Abstract B01: Testing a multilevel risk prediction model in high risk men enrolled in a prostate cancer early detection program

Abstract

Background: Health disparities play a major role in prostate cancer (PCa). African American (AA) compared to European American (EA) men are twice as likely to die of and be diagnosed with PCa. Multilevel factors from societal/neighborhood exposures down to genetics likely contribute to racial disparities, but few PCa risk prediction models include multilevel factors and consider race/ethnic differences. Objective: We sought to: 1) develop a multilevel risk prediction model for time to PCa diagnosis, that includes neighborhood variables, individual-level socioeconomic and clinical factors (education, race, digital rectal exam or DRE), and biologic variables (prostate specific antigen or PSA level, and percent West African genetic ancestry) in men at high risk for prostate cancer (defined as AA men and/or men with a PCa family history); 2) compare our multilevel model to a more standard prediction model that includes only age, race, PSA, and DRE (abnormal/normal). Methods: A total of 443 high risk, cancer-free men between 35 and 69 years of age with complete socioeconomic, racial, and genetic ancestry data were identified from the Prostate Risk Assessment Program (PRAP) at Fox Chase Cancer Center. Their data were geocoded and linked to 17 neighborhood variables at the census tract level (from the Year 2000 U.S. Census) that were previously associated with advanced PCa in EA men in a novel neighborhood-wide association study(NWAS) our study team developed. These variables generally represent neighborhood transportation, poverty, income, social support, immigration, renting/owning a house, and employment. Men were followed from time of program (PRAP) entry to PCa diagnosis or censoring, with annual follow-up visits that included PSA and DRE screening. Men with elevated PSA or other indications for PCa were referred to Urology for evaluation and potential biopsy according to PRAP protocols. Univariate analyses of neighborhood variables, and the interaction of each variable with PSA and race, were evaluated in Cox regression models, using robust standard errors to adjust for clustering by census tract, in order to inform the final multivariate, multilevel model. Harrell9s C Index (C Statistic) was used to compare the multilevel risk prediction model to a standard prediction model. Results: With a median follow-up time of 71 months, PCa diagnosis occurred in 69 participants. The final multilevel risk prediction model included 3 neighborhood variables related to transportation, social support, and poverty, along with education, age, race, baseline PSA, baseline DRE, and PCa family history. Significant interactions between the top hit from the NWAS and PSA were noted in the full study population (neighborhood mode of transportation to work X PSA, p-value Conclusion: This study is the first to investigate the role of neighborhood in PCa risk prediction. While risk prediction models show little change, significant neighborhood effects in multilevel models warrant additional study and could inform future health disparity studies. Citation Format: Shannon M. Lynch, Elizabeth Handorf, Elizabeth Blackman, Lisa Bealin, Shiju Daniel, Veda N. Giri, Elias Obeid, Camille Ragin, Mary B. Daly. Testing a multilevel risk prediction model in high risk men enrolled in a prostate cancer early detection program. [abstract]. In: Proceedings of the AACR Special Conference: Improving Cancer Risk Prediction for Prevention and Early Detection; Nov 16-19, 2016; Orlando, FL. Philadelphia (PA): AACR; Cancer Epidemiol Biomarkers Prev 2017;26(5 Suppl):Abstract nr B01.