Characterising disease heterogeneity in Crohn's disease and ulcerative colitis: leveraging longitudinal biomarker profiles

Abstract

Inflammatory bowel disease, primarily consisting of Crohn’s disease and ulcerative colitis, is highly heterogenous with much of this heterogeneity being unexplained. In this thesis, I explore different components of this heterogeneity and propose a novel approach to its characterisation. I demonstrate the heterogeneity observed in treatment response by conducting a meta-analysis of the real-world effectiveness and safety of tofacitinib, a small molecule therapeutic approved for the treatment of ulcerative colitis. Tofacitinib is found to be effective and safe; 40% of patients exhibited a clinical response and 29% were in remission 16–26 weeks after treatment commencement. To improve the characterisation of unexplained heterogeneity, I propose the inflammatory bowel disease population likely consists of homogeneous subgroups defined by different disease trajectories which are characterised by distinctive biomarker profiles. As a proof-of-concept, I explore characterising heterogeneity in Crohn’s disease using longitudinal measurements of faecal calprotectin, a biomarker of gastrointestinal inflammation. I find and describe four distinct clusters. Cluster membership is found to be significantly associated with smoking (p = 0.015) and early biologic therapy (p < .001) and is unable to be independently predicted using traditional clinical approaches to disease classification. Whilst I find the approach feasible, the study cohort in the pilot study was restricted in size (n = 356) and susceptible to inclusion bias. I then apply the approach to all inflammatory bowel disease patients treated by NHS Lothian meeting less stringent criteria than previously used (n = 1343) and consider both faecal calprotectin and C-reactive protein observations. I compare alternative model specifications to determine an optimal approach. I discuss the challenges faced during this work such as data availability and the distribution of faecal calprotectin data which violate the assumptions of common modelling techniques. This work has future potential in predictive models which assign the risk of a patient experiencing a poor outcome.