# Interpretable Machine Learning Model for Predicting 30‐Day Readmission in Advanced Heart Failure Patients: Synergistic Assessment of Inflammatory and Metabolic Biomarkers

## TL;DR

A machine learning model predicts 30-day readmission in advanced heart failure patients using inflammatory and metabolic biomarkers.

## Contribution

A robust, interpretable model using SHAP values and a web-based nomogram for predicting readmission in advanced heart failure patients.

## Key findings

- Random Forest achieved the highest AUC of 0.85 for predicting 30-day readmission.
- Key predictors included CRP, TYG-BMI, NLR, age, NYHA class, AF, comorbidity count, and ACEI/ARB/ARNI use.
- SHAP analysis confirmed the importance of inflammatory and metabolic markers in readmission prediction.

## Abstract

Patients with advanced heart failure (AdHF) face a high risk of early readmission, leading to poor outcomes and increased healthcare burden. Early identification of high‐risk individuals remains a clinical challenge.

This retrospective study included 769 AdHF patients from the Affiliated Hospital of Xuzhou Medical University, with an independent external validation performed on a cohort of 495 AdHF patients from Shanghai Tenth People′s Hospital. After handling missing data via multiple imputation, the dataset was randomly split into training and validation sets in a 7:3 ratio. Class imbalance was addressed by applying the synthetic minority oversampling technique (SMOTE) exclusively within the training set, followed by the least absolute shrinkage and selection operator (LASSO)‐based feature selection. Seven machine learning models were developed. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), calibration curve, accuracy, sensitivity, specificity, F1 score, and Brier score, among others. SHAP values and a web‐based dynamic nomogram were used for model interpretation and clinical application.

The 30‐day readmission rate was 30.9%. RF achieved the highest AUC (0.85), accuracy (0.79), and the lowest Brier score (0.159). Key predictors included C‐reactive protein (CRP), triglyceride‐glucose‐body mass index (TYG‐BMI), neutrophil‐to‐lymphocyte ratio (NLR), age, New York Heart Association (NYHA) class, atrial fibrillation (AF), comorbidity count, and angiotensin‐converting enzyme inhibitors/angiotensin receptor blockers/angiotensin receptor–neprilysin inhibitors (ACEI/ARB/ARNI) use. SHAP analysis confirmed the importance of inflammatory and metabolic markers. A web‐based nomogram was constructed to allow interactive risk prediction.

This study presents a robust and interpretable model for predicting a 30‐day readmission in AdHF patients, highlighting the role of inflammation, metabolism, and comorbidity burden. The model can assist clinicians in risk stratification and personalized postdischarge management.