Predicting frequent ED use by people with epilepsy with health information exchange data.

Abstract

To describe (1) the predictability of frequent emergency department (ED) use (a marker of inadequate disease control and/or poor access to care), and (2) the demographics, comorbidities, and use of health services of frequent ED users, among people with epilepsy. We obtained demographics, comorbidities, and 2 years of encounter data for 8,041 people with epilepsy from a health information exchange in New York City. Using a retrospective cohort design, we explored bivariate relationships between baseline characteristics (year 1) and subsequent frequent ED use (year 2). We then built, evaluated, and compared predictive models to identify frequent ED users (≥4 visits year 2), using multiple techniques (logistic regression, lasso, elastic net, CART [classification and regression trees], Random Forests, AdaBoost, support vector machines). We selected a final model based on performance and simplicity. People with epilepsy who, in year 1, were adults (rather than children or seniors), male, Manhattan residents, frequent users of health services, users of multiple health systems, or had medical, neurologic, or psychiatric comorbidities, were more likely to frequently use the ED in year 2. Predictive techniques identified frequent ED visitors with good positive predictive value (approximately 70%) but poor sensitivity (approximately 20%). A simple strategy, selecting individuals with 11+ ED visits in year 1, performed as well as more sophisticated models. People with epilepsy with 11+ ED visits in a year are at highest risk of continued frequent ED use and may benefit from targeted intervention to avoid preventable ED visits. Future work should focus on improving the sensitivity of predictions.