Referral bias in diabetes research.

Of all the known diabetic individuals residing in the community of Rochester, Minnesota, only about one-half would have been recognized through a review of the medical records of 1 yr. Only one-fifth of the diabetic residents were hospitalized during the index year and a similarly small proportion were attended in a diabetes clinic. This selective process produced distortion in the apparent clinical spectrum of diabetes observed at different levels in the medical care system. Notable was overestimation of the relative importance of insulin-dependent diabetes and of the vascular complications of diabetes among hospital and diabetes clinic patients. Referral bias should be taken into account when differences are noted in the epidemiologic features of diabetes as reported from different types of medical care settings.