The hazards of using active clinic patients as a source of subjects for clinical studies

Abstract

The authors describe and empirically demonstrate a form of bias that results from deriving subjects for clinical studies from available patients currently being followed in specific disease clinics instead of inception cohorts (patients enrolled at a uniform and early point in their disease). They label this effect “clinic patient bias.” It is a variation of prevalence-incidence (Neyman) bias in that it also results from the time gap between the onset of a specific characteristic (a risk factor, exposure or disease) and enrollment in the study, causing selective exclusion of fatal or short episodes, or mild or silent cases. Clinic patient bias may distort an estimate of relative risk in either direction. The empirical example is derived from a study of risk factors for developing complications such as peritonitis among end-stage renal disease patients treated with continuous ambulatory peritoneal dialysis (CAPD). The use of available clinic patients rather than an inception cohort (patients newly beginning CAPD) resulted in the demonstration of false apparent risk relationships for two variables: the calendar date when patients began CAPD (with those enrolled at an earlier time appearing to be at lower risk), and serum albumin level at the start of CAPD (with those having lower albumin levels appearing to be at higher risk). This example demonstrates one of the potential hazards of using active or available clinic patients as a source of subjects for clinical studies.