Browse the corpus

Introduction Palliative care plays a central role in the management of advanced cancer. Despite advances in targeted chemotherapy and immunotherapy, cancer remains the second leading cause of death in the United States,1 and most patients with metastatic cancer will ultimately die of their disease. For these patients, receipt of palliative care is associated with improved quality of life and prolonged survival.2 The presence of race/ethnicity-based disparities in health care and health outcomes is well documented. Racial/ethnic minorities often receive worse care and have worse outcomes.3 In cancer specifically, there are disparities in screening,4 treatment,5,6 and survival.7 Race/ethnicity-based differences have also been found in receipt of end-of-life care.8,9 Although much research on racial/ethnic differences in care has focused on patient characteristics10 and physician bias,11,12 there is an increasing effort to also investigate the role of the site of care.13,14,15,16,17 Because hospital care for most minority patients is concentrated at a comparatively small number of facilities,18 differences in care at these minority-serving hospitals (MSHs) could explain worse population-level outcomes for minorities overall. If so, policies to improve care at these hospitals represent a potential strategy to address race/ethnicity-based disparities.

oncentrated at a comparatively small number of facilities,18 differences in care at these minority-serving hospitals (MSHs) could explain worse population-level outcomes for minorities overall. If so, policies to improve care at these hospitals represent a potential strategy to address race/ethnicity-based disparities. We assessed racial/ethnic differences in receipt of palliative care for individuals diagnosed with metastatic prostate, lung, colon, and breast cancer. We examined whether receipt of palliative care differed by site of care and whether racial/ethnic disparities in receipt of palliative care are associated with minority patients receiving treatment in a subset of hospitals where palliative care is less often provided.

etastatic prostate, lung, colon, and breast cancer. We examined whether receipt of palliative care differed by site of care and whether racial/ethnic disparities in receipt of palliative care are associated with minority patients receiving treatment in a subset of hospitals where palliative care is less often provided. Methods Data Source The data for this study were abstracted from the Participant Use Files of the National Cancer Database (NCDB), a US cancer registry combining data on patients seen at any 1 of 1500 Commission on Cancer–accredited institutions in the United States.19 The NCDB registry is a joint project of the American Cancer Society and the Commission on Cancer of the American College of Surgeons, comprising more than 29 million unique cases. Trained data abstractors use standardized methods to collect sociodemographic and clinical data, including tumor type, stage, grade, and treatments.20 The NCDB captures 50.8% of all prostate cancers, 82.1% of all lung cancers, 62.5% of all colon cancers, and 66.6% of all breast cancers diagnosed in the United States.21 Data were accessed in October 2017, and the analysis was performed in July 2018. The study was approved by the Brigham and Women’s Hospital Institutional Review Board under a general study protocol for analyses using NCDB data, which included a waiver of informed consent because the information in the Commission on Cancer’s NCDB is deidentified. This study conformed to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for reporting observational research.22

ocol for analyses using NCDB data, which included a waiver of informed consent because the information in the Commission on Cancer’s NCDB is deidentified. This study conformed to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for reporting observational research.22 Study Cohort We chose to focus on men and women 40 years and older with metastatic prostate, non–small cell lung, colon, and breast cancer. These 4 cancer types were chosen because they represented the most common and most lethal cancers for men and women during the study period.23 We chose individuals diagnosed with each cancer from January 1, 2004, to December 31, 2015, using the following International Classification of Diseases for Oncology, Third Edition topography codes: prostate C619, lung C340 to C349, colon C180 to C189 and C260, and breast C500 to C509. We selected men and women with confirmed distant metastases based on the American Joint Committee on Cancer staging system.24 We excluded individuals who had missing follow-up information as well as those diagnosed when younger than 40 years because facility information on these patients is censored by the NCDB for confidentiality purposes.

men with confirmed distant metastases based on the American Joint Committee on Cancer staging system.24 We excluded individuals who had missing follow-up information as well as those diagnosed when younger than 40 years because facility information on these patients is censored by the NCDB for confidentiality purposes. Receipt of Palliative Care The main outcome measure was receipt of any palliative care services. Receipt of palliative care is a variable included with the Participant Use Files of the NCDB. Receipt of palliative care is determined by NCDB data abstractors based on patients’ clinical medical records at participating institutions. Treatments are coded as palliative only if it is explicitly mentioned that the goal of treatment is palliation and not cure (eg, pain control after a routine surgical procedure would not be coded as palliative care). Palliative care encompasses surgical treatment, radiation therapy, and systemic chemotherapy administered to alleviate symptoms but not to cure.25 For the purposes of this study, palliative care was treated as a dichotomous variable.

ain control after a routine surgical procedure would not be coded as palliative care). Palliative care encompasses surgical treatment, radiation therapy, and systemic chemotherapy administered to alleviate symptoms but not to cure.25 For the purposes of this study, palliative care was treated as a dichotomous variable. MSH Status The site of care was the facility reporting the case to the NCDB. This facility is typically the site of diagnosis. For those who received care at multiple institutions, the site of care was the facility where they received definitive cancer care. The MSH status was calculated for each facility based on the proportion of minority patients as follows. First, hospitals were ranked in terms of the proportion of minority patients (black or Hispanic). Second, we identified hospitals in the top decile when ranked from least to greatest proportion of minority patients.26,27 Hospitals in the top decile were considered MSHs. We used the entire population with a diagnosis, not limiting to metastatic cancer only (eg, number of black and Hispanic men with prostate cancer [any stage] at an institution as a portion of the total number of men with prostate cancer [any stage] at that institution and so forth).

the top decile were considered MSHs. We used the entire population with a diagnosis, not limiting to metastatic cancer only (eg, number of black and Hispanic men with prostate cancer [any stage] at an institution as a portion of the total number of men with prostate cancer [any stage] at that institution and so forth). Covariates Baseline sociodemographic covariates included age at diagnosis, sex, race/ethnicity (white non-Hispanic, black non-Hispanic [henceforth referred to as white and black], Hispanic, Asian, other, or unknown), and year of diagnosis. Sociodemographic variables include primary insurance carrier (private, Medicaid or other government payer, Medicare, uninsured, and unknown), educational level (estimated from the percentage of adults within the patient’s zip code without a high school diploma [<7%, 7%-12.9%, 13%-20.9%, or ≥21%]), and zip code–level median household income (<$38 000, $38 000-$47 999, $48 000-$62 999, or ≥$63 000). Clinical covariates included clinical comorbidity (based on the Charlson-Deyo Comorbidity Index, categorized into 0, 1, or ≥2) and cancer type. Because all patients in the cohort had distant metastases (stage IV), we did not adjust by clinical stage. Facility caseload was defined for each cancer as the mean of the total volume of patients with any stage disease treated at the facility for each cancer type in the year of the patient’s diagnosis. This calculation was performed using a previously defined method for NCDB data to account for some facilities leaving and entering the NCDB during the study.28

h cancer as the mean of the total volume of patients with any stage disease treated at the facility for each cancer type in the year of the patient’s diagnosis. This calculation was performed using a previously defined method for NCDB data to account for some facilities leaving and entering the NCDB during the study.28 Statistical Analysis For each of the 4 cancer types, clinical covariates were compared between patients treated at MSHs and non-MSHs. Clustering was performed at the level of the facility to account for correlation of patient characteristics within hospitals. Means (SDs) were calculated for all continuous variables and proportions for all categorical variables. Given less than 5% of missing data in variables, missing values for covariates were ignored because this has a low probability of skewing results.29 Missing outcome variables (unknown whether palliative care was performed) were assumed to be nonignorable, and a maximum likelihood technique for our multilevel model was used to address this.30 We used χ2 tests with a Rao-Scott adjustment to account for clustering to compare the distribution of covariates between patients treated at MSHs and non-MSHs.31,32 We then performed a univariate analysis, again clustering by facility, to compare the proportion of patients receiving palliative care based on race/ethnicity and other baseline characteristics (eg, site of care, cancer type).

to compare the distribution of covariates between patients treated at MSHs and non-MSHs.31,32 We then performed a univariate analysis, again clustering by facility, to compare the proportion of patients receiving palliative care based on race/ethnicity and other baseline characteristics (eg, site of care, cancer type). To assess the association among site of care, patient characteristics, cancer type, and palliative care, a multilevel logistic regression model was fit using the entire study population. This model included fixed-effect terms for patient clinical and demographic covariates (including race/ethnicity and cancer type) and site of care (MSH vs non-MSH). We included an interaction term between cancer type and MSH status to assess whether the effect of MSHs differed in a statistically significant fashion among the 4 cancer types. A facility-level random intercept was included to account for unmeasured hospital-level characteristics that might cut across multiple cancers.33 For example, some hospitals may have palliative care departments, whereas others may not. Finally, based on a significant interaction term (between MSH and cancer type), we performed subgroup analyses by cancer type. For each cancer type, we fit separate multilevel models that assessed the association of clinical and demographic variables as well as site of care on the odds of receiving palliative care. All analyses were performed with Stata statistical software, version 14.0 (StataCorp). Statistical significance was defined as a 2-sided P < .05.

Finally, based on a significant interaction term (between MSH and cancer type), we performed subgroup analyses by cancer type. For each cancer type, we fit separate multilevel models that assessed the association of clinical and demographic variables as well as site of care on the odds of receiving palliative care. All analyses were performed with Stata statistical software, version 14.0 (StataCorp). Statistical significance was defined as a 2-sided P < .05. Results The study cohort consisted of 601 680 individuals (mean [SD] age, 67.4 [11.4] years; 95% CI, 67.2-67.6 years; 314 279 [52.2%] male; 475 039 [78.9%] white) with metastatic cancer diagnosed from January 1, 2004, to December 31, 2015. There were 44 521 men with metastatic prostate cancer, of whom 7096 (15.9%) were treated at MSHs. There were 402 912 men and women with metastatic non–small cell lung cancer, of whom 43 882 (9.4%) were treated at MSHs. There were 89 826 men and women with metastatic colon cancer, of whom 10 570 (11.8%) were treated at MSHs. Finally, of the 65 380 women and men with metastatic breast cancer, 9166 (14.0%) were treated at MSHs. For all 4 cancer types, those treated at MSHs had lower educational levels, had lower income, and were less likely to have public insurance. The baseline characteristics of men and women treated for each of the 4 cancer types at MSHs and non-MSHs are summarized in Table 1. Table 1. Baseline Characteristics of Patients With Metastatic Prostate, Lung, Colon, and Breast Cancer in the National Cancer Database Characteristic No.

For all 4 cancer types, those treated at MSHs had lower educational levels, had lower income, and were less likely to have public insurance. The baseline characteristics of men and women treated for each of the 4 cancer types at MSHs and non-MSHs are summarized in Table 1. Table 1. Baseline Characteristics of Patients With Metastatic Prostate, Lung, Colon, and Breast Cancer in the National Cancer Database Characteristic No. (%) of Patients Prostate Cancer Non–Small Cell Lung Cancer Colon Cancer Breast Cancer MSHs Non-MSHs P Valuea MSHs Non-MSHs P Valuea MSHs Non-MSHs P Valuea MSHs Non-MSHs P Valuea Total patients 7095 (15.9) 37 426 (84.1) NA 43 882 (9.4) 359 030 (90.6) NA 10 570 (11.8) 79 256 (88.2) NA 9166 (14.0) 56 214 (86.0) NA Palliative care Yes 831 (11.7) 5962 (16.0) <.001 9452 (21.5) 92567 (25.8) .02 1036 (9.8) 8930 (11.3) .17 1373 (15.0) 10662 (19.0) .003 No 6263 (88.3) 31405 (84.0) 34420 (78.5) 265841 (74.2) 9532 (90.2) 70260 (88.7) 7792 (85.0) 45354 (81.0) Sex Male NA NA NA 25 320 (57.7) 198 359 (55.2) <.001 5323 (50.4) 39 810 (50.2) .84 150 (1.6) 796 (1.4) .11 Female NA NA 18 562 (42.3) 160 671 (44.8) 5247 (49.6) 39 446 (49.8) 9016 (98.4) 55 418 (98.6) Age group, y 40-50 289 (4.1) 1203 (3.2) <.001 4080 (9.3) 24 620 (6.9) <.001 1432 (13.5) 8682 (11.0) <.001 1908 (20.8) 8648 (15.4) <.001 51-60 1572 (22.2) 5628 (15.0) 11 282 (25.7) 71 508 (19.9) 2802 (26.5) 16 202 (20.4) 2816 (30.7) 14 939 (26.6) 61-70 2227 (31.4) 9943 (26.6) 13 502 (30.8) 110 694 (30.8) 2818 (26.7) 19 763 (24.9) 2229 (24.3) 14 866 (26.4) 71-80 1780 (25.1) 10 590 (28.3) 10 549 (24.0) 104 514 (29.1) 2047 (19.4) 18 839 (23.8) 1361 (14.8) 10 465 (18.6) ≥81 1227 (17.3) 10 062 (26.9) 4469 (10.2) 47 694 (13.3) 1471 (13.9) 15 770 (19.9) 852 (9.4) 7296 (13.0) Race/ethnicity White 2137 (30.1) 29 250 (78.2) <.001 20 542 (46.8) 306 885 (85.5) <.001 3955 (37.4) 63 648 (80.3) <.001 3368 (36.8) 45 254 (80.5) <.001 Black 3463 (48.8) 5244 (14.0) 16 977 (38.7) 31 507 (8.8) 4425 (41.9) 9630 (12.2) 3916 (42.7) 7100 (12.6) Hispanic 1196 (16.9) 1538 (4.1) 4255 (9.7) 7865 (2.2) 1727 (16.3) 2723 (3.4) 1407 (15.4) 1728 (3.1) Asian 164 (2.3) 750 (2.0) 1408 (3.2) 8206 (2.3) 306 (2.9) 2017 (2.5) 280 (3.1) 1231 (2.2) Other 61 (0.9) 281 (0.7) 287 (0.7) 1929 (0.5) 83 (0.8) 549 (0.7) 76 (0.8) 377 (0.7) Unknown 74 (1.0) 363 (1.0) 413 (0.9) 2638 (0.7) 74 (0.7) 689 (0.9) 119 (1.3) 524 (0.9) Year of diagnosis 2004-2006 1717 (24.2) 8350 (22.3) .007 10 716 (24.4) 85 646 (23.9) .30 2081 (19.7) 14 583 (18.4) .19 2027 (22.1) 11 422 (20.3) .02 2007-2009 2048 (28.9) 10 389 (27.8) 12 390 (28.2) 102 357 (28.5) 3101 (29.3) 23 001 (29.0) 2724 (29.7) 16

