expression of thyroid cell markers was below the established acceptable level.32 Thus, 257 (90%) samples from 232 patients were informative for molecular analysis comprising the final study set. It included samples from 154 Bethesda III, 93 Bethesda IV, and 10 Bethesda V nodules. Molecular analysis yielded a negative test result in 152 (59%) samples and a positive result in 105 (41%) samples (eTable 4 in the Supplement). Among all 318 patients with indeterminate cytology, baseline characteristics of the included and excluded (Figure 1) patients and nodules were similar (eTable 3 in the Supplement). Overall Test Performance The primary outcome of this study was the accurate separation of histopathological benign nodules from cancer and NIFTP in samples with Bethesda III and IV cytology. Table 1 summarizes the test sensitivity, specificity, NPV, and PPV in these cytologic groups. Overall, in Bethesda III and IV nodules, a negative or benign call rate was 61%.

of this study was the accurate separation of histopathological benign nodules from cancer and NIFTP in samples with Bethesda III and IV cytology. Table 1 summarizes the test sensitivity, specificity, NPV, and PPV in these cytologic groups. Overall, in Bethesda III and IV nodules, a negative or benign call rate was 61%. Table 1. Performance of the Genomic Classifier Test in Cytologically Indeterminate Thyroid Nodules Performance in Bethesda III nodules (n = 154; disease prevalence 23%) Result Cancer+NIFTP (n = 35) Benign (n = 119) Test performance, % (95% CI) Positive 32 18 Sensitivity, 91 (77-97) Specificity, 85 (77-90) NPV, 97 (92-99) PPV, 64 (50-77) Negative 3 101 Performance in Bethesda IV nodules (n = 93; disease prevalence 35%) Result Cancer+NIFTP (n = 33) Benign (n = 60) Test performance, % (95% CI) Positive 32 15 Sensitivity, 97(85-100) Specificity, 75(63-84) NPV, 98(89-100) PPV, 68 (54-80) Negative 1 45 Performance in Bethesda III and IV nodules (n = 247; disease prevalence 28%) Result Cancer+NIFTP (n = 68) Benign (n = 179) Result Positive 64 33 Sensitivity, 94 (86-98) Specificity, 82 (75-87) NPV, 97 (93-99) PPV, 66 (56-75) Negative 4 146 Performance Across the Entire Cohort (n = 257; Disease Prevalence 30%) Result Cancer+NIFTP (n = 76) Benign (n = 181) Test performance, % (95% CI) Positive 71 34 Sensitivity, 93 (86-97) Specificity, 81 (75-86) NPV, 97 (93-99) PPV, 68 (58-76) Negative 5 147 Abbreviations: NIFTP, Noninvasive follicular thyroid neoplasm with papillary-like nuclear features; NPV, negative predictive value; PPV, positive predictive value.

Benign (n = 181) Test performance, % (95% CI) Positive 71 34 Sensitivity, 93 (86-97) Specificity, 81 (75-86) NPV, 97 (93-99) PPV, 68 (58-76) Negative 5 147 Abbreviations: NIFTP, Noninvasive follicular thyroid neoplasm with papillary-like nuclear features; NPV, negative predictive value; PPV, positive predictive value. Test performance in specific histopathologic types of thyroid nodules is presented in Table 2. Among nodules found to be benign after surgery, the test correctly classified as negative 84 of 95 (88%) hyperplastic follicular cell nodules, 5 of 5 (100%) hyperplastic Hürthle cell nodules, 37 of 47 (79%) follicular adenomas, and 21 of 34 (62%) Hürthle cell adenomas. The GC scores in these benign nodules were 0 in 86% and 1 in 14% (eFigure in the Supplement). In a subgroup of histologically benign nodules with Bethesda III-IV cytology, 146 of 179 (82%) were classified as negative.

ic Hürthle cell nodules, 37 of 47 (79%) follicular adenomas, and 21 of 34 (62%) Hürthle cell adenomas. The GC scores in these benign nodules were 0 in 86% and 1 in 14% (eFigure in the Supplement). In a subgroup of histologically benign nodules with Bethesda III-IV cytology, 146 of 179 (82%) were classified as negative. Table 2. Test Performance in Specific Histopathologic Types of Thyroid Lesions Histopathologic Diagnosis Nodules, No. (%) Test Correctly Classified, % (95% CI) Positive Negative Benign Hyperplastic follicular cell nodule 95 (37) 11 84 88 (80-93) Hyperplastic Hürthle cell nodule 5 (2) 0 5 100 (57-100) Follicular adenoma 47 (18) 10 37 79 (65-88) Hürthle cell adenoma 34 (13) 13 21 62 (45-76) NIFTP 11 (4) 11 0 100 (74-100)a Malignant Papillary thyroid carcinoma 49 (19) 45 4 92 (81-97) Follicular thyroid carcinoma 4 (2) 3 1 75 (30-99) Hürthle cell carcinoma 10 (4) 10 0 100 (72-100) Medullary thyroid carcinoma 1 (0.5) 1 0 100 (5-100) Metastatic carcinomab 1 (0.5) 1 0 100 (5-100) Total 257 (100) 105 152 85 (80-89) Abbreviation: NIFTP, Noninvasive follicular thyroid neoplasm with papillary-like nuclear features. a Considering positive test result for NIFTP as correct classification. b Metastatic renal cell carcinoma. All 11 NIFTP nodules were correctly classified as positive. Among malignant nodules, 45 of 49 (92%) papillary carcinomas, 3 of 4 (75%) follicular carcinomas, and 10 of 10 (100%) Hürthle cell carcinomas were correctly classified as positive. A medullary thyroid carcinoma and a metastatic renal cell carcinoma were also correctly identified.

were correctly classified as positive. Among malignant nodules, 45 of 49 (92%) papillary carcinomas, 3 of 4 (75%) follicular carcinomas, and 10 of 10 (100%) Hürthle cell carcinomas were correctly classified as positive. A medullary thyroid carcinoma and a metastatic renal cell carcinoma were also correctly identified. Of 152 test-negative samples in the study cohort, 5 (3%) were found to be false-negative, all having a GC score of 0. They included samples from 3 Bethesda III cytology nodules, 1 Bethesda IV, and 1 Bethesda V (eTable 5 in the Supplement). Among them, there were 4 papillary carcinomas and 1 minimally invasive follicular carcinoma. These were all T1 or T2 tumors (1-4 cm), intrathyroidal and without vascular invasion or clinical evidence of nodal or distant metastasis.

esda III cytology nodules, 1 Bethesda IV, and 1 Bethesda V (eTable 5 in the Supplement). Among them, there were 4 papillary carcinomas and 1 minimally invasive follicular carcinoma. These were all T1 or T2 tumors (1-4 cm), intrathyroidal and without vascular invasion or clinical evidence of nodal or distant metastasis. Cancer Probability in Specific Genetic Alteration Groups Among 105 cases with positive GC results, the probability of surgery-requiring disease, defined as cancer or NIFTP, varied depending on specific genetic alterations (Table 3). Two nodules had high-risk TERT or TP53 mutations, of which 1 was a widely invasive follicular carcinoma and the other was a multifocal papillary carcinoma on surgical pathology. Thirteen nodules were positive for either BRAF V600E mutation or NTRK3, BRAF, or RET fusion. Histopathologically, these were all cancers, primarily classical papillary carcinomas. Another 60 nodules were positive for RAS, BRAF K601E, PTEN, IDH2, or DICER1 mutation, or PPARG-THADA fusion. In this group, cancer/NIFTP was found in 37 of 60 (62%) cases and histologically benign nodules in 23 of 60 (38%); most of the cancers were follicular patterned, either follicular variant papillary or follicular carcinomas. Most common mutations (n = 45) involved RAS genes, which were associated with a diagnosis of cancer or NIFTP in 72% for HRAS, 52% for NRAS, and 40% for KRAS. Twenty-two nodules were positive for copy number alterations alone. Cancer/NIFTP was found in 13 (59%) of those, and this group was enriched in Hürthle cell carcinoma and follicular variant papillary carcinoma. Finally, 8 samples were positive for gene expression alterations alone (Table 3).

for NRAS, and 40% for KRAS. Twenty-two nodules were positive for copy number alterations alone. Cancer/NIFTP was found in 13 (59%) of those, and this group was enriched in Hürthle cell carcinoma and follicular variant papillary carcinoma. Finally, 8 samples were positive for gene expression alterations alone (Table 3). Table 3. Probability of Cancer/NIFTP in Specific Molecular Alteration Groups Group Molecular Alterations, No. Prevalence in Test-Positive Samples, No. (%) Histopathologic Diagnosis, % Cancer Type/NIFTP (%) Cancer/NIFTP Benign High-risk group TERT (and HRAS) (1) TP53 (and MEN1) (1) 2 (2) 100 0 Papillary carcinoma (50) Follicular carcinoma (50) BRAF-like group BRAF V600E (9) NTRK3 fusions (2) RET fusions (1) BRAF fusions (1) 13 (12) 100 0 Classical papillary carcinoma (92) Follicular variant papillary carcinoma (8) RAS-like group NRAS (21) HRAS (18) KRAS (5) EIF1AX (5) BRAF K601E (3) PTEN (1) IDH2 (1) DICER1 (1) PPARG fusions (4) THADA fusions (4) 60 (57) 62 38 Follicular variant papillary carcinoma (22) Papillary carcinoma, other variants (17) NIFTP (15) Follicular carcinoma (3) Hürthle cell carcinoma (5) Copy number alterations group Copy number alterations 22 (21) 59 41 Hürthle cell carcinoma (32) Follicular variant papillary carcinoma (14) Papillary carcinoma, other variants (9) NIFTP (5) Gene expression alterations group Gene expression alterations 8 (8) 75 25 Classical papillary carcinoma (37) NIFTP (13) Other cancers (MTC, mRCC) (25) Abbreviations: mRCC, metastatic renal cell carcinoma; MTC, medullary thyroid carcinoma; NIFTP, noninvasive follicular thyroid neoplasm with papillary-like nuclear features.

e expression alterations group Gene expression alterations 8 (8) 75 25 Classical papillary carcinoma (37) NIFTP (13) Other cancers (MTC, mRCC) (25) Abbreviations: mRCC, metastatic renal cell carcinoma; MTC, medullary thyroid carcinoma; NIFTP, noninvasive follicular thyroid neoplasm with papillary-like nuclear features. Among 34 test-positive nodules that were pathologically benign on surgery, 23 (67%) were adenomas and 11 (32%) hyperplastic nodules (eTable 6 in the Supplement). However, 32 of 34 (94%) of them showed 1 or more clonal molecular alterations (point mutation, gene fusion, or DNA copy number alterations) present in a large proportion of cells in the nodule, indicating that these nodules represented neoplasia and not hyperplasia.

%) hyperplastic nodules (eTable 6 in the Supplement). However, 32 of 34 (94%) of them showed 1 or more clonal molecular alterations (point mutation, gene fusion, or DNA copy number alterations) present in a large proportion of cells in the nodule, indicating that these nodules represented neoplasia and not hyperplasia. Discussion The main goal of molecular tests for thyroid FNA samples with indeterminate cytology is to correctly identify most of the benign nodules so that diagnostic thyroid surgery can be avoided in these patients. The safety of the approach is predicated by the test ability to detect all types of thyroid tumors and not miss high-risk cancers. The results of this prospective, blinded, multicenter study demonstrate that in nodules with Bethesda III or IV indeterminate cytology, the multigene GC test was highly sensitive (94%) and reasonably specific (82%) for discriminating benign from malignant/NIFTP nodules. With a baseline disease prevalence of 28%, the test yielded an NPV of 97% and a residual cancer risk of 3% in test-negative nodules, which is similar to an average 3.7% cancer risk in nodules diagnosed as benign by FNA cytology.10 Although no test has perfect accuracy, it is reassuring that all false-negative cases in the study were low-stage and low-risk cancers by the American Thyroid Association criteria.3

cer risk of 3% in test-negative nodules, which is similar to an average 3.7% cancer risk in nodules diagnosed as benign by FNA cytology.10 Although no test has perfect accuracy, it is reassuring that all false-negative cases in the study were low-stage and low-risk cancers by the American Thyroid Association criteria.3 Whereas sensitivity and specificity characterize a test independently of disease prevalence, NPV and PPV depend on the prevalence of disease in the studied population. Based on the fixed sensitivity and specificity, Bayes theorem can predict the test NPV and PPV along the spectrum of disease prevalence.35 For the GC test, it predicts a robust NPV of 95% or higher, required to consider nonsurgical treatment by the NCCN guidelines,36 up to a disease prevalence of 40% in Bethesda III and 60% in Bethesda IV nodules (Figure 2). This is within the range of cancer/NIFTP probability expected based on the Bethesda reporting system12,13 and observed in most clinical studies.10 Figure 2. Predicted Performance of Genomic Classifier (GC) Test in Populations With Different Cancer/NIFTP Prevalence Predicted negative predictive value (NPV) (solid blue lines) and positive predictive value (PPV) (solid orange lines) with 95% CIs (dotted lines) based on sensitivity and specificity of the multigene GC test established in this study for Bethesda III and IV cytology thyroid nodules. NIFTP indicates noninvasive follicular thyroid neoplasm with papillary-like nuclear features.

es) and positive predictive value (PPV) (solid orange lines) with 95% CIs (dotted lines) based on sensitivity and specificity of the multigene GC test established in this study for Bethesda III and IV cytology thyroid nodules. NIFTP indicates noninvasive follicular thyroid neoplasm with papillary-like nuclear features. Another commonly used molecular test for thyroid FNA samples is based on measuring expression of multiple genes either by the microarray assay (Gene Expression Classifier; GEC)27 or RNA-Seq (Gene Sequencing Classifier; GSC).37 In a comparable size validation studies, ThyroSeq GC shows an overall similar sensitivity (94% ThyroSeq vs 90% GEC and 91% GSC) but a specificity of 82% vs 52% in GEC and 68% in GSC (eTable 7 in the Supplement). Furthermore, ThyroSeq GC had a benign or negative call rate of 61% in indeterminate Bethesda III and IV nodules, with 82% of all histologically benign nodules yielding a negative test result. This indicates that ThyroSeq GC can prevent diagnostic surgeries for up to 61% of all of indeterminate Bethesda III to IV cytology nodules and as many as 82% of all benign nodules that yielded indeterminate cytology diagnosis. This should maximize the effect of molecular testing on the avoidance of surgery, reduction of health care costs, and improvement of patient quality of life. This is particularly important during what is widely considered as the era of thyroid cancer overdiagnosis38 and overtreatment.3

indeterminate cytology diagnosis. This should maximize the effect of molecular testing on the avoidance of surgery, reduction of health care costs, and improvement of patient quality of life. This is particularly important during what is widely considered as the era of thyroid cancer overdiagnosis38 and overtreatment.3 The multigene GC test showed robust performance in detecting all types of thyroid cancer, including Hürthle cell carcinoma. To date, the performance of existing molecular FNA tests in Hürthle cell nodules has been either not specifically reported,28,29,30 not validated at all,31 or observed to have very low specificity.39,40 In this study, all 10 Hürthle cell carcinomas were correctly classified, whereas in all types of Hürthle cell nodules the GC test negative call rate was 53%. This should allow the avoidance of diagnostic surgery in more than half of biopsied Hürthle cell nodules.

at all,31 or observed to have very low specificity.39,40 In this study, all 10 Hürthle cell carcinomas were correctly classified, whereas in all types of Hürthle cell nodules the GC test negative call rate was 53%. This should allow the avoidance of diagnostic surgery in more than half of biopsied Hürthle cell nodules. Another potential advantage of the GC test is that it provides a molecular profile of the test-positive nodules, which may help clinicians to refine the treatment of patients with Bethesda III, IV, and V nodules and a positive test. Indeed, the finding of BRAF V600E and similar alterations as well as high-risk (TERT, TP53) mutations conferred a 100% probability of cancer in this study, in keeping with previous reports.41,42,43 Tumors harboring a BRAF V600E mutation are classic papillary carcinoma with a higher rate of regional lymph node metastasis.3,19 On the contrary, RAS and RAS-like alterations were associated with a spectrum of follicular-pattern thyroid tumors, from pathologically benign adenomas to borderline NIFTP and fully invasive cancers, with a roughly 60% probability of cancer/NIFTP. These cancers are frequently encapsulated and if spread, they typically skip regional lymph nodes and metastasize hematogenously.19 However, most thyroid cancers driven by single RAS and RAS-like mutations are minimally invasive and low risk. The histologically benign nodules carrying these mutations are monoclonal tumors, in contrast to polyclonal hyperplastic nodules which are the most common type of benign thyroid nodules. Finally, the GC test correctly classified nodules composed of nonthyroid follicular cells, including medullary carcinoma and a metastatic tumor. This additional information on the test-positive nodules along with clinical factors may help to further individualize patient treatment.

ommon type of benign thyroid nodules. Finally, the GC test correctly classified nodules composed of nonthyroid follicular cells, including medullary carcinoma and a metastatic tumor. This additional information on the test-positive nodules along with clinical factors may help to further individualize patient treatment. As genetic information becomes available preoperatively, future studies are required to better understand how this information should be integrated with ultrasound and other clinical data to inform more tailored treatment of patients with thyroid nodules and cancers that have different molecular profiles. Furthermore, prospective studies will be needed to determine whether patients with the molecular signature of low-risk cancer or NIFTP can have surgery safely delayed or replaced by medical surveillance, as is currently under consideration for small thyroid cancers.44,45

cancers that have different molecular profiles. Furthermore, prospective studies will be needed to determine whether patients with the molecular signature of low-risk cancer or NIFTP can have surgery safely delayed or replaced by medical surveillance, as is currently under consideration for small thyroid cancers.44,45 Limitations This study has several limitations. By selecting patients based on the Bethesda reporting system for thyroid cytology, the applicability of the findings is limited to practices that use this reporting system. The observed small number of samples from Bethesda V nodules did not allow meaningful test validation in this subset of nodules. By surgically removing nodules with low cancer probability genetic alterations (GC score 1) for final histological diagnosis, the long-term clinical impact of these alterations could not be established. Finally, this study was performed at moderate- to high-volume centers with established thyroid nodule imaging and clinical expertise. Thus, the results may differ for practices that have a different setting and diagnostic approaches to thyroid nodules. Conclusions The study documents a high sensitivity and correspondingly high NPV of the ThyroSeq GC test for Bethesda III and IV indeterminate cytology nodules, which together with high specificity may prevent diagnostic surgeries in the majority of such patients. The availability of detailed genetic information in test-positive cases may help to further inform individualized treatment for these patients after integration with imaging and other clinical information.

y nodules, which together with high specificity may prevent diagnostic surgeries in the majority of such patients. The availability of detailed genetic information in test-positive cases may help to further inform individualized treatment for these patients after integration with imaging and other clinical information. Supplement. eMethods 1. Central pathology review eMethods 2. Supplemental statistical analysis eTable 1. Agreement between local pathology diagnosis and central pathology review diagnosis eTable 2. Distribution of samples and histopathology diagnoses contributed from 10 study sites eTable 3. Characteristics of patients and nodules in the study cohort eTable 4. Genomic classifier (GC) scores and alteration types detected in samples with negative and positive test results eTable 5. Demographic, clinical and pathological characteristics of cases with false negative test result eTable 6. Characteristics of 34 cases with false-positive test results eTable 7. Study characteristics and performance of ThyroSeq GC and Afirma GEC and GSC in Bethesda III and IV indeterminate cytology thyroid nodules eFigure 1. Distribution of GC scores in nodules with different histopathology Click here for additional data file.

