Abstract:

Objective To compare the predictive performance of multivariable logistic regression and XGBoost models for the risk of 30-day post-discharge readmission in patients with chronic heart failure (CHF). Methods This retrospective study included 288 patients with CHF who were hospitalized in our institution from November 2022 to November 2025. The patients were randomly assigned in a 7∶3 ratio to a training set (n=202) and a validation set (n=86). According to whether readmission occurred within 30 days after discharge because of worsening heart failure, patients were divided into a readmission group (n=62) and a non-readmission group (n=226). Univariate factors associated with 30-day readmission after discharge were analyzed, and multivariable logistic regression and XGBoost models were developed. Internal validation was performed using the validation set. Receiver operating characteristic (ROC) curves were used to assess the predictive performance of the two models. Results Compared with the non-readmission group, the readmission group had higher age, a higher proportion of patients with New York Heart Association (NYHA) class Ⅲ-Ⅳ, higher N-terminal pro-B-type natriuretic peptide (NT-proBNP) levels, longer length of hospital stay, and a higher proportion of patients with irregular medication use after discharge, as well as lower left ventricular ejection fraction (LVEF), serum sodium, and estimated glomerular filtration rate (eGFR) levels (P<0.05). Multivariable logistic regression confirmed that age, NT-proBNP, serum sodium, and irregular medication use after discharge were all factors associated with 30-day readmission after discharge in patients with CHF (P<0.05). The top 8 feature variables ranked by importance in the XGBoost model were NT-proBNP, LVEF, serum sodium, age, eGFR, NYHA class, length of hospital stay, and irregular medication use after discharge. ROC analysis showed that the XGBoost model outperformed the multivariable logistic regression model. Further comparison using DeLong’s test showed that the areas under the curve (AUCs) of the two models were 0.793 and 0.886, respectively, and the difference was statistically significant (Z=2.415, P=0.016). Conclusion Both multivariable logistic regression and XGBoost models can effectively predict the risk of 30-day readmission after discharge in patients with CHF. However, the XGBoost model demonstrated superior predictive performance and may serve as an auxiliary tool for early identification of high-risk patients and clinical intervention decisions.

Key words: heart failure, readmission, risk prediction, logistic regression, XGBoost model