Mayo Clinic Proceedings Home

Referral Bias in Elderly Patients

      When surgeons and anesthesiologists are asked to estimate the probability of perioperative morbidity or mortality in their preoperative counseling of patients, they often quote published statistics from large referral centers such as the Mayo Clinic. Elderly patients are often more interested than other patients in accurate appraisals of risk because their chances of untoward outcome are higher and the option of not undergoing a surgical procedure—that is, “living with the condition”—may become more attractive with advancing age. Therefore, it is important to note that referral bias of elderly patients to such tertiary medical centers may result in a nonrepresentative sample of patients from which outcome is determined.
      In this issue of the Proceedings (pages 1185 to 1191), Warner and his collaborators report an analysis of perioperative morbidity and mortality for an 11-year experience with 795 Mayo Clinic patients 90 years of age or older. The excellent medical record system at the Mayo Clinic allowed a thorough analysis and comparison of the preoperative conditions of these patients. Postoperative follow-up was often possible until death. Their analysis compared patients from (1) the local area of Olmsted County, Minnesota, (2) within a radius of 250 miles but not from Olmsted County, and (3) a distance of 250 miles or more.
      Olmsted County patients were found more likely to be older, to be of female sex, and to have a higher prevalence of central nervous system and cardiovascular diseases than patients from the other groups. In addition, Olmsted County patients were more likely to require emergency surgical procedures for traumatic injuries, orthopedic operations, and surgical procedures of longer duration. Patients from a distance were more likely to be men in their early 90s with fewer coexisting diseases. Perioperative morbidity was similar among the groups, except that the Olmsted County patients were more likely to need ventilator therapy. A lower long-term survival rate was found in non-Olmsted County patients from within 250 miles than in the other two comparison groups.
      When various preoperative conditions, patient demographic information, and type of surgical procedure were used in a model to predict outcome, the distance from the Mayo Clinic was found not to be a significant predictive factor for perioperative mortality or morbidity. This finding suggests that once the patients arrived at the Mayo Clinic their outcomes depended on only the prognostic factors identified by Warner and associates in Tables 1 and 2 of their article. In a specific patient who needed, for example, a prostatic surgical procedure, the probability of postoperative morbidity or mortality was the same whether he came from the local area or from a distance. He apparently had less chance of being referred to the Mayo Clinic if he was older or had coexisting diseases.
      Although systematic differences existed in the selection of very elderly patients from the local population and the population at a distance, no additional risk of morbidity or mortality was attributable to distance itself. These findings suggest that estimates of perioperative mortality and morbidity need to be adjusted for known prognostic factors.
      • Lilienfeld AM
      • Lilienfeld DE
      It would be interesting to know whether differences in the composition of referrals based on distance would also be present in younger patients. One can speculate that a very elderly patient with a coexisting disease, such as the neurologic residual of a stroke, would not be referred a considerable distance to the Mayo Clinic. Would that also be true for a 65-year-old patient? Are all reports of studies from tertiary medical centers somewhat biased because they represent samples from generally more healthy patients with the resources and direction necessary for referral?
      Predictions for mortality and morbidity are useful only when they are adjusted for appropriate prognostic factors. These factors should be quoted as part of any study. Investigators should be aware of possible referral bias in large tertiary medical centers that may substantially affect the composition of the study sample with respect to important prognostic factors. The referral bias pointed out by Warner's group emphasizes the need for careful consideration of referral data in conjunction with other factors when risk and outcome are assessed in elderly surgical patients.


        • Lilienfeld AM
        • Lilienfeld DE
        Foundations of Epidemiology. Second edition. Oxford University Press, New York1980