Development and Application of a Machine Learning-Based Prediction Model for 6-Month Unplanned Readmission in Heart Failure Patients

Accurately predicting the risk of readmission for patients with heart failure is of utmost importance for their prognostic management. In this study, prediction models for 6-month unplanned readmission in heart failure patients were developed using a cohort of 1888 individuals. The cohort was divided into training and testing sets in an 8:2 ratio. Variable selection was performed using Lasso regression, and the prediction models were trained using logistic regression, random forest, decision tree, Bayesian classifier, and support vector machines algorithms. A set of 17 predictive indicators were identified, including age, gender, basophil ratio, monocyte count, neutrophil count, calcium, sodium, glomerular filtration rate, uric acid, prothrombin activity, dementia, type of heart failure, consciousness, cardiac function classification, diabetes, chronic kidney disease and length of hospital stay. Among the models tested, the Bayesian classifier model exhibited a relatively stronger predictive performance (Area under the curve: 0.60, sensitivity: 0.81, specificity: 0.38). Additionally, a user-friendly software application based on the R Shiny package was developed to facilitate the practical implementation of the prediction models.