Discontinuity in the Health Care Financing Administration's published standardized mortality ratios due to underestimation bias across multiple admissions case selection methods.

Case counts below 50 can compromise the careful use of disease- or procedure-specific standardized mortality ratio comparisons across small hospitals in predominantly rural areas. Linking data series where data are of the same quality and are continuous across the series is one way of handling the problem of small case sizes. Examination of three different annual reports of Health Care Financing Administration (HCFA) mortality findings for 1987 of one state's providers showed significantly lower actual and predicted mortality rates reported in 1989, compared with the same mortality rates reported about 1987 published in 1988 and 1987. One hypothesis for the downward trend is the change in case selection method used for constructing the mortality model from last discharge (used in 1987 and 1988 reports) to random selection of cases with multiple admissions (initiated in 1989). A test of both case selection methods is presented and shows the consequences of mortality model changes that create discontinuities with previously reported findings, changes that limit or at least need to be taken into account when linking current HCFA mortality data series with historic series.