Prediction of prognosis in elderly patients with chronic heart failure based on random survival forest

There is a lack of tools to identify the risk of poor prognosis in elderly patients with chronic heart failure (CHF). This study aimed to develop a random survival forest (RSF) model to predict the prognosis of elderly CHF patients.

The primary endpoint of this was all-cause mortality. The secondary endpoint was the combined outcome of unplanned readmissions and all-cause mortality. Patients were divided into a training set and a test set at a ratio of 7:3. We established and compared the performance of the RSF model with that of the New York Heart Association (NYHA) functional classes, left ventricular ejection fraction (LVEF) and B-type natriuretic peptide (BNP) level in evaluating the prognosis of elderly CHF patients. Harrell's C-index, decision curve analysis (DCA) and calibration curves were the main evaluation metrics for the model.

A total of 525 patients were enrolled. At a median follow-up of 60.1 (46.2, 63.5) months, 168 (32.0%) patients reached the primary endpoint and 219 (41.7%) patients reached the secondary endpoint. The C-index of the RSF model for predicting the primary endpoint was 0.747 in the training set and 0.714 in the test set. For the secondary endpoint, the C-index of the RSF model was 0.707 in the training set and 0.641 in the test set. DCA and calibration curves demonstrated that the RSF model showed good clinical usefulness and calibration.

The RSF model showed good discrimination, clinical usefulness and calibration in predicting the prognosis of elderly CHF patients.