Predictive Analysis of Heterogenous Data for Hospital Readmission

Abstract

Hospital readmission is a high priority health care quality measure. Diabetes patient readmission rate is increasing to such an extent that it becomes one of the major concerns for many hospitals. Many studies are conducted for finding the possible causes and risks of diabetes patient’s readmission. Reducing readmission rates of diabetes patients reduce health care costs. Here the relationship between diabetes and the various patient attributes are examined. Different prediction models were developed to predict the risk of readmission within 30 days among hospitalized patients with diabetes. The dataset used here contains more than 1 lakh observations and 56 features. They include a set of numerical attributes such as number of outpatient visits, number of emergency visits and time spent in hospital etc and a set of categorical data such as what type of admission the encounter faced , sets of drugs that the patient took etc. In this study we presented a scheme to identify high-risk patients and evaluated different machine learning algorithms. Results indicate that Adaboost with hyperparameter tuning is optimal for this task The results from the study help health care providers to improve diabetic care.