A LASSO regression model using continuous glucose monitor data estimated insulin sensitivity with a mean absolute error of 0.09 compared to the gold-standard intravenous test.
Observational (n=8)
Does continuous glucose monitoring accurately estimate insulin resistance compared to FSIGT in healthy adults?
Continuous glucose monitoring data analyzed with LASSO regression provides a feasible, low-cost, and minimally invasive method to estimate insulin resistance.
Introduction and Objective: Insulin resistance (IR) is cited as a contributor to common diseases; however, with the exception of diabetes-related research, IR is infrequently measured and rarely measured well. Gold-standard assessments, e.g. frequently sampled intravenous glucose tolerance test (FSIGT), are expensive, time-intensive, and invasive, which restricts use in large-scale or high-throughput settings. Our objective was to develop an accurate, cost-effective method for estimating IR that minimizes burden and is suitable for scalable use. Methods: Eight healthy adults were enrolled to wear a Dexcom G6 continuous glucose monitor (CGM) and subsequently undergo an FSIGT to derive insulin sensitivity (SI), a dynamic measure of IR. Raw CGM data were cleaned and trimmed to allow for device calibration and exclude glucose measurements collected during the FSIGT. Only full days of data were included. We used the R iglu package to compute 40 CGM-derived glucose control and variability metrics, including mean, coefficient of variation (CV), glucose management indicator (GMI), glucose risk index (GRI), and mean amplitude of glycemic excursions (MAGE). All metrics were scaled prior to model fitting. We applied regularized regression approaches—Ridge, LASSO, and Elastic Net—to develop estimation models, with mean absolute error (MAE) evaluating performance. Results: CGM wear ranged from 3-7 days, and SI ranged from 1.11-11.5 L·mU-¹·min-¹. Among the models, LASSO demonstrated the strongest predictive performance and lowest error (MAE=0.09) at λ=0.01. Additionally, we observed a consistent reduction in MAE as CGM wear days increased, indicating improved estimation accuracy with longer monitoring periods. Conclusion: This pilot provides evidence supporting feasibility of estimating IR from CGM as a low cost, minimally invasive alternative to direct measurement. Future studies will focus on improving generalizability, evaluating clinical relevance across common diseases, validating findings in other cohorts, and expanding the methodology to incorporate additional measures of IR. Disclosure N. Palmer: None. S. Chen: None. F. Hsu: None. S.K. Das: None. M. Mongraw-Chaffin: None. Funding P30 DK124723
PALMER et al. (Fri,) conducted a observational in Healthy (n=8). Continuous glucose monitor (CGM) derived metrics vs. Frequently sampled intravenous glucose tolerance test (FSIGT) was evaluated on Mean absolute error (MAE) of insulin sensitivity estimation. A LASSO regression model using continuous glucose monitor data estimated insulin sensitivity with a mean absolute error of 0.09 compared to the gold-standard intravenous test.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: