Building trajectories through clinical data to model disease progression

Abstract

Clinical trials are typically conducted over a population within a defined time period in order to illuminate certain characteristics of a health issue or disease process. These cross-sectional studies provide a snapshot of these disease processes over a large number of people but do not allow us to model the temporal nature of disease, which is essential for modeling detailed prognostic predictions. Longitudinal studies, on the other hand, are used to explore how these processes develop over time in a number of people but can be expensive and time-consuming, and many studies only cover a relatively small window within the disease process. This thesis describes the application of intelligent data analysis techniques for extracting information from time series generated by different diseases. The aim of this thesis is to identify intermediate stages in a disease process and sub-categories of the disease exhibiting subtly different symptoms. It explores the use of a bootstrap technique that fits trajectories through the data generating “pseudo time-series”. It addresses issues including: how clinical variables interact as a disease progresses along the trajectories in the data; and how to automatically identify different disease states along these trajectories, as well as the transitions between them. The thesis documents how reliable time-series models can be created from large amounts of historical cross-sectional data and a novel relabling/latent variable approach has enabled the exploration of the temporal nature of disease progression. The proposed algorithms are tested extensively on simulated data and on three real clinical datasets. Finally, a study is carried out to explore whether we can “calibrate” pseudo time-series models with real longitudinal data in order to improve them. Plausible directions for future research are discussed at the end of the thesis.