Predicting Hospital safety: A Comparison of Four Models

Choosing the best hospitals or avoiding the worst hospitals is important to patients, physicians, and insurers. There is a wealth of publicly reported quality data to use in making those choices, but that data is presented in ways that may conceal important information. We develop three models of how individuals might use the available data in choosing a hospital and present a model that relies on a group-based alternative to the current reporting methods. We compare these four models on their ability to both predict hospital performance on quality and to identify the best and worst hospitals using three measures of patient safety calculated for California hospitals 1997-2006. We find that there are no clear winners in predicting the levels of quality, but we do find evidence that groupbased models may be better at identifying the worst hospitals.