Developing Statistical Models to Monitor Clinical Outcomes from Large Clinical Registries at the Small-Area Level

This thesis combines advanced Bayesian spatial models with machine learning techniques to study health problems across regions and over time. Within the applied field, it examines two technical challenges in spatial statistics: the choice of geographic units (modifiable areal unit problem) and how to define neighbourhood relationships. Real-life applications included childhood anaemia in Ethiopia and mental health and quality of life post-prostate cancer surgery in Australia, identifying key social, clinical, and environmental risk factors associated with clinical outcomes. In addition, the developed tools were also used to predict Pseudomonas aeruginosa infection in cystic fibrosis patients. Findings guide better public health decisions by improving model accuracy, targeting high-risk areas, and refining research methods.