(1.0) 413 (0.9) 2638 (0.7) 74 (0.7) 689 (0.9) 119 (1.3) 524 (0.9) Year of diagnosis 2004-2006 1717 (24.2) 8350 (22.3) .007 10 716 (24.4) 85 646 (23.9) .30 2081 (19.7) 14 583 (18.4) .19 2027 (22.1) 11 422 (20.3) .02 2007-2009 2048 (28.9) 10 389 (27.8) 12 390 (28.2) 102 357 (28.5) 3101 (29.3) 23 001 (29.0) 2724 (29.7) 16  393 (29.2) 2010-2012 2376 (33.5) 13 172 (35.2) 15 596 (35.5) 126 400 (35.2) 3985 (37.7) 30 695 (38.7) 3262 (35.6) 20 910 (37.2) 2013-2015 954 (13.4) 5515 (14.7) 5180 (11.9) 44 627 (12.4) 1403 (28.7) 10 977 (13.9) 1153 (12.6) 7489 (13.3) Charlson-Deyo Comorbidity Index 0 5448 (76.8) 28 556 (76.3) .81 28 493 (64.9) 224 356 (62.5) .10 7780 (73.6) 56 907 (71.8) .20 7408 (80.8) 44 836 (79.7) .25 1 1129 (15.9) 6075 (16.2) 10 524 (24.0) 92 953 (25.9) 2054 (19.4) 16 161 (20.4) 1320 (14.4) 8280 (14.8) ≥2 518 (7.3) 2795 (7.5) 4865 (11.1) 41 721 (11.6) 736 (7.0) 6188 (7.8) 438 (4.8) 3098 (5.5) Insurance Private 1381 (19.5) 9298 (24.8) <.001 10 721 (24.4) 105 008 (29.3) <.001 2883 (27.3) 26 823 (33.8) <.001 2752 (30.0) 22 142 (39.4) <.001 Medicare 3335 (47.0) 23 539 (62.9) 20 846 (47.5) 205 995 (57.4) 4423 (41.8) 42 910 (54.1) 3061 (33.4) 24 636 (43.8) Medicaid 1011 (14.2) 1920 (5.1) 5928 (13.5) 22 778 (6.3) 1423 (13.5) 4624 (5.8) 1763 (19.2) 5175 (9.2) Other governmental 57 (0.8) 423 (1.1) 507 (1.2) 4774 (1.3) 68 (0.6) 654 (0.8) 59 (0.7) 419 (0.7) None 1011 (14.2) 1518 (4.1) 4328 (9.9) 14 125 (3.9) 1372 (13.0) 3057 (3.9) 1108 (12.1) 2702 (4.8) Unknown 300 (4.2) 728 (2.0) 1552 (3.5) 6350 (1.8) 401 (3.8) 1188 (1.5) 423 (4.6) 1140 (2.0) Family income, $b >63 000 1203 (17.0) 11 159 (29.8) <.001 7252 (16.5) 98 561 (27.4) <.001 1861 (17.6) 24 521 (30.9) <.001 1662 (18.1) 18 222 (32.4) <.001 49 000-63 000 1471 (20.7) 10 069 (26.9) 9047 (20.6) 95 824 (26.7) 2152 (20.4) 20 713 (26.1) 2003 (21.9) 14 918 (26.5) 38 000-48 999 1498 (21.1) 8887 (23.7) 9492 (21.7) 90 470 (25.2) 2381 (22.5) 18 684 (23.6) 1996 (21.8) 12 729 (22.7) <38 000 2830 (39.9) 6679 (17.9) 17160 (39.1) 65337 (18.2) 3976 (37.6) 13 643 (17.2) 3357 (36.6) 9340 (16.6) Unknown 93 (1.3) 632 (1.7) 931 (2.1) 8838 (2.5) 200 (1.9) 1695 (2.2) 148 (1.6) 1005 (1.8) Educational level, % without high school diplomab Unknown 89 (1.3) 601 (1.6) <.001 910 (2.1) 8645 (2.4) <.001 198 (1.9) 1654 (2.1) <.001 144 (1.6) 973 (1.7) <.001 <7 595 (8.4) 9018 (24.1) 4024 (9.2) 73 948 (20.6) 1032 (9.8) 17 938 (22.6) 934 (10.2) 13 331 (23.7) 7-12.9 1211 (17.0) 12 328 (32.9) 8787 (20.0) 119 752 (33.4) 2068 (19.6) 25 854 (32.6) 1756 (19.2) 18 982 (3

Unknown 89 (1.3) 601 (1.6) <.001 910 (2.1) 8645 (2.4) <.001 198 (1.9) 1654 (2.1) <.001 144 (1.6) 973 (1.7) <.001 <7 595 (8.4) 9018 (24.1) 4024 (9.2) 73 948 (20.6) 1032 (9.8) 17 938 (22.6) 934 (10.2) 13 331 (23.7) 7-12.9 1211 (17.0) 12 328 (32.9) 8787 (20.0) 119 752 (33.4) 2068 (19.6) 25 854 (32.6) 1756 (19.2) 18 982 (3 3.8) 13-20.9 2108 (29.7) 9515 (25.5) 13 629 (31.0) 98 698 (27.5) 3241 (30.6) 21 094 (26.6) 2865 (31.3) 14 218 (25.3) >30 3092 (43.6) 5964 (15.9) 16 532 (37.7) 57 987 (16.2) 4031 (38.1) 12 716 (16.1) 3467 (37.8) 8710 (15.5) Abbreviations: MSH, minority-serving hospital; NA, not applicable. a Hospital-level clustering with Taylor series linearization; the Pearson χ2 test was used to test significance. b Both estimated using patients’ county of residence.

3.8) 13-20.9 2108 (29.7) 9515 (25.5) 13 629 (31.0) 98 698 (27.5) 3241 (30.6) 21 094 (26.6) 2865 (31.3) 14 218 (25.3) >30 3092 (43.6) 5964 (15.9) 16 532 (37.7) 57 987 (16.2) 4031 (38.1) 12 716 (16.1) 3467 (37.8) 8710 (15.5) Abbreviations: MSH, minority-serving hospital; NA, not applicable. a Hospital-level clustering with Taylor series linearization; the Pearson χ2 test was used to test significance. b Both estimated using patients’ county of residence. In the combined cohort, 130 813 patients (21.7%) received any palliative care and 470 867 (78.1%) did not. The number of patients receiving palliative care differed based on cancer type. The number of patients receiving palliative care was 6793 (15.3%) of those with metastatic prostate cancer, 102 019 (25.4%) of those with metastatic lung cancer, 9966 (11.1%) of those with metastatic colon cancer, and 120 035 (18.5%) of those with metastatic breast cancer (P < .001). In terms of race/ethnicity, whereas 106 603 white patients (22.5%) received palliative care, only 16 435 black patients (20.0%) and 3551 Hispanic patients (15.9%) received palliative care (P < .001 for all). Patients treated at an MSH were less likely than patients treated at a non-MSH to receive palliative care regardless of race/ethnicity (12 692 [18.0%] vs 118 121 [22.3%], P = .002). Receipt of palliative care based on other baseline characteristics is summarized in Table 2.

received palliative care (P < .001 for all). Patients treated at an MSH were less likely than patients treated at a non-MSH to receive palliative care regardless of race/ethnicity (12 692 [18.0%] vs 118 121 [22.3%], P = .002). Receipt of palliative care based on other baseline characteristics is summarized in Table 2. Table 2. Unadjusted Proportions of Patients With Metastatic Cancer Receiving Palliative Care in Overall Cohort by Baseline Characteristics Characteristic No. (%) of Patients P Valuea No Palliative Care Any Palliative Care Total patients 470 867 (78.1) 130 813 (21.7) NA Hospital type MSH 58 007 (82.1) 12 692 (18.0) .002 Non-MSH 412 860 (77.8) 118 121 (22.3) Cancer type Prostate 37 668 (84.7) 6793 (15.3) <.001 Non–small cell lung 300 261 (74.6) 102 019 (25.4) Colon 79 792 (88.9) 9966 (11.1) Breast 53 146 (81.5) 12 035 (18.5) Sex Male 244 246 (77.8) 69 571 (22.2) <.001 Female 226 621 (78.7) 61 242 (21.3) Age group, y ≤50 39 413 (77.7) 11 305 (22.3) <.001 51-60 97 147 (76.8) 29 331 (23.2) 61-70 136 217 (77.5) 39 545 (22.5) 71-80 125 971 (78.8) 33 977 (21.2) ≥81 72 119 (81.2) 16 655 (18.8) Race/ethnicity White 367 695 (77.5) 106 603 (22.5) <.001 Black 65 716 (80.0) 16 435 (20.0) Hispanic 18 814 (84.1) 3551 (15.9) Asian 11 782 (82.1) 2572 (17.9) Other 2879 (79.5) 741 (20.5) Unknown 3981 (81.4) 911 (18.6) Year of diagnosis 2004-2006 108 557 (79.7) 27 602 (20.3) <.001 2007-2009 135 234 (78.6) 36 847 (21.4) 2010-2012 168 279 (77.8) 47 907 (22.2) 2013-2015 58 797 (76.1) 18 457 (23.9) Charlson-Deyo Comorbidity Index 0 318 898 (79.1) 84 038 (20.9) <.001 1 105 984 (76.6) 32 434 (23.4) ≥2 45 985 (76.2) 14 341 (23.8) Insurance Private 141 937 (78.5) 38 986 (21.6) <.001 Medicare 257 850 (78.5) 70 705 (21.5) Medicaid 33 980 (76.2) 10 624 (23.8) Other governmental 5143 (73.9) 1816 (26.1) None 22 424 (76.8) 6784 (23.2) Unknown 9533 (83.4) 1898 (23.4) Family income, $a >63 000 130 096 (79.2) 34 101 (20.8) .02 49 000-63 000 122 089 (78.3) 33 879 (21.7) 38 000-48 999 113 059 (77.5) 32 849 (22.5) <38 000 95 269 (78.0) 26 824 (22.0) Unknown 10 354 (76.6) 3160 (16.6) Educational level, % without high school diplomaa <7 94 022 (77.9) 26 614 (22.1) <.001 7-12.9 147 688 (77.5) 42 865 (22.5) 13-20.9 129 068 (78.2) 36 018 (21.8) >30 89 998 (80.2) 22 220 (19.8) Unknown 10 091 (76.5) 3096 (23.5) Abbreviations: MSH, minority-serving hospital; NA, not applicable.

n 10 354 (76.6) 3160 (16.6) Educational level, % without high school diplomaa <7 94 022 (77.9) 26 614 (22.1) <.001 7-12.9 147 688 (77.5) 42 865 (22.5) 13-20.9 129 068 (78.2) 36 018 (21.8) >30 89 998 (80.2) 22 220 (19.8) Unknown 10 091 (76.5) 3096 (23.5) Abbreviations: MSH, minority-serving hospital; NA, not applicable. a Estimated from patients’ county of residence. In our adjusted multilevel logistic regression model adjusting for age, race/ethnicity, comorbidity, cancer type, and patient demographics and including an interaction term between MSH status and cancer type, patients who received care at an MSH had two-thirds the odds of receiving palliative care compared with those who received care at a non-MSH (odds ratio [OR], 0.67; 95% CI, 0.53-0.84). Later study year was also associated with increased odds of receiving palliative care (first vs last period: OR, 1.30; 95% CI, 1.27-1.33). Patients with Medicaid and uninsured patients were more likely to receive palliative care compared with those with private insurance (Medicaid vs private: OR, 1.16 [95% CI, 1.13-1.19]; uninsured vs private: OR, 1.17 [95% CI, 1.13-1.21]).

