Consumers of outcomes research may assume that risk-adjustment procedures based on patients' comorbid conditions will control for baseline prognostic differences between comparison groups, so that differences in risk-adjusted outcomes represent effects other than those due to differences in comorbidity severity. However, surgeons may differ in their threshold to operate on patients with different intensities of the same comorbidity, which may not be accounted for using commonly employed risk-adjustment methods. We developed a model to explore the effect that selection based on comorbidity severity could have on estimates of the risk-adjusted relative risk (RR) of operative death. Larger effects on the apparent RR of operative death were observed when both the proportion of patients in the high-risk ("selected") stratum and the relative increase in the risk of death due to being in the high-risk stratum were large. Biased estimates of the risk-adjusted RR of operative death will be observed if surgeons differentially select patients based on comorbidity severity and if differences in comorbidity severity are not captured by the risk-adjustment methodology.