Machine learning–based prediction of mortality and hospitalization in diabetic patients with heart failure with preserved ejection fraction: the GUARDIAN-P risk score

Diabetes mellitus (DM) is a major contributor to adverse outcomes in patients with heart failure with preserved ejection fraction (HFpEF). We aim to develop and externally validate a machine learning-based model using a random survival forest (RSF) approach for predicting the composite outcome of hospitalization for heart failure (HHF) and cardiovascular (CV) death in patients with DM and HFpEF.

This retrospective cohort study included 1450 adult patients with coexisting DM and HFpEF identified from the National Taiwan University Hospital-Integrated Medical Database. An initial RSF model was trained using 27 clinical variables, and the top 9 predictors were selected to construct a parsimonious final model. Predictive performance was evaluated using the area under the receiver operating characteristic curve (AUC), and external validation was conducted in an independent cohort (n = 729) from MacKay Memorial Hospital. Over a mean follow-up of 3.6 ± 3.0 years, 327 patients (22.6%) experienced the composite outcome. The final RSF model achieved an AUC of 88.2% in the training cohort and 79.8% in the validation cohort. The nine selected predictors were age, N-terminal pro-brain natriuretic peptide, serum albumin, fasting glucose, estimated glomerular filtration rate, uric acid, left atrial diameter, peripheral artery disease, and left ventricular ejection fraction. Risk increased progressively with the number of risk factors present.

The RSF-based model incorporating nine routinely available variables accurately predicts HHF and CV death in patients with DM and HFpEF. This tool may support personalized risk assessment and guide clinical decision-making.