Abstract / Summary
Purpose: Adjuvant endocrine therapy (AET) substantially reduces recurrence and mortality in hormone receptor-positive breast cancer but requires sustained daily adherence over 5 to 10 years. Approximately one-third of patients fall short of recommended adherence in the first year alone, largely due to distinct combinations of attitudes, barriers, and circumstances. Existing studies have catalogued individual risk factors but lack the scale and breadth to characterize how these factors co-occur within patients, or to distinguish behavioral drivers from confounding by clinical and demographic context. We sought to characterize the behavioral and social heterogeneity underlying AET adherence in a national real-world cohort. Moving beyond population-average risk factors, we identify the distinct patient profiles, and the differing drivers within them, that any effective adherence strategy must address. Methods: We conducted a retrospective cohort study of US women with invasive breast cancer diagnosed between 2016 and 2025, linking two large-scale, individual-level datasets through privacy-preserving tokenization: Surgo Health's BehavioralPulse, which provides modeled individual-level behavioral and attitudinal risk scores together with consumer sociodemographic attributes, and longitudinal medical and pharmacy claims from a claims data provider. Eligible patients underwent 1 to 2 breast surgeries, initiated oral AET (tamoxifen or aromatase inhibitors), and maintained continuous insurance coverage for 365 days following therapy initiation. The primary outcome was adherence, defined as medication possession ratio (MPR) [≥]80% in the first year. Mixed-effects logistic regression with a random intercept for ZIP3 estimated adjusted associations across behavioral, sociodemographic, and clinical predictors. To characterize how behavioral factors co-occur within patients, we identified the most prevalent configurations of the statistically significant behavioral predictors and estimated their relative association with adherence, holding clinical and demographic factors constant. Results: The final cohort included 401,450 women, of whom 280,595 (69.9%) achieved MPR [≥]80%. Several behavioral factors were independently associated with adherence after adjustment for clinical and demographic covariates, including comfort following medication instructions (aOR, 1.15; 95% CI, 1.06-1.24), geographic proximity to breast oncologists (aOR, 1.17; 95% CI, 1.04-1.32), tangible instrumental social support (aOR, 1.06; 95% CI, 1.00-1.13), religiosity (aOR, 1.04; 95% CI, 1.01-1.08), concern about sexual side effects (aOR, 0.96; 95% CI, 0.93-0.99), and cost-related access barriers (aOR, 0.97; 95% CI, 0.95-1.00). The 10 most common configurations of significant behavioral predictors accounted for over 70% of the cohort, with the two most prevalent representing more than 40% of patients. The most common profile, defined by an absence of behavioral barriers and the presence of social support, was associated with a positive behavioral contribution to adherence propensity (behavioral linear predictor OR = 1.18; 95% CI: 1.04-1.36) comparable in magnitude to several established clinical predictors. Compared against this referent profile, six of the nine remaining profiles had lower adherence, with relative odds ranging from approximately 0.92 (95% CI: 0.89-0.95) to 0.97 (95% CI: 0.94-0.99). One profile, similar to the reference but including high trust in doctors, was associated with higher adherence odds (1.04, 95% CI: 1.01-1.07). These profiles arose from substantively different underlying combinations of factors: segments dominated by cost barriers, by side-effect concerns, or by limited social support produced comparable overall adherence risk but through distinct pathways. Conclusion: In this national cohort, nearly one-third of women did not achieve recommended first-year adherence to AET. The pathways to non-adherence were heterogeneous, structured into recurring behavioral profiles rather than randomly distributed across patients. This heterogeneity is clinically meaningful: patients with similar adherence risk may benefit from substantially different forms of support, ranging from financial navigation to side-effect management to social support resources. Surfacing this structure required linking individual-level behavioral data to large-scale claims data, offering a practical foundation for optimal design of patient-centered adherence interventions that are tailored to the specific configurations of barriers patients actually face.