Clinical indexes of insulin resistance (IR) have acquired increasing importance with the development of various drugs that improve endogenous insulin action (1). Recently, the largest database on insulin clamp studies has been established. This database includes 2,321 subjects, of whom 2,138 are nondiabetic (92%), from 19 sites worldwide (2). Using classification trees, three models have been derived. Model 1 is based on homeostasis model assessment of insulin resistance (HOMA-IR) 4.65, BMI 28.9 kg/m2, or HOMA-IR 3.60 and BMI 27.5 kg/m2. Model 2 is based on BMI 28.7 kg/m2, or BMI 27.0 kg/m2 and a positive family history of diabetes. Model 3 is based on BMI 28.7 kg/m2, BMI 27.0 kg/m2 and a positive diabetes family history, or triglycerides 2.44 mmol/l and a negative family history of diabetes. These three models should all accurately identify insulin-resistant individuals (2). We have evaluated the prevalence and characteristics of subjects with IR based on these models using data from the KORA Survey 2000, an oral glucose tolerance test (OGTT)-based, populationbased survey in Germany (n 1,352 individuals aged 55–74 years without previously known diabetes)(3). In the KORA Survey, proportions (95% CI) with IR were 47.4%(44.7–50.1), 45.8%(43.1–48.5), and 49.1%(46.4–51.8) for models 1, 2, and 3, respectively. Agreement of the models was high (coefficients 0.78–0.94). Although HOMA-IR significantly increased with worsening glucose tolerance (geometric means [SDF]: normal glucose tolerance 2.17 [1.83], impaired glucose tolerance [IGT] 3.39 [2.12], and newly diagnosed diabetes 4.67 [2.16]; all P 0.05), the sensitivities (0.67, 0.60, and 0.64 for models 1, 2, and 3, respectively) and specificities (0.60, 0.59, and 0.56) of the IR models for detecting IGT or diabetes were only moderate. Model 1 included

HOMA-IR as a surrogate measure of IR. In multiple age-sex–adjusted logistic regression including all three models, model 1 (odds ratio 4.3, 95% CI 2.8–6.8) was more closely related to IGT/diabetes (dependent variable) than the others (model 2: 0.5, 0.2–1.1; model 3: 1.6, 0.7–3.4). Overall, about one-third of the subjects with IGT or previously undiagnosed diabetes would not have been included when applying the proposed rules for identifying individuals with IR in our elderly population. This may indicate a limited diagnostic validity of the IR models, because a large body of evidence shows that IGT and type 2 diabetes are characterized by moderate-to-severe IR (4). On the other hand, defective insulin secretion rather than IR may be present in some subjects with IGT and type 2 diabetes (4). This could partly explain the low sensitivities of the models to detect glucose disorders in our population. In conclusion, the recently proposed IR models need to be further validated, using measures of-cell function across the whole range of glucose intolerance, before they should be incorporated into clinical trials and clinical practice (2).