of receiving palliative care (first vs last period: OR, 1.30; 95% CI, 1.27-1.33). Patients with Medicaid and uninsured patients were more likely to receive palliative care compared with those with private insurance (Medicaid vs private: OR, 1.16 [95% CI, 1.13-1.19]; uninsured vs private: OR, 1.17 [95% CI, 1.13-1.21]). After adjusting for MSH status and other covariates, the difference in receipt of palliative care between white and black individuals was no longer statistically significant (OR, 1.02; 95% CI, 0.99-1.04). Hispanic patients had higher odds of palliative care compared with white patients (OR, 1.06; 95% CI, 1.01-1.10). Compared with non-Hispanic white patients, a lower proportion of Asian patients received palliative care (2572 [17.9%] vs 106 603 [22.5%], P < .001). This finding was also true on adjusted analyses (OR, 0.93; 95% CI, 0.88-0.98). Table 3 provides a summary of the adjusted analyses.

atients (OR, 1.06; 95% CI, 1.01-1.10). Compared with non-Hispanic white patients, a lower proportion of Asian patients received palliative care (2572 [17.9%] vs 106 603 [22.5%], P < .001). This finding was also true on adjusted analyses (OR, 0.93; 95% CI, 0.88-0.98). Table 3 provides a summary of the adjusted analyses. Table 3. Factors Associated With Palliative Care in an Adjusted Multilevel Model Including a Hospital-Level Random Intercept Indicator Odds Ratio (95% CI) P Valuea Hospital type Non-MSH 1 [Reference] NA MSH 0.67 (0.53-0.84) .001 Metastatic cancer type Prostate 1 [Reference] NA Non–small cell lung 1.69 (1.51-1.88) <.001 Colon 0.56 (0.50-0.63) <.001 Breast 1.10 (0.98-1.23) .08 Sex Male 1 [Reference] NA Female 0.95 (0.94-0.97) <.001 Age group, y ≤50 1 [Reference] NA 51-60 1.00 (0.97-1.03) .89 61-70 0.93 (0.90-0.95) <.001 71-80 0.86 (0.83-0.88) <.001 ≥81 0.79 (0.77-0.82) <.001 Race/ethnicity White 1 [Reference] NA Black non-Hispanic 1.02 (0.99-1.04) .19 Hispanic 1.06 (1.01-1.10) .01 Asian 0.93 (0.88-0.98) .008 Other 0.92 (0.85-1.01) .08 Unknown 0.78 (0.72-0.84) <.001 Year of diagnosis 2004-2006 1 [Reference] NA 2007-2009 1.10 (1.08-1.12) <.001 2010-2012 1.16 (1.14-1.18) <.001 2013-2015 1.30 (1.27-1.33) <.001 Charlson-Deyo Comorbidity Index 0 1 [Reference] NA 1 1.01 (0.99-1.03) .18 ≥2 1.00 (0.98-1.03) .70 Insurance Private 1 [Reference] NA Medicare 1.01 (0.99-1.03) .14 Medicaid 1.16 (1.13-1.19) <.001 Other governmental 1.20 (1.13-1.27) <.001 None 1.17 (1.13-1.21) <.001 Unknown 0.87 (0.82-0.92) <.001 Family income, $a >63 000 1 [Reference] NA 49 000-63 000 0.99 (0.97-1.01) .39 38 000-48 999 0.97 (0.95-1.00) .06 <38 000 0.99 (0.96-1.02) .46 Unknown 0.87 (0.65-1.16) .34 Educational level, % without high school diplomaa >30 1 [Reference] NA 13-20.9 1.00 (0.98-1.02) .94 7-12.9 1.00 (0.97-1.03) .99 <7 1.00 (0.97-1.03) .98 Unknown 1.24 (0.92-1.66) .16 Abbreviations: MSH, minority-serving hospital; NA, not applicable.

0.95-1.00) .06 <38 000 0.99 (0.96-1.02) .46 Unknown 0.87 (0.65-1.16) .34 Educational level, % without high school diplomaa >30 1 [Reference] NA 13-20.9 1.00 (0.98-1.02) .94 7-12.9 1.00 (0.97-1.03) .99 <7 1.00 (0.97-1.03) .98 Unknown 1.24 (0.92-1.66) .16 Abbreviations: MSH, minority-serving hospital; NA, not applicable. a Estimated from county of residence. The interaction term between cancer type and MSH status was associated with receipt of palliative care. Thus, we performed a subgroup analysis stratifying by cancer type. In the metastatic prostate cancer subgroup, the odds of receiving palliative care at MSHs were approximately 33% lower (OR, 0.67; 95% CI, 0.55-0.82); in the lung cancer subgroup, the odds of palliative care were 27% lower at MSHs (OR, 0.73; 95% CI, 0.57-0.93); in colon cancer, the odds of palliative care at MSHs were not significantly lower (OR, 0.86; 95% CI, 0.67-1.09); and in breast cancer, the odds of palliative care were 27% lower (OR, 0.73; 95% CI, 0.59-0.89). As in the combined cohort, adjustment for MSH status in all cancers attenuated the association between race/ethnicity and odds of receiving palliative care toward the null.

significantly lower (OR, 0.86; 95% CI, 0.67-1.09); and in breast cancer, the odds of palliative care were 27% lower (OR, 0.73; 95% CI, 0.59-0.89). As in the combined cohort, adjustment for MSH status in all cancers attenuated the association between race/ethnicity and odds of receiving palliative care toward the null. Discussion In this retrospective, registry-based study of adults diagnosed with metastatic prostate, lung, breast, and colon cancer, there were significantly lower odds of receiving palliative care among patients treated at MSHs compared with non-MSHs. Although it has been previously reported that minority patients are less likely to receive palliative care services at the end of life,8,9 the present findings suggest that site of care may be a significant factor associated with race/ethnicity-based differences in palliative care. The policy implications of this finding are significant. Given that care for minority patients is concentrated at a comparatively small number of hospitals in the United States, it is likely that one important strategy to address racial/ethnic disparities in palliative care is to focus on improving access to palliative care at the small number of hospitals that care for most minority patients. If initiatives to target palliative care use at MSHs are successful, national disparities in palliative care may be reduced.

mportant strategy to address racial/ethnic disparities in palliative care is to focus on improving access to palliative care at the small number of hospitals that care for most minority patients. If initiatives to target palliative care use at MSHs are successful, national disparities in palliative care may be reduced. Overall, this fits with an increasing understanding that the site of care is a determinant of health outcomes for minority patients. Although there are data that physicians may systematically treat black and white patients differently,11,12 that minority patients tend to receive care at different facilities is also important. More than being a function of individual behavior, there is increasing recognition that disparities in outcomes depend on different treatment of white and minority patients within the same hospital and systemic differences in where minority patients receive care.14,15

ive care at different facilities is also important. More than being a function of individual behavior, there is increasing recognition that disparities in outcomes depend on different treatment of white and minority patients within the same hospital and systemic differences in where minority patients receive care.14,15 A previous study18 found that MSHs have higher readmission rates and worse performance in many clinical scenarios, for example, when treating acute myocardial infarctions and pneumonia. A study34 of emergency general surgery at MSHs found that hospital-level factors accounted for approximately 40% of increased odds for readmission, and inpatient mortality was significantly greater. Hospital leadership can also play an important role. A survey of chairmen at black-serving hospitals found that, when compared with non–black-serving hospital boards, these chairpersons report less expertise with quality-of-care issues and are less likely to give high priority to quality of care.35 An analysis36 of racial disparity in surgical mortality found that although gaps between black and white patients have narrowed overall, improvements were less likely among hospitals that served the highest proportion of minority patients. Overall, our findings suggest that similar systemic differences between MSHs and non-MSHs may be associated with the differences in receipt of palliative care among underserved minority patients.

ave narrowed overall, improvements were less likely among hospitals that served the highest proportion of minority patients. Overall, our findings suggest that similar systemic differences between MSHs and non-MSHs may be associated with the differences in receipt of palliative care among underserved minority patients. Although Asian patients composed a small proportion of our population, they were less likely to receive palliative care after adjusting for MSH status. Asian individuals are a heterogeneous group and may in some cases have better access to health care compared with Hispanic patients and black patients; Asian individuals have population-level health outcomes that exceed most of the other racial/ethnic groups.37 Thus, as has been done in a prior study,27 we did not include Asian patients in our definition of MSHs. The lower odds of palliative care among Asian patients could reflect cultural differences, differences in familial characteristics among this population, and other economic or health systems factors. The finding that palliative care is more common in Medicaid patients and uninsured patients was similarly surprising given that these patients seem to receive worse care based on many other health metrics.38 Perhaps these patients were presenting at a more advanced stage of disease, when palliative care is the only good option. Alternatively, perhaps the absence of a strong fee-for-service incentive toward doing more reduced the barrier for palliative care for the Medicaid and uninsured patients.

ny other health metrics.38 Perhaps these patients were presenting at a more advanced stage of disease, when palliative care is the only good option. Alternatively, perhaps the absence of a strong fee-for-service incentive toward doing more reduced the barrier for palliative care for the Medicaid and uninsured patients. Strengths and Limitations Strengths of our study include our use of a large, accurate national registry, which captures most US patients diagnosed with 4 highly prevalent types of cancers. Another strength is that our study included patients from all payers. Our work therefore improves on earlier definitions of minority serving, which often used Medicare claims and therefore involved only the proportion of Medicare beneficiaries who were racial/ethnic minorities not the proportion of patients with a given condition.26

th is that our study included patients from all payers. Our work therefore improves on earlier definitions of minority serving, which often used Medicare claims and therefore involved only the proportion of Medicare beneficiaries who were racial/ethnic minorities not the proportion of patients with a given condition.26 Despite these strengths, this work has limitations. Data on palliative care services are of uncertain accuracy. The data on receipt of palliative care in the NCDB were collected from medical records by trained data abstractors at each institution. Intent must be inferred from clinical records. Although we believe that record review may be more accurate than insurance claims, which have been reported to often have only moderate accuracy for ascertaining the intensity of end-of-life care,39 the accuracy may be lower than some prospective trials that have specifically assigned patients to palliative care interventions.2 Additional studies that specifically address interrater variability and validate this variable against other end points (eg, inappropriately aggressive end-of-life care, such as chemotherapy in the last 14 days of life, death in hospital, or death in the intensive care unit) would be useful. Another limitation is the possibility of unmeasured patient confounders, which are always a factor in retrospective research. Our use of a multilevel model with a hospital-level random intercept should account for unmeasured hospital characteristics at the level of the hospital (eg, some hospitals may have palliative care departments, whereas others may not).

unmeasured patient confounders, which are always a factor in retrospective research. Our use of a multilevel model with a hospital-level random intercept should account for unmeasured hospital characteristics at the level of the hospital (eg, some hospitals may have palliative care departments, whereas others may not). Although the NCDB captures most patients with each of these 4 cancer types in the United States, data are not population based. Thus, certain patients who did not receive care at Commission on Cancer–accredited US hospitals may have been underrepresented. For example, if the database underrepresents poor-performing, rural non-MSHs, the disparities among MSHs could be inflated. Conclusions These findings suggest that there are significant racial/ethnic disparities in receipt of palliative care for metastatic cancer within a large cohort of US patients with cancer. After controlling for race/ethnicity and MSH status, we found that treatment at MSHs was associated with significantly lower odds of receiving palliative care, but black and Hispanic race/ethnicity was not. Strategies that focus on improving palliative care use at MSHs may be an effective strategy to increase the receipt of palliative care for this population.

ach is likely to compound this trend. In addition to acquiescing to patient demand, another mechanism that may be associated with the rapid acquisition and implementation of robotic surgery could be the ability of nonprofit hospitals to access tax-exempt financial instruments toward using debt for operational growth.30 The adoption of robotic surgery has coincided with the centralization of surgical procedures, most notably for RP in the United States. Investigators have shown that institutions acquiring surgical robotics have seen a dramatic increase in their surgical volume. Specifically, Riikonen et al31 demonstrated that the principal outcome of national-level adoption of robotic surgery in Finland led to the immediate and unpremeditated centralization for prostate cancer surgery. It has also been reported that high-volume surgeons in the United States at teaching and large hospitals have swiftly adopted the robotic approach.32 Evidence also shows that regionalization has been observed at a higher rate for the robotic approach compared with the open approach.33 In the United Kingdom, Aggarwal et al27 showed that competitive factors and centralization of services have led to greater investment in building cancer surgery units that use robotic surgery as a primary treatment modality. Several studies have shown the conditions in which the excess costs of robotic surgery are mitigated: when hospital volume is high and operative time is low, robotic surgery can cost less.34,35,36

