Reliability of error estimates from the minimal model: implications for measurements in physiological studies

MINMOD provides an estimate of the error in the insulin sensitivity index (SI) and glucose effectiveness at basal insulin (Sg) as the fractional standard deviation (FSD). The validity of the FSD estimate has not been assessed in a large number of human studies, nor has a comparison of the accuracies achievable using the two different intravenous glucose tolerance test (IVGTT) protocols (glucose only and tolbutamide protocol) been performed. To address these two issues, we obtained the FSD value and performed Monte Carlo simulations for 237 IVGTT studies. The FSD underestimated the true error as determined as the coefficient of variation from Monte Carlo simulation (COV-MC) with the ratio of COV-MC to FSD being 3.07 +/- 0.20 (mean +/- SE) for SI using the tolbutamide protocol. Additionally, the mean COV-MC for glucose-only protocol was approximately two to three times that for the tolbutamide protocol for both SI and Sg. We conclude that FSD underestimates the true error in SI and Sg. Additionally, more accurate results are obtained from the tolbutamide protocol than with the glucose-only protocol.