Is in-vehicle technology better than humans at identifying alcohol impairment in simulated driving?

Eichelberger, Angela H. / Brown, Timothy L. / Schmitt, Rose / Marshall, Dawn C. / Kuo, Jonny / Lenné, Michael G.
Insurance Institute for Highway Safety
February 2026

Abstract
Objective: This study focuses on the accuracy of eye metrics in identifying alcohol impairment among drivers, compared with observations and ratings by law enforcement officers.
Method: Thirty-six adults participated in a baseline-controlled study with five breath alcohol concentration (BrAC) conditions (0.000, 0.100, 0.085, 0.070, and 0.055 g/210 L). In each condition, participants drove on a simulator integrated with a driver monitoring system (DMS) camera that measured several eye metrics (e.g., eye opening, pupil size, gaze). Officers reviewed and evaluated videos of participants for indicators of impairment. Logistic regression models predicting alcohol thresholds (>= 0.05 and >= 0.08 g/210 L) and general linear models predicting driving performance, measured as standard deviation of lane position (SDLP), were used to evaluate DMS eye metrics and officers’ observations and ratings.
Results: Both the officer and DMS models provided a fair level of accuracy when predicting BrAC >= 0.05 (area under the receiver operating characteristic curve [AUROC] = 0.75), and both models had more difficulty predicting BrAC >=0.08 (AUROC = 0.64 for officer and 0.63 for DMS). Regarding SDLP, the DMS model accounted for more variance compared with the officer model (36% vs. 22%).
Conclusions: Regarding the prediction of alcohol thresholds, eye metrics from a DMS camera provided a similar level of accuracy compared with officers’ observations and ratings. If a high level of accuracy is desired in identifying alcohol impairment, none of the models tested would be sufficient without additional confirmation. Public health significance statement: Alcohol-impaired driving continues to be a significant problem on our roads. This research highlights eye metrics as potential indicators of alcohol impairment among drivers.
Abstract Objective: This study focuses on the accuracy of eye metrics in identifying alcohol impairment among drivers, compared with observations and ratings by law enforcement officers.
Method: Thirty-six adults participated in a baseline-controlled study with five breath alcohol concentration (BrAC) conditions (0.000, 0.100, 0.085, 0.070, and 0.055 g/210 L). In each condition, participants drove on a simulator integrated with a driver monitoring system (DMS) camera that measured several eye metrics (e.g., eye opening, pupil size, gaze). Officers reviewed and evaluated videos of participants for indicators of impairment. Logistic regression models predicting alcohol thresholds (>= 0.05 and >= 0.08 g/210 L) and general linear models predicting driving performance, measured as standard deviation of lane position (SDLP), were used to evaluate DMS eye metrics and officers’ observations and ratings.
Results: Both the officer and DMS models provided a fair level of accuracy when predicting BrAC >= 0.05 (area under the receiver operating characteristic curve [AUROC] = 0.75), and both models had more difficulty predicting BrAC >=0.08 (AUROC = 0.64 for officer and 0.63 for DMS). Regarding SDLP, the DMS model accounted for more variance compared with the officer model (36% vs. 22%).
Conclusions: Regarding the prediction of alcohol thresholds, eye metrics from a DMS camera provided a similar level of accuracy compared with officers’ observations and ratings. If a high level of accuracy is desired in identifying alcohol impairment, none of the models tested would be sufficient without additional confirmation. Public health significance statement: Alcohol-impaired driving continues to be a significant problem on our roads. This research highlights eye metrics as potential indicators of alcohol impairment among drivers.