Tail-Risk-Aware Model Predictive Control for Hypoglycemia Mitigation in Automated Insulin Delivery
- 1Faculty of Computing, Harbin Institute of Technology, Harbin 150001, China
- 2Medical College, Qinghai University, Xining 810016, China
Abstract
Background: Automated insulin delivery must maintain glycemic control despite partial observability, physiological heterogeneity, and sensing and actuation errors. Cohort-average metrics may nevertheless conceal hypoglycemia in vulnerable virtual patients. Methods: We developed a tail-risk-aware model predictive control framework that combines a Transformer-based latent world model, unguided iterative cross-entropy method planning over perturbed rollouts, and a deterministic Safety Layer that projects insulin actions onto programmed bounds. Reinforcement-learning-guided proposal sampling was evaluated separately as an exploratory search variant. The primary controller was evaluated in Simglucose using 10 virtual patients, three random seeds, and five prespecified operating conditions. Results: Under nominal conditions, it achieved a cohort-average time in range (TIR) of 77.6% and time below range (TBR) of 3.8%, but worst-k TBR was 11.0% and increased to 37.4% under 5% random CGM missingness. In the risk-objective ablation, combining variance penalization with TailMean increased TIR by 2.8 percentage points and reduced cohort-average and worst-k TBR by 2.3 and 7.4 percentage points, respectively, relative to mean-only planning. The Safety Layer modified every proposed action in the analyzed run, indicating strong dependence on the programmed bounds. Conclusions: These results show that average target attainment can coexist with substantial patient-level risk. They support tail-aware evaluation and identify patient-specific safeguards as a priority for future validation, but do not establish clinical safety or efficacy.
Related
- The Impact of Bias Mitigation on Fairness and Accuracy in Automated Skin Lesion Classification
- Continuous glucose monitoring, automated insulin delivery, and remote care in adolescents with type 1 diabetes: advances toward integrated digital management
- An international position statement on practical approaches for inpatient continuous glucose monitoring, insulin pumps and automated insulin delivery systems in adults
- Extended Use of the Omnipod 5 Automated Insulin Delivery System in Adults With Type 1 Diabetes: 12‐Month Extension of a Randomized Controlled Trial
- Cost‐effectiveness analysis comparing advanced hybrid closed‐loop automated insulin delivery with multiple daily insulin injections and continuous glucose monitoring in adults with type 1 diabetes in Japan
BibTeX
@article{zhang2026tailriskaware,
title = {{Tail-Risk-Aware Model Predictive Control for Hypoglycemia Mitigation in Automated Insulin Delivery}},
author = {Kaiyuan Zhang and Congyu Han and Hexing Guo and Shuo Kang and Shuji Jing and Chunrui Wang and Tianyi Zang and Yanli Zhao},
journal = {Metabolites},
year = {2026},
month = oct,
doi = {10.3390/metabo16100759},
url = {https://www.mdpi.com/2218-1989/16/10/759},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Tail-Risk-Aware Model Predictive Control for Hypoglycemia Mitigation in Automated Insulin Delivery
AU - Kaiyuan Zhang
AU - Congyu Han
AU - Hexing Guo
AU - Shuo Kang
AU - Shuji Jing
AU - Chunrui Wang
AU - Tianyi Zang
AU - Yanli Zhao
PY - 2026
DA - 2026/10/01
JO - Metabolites
DO - 10.3390/metabo16100759
UR - https://www.mdpi.com/2218-1989/16/10/759
AB - Background: Automated insulin delivery must maintain glycemic control despite partial observability, physiological heterogeneity, and sensing and actuation errors. Cohort-average metrics may nevertheless conceal hypoglycemia in vulnerable virtual patients. Methods: We developed a tail-risk-aware model predictive control framework that combines a Transformer-based latent world model, unguided iterative cross-entropy method planning over perturbed rollouts, and a deterministic Safety Layer that projects insulin actions onto programmed bounds. Reinforcement-learning-guided proposal sampling was evaluated separately as an exploratory search variant. The primary controller was evaluated in Simglucose using 10 virtual patients, three random seeds, and five prespecified operating conditions. Results: Under nominal conditions, it achieved a cohort-average time in range (TIR) of 77.6% and time below range (TBR) of 3.8%, but worst-k TBR was 11.0% and increased to 37.4% under 5% random CGM missingness. In the risk-objective ablation, combining variance penalization with TailMean increased TIR by 2.8 percentage points and reduced cohort-average and worst-k TBR by 2.3 and 7.4 percentage points, respectively, relative to mean-only planning. The Safety Layer modified every proposed action in the analyzed run, indicating strong dependence on the programmed bounds. Conclusions: These results show that average target attainment can coexist with substantial patient-level risk. They support tail-aware evaluation and identify patient-specific safeguards as a priority for future validation, but do not establish clinical safety or efficacy.
SN - 2218-1989
ER -