eFigure 6. Comparison of Progression-Free Survival Between Patients in the bTMB-H and bTMB-L Groups With Different Cut-Points for bTMB eFigure 7. Comparison of Progression-Free Survival Between bTMB-H and bTMB-L With Different Cut-Points for bTMB in Patients With Anti-PD1/PD-L1 Therapy as First- or Second-Line Treatment eFigure 8. Association Between bTMB and Clinical Outcomes in Patients With Anti-PD1/PD-L1 Therapy as First- or Second-Line Treatment eTable 1. Detailed list of Genes in the NCC-GP150 Panel eTable 2. Summary of Pearson Correlation Between Different Public Panels and WES for Different Tumor Types eTable 3. Detailed Clinicopathologic Features of 48 NSCLC Patients for bTMB Technical Validation eTable 4. Summary of Sensitivity, Specificity and Youden's Index at Different Cutoffs for bTMB-H in the Matched Tissue/Blood TMB Comparison eTable 5. Detailed Clinicopathologic Features of 50 NSCLC Patients for bTMB Clinical Validation by bTMB Status eReferences Click here for additional data file.

Introduction High tumor mutational burden (TMB), which represents genomic instability, has the potential to induce neoantigen production and further immunogenicity improvement.1 Recent studies have confirmed that TMB measured by whole-exome sequencing (WES) or a next-generation sequencing (NGS) cancer gene panel (CGP) can serve as a candidate biomarker of clinical outcome from immune checkpoint blockades (ICBs)2,3 in melanoma,4,5 lung cancer,6,7,8,9 and urothelial carcinoma.10 However, a considerable proportion of patients with advanced cancer could not provide sufficient tumor tissue for molecular testing.8,11 Therefore, whether TMB can be measured using circulating tumor DNA (ctDNA) (namely, blood TMB [bTMB]) as a noninvasive approach to guide ICB therapies has attracted widespread attention from practitioners. A previous study12 found that blood-derived variants of unknown significance were correlated with ICB responses. More recently, Gandara et al13 reported that bTMB is associated with progression-free survival (PFS) benefit from atezolizumab over docetaxel in non–small cell lung cancers (NSCLCs). Further evaluation of bTMB will be performed in patients undergoing first-line treatment in a prospective, phase 3 randomized clinical trial (A Study of Atezolizumab as First-line Monotherapy for Advanced or Metastatic Non–Small Cell Lung Cancer [B-F1RST]).14 However, the reliability of ctDNA detection is still under debate15,16; thus, more supported evidence of bTMB is needed to promote its clinical value on guiding ICB delivery. In this study, we aimed to explore the optimal gene panel size and algorithm to design a CGP for TMB estimation, evaluate the panel reliability, and further validate the feasibility of bTMB as a clinically actionable biomarker for immunotherapy.

e of bTMB is needed to promote its clinical value on guiding ICB delivery. In this study, we aimed to explore the optimal gene panel size and algorithm to design a CGP for TMB estimation, evaluate the panel reliability, and further validate the feasibility of bTMB as a clinically actionable biomarker for immunotherapy. Methods Study Design This study contained 4 sections (eFigure 1 in the Supplement), including panel design (named NCC-GP150), virtual validation (eMethods 1 in the Supplement), technical validation, and clinical validation (eMethods 2 in the Supplement). The WES data from The Cancer Genome Atlas (TCGA) were used for panel design and virtual validation. Tumor mutational burden estimated by NCC-GP150 was compared with those by established gene panels, including Memorial Sloan Kettering Cancer Center’s Integrated Mutation Profiling of Actionable Cancer Targets (MSK-IMPACT), FoundationOne CDx (F1CDx), Guardant360, PlasmaSELECT 64, and FoundationACT (Assay for Circulating Tumor DNA). A public NSCLC cohort from Rizvi et al6 was used to evaluate the performance of NCC-GP150–based TMB to stratify ICB survival outcomes. Patients with NSCLC with sufficient tumor tissue samples and matched plasma samples were enrolled for technical validation to investigate the correlation between bTMB from the NCC-GP150 panel and tTMB from WES (eFigure 2A in the Supplement). Last, an independent cohort of patients with advanced NSCLC who were undergoing on-study anti–programmed cell death 1 (anti–PD-1) and anti–programmed cell death ligand 1 (anti–PD-L1) therapy was analyzed to validate the utility of bTMB by NCC-GP150 in identifying patients who could benefit from anti–PD-1 and anti–PD-L1 therapy (eFigure 2B and eMethods 3 in the Supplement). The study was performed from July 19, 2016, to April 20, 2018. For more details about the methods for DNA extraction, library preparation, target capture and DNA sequencing, WES analysis pipeline, and bTMB detection pipeline, see eMethods 4 to 7 in the Supplement. This study was approved by the ethics committees of the National Cancer Center, and all patients provided written informed consent. All data were deidentified.

A extraction, library preparation, target capture and DNA sequencing, WES analysis pipeline, and bTMB detection pipeline, see eMethods 4 to 7 in the Supplement. This study was approved by the ethics committees of the National Cancer Center, and all patients provided written informed consent. All data were deidentified. Statistical Analysis Correlations of TMB between WES and gene panels with different gene counts and algorithms were examined by the Pearson correlation coefficient (r2). The correlation between tissue-based TMB by WES and ctDNA-based bTMB by NCC-GP150 was determined by the Spearman rank correlation coefficient. Categorical variables were expressed as percentages, means (SDs) were provided for normally distributed data, and medians (interquartile ranges) were provided for data that are not normally distributed. Differences between the 2 groups were examined by the 2-tailed, unpaired t test for normally distributed variables or by the Mann-Whitney test for nonnormally distributed variable. The χ2 test or Fisher exact test was used to test the difference of categorical variables between the 2 groups. For PFS analysis, Kaplan-Meier curves were compared by using a log-rank test, and the hazard ratio (HR) was determined through a Cox proportional hazards regression model. The proportionality assumption was verified. Logistic regression was used to test the correlations between different variables and the objective response rate (ORR), with the results presented as odds ratios (ORs) and 95% CIs. Baseline variables that achieved a level of significance of P < . 05 in the univariable analysis were entered into multivariable models. All reported P values were 2-tailed, and P < .05 was considered to be statistically significant. Statistical analyses were performed with GraphPad Prism software, version 5.0 (GraphPad Software Inc) and R software, version 3.5.0 (R Foundation for Statistical Computing).

