Validation of the UVA Simulation Replay Methodology Using Clinical Data: Reproducing a Randomized Clinical Trial

Area of Science:

• Biomedical Engineering
• Computational Biology
• Endocrinology

Background:

• Computer simulators are vital for developing and testing diabetes treatments.
• Existing simulators often lack the ability to replicate diverse patient behaviors and glycemic conditions.
• Replay methodologies use real-world data to create more realistic simulation environments.

Purpose of the Study:

• To validate a replay methodology by demonstrating its capacity to reproduce a recent clinical trial comparing hybrid closed loop (HCL) and fully closed loop (FCL) automated insulin delivery (AID).
• To assess the accuracy of simulated glycemic outcomes against observed data for key glycemic metrics.

Main Methods:

• Generated an ensemble of replay simulators using data from a randomized crossover clinical trial.
• Created 64 subject/modality pairs, exposing each virtual subject to the alternate AID modality.
• Performed equivalence tests for time in range (TIR), time below range (TBR), time above range (TAR), high blood glucose index (HBGI), and low blood glucose index (LBGI).

Main Results:

• Time in range (TIR), time above range (TAR), low blood glucose index (LBGI), and high blood glucose index (HBGI) showed statistical and clinical equivalence between simulated and observed data.
• TBR failed the equivalence test.
• For example, simulated TIR in HCL mode was 84.89% vs. observed 84.31% (P = 0.0170).

Conclusions:

• The UVA replay method accurately predicts the glycemic impact of insulin treatments, as confirmed by clinical trial data.
• This in vivo validation supports the use of the replay method for personalizing and adapting type 1 diabetes treatment strategies.