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

Bias Estimates of Mortality Using MedisGroups-Reply

Author Affiliations

Kaiser Foundation Health Plan, Inc Oakland, Calif

Kaiser Foundation Health Plan, Inc Oakland, Calif

JAMA. 1991;266(19):2705. doi:10.1001/jama.1991.03470190052025

In Reply.  —The Pennsylvania Hospital Effectiveness Reports are presented by hospital, but their risk models are patient-specific. Thus, it was appropriate for me to test for biases using the patient (not the hospital) as the unit of observation. Nonetheless, Dr Brewster and Ms Throop and Drs Hartz and Van Ruiswyk wish to know how important my findings are when making interhospital comparisons. This depends on the patient characteristics at the hospitals being compared. If patients with a given condition (such as AMI) were randomly admitted to different hospitals, the average risks of these cases would be the same at each hospital. Risk adjustment would then be unnecessary and bias inconsequential. The need for models to make unbiased estimates of the risk of a given outcome for patients at the hospital level increases as patients with a given condition at hospitals become more diverse in their patient characteristics.Brewster and Throop