Differences among hospitals in Medicare patient mortality.

Using hospital discharge abstract data for fiscal year 1984 for all acute care hospitals treating Medicare patients (age greater than or equal to 65), we measured four mortality rates: inpatient deaths, deaths within 30 days after discharge, and deaths within two fixed periods following admission (30 days, and the 95th percentile length of stay for each condition). The metric of interest was the probability that a hospital would have as many deaths as it did (taking age, race, and sex into account). Differences among hospitals in inpatient death rates were large and significant (p less than .05) for 22 of 48 specific conditions studied and for all conditions together; among these 22 "high-variation" conditions, medical conditions accounted for far more deaths than did surgical conditions. We compared pairs of conditions in terms of hospital rankings by probability of observed numbers of inpatient deaths; we found relatively low correlations (Spearman correlation coefficients of 0.3 or lower) for most comparisons except between a few surgical conditions. When we compared different pairs of the four death measures on their rankings of hospitals by probabilities of the observed numbers of deaths, the correlations were moderate to high (Spearman correlation coefficients of 0.54 to 0.99). Hospitals with low probabilities of the number of observed deaths were not distributed randomly geographically; a small number of states had significantly more than their share of these hospitals (p less than .01). Information from hospital discharge abstract data is insufficient to determine the extent to which differences in severity of illness or quality of care account for this marked variability, so data on hospital death rates cannot now be used to draw inferences about quality of care. The magnitude of variability in death rates and the geographic clustering of facilities with low probabilities, however, both argue for further study of hospital death rates. These data may prove most useful as a screening mechanism to identify patterns of potentially poor quality of care. Careful choice of the mortality measure used is needed, however, to maximize the probability of identifying those hospitals, and only those hospitals, warranting more in-depth review.