Abstract 16812: Value of Echocardiographic Variables to Predict Cardiotoxicity in Patients Using Machine Learning

Introduction: Early identification of subclinical signs of cardiotoxicity in patients with breast cancer submitted to echocardiographic exam may change the natural history of patients undergoing treatment with anthracyclines. Objectives: The aim of this study was to identify echocardiographic and clinical variables related to cardiotoxicity and death analyzed using machine learning.

Methods: The study consisted of a historical cohort of patients with breast cancer, submitted to chemotherapy treatment with anthracyclines, with clinical evaluation, electrocardiogram and echocardiogram performed before the first session,3 ,6 , 12 and 24 months later, with the intention to analyze echocardiographic markers to identify prognostic parameters related to cardiotoxicity.. Among various classifier models, the Extreme Gradient Booster (XGB) model was selected as the best instrument for predicting cardiotoxicity in this study. The SHAP (SHapley Additive exPlanations) approach was employed to assess the importance of each variable in the model.

Results: The sample consisted of 58 patients, with a mean age of 50.29 (standard deviation = 13.32), who completed treatment between February 2010 and June 2011. The outcome measure was a composite of death or cardiotoxicity, with 11 patients in the positive group (mean age: 43.73, std: 13.08) and 47 in the negative group (mean age: 51.83, std: 13.05) The XGB model achieved an accuracy of 0.89, recall and precision of 0.67. The Area Under the Receiver Operating Characteristic Curve (ROC AUC) was 0.91 . Notably, echocardiographic variables such as Myocardial Performance Index (MPI), End Systolic Volume (ESV), End Diastolic Volume (EDV) and Systolic Blood Pressure (SBP) emerged as relevant predictors of the outcome figure 1.

Conclusions: Echocardiographic indices may predict subclinical cardiotoxicity and which and may be confirmed in larger exploratory studies.