Big data and predictive modeling using a large administrative dataset to understand population-level changes in emergency department use

Abstract

Research has shown that many emergency department visits could be avoided with better access to primary and ambulatory care. Thus, emergency department use has long been a barometer for researchers and policymakers looking to monitor health systems. In this study, a large administrative database containing insurance and health services information for millions of people allowed us to look for new insights into population-level changes in care patterns after health reforms in Massachusetts in 2006. We applied multivariable generalized linear regression models to predict a new measure of emergency department use that offers several advantages over other measures, giving us insights into the characteristics of individuals with potentially preventable visits. In this case study, we describe the measure, our approach to modeling, a number of unexpected issues we encountered along the way, and the solutions we used to resolve them.