ABSTRACT Large real‐world breast cancer datasets with molecular profiling are increasingly used to estimate the benefit of adjuvant chemotherapy in hormone receptor‐positive, HER2‐negative (HR+/HER2−) disease and are often interpreted alongside randomized trial data. Since chemotherapy usage in routine practice is strongly influenced by baseline clinical risk and tumor biology, observed survival associations may reflect treatment selection rather than treatment effect. In this study, we analyzed the population‐based RNA‐sequencing dataset GSE96058, including 3409 primary breast cancers with long‐term follow‐up. Among 2536 patients with HR+/HER2− disease treated with adjuvant endocrine therapy, we evaluated overall survival by chemotherapy exposure within PAM50‐defined Luminal A and Luminal B subtypes. In unadjusted analyses, chemotherapy was associated with improved overall survival, particularly in Luminal B tumors (hazard ratio (HR) 0.29 and 95% CI 0.16–0.53). After inverse probability of treatment weighting with trimming, this association remained large (HR 0.25 and 95% CI 0.12–0.55). However, after restricting analyses to patients with comparable likelihood of receiving chemotherapy and applying overlap‐restricted doubly robust adjustment, chemotherapy was no longer significantly associated with overall survival in either subtype. In the overlap‐restricted cohort ( n = 1194; 80 deaths), adjusted HRs were 1.06 (95% CI 0.49–2.28) for Luminal A and 0.60 (95% CI 0.26–1.39) for Luminal B disease. These findings highlight the limitations of real‐world molecular cohorts for estimating chemotherapy benefit without careful attention to treatment selection.
Shrestha et al. (Wed,) studied this question.