Abstract C022: Integration of clinical models based on machine learning to predict pathological response to neoadjuvant chemotherapy in Colombian women with triple-negative breast cancer

Abstract Introduction: Triple-negative breast cancer (TNBC) is a highly heterogeneous subtype with a high recurrence rate. Pathological complete response (pCR) is a strong predictor of disease prognosis. Clinical and pathological features can influence the response to neoadjuvant chemotherapy (NAC), making them valuable for treatment decision-making. Objective: To develop a predictive model based on clinical and pathological features to estimate the pathological response in Colombian patients with TNBC. Methods: 204 TNBC patients treated with NAC at three Colombian institutions were included. Pathological response was assessed using the Residual Cancer Burden classification: pCR (n=50, 24.5%) and residual disease (RD) (n=154, 75.5%). Data were preprocessed and transformed for the selection of the most relevant predictive variables. The dataset was split into training (80%, n=163) and testing sets (20%, n=41). Five classification models were evaluated using cross-validation: logistic regression (LR), support vector machine (SVM), random forest (RF), XGBoost, and Naive Bayes (NB). The best-performing models were selected for integration using a voting classifier. Results: The analysis identified six clinical and pathological variables with the highest predictive stability for pCR: clinical stage, axillary lymph node involvement, genetic testing criteria, tumor size, age, and weight at diagnosis. Individual models were optimized through hyperparameter tuning and evaluated on the test set using the area under the ROC curve (AUC): LR (0.75), SVM (0.71), RF (0.81), XGBoost (0.83), and NB (0.82). XGBoost and NB were selected to build an integrated model through soft voting classification, achieving an AUC of 0.84. This improved the identification of patients who may benefit from alternative therapeutic strategies. Conclusions: The results suggest that clinical models can enhance the prediction of pathological response to NAC in TNBC patients. The proposed integrated model may serve as the basis for developing a useful clinical decision-making tool. Citation Format: Carlos Alexander. Huertas-Caro, Maria Jose. Peña, Angela Maria. Barrera, Andrea Marcela. Zuluaga, Diego Felipe. Ballen, Juan Carlos. Mejía, Luz Fernanda. Sua-Villegas, Alicia Cock-Rada, Silvia J. Serrano-Gomez. Integration of clinical models based on machine learning to predict pathological response to neoadjuvant chemotherapy in Colombian women with triple-negative breast cancer [abstract]. In: Proceedings of the 18th AACR Conference on the Science of Cancer Health Disparities; 2025 Sep 18-21; Baltimore, MD. Philadelphia (PA): AACR; Cancer Epidemiol Biomarkers Prev 2025;34(9 Suppl):Abstract nr C022.