—To demonstrate theoretically and empirically the existence of systematic bias in commonly reported standardized hospital mortality ratios when variation in hospital admission practice is not adjusted for in the analysis. The underlying analytic model used in hospital mortality analyses is specified and the confounding effect of selection bias arising from variation in admission practice is shown.
—An empirical example is presented using state-level data from the Health Care Financing Administration's Medicare Hospital Information Report for 1988 to 1990.
—The Medicare Hospital Information data are used to demonstrate the effects of the bias because they contain population-based admission rates and mortality rates.
—Selection bias arising from variation in admission practice causes the expected mortality rate to be overestimated for all hospitals, but especially for hospitals with more lenient admission practices. Using the Medicare Hospital Information Report, the resulting standardized hospital mortality ratios are shown to be significantly inversely correlated with higher relative risks of hospitalization (P<.01).
—Standardized hospital mortality ratios based on analyses that do not account for variation in admission practice among hospitals are biased. Variation in admission practice will cause any outcome measure based solely on hospitalized patients to be similarly biased. Correction for selection bias is required to produce valid measures of hospital quality.(JAMA. 1994;271:1493-1498)