Reduced Sampling Protocols in Estimation of Insulin Sensitivity and Glucose Effectiveness Using the Minimal Model in NIDDM

Recent work in healthy subjects, the aged, and subjects with gestational diabetes or drug-induced insulin resistance using minimal model analysis of the tolbutamide-modified frequently sampled intravenous glucose tolerance test suggested that a reduced sampling regimen of 12 time points produced unbiased and generally acceptable estimates of insulin sensitivity (S_I) and glucose effectiveness (S_G) compared with a full sampling schedule of 30 time points. We have used data from 26 insulin-modified frequently sampled intravenous glucose tolerance tests in 21 subjects with NIDDM to derive and compare estimates of S_I and S_G from the full sampling schedule (S_I(30), S_G(30)) with those estimated from the suggested 12 time points (S_I(12), S_G(12)) and those estimated with the addition of a 25-min time point (S_I(13), S_G(13)). Percentage relative errors were calculated relative to the corresponding 30 time-point values. A statistically significant bias of 15% (97% confidence interval from 7.4 to 25.6%, interquartile range 25%) was introduced by the estimation of S_I(12) but not S_I(13) (1%, 97% confidence interval from –9.4 to 9.3%, interquartile range 21%). Results for S_G(12) (–12%, 97% confidence interval from –46.7 to 1.2%, interquartile range 49%) and S_G(13) (–5%, 97% confidence interval from –27.8 to 6.8%, interquartile range 37%) were statistically equivocal. The precision of estimation of S_I(12), S_G(12), and S_G(13) measured by the interquartile range of the percentage relative errors was poor. The precision of determination measured by the median minimal model coefficient of variation was 18, 29, and 27% for S_I(30), S_I(12), and S_I(13) and 9, 11, and 11% for S_G(30), S_G(12), and S_G(13), respectively. Thus, the application of minimal model analysis to the 12 time-point protocol of the insulin-modified IVGTT for the estimation of S_I and S_G in NIDDM may necessitate an inordinately large number of subjects. Although the 13 time-point protocol may be more acceptable for the assessment of S_G in population studies, we recommend retention of the full sampling schedule where feasible.