Risk prediction in asymptomatic left ventricular dysfunction using machine learning

Abstract Background Asymptomatic left ventricular systolic dysfunction (ALVSD) is categorized as stage B heart failure and when present it indicates a worse prognosis compared to individuals with normal left ventricular contraction. Consequently, prognostic prediction for patients with ALVSD is crucial; however, no models specifically tailored to risk prediction in patients with ALVSD are currently available. Therefore, we developed a machine learning (ML)-based model to predict 3-year mortality in patients with ALVSD. Methods We conducted an analysis using data from a single-center registry, which included patients who underwent echocardiography between September 2013 and March 2020. ALVSD was defined by either left ventricular ejection fraction (LVEF) 50% or an average global longitudinal strain (GLS) greater than -16 in asymptomatic patients, consistent with prior research and the guideline. Extreme Gradient Boosting algorithm was applied to derive the risk score (BOOST-LV) for ALVSD patients. We identified 10,529 ALVSD patients categorized as high-risk and low-risk and used this cohort for model creation. We created a risk score that incorporates six critical variables: age, LVEF, average GLS, hemoglobin, albumin, and red blood cell distribution width. The performance of BOOST-LV score was then evaluated in the validation cohort and compared with that of LVEF and GLS using receiver-operating-characteristic (ROC) curve analysis. The BOOST-LV risk score was also divided into seven intervals and Kaplan-Meier estimates were used to analyze and plot the 3-year survival rates for each score interval, along with 95% confidence intervals (CI). Results BOOST-LV score exhibited strong predictive performance, achieving an area under the receiver operating characteristic curve (AUC) of 0.84 (95% CI: 0.83–0.85) in ROC curve analysis, which was superior to that of LVEF (AUC 0.55) and GLS (AUC 0.60). In the plot of 3-year survival rates, patients with very low scores experienced almost no mortality events within 3 years, whereas the likelihood of survival decreased sharply with higher BOOST-LV scores. Conclusions The BOOST-LV score, developed from ALVSD patients, effectively predicts mortality and surpasses LVEF and GLS in 3-year risk prediction. This novel risk score may provide a valuable tool for the personalized management of ALVSD patients.Figure 1 Figure 2, Figure 3