Scientific Rationale:
Type 2 Diabetes is a chronic condition requiring personalized treatment adjustments. Existing rule-based approaches lack adaptability to individual patients’ biomedical and lifestyle changes.
Reinforcement Learning (RL) is a promising AI approach for sequential decision-making, but current models struggle with long-term patient outcomes. Hierarchical RL (HRL) improves upon this by incorporating short-term interventions (e.g., medication, exercise) and long-term health goals (e.g., HbA1c reduction, complication prevention).
UK Biobank provides rich longitudinal data (biometrics, lifestyle, medications, lab results) essential for training and validating an HRL-based personalized treatment model. This research will help develop an AI-driven system to assist clinicians in optimizing T2D treatment plans, ultimately improving patient health outcomes.
Research Questions:
1. How can machine learning models be used to optimise both short-term and long-term decision making in Type 2 diabetes management using longitudinal health data?
2. Which features from longitudinal datasets (e.g., UK Biobank) are critical for supporting personalised decision-making in Type 2 diabetes management?
3. Can machine learning models perform better compared to standard clinical approaches in managing both glycaemic control and long-term outcomes, and how well do they generalise across diverse patient subgroups?
4. How can machine learning models be made interpretable and actionable for healthcare providers to ensure effective real-world application in Type 2 diabetes management