Functional Survival Models for Detecting Recent Cannabis Intoxication using Wearable Pupillometers

JULIA WROBEL – EMORY UNIVERSITY

ABSTRACT

The legalization of cannabis has created an urgent need for objective, real-time markers of recent intoxication that remain informative even for frequent users with high tolerance. Pupil light response curves, measured by portable pupillometers, provide a rich functional signal that changes measurably following cannabis consumption. We analyze these curves within a functional survival framework, treating time since cannabis use as a right-censored outcome and using the full response trajectory as a functional predictor. To this end, we develop novel functional linear and additive accelerated failure time (AFT) models to predict time since last cannabis use. Simulation studies further show strong estimation and prediction performance across a range of scenarios, with robustness to moderate model misspecification. Applied to data from the Colorado Cannabis & Driving Study, our methods demonstrate that pupil light responses contain meaningful signal about recency of intoxication. These results highlight the potential of functional survival models for cannabis impairment detection, with broader implications for traffic safety and biomedical monitoring.