Naturalistic driving data can identify elevated driving risk in adults with Type 1 diabetes

Individuals with Type 1 Diabetes Mellitus (T1DM) may experience acute glucose events, such as hypoglycemia and severe hyperglycemia, that can impair attention, reaction time, and decision-making during activities of daily living (ADL), such as driving. While continuous glucose monitoring (CGM) detects such events, less is known about whether everyday driving behavior reflects days when these disturbances occur during active vehicle operation. We evaluate whether daily driving behavior differs on days when acute glucose-related impairment occurs during driving among individuals with T1DM.

This study followed 18 adults with T1DM (mean age = 30.78; 11 females) over 4 wk. Real-world driving data were continuously recorded, and sleep was assessed using wrist-worn actigraphy as a background daily contextual factor. Glucose levels were measured using CGM. The primary outcome was a high-risk driving day, defined as a day on which hypoglycemia or severe hyperglycemia occurred during active driving, as determined from CGM data. Gradient-boosted decision tree models (XGBoost) were evaluated using a nested leave-one-subject-out cross-validation framework to distinguish high-risk driving days based on behavioral features. Models were trained using driving-only, sleep-only, and combined driving-and-sleep feature sets.

Models trained using driving features alone demonstrated the most consistent discriminative performance in identifying high-risk driving days (AUROC = 0.67; F1 = 0.65), outperforming the sleep-only and sleep + driving combined model. Influential driving features reflect exposure and temporal patterns, including afternoon peak trips, trips near home, and total driving time. Sleep features alone showed limited ability to distinguish high-risk driving days (AUROC = 0.48).

Driving exposure features, particularly those capturing trip frequency, timing, and duration, were informative for distinguishing days when glucose-related impairment occurred during driving in individuals with T1DM. These findings demonstrate the feasibility of behavior-based approaches for flagging elevated driving risk when glucose-related impairment occurs. Such approaches may complement physiological monitoring in transportation safety and injury-prevention contexts.