Confounding by indication in the evaluation of neoadjuvant chemotherapy: a sequential Cox analysis of an eight-year single-centre breast cancer cohort.

Purpose

Neoadjuvant chemotherapy (NACT) is reserved for tumours with a worse prognosis, introducing confounding by indication. Using sequentially adjusted Cox models to display confounding step by step, we assessed prognostic factors and the role of neoadjuvant therapy in a single-centre breast cancer cohort.

Methods

Retrospective study of 461 women treated at a single Spanish hospital (2016-2018) and followed to March 2026 (median 7.3 years; 456 in survival analyses). Overall (OS) and disease-free survival (DFS) were estimated by the Kaplan-Meier and Cox regression methods; proportional hazards were checked with Schoenfeld residuals and missing data handled by multiple imputation.

Results

Five-year OS and DFS were 89.7% and 86.0%. Independent prognostic factors were stage III-IV (HR 3.9), triple-negative subtype (HR 2.4) and adjuvant radiotherapy (HR 0.32). The apparent adverse effect of neoadjuvant therapy (univariable HR 2.35) disappeared after stage adjustment (HR 1.6; p = 0.17), with the same pattern for DFS; treated women differed chiefly in stage III-IV (60% vs. 14%; standardised mean difference 1.09). Recovering missing histological grade from source reports abolished its apparent prognostic effect (HR 1.40, 0.82-2.39). The triple-negative excess risk was concentrated in the first three years (HR 4.8 vs 0.4; interaction p = 0.020).

Conclusions

After adjustment for stage and tumour burden, the apparent adverse effect of neoadjuvant therapy is largely explained by confounding by indication rather than treatment although the small treated subgroup cannot exclude a modest residual effect. Sequentially adjusted Cox models display the confounding pathway explicitly, complementing propensity-score approaches.