Introduction As of 2017, US national health expenditures stood at $3.5 trillion.1 Despite recent reforms aimed at containing increasing US health care expenditures, overall US health care spending remains on an unsustainable course.2,3,4 Consequently, renewed focus has been placed on the value of medical services rendered. Although value-driven initiatives in the United States have traditionally emphasized eliminating excessive administrative costs and/or physician reimbursements, the role of innovative—and costly—technologies such as robotic surgery in increased health care spending has not been well studied, to our knowledge.5 In the past 2 decades, an exponential increase in the adoption of minimally invasive surgery for the management of common malignant neoplasms has occurred.6 This adoption is due in part to early evidence of lower morbidity, hospital length of stay (LOS), and blood loss, as well as reduced postoperative analgesia requirements, associated with minimally invasive surgery.7,8,9 As a result, minimally invasive procedures have secured a more integral role in oncologic surgery. However, this change has occurred in the context of research demonstrating higher associated surgical costs and equivocal evidence of improved clinical outcomes.10,11,12

s, associated with minimally invasive surgery.7,8,9 As a result, minimally invasive procedures have secured a more integral role in oncologic surgery. However, this change has occurred in the context of research demonstrating higher associated surgical costs and equivocal evidence of improved clinical outcomes.10,11,12 Most studies that have examined perioperative outcomes and costs associated with robotic surgery have been limited by a dearth of granular cost data, thereby precluding a systematic assessment of the true financial cost—and, by extension, value—associated with the rapid adoption of robotic surgery. The literature to date has focused on total health care spending associated with robotic surgery (usually estimated by using total charges), generally showing robotic surgery to be associated with higher mean direct hospital costs and lower health plan spending, and there has not been a comprehensive scientific inquiry into out-of-pocket (OOP) costs for patients, to our knowledge.6,13 To truly understand whether robotic surgery is beneficial compared with open surgery, it is important to capture all costs borne by the patient, not just those covered by payers. Furthermore, understanding the specific segment of patients affected by the costs of a particular procedure may help better elucidate the factors associated with growing health inequity.14 To examine this question, we used a large, nationally representative sample of patients to assess OOP costs and total payments for 5 types of common oncologic procedures that can be performed using an open or robotic approach.

ar procedure may help better elucidate the factors associated with growing health inequity.14 To examine this question, we used a large, nationally representative sample of patients to assess OOP costs and total payments for 5 types of common oncologic procedures that can be performed using an open or robotic approach. Methods Data Source We queried the IBM Watson Health (formerly Truven Health Analytics) MarketScan Commercial Claims and Encounters database. As a Health Insurance Portability and Accountability Act–compliant database, it assembles information on insurance enrollment along with medical and drug claims for millions of individuals who receive health insurance coverage from their employers in the form of various health plans. The database captures unique information on inpatient, outpatient, and emergency department encounters, including OOP charges and claims on prescription drugs, using unique patient identifiers. We analyzed data collected from January 1, 2012, to December 31, 2017, which contained deidentified claims for 1.9 million enrollees, representing 260 employers, spread over 40 health plans with 350 unique carriers in the United States. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.15 As the data were deidentified, the study was deemed exempt by the Brigham and Women’s Hospital (Partners Healthcare) Institutional Review Board.

ans with 350 unique carriers in the United States. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.15 As the data were deidentified, the study was deemed exempt by the Brigham and Women’s Hospital (Partners Healthcare) Institutional Review Board. Study Population With a previously validated approach,13 we used International Classification of Diseases, Ninth Revision (ICD-9), International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10), and Current Procedural Terminology (CPT) codes to identify a sample of adults aged 18 to 64 years enrolled in an employer-sponsored health plan who underwent either an open or robotic radical prostatectomy (RP), hysterectomy (HYS), partial colectomy (PC), radical nephrectomy (RN), or partial nephrectomy (PN) for a solid-organ malignant neoplasm. For PC, we included colostomy and anterior resection. Adults older than 64 years were excluded because they are eligible for Medicare. Inclusion criteria were based on inpatient insurance claims for one of the aforementioned procedures between January 1, 2012, and December 31, 2017. To calculate the index surgical date, the earliest available date was considered, especially when multiple claims were available. Exclusion criteria included lack of 12 months of continuous insurance coverage in the same health plan before and after the index date, incurring a total expenditure of less than $1 (implying erroneous data collection), or incomplete demographic data (Figure 1).

nsidered, especially when multiple claims were available. Exclusion criteria included lack of 12 months of continuous insurance coverage in the same health plan before and after the index date, incurring a total expenditure of less than $1 (implying erroneous data collection), or incomplete demographic data (Figure 1). Figure 1. Flowchart of Patient and Procedure Selection in the Study Outcomes Our primary outcome of interest was OOP costs associated with robotic and open surgery; our secondary outcome of interest was total payments for patients who underwent 1 of the study’s 5 procedures. In accordance with previously published literature, we framed 3 time periods around the index surgical date: the baseline (−380 to −15 days), perioperative (−14 to 28 days), and postoperative (29 to 352 days) periods.13,16 Total payments associated with each surgical procedure were calculated by adding gross payments of all inpatient, outpatient, and pharmacy claims in the perioperative and postoperative periods (−14 to 352 days). Out-of-pocket costs were calculated by adding the coinsurance, copayment, and deductible of all inpatient, outpatient, and pharmacy claims during the perioperative and postoperative periods. We adjusted total payments to 2018 US dollars, relying on the general Consumer Price Index.

toperative periods (−14 to 352 days). Out-of-pocket costs were calculated by adding the coinsurance, copayment, and deductible of all inpatient, outpatient, and pharmacy claims during the perioperative and postoperative periods. We adjusted total payments to 2018 US dollars, relying on the general Consumer Price Index. Robotic Surgery We used ICD-9, ICD-10, and CPT codes to identify the different types of open and laparoscopic surgery. We considered patients to be receiving robotic surgery if they had an open or laparoscopic surgery code, plus a robotic modifier (available in the eAppendix in the Supplement). Nonrobotic laparoscopic procedures were excluded from further analysis.6,13 Statistical Analysis Statistical analysis was performed from December 18, 2018, to June 5, 2019. Patient-level covariates included age, sex, Elixhauser comorbidity score, US Census region, urban vs rural residence, year of surgery, and health plan type (less restrictive vs more restrictive).17 Less restrictive plans included a comprehensive or preferred provider organization. More restrictive plans included an exclusive provider organization, a health maintenance organization, a noncapitated point-of-service, consumer-driven health plan, and a high-deductible health plan. The Elixhauser comorbidity score excluded the primary diagnosis for each cancer. To differentiate baseline characteristics between patients undergoing robotic surgery and patients undergoing open surgery, t tests were used for continuous variables and χ2 tests were used for categorical variables.

eductible health plan. The Elixhauser comorbidity score excluded the primary diagnosis for each cancer. To differentiate baseline characteristics between patients undergoing robotic surgery and patients undergoing open surgery, t tests were used for continuous variables and χ2 tests were used for categorical variables. We first performed an inverse probability of treatment weighting (IPTW) propensity score analysis to address inherent differences in the covariates between the open and robotic surgery cohorts. We conducted IPTW propensity score analyses to balance the open vs robotic approach based on the patient-level covariates separately for each procedure cohort. Next, multivariable linear regression—weighted by the inverse of the probability of receiving robotic surgery based on baseline covariates and adjusting for baseline OOP or total payments—was used to estimate the independent association of surgical approach with OOP costs or total payments within the entire perioperative and postoperative period for each of the 5 procedure cohorts. The gamma distribution, which provides an accurate estimation of population means of health care costs,18 was used to report OOP costs and total payments.

ndent association of surgical approach with OOP costs or total payments within the entire perioperative and postoperative period for each of the 5 procedure cohorts. The gamma distribution, which provides an accurate estimation of population means of health care costs,18 was used to report OOP costs and total payments. In addition, we calculated IPTW-adjusted differences in hospital LOS for patients undergoing open or robotic procedures. Given that many of the robotic and minimally invasive procedures are increasingly performed on a short-stay basis, we wanted to understand the association of this status with the amount paid by the insurance company and to what extent longer hospital stays were associated with higher payments for open procedures.

ures. Given that many of the robotic and minimally invasive procedures are increasingly performed on a short-stay basis, we wanted to understand the association of this status with the amount paid by the insurance company and to what extent longer hospital stays were associated with higher payments for open procedures. We also conducted an outlier analysis for OOP costs and total payments between patients undergoing open surgery and patients undergoing robotic surgery separately for each procedure cohort, to examine the range of these expenses. This analysis was performed to address any observable variations in total payments and to examine whether these variations were artificially shifting the differences in mean payments between open and robotic procedures. In addition, because our main analysis examined the total payments associated with each surgical procedure in the entire perioperative and postoperative period (−14 to 352 days), to understand the broader overview of the association of the procedure with patient expenditures, we also wanted to understand the direct outcome in the shorter timeframe as reference points. Therefore, we conducted additional cost analyses for OOP and total payments at perioperative (−14 to 28 days) and 3-month (−14 to 90 days) periods. All analyses were conducted using SAS, version 9.4 (SAS Institute Inc). A 2-tailed P < .05 was considered statistically significant.

We also conducted an outlier analysis for OOP costs and total payments between patients undergoing open surgery and patients undergoing robotic surgery separately for each procedure cohort, to examine the range of these expenses. This analysis was performed to address any observable variations in total payments and to examine whether these variations were artificially shifting the differences in mean payments between open and robotic procedures. In addition, because our main analysis examined the total payments associated with each surgical procedure in the entire perioperative and postoperative period (−14 to 352 days), to understand the broader overview of the association of the procedure with patient expenditures, we also wanted to understand the direct outcome in the shorter timeframe as reference points. Therefore, we conducted additional cost analyses for OOP and total payments at perioperative (−14 to 28 days) and 3-month (−14 to 90 days) periods. All analyses were conducted using SAS, version 9.4 (SAS Institute Inc). A 2-tailed P < .05 was considered statistically significant. Results Baseline Demographic Characteristics A total of 15 893 patients (11 102 men; mean [SD] age, 55.4 [6.6] years) underwent 1 of 5 surgical procedures: 8260 underwent robotic procedures, and 7633 underwent open procedures. From 2012 to 2017, there were 7521 patients who met inclusion criteria and underwent either open or robotic RP, 1208 patients who underwent either open or robotic RN, 1996 patients who underwent either open or robotic PN, 1978 patients who underwent either open or robotic HYS, and 3190 patients who underwent either open or robotic PC (Table 1). Robotic procedures represented 78.0% of the RP cohort (n = 5869), 28.7% of the RN cohort (n = 347), 59.7% of the PN cohort (n = 1192), 22.8% of the HYS cohort (n = 450), and 12.6% of the PC cohort (n = 402). In the RP cohort, both open (48.4%) and robotic (42.7%) procedures were observed in higher proportion in the South. Patients undergoing robotic HYS were older than those undergoing open HYS (mean [SD] age, 55.7 [6.7] vs 54.6 [7.2] years; P = .004). Patients undergoing open RN had more comorbidities than those undergoing robotic RN (≥2 comorbidities, 658 of 861 [76.4%] vs 244 of 347 [70.3%]; P = .01). Differences in baseline characteristics between patients undergoing open surgery and patients undergoing robotic surgery are described in Table 1.

years; P = .004). Patients undergoing open RN had more comorbidities than those undergoing robotic RN (≥2 comorbidities, 658 of 861 [76.4%] vs 244 of 347 [70.3%]; P = .01). Differences in baseline characteristics between patients undergoing open surgery and patients undergoing robotic surgery are described in Table 1. Table 1. Baseline Demographic Characteristics of Patients Undergoing Open or Robotic Surgery in the MarketScan Database, 2012-2017 Characteristic Radical Prostatectomy Hysterectomy Partial Colectomy Radical Nephrectomy Partial Nephrectomy Open (n = 1652) Robotic (n = 5869) P Value Open (n = 1528) Robotic (n = 450) P Value Open (n = 2788) Robotic (n = 402) P Value Open (n = 861) Robotic (n = 347) P Value Open (n = 804) Robotic (n = 1192) P Value Age, mean (SD), y 57.2 (4.8) 57.0 (4.7) .14 54.6 (7.2) 55.7 (6.7) .004 53.5 (7.6) 53.4 (7.0) .73 53.8 (7.5) 54.3 (7.9) .29 53.1 (8.1) 53.3 (8.1) .63 Age group, No. (%) 18-34 0 2 (0.03) .16 21 (1.4) 9 (2.0) <.001 57 (2.0) 6 (1.5) .007 14 (1.6) 11 (3.2) .24 23 (2.9) 35 (2.9) .34 35-44 28 (1.7) 81 (1.4) 142 (9.3) 21 (4.7) 310 (11.1) 37 (9.2) 89 (10.3) 28 (8.1) 86 (10.7) 140 (11.7) 45-54 384 (23.2) 1504 (25.6) 470 (30.8) 108 (24.0) 963 (34.5) 174 (43.3) 282 (32.8) 113 (32.6) 293 (36.4) 388 (32.6) 55-64 1240 (75.1) 4282 (73.0) 895 (58.6) 312 (69.3) 1458 (52.3) 185 (46.0) 476 (55.3) 195 (56.2) 402 (50.0) 629 (52.8) Sex, No. (%) Male 1652 (100) 5869 (100) NA NA NA NA 1375 (49.3) 210 (52.2) .27 532 (61.8) 229 (66.0) .17 516 (64.2) 726 (60.9) .13 Female NA NA 1528 (100) 450 (100) 1413 (50.7) 192 (47.8) 329 (38.2) 118 (34.0) 288 (35.8) 466 (39.1) Comorbidities, No. (%) 0 313 (19.0) 1112 (19.0) .94 151 (9.9) 44 (9.8) .17 237 (8.5) 39 (9.7) .26 64 (7.4) 43 (12.4) .01 74 (9.2) 112 (9.4) .91 1 507 (30.7) 1777 (30.3) 310 (20.3) 71 (15.8) 538 (19.3) 86 (21.4) 139 (16.1) 60 (17.3) 155 (19.3) 221 (18.5) ≥2 832 (50.4) 2980 (50.8) 1067 (69.8) 335 (74.4) 2013 (72.2) 277 (68.9) 658 (76.4) 244 (70.3) 575 (71.5) 859 (72.1) Geographical region, No.

