Computer-based models to identify high-risk adults with asthma: Is the glass half empty or half full?

This study developed and evaluated the performance of prediction models for asthma-related adverse outcomes based on the computerized hospital, clinic , and pharmacy utilization databases of a large health maintenance organiza tion. Prediction models identified patients at three- to four-fold increase d risk of hospitalization and emergency department visits, and were valid f or test samples from the same population. A model that identified 19% of pa tients as high risk had a sensitivity of 49%, a specificity of 84%, and a p ositive predictive value of 19%. We conclude that prediction models that ar e based on computerized utilization data can identify adults with asthma at elevated risk, but may have limited sensitivity and specificity in actual populations.