Document resource
Background Cardiotoxicity is a significant concern in breast cancer patients receiving anthracycline or HER2-targeted therapies. Recent ESC 2022 guidelines recommend systematic cardiac biomarker assessment, yet real-world implementation remains unclear. Machine learning (ML) may support early risk stratification to identify patients at higher risk of cancer therapy-related cardiac dysfunction (CTRCD).Objectives This study aimed to (1) evaluate the use of cardiac biomarkers in routine cardio-oncology services following the ESC 2022 guidelines and (2) explore the predictive performance of ML in classifying CTRCD risk among HER2+ breast cancer patients.Methods A retrospective service evaluation included 203 female patients treated at the Royal Devon University Hospital between April 2022 and April 2024. Biomarker measurement frequency, timing relative to echocardiography, and variation by treatment type were assessed. Separately, an ML sub-study utilised an open-source dataset of 431 HER2+ patients (10% CTRCD incidence) to develop a Logistic Regression model with engineered interaction features. Model performance was evaluated using sensitivity, specificity, precision, F1 score, and ROC-AUC, with SHAP analyses to interpret feature importance.Results Biomarker measurement was infrequent; 93.6% of patients had no biomarkers recorded. Of the 6.4% measured, Troponin was most common, typically during therapy. Logistic Regression achieved high recall (0.83) on the validation set, identifying 5 of 6 CTRCD events, with moderate specificity (0.68) and low precision (0.21). Risk stratification by ML aligned with HFA-ICOS scoring in 70% of cases. Age, heart rate, and LVEF were the strongest predictors of CTRCD.Conclusions Cardiac biomarkers remain underutilised in real-world cardio-oncology practice despite guideline recommendations. ML approaches demonstrate potential to complement traditional risk stratification, offering clinically meaningful predictions and supporting early identification of patients at risk of cardiotoxicity. Future work should integrate biomarker data with AI models to enhance proactive cardio-oncology care.