The current approach for provider profiling has significant limitations that deserve immediate attention due to the extensive repercussions profile reports have on the health care system. The long-term goal is to improve the current methodology for profiling providers by addressing some of its most pressing limitations. Since performance estimates are only valid when there exists sufficient overlap in patient characteristics, a substantial limitation with the current approach is its failure to assess the extent of patient covariate overlap among providers being profiled. The overall objective for this application is to implement a methodology that identifies similar groups of providers based on the admission characteristics of the patients they treat. The conjecture is that providers will need to be assigned to multiple groups to achieve sufficient overlap in patient characteristics and that, upon grouping providers, conclusions regarding provider performance will differ from conclusions under the current approach. Addressing this significant limitation provides a novel framework from which other researchers can build from to continue refining the current approach. The overall objective will be attained by pursuing three specific aims: 1) develop a method for grouping providers based on patient admission characteristics; 2) perform balance checks informing the number of groups needed for within-group patient covariate balance; and 3) apply methods to nursing home data to compare performance of homes within each group. For the first aim, a Bayesian hierarchical mixture model will be used to estimate each provider?s posterior probability of belonging to the different groups based on their patient?s admission characteristics. These posterior probabilities will be used to assign each provider to a group. For the second aim, balance checks assessing within-group balance in patient admission covariates will be designed to ensure that the distribution of patient covariates within each group resembles what would be expected in a randomized setting. This mirrors what is done in causal inference prior to comparing multiple treatments.
The third aim applies the developed methodology by grouping nursing homes across the US using patient characteristics like age and gender. Upon using balance checks to inform the number of groups homes should be assigned to, home readmission rates will be compared within each group. The proposed dissertation research is innovative, in the applicant?s opinion, because it develops a promising link between causal inference and provider profiling while applying novel Bayesian statistical methodology to the profiling of providers. The proposed project is significant because it reformulates the current statistical methodology and thus has the potential to improve the accuracy of profiling reports. This is of utmost importance, as accurate performance estimates help patients make informed decisions, motivate providers to improve quality, and guide policymakers as they develop policy. .
The proposed dissertation research is relevant to public health because it addresses a significant limitation in the current methodology used to evaluate provider performance. Reports profiling providers are publicly available and so the conclusions presented in these have significant impacts in the behaviors and decisions of those in the health care system. The dissertation project is highly relevant to AHRQ's mission to link research and practice: developing a method that properly identifies a hospital?s quality will lead to modifications in the health care system that improve health care quality and patient safety.
