Association between area deprivation index (ADI) and treatment persistence (TP) with CDK4/6 inhibitors (CDK4/6i)in HR+/HER2- breast cancer

Background: CDK4/6i improve invasive disease-free and progression-free survival in HR-positive/HER2-negative breast cancer, yet real-world treatment persistence is suboptimal. ADI is a validated composite of neighborhood socioeconomic disadvantage that has been linked to worse medication adherence and health outcomes. Evaluating the association between ADI and persistence with CDK4/6i may identify equity-sensitive targets for intervention. The primary objective was to assess the association between ADI and TP in early (EBC) and metastatic (MBC) breast cancer. Methods: We conducted a retrospective cohort study of patients prescribed CDK4/6i at Yale New Haven Health from February 2018 through December 2024. Artificial intelligence and natural language processing were used to extract structured electronic health record data. TP was defined using Epic Beacon treatment plan start and end dates and analyzed separately for EBC and MBC cohorts. Analyses used Python and R. Residential addresses at initiation were batch processed in R (ezADI), geocoded to 12-digit census block group GEOIDs via the U.S. Census Geocoder, and linked to 2020 Neighborhood Atlas ADI percentiles. Records with suppressed or missing ADI were excluded. ADI was dichotomized at the cohort median (above-median = more deprived; below-median = less deprived). Cox proportional hazards models, adjusting for medication, age, and gender were used to estimate hazard ratios (HR) for therapy discontinuation. Kaplan-Meier analyses summarized TP by ADI and CDK4/6i. Results: A total of 1,363 patients were included in the MBC cohort and 298 in the EBC cohort. The median ADI across both cohorts was 30, which was used as the deprivation cutoff. In MBC, median ADI values were 19 (IQR 11-25) for the less-deprived groups and 47 (IQR 39-62) for the more-deprived groups across agents (abemaciclib, palbociclib, ribociclib). In multivariable Cox models adjusting for medication, age, and gender, higher deprivation was not significantly associated with earlier therapy discontinuation (HR = 0.97; 95 % CI, 0.86-1.08; p = 0.56). Older age was associated with shorter TP (HR = 1.01 per year; 95% CI, 1.00-1.01; p < 0.001). In EBC, median ADI values were 19 (IQR 14-25) versus 47 (IQR 39-63) among abemaciclib-treated patients, and median ADI was 16 (IQR 10-22) versus 50 (IQR 41-59) among ribociclib-treated patients. ADI was not associated with increased discontinuation (HR = 0.98; 95 % CI, 0.78-1.22; p = 0.83). Conclusions: In this large, real-world cohort, TP with CDK4/6i did not differ by neighborhood socioeconomic deprivation. Further work incorporating individual-level social determinants of health may better elucidate drivers of adherence and persistence disparities. Acknowledgment: ChatGPT (OpenAI) was used to assist with text revision and editing for clarity.