Cohort Studies with Complex Sampling Scheme and Loss to Followup
In an observational cohort study, a group of individuals (a cohort) is sampled and their subsequent status with respect to certain chronic disease are followed over time. Such studies provide information on rates of disease progression as well as the effects of associated risk factors, and multistate models can be used as a natural framework for the analysis. Outcome dependent sampling is often utilized for cohort studies, for example a prevalent cohort is obtained by selecting individuals who have been diagnosed with the disease and still alive (potentially in any of the transient disease stage) at the time of screening. It can be shown that the transition intensities in such cohorts do not coincide with those of the population. A hierarchy of conditional likelihoods relying on different assumptions about the available information can be used to deal with the problem. The observation process in the cohort studies can be complex as often the transient stages can only be monitored intermittently at clinic visits, and individuals may miss visits, make unscheduled visits due to worsening, withdrawal or die before the end of follow up. Without further investigation it may not be clear if and when either of these events occurred. We consider the design of tracing studies in which effort is made to trace individuals who have lost to follow-up and upon the contact at the end of the study their disease and survival status may be determined. Selection models are used to sample individuals for tracing depending on their history observed up to the time of last contact, and the attempt is to identify a selection model that leads to the optimal efficiency gain from the tracing efforts with respect to a particular estimate of interest. These research are motivated by the studies of psoriasis and psoriatic arthritis at a rheumatology clinic at Toronto Western Hospital.