analysis were entered into multivariable models. All reported P values were 2-tailed, and P < .05 was considered to be statistically significant. Statistical analyses were performed with GraphPad Prism software, version 5.0 (GraphPad Software Inc) and R software, version 3.5.0 (R Foundation for Statistical Computing). Results This study used 2 independent cohorts of patients with NSCLC (cohort 1: 48 patients; mean [SD] age, 60 [13] years; 15 [31.2%] female; cohort 2: 50 patients; mean [SD] age, 58 [8] years; 15 [30.0%] female) to examine the correlation between bTMB estimated by NCC-GP150 and tTMB measured by WES and to identify the utility of bTMB estimated by NCC-GP150 in distinguishing patients who would benefit from anti–PD-1 and anti–PD-L1 therapy. Correlation of the NCC-GP150 Panel With TCGA-Based WES Data for TMB Estimation With the WES data of 9205 samples from TCGA, randomized genes were extracted to generate panels for TMB estimation and compared with WES-based TMB. With the increase in randomized gene number included in TMB calculation, a gradually increasing correlation between the panel- and WES-based TMB was observed along with a decreasing SD, reaching a plateau when 150 genes were included (Figure 1A). An NGS CGP, named NCC-GP150, was designed that covered whole exon regions of 150 selected cancer-related genes (eTable 1 in the Supplement). NCC-GP150 exhibited a better performance than most of the randomly sampled panels in most cancer types based on TCGA data (Figure 1B).

au when 150 genes were included (Figure 1A). An NGS CGP, named NCC-GP150, was designed that covered whole exon regions of 150 selected cancer-related genes (eTable 1 in the Supplement). NCC-GP150 exhibited a better performance than most of the randomly sampled panels in most cancer types based on TCGA data (Figure 1B). Figure 1. Panel Design and Virtual Validation of the Association Between Blood and Tissue Tumor Mutational Burden (TMB) A, Distribution of Pearson correlations between whole-exome sequencing (WES) and different numbers of genes randomly chosen (50 times) from gene panels. Syn + and Syn − indicate that synonymous mutations were included or excluded, respectively. B, Comparison of the performance of the NCC-GP150 panel and 150 randomly extracted genes (RAN150) among different tumor types. C, Pearson correlation of TMB between different public cancer gene panels and WES. D, Progression-free survival (PFS) by TMB status based on NCC-GP150 genes in the cohort studied by Rizvi et al.6 F1CDx indicates FoundationOne CDx; MSK-IMPACT, Memorial Sloan Kettering Cancer Center’s Integrated Mutation Profiling of Actionable Cancer Targets.

of TMB between different public cancer gene panels and WES. D, Progression-free survival (PFS) by TMB status based on NCC-GP150 genes in the cohort studied by Rizvi et al.6 F1CDx indicates FoundationOne CDx; MSK-IMPACT, Memorial Sloan Kettering Cancer Center’s Integrated Mutation Profiling of Actionable Cancer Targets. Subsequently, NCC-GP150 was compared with 5 established NGS gene panels, including MSK-IMPACT (468 cancer-related genes), F1CDx (324 cancer-related genes), Guardant360 (73 cancer-related genes), PlasmaSELECT 64 (64 cancer-related genes), and FoundationACT (62 cancer-related genes). The analysis of overlapping genes among these panels is shown in eFigure 3 in the Supplement. Among these gene panels, MSK-IMPACT exhibited a leading correlation with WES-based TMB (r2 = 0.97), followed by F1CDx (r2 = 0.96) and NCC-GP150 (r2 = 0.96) (Figure 1C). The performance of NCC-GP150 for TMB estimation in various cancers by virtual validation is given in eTable 2 in the Supplement. Considering the different prevalence of oncogenic driver mutations in NSCLC between Asian and white populations,17 we conducted another virtual validation of NCC-GP150 with established panels between TCGA NSCLC populations with and without EGFR (OMIM 131550) and/or KRAS (OMIM 190070) driver mutation. The NCC-GP150 demonstrated a consistently satisfactory performance for bTMB estimation vs TMB from TCGA WES data (eFigure 4 in the Supplement).

ed another virtual validation of NCC-GP150 with established panels between TCGA NSCLC populations with and without EGFR (OMIM 131550) and/or KRAS (OMIM 190070) driver mutation. The NCC-GP150 demonstrated a consistently satisfactory performance for bTMB estimation vs TMB from TCGA WES data (eFigure 4 in the Supplement). Next, we used a published clinical data set6 that included 34 patients with NSCLC treated with PD-1 to further test the practicability of NCC-GP150. The PFS was significantly longer in patients with high TMB (TMB greater than the median: PFS, 14.5 months; 95% CI, 8.3 months to not reached [NR]) than in patients with low TMB (TMB less than the median: PFS, 5.2 months; 95% CI, 2.1-8.3 months), with an HR of 0.36 (95% CI, 0.14-0.93, log-rank P = .03) (Figure 1D). Correlation of bTMB Estimated by NCC-GP150 With tTMB Calculated by WES To investigate the reliability of bTMB from ctDNA-derived sequencing by the NCC-GP150 panel, 48 patients with advanced NSCLC with qualified tumor tissue samples and matched plasma samples provided for synchronous WES and NCC-GP150 sequencing, respectively, were enrolled for technical validation (cohort 1) (eFigure 2A and eTable 3 in the Supplement). The Spearman correlation coefficient between NCC-GP150–based bTMB and WES-based TMB reached 0.62 (eFigure 5A in the Supplement). With the TMB median (75 for WES) as the cut point, we found that a bTMB of 6 or higher had an optimal Youden index of 0.59, with a sensitivity of 0.88 and a specificity of 0.71 (eFigure 5B and eTable 4 in the Supplement).

icient between NCC-GP150–based bTMB and WES-based TMB reached 0.62 (eFigure 5A in the Supplement). With the TMB median (75 for WES) as the cut point, we found that a bTMB of 6 or higher had an optimal Youden index of 0.59, with a sensitivity of 0.88 and a specificity of 0.71 (eFigure 5B and eTable 4 in the Supplement). bTMB Estimated by the NCC-GP150 Panel and Clinical Outcomes of NSCLC Treated With ICBs To unravel whether bTMB could identify patients benefiting from ICB therapy, another independent cohort of 50 patients with advanced NSCLC treated with anti–PD-1 and anti–PD-L1 agents (cohort 2) (eFigure 2B and eTable 5 in the Supplement) was used for analysis. Methods for the assessment of clinical outcomes are provided in eMethods 3 in the Supplement.

