Detecting benzodiazepine use through induced eye convergence inability with a smartphone app: a proof-of-concept study

Abstract

Background: Benzodiazepines (BZDs) are readily available potent drugs that act as central depressants. These drugs are widely used, misused, and abused. For patients with BZD use disorder, the traditional sobriety monitoring method is periodic urine tests.

Methods: The utility of eye-scanning data related to non-convergence (the ability to cross eyes) collected using smartphones with the Previct Drugs app before and after ingestion of the BZD lorazepam for detecting BZD-driven effects was evaluated using data from 12 individuals from a historic clinical study (NCT05731999). Using a novel metric that represents the change in distance between irises when converging eyes, either in absolute terms (NCdiff) or individualized (NCdiffInd), classifiers were built using logistic regression.

Results: The ability to converge eyes is a strongly individual and acquired skill that is impaired after ingesting lorazepam. The maximum NCdiff for a BZD-sober individual may be smaller than the impaired NCdiff for another individual. Using the NCdiff measured in a sober condition after approximately 1 week of regular eye-scanning as the individual baseline to form NCdiffInd produced a highly functional classifier with an area under the curve (AUC) = 0.88, which was superior to a classifier based on NCdiff with an AUC = 0.79.

Conclusions: The loss of eye convergence induced by lorazepam is continuous, individual, and can be partial. Smartphone-based eye-scanning technology combined with a classifier adapted to the ability of eye convergence