Big data to study big questions: the application of big data and machine learning techniques perform broad-scale rheumatology studies

Abstract

This thesis investigates the substantial heterogeneity present across rheumatology patient populations by integrating electronic health records, genetic data, and cohort studies through advanced machine learning methodologies. Rheumatoid arthritis (RA), a chronic autoimmune disease characterised by synovial inflammation, systemic manifestations, and marked variation in serological status, comorbidities, and treatment response, forms a central focus. Genetic predisposition—particularly HLA Shared Epitope alleles—environmental factors such as smoking, and emerging influences including the microbiome contribute to RA risk and progression.

Large scale datasets, including ICD/PheCode based EHR phenotypes, whole genome sequencing, and inception cohorts such as Clinically Suspect Arthralgia and Systemic Sclerosis care pathways, enable detailed exploration of disease mechanisms and patient stratification. Machine learning is employed for prediction, hypothesis generation, cohort construction, and identification of latent clinical structure, demonstrating its value in processing high dimensional medical data.

Across three interconnected parts, the thesis develops frameworks for EHR deconvolution, examines genetic architecture and gene–environment interactions, and demonstrates clinical applications of machine learning for personalised risk assessment and healthcare optimisation within rheumatology.