237 (8.5) 39 (9.7) .26 64 (7.4) 43 (12.4) .01 74 (9.2) 112 (9.4) .91 1 507 (30.7) 1777 (30.3) 310 (20.3) 71 (15.8) 538 (19.3) 86 (21.4) 139 (16.1) 60 (17.3) 155 (19.3) 221 (18.5) ≥2 832 (50.4) 2980 (50.8) 1067 (69.8) 335 (74.4) 2013 (72.2) 277 (68.9) 658 (76.4) 244 (70.3) 575 (71.5) 859 (72.1) Geographical region, No. (%) Northeast 281 (17.0) 998 (17.0) <.001 320 (20.9) 95 (21.1) <.001 497 (17.8) 107 (26.6) <.001 111 (12.9) 51 (14.7) <.001 210 (26.1) 274 (23.0) <.001 North central 313 (19.0) 1475 (25.1) 371 (24.3) 108 (24.0) 626 (22.5) 80 (19.9) 172 (20.0) 101 (29.1) 140 (17.4) 300 (25.2) South 800 (48.4) 2506 (42.7) 602 (39.4) 143 (31.8) 1315 (47.2) 169 (42.0) 466 (54.1) 137 (39.5) 321 (39.9) 481 (40.4) West 229 (13.9) 810 (13.8) 218 (14.3) 100 (22.2) 314 (11.3) 45 (11.2) 101 (11.7) 56 (16.1) 116 (14.4) 129 (10.8) Unknown 29 (1.8) 80 (1.4) 17 (1.1) 4 (0.9) 36 (1.3) 1 (0.3) 11 (1.3) 2 (0.6) 17 (2.1) 8 (0.7) Residence, No. (%) Rural 343 (20.8) 943 (16.1) <.001 235 (15.4) 53 (11.8) .05 579 (20.8) 48 (11.9) <.001 176 (20.4) 56 (16.1) .08 127 (15.8) 175 (14.7) .49 Urban 1309 (79.2) 4926 (83.9) 1293 (84.6) 397 (88.2) 2209 (79.2) 354 (88.1) 685 (79.6) 291 (83.9) 677 (84.2) 1017 (85.3) Health plan type, No. (%) Less restrictivea 1124 (68.0) 4000 (68.2) .92 1019 (66.7) 296 (65.8) .71 1877 (67.3) 270 (67.2) .94 565 (65.6) 239 (68.9) .27 542 (67.4) 825 (69.2) .39 More restrictiveb 528 (32.0) 1869 (32.0) 509 (33.3) 154 (34.2) 911 (32.7) 132 (32.8) 296 (34.4) 108 (31.1) 262 (32.6) 367 (30.8) Colostomy, No. (%) Yes NA NA NA NA NA NA 356 (12.8) 22 (5.5) <.001 NA NA NA NA NA NA No NA NA NA NA 2432 (87.2) 380 (94.5) NA NA NA NA Low anterior resection, No. (%) Yes NA NA NA NA NA NA 593 (21.3) 152 (37.8) <.001 NA NA NA NA NA NA No NA NA NA NA 2195 (78.7) 250 (62.2) NA NA NA NA Abbreviation: NA, not applicable.

) Colostomy, No. (%) Yes NA NA NA NA NA NA 356 (12.8) 22 (5.5) <.001 NA NA NA NA NA NA No NA NA NA NA 2432 (87.2) 380 (94.5) NA NA NA NA Low anterior resection, No. (%) Yes NA NA NA NA NA NA 593 (21.3) 152 (37.8) <.001 NA NA NA NA NA NA No NA NA NA NA 2195 (78.7) 250 (62.2) NA NA NA NA Abbreviation: NA, not applicable. a Less restrictive health plans: comprehensive, preferred provider organization. b More restrictive health plans: basic or major medical, exclusive provider organization, health maintenance organization, noncapitated point of service, point of service with capitation or partially capitated point of service, consumer-driven health plan, and high-deductible health plan. OOP Costs In IPTW-adjusted analyses accounting for the OOP costs in the baseline period (−380 to −15 days), the robotic approach was associated with lower OOP costs for all procedures examined: –$137.75 (95% CI, −$240.24 to −$38.63) for RP (P = .006); –$640.63 (95% CI, −$933.62 to −$368.79) for HYS (P < .001); –$1140.54 (95% CI, −$1397.79 to −$896.54) for PC (P < .001); –$728.32 (95% CI, −$1126.90 to −$366.08) for RN (P < .001); and –$302.74 (95% CI, −$523.14 to −$97.10) for PN (P = .003) (Table 2).

examined: –$137.75 (95% CI, −$240.24 to −$38.63) for RP (P = .006); –$640.63 (95% CI, −$933.62 to −$368.79) for HYS (P < .001); –$1140.54 (95% CI, −$1397.79 to −$896.54) for PC (P < .001); –$728.32 (95% CI, −$1126.90 to −$366.08) for RN (P < .001); and –$302.74 (95% CI, −$523.14 to −$97.10) for PN (P = .003) (Table 2). Table 2. Adjusted Differences in OOP Costs Between Patients Undergoing Open and Patients Undergoing Robotic Surgery Surgery Mean OOP Costs, $ Adjusted Difference in OOP, $ (95% CI)a P Value Radical prostatectomy Open 3151.43 137.75 (38.63-240.24) .006 Robotic 2888.57 Hysterectomy Open 3769.22 640.63 (368.79-933.62) <.001 Robotic 3011.26 Partial colectomy Open 4620.09 1140.54 (896.54-1397.79) <.001 Robotic 3435.48 Radical nephrectomy Open 4002.82 728.32 (366.08-1126.90) <.001 Robotic 3371.95 Partial nephrectomy Open 3177.02 302.74 (97.10-523.14) .003 Robotic 2816.95 Abbreviation: OOP, out-of-pocket. a Adjusted for OOP costs in the baseline period and weighted by the inverse probability of receiving robotic surgery based on baseline covariates.

Table 2. Adjusted Differences in OOP Costs Between Patients Undergoing Open and Patients Undergoing Robotic Surgery Surgery Mean OOP Costs, $ Adjusted Difference in OOP, $ (95% CI)a P Value Radical prostatectomy Open 3151.43 137.75 (38.63-240.24) .006 Robotic 2888.57 Hysterectomy Open 3769.22 640.63 (368.79-933.62) <.001 Robotic 3011.26 Partial colectomy Open 4620.09 1140.54 (896.54-1397.79) <.001 Robotic 3435.48 Radical nephrectomy Open 4002.82 728.32 (366.08-1126.90) <.001 Robotic 3371.95 Partial nephrectomy Open 3177.02 302.74 (97.10-523.14) .003 Robotic 2816.95 Abbreviation: OOP, out-of-pocket. a Adjusted for OOP costs in the baseline period and weighted by the inverse probability of receiving robotic surgery based on baseline covariates. Total Payments In IPTW-adjusted analyses accounting for the total payments in the baseline period (−380 to −15 days), the robotic approach was associated with lower total payments for all procedures examined: –$3872.62 (95% CI, −$5385.49 to −$2399.04) for RP (P < .001); –$29 640.69 (95% CI, −$36 243.82 to −$23 465.94) for HYS (P < .001); –$38 151.74 (95% CI, −$46 386.16 to −$30 346.22) for PC (P < .001); –$33 394.15 (95% CI, −$42 603.03 to −$24 955.20) for RN (P < .001); and –$9162.52 (95% CI, −$12 728.33 to −$5781.99) for PN (P < .001) (Table 3).

5.49 to −$2399.04) for RP (P < .001); –$29 640.69 (95% CI, −$36 243.82 to −$23 465.94) for HYS (P < .001); –$38 151.74 (95% CI, −$46 386.16 to −$30 346.22) for PC (P < .001); –$33 394.15 (95% CI, −$42 603.03 to −$24 955.20) for RN (P < .001); and –$9162.52 (95% CI, −$12 728.33 to −$5781.99) for PN (P < .001) (Table 3). Table 3. Adjusted Differences in Total Costs Between Patients Undergoing Open and Patients Undergoing Robotic Surgery Surgery Total Costs, Mean, $ Adjusted Difference in Total Costs, $ (95% CI)a P Value Radical prostatectomy Open 54 529.42 3872.62 (2399.04-5385.49) <.001 Robotic 49 406.32 Hysterectomy Open 98 045.31 29 640.69 (23 465.94-36 243.82) <.001 Robotic 68 503.97 Partial colectomy Open 158 911.64 38 151.74 (30 346.22-46 386.16) <.001 Robotic 113 033.10 Radical nephrectomy Open 105 899.26 33 394.15 (24 955.20-42 603.03) <.001 Robotic 77 434.54 Partial nephrectomy Open 66 057.34 9162.52 (5781.99-12 728.33) <.001 Robotic 55 791.82 a Adjusted for total costs in the baseline period and weighted by the inverse probability of receiving robotic surgery based on baseline covariates. Length of Stay In IPTW-adjusted analyses, the robotic approach was associated with shorter LOS for all procedures examined: −0.94 days (95% CI, −1.02 to −0.85 days) for RP (P < .001); −2.28 days (95% CI, −2.53 to −2.04 days) for HYS (P < .001); −3.18 days (95% CI, −3.52 to −2.83 days) for PC (P < .001); −2.34 days (95% CI, −2.66 to −2.03 days) for RN (P < .001); and −1.59 days (95% CI, −1.77 to −1.41 days) for PN (P < .001) (eTable 1 in the Supplement).

5% CI, −1.02 to −0.85 days) for RP (P < .001); −2.28 days (95% CI, −2.53 to −2.04 days) for HYS (P < .001); −3.18 days (95% CI, −3.52 to −2.83 days) for PC (P < .001); −2.34 days (95% CI, −2.66 to −2.03 days) for RN (P < .001); and −1.59 days (95% CI, −1.77 to −1.41 days) for PN (P < .001) (eTable 1 in the Supplement). Additional Cost and Outlier Analysis For the perioperative period (−14 to 28 days), adjusted OOP costs were significantly lower for the robotic option for PC (–$471.90 [95% CI, −$651.84 to −$305.81]; P < .001) and RN (–$570.46 [95% CI, −$855.66 to −$320.35]; P < .001) but not for RN, HYS, and PN (eTable 2A in the Supplement). In the same period, adjusted total payments were significantly lower for all robotic procedures except RP (eTable 2B in the Supplement). Last, at 3 months (−14 to 90 days), adjusted OOP costs as well as adjusted total payments were significantly lower for all robotic procedures except RP (eTable 3A and 3B in the Supplement). Our outlier analysis also demonstrated that, apart from infrequent values for OOP costs for RP, PC, RN, and PN and total payments for RP, HYS, PC, and RN, the variations remained generally narrow (eFigure in the Supplement).

significantly lower for all robotic procedures except RP (eTable 3A and 3B in the Supplement). Our outlier analysis also demonstrated that, apart from infrequent values for OOP costs for RP, PC, RN, and PN and total payments for RP, HYS, PC, and RN, the variations remained generally narrow (eFigure in the Supplement). Discussion In this study of 15 893 adults within a large nationally representative cohort of privately insured patients, we found significantly lower OOP and total payments associated with the robotic approach for all 5 studied oncologic procedures (Figure 2). Notwithstanding the equivocal evidence regarding clinical benefit12 and a contentious debate on the value rendered by robotic oncologic surgery, evidence suggests that the robotic approach is assuming a greater role in urologic, gynecologic, and general surgery procedures.6,19 This increase is in spite of evidence that patients express greater disillusionment with robotic surgery after the procedure,20 of gaps in the literature on long-term cost and quality-of-life implications for patients who may not benefit from robotic procedures, and of a recent US Food and Drug Administration warning against using the robotic approach in several cancer-related surgical procedures.21 Although previous investigations have focused on total health care spending associated with robotic surgery to determine the value of robotic surgery, it is necessary to understand whether the burden of these costs falls on the patients directly (in the form of higher OOP costs) or on the hospitals where patients seek care. This report provides the first comprehensive economic assessment, to our knowledge, of variations in total and OOP costs when comparing robotic and open surgery.

