Grant: Operating Grant

Many complex diseases are of a dynamic nature and understanding them often require of expensive and time consuming designs that collect phenotype data as well as other relevant repeated measurements on a group of individuals. Modelling data from these cohort studies requires to consider at least two kinds of dependence: i) individuals may have common ancestry and/or shared environment; and ii) the correlated nature of repeated measures. Ignoring the dependence structure or replacing it with a simplified 'working correlation' for computational purposes could greatly preclude the advantages conferred by the cohort design and even the validity of the inference. In spite of this, most studies of longitudinal traits use over-simplified models that fail not only to capture the dynamics of the disease but also may yield misleading results regarding the nature and extent of genetic effects. Furthermore, in longitudinal studies there are events that happen (like death, drop-out or other milestone stage) that cannot be ignored in the analysis because these events affect data collection and the longitudinal outcomes are likely to affect the time on which these events happen. Thus, the genetic analysis of data from cohorts requires of the development of methods for models able to deal with these issues. Although no in a human genetics context, joint modelling of continuous longitudinal outcomes and time-to-event has received considerable attention in the last decade. However, methods considering discrete longitudinal outcomes, quite common in medical studies, are almost absent in the statistical literature. The objectives of this proposal are to develop methods for genetic models of longitudinal and event history data that use the true dependence structure of the data under study and are computationally sound. Particular attention will be given to models for discrete longitudinal traits. These methods will be of immediate utility in the analysis of on-going medical research.