Ascaso et al. (1) evaluated insulin sensitivity in 65 healthy subjects using minimal model analysis of an insulin-modified frequently sampled intravenous glucose tolerance test (FSIVGTT) to estimate the insulin sensitivity parameter derived from minimal model (SIMM) (2), as well as simple surrogates including fasting insulin, homeostasis model assessment (3), QUICKI (4), and the McAuley index (5). A statistically significant correlation was observed between SIMM and all of the other surrogate indexes examined. However, using SIMM as the reference method, the authors claim that the McAuley index has the best correlation with SIMM and the highest sensitivity and specificity (0.75 and 0.91, respectively). Therefore, the authors conclude that among several simple surrogate indexes, the McAuley index is the most sensitive and specific measure of insulin sensitivity.
While the data presented by Ascaso et al. may be sound, the conclusions drawn are not supported by the data. The main problem with the interpretation of the data comes from not including a true reference standard that is a direct measure of insulin sensitivity (e.g., euglycemic-hyperinsulinemic glucose clamp). SIMM derived from an FSIVGTT is not a direct measure of insulin sensitivity. Moreover, SIMM has well-documented systematic and random errors that are problematic even when insulin-modified FSIVGTTs are used (4,6–9). When compared with glucose clamp–derived measurements of insulin sensitivity (SIClamp), QUICKI is a substantially and significantly better estimate of insulin sensitivity than SIMM (4). Moreover, in hypertensive subjects, changes in QUICKI after therapeutic intervention are significantly correlated with changes in SIClamp, while changes in SIMM are unrelated (9). Thus, discordance between QUICKI and SIMM most likely reflects problems associated with the minimal model approach rather than inaccuracies manifested by QUICKI. Indeed, a number of independent groups have found excellent correlations between QUICKI and reference glucose clamp measurements in normal, obese, and diabetic populations (10–16), as well as in pregnant women and women with gestational diabetes (17). Because Ascaso et al. used SIMM as their reference method and they do not include a direct measure of insulin sensitivity in their analysis, it is problematic to make sound conclusions regarding the relative merits of various surrogate indexes for insulin sensitivity. An additional problem with the analysis presented by Ascaso et al. is that they do not use statistical methods to compare the differences in sensitivity and specificity among the various indexes. For example, it is uncertain whether the sensitivity and specificity of the McAuley index (0.75 and 0.91, respectively) are significantly better than that reported for QUICKI (0.65 and 0.87, respectively). In summary, when evaluating the relative merits of simple surrogate measures for insulin sensitivity, it is important to include a direct measurement of insulin sensitivity as a reference standard and to use appropriate statistical analysis so that valid conclusions may be made.