252 SHORT-TERM ANTI-REMODELING EFFECTS OF GLIFLOZINS IN DIABETIC PATIENTS WITH REDUCED EJECTION FRACTION: AN EXPLAINABLE ARTIFICIAL INTELLIGENCE APPROACH.

Abstract

Abstract Background Sodium glucose cotransporter type 2 inhibitors (SGLT2i), also called gliflozins, are playing an emerging role for the treatment of heart failure with reduced left ventricle ejection fraction (HFrEF). However, the direct effects of SGLT2i on left and right ventricular remodeling and function have not been completely clarified. We therefore aimed to assess clinical response to gliflozins focusing on echocardiographic evaluation and identify any predictive factors with a machine learning approach. Methods Based on Random Forest, a robust and consolidated machine learning approach, we carried out a single subject analysis to evaluate to which extent patients treated with gliflozins can effectively be distinguished from patients undergoing non-gliflozins treatments. Besides, we carried out an eXplainability analysis using Shapley values to outline the clinical parameters which mostly took advantage by gliflozins. Finally, machine learning experiments were designed to highlight the presence of specific clinical patterns undermining the gliflozins effectiveness. Results 5-fold cross-validation analyses showed that gliflozin treatment was identified with a 0.70 ± 0.03% accuracy; the most important parameters supporting such accuracy were Right Ventricle S’ Velocity (RV S’), Left Ventricle End Systolic Diameter (LVESD) and E/e’ ratio. Low Tricuspid Annular Plane Systolic Excursion (TAPSE) values along with high LVESD and End Diastolic Volume (EDV) values are likely to impair the gliflozin effectiveness. Conclusions Treatment with gliflozins resulted in an improvement of several echocardiographic parameters related to biventricular function and left ventricle remodeling. Several clinical parameters, as simple echocardiographic parameters, may accurately predict the cardiovascular response to gliflozins treatment.