|
Abstract:
|
Sometimes measured data are subject to length-biased sampling (e.g., gas lines or hospitalization stays). When the variable, which is length-biased sampled, is observable, the distribution of the observations is known (Cox and Lewis 1972). Length bias exists in screening programs where disease may be detected during the preclinical stage, as a longer sojourn time yields a greater likelihood of being screen-detected (Zelen 1976). This paper quantifies the effect of length-biased sampling on clinical duration when cases are subject to periodic screening. Underlying bivariate distribution, correlation (between sojourn and clinical time), mixing proportion of slow versus fast progressing disease, and the ratio of screening interval length to mean sojourn time all influence the magnitude of the effect length-biased sampling has on the distribution of clinical duration.
|
