Computer-based Models to Identify High-risk Children with Asthma

Abstract

Effective management of populations with asthma requires methods for identifying patients at high risk for adverse outcomes. The aim of this study was to develop and validate prediction models that used computerized utilization data from a large health-maintenance organization (HMO) to predict asthma-related hospitalization and emergency department (ED) visits. In this retrospective cohort design with split-sample validation, variables from the baseline year were used to predict asthma-related adverse outcomes during the follow-up year for 16,520 children with asthma-related utilization. In proportional-hazard models, having filled an oral steroid prescription (relative risk [RR]: 1.9; 95% confidence interval [CI]: 1.3 to 2.8) or having been hospitalized (RR: 1.7; 95% CI: 1.1 to 2.7) during the prior 6 mo, and not having a personal physician listed on the computer (RR: 1.6; 95% CI: 1.1 to 2.3) were associated with increased risk of future hospitalization. Classification trees identified previous hospitalization and ED visits, six or more beta-agonist inhalers (units) during the prior 6 mo, and three or more physicians prescribing asthma medications during the prior 6 mo as predictors. The classification trees performed similarly to proportional-hazards models, and identified patients who had a threefold greater risk of hospitalization and a twofold greater risk of ED visits than the average patient. We conclude that computer-based prediction models can identify children at high risk for adverse asthma outcomes, and may be useful in population-based efforts to improve asthma management.