Predicting Mortality and Hospitalization in Heart Failure With Preserved Ejection Fraction by Using Machine Learning (JACC Asia December 2024)-Article: Predicting Mortality and Hospitalization in Heart Failure With Preserved Ejection Fraction by Using Machine Learning

Article: Predicting Mortality and Hospitalization in Heart Failure With Preserved Ejection Fraction by Using Machine Learning

Back to course

Pdf Summary

The study aimed to identify predictors for heart failure (HF) hospitalization and cardiovascular (CV) death specifically in patients with heart failure with preserved ejection fraction (HFpEF) using machine learning techniques. A retrospective analysis utilized the Chang Gung Research Database in Taiwan, involving 6,092 HFpEF patients with data spanning from 2008 to 2019. Researchers developed a predictive model using a random survival forest (RSF) algorithm, incorporating 58 variables from clinical, laboratory, and echocardiographic data.<br /><br />The study identified 15 significant predictors of HF hospitalization and CV death: age 65 years, high B-type natriuretic peptide levels, enlarged left atrium size, presence of atrial fibrillation, prior HF hospitalizations, body mass index (BMI) 30 kg/m², moderate or severe mitral regurgitation, left ventricular (LV) posterior wall thickness, dysnatremia, LV end-diastolic dimension, elevated uric acid levels, abnormal triglyceride and blood urea nitrogen (BUN) levels, interventricular septum thickness, and glycated hemoglobin level. The model exhibited strong predictive performance, achieving an 86.9% area under the curve (AUC) in the validation cohort.<br /><br />This approach allows for high-risk patient identification and personalized treatment planning in clinical practice, offering a more nuanced understanding of HFpEF risk factors compared to previous models. The study highlighted the importance of integrating comprehensive clinical and echocardiographic data in developing predictive models for HFpEF, suggesting that implementing this model in electronic medical records could facilitate better heart failure management. Despite its strong findings, the study acknowledged limitations like the absence of data on some lifestyle factors, and broader validation across different demographics is recommended for future research.

Keywords

heart failure

HFpEF

machine learning

predictive model

random survival forest

cardiovascular death

risk factors

clinical data

echocardiographic data

personalized treatment