SMART-HF

SMART-HF predicts one-year mortality after hospitalization for acute decompensated heart failure using machine learning on administrative claim data from the Japanese Registry of Acute Decompensated Heart Failure (10,175 patients).

Key Features:

• Dataset: Uses administrative claim data (ACD) derived from the Japanese Registry of Acute Decompensated Heart Failure comprising 10,175 heart failure patients.
• Machine learning algorithms: Evaluated six different ML algorithms and reported a voting classifier algorithm (ACD-VC) that achieved a c-statistic of 0.777.
• Predictor identification: Applied permutation feature importance to identify key predictors: Barthel index (0.054), age (0.025), body mass index (0.010), duration of hospitalization (0.005), last hospitalization (0.005), renal disease (0.004), and non-loop diuretics use (0.004).
• Performance metrics: Reported overall SMART-HF performance with a c-statistic of 0.765 (95% CI 0.739–0.791) and a Brier score of 0.124, and compared these metrics against the Seattle Heart Failure Model (SHFM; c-statistic 0.713) and MAGGIC (c-statistic 0.726).

Scientific Applications:

• Risk stratification: Predicts one-year mortality risk after hospitalization for acute decompensated heart failure to support prognostic assessment.
• Model comparison and improvement: Provides a machine-learning–based prognostic alternative benchmarked against conventional risk models SHFM and MAGGIC.

Methodology:

Analysis of administrative claim data from the Japanese Registry of Acute Decompensated Heart Failure (n=10,175); evaluation of six machine learning algorithms including a voting classifier (ACD-VC); permutation feature importance for predictor ranking; performance assessed by c-statistics and Brier scores with comparisons to SHFM and MAGGIC.