For the roughly 15% of breast cancer patients diagnosed with triple-negative disease — the subtype with the fewest targeted treatment options and highest recurrence risk — knowing who truly needs chemotherapy has long been a clinical judgment call plagued by over- and under-treatment. Incorporating a validated immune biomarker into an established prediction tool may meaningfully change that calculus.

The updated model, designated PREDICT_sTILs, integrates stromal tumor-infiltrating lymphocyte (sTIL) scores into the widely used PREDICT v2.3 prognostic framework using a Cox regression approach applied to a pooled cohort of 3,698 women with early-stage triple-negative breast cancer (TNBC), diagnosed across nearly four decades (1979–2017). The dataset was divided between 1,806 chemotherapy-treated and 1,892 chemotherapy-naïve patients, with 860 breast-cancer-specific deaths recorded within 10 years. sTIL levels — scored per international standardized guidelines — were incorporated alongside the existing PREDICT prognostic index to model breast-cancer-specific survival. Internal-external validation was performed using leave-one-region-out cross-validation, with discrimination assessed by area under the curve and calibration by observed-to-expected mortality ratios. Decision curve analysis then benchmarked the model's net clinical benefit across 10-year mortality thresholds of 8%–15%.

This work represents a clinically meaningful step forward rather than a paradigm shift, but its implications are consequential. sTILs have accumulated strong prognostic evidence in TNBC over the past decade — high immune infiltration consistently correlates with better survival — yet biomarkers rarely migrate into routine decision-support tools at scale. Embedding sTILs into PREDICT bridges that gap. The key limitation is the cohort's historical breadth: diagnoses dating to 1979 predate modern chemotherapy regimens and immunotherapy, raising questions about transportability to contemporary practice. Additionally, as an observational, non-randomized dataset, causal inference about chemotherapy benefit remains constrained. The model performs best as a stratification aid, potentially sparing low-risk, high-sTIL patients from unnecessary cytotoxic therapy — but prospective validation in modern cohorts will be essential before wide clinical adoption.