to understand whether the burden of these costs falls on the patients directly (in the form of higher OOP costs) or on the hospitals where patients seek care. This report provides the first comprehensive economic assessment, to our knowledge, of variations in total and OOP costs when comparing robotic and open surgery. Figure 2. Differences in Out-of-Pocket Costs The circles outside the bars represent outlier values. The horizontal line within each box represents the median value. The circles inside the boxes represent the mean values of each group. The error bars below and above the boxes represent the minimum and maximum values, respectively.

ifferences in Out-of-Pocket Costs The circles outside the bars represent outlier values. The horizontal line within each box represents the median value. The circles inside the boxes represent the mean values of each group. The error bars below and above the boxes represent the minimum and maximum values, respectively. Our results of lower OOP and total payments for robotic surgery should be interpreted carefully, given the scope of our analysis. First, these analyses do not account for the costs of procuring and maintaining a robotic system (ranging from $0.5 to $2.5 million).6 Second, previous economic analyses have demonstrated that robotic surgery could be more expensive perioperatively than open surgery,19,22 when considering the costs of robotic maintenance, as well as disposable instruments (costs range from $600 to $1000, and the instruments are generally limited to 10 uses). Another factor that could be associated with these differences is hospital LOS, which has been demonstrated to be significantly reduced for patients undergoing robotic procedures.9 In our analysis, we found that the mean LOS was shorter for the robotic approach in all procedures examined. It is possible that these differences in LOS may also be a factor associated with higher payments for open surgery and may explain the differences in total payments, given that hospital-related costs likely exceed those of the other categories comprising total payments (eg, pharmaceuticals).

in all procedures examined. It is possible that these differences in LOS may also be a factor associated with higher payments for open surgery and may explain the differences in total payments, given that hospital-related costs likely exceed those of the other categories comprising total payments (eg, pharmaceuticals). Our findings also contrast with those of previous studies. Yu et al23 reported an approximately $1000 additional cost for those undergoing robotic prostatectomy. However, their analysis relied on inpatient costs for all payers, whereas our analysis examines perioperative costs extending beyond the inpatient stay, for private payers. Also, Nguyen et al24 found that minimally invasive prostatectomy (mostly robotic assisted) costs $236 more than open prostatectomy. Their analysis included older patients with Medicare coverage and calculated costs over the course of a year, which would understandably include health care use not observed among younger, privately insured patients (the population of our study). In our outlier analysis, we observed that, apart from occasional values for OOP costs for RP, PC, RN, and PN and total payments for RP, HYS, PC, and RN, the variations generally remained narrow. These outlier values, however, could be a result of patient-level differences or other factors. However, these outlier values are unlikely to shift the differences in mean payments between open and robotic procedures, and because we have accounted for patient-level baseline characteristics, we decided to include those values in our analysis.

owever, could be a result of patient-level differences or other factors. However, these outlier values are unlikely to shift the differences in mean payments between open and robotic procedures, and because we have accounted for patient-level baseline characteristics, we decided to include those values in our analysis. Additional cost calculations demonstrated that, with the increase in duration of care, the differences in costs became increasingly pronounced. For the perioperative periods (−14 to 28 days), adjusted OOP costs were significantly lower for the robotic option for PC and RN but not for RN, HYS, and PN. In the same period, adjusted total payments were significantly lower for all robotic procedures except RP. At 3 months (−14 to 90 days), adjusted OOP costs as well as adjusted total payments were significantly lower for all robotic procedures except RP.

for the robotic option for PC and RN but not for RN, HYS, and PN. In the same period, adjusted total payments were significantly lower for all robotic procedures except RP. At 3 months (−14 to 90 days), adjusted OOP costs as well as adjusted total payments were significantly lower for all robotic procedures except RP. Our analyses indicate that the additional costs of robot acquisition and maintenance are seemingly not paid by private health insurers (approximated here by total payments) or patients (approximated here by OOP costs); if such is the case, by extension, these costs appear to be absorbed by the hospitals. Although the exact reasons why hospitals have been willing to absorb or subsidize costs associated with robotic surgery remain unclear, there are some plausible explanations. In recent years, there has been a rapid diffusion in the adoption of robotic surgery.25 Given this trend, it is possible that the marginal cost of undergoing robotic vs open surgery is lower—that is, while the total reimbursements are lower for the hospitals, the net profit is still higher. Our analyses of the most recent years available (2012-2017) provide a more thorough understanding of this trend because previous investigations relied on data from the last decade.13

ing robotic vs open surgery is lower—that is, while the total reimbursements are lower for the hospitals, the net profit is still higher. Our analyses of the most recent years available (2012-2017) provide a more thorough understanding of this trend because previous investigations relied on data from the last decade.13 Since the last decade, the adoption of robotic surgery has increased considerably.25 Although profitability remains an important motivator for rapidly adopting robotic surgery,26 a key reason why hospitals are willing to absorb the high upfront costs of robotic surgery is patient demand.27 Evidence supports the finding that direct-to-consumer advertising of robotic surgery increases demand.28,29 This higher demand could influence hospitals offering robotic surgery as a viable option to retain market share and stay competitive. Our finding of significantly lower OOP costs associated with the robotic approach is likely to compound this trend. In addition to acquiescing to patient demand, another mechanism that may be associated with the rapid acquisition and implementation of robotic surgery could be the ability of nonprofit hospitals to access tax-exempt financial instruments toward using debt for operational growth.30

s have led to greater investment in building cancer surgery units that use robotic surgery as a primary treatment modality. Several studies have shown the conditions in which the excess costs of robotic surgery are mitigated: when hospital volume is high and operative time is low, robotic surgery can cost less.34,35,36 Furthermore, another reason why hospitals may be willing to absorb the costs of the acquisition of surgical robotics could be a consequence of changes in residency training programs. Recent evidence points to surgical residency programs increasingly training their residents in operating on robotic platforms,37 leading trainees to be better equipped for the robotic approach. It has also been suggested that the laparoscopic approach has a steep learning curve, which could further augment training on a robotic platform.38 This change is compounded by reports that high-volume centers are more likely to use a robotic platform.39 Because teaching hospitals tend to be high-volume centers, it is possible there is an increased focus—albeit unintended—on learning common surgical procedures on a robotic platform, especially in urology and gynecology. Residency training programs and hospitals should be cautious about this trend because it may be associated with long-term harmful consequences, such as the development of surgeons who are not well trained in open procedures, which may impede timely access to surgical care in low-resource settings that have yet to acquire robotic surgical platforms.

d hospitals should be cautious about this trend because it may be associated with long-term harmful consequences, such as the development of surgeons who are not well trained in open procedures, which may impede timely access to surgical care in low-resource settings that have yet to acquire robotic surgical platforms. Given that the cost of the acquisition and maintenance of surgical robotics are not accounted for in this analysis, it is plausible that robotic surgery exhibits small gains compared with the conventional open approaches through shorter LOS, use of pain medication, and use of laboratory tests, among other factors. From an accounting perspective, total cost differences between robotic and open approaches could also be associated with fewer postacute care days and lower morbidity.7,8,13 Our findings are concordant with a recent investigation by Motz et al,40 who found that transoral robotic surgery was associated with significantly lower total treatment-related costs (–$22 724). Previous economic analyses have explored the possibility that inclusion of robotic surgery has discrete benefits for hospitals in terms of revenue because the costs of procuring and maintaining a robotic system (ranging from $0.5 to $2.5 million) can be recuperated when a steady volume of 100 to 150 procedures per year is maintained.35,41

alyses have explored the possibility that inclusion of robotic surgery has discrete benefits for hospitals in terms of revenue because the costs of procuring and maintaining a robotic system (ranging from $0.5 to $2.5 million) can be recuperated when a steady volume of 100 to 150 procedures per year is maintained.35,41 Limitations Although our analysis accounted for potential confounders, our study has certain limitations. The MarketScan database does not include details on patient race/ethnicity and clinical factors such as stage of cancer, grade of cancer, preceding abdominal surgical procedures, or body mass index, factors that could influence the decision to undergo open or robotic surgery. As such, clinically meaningful differences among patients undergoing these procedures may not be satisfactorily captured; while we have adjusted for baseline characteristics, it is possible that omitted-variable bias could affect our analysis. Also, administrative data are limited in their ability to control for unknown confounders that could explicate these differences. Our analysis is limited by calculating medical expenditure data for employees of self-insured firms alone. Although it is unlikely that this limitation would be associated with the type of procedure that a given patient would undergo, it could limit the generalizability of our analysis. Another limitation could be the exclusion of Medicare beneficiaries, who are covered under a different reimbursement structure. However, OOP costs and total payments are much more relevant for those covered through private insurance because this type of plan tends to have a tier-based structure that can greatly limit the ability and affordability of patients to choose between robotic and open surgery options. We also recognize the absence of measurement and comparison of clinical outcomes under the open and robotic approach. This limitation was mitigated by undertaking an assessment of 5 different types of procedures to allow for variability because each of these procedures has previously reported similar clinical outcomes between the open and robotic options.

urement and comparison of clinical outcomes under the open and robotic approach. This limitation was mitigated by undertaking an assessment of 5 different types of procedures to allow for variability because each of these procedures has previously reported similar clinical outcomes between the open and robotic options. Conclusions We observed significant variation in perioperative costs according to surgical technique for both patients (OOP costs) and payers (total payments), with the robotic approach associated with significantly lower OOP costs for all studied oncologic procedures. These results highlight the complexity of economic factors that are associated with the rapid adoption and possible subsidization of the robotic approach for common surgically amenable conditions and lay a foundation for future work on this issue. Supplement. eFigure. Outlier Analysis for Out-of-Pocket (OOP) Costs and Total Payments Between Patients Undergoing Open and Robotic Radical Prostatectomy, Hysterectomy, Partial Colectomy, Radical Nephrectomy, and Partial Nephrectomy eTable 1. Adjusted Differences in Length of Stay (LoS) for Patients Undergoing Open and Robotic Radical Prostatectomy, Hysterectomy, Partial Colectomy, Radical Nephrectomy, and Partial Nephrectomy – Weighted by the Inverse Probability of Receiving Robotic Surgery Based on Baseline Covariates

Supplement. eFigure. Outlier Analysis for Out-of-Pocket (OOP) Costs and Total Payments Between Patients Undergoing Open and Robotic Radical Prostatectomy, Hysterectomy, Partial Colectomy, Radical Nephrectomy, and Partial Nephrectomy eTable 1. Adjusted Differences in Length of Stay (LoS) for Patients Undergoing Open and Robotic Radical Prostatectomy, Hysterectomy, Partial Colectomy, Radical Nephrectomy, and Partial Nephrectomy – Weighted by the Inverse Probability of Receiving Robotic Surgery Based on Baseline Covariates eTable 2. Adjusted Differences in Perioperative (-14 to +28 days) Out-of-Pocket Costs and Total Payments for Patients Undergoing Open and Robotic Radical Prostatectomy, Hysterectomy, Partial Colectomy, Radical Nephrectomy, and Partial Nephrectomy – Adjusted for OOP Costs in Baseline Period and Weighted by the Inverse Probability of Receiving Robotic Surgery Based on Baseline Covariates eTable 3. Adjusted Differences in 3 Month (-14 to +90 days) Out-of-Pocket Costs and Total Payments for Patients Undergoing Open and Robotic Radical Prostatectomy, Hysterectomy, Partial Colectomy, Radical Nephrectomy, and Partial Nephrectomy – Adjusted for OOP Costs in Baseline Period and Weighted by the Inverse Probability of Receiving Robotic Surgery Based on Baseline Covariates eAppendix. ICD-9, ICD-10, and CPT Codes for Disease States and Procedures Used in this Analysis Click here for additional data file.

