The overall goals of this project are to derive and evaluate empirically computerized algorithms to screen for potentially substandard hospital care and poor outcomes using readily available administrative data, such as the Uniform Hospital Discharge Data Set (UHDDS). The screens must have good face validity as outcome or quality indicators within the limitations imposed by the UHDDS. The first task is to delineate the screens, in the following steps: (1) refining existing preliminary versions of the screens, focusing on five areas (surgical complications; obstetrical complications, incompatible, contraindicated, or questionable combinations of surgical procedures; outmoded procedures; and """"""""sentinel deaths"""""""", or in-hospital deaths with diagnoses and/or procedures which characteristically have a low death rates); (2) exploring additional screening strategies and developing the computer algorithms; and (3) creating an adjustment for patients' a priori risks of poor outcomes and complications related to the burden of chronic disease (BCD). The second task is to evaluate empirically the results of the screens using UHDDS data from Massachusetts and California, as follows: (1) documenting the yields of the screens across hospitals and at the individual hospital level; (2) identifying patient-level factors which are related to positive findings on particular screens (e.g., BCD, payor source, admission source, race); (3) exploring which hospital-level factors are related to positive findings on individual screens (e.g., bedsize, teaching status, location, ownership, volume of specified surgical procedures); (4) determining whether specific clinical services within individual hospitals (e.g., orthopedics, cardiothoracic surgery) perform differently on the screens than the overall hospital performance; (5) comparing the hospital-level results on the screens to the Medicare mortality ratings produced by the Health Care Financing Administration; and (5) comparing hospital charges of flagged cases with those of cases which pass the screens. The ultimate, potential uses of these screens are twofold: first, to provide a tool for research studies aiming to examine hospital quality and patient outcomes using large administrative databases; and second, to assist in targeting more efficiently, expensive chart-based reviews of quality and patient outcomes. If the empirical evaluations performed here suggest that these screens might prove useful, a subsequent project using medical record reviews will be required to validate these screens as true indicators of quality of care.
