Modeling the oral glucose tolerance test in normal and impaired glucose tolerant states: a population approach

The conventional approach to analyzing data from oral glucose tolerance testing (OGTT) requires model identification in each individual separately (standard two stage, STS), ignoring knowledge about the population as a whole. In practice, however, the OGTT is sparsely sampled and individual estimates are often not resolvable from available data. This weakness is often encountered in large scale trials or epidemiological studies, leading to either multiple imputations or simply much less data available for analysis.

We have applied a population approach, nonlinear mixed effects modeling, to plasma glucose, insulin and C-peptide data obtained from a 120 minute OGTT undertaken by 106 subjects with varying glucose tolerance. This method provides estimates of population means, variances and covariances of model parameters and empirical Bayes estimates of individual parameter values, as well as measures of intra-individual (within-subject) and inter-individual (between-subject) variability. The recently developed oral glucose minimal model was used to evaluate insulin sensitivity, and a combined model approach was used to assess β-cell secretion.

Applying these models allowed for the reconstruction of insulin secretion and glucose absorption profiles and gave population indexes of insulin sensitivity (S_I = 6.51 ± 1.20 × 10^-4 min^-1·μU^-1·ml), fractional hepatic extraction of insulin (F = 0.522 ± 0.291) and fractional insulin clearance (k_I = 0.258 ± 0.151 min^-1). Whereas the traditional approach to parameter estimation failed to recover estimates in more than one third of the population, the population approach provided individual estimates in all subjects. Examination of the empirical Bayes estimates showed that individual parameter estimates were able to differentiate well between individuals at glucose tolerant states ranging from euglycemia to overt type 2 diabetes.

Our findings suggest that population analysis is a powerful tool for obtaining accurate assessments of indexes of insulin sensitivity and β-cell function from the OGTT, especially in epidemiological studies with large numbers of sparsely sampled subjects.