Exploring heterogeneity in survival benefit of neoadjuvant therapy for triple-negative breast cancer using causal survival forests.

INTRODUCTION: The individual survival benefit of neoadjuvant therapy (NAT) for patients with triple-negative breast cancer (TNBC) remains uncertain. This study aimed to evaluate the individualized treatment effect (ITE) of NAT on event-free survival (EFS) in patients with TNBC using causal survival forests. METHODS: This study included 803 patients with TNBC from a retrospective clinical cohort at a tertiary hospital in Henan Province, China. The outcome was 3-year EFS. ITEs were estimated on the restricted mean survival time (RMST) scale. A positive ITE indicated predicted benefit from NAT. Model performance was evaluated using C-for-benefit, rank weighted average treatment effects, and calibration. Patients were grouped into ITE quartiles, and survival outcomes were compared according to predicted benefit group and treatment concordance. RESULTS: Among 803 patients, 250 received NAT. The average treatment effect was -1.62 months. The predicted ITEs ranged from -7.92 months favoring No NAT to 3.67 months favoring NAT, indicating substantial heterogeneity in treatment benefit. The model showed positive ranking performance with an AUTOC of 5.67 months and the C-for-benefit was 0.537. Tumor size, clinical axillary lymph node status, and clinical N stage were the main contributors to predicted ITEs. In the Q1 and Q4 groups, concordant treatment was associated with an average RMST gain of 5.23 months. CONCLUSIONS: Among patients with TNBC, CSF identified subgroups with different predicted directions and magnitudes of 3-year EFS benefit. Stratification based on ITEs may support individualized assessment of treatment benefit and provide a basis for future validation.