Giorgi, EmanueleAlqahtani, Nouf2026-01-042025Dickson, E., Grambsch, P., Fleming, T., Fisher, L., & Langworthy, A. (1989). Cirrhosis Patient Survival Prediction [Dataset]. UCI Machine Learning Repository. https://doi.org/10.24432/C5R02Ghttps://hdl.handle.net/20.500.14154/77785Primary biliary cholangitis (PBC) is a chronic autoimmune liver disease with a highly variable clinical course, making accurate survival prediction essential for patient management and transplant decision-making. This dissertation presents a re-analysis of the Mayo Clinic PBC cohort, a landmark dataset that has underpinned prognostic modelling for several decades, with the aim of developing a robust and interpretable survival model while addressing methodological limitations identified in the literature. The analysis is based on 418 patients with long-term follow-up and detailed baseline demographic, clinical, and biochemical measurements. Missing data mechanisms were formally assessed, and appropriate handling strategies were implemented prior to model fitting. Cox proportional hazards models were fitted and rigorously evaluated, with particular attention to violations of the proportional hazards assumption. Where necessary, stratified Cox models were employed to accommodate non-proportional effects while preserving interpretability. Model performance was assessed using discrimination, calibration, and prediction error metrics, and alternative parametric survival models were explored as sensitivity checks. The results demonstrate that stratification substantially improves model adequacy without compromising clinical relevance, and that key biochemical markers remain strong predictors of mortality. Overall, this study highlights the continued value of carefully specified classical survival models and emphasises the importance of assumption checking and validation when developing prognostic tools for chronic liver disease.59enPrimary biliary cholangitis (PBC)Survival analysisCox proportional hazards modelStratified Cox regressionPrognostic modellingMissing data imputationInternal validationModel calibration and discriminationClinical risk predictionMayo Clinic PBC datasetThesis