Abstract
Background: Racial disparities in breast cancer outcomes appear to be widening in the US despite multiple advancements in breast cancer (BC) management. Although BC incidence rates are similar between European-American (EA) and African American (AA) women in the US, striking differences exist in age-adjusted mortality rates. Moreover, AAs only represent 5-10% of BC clinical trials is the US, and there is a dearth of studies that stratify outcomes by race. Unfortunately, there is no reliable way to identify AA BC patients at high risk of poor outcomes, which would require more aggressive treatment. In this study, we aimed to develop a prognostic model for AA women that will allow for a deeper segmentation and an accurate patient stratification into high and low risk subgroups to aid clinical decision making.
Methods: We obtained protein data from The Cancer Proteome Atlas (TCPA) dataset, which consists of reverse phase protein array (RPPA) expression levels of 224 proteins measured from BC TCGA cohort. Protein expression information were available for a total of 754 BC patients (134 AA and 620 EA). Sequential forward selection alongside cross-validation was used to select proteins, fit into a random forest model, based on their combined accuracy in predicting patient prognosis. Models were evaluated on the combined cross validated test sets either univariately through the Kaplan-Meier log-rank test or via multivariate analysis after adjusting for stage, age, and positive lymph nodes through Cox regression model.
Results: The selection process resulted in combination of four proteins that optimized prognostic prediction: bcl2- like protein (BAX), Inositol polyphosphate-4-phosphatase, type II (INPP4B), X-ray repair cross-complementing protein 1 (XRCC1) and Cleaved Poly (ADP-ribose) polymerase (c-PARP)). Alone, these proteins did not have significant prognostic value in AA BC patients [BAX (HR=0.676, p=0.6276), INPP4B (HR=0.9935, p=0.9772), XRCC1(HR=0.2613, p=0.1455), c-PARP (HR=0.6375, p=0.6186)]; within random forest, these variables were able to stratify high-risk group patients with 86% accuracy and Hazard Ratio (HR) of 5 (p <0.001). The model retained its significant prognostic ability (HR=10.741, p=0.0006) when controlling for clinicopathologic variables like stage, age, and positive lymph nodes. Finally, we retrained our model to risk stratify BC patients in the EA cohort; the magnitude of risk stratification was very low (HR=1.33) compared to AA cohort, confirming its prognostic role specifically in AA BC patients.
Conclusions: Based on statistical modeling and ML-based approaches, our data show that assessment of expression of a quartet of biomarkers—BAX, XRCC1, INPP4B and c-PARP—can be used to robustly stratify AA BC patients into high- and low-risk categories. We believe our model plays an important prognostic role in AA BC patients and could inform clinicians to prioritize AA patients for appropriate clinical trials and also help patients make decisions about enrolling in such trials.
Citation Format: Shristi Bhattarai, Sergey Klimov, Ritu Aneja. A prognostic risk model for African American women with breast cancer: Implications for meaningful accrual in clinical trials [abstract]. In: Proceedings of the Eleventh AACR Conference on the Science of Cancer Health Disparities in Racial/Ethnic Minorities and the Medically Underserved; 2018 Nov 2-5; New Orleans, LA. Philadelphia (PA): AACR; Cancer Epidemiol Biomarkers Prev 2020;29(6 Suppl):Abstract nr C101.