Development of a New Model of Insulin Sensitivity in Patients With Type 2 Diabetes and Association With Mortality

Abstract

Abstract Context Despite being one of the major drivers of diabetes incidence, the degree of insulin resistance in patients with type 2 diabetes (T2D) is not usually evaluated in clinical practice or in large epidemiologic studies. Objective To identify a model of insulin sensitivity using widely available clinical and laboratory parameters in patients with T2D and evaluate its association with all-cause and cardiovascular mortality. Methods One hundred forty patients with T2D underwent a euglycemic hyperinsulinemic clamp to measure total body glucose disposal rate (mg kg−1 minute−1). We used demographic, clinical, and common laboratory parameters to estimate insulin sensitivity (IS) via stepwise linear regression on 85 patients (training cohort) and validated it in the remaining 55 (validation cohort). The identified equation was then applied to 3553 patients with T2D from the 1999-2010 cycles of the National Health and Nutrition Examination Survey (NHANES) to evaluate its association with all-cause and cardiovascular mortality up to December 2015. Results The best model included triglycerides, gamma glutamyl transpeptidase, albumin excretion rate, and body mass index. The identified IS score correlated well with the clamp-derived glucose disposal rate in both the training (r = 0.77, P < .001) and the validation (r = 0.74, P < .001) cohorts. In the NHANES cohort, after a median follow-up of 8.3 years, 1054 patients died, 265 of cardiovascular causes. In a multivariable Cox proportional hazard model adjusted for age, sex, race-ethnicity, education, cigarette smoke, total cholesterol, chronic kidney disease, blood pressure, prevalent cardiovascular disease, and alcohol consumption, a higher estimated IS was associated with a lower risk of both all-cause and cardiovascular mortality. Conclusion We propose a new model of IS in patients with T2D based on readily available clinical and laboratory data. Its potential applications are in both diagnosis as well as prognostication.