Introduction Black men are more likely than white men to be diagnosed with and die of prostate cancer.1 Current evidence attributes this to racial differences in both tumor biology and access to care.2,3,4 These differences may also be greatest in specific disease states, such as low-risk prostate cancer.5,6 While race-based prostate cancer survival differences have been identified in population-based samples,7 little is known about how these differences vary geographically within the US. Because populations of racial and ethnic minority groups are concentrated in certain geographic areas of the US, it is possible that geographic differences in cancer survival may yield apparent race-based differences in survival and vice versa. Indeed, the mortality for many conditions, ranging from cardiovascular disease to cancer, varies significantly among counties. While it is possible that the underlying geographic variation may be associated with biology and clusters of biologically related individuals, it is also plausible that such variation is associated with other factors, such as access to insurance and high-quality care.

ase to cancer, varies significantly among counties. While it is possible that the underlying geographic variation may be associated with biology and clusters of biologically related individuals, it is also plausible that such variation is associated with other factors, such as access to insurance and high-quality care. In a 2018 study,2 we found that there were significant differences in prostate cancer care among hospitals, wherein nearly half the examined hospitals were more likely to provide definitive therapy to white men compared with black men. Moreover, we found that hospitals that primarily treated people in racial minority groups were associated with lower odds of definitive therapy and longer time to definitive therapy, regardless of patient race.2,4 Additionally, we found that when patient and sociodemographic factors are taken into account, the differences were actually reversed, that is, black men may have had better survival than white men.3 Taken together, these findings may provide more evidence of a racial disparity, suggesting a form of social injustice rather than racial difference attributable to biology.

emographic factors are taken into account, the differences were actually reversed, that is, black men may have had better survival than white men.3 Taken together, these findings may provide more evidence of a racial disparity, suggesting a form of social injustice rather than racial difference attributable to biology. To better understand the extent to which geographic variability may underlie observed differences in survival associated with race, we used a nationally representative cancer database to characterize geographic registry-based variation in prostate cancer–specific mortality differences between black and white men. We hypothesized that differences in prostate cancer outcomes associated with race would be greatest in areas with large populations of racial minority groups and significant social and economic barriers to health care.

gistry-based variation in prostate cancer–specific mortality differences between black and white men. We hypothesized that differences in prostate cancer outcomes associated with race would be greatest in areas with large populations of racial minority groups and significant social and economic barriers to health care. Methods Data Source This cohort study used the Surveillance, Epidemiology, and End Results (SEER) database, a collection of 18 defined geographic areas (ie, registries) sponsored by the US National Cancer Institute. These registries include Alaska Native Tumor Registry; Connecticut; Detroit (Metropolitan), Michigan; Atlanta (Metropolitan), Georgia; Greater Georgia; Rural Georgia; San Francisco-Oakland metropolitan statistical area, California; San Jose-Monterey, California; Greater California; Hawaii; Iowa; Kentucky; Los Angeles, California; Louisiana, New Mexico, New Jersey, Seattle (Puget Sound), Washington; and Utah. The database covers about one-third of the US population, and the registries are chosen to be nationally representative.8 The SEER database contains detailed data on patient sociodemographic and tumor characteristics as well as survival and treatment patterns. The reporting of our methods are consistent with Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline. The Brigham and Women’s Hospital institutional review board deemed this study exempt from review and informed consent because data were deidentified.

nd treatment patterns. The reporting of our methods are consistent with Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline. The Brigham and Women’s Hospital institutional review board deemed this study exempt from review and informed consent because data were deidentified. Cohort Selection We identified men 18 years and older with biopsy-confirmed prostate cancer, including all stages and grades, between January 1, 2007, and December 31, 2014. We chose this time frame because 2007 was the first year that the full complement of covariates was available, including insurance status, and 2014 was the last year of available follow-up data at the time of analysis. We excluded men who were missing information on cancer stage, grade, prostate-specific antigen level, and survival follow-up data. We also excluded men in the Alaska Native Tumor Registry, as there are neither black nor white individuals with prostate cancer collected in this registry. Outcomes The outcome of interest was prostate cancer–specific mortality, based on the SEER variable for cause-specific death classification. We investigated mortality differences between races in the entire cohort. Given that a recent study showed important differences in mortality between black and white men with low-risk prostate cancer,5 we also performed the analysis stratified by Gleason grade group, as grade group 1 (low risk) vs grade groups 2 through 5 (intermediate to high risk).9,10

s between races in the entire cohort. Given that a recent study showed important differences in mortality between black and white men with low-risk prostate cancer,5 we also performed the analysis stratified by Gleason grade group, as grade group 1 (low risk) vs grade groups 2 through 5 (intermediate to high risk).9,10 Exposure Variable The exposure of interest was self-reported patient race, classified as black, white, or other. We fit an interaction term between race and registry to test whether the association of race with the study outcome was statistically significantly different across 18 component registries that compose SEER. Covariates Covariates included demographic characteristics, such as age, county-level median household income, county-level education level (based on percentage of adults without a high school degree), year of diagnosis, and insurance status (insured vs uninsured). We also included clinical characteristics, such as prostate-specific antigen value, Gleason grade group, clinical TNM stage, and receipt of definitive treatment (ie, radical prostatectomy or any form of radiation therapy).

ts without a high school degree), year of diagnosis, and insurance status (insured vs uninsured). We also included clinical characteristics, such as prostate-specific antigen value, Gleason grade group, clinical TNM stage, and receipt of definitive treatment (ie, radical prostatectomy or any form of radiation therapy). Statistical Analysis We reported descriptive statistics using frequencies and proportions for categorical variables; medians and interquartile ranges were used for continuous variables. For baseline patient characteristics, we used the χ2 test for categorical variables and t test for continuous variables. We constructed Fine and Gray competing risks regression models, including an unadjusted model and an adjusted model with an interaction term between race and registry to assess the association of race with cancer-specific mortality in each registry. Death due to prostate cancer was the primary event of interest, and death due to any other cause was the competing event. We ranked registries by the estimated adjusted hazard ratio (AHR) of prostate cancer–specific mortality for black vs white men. We defined 2-sided statistical significance as P < .05. Analyses were performed using Stata statistical software version 14.0 (StataCorp) and R statistical software version 3.4.1 (R Project for Statistical Computing). Analysis was performed from September 5 to December 25, 2018.

ecific mortality for black vs white men. We defined 2-sided statistical significance as P < .05. Analyses were performed using Stata statistical software version 14.0 (StataCorp) and R statistical software version 3.4.1 (R Project for Statistical Computing). Analysis was performed from September 5 to December 25, 2018. Results A total of 406 628 men fit our initial inclusion criteria. We excluded 176 762 men (43.5%) for missing data and 95 men (<0.1%) from the Alaska Native cancer registry because this registry contained no black or white men. Our final study sample included 229 771 men (mean [SD] age at diagnosis, 64.9 [8.8] years), of whom 35 006 (15.2%) were black and 178 204 (77.6%) were white. Median (interquartile range) follow-up time was 56 (31-78) months for white men and 53 (27-75) months for black men. Mean (SD) age at diagnosis was 65.2 (8.7) years for white men and 62.6 (8.8) years for black men. Black men were more likely than men of other races to be uninsured (black: 1057 men [3.0%]; white: 2059 men [1.2%]; other: 204 men [1.2%]), have low education level (black: 21 731 men [62.1%]; white: 83 655 men [47.0%]; other: 7261 men [43.9%]), and have low income (black: 21 318 men [60.9%]; white: 85 582 men [48.0%]; other: 6047 men [36.5%]) (Table 1). There were 4773 prostate cancer deaths among white men (2.7%) and 1250 prostate cancer deaths among black men (3.6%). Receipt of definitive treatment (ie, prostatectomy or radiation treatment) was recorded for 125 493 white men (70.4%) and 23 699 black men (67.7%). In the competing-risks regression, black men had the highest risk of mortality (AHR, 1.39; 95% CI, 1.30-1.48). Table 2 details the population of patients by race within each registry. The eTable in the Supplement presents additional characteristics of each registry.

5 493 white men (70.4%) and 23 699 black men (67.7%). In the competing-risks regression, black men had the highest risk of mortality (AHR, 1.39; 95% CI, 1.30-1.48). Table 2 details the population of patients by race within each registry. The eTable in the Supplement presents additional characteristics of each registry. Table 1. Descriptive Characteristics of Men With Prostate Cancer in the Surveillance, Epidemiology, and End Results Database From 2007 to 2014 Characteristic Men, No. (%) P value Total (N = 229 771) White (n = 178 204) Black (n = 35 006) Other or unknown (n = 16 561) Age at diagnosis, y <50 8267 (3.6) 5559 (3.1) 2278 (6.5) 430 (2.6) <.001 50-59 55 571 (24.2) 41 460 (23.3) 10 692 (30.5) 3419 (20.6) 60-69 98 880 (43.0) 77 507 (43.5) 14 484 (41.4) 6889 (41.6) 70-79 55 234 (24.0) 44 143 (24.8) 6465 (18.5) 4626 (27.9) ≥80 11 819 (5.1) 9535 (5.4) 1087 (3.1) 1197 (7.2) Incomea Low 112 947 (49.2) 85 582 (48.0) 21 318 (60.9) 6047 (36.5) <.001 High 116 773 (50.8) 92 575 (52.0) 13 686 (39.1) 10 512 (63.5) Education levelb Low 112 647 (49.0) 83 655 (47.0) 21 731 (62.1) 7261 (43.9) <.001 High 117 073 (51.0) 94 502 (53.0) 13 273 (37.9) 9298 (56.2) Insurance status Insured 210 456 (91.6) 164 307 (92.2) 31 721 (90.6) 14 428 (87.1) <.001 Uninsured 3320 (1.4) 2059 (1.2) 1057 (3.0) 204 (1.2) Unknown 15 995 (7.0) 11 838 (6.6) 2228 (6.4) 1929 (11.7) Gleason grade group 1 52 452 (22.8) 41 802 (23.5) 7368 (21.1) 3282 (19.8) <.001 2 36 569 (15.9) 28 596 (16.1) 5704 (16.3) 2269 (13.7) 3 13 823 (6.0) 10 708 (6.0) 2136 (6.1) 979 (5.9) 4 95 934 (41.8) 73 755 (41.4) 14 596 (41.7) 7583 (45.8) 5 30 993 (13.5) 23 343 (13.1) 5202 (14.9) 2448 (14.8) Receipt of definitive treatmentc No 70 691 (30.8) 52 711 (29.6) 11 307 (32.3) 6673 (40.3) <.001 Yes 159 080 (69.2) 125 493 (70.4) 23 699 (67.7) 9888 (59.7) a High income was defined as income above the 50th percentile of county-level median household income, and low income was defined as income at or below the 50th percentile of county-level median household income.

) 52 711 (29.6) 11 307 (32.3) 6673 (40.3) <.001 Yes 159 080 (69.2) 125 493 (70.4) 23 699 (67.7) 9888 (59.7) a High income was defined as income above the 50th percentile of county-level median household income, and low income was defined as income at or below the 50th percentile of county-level median household income. b Low education was defined as individuals with less than a high school diploma or equivalent, and high education included individuals with a high school diploma or equivalent or higher. c Defined as prostatectomy or radiation treatment.

) 52 711 (29.6) 11 307 (32.3) 6673 (40.3) <.001 Yes 159 080 (69.2) 125 493 (70.4) 23 699 (67.7) 9888 (59.7) a High income was defined as income above the 50th percentile of county-level median household income, and low income was defined as income at or below the 50th percentile of county-level median household income. b Low education was defined as individuals with less than a high school diploma or equivalent, and high education included individuals with a high school diploma or equivalent or higher. c Defined as prostatectomy or radiation treatment. Table 2. Population of Men With Prostate Cancer by Race Within Each Registry of the Surveillance, Epidemiology, and End Results Database From 2007-2014 Registry Men, No. (%) White (n = 178 204) Black (n = 35 006) Other or unknown (n = 16 561) Total (N = 229 771) Connecticut 8348 (4.7) 1047 (3.0) 353 (2.1) 9748 (4.2) Detroit (Metropolitan), Michigan 8580 (4.8) 3609 (10.3) 452 (2.7) 12 641 (5.5) Atlanta (Metropolitan), Georgia 4992 (2.8) 4029 (11.5) 229 (1.4) 9250 (4.0) Greater Georgia 12 203 (6.9) 5369 (15.3) 162 (1.0) 17 734 (7.7) Rural Georgia 305 (0.2) 201 (0.6) 3 (<0.1) 509 (0.2) San Francisco-Oakland metropolitan statistical area, California 9393 (5.3) 1797 (5.1) 2336 (14.1) 13 526 (5.9) San Jose-Monterey, California 5840 (3.3) 317 (0.9) 1250 (7.6) 7407 (3.2) Greater California 41 709 (23.4) 3589 (10.3) 4414 (26.7) 49 712 (21.6) Hawaii 894 (0.5) 84 (0.2) 1883 (11.4) 2861 (1.3) Iowa 8207 (4.6) 188 (0.5) 117 (0.7) 8512 (3.7) Kentucky 9661 (5.4) 900 (2.6) 151 (0.9) 10 712 (4.6) Los Angeles, California 13 230 (7.4) 3474 (9.9) 2511 (15.2) 19 215 (8.4) Louisiana 10 287 (5.8) 5318 (15.2) 129 (0.8) 15 734 (6.9) New Mexico 4183 (2.4) 115 (0.3) 248 (1.5) 4546 (2.0) New Jersey 22 344 (12.5) 4116 (11.8) 1414 (8.5) 27 874 (12.1) Seattle (Puget Sound), Washington 11 986 (6.7) 792 (2.3) 777 (4.7) 13 555 (5.9) Utah 6042 (3.4) 61 (0.2) 132 (0.8) 6235 (2.7) The stratified multivariable analyses of individuals with Gleason grade group 1 disease and individuals with Gleason grade groups 2 through 5 disease revealed that the prostate cancer–specific mortality difference between black and white men was present in both groups but with a larger effect size for Gleason grade group 1 disease. Within this group, statistically significantly worse survival for black men was seen in 4 of the 17 included registries, and the greatest race-based survival difference for those with Gleason grade group 1 disease was in the Atlanta registry (AHR, 5.49; 95% CI, 2.03-14.87), followed by New Jersey (AHR, 2.60; 95% CI, 1.53-4.40), Greater Georgia (AHR, 1.88; 95% CI, 1.10-3.22), and Louisiana (AHR, 1.80; 95% CI, 1.06-3.07).

luded registries, and the greatest race-based survival difference for those with Gleason grade group 1 disease was in the Atlanta registry (AHR, 5.49; 95% CI, 2.03-14.87), followed by New Jersey (AHR, 2.60; 95% CI, 1.53-4.40), Greater Georgia (AHR, 1.88; 95% CI, 1.10-3.22), and Louisiana (AHR, 1.80; 95% CI, 1.06-3.07). There were no race-based survival differences for men with Gleason grade group 1 disease in the other registries. For men with Gleason grade groups 2 through 5 disease, survival differences were seen in Detroit (AHR, 1.65; 95% CI, 1.32-2.08), Atlanta (AHR, 1.88; 95% CI, 1.46-2.45), New Jersey (AHR, 1.52; 95% CI, 1.24-1.87), Greater Georgia (AHR, 1.29; 95% CI, 1.07-1.56), and Louisiana (AHR, 1.28; 95% CI, 1.07-1.54); however, the effect sizes were smaller than that seen in Gleason grade group 1 disease for all registries besides Detroit (Table 3).

