A Bayesian Framework to Identify Type 1 Diabetes Physiological Models Using Easily Accessible Patient Data.

Area of Science:

• Biomedical Engineering
• Computational Physiology
• Endocrinology

Background:

• Mathematical models are crucial for developing glucose control algorithms in type 1 diabetes (T1D).
• Individualized model parameterization is challenging due to identifiability issues.
• Bayesian techniques and clinical data have enabled model fitting for T1D patients.

Purpose of the Study:

• To develop a methodology for estimating T1D physiological model parameters using readily available data.
• Utilize continuous glucose monitoring (CGM), carbohydrate intake (CHO), and exogenous insulin (I) data.
• Address identifiability problems in T1D glucose-insulin dynamics modeling.

Main Methods:

• Employed a Bayesian approach with Markov Chain Monte Carlo (MCMC) for parameter estimation.
• Leveraged a priori knowledge from literature to constrain parameter uncertainty.
• Tested the methodology on synthetic data from 100 virtual T1D patients.

Main Results:

• Demonstrated good model fit and acceptable precision for parameter estimates.
• Successfully reconstructed non-accessible glucose-insulin fluxes, including glucose rate of appearance and plasma insulin.
• Validated the methodology's effectiveness on synthetic patient data.

Conclusions:

• The developed methodology enables parameter estimation for T1D physiological models using accessible CGM, CHO, and I data.
• This approach offers a promising framework for personalized diabetes management and algorithm development.
• Further research and assessment in more complex scenarios are warranted.