ify patients benefiting from ICB therapy, another independent cohort of 50 patients with advanced NSCLC treated with anti–PD-1 and anti–PD-L1 agents (cohort 2) (eFigure 2B and eTable 5 in the Supplement) was used for analysis. Methods for the assessment of clinical outcomes are provided in eMethods 3 in the Supplement. The bTMB and PD-L1 expression were not correlated (eTable 5 in the Supplement), which is consistent with previous results.8,13,18,19 When the bTMB cut point was set to 6, both HRs and P values reached a minimum (eFigure 6 in the Supplement). Compared with patients with low bTMB (bTMB<6, n = 22), patients with high bTMB (bTMB≥6, n = 28) demonstrated superior PFS (high bTMB: PFS, NR; 95% CI, 2.8 months to NR; low bTMB: PFS, 2.9 months; 95% CI, 2.7 months to NR; HR, 0.39; 95% CI, 0.18-0.84; log-rank P = .01) (Figure 2A) and were more likely to undergo tumor shrinkage (Figure 2B). In addition, high bTMB (39.3%; 95% CI, 23.9%-56.5%) was associated with a higher ORR than was low bTMB (9.1%; 95% CI, 1.6%-25.9%; P = .02) (Figure 2C). Similarly, responders had significantly higher bTMB levels than did nonresponders (Mann-Whitney P = .02) (Figure 2D).

ore likely to undergo tumor shrinkage (Figure 2B). In addition, high bTMB (39.3%; 95% CI, 23.9%-56.5%) was associated with a higher ORR than was low bTMB (9.1%; 95% CI, 1.6%-25.9%; P = .02) (Figure 2C). Similarly, responders had significantly higher bTMB levels than did nonresponders (Mann-Whitney P = .02) (Figure 2D). Figure 2. Clinical Validation of the Association Between NCC-GP150–Derived Blood Tumor Mutational Burden (bTMB) and Clinical Benefit in Patients With Non–Small Cell Lung Cancer A, Progression-free survival (PFS) by bTMB status. B, Waterfall plot of observed best response from anti–programmed cell death 1 (anti–PD-1) and anti–programmed cell death ligand 1 (anti–PD-L1) checkpoint inhibitors. C, Comparison of objective response rates (ORRs) between the high and low bTMB groups (P = .02). D, Comparison of bTMB level between nonresponse and response groups (P = .02). HR indicates hazard ratio.

anti–programmed cell death 1 (anti–PD-1) and anti–programmed cell death ligand 1 (anti–PD-L1) checkpoint inhibitors. C, Comparison of objective response rates (ORRs) between the high and low bTMB groups (P = .02). D, Comparison of bTMB level between nonresponse and response groups (P = .02). HR indicates hazard ratio. In the univariable Cox proportional hazards regression model, the Eastern Cooperative Oncology Group (ECOG) and treatment lines were also associated with PFS (ECOG: HR, 2.67; 95% CI, 1.21-5.88; P = .02; treatment lines: HR, 4.50; 95% CI, 2.05-9.89; P < .001), and the association between PD-L1 of 1% or higher and benefits for immunotherapy tended to be significant (HR, 0.49; 95% CI, 0.21-1.15; P = .10) (Table). In the multivariate Cox proportional hazards regression model that included bTMB, ECOG, and treatment lines, the association between bTMB and PFS remained significant (HR, 0.44; 95% CI, 0.20-0.99; P = .05). In the multivariate logistic regression analysis that included ECOG and treatment lines, bTMB status was also positively associated with an ORR (odds ratio, 11.69; 95% CI, 2.16-111.6; P = .01) (Table).

and treatment lines, the association between bTMB and PFS remained significant (HR, 0.44; 95% CI, 0.20-0.99; P = .05). In the multivariate logistic regression analysis that included ECOG and treatment lines, bTMB status was also positively associated with an ORR (odds ratio, 11.69; 95% CI, 2.16-111.6; P = .01) (Table). Table. Univariable and Multivariable Analysis of Progression-Free Survival and Objective Response Ratesa Parameter Progression-Free Survival Objective Response Rate Univariable Analysis Multivariable Analysis Univariable Analysis Multivariable Analysis HR (95% CI) P Value HR (95% CI) P Value OR (95% CI) P Value OR (95% CI) P Value Age ≥65 vs <65 y 0.62 (0.21-1.79) .37 NA NA 2.84 (0.60-13.12) .17 NA NA Male vs female 0.62 (0.28-1.34) .22 NA NA 2.98 (0.67-21.17) .20 NA NA ECOG performance status ≥2 vs 1 or 0 2.67 (1.21-5.88) .02 2.31 (1.08-4.95) .03 0.46 (0.12-1.57) .23 0.35 (0.04-1.89) .25 ≥3 vs <3 Metastatic sites 0.83 (0.39-1.75) .62 NA NA 1.23 (0.34-4.51) .75 NA NA LDH≥250 vs <250 U/L 1.19 (0.55-2.55) .66 NA NA 1.30 (0.33-4.80) .69 NA NA PD-L1 status ≥1% vs <1% 0.49 (0.21-1.15) .10 NA NA 2.47 (0.49-18.6) .31 NA NA Current or former vs never smoker 0.86 (0.41-1.80) .69 NA NA 1.69 (0.47-6.51) .43 NA NA bTMB≥6 vs <6 0.39 (0.18-0.84) .02 0.44 (0.20-0.99) .05 6.47 (1.48-45.72) .03 11.69 (2.16-111.6) .01 ≥3 vs 1 or 2 Lines of PD-1/PD-L1 blocked therapy 4.50 (2.05-9.89) <.001 3.34 (1.50-7.43) .003 0.11 (0.01-0.64) .04 0.11 (0.006-0.79) .06 Abbreviations: bTMB, blood tumor mutational burden; ECOG, Eastern Cooperative Oncology Group; HR, hazard ratio; LDH, lactate dehydrogenase; NA, not applicable; OR, odds ratio; PD-1, programmed cell death 1; PD-L1, programmed cell death ligand 1.

89) <.001 3.34 (1.50-7.43) .003 0.11 (0.01-0.64) .04 0.11 (0.006-0.79) .06 Abbreviations: bTMB, blood tumor mutational burden; ECOG, Eastern Cooperative Oncology Group; HR, hazard ratio; LDH, lactate dehydrogenase; NA, not applicable; OR, odds ratio; PD-1, programmed cell death 1; PD-L1, programmed cell death ligand 1. SI conversion factor: To convert lactate dehydrogenase to microkatals per liter, multiply by 0.0167. a Baseline variables that achieved a level of significance of P < .05 in the univariable analysis were entered into multivariable models.

