Predicting Pathological Complete Response in HER2+ Breast Cancer: An AI-Driven Model Using Standard Clinical Practice Parameters

Abstract IntroductionIn breast carcinoma (BC), particularly the HER2 + subtype, achieving a complete pathological response (pCR) after neoadjuvant therapy (NAT) is vital for prognosis and disease-free survival. While dual HER2 blockade with trastuzumab and pertuzumab is standard for HER2 + BC, variability in pCR rates necessitates refined treatment strategies. Predictive models for pCR can optimize patient management, reduce unnecessary toxicity, and identify candidates for alternative therapies, enhancing personalized cancer care.Aim: to develop a predictive model for pCR in HER2 + BC.MethodsData was sourced from an international, multicenter, retrospective study focused on patients with early HER2-positive breast cancer from January 2017 to December 2023. Eligible patients for this study were those who received NAT that included trastuzumab and pertuzumab, followed by surgical intervention. Logistic regression analyses and machine learning techniques assessed associations with pCR, while receiver operating characteristic (ROC) curves measured the performance of the predictive model.ResultsA total of 545 patients with early HER2 + BC were included in the dataset. The predictive model demonstrated a ROC AUC of 0.63 to 0.66, with an accuracy between 62.4% and 65.9%. Precision ranged from 0.66 to 0.69, indicating a strong ability to avoid false positives, while recall was between 0.74 and 0.79, effectively capturing true positives. The F1 score ranged from 0.70 to 0.73, showing a good balance between precision and recall.ConclusionThis work shows that machine learning can create a predictive model integrating clinical and pathological parameters to forecast pCR in patients treated with trastuzumab, pertuzumab, and chemotherapy. While promising, the model requires validation in a larger patient cohort to confirm its robustness.