Any effort to improve the science of public reporting must include an approach that ensures that guidance provided to the public is accurate, informative, relevant and understandable. Our approach will be to improve upon these elements in the context of outcomes measurement across hospitals using a fully Bayesian framework. In previous work we have demonstrated that the Hospital Compare random effects mortality model provides predictions that may be misleading when evaluating small hospitals. In this application we will develop a more realistic approach to modeling hospital outcomes that is both more accurate (less misleading) than Hospital Compare, and more informative from the perspective of the individual patient seeking guidance on which hospital to choose for care. Furthermore, in order for the public to benefit, not only do the models need improvement, but the public will need to increase their use of these models. To accomplish this latter goal, our approach will address barriers to the general use of these reports. We will develop models that are personalized to specific patient characteristics (making them more relevant for the individual patient), and, by making use of the Bayesian framework, we will introduce new methods for presenting results that adapt to common mistakes surrounding the interpretation of probabilities. Thus, patient error in the interpretation of results will be both less likely to occur and less likly to lead to mistaken hospital selection. Finally, as models will inevitably change and improve, we will develop a framework for future model comparisons, in order to assess whether new models should be adopted. In the end, we hope to develop a better approach with respect to (1) presenting results so that ease of use and understanding is improved, and use is increased;(2) the model predictions are improved;and (3) the process for adoption of future models are made more transparent and rational. Specifically, we will develop a fully Bayesian approach to model development for the conditions of AMI, Pneumonia, and Congestive Heart Failure.
AIM 1 will construct a new approach to the presentation of public reporting results, using Bayesian derived probabilities, constructed so as to reduce errors in selection.
AIM 2 will develop fully probabilistic predictive models of hospital outcomes.
AIM 3 will develop a framework for evaluating any new model for public reporting. We will base our evaluation on two principles: (1) That a population following the recommendations of the improved model should have a higher predicted survival than a population using another model;and (2) We will use data to compare models using, for example, Bayes factors. At the conclusion of this project, we will have developed a better method to present information to the public, a better model for predicting and comparing outcomes across hospitals, and better methods to select and improve future models that may be used to aid the public in hospital selection.
Any effort to improve the science of public reporting must include an approach that ensures that guidance provided to the public is accurate, informative, relevant and understandable. Through the use of Bayesian models (a form of statistical modeling that is ideally suited to compare risks), this project will develop a better method to present information to the public, a better model for predicting and comparing outcomes across hospitals, and better methods to select and improve future models that may be used to aid the public in hospital selection.