Assuring AI safety for managing risk of myocardial infarction for patients with type 2 diabetes

Type 2 Diabetes (T2D) is a highly prevalent health condition, affecting hundreds of millions of people worldwide. As this health condition progresses, it causes various serious comorbidities. One of the critical T2D-related comorbidities is Myocardial Infarction (MI), which is also known as a heart attack. However, with timely and accurate interventions, the development of T2D-related MI can be prevented. Different methods have been used to support the management of T2D-related MI, including the use of Artificial Intelligence (AI) to predict the risk of MI development. While AI-based methods have shown promising results, most of the existing research has focused primarily on improving the prediction performance of the models. However, in safety-critical domains like healthcare, the performance of AI models alone is not sufficient. These systems also need to be developed by considering safety to prevent harmful clinical outcomes. This thesis makes novel contributions by embedding AI safety systematically across the modelling pipeline and by developing a holistic safety case for T2D-related MI prediction using a large-scale real-world healthcare dataset. Clinical hazards were proactively identified and directly translated into safety requirements, ensuring that risk considerations were addressed in the entire process. These requirements were embedded within a structured safety case that combined SHARD, Bow-tie, and Goal Structuring Notation (GSN). Importantly, the safety case was not treated as external documentation but actively shaped design and modelling choices, guiding the application of class imbalance handling, ensemble learning, model optimization, and explainability methods. Therefore, this research addresses a critical gap in clinical AI by linking hazard analysis, safety assurance, and model optimisation within a unified methodology. Beyond T2D-related MI, this research establishes a pathway that can inform the development of safe and predictive AI in healthcare, highlighting its potential transferability to other clinical domains.