Digital Twin Workflow for Continuous Glucose Forecasting: Real-Time Data Assimilation Using Reinforcement Learning
- 1University of South Carolina
Abstract
Digital twins for physiological forecasting require robust mechanisms for continuously assimilating streaming data and adapting to patient-specific dynamics. Threshold-triggered model updates can restore predictive performance after substantial drift, but their reactive nature limits continuous adaptation. We propose a proactive digital twin framework that couples a pretrained PatchTST foundation model with an Actor-Critic reinforcement learning (RL) agent for residual correction. Using a 30-minute look-back window and recent residual history, the agent generates bounded corrections ({+/-}0.05) to the first-step baseline forecast, with corrected predictions recursively propagated over a 15- step horizon. We evaluate both an RL-only workflow and a hybrid workflow combining RL correction with threshold-triggered fine-tuning using minute-by-minute continuous glucose monitoring data from 10 critically ill patients with sepsis. Across all patients, the pretrained RL agent improves forecasting accuracy without patient-specific retraining. Both workflows reduce trajectory-averaged mean absolute percentage error, ensemble relative error, and maximum errors compared with the uncorrected baseline and previously developed reactive digital twin models. These findings support residual reinforcement learning as a lightweight mechanism for continuous, real-time data assimilation that preserves the foundation models global predictive prior while enabling patient-specific forecast adaptation, with potential applications in critical care decision support.
Related
- Continuous control deep reinforcement learning for joint resource optimization in STAR-RIS assisted mmWave-NOMA systems with hardware impairments
- Short-Term Continuous Glucose Forecasting with Large-Language-Model-Derived Nutrient Estimates from Real-World Chinese Dietary Records.
- Enhancing prenatal diagnostic workflow: a deep learning-based system for multi-class fetal ultrasound classification across all gestational stages
- Meteorological forecasting of respiratory diseases incidence with geospatial and machine learning methods
- Reinforcement learning–coupled neural ode modelling reveals ferroptotic resolution failure as a shared dynamical axis linking periodontitis and coronary atherogenesis
BibTeX
@misc{cao2026digital,
title = {{Digital Twin Workflow for Continuous Glucose Forecasting: Real-Time Data Assimilation Using Reinforcement Learning}},
author = {X. Cao and Q. Wang},
howpublished = {medRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.23.26366163},
url = {https://www.medrxiv.org/content/10.64898/2026.10.23.26366163v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - Digital Twin Workflow for Continuous Glucose Forecasting: Real-Time Data Assimilation Using Reinforcement Learning
AU - X. Cao
AU - Q. Wang
PY - 2026
DA - 2026/10/23
PB - medRxiv
DO - 10.64898/2026.10.23.26366163
UR - https://www.medrxiv.org/content/10.64898/2026.10.23.26366163v1
AB - Digital twins for physiological forecasting require robust mechanisms for continuously assimilating streaming data and adapting to patient-specific dynamics. Threshold-triggered model updates can restore predictive performance after substantial drift, but their reactive nature limits continuous adaptation. We propose a proactive digital twin framework that couples a pretrained PatchTST foundation model with an Actor-Critic reinforcement learning (RL) agent for residual correction. Using a 30-minute look-back window and recent residual history, the agent generates bounded corrections ({+/-}0.05) to the first-step baseline forecast, with corrected predictions recursively propagated over a 15- step horizon. We evaluate both an RL-only workflow and a hybrid workflow combining RL correction with threshold-triggered fine-tuning using minute-by-minute continuous glucose monitoring data from 10 critically ill patients with sepsis. Across all patients, the pretrained RL agent improves forecasting accuracy without patient-specific retraining. Both workflows reduce trajectory-averaged mean absolute percentage error, ensemble relative error, and maximum errors compared with the uncorrected baseline and previously developed reactive digital twin models. These findings support residual reinforcement learning as a lightweight mechanism for continuous, real-time data assimilation that preserves the foundation models global predictive prior while enabling patient-specific forecast adaptation, with potential applications in critical care decision support.
ER -