Including tumor-infiltrating lymphocytes into the PREDICT prognostic model for triple-negative breast cancer survival

Background: Stromal tumor infiltrating lymphocytes (sTILs) are a strong prognostic marker in triple-negative breast cancer (TNBC) patients. We aimed to incorporate sTILs into the PREDICT model for women with early-stage TNBC to guide chemotherapy decisions. Patients and methods: We included 3698 women with early-stage TNBC, diagnosed between 1979 and 2017, from two pooled cohorts with sTILs scored according to international guidelines to update PREDICT version 2.3. The updated model, PREDICT_sTILs, is a Cox regression model with sTILs and the PREDICT prognostic index as predictors, and breast cancer-specific survival as the outcome. Internal-external validation was conducted using leave-one-region-out cross-validation, with calibration assessed by the observed-to-expected (O/E) ratio and discrimination by area under the curve (AUC). We compared the clinical values of PREDICT_sTILs and PREDICT through decision curve analysis, examining net true high-risk and low-risk classifications across risk thresholds of 10-year breast cancer mortality above 8-15%. Results: Among the 3698 patients, 1806 received chemotherapy while 1892 were chemotherapy-naïve. Chemotherapy-treated patients had a median age of 50 years, tumor size of 25mm, and 1 positive lymph node (pN). The chemotherapy-naïve patients had a median age of 55 years, tumor size of 20mm, and 0 pN. Within 5 and 10 years, 741 and 860 deaths were attributed to breast cancer, respectively. PREDICT_sTILs showed strong internal-external validity, with comparable calibration and discrimination at both time points: pooled O/E ratios of 0.98 (95% CI: 0.69-1.41) at 5 years and 0.99 (95%CI: 0.72-1.35) at 10 years, and AUCs of 0.74 (95%CI: 0.72-0.77) and 0.74 (95%CI: 0.70-0.78), respectively. Compared to PREDICT, PREDICT_sTILs identified 19-60 additional net true low-risk patients, and 3-10 additional net true high-risk patients per 1000 chemotherapy-naïve patients at the risk thresholds of 8-15% at 10 years. Conclusion: Adding sTILs to PREDICT improves its ability to inform chemotherapy decisions for early-stage TNBC patients, especially in identifying low-risk individuals who may safely forgo chemotherapy.