Biomarkers of Insulin Resistance and Their Performance as Predictors of Treatment Response in Overweight Adults

Abstract

Context: Insulin resistance (IR) contributes to the pathogenesis of type 2 diabetes mellitus and is a risk factor for cardiovascular and neurodegenerative diseases. Amino acid and lipid metabolomic biomarkers associate with future type 2 diabetes mellitus risk in several epidemiological cohorts. Whether these biomarkers can accurately monitor changes in IR status following treatment is unclear.

Objective: Herein we evaluated the performance of clinical and metabolomic biomarker models to forecast altered IR, following lifestyle-based interventions.

Design: We contrasted the performance of two distinct insulin assay types (high-sensitivity ELISA and immunoassay) and built IR diagnostic models using cross-sectional clinical and metabolomic data. These models were used to stratify IR status in preintervention fasting samples, from 3 independent cohorts (META-PREDICT (n = 179), STRRIDE-AT/RT (n = 116), and STRRIDE-PD (n = 149)). Linear and Bayesian projective prediction strategies were used to evaluate models for fasting insulin and homeostatic model assessment 2 for insulin resistance and change in fasting insulin with treatment.

Results: Both insulin assays accurately quantified international standard insulin (R2 > 0.99), yet agreement between fasting insulins was less congruent (R2 = 0.65). A mean treatment effect on fasting insulin was only detectable using the ELISA. Clinical-metabolomic models were statistically related to fasting insulin (R2 0.33-0.39) but with modest capacity to classify IR at a clinically relevant homeostatic model assessment 2 for insulin resistance threshold. Furthermore, no model predicted treatment responses in any cohort.

Conclusion: We demonstrate that the choice of insulin assay is critical when quantifying the influence of treatment on fasting insulin, whereas none of the clinical-metabolomic biomarkers, identified in cross-sectional studies, are suitable for monitoring longitudinally changes in IR status.

Keywords: Bayesian projective prediction; exercise; obesity.