GlyMan: Glycemic Management using Patient-Centric Counterfactuals

Prolonged and frequent exposure to elevated blood glucose levels (hyperglycemia) significantly increases the likelihood of developing chronic complications, such as neuropathy, nephropathy, and cardiovascular disease, along with acute symptoms like fatigue and blurry vision. While current technologies, such as continuous subcutaneous insulin infusion (CSII) and continuous glucose monitors (CGMs), can forecast adverse events like hypoglycemia and deliver small insulin doses to counteract hyperglycemia, progress in developing tailored AI-driven interventions remains limited, which poses a barrier to optimal diabetes care. To address this gap, we propose leveraging counterfactual explanations that guide patients in making targeted adjustments to their carbohydrate intake and insulin dosing to avoid abnormal glucose levels. We introduce GlyMan<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">4</sup>, a novel method that generates counterfactual behavioral recommendations aimed at helping patients and caregivers make small, informed changes to prevent hyperglycemia, thus substantially reducing both its frequency and duration. Additionally, GlyMan incorporates user preferences into its intervention process and ensures more customized and patient-centered guidance. We rigorously evaluated GlyMan using real-world data from 21 type 1 diabetes (T1D) patients using automated insulin delivery (AID) systems. Results indicate that GlyMan surpasses existing methods, delivering 76.6% valid explanations and 86% effectiveness when assessed against historical data.