Risk Prediction of Early Readmission in Patients with Diabetes Using Machine Learning Algorithms: A Retrospective Clinical Study

Abstract

Patients with diabetes and early hospital readmissions have been a major problem to healthcare systems, which leads to higher costs of healthcare and poor patient outcomes. Proper identification of high-risk patients is critical to enhancing the quality of care and resource use. The purpose of the study was to design and test machine learning models to predict early hospital readmission (within 30 days) in patients with diabetes and their possible applicability to clinical decision-making. The Diabetes 130-US hospitals dataset (more than 100,000 patient encounters) was used to carry out a retrospective clinical study. The measure of outcome was early readmission within 30 days. Several machine learning models, such as Logistic Regression, Random Forest, and Gradient Boosting (XGBoost) were created and tested. Improvements in model performance were made using data preprocessing and feature selection techniques as well as class imbalance handling techniques. The models were evaluated in terms of accuracy, precision, recall, F1-score, and area under the ROC curve (AUC) and interpretability in terms of feature importance and SHAP analysis. The findings showed that ensemble models were more effective than the traditional ones, with XGBoost being the best (AUC ≈ 0.81). The most crucial predictors of readmission were previous inpatient admissions, emergency admissions, length of stay, and medication switch, which all presented clinically significant information. The results show that machine learning models are capable of predicting early readmission of diabetes patients with real-world clinical data. Clinical applicability is maximised by combining predictive modelling with interpretability, and is used in everyday healthcare practice to support decision-making. These findings demonstrate the promise of artificial intelligence to enhance patient outcomes and health care efficiency by enhancing the stratification of risks and targeted interventions.