Cancer treatments have improved substantially over the past decades, but some effective therapies such as anthracyclines and HER2-targeted agents are associated with severe cardiovascular adverse effects, including heart failure. Existing cardiovascular risk prediction scores have limited evidence in this setting. The ML-CardioTox study is a prospective, multicenter, observational cohort conducted in 15 centers in France. The primary objective is to develop a one-year prediction score for cancer therapy-related cardiotoxicity using machine learning methods. A dedicated software platform will be used to standardize data collection and support integration of artificial intelligence tools. A total of 600 patients treated with anthracyclines or HER2-targeted therapies in cardio-oncology clinics will be enrolled over a one-year inclusion period starting in December 2024, with a 12-month follow-up. The primary endpoint is the occurrence of cardiotoxicity as defined by the 2022 European Society of Cardiology guidelines (hospitalization for heart failure, initiation or escalation of diuretic therapy, decline in cardiac function on imaging, or increase in cardiac biomarkers such as troponin or natriuretic peptides). Secondary objectives include comparison of the predictive performance of the machine learning-derived score with the established HFA-ICOS risk score. Patients will be managed according to routine clinical practice. This study aims to improve prognostic stratification tools for patients receiving anthracyclines or HER2-targeted therapies, with the goal of better identifying those at high risk of developing cardiotoxicity during follow-up.
Study Type
OBSERVATIONAL
Enrollment
600
Incidence of cardiotoxicity at 12 months
Cardiotoxicity defined as a composite outcome, according to the 2022 ESC guidelines: hospitalization for heart failure OR initiation or escalation of diuretic therapy for signs or symptoms of heart failure OR decrease in left ventricular ejection fraction (LVEF) by \>10 percentage points from baseline or to a value \<40% OR relative decrease in global longitudinal strain (GLS) ≥15% from baseline OR new increase in cardiac biomarkers (troponin or NT-proBNP/BNP). Events will be adjudicated by an independent committee.
Time frame: 12 months after initiation of anthracycline or HER2-targeted therapy.
Predictive performance of machine learning-derived cardiotoxicity risk score
Performance assessed by AUC-ROC, AUC-PR, balanced accuracy, F1 score, PPV, NPV, McNemar's test, and Brier score. Unit of Measure: AUC values, accuracy (%), predictive values (%).
Time frame: 12 months after initiation of anthracycline or HER2-targeted therapy.
Comparison of predictive performance with HFA-ICOS risk score and traditional Cox model
Direct comparison of ML-derived risk score vs. HFA-ICOS and Cox model. Unit of Measure: Difference in AUC values (%), statistical comparison (DeLong test).
Time frame: 12 months after initiation of anthracycline or HER2-targeted therapy.
Correlation between baseline prognostic factors and cardiotoxicity at 12 months
Evaluation of demographic, clinical, biological (including biomarkers), and echocardiographic factors for correlation with cardiotoxicity occurrence. Unit of Measure: Odds ratios (OR) or hazard ratios (HR) with 95% confidence intervals.
Time frame: 12 months after initiation of anthracycline or HER2-targeted therapy.
This platform is for informational purposes only and does not constitute medical advice. Always consult a qualified healthcare professional.