Referral selection bias in the Medicare hospital mortality prediction model: are centers of referral for Medicare beneficiaries necessarily centers of excellence?

OBJECTIVE

Although the Health Care Financing Administration (HCFA) uses Medicare hospital mortality data as a measure of hospital quality of care, concerns have been raised regarding the validity of this concept. A problem that has not been fully evaluated in these data is the potential confounding effect of illness severity factors associated with referral selection and hospital mortality on comparisons of risk-adjusted hospital mortality. We address this issue.

DATA SOURCES AND STUDY SETTING

We analyzed the 1988 Medicare hospitalization data file (MEDPAR). We selected data on patients treated at the two Mayo Clinic-associated hospitals in Rochester, Minnesota, and a group of seven other hospitals that treat many patients from large geographic areas. These hospitals have had observed mortality rates substantially lower than those predicted by the HCFA model for the period 1987-1990.

STUDY DESIGN

Using the multiple logistic regression model applied by HCFA to the 1988 data, we evaluated the relationship between distance from patient residence to the admitting hospital and risk-adjusted hospital mortality.

PRINCIPAL FINDINGS

Among patients admitted to Mayo Rochester-affiliated hospitals, residence outside Olmsted County, Minnesota was independently associated with a 33 percent lower 30-day mortality rate (p < .001) than that associated with residence in Olmsted County. When patients at Mayo hospitals were stratified by residence (Olmsted County versus non-Olmsted County), the observed mortality was similar to that predicted for community patients (9.6 percent versus 10.2 percent, p = .26), whereas hospital mortality for referral patients was substantially lower than predicted (5.0 percent versus 7.5 percent, p = < .001). After incorporation of the HCFA risk adjustment methods, distance from patient residence to the hospitals was also independently associated with mortality among the Mayo Rochester-affiliated hospitals and seven other referral center hospitals.

CONCLUSIONS

The HCFA Medicare hospital mortality model should be used with extreme caution to evaluate hospital quality of care for national referral centers because of residual confounding due to severity of illness factors associated with geographic referral that are inadequately captured in the extant prediction model.