USING MEDICARE CLAIMS DATA TO ASSESS PROVIDER QUALITY FOR CABG SURGERY - DOES IT WORK WELL ENOUGH

Objectives. To assess the relative abilities of clinical and administr ative data to predict mortality and to assess hospital quality of care for CABG surgery patients. Data Sources/Study Setting. 1991-1992 data from New York's Cardiac Surgery Reporting System (clinical data) and HCFA's MEDPAR (administrative data). Study Design/Setting/Sample. This is an observational study that identifies significant risk factors fo r in-hospital mortality and that risk-adjusts hospital mortality rates using these variables. Setting was all 31 hospitals in New York State in which CABG surgery was performed in 1991-1992. A total of 13,577 p atients undergoing isolated CABG surgery who could be matched in the t wo databases made up the sample. Main Outcome Measures. Hospital risk- adjusted mortality rates, identification of ''outlier'' hospitals, and discrimination and calibration of statistical models were the main ou tcome measures. Principal Findings. Part of the discriminatory power o f administrative statistical models resulted from the miscoding of pos toperative complications as comorbidities. Removal of these complicati ons led to deterioration in the model's C index (from C = .78 to C = . 71 and C = .73). Also, provider performance assessments changed consid erably when complications of care were distinguished from comorbiditie s. The addition of a couple of clinical data elements considerably imp roved the fit of administrative models. Further, a clinical model base d on Medicare CABG patients yielded only three outliers, whereas eight were identified using a clinical model for all CABG patients. Conclus ions. If administrative databases are used in outcomes research, (1) e fforts to distinguish complications of care from comorbidities should be undertaken, (2) much more accurate assessments may be obtained by a ppending a limited number of clinical data elements to administrative data before assessing outcomes, and (3) Medicare data may be misleadin g because they do not reflect outcomes for all patients.