Machine Learning Models Incorporating Nursing Care Needs to Predict 180-Day Prognosis in Patients With Heart Failure　― Validation With Discrimination and Calibration Analyses ―

Abstract

Background: The number of patients with heart failure (HF) is increasing with aging of the population, resulting in a shift in care from hospitals to community settings. Although predicting medium-term prognosis after discharge could improve community-based management and reduce readmissions, no established model has integrated structured multidimensional assessments into HF prognostic modeling.

Methods and Results: This multicenter study developed and validated machine learning (ML) models (i.e., logistic regression, random forest, extreme gradient boosting, and light gradient boosting) to predict 180-day mortality or emergency hospitalization in 4,904 patients with HF. Patients were randomly divided into training and validation sets (8 : 2). Nursing care needs, derived from structured nursing assessments that capture patients’ physical status and care dependency, were included as a predictive feature. All models demonstrated acceptable discriminative performance based on the area under the precision-recall curve, favorable calibration assessed by the calibration slope and Brier score, and effective risk stratification. The Shapley additive explanations algorithm identified nursing care needs as an important prognostic factor, alongside established laboratory variables for HF prognosis.

Conclusions: ML models incorporating nursing care needs effectively predicted the 180-day prognosis of patients with HF. The prominent contribution of nursing care needs underscores the value of incorporating structured multidimensional care-related information into prognostic modeling and highlights the importance of team-based post-discharge HF management.