.32-2.08), Atlanta (AHR, 1.88; 95% CI, 1.46-2.45), New Jersey (AHR, 1.52; 95% CI, 1.24-1.87), Greater Georgia (AHR, 1.29; 95% CI, 1.07-1.56), and Louisiana (AHR, 1.28; 95% CI, 1.07-1.54); however, the effect sizes were smaller than that seen in Gleason grade group 1 disease for all registries besides Detroit (Table 3). Table 3. Adjusted Hazard Ratio Analysis With Gleason Grade 1 and Grades 2 Through 5 Disease for Prostate Cancer–Specific Mortality Among Black Men Stratified by Region Registry AHR (95% CI)a P value Gleason grade group 1 Connecticut 1.58 (0.47-5.39) .46 Detroit (Metropolitan) 1.61 (0.52-4.92) .41 Atlanta (Metropolitan) 5.49 (2.03-14.87) .001 Greater Georgia 1.88 (1.10-3.22) .02 Rural Georgiab NA NA San Francisco-Oakland metropolitan statistical area, California 0.94 (0.27-3.20) .93 San Jose-Monterey, Californiab NA NA Greater California 1.17 (0.65-2.11) .60 Hawaiib NA NA Iowab NA NA Kentuckyb NA NA Los Angeles, California 1.63 (0.86-3.10) .14 Louisiana 1.80 (1.06-3.07) .03 New Mexico 2.85 (0.37-21.70) .31 New Jersey 2.60 (1.53-4.40) .001 Seattle (Puget Sound), Washington 1.87 (0.44-8.06) .40 Utahb NA NA Gleason grade groups 2-5 Connecticut 1.36 (0.93-1.99) .11 Detroit (Metropolitan), Michigan 1.65 (1.32-2.08) <.001 Atlanta (Metropolitan), Georgia 1.88 (1.46-2.45) <.001 Greater Georgia 1.29 (1.07-1.56) .008 Rural Georgia 3.69 (0.96-14.20) .06 San Francisco-Oakland metropolitan statistical area, California 1.25 (0.92-1.68) .15 San Jose-Monterey, California 1.63 (0.88-3.01) .12 Greater California 1.10 (0.91-1.36) .32 Hawaiib NA NA Iowa 0.72 (0.26-1.95) .52 Kentucky 1.08 (0.75-1.57) .68 Los Angeles, California 1.16 (0.93-1.46) .17 Louisiana 1.28 (1.07-1.54) .008 New Mexico 0.91 (0.29-2.84) .88 New Jersey 1.52 (1.24-1.87) <.001 Seattle (Puget Sound), Washington 1.42 (0.96-2.13) .08 Utah 1.07 (0.15-7.85) .95 Abbreviations: AHR, adjusted hazard ratio; NA, not applicable.

.52 Kentucky 1.08 (0.75-1.57) .68 Los Angeles, California 1.16 (0.93-1.46) .17 Louisiana 1.28 (1.07-1.54) .008 New Mexico 0.91 (0.29-2.84) .88 New Jersey 1.52 (1.24-1.87) <.001 Seattle (Puget Sound), Washington 1.42 (0.96-2.13) .08 Utah 1.07 (0.15-7.85) .95 Abbreviations: AHR, adjusted hazard ratio; NA, not applicable. a White men were used as the reference in each group. b Unable to calculate owing to insufficient events between races. Discussion In this cohort study of the nationally representative SEER cancer registry, we found that statistically significant race-based differences in cancer-specific survival were confined mainly to low-grade prostate cancer and present in less than one-third of included SEER registries. These findings are consistent with the idea that racial differences in prostate cancer survival are subject to geographic variation. Historically, racial differences in prostate cancer survival have been believed to be associated with a combination of differences in host genotype, differences in risk factors, and tumor biological differences, as well as differences in access to appropriate care.11,12,13,14 The relative contributions of these factors remain a topic of debate with substantial clinical and policy implications. For example, Jiang et al15 suggest that the differences are mostly associated with biological differences; therefore, they advocate for earlier and more rigorous prostate cancer screening in black men and early treatment of prostate cancer even at the low-risk stage.

e with substantial clinical and policy implications. For example, Jiang et al15 suggest that the differences are mostly associated with biological differences; therefore, they advocate for earlier and more rigorous prostate cancer screening in black men and early treatment of prostate cancer even at the low-risk stage. Regarding low–Gleason grade group cancer, a 2018 study by Mahal et al5 reported that the greatest survival difference between black and white men was seen in patients presenting with low-risk prostate cancer. Our study adds to that evidence; after adjusting for patient, disease, and treatment characteristics, analyses demonstrated no significant racial difference in prostate cancer survival for most registries, regardless of disease grade at presentation.

te men was seen in patients presenting with low-risk prostate cancer. Our study adds to that evidence; after adjusting for patient, disease, and treatment characteristics, analyses demonstrated no significant racial difference in prostate cancer survival for most registries, regardless of disease grade at presentation. Our analysis found that the differences in prostate cancer survival between black and white men for intermediate- and high-risk prostate cancer were present but lesser in magnitude compared with the difference in low-risk prostate cancer. These findings are in line with our 2019 study3 using data from the National Cancer Database, which found that when access to care, treatment, and cancer characteristics were accounted for, black race was not associated with a difference in overall survival for advanced prostate cancer. In fact, in that analysis, black race was associated with better overall survival.3 Data from several clinical trials point in the same direction. George et al16 reported that black men had better a progression-free survival rate compared with white men (16.6 months vs 11.5 months) when treated with abiraterone and prednisone. Moreover, a pooled analysis by Halabi et al17 of 9 randomized clinical trials for advanced prostate cancer found that overall survival was the same for black and white men. Taken together, these data may suggest that when black men are treated, they do equally well compared with white men. However, current evidence suggests that there is a significant gap in treatment rates between black and white men. For example, Underwood et al,18 as well as Moses et al,19 found that black and Hispanic men were significantly less likely to receive definitive therapy compared with white men. Surprisingly, higher tumor grade was associated with decreasing odds of definitive therapy for black and Hispanic men. In a 2018 study, Friedlander et al2 examined the facility-level variation in the use of definitive therapy among black and white men with intermediate- and high-risk prostate cancer and found that nearly half of the included institutions were more likely to provide definitive therapy to white men than black men. Similarly, a 2019 study4 found that men presenting with prostate cancer at hospitals that primarily treated people in racial/ethnic minority groups were less likely to receive definitive therapy and more likely to encounter delays to treatment, despite adjustment for race.

efinitive therapy to white men than black men. Similarly, a 2019 study4 found that men presenting with prostate cancer at hospitals that primarily treated people in racial/ethnic minority groups were less likely to receive definitive therapy and more likely to encounter delays to treatment, despite adjustment for race. For low-risk prostate cancer, we found a statistically significant difference in prostate cancer–specific mortality between black and white men within 4 SEER registries: 2 from Georgia plus Louisiana and New Jersey. These findings provide clarification points to the 2018 study by Mahal et al5 that found worse outcomes in black men with low-risk prostate cancer. Many experts have interpreted those findings as a call for more aggressive treatment in black men presenting with low-risk disease. Our findings suggest that there may be specific demographic characteristic differences in prostate cancer care within specific areas of the US rather than a generalized association of worse survival for black men presenting with low-risk prostate cancer. Considering the complexity of low-risk prostate cancer management, for which the standard of care option is now active surveillance, requiring rigorous follow-up visits with repeated blood tests, physical examinations, biopsies, and optionally, imaging as well as genomic tests, gaps in access to quality care may exacerbate difference between demographic groups, whether stratified according to race, income, or education. For example, Krishna et al20 found that black men presenting with low-risk prostate cancer were followed less closely after an initial period of active surveillance than white men.

in access to quality care may exacerbate difference between demographic groups, whether stratified according to race, income, or education. For example, Krishna et al20 found that black men presenting with low-risk prostate cancer were followed less closely after an initial period of active surveillance than white men. The findings of worse outcomes for black men in a wide spectrum of medical conditions specifically in the Atlanta area and in Georgia as a whole have been replicated in other non–prostate cancer–focused studies. Among large US cities, the widest gap in breast cancer mortality between black and white women was seen in Atlanta. These differences extend far beyond the scope of cancer care: the widest gap in US maternal mortality was seen in the state of Georgia.21 While our data cannot further characterize the causes of these differences, several hypotheses may be considered. Regardless of race, the mortality rates of cardiovascular disease, lower respiratory infections, meningitis, and asthma are also disproportionately high in Georgia.22,23,24,25 Translating these findings to prostate cancer, part of the reason may be associated with higher predisposition among inhabitants of Georgia. However, a combination of poor access to care cannot be discounted as a potential factor associated with racial differences for Georgia. In fact, a recent study showed that higher prostate cancer mortality incidence ratios in the state of Georgia were associated with worse health behavior and worse medical care.26 Our finding of a smaller magnitude but statistically significantly worse survival outcome for black men in New Jersey is also supported by prior literature.27 The cause for a stronger effect size seen in this registry is likely complex and multifactorial but may be associated with worse access to care among black men.

Our finding of a smaller magnitude but statistically significantly worse survival outcome for black men in New Jersey is also supported by prior literature.27 The cause for a stronger effect size seen in this registry is likely complex and multifactorial but may be associated with worse access to care among black men. Although examining racial mortality differences according to registry cannot pinpoint causative or mediating factors or identify specific mechanisms accountable for these differences, our findings may help identify target areas for future research or interventions in both primary and secondary prevention. Specifically, if racial/ethnic differences in cancer outcomes are being driven by a small number of areas with large populations of racial/ethnic minority groups, then these represent important targets for addressing racial inequality. On the basis of previous research,20,21,22,23,24,25 such efforts should focus on identifying characteristics present in these areas, such as environmental risk factors and barriers to health care access. For example, Servadio et al28 showed that black individuals in the Atlanta area experience disproportionately high exposure to air pollution and lower access to green space compared with white individuals, both of which are associated with a higher prevalence of chronic obstructive pulmonary disease, chronic heart disease, and stroke. A similar study may help better understand the associations of such determinants with prostate cancer mortality. Regarding interventions, given such a large difference in mortality found in Georgia and New Jersey, a randomized clinical trial may help inform the value of systematic prostate cancer screening in at-risk populations, which remains an unresolved issue in the latest report from the US Preventive Services Task Force.29

Regarding interventions, given such a large difference in mortality found in Georgia and New Jersey, a randomized clinical trial may help inform the value of systematic prostate cancer screening in at-risk populations, which remains an unresolved issue in the latest report from the US Preventive Services Task Force.29 Limitations Our study has several limitations that should be considered when interpreting our findings. This analysis is limited to the SEER registries, and it is plausible that other geographic areas of significant racial differences exist and would be worthy of further study. For example, there are several studies suggesting that similar racial differences in cancer mortality exist in South Carolina30; however, this possibility could not be examined, as it is not a SEER registry. Moreover, the significant disproportionate representation of white men in several registries poses a challenge, and for that reason, we adopted a methodological approach based on interaction term analysis rather than a separate registry-by-registry analysis. The SEER registries also lack data on patient comorbidities as well as quality of care processes (eg, caseload volume, provision of guideline-directed therapy), which can influence long-term mortality outcomes.

hodological approach based on interaction term analysis rather than a separate registry-by-registry analysis. The SEER registries also lack data on patient comorbidities as well as quality of care processes (eg, caseload volume, provision of guideline-directed therapy), which can influence long-term mortality outcomes. Conclusions After adjusting for patient, disease, and treatment characteristics, this cohort study found that population-level differences in prostate cancer survival among black and white men in the US were associated with a small set of geographic areas and with low-risk prostate cancer. While the cause of racial disparities in prostate cancer survival remains a topic of ongoing study, future studies and interventions should be targeted at settings where racial disparities are most pronounced. Supplement. eTable. Patient Characteristics by Registry Click here for additional data file.