Predictive Modeling of Inpatient Fall of Stroke Patients Using electronic medical records data

• Activity Number: 323
• Type: Contributed
• Date/Time: Tuesday, August 2, 2016 : 8:30 AM to 10:20 AM
• Sponsor: Health Policy Statistics Section
• Author(s): Yin Liu* and Cindy Jin and Jeffrey Yangang Zhang
• Companies: Princeton Pharmatech and Lawrenceville School and Princeton Pharmatech
• Keywords: discrete-time survival data ; multi-task learning ; ensemble algorithm ; electronic medical record ; patient fall

Abstract

Predictive modeling of inpatient fall of stroke patients is challenging. Traditionally, logistic regression or survival models could be applied, but . This study presents a framework of analyzing discrete-time data as a binary classification problem. With its great flexibility of the novel framework, the potential relationship between clusters can be incorporated. Comparisons in experiments suggest that the proposed models consistently yield better predictive performances than classical statistical modeling for survival data. The proposed method is applied to electronic patient records data from 2007 to 2014 collected by Kessler foundation.