Background

Mortality and length of stay are frequently used as performance measures for hospitals. If they are valid measures, they should be reproducible from year to year with attributable variation rather than random variation.

Methods

We compared hospitals on 2 outcomes, mortality and length of stay, in pneumonia in Medicare patients. The database was from 20 Illinois hospitals with the largest number of discharges for diagnosis-related group 89 (pneumonia with complications/comorbidities) for the years 1989 through 1992. This comprised 16,249 claims for hospitalization in patients 65 years of age or older.

Results

The distributions showed trends toward lower mortality and shorter stays over the 4 years. Correlation of performance from year to year at each hospital for mortality was low with none of the calculated correlation coeffi-cients significant at p < .05. Correlations for length of stay were higher (all coefficients significant at p < .01). For length of stay, the correlation between 1991 and 1992 was .766 (p < .00005, r² = .587), showing that nearly 60% of differences (variance) were caused by differences in performance. In contrast, for mortality in 1991 and 1992, the correlation was .301 (p = .0986, r² = .091), showing that less than 10% of differences (variance) between hospitals were caused by differences in performance. Similar results were obtained when the 20 hospitals were ranked and their rank correlations calculated.

Conclusion

For pneumonia in Medicare patients, differences in length of stay between hospitals are caused by differences in performance, while differences in mortality are random.