Minimal model S(I)=0 problem in NIDDM subjects: nonzero Bayesian estimates with credible confidence intervals

Area of Science:

• Metabolic research
• Biomedical modeling
• Statistical analysis

Background:

• The minimal model and insulin-modified intravenous glucose tolerance test are standard for estimating insulin sensitivity (S(I)).
• Nonlinear least squares (NLS) is commonly used for parameter estimation, providing point estimates and standard deviations.
• NLS faces challenges with type 2 diabetic subjects, sometimes yielding S(I)=0 or unrealistic confidence intervals.

Purpose of the Study:

• To implement Bayesian parameter estimation via Markov chain Monte Carlo (MCMC) to address NLS limitations in S(I) estimation.
• To provide a more robust method for calculating S(I) and its confidence intervals in individuals with type 2 diabetes.

Main Methods:

• Bayesian parameter estimation using a Markov chain Monte Carlo (MCMC) method.
• Application to the minimal model of glucose kinetics and insulin-modified intravenous glucose tolerance test data.
• Comparison of MCMC results with traditional Nonlinear Least Squares (NLS) methods.

Main Results:

• Bayesian estimation using MCMC successfully determined a non-zero S(I) point estimate with a credible confidence interval in all subjects.
• NLS methods failed or produced unacceptable results in 40% of the studied type 2 diabetic subjects.
• MCMC provides a posteriori probability density function for S(I), allowing for more comprehensive uncertainty quantification.

Conclusions:

• Bayesian parameter estimation with MCMC offers a superior alternative to NLS for estimating insulin sensitivity in type 2 diabetes.
• This approach reliably quantifies S(I) and its uncertainty, overcoming critical limitations of conventional methods.
• The MCMC approach is recommended for reanalyzing large epidemiological datasets involving insulin sensitivity.