P-163: Profile, challenges, expectations and attitudes toward emerging technology in patients with type 1 diabetes mellitus and in-silico trials
E Malesiou, S Lekka, I Kosteria, E Dikaiakou, E -A Vlachopapadopoulou, I EmirisAbstract
Introduction
Managing type 1 diabetes mellitus (T1DM) with continuous subcutaneous insulin infusion (CSII) and continuous glucose monitoring (CGM) imposes a significant daily burden on young patients and their caregivers. Real-world glycemic outcomes, technology attitudes, and the potential of reinforcement learning models (RLMs) for personalized insulin delivery support remain incompletely explored in this population.
Objectives
To evaluate glycemic control, management burden, and technology attitudes in youth with T1DM on CSII/CGM, and to develop and evaluate RLMs for automated insulin delivery.
Methods
TSeventeen pediatric patients (12 male, 5 female; mean age 12.0±5.1 years; disease duration 3.7±0.68 years) contributed up to 8 weeks of device-derived data and were asked to complete a questionnaire. In-silico simulations across age-stratified meal scenarios were performed for RLM training and evaluation.
Results
Device-derived glycemic metrics: TIR 74.5%, TBR 2%, TAR 24%, mean glucose 150 mg/dL, HbA1c 6.9±0.3%. Management was rated challenging on 38% of days, with 149 min/day devoted to glycemic decisions. CGM was rated 9.4/10 in importance but only 7.4/10 in accuracy confidence. Despite an attainable TIR of 89%, 60% accepted sub-optimal glycemic outcomes. Technology improved quality of life (4.25/5). In silico, age-stratified subgroups (children/adolescents/adults) proved sensitive to bolus timing delays while hyperglycemia risk and TBR were primary differentiating features. Risk index-based reward functions outperformed alternatives; additional decision constraints modulated RLM efficacy
Conclusions
A meaningful gap exists between achievable and accepted glycemic targets. Integrating risk-sensitive RL models into a safe, automated framework is essential to reduce the management burden and support clinically realistic improvements.