Keywords: latent class growth analysis, hidden heterogeneity, statistical modelling for precision medicine, statistical classification

"Hidden Depths: Robustness of modelling approaches for uncovering latent classes in longitudinal data"

This thesis investigated the performance of a class of statistical models, known as finite mixture models, in uncovering hidden classes of subjects following distinct temporal patterns of development. Such models have great utility when a grouping variable is either unknown (such as true disease diagnosis given clinical measurements) or is expensive to measure (a rare genetic marker given an observable trait). By studying potential data conditions and statistical model specifications which could influence the accuracy of uncovering these hidden groups, this thesis strove to increase the applicability of such models in applied research and precision medicine. This is important as the proper identification of hidden classes of temporal development could assist practitioners in the early diagnosis of disease and/or treatments tailored for the individual. To this end, practical guidelines for practitioners to properly employ these models in their research are developed.

Click here for the live stream.

Language: English