Predictive Modeling for Preventable Hospital Readmissions: A Data Mining Approach

• Activity Number: 356
• Type: Contributed
• Date/Time: Tuesday, July 31, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Jingjing Gao*+ and Xia Lin and Andrew Post and Sharath Cholleti and Joel Saltz
• Companies: Emory University and Emory University and Emory University and Emory University and Emory University
• Keywords: classification and prediction ; hospital readmission rate ; decision trees ; classification and regression trees ; random forest ; support vector machine

Abstract

The accelerated use of electronic health records (EHR) has allowed researcher to integrate and retrieve longitudinal and high-dimensional electronic patient data in a more efficiently way, which leads to extensive retrospective data analysis on patients' healthcare history. The hospital 30-day readmission rate is one of the arising topics attracted great attention. Previous studies constructed generalized linear models for classification and prediction, which have limited feasibility in handling remarkable quantities of EHR data mainly due to inevitable assumptions of parametric models, whereas data mining provides methods and tools to extract meaningful patterns and rules without making assumptions on underlying distribution. In this study, we apply several classification techniques including CART, Random Forest and SVM to identify leading medical and clinical factors that distinguish those patients who are highly likely to be readmitted to hospital within 30 days of initial visit from those who are not. K-fold cross-validation is used for evaluation. High accuracy has been achieved especially in classifying patients who are highly unlikely to be readmitted within 30 days.