Statistical methods for complex multi-modality kidney disease data

Abstract

The development of statistical and data science approaches is essential for the propermanagement and analysis of “Big Data” in healthcare. Motivated by the over ten years ofkidney transplantation records in Australia, we develop several statistical and data scienceapproaches that not only provide insights into method developments for complex data butalso carry promise to contribute to society by supporting policy-making, assisting decision-making and providing personalised healthcare. To comprehensively examine how differentsurvival analysis methods together with feature selection approaches perform on real-lifedatasets, we develop a benchmark framework for survival models with 320 comparisonsincluding over 20 survival approaches (classical approaches and modern machinelearning approaches) with 16 real-life datasets (clinical and omics datasets). This workcontributes to providing practical guidelines for translational scientists and clinicians,insights into developing survival analysis techniques and protocols for future benchmarkstudies. To provide personalized prediction and identify associated risk factors in aheterogeneous cohort, this thesis develops a prediction precision pathway, which is adata-driven multi step modelling framework. The key innovation lies in that it uses thesurvival model performance to guide subgroup identification. With the capacity of thisframework to be applied to any risk population, it creates a strong methodology foundationfor individualized healthcare. To examine the real-life implementation effect of allocationalgorithms, this thesis further develops an evaluation framework using a simulationalgorithm. This is the first kidney transplant allocation simulation process that providesend-to-end modeling from the arrival of recipients to shared decision-making. We highlightthe capacity of this process for policymakers in any transplant community to evaluate anyproposed allocation algorithm using in-silico simulation.