Parametric estimation procedures for screening programmes: Stable and nonstable disease models for multimodality case finding

This paper develops methods for estimating the parameters associated with e arly detection programmes. Disease is considered to have three states: a di sease-free state or a state in which the disease cannot be detected, a prec linical state and a clinical state. The natural history of the disease is a ssumed to be progressive. The parameters to be estimated, are the sensitivi ty of one or two disease detection modalities and the characteristics of th e preclinical sojourn time distribution under both the stable disease and n onstable disease models. The stable-disease model assumes that the incidenc e or prevalence of a disease is independent of age or chronological time, w hile the nonstable disease model allows these quantities to depend on time. With the nonstable disease model, the relevant parameters can be jointly e stimated by a two-step iteration procedure from the likelihood function. Fo r the stable disease model, the sensitivity and the parameters of the sojou rn time distribution of the preclinical state can be obtained directly from a conditional likelihood function. Applications are made to recent clinica l trials for the early detection of breast cancer.