Abstract Purpose: According to population-based studies, prostate cancer (PC) is a more common, more aggressive, and more often fatal disease in African American (AA) men compared to American men of European decent (W). Factors hypothesized to contribute to these disparities include differences in cancer screening and treatment, risk factors such as obesity, and PC risk gene variants. Radical prostatectomy (RP) datasets have been used to examine whether racial disparities in PC outcomes might stem from differences in the biology of the tumors themselves. We sought to test whether selection bias (stemming from imbalances among factors determining who receives RP vs. other non-surgical treatment) could alter the associations between patient race and adverse pathological findings in RP specimens. We present an application of marginal structural modeling using inverse probability weights (IPW) as a means of identifying and counteracting potential treatment selection biases in PC. Materials and Methods: Among 259 AA and 343 W men with prostate cancer recruited at Henry Ford Health System from July 2001 to January 2005, we tested associations between clinical and demographic characteristics with race and receipt of RP. We used logistic regression to estimate the odds of adverse clinical findings (diagnostic PSA>10ng/ml, biopsy Gleason score≥7, or intermediate/high clinical risk) in AA vs. W cases at the time of biopsy (n=602), and compared these estimates to those obtained from cases treated with RP (n=395). We modeled patient factors associated with receipt of RP, and computed an IPW for each case. We report weighted and un-weighted, multivariable-adjusted generalized estimating equation (GEE) models examining the associations between race and adverse pathological findings in RP specimens. Results: At the time of biopsy, compared to W, AA men were 1.66(95%CI 1.09-2.52) times more likely to have intermediate/high clinical risk disease (diagnostic PSA>10ng/ml, biopsy Gleason score ≥7, or clinical stage ≥T2b). AA patients were younger at diagnosis (61.9 vs. 62.9 years), had a greater degree of co-morbidity (p<0.001), lower educational attainment (p<0.001), and lower median household income level (p<0.001), compared to W. When modeling potential treatment selection factors among 602 biopsy confirmed cases, AA men with low clinical risk (biopsy Gleason≤6, PSA<10 and clinical stage ≤T2a) were 2.73(95%CI 1.23-6.08) times as likely to receive RP as W, adjusting for age, family history, co-morbidities, educational attainment, and census tract income level, whereas no such difference existed between AA and W with higher clinical risk (OR=0.918, 95%CI 0.48-1.76)(p<0.05 for interaction term). When testing for racial disparities in tumor characteristics among the 395 cases who received RP, in conventional multivariate models (without IPW), there were no statistically significant differences by race for any adverse clinical or pathological features. However, after weighting each of the 395 cases by their inverse probability of receiving RP, AA men had greater odds of intermediate/high clinical risk (OR=1.62, 95%CI 1.06-2.50), and pathologic Gleason grade (OR=1.86, 95%CI 1.16-2.99). Conclusions: In the current dataset, treatment selection bias nullified the associations between race and adverse clinical and pathological characteristics, in multivariate adjusted models. After applying IPW, our data support an association between AA race and adverse clinical and pathological findings among men treated with RP. Our results suggest that, if unaddressed, treatment selection bias may obscure estimated disparities between AA and W PCa patients, and that the use of marginal structural modeling may benefit cancer risk factor analyses utilizing treatment series datasets. Citation Format: Russell B. McBride, Carlos Cordon-Cardo, Benjamin A. Rybicki. Race and aggressive prostate cancer: Resolving bias introduced by treatment selection. [abstract]. In: Proceedings of the Fifth AACR Conference on the Science of Cancer Health Disparities in Racial/Ethnic Minorities and the Medically Underserved; 2012 Oct 27-30; San Diego, CA. Philadelphia (PA): AACR; Cancer Epidemiol Biomarkers Prev 2012;21(10 Suppl):Abstract nr A08.
Read full abstract