Modeling and Analysis of Chronic Disease Processes Under Intermittent Observation

A wealth of information is available from clinic-based registries of individuals with chronic diseases but modeling and analysis of disease processes can be challenging because of the ways individuals are enrolled in the registry and because they are only seen at intermittent clinic visits post-enrolment. Such features are common in observational cohort studies in general, but even in randomized trials, intermittent observation of individuals is common, resulting in incomplete information about transitions among disease states and about time-dependent covariates. In this chapter we describe independence conditions needed for valid likelihood-based inference about multistate disease processes under intermittent observation schemes. In addition, we describe how joint models for disease and observation processes can be used to address disease-related clinic visits. We also describe how joint models can be used to deal with internal time-dependent markers when marker values are observed only at clinic visits, and investigate the limiting values of regression coefficients of marker effects when the common approach of carrying forward the most recently recorded value is used. In addition to failure time processes, we consider more general multistate disease processes, and partially specified models for state occupancy probabilities. An application to a study of cancer metastatic to bone is considered, involving a serum marker of bone activity and its association with risk of skeletal complications.