Prediction of Hospital Readmission through Survival Analysis

This research article presents a comparative study on predicting 30-day unplanned hospital readmissions using survival analysis methods. The objective is to address the challenges associated with estimating readmission risks, which are influenced by various heterogeneous factors. The study evaluates different statistical and machine learning survival analysis models, particularly focusing on tree-ensemble regression methods based on proportional hazards. The performance of these models is assessed using right-censored all-cause hospital admission data, with results indicating that survival models can effectively address the hospital readmission problem. The findings reveal that machine learning models, especially the XGBoost Regression model, outperform traditional statistical methods in predicting readmissions. The article also discusses methodological steps that enhance model performance, such as limiting the observation period to 90 days post-discharge. The document outlines the dataset, data preparation, and the models employed, contributing to the understanding of hospital readmission prediction.