Fasting-based estimates of insulin sensitivity in overweight and obesity: A critical appraisal

Objective: To identify simple methods to estimate the degree of insulin resistance. Research Methods and Procedures: The performance of a wide range of fasting-based index estimates of insulin sensitivity was compared by receiver operating characteristic analysis (area under curves and their 95% confidence intervals) against the M value from euglycemic insulin clamp studies collected in the San Antonio (non-Hispanic whites and Hispanic residents of San Antonio, TX) and European Group for the Study of Insulin Resistance (non-diabetic white Europeans) databases (n = 638). Results: Insulin resistance differed substantially between lean (BMI < 25 kg/m(2)), overweight or obese (BMI >= 25 kg/m(2)), and type 2 diabetic individuals. Estimates of insulin resistance were, therefore, assessed in each group separately. In the overweight and obese subgroup (n = 302), the receiver operating characteristic performance of fasting-based indices varied from 0.72 (0.62 to 0.82), in the case of the insulin/glucose ratio, to 0.80 (0.72 to 0.88) in the case of Belfiore free fatty acids. One superior method could not be identified; the confidence intervals overlapped, and no statistically significant differences emerged. All indices performed better when using the whole study population, with fasting plasma insulin, homeostatic model assessment insulin/glucose ratio, quantitative insulin sensitivity check index, glucose/insulin ratio, Belfiore glycemia, revised quantitative insulin sensitivity check index, McAuley index, and Belfiore free fatty acids showing area under curves of 0.83, 0.90, 0.66, 0.90, 0.66, 0.90, 0.85, 0.83, and 0.86, respectively, because of the inclusion of very insulin sensitive (lean) and very insulin resistant cases (diabetic subjects). Discussion: In conclusion, a superior fasting-based index estimate to distinguish between the presence and absence of insulin resistance in overweight and obesity could not be identified despite the use of the large datasets.