Estimation of the duration of a pre-clinical disease state using screening data

In this paper, the authors show how data on the observed prevalence of disease at a screen and on the incidence of disease during intervals between screens may be used to estimate jointly the distribution of the length of time during which individuals remain in the pre-clinical state and the sensitivity of the screen. Apart from being of biologic interest, such estimates may be used to evaluate the length of time by which the date of diagnosis could be advanced by screening (the lead time) as well as to predict the relative effectiveness of various alternative screening strategies. The methodology uses only information which should be routinely available in the course of a typical screening program, and makes only rather mild statistical assumptions. The authors illustrate the methods with breast cancer screening data from the Health Insurance Plan of Greater New York (HIP). Although these data have been analyzed by several other authors, the present approach is the first which simultaneously gives estimates of the pre-clinical state duration, the sensitivity of the screening method, and the underlying incidence rate in the screened group, while also taking into account the problem of length-biased sampling.