Population and individual minimal modeling of the frequently sampled insulin-modified intravenous glucose tolerance test.

Population approaches are more robust estimators of insulin sensitivity (SI) and glucose effectiveness (SG) with the minimal model of glucose kinetics during an intravenous glucose tolerance test (IVGTT). We assessed the performance of 3 population methods, iterative two-stage (ITS), Bayesian hierarchical Markov chain Monte Carlo (MCMC), and NONMEM first-order conditional estimation (FOCE) with interaction (NM), and made a comparison with the standard two-stage method (STS) employing the weighted nonlinear regression analysis. To evaluate accuracy of individual and population estimates, 40 simulated insulin-modified frequently sampled IVGTTs (IM-FSIVGTT) were derived from real IM-FSIVGTTs (0.3 g glucose per kg body weight with 0.02 U/kg insulin at 20 minutes; 30 samples over 180 minutes) performed in 40 healthy Caucasian subjects (male/female, 22/18; age, 46 +/- 9 years; body mass index [BMI], 26.7 +/- 5.7 kg. m(-2); mean +/- SD). The population methods assumed a log-normal population distribution of parameters. All methods gave a similar but overestimated population SG by 9% to 13%. Population SI was underestimated to a different degree by the methods (STS 6%, ITS 10%, MCMC 13%, and NM 7%). The between-subject variability of SG was overestimated by STS and underestimated by the population methods (true 33%, STS 40%, ITS 19%, MCMC 24%, NM 24%; coefficient of variation). For SI, this quantity was well estimated by all methods (true 79%, STS 80%, ITS 82%, MCMC 83%, NM 82%). The results for individual estimates indicate that STS performs better than the population methods when estimating SI (STS 12%, ITS 16%, MCMC 16%, NM 16%; 1 outlying subject excluded; root mean squared error expressed as percent of mean) but worse for SG (STS 28%, ITS 21%, MCMC 20%, NM 19%). We conclude that the robust performance of population approaches, preventing parameter estimation failures associated with the nonlinear regression analysis, is not required with IM-FSIVGTT in subjects with normal glucose tolerance. The standard two-stage technique is the preferred method under such circumstances.