Using Real-World Hospital Data To Develop And Evaluate Readmission Risk Prediction Models For Clinical Practice

Abstract

Hospital readmissions remain a critical challenge in healthcare systems, impacting both patient outcomes and resource utilization. This study aims to develop and evaluate readmission risk prediction models using real-world hospital data to enhance clinical decision-making. A comprehensive dataset integrating clinical, demographic, administrative, and billing information was utilized to construct predictive models, including logistic regression, random forest, and gradient boosting. The findings demonstrate that ensemble-based models outperform traditional approaches, achieving higher discrimination, improved classification performance, and better calibration. Key predictors of readmission included comorbidity burden, prior admissions, laboratory abnormalities, and length of stay, reflecting the multifactorial nature of readmission risk. Subgroup analysis indicated stronger model performance among high-risk and older populations, highlighting the influence of patient heterogeneity. The study underscores the importance of leveraging multidimensional real-world data to improve predictive accuracy and support targeted interventions at discharge. While the results demonstrate strong potential for clinical application, challenges related to model interpretability, generalizability, and data quality remain. Overall, this research reinforces the role of predictive analytics in reducing hospital readmissions and improving healthcare efficiency.