The relationship of measures of eye behavior and driving performance under the influence of alcohol

Brown, Timothy L. / Schmitt, Rose / Marshall, Dawn C. / Eichelberger, Angela H. / Kuo, Jonny / Lenne´, Michael G.
Insurance Institute for Highway Safety
June 2025

Abstract
Objective: The persistence of alcohol-involved crashes is a significant and persistent challenge to traffic safety that requires new strategies to combat the annual loss of life. This study examined the extent to which a modern vision-based driver monitoring system (DMS) could be used to detect driver-based measures of alcohol impairment.
Method: Thirty-six participants completed the study protocol. This baseline-controlled within-subject study involved testing on the decline at four sequential breath alcohol concentrations (BrACs) of 0.100 , 0.085, 0.070, and 0.055 g/210L on a quarter-cab driving simulator with an integrated DMS. Visual measures from the DMS were used to predict alcohol impairment and driving performance as measured by standard deviation of lateral position (SDLP).
Results: In models predicting alcohol level (> 0.00, >= 0.05, and >= 0.08), two variables were significantly associated with the odds that a driver was above a given threshold: median eye opening (OR = 0.47, 0.62, and 0.68) and median percentage of time focused on the road center (OR = 1.05, 1.07, and 1.04). There were no other consistent predictors across the three models. Two models predicting driving performance (SDLP) included a general model (adjusted R-squared = 0.36) and an individualized model accounting for individual differences (adjusted R-squared = 0.56). In the general model, standard deviation of eye opening, median eye opening, median percentage of time focused on the road center, standard deviation of percentage of time focused on the road center, and standard deviation of horizontal gaze predicted SDLP. In the individualized model, median eye opening, standard deviation of eye opening, and median pupil diameter predicted SDLP.
Conclusion: Visual measures were used to predict alcohol-impaired drivers, but the summary measures alone lack the sensitivity and specificity to be used to identify them without including additional measures or context to the prediction. Models that accounted for individual differences between drivers were superior compared with those that did not account for these differences. Of the visual measures considered, decreasing median eye opening was particularly sensitive to alcohol-impaired driving. Increasing glance toward the forward roadway was predictive of increased levels of alcohol, especially when trying to predict alcohol above the legal limit of 0.08% blood alcohol concentration.
Abstract Objective: The persistence of alcohol-involved crashes is a significant and persistent challenge to traffic safety that requires new strategies to combat the annual loss of life. This study examined the extent to which a modern vision-based driver monitoring system (DMS) could be used to detect driver-based measures of alcohol impairment.
Method: Thirty-six participants completed the study protocol. This baseline-controlled within-subject study involved testing on the decline at four sequential breath alcohol concentrations (BrACs) of 0.100 , 0.085, 0.070, and 0.055 g/210L on a quarter-cab driving simulator with an integrated DMS. Visual measures from the DMS were used to predict alcohol impairment and driving performance as measured by standard deviation of lateral position (SDLP).
Results: In models predicting alcohol level (> 0.00, >= 0.05, and >= 0.08), two variables were significantly associated with the odds that a driver was above a given threshold: median eye opening (OR = 0.47, 0.62, and 0.68) and median percentage of time focused on the road center (OR = 1.05, 1.07, and 1.04). There were no other consistent predictors across the three models. Two models predicting driving performance (SDLP) included a general model (adjusted R-squared = 0.36) and an individualized model accounting for individual differences (adjusted R-squared = 0.56). In the general model, standard deviation of eye opening, median eye opening, median percentage of time focused on the road center, standard deviation of percentage of time focused on the road center, and standard deviation of horizontal gaze predicted SDLP. In the individualized model, median eye opening, standard deviation of eye opening, and median pupil diameter predicted SDLP.
Conclusion: Visual measures were used to predict alcohol-impaired drivers, but the summary measures alone lack the sensitivity and specificity to be used to identify them without including additional measures or context to the prediction. Models that accounted for individual differences between drivers were superior compared with those that did not account for these differences. Of the visual measures considered, decreasing median eye opening was particularly sensitive to alcohol-impaired driving. Increasing glance toward the forward roadway was predictive of increased levels of alcohol, especially when trying to predict alcohol above the legal limit of 0.08% blood alcohol concentration.