What question did this study set out to answer?

The aim is to develop a mechanistic forecaster to predict glucose management indicator values for individuals on basal insulin in type 2 diabetes.

June 7, 2026

1955-P: Predicting Glucose Management Indicator Values under Once-Daily Basal Insulin Therapy in Type 2 Diabetes

Key Points

The aim is to develop a mechanistic forecaster to predict glucose management indicator values for individuals on basal insulin in type 2 diabetes.
Secondary analysis of a 16-week basal insulin titration study (NCT06111508) with n=30 participants.
Data arranged in 28-day segments, using a 14-day training and a 14-day test window.
Daily plasma insulin concentration was estimated from recorded doses and a pharmacokinetic model.
The prediction model showed a root mean square error of 0.3% for test-window GMI.
Correlation coefficient of 0.86 and R² = 0.74 demonstrated good predictive agreement.
Individualized prediction tools support improved insulin titration for better glycemic control.

Abstract

Introduction and Objective: Once-daily basal insulin can improve glycemic control in type 2 diabetes (T2D), but dose titration remains largely a trial-and-error procedure. This work develops a mechanistic dose-response forecaster to predict near-term glucose management indicator (GMI) in individuals using a continuous glucose monitor (CGM) and once-daily basal insulin therapy. Methods: A secondary analysis of data from a 16-week basal insulin titration study with CGM (NCT06111508) was performed (n=30). For each participant, data were arranged into rolling 28-day segments (14-day training window followed by 14-day test window), advanced by one day. CGM was smoothed and daily minimum glucose (DMG) was computed on days with at least 75% CGM coverage. Daily plasma insulin concentration (I) was estimated from recorded basal doses using an insulin degludec pharmacokinetic model. In each training window, DMG was modeled as DMG = Gmax / (1 + Si·I) and participant-specific parameters (Gmax, Si) were estimated. Test-window CGM was predicted by scaling training CGM using the ratio CGMpred/CGMtrain = (1 + Si·Itrain)/(1 + Si·Itest), using the individualized parameters and insulin concentration estimated from training (Itrain) and test (Itest) window basal records. Predicted GMI was computed from predicted CGM and compared against observed test-window GMI. Results: Predicted versus observed test-window GMI showed a root mean square error of 0.3% (per-subject mean±SD: 0.29±0.142%), a correlation coefficient of 0.86, and R² = 0.74 across rolling windows and participants. Conclusion: A mechanistic dose-response forecaster using CGM and basal insulin data predicted near-term GMI with good agreement, supporting its potential as a tool to inform individualized basal insulin titration in T2D. Disclosure M. Ganji: None. A. El Fathi: Research Support; Current; Tandem Diabetes Care, Inc. Research Support; Ended; Novo Nordisk A/S. C. Fabris: Research Support; Current; Dexcom, Inc., Tandem Diabetes Care, Inc. Other - Royalties; Current; Dexcom, Inc., Tandem Diabetes Care, Inc. Other - Husband is Dexcom employee; Current; Dexcom, Inc. Funding This work was supported by a grant from the UVA Office of the VicePresident for Research LaunchPad for Diabetes.

Ask AI

Helpful

Bookmark