89) <.001 3.34 (1.50-7.43) .003 0.11 (0.01-0.64) .04 0.11 (0.006-0.79) .06 Abbreviations: bTMB, blood tumor mutational burden; ECOG, Eastern Cooperative Oncology Group; HR, hazard ratio; LDH, lactate dehydrogenase; NA, not applicable; OR, odds ratio; PD-1, programmed cell death 1; PD-L1, programmed cell death ligand 1. SI conversion factor: To convert lactate dehydrogenase to microkatals per liter, multiply by 0.0167. a Baseline variables that achieved a level of significance of P < .05 in the univariable analysis were entered into multivariable models. We further explored the association between bTMB and PFS in first-line or second-line NSCLC as a subgroup analysis. The HRs and P values still reached a minimum, with a high bTMB cut point of 6 (HR, 0.08; 95% CI, 0.02-0.36; P = .001) (eFigure 7 in the Supplement). Patients with a bTMB of 6 or higher in this subgroup yielded significantly prolonged PFS (NR) compared with patients with a bTMB of less than 6 (2.9 months; 95% CI, 2.7 months to NR; HR, 0.08; 95% CI, 0.02-0.36; log-rank P < .001) (eFigure 8A in the Supplement). In addition, patients with a bTMB of 6 or higher had an increased ORR (61.1%; 95% CI, 39.2%-80.1%) compared with patients with a bTMB less than 6 (6.7%, 95% CI, 0.3%-27.9%; P = .003) (eFigure 8B in the Supplement). Responders had significantly higher bTMB levels (median, 10; interquartile range, 7-13) than did nonresponders (median, 5; interquartile range, 3-9; Mann-Whitney P = .008) (eFigure 8C in the Supplement). Collectively, bTMB measured by the NCC-GP150 panel was confirmed to be a potential clinical actionable biomarker for ICB therapy in patients with NSCLC.

bTMB levels (median, 10; interquartile range, 7-13) than did nonresponders (median, 5; interquartile range, 3-9; Mann-Whitney P = .008) (eFigure 8C in the Supplement). Collectively, bTMB measured by the NCC-GP150 panel was confirmed to be a potential clinical actionable biomarker for ICB therapy in patients with NSCLC. Discussion Our findings suggest that NCC-GP150, a panel through rational design on TCGA, could be used for bTMB estimation as a surrogate for WES-based TMB. We also validated bTMB as a potential biomarker to identify patients with NSCLC who could obtain significant improvements from immunotherapy. Blood TMB profiled with ctDNA sequencing is a promising strategy for TMB and ICB response estimation. Most recently, Gandara et al13 published the first study, to our knowledge, to identify bTMB of 16 or higher as an indicator of PFS benefit in patients with NSCLC treated with atezolizumab vs docetaxel. However, several key questions remained to be answered, including the suitable panel size and the kind of variants that should be included for TMB calling. In addition, in the reported study, analyses were made between bTMB and tTMB that were derived from the same gene panel but not with WES-based TMB as the criterion standard.

several key questions remained to be answered, including the suitable panel size and the kind of variants that should be included for TMB calling. In addition, in the reported study, analyses were made between bTMB and tTMB that were derived from the same gene panel but not with WES-based TMB as the criterion standard. We present the first study, to our knowledge, to systematically explore the optimal gene panel size and algorithm of a CGP design for TMB (especially bTMB) estimation. With WES data from TCGA, we found that a minimum gene panel size of 150 was sufficient for TMB estimation. The NCC-GP150 panel was then established through rational design with 150 selected genes and performed at the forefront for TMB estimation among most random-sampling models. We also found that the incorporation of synonymous mutations into panel-derived TMB calculation enhanced its correlation with WES, which agrees with the previous finding.20 Furthermore, in our independent cohort 1, we validated the satisfactory correlation of NCC-GP150–based bTMB with WES-based tTMB. Taken together, NCC-GP150 with a smaller panel size and satisfactory performance may be more accessible for clinic use with superior cost-effectiveness.

ch agrees with the previous finding.20 Furthermore, in our independent cohort 1, we validated the satisfactory correlation of NCC-GP150–based bTMB with WES-based tTMB. Taken together, NCC-GP150 with a smaller panel size and satisfactory performance may be more accessible for clinic use with superior cost-effectiveness. Our study also found that bTMB may be a potential biomarker to differentiate patients benefiting from anti–PD-1 and anti–PD-L1 therapy. The utility of the NCC-GP150 panel seemed to be more significant when ICBs were used as a first- or second-line rather than a later-line treatment. In addition, ECOG performance status and treatment lines were found to be negatively correlated with PFS, suggesting that patients with previous treatment might receive limited benefit from ICB therapy, which agrees with previous reports.21,22 Therefore, early use of therapy with ICBs should be considered in clinical practice.

ion, ECOG performance status and treatment lines were found to be negatively correlated with PFS, suggesting that patients with previous treatment might receive limited benefit from ICB therapy, which agrees with previous reports.21,22 Therefore, early use of therapy with ICBs should be considered in clinical practice. Limitations There are several limitations to our study. First, the clinical validation was retrospective, and the limited sample sizes might yield statistical bias. Second, the clinical cohort was obtained from different trials, including anti–PD-1 and anti–PD-L1 treatment, which might potentially influence the ultimate survival outcomes. Third, we used TCGA data for virtual validation but a Chinese cohort for technical and clinical validation. The different population-based prevalence in oncogenic driver mutations may have a potential confounding effect on the validity of analyses,17 although NCC-GP150 still demonstrated consistently satisfactory performance for TCGA based-TMB estimation and Chinese cohort–based bTMB estimation after excluding patients with EGFR/KRAS driver mutations. Conclusions We developed and evaluated the clinical feasibility of the NCC-GP150 panel for TMB estimation through virtual, technical, and clinical validation. The findings suggest that a ctDNA-based bTMB measured by the NCC-GP150 panel could be used as a potential biomarker for anti–PD-1 and anti–PD-L1 treatment in patients with NSCLC. Supplement. eMethods 1. Gene Panel Design and Virtual Validation eMethods 2. Technical Validation and Clinical Validation

Conclusions We developed and evaluated the clinical feasibility of the NCC-GP150 panel for TMB estimation through virtual, technical, and clinical validation. The findings suggest that a ctDNA-based bTMB measured by the NCC-GP150 panel could be used as a potential biomarker for anti–PD-1 and anti–PD-L1 treatment in patients with NSCLC. Supplement. eMethods 1. Gene Panel Design and Virtual Validation eMethods 2. Technical Validation and Clinical Validation eMethods 3. Assessment of Clinical Outcomes eMethods 4. DNA Extraction eMethods 5. Library Preparation, Target Capture and DNA Sequencing eMethods 6. WES Analysis Pipeline eMethods 7. bTMB Detection Pipeline eFigure 1. Flow Diagram of the Study eFigure 2. Flow Diagram for the Technical and Clinical Validation Cohorts eFigure 3. Venn Diagram and Flow Diagram Showing the Numbers and Percentages of Overlapping Gene Numbers Among NCC-GP150, F1CDx, MSK-IMPACT, and Guardant360 eFigure 4. Comparison of Panel Performance Between TCGA NSCLC Populations With and Without EGFR or KRAS Driver Mutation eFigure 5. Technical Validation of the Association Between bTMB Estimated by NCC-GP150 and WES eFigure 6. Comparison of Progression-Free Survival Between Patients in the bTMB-H and bTMB-L Groups With Different Cut-Points for bTMB eFigure 7. Comparison of Progression-Free Survival Between bTMB-H and bTMB-L With Different Cut-Points for bTMB in Patients With Anti-PD1/PD-L1 Therapy as First- or Second-Line Treatment eFigure 8. Association Between bTMB and Clinical Outcomes in Patients With Anti-PD1/PD-L1 Therapy as First- or Second-Line Treatment

Immune checkpoint inhibitors (ICIs) that target the programmed death 1 receptor (anti–programmed cell death 1 [PD-1] therapy) have ushered in a new era of cancer therapy. However, their application has been curtailed by serious immune-related adverse events (irAEs), such as colitis, pneumonitis, and myocarditis, that remain largely unpredictable. Although the use of tumor mutational burden (TMB) as a biomarker for expected therapy response has been advocated, a similar parameter for irAEs is lacking. In an attempt to fill this clinically relevant knowledge gap, we investigated the association between irAEs reported during anti–PD-1 therapy and TMB by comparing large-scale surveillance data of irAEs with the median TMB across multiple cancer types.

sponse has been advocated, a similar parameter for irAEs is lacking. In an attempt to fill this clinically relevant knowledge gap, we investigated the association between irAEs reported during anti–PD-1 therapy and TMB by comparing large-scale surveillance data of irAEs with the median TMB across multiple cancer types. Methods We retrieved postmarketing data of adverse events from the US Food and Drug Administration Adverse Event Reporting System (FAERS) from July 1, 2014, to March 31, 2019. According to the ethics committee policy of the EKOS (Ethikkommission Ostschweiz, Switzerland), this study was exempt from ethical review because all analyzed datasets are deidentified and publicly available. We considered only reports for which the anti–PD-1 agents nivolumab or pembrolizumab were the suspected cause of adverse events. Anti–PD-1 and anti-cytotoxic T-lymphocyte–associated protein 4 combination treatment was excluded. Closely related indications were aggregated to unified terms; for example, “malignant melanoma” was aggregated to “melanoma.” To limit our analysis to irAEs, we filtered terms to match broadly accepted diagnoses that were outlined in peer-reviewed irAE management guidelines. The median TMB in tumor tissue was obtained from previously published comprehensive genomic profiling. Lastly, we only considered cancers for which there were at least 100 cases of adverse events during anti–PD-1 therapy reported in FAERS. To assess the risk of a patient developing any irAE, we estimated reporting odds ratios (RORs) by comparing the odds of reporting these irAEs rather than others for the anti–PD-1 agents with the odds for all other drugs in the database, which represents standard practice for quantitative analyses of data in FAERS and similar databases.

k of a patient developing any irAE, we estimated reporting odds ratios (RORs) by comparing the odds of reporting these irAEs rather than others for the anti–PD-1 agents with the odds for all other drugs in the database, which represents standard practice for quantitative analyses of data in FAERS and similar databases. Results Our search strategy identified a total of 47 304 adverse events (AEs) in 16 397 patients reported as treated with anti–PD-1 monotherapy for 19 different cancer types. Of these patients, 3661 had at least 1 irAE (22.3%; 95% CI, 21.7-23.0). The comparator group comprised 16 411 749 AE reports from 5 160 064 patients. Our analysis revealed a significant positive correlation between the ROR of reporting an irAE during anti–PD-1 therapy and the corresponding TMB across multiple cancer types, with a higher ROR of irAE associated with a higher median number of coding somatic mutations per megabase of DNA (Figure; Pearson correlation coefficient R = 0.704; P < .001). The correlation coefficient suggests that 50% of the differences in the irAE risk across cancer types may be attributed to the TMB.

ancer types, with a higher ROR of irAE associated with a higher median number of coding somatic mutations per megabase of DNA (Figure; Pearson correlation coefficient R = 0.704; P < .001). The correlation coefficient suggests that 50% of the differences in the irAE risk across cancer types may be attributed to the TMB. Figure. Association of Tumor Mutational Burden With Immune-Related Adverse Events During Anti–PD-1 Therapy Across Multiple Cancers The x-axis indicates the tumor mutational burden (TMB)—defined as the median number of coding somatic mutations per megabase of DNA—across 19 cancer types. Data on the x-axis are presented on a logarithmic scale. The y-axis shows the reporting odds ratio of any immune-related adverse event (irAE) across cancer types, calculated using data from the US Food and Drug Administration Adverse Event Reporting System (FAERS) database. The dashed line represents the 95% CI of the linear fit. Circle size represents the total number of FAERS cases for each cancer type, and the color indicates the total number of tumor samples used to measure TMB for each cancer type.

m the US Food and Drug Administration Adverse Event Reporting System (FAERS) database. The dashed line represents the 95% CI of the linear fit. Circle size represents the total number of FAERS cases for each cancer type, and the color indicates the total number of tumor samples used to measure TMB for each cancer type. Discussion Our analysis indicates that cancers with a high TMB, such as melanoma and non–small cell lung cancer, are associated with a higher irAE ROR during anti–PD-1 therapy, strongly suggesting that these cancers are associated with a higher risk of irAEs than cancers with a low TMB. A possible explanation for this finding may be the different neoantigenic load across cancer types. Additionally, studies have shown that T cells that react against a neoantigen can crossreact against the corresponding wild-type protein. Another contributing mechanism may be antigen spreading, where tumor cell death releases antigens, including neoantigens, that prime lymphocytes against the wild-type antigens in healthy tissue. Given the results of the analysis, we propose that the association between irAEs and improved response to anti–PD-1 treatment are linked via an underlying neoantigenic potential that stems from a high TMB. A limitation of the study is the use of spontaneous reports for indirectly measuring the risk of irAE. Furthermore, patients with cancers with a high TMB may receive a longer course of anti–PD-1 treatment. However, most irAEs reported during anti–PD-1 therapy develop within the first few weeks of treatment. This finding suggests that therapy duration is unlikely to influence the statistical outcome. In conclusion, a high TMB may be a useful biomarker for assessing patients’ risk of irAEs during anti–PD-1 therapy, which has particular relevance for vulnerable patient groups.