Accurate and practical methods for performing risk adjustment are crucial for studies using administrative data to assess health care outcomes and quality of care. This research will use California hospital discharge abstract data, which identifies whether or not reported diagnoses were present on admission, to develop risk adjustment methods that more accurately and completely assess the importance of preexisting disease. We will develop and apply our improved risk adjustment methods to assess inpatient mortality outcomes in the following 7 patient populations: 1. respiratory failure, 2. septicemia, 3. acute cerebrovascular disease, 4. acute myocardial infarction, 5. pneumonia, 6. congestive heart failure, and 7. Chronic obstructive pulmonary disease. These diseases groups represent common primary diagnoses for hospitalized patients and substantial numbers of the patients in these disease groups die during their hospital stay. The research is organized within 4 stages. In the first stage, criteria will be developed for use in defining and refining diagnosis classification categories for use as predictor variables in the models. The criteria will be developed using a process that combines empirical assessment with a physician panel review of the selected diagnoses. In the second stage, parameters for predictor variables in the developed risk adjustment models will be estimated using California hospital discharge abstract data for calendar years 1996 and 1997. Models that assess the effects of the severity of the primary diagnosis and of preexisting disease will be estimated in each of the 7 study populations. In the third stage, the developed risk adjustment models will be rigorously validated by applying them to identically defined patient groups identified using California hospital discharge abstract data for the 1998 calendar year, and to identically defined patient groups identified using the Nationwide Inpatient Sample of the Healthcare Cost and Utilization Project (HCUP) for the calendar years 1993-1997. In the fourth stage, we will compare the statistical performance of the developed models to that of other comparable methods (Deyo/Charlson method and the Elixhauser et al. Method) in each study population.
