Investigating the Physiology of Cannabis Intoxication and Impairment Using Wearable Sensors

Introduction Cannabis decriminalization/legalization has increased the availability and recreational use of cannabis products. Unlike ethanol, there are no standard validated assessments to identify cannabis-induced intoxication and impairment. The objective of this study was to show that wearable sensors can objectively identify cannabis intoxication. Methods This observational study evaluated participants who smoked their personal cannabis products. Heart rate (HR), skin temperature, electrodermal activity, and accelerometry data were continuously collected using a wrist-worn sensor (Empatica EmbracePlus). Participants were assessed using the DRUID© impairment application, the Subjective High Assessment Scale (SHAS) surveys, and 2-minute periods each of sit/stand/walk exercises. Participants performed the study protocol at home while monitored remotely by study staff. The biometric data, DRUID© scores, and subjective intoxication scores were analyzed using descriptive statistics, paired t-tests, Welch tests, and repeated measures analysis of variance (ANOVA). This data was used to train multiple machine learning (ML) models. Results Statistical analysis was performed on 17/20 enrolled participants who provided complete data sets. In this cohort, cannabis intoxication caused significant changes in HR and aggregate accelerometry. Non-significant decreases in limb electrodermal activity and increases in limb temperature were recorded. Significant changes in subjective intoxication (SHAS) and objective impairment (DRUID©) were observed. The best performing proof-of-concept ML models were better at detecting subjective intoxication (balanced accuracy of 0.864) than impairment (balanced accuracy of 0.763). Conclusions Wearable sensors can detect physiologic changes in participants intoxicated with cannabis. This biometric data can be used to create ML models to detect cannabis-induced impairment and subjective intoxication.