Risk feature assessment of readmission for diabetes

Abstract

About 382 million people have Diabetes in 2013, and the International Diabetes Federation estimated that there are 4.9 million people died from Diabetes in 2014. Diabetes continues to be a chronic disease plagued by frequent hospital readmissions. In order to better understand the risk features impacting readmissions for future prevention and management, in this study, we programmatically analyzed a large clinical dataset containing more than 100,000 clinical records for diabetes patients from 130 US hospitals. Specifically, we developed three different machine learning algorithms, Logistic Regression, Random Forest and manipulated Random Forest to identify and prioritize the most significant risk features. By comparing the results generated by these three methods, the manipulated Random Forest illustrates greater capacity of generating a more complete and concrete list of readmission related risk features. Such method is generalizable and can be applied in other disease oriented studies.