[Skip to Content]
[Skip to Content Landing]
November 20, 1991

Bias Estimates of Mortality Using MedisGroups

Author Affiliations

Medical College of Wisconsin Milwaukee

Medical College of Wisconsin Milwaukee

JAMA. 1991;266(19):2704-2705. doi:10.1001/jama.1991.03470190052024

To the Editor.  —Dr Blumberg1 used the term bias in an unconventional and misleading way. Bias "occurs when comparisons are made between groups of patients that differ with respect to determinants of the outcome other than those under study. "2 Blumberg used the word bias to describe the relationships between several patient characteristics (age, AMI location, CHF status) and the ratio of the observed mortality rate to the mortality rate predicted using the MedisGroups admission severity groups (ASG) score. These relationships do not illustrate bias. To show that the ASG score provides biased estimates of mortality, it is necessary to show that two patient populations (eg, patients treated at teaching hospitals vs nonteaching hospitals) systematically differ with respect to risk in ways not accounted for by the ASG score. The relationships in the Blumberg study illustrate only that the patient characteristics are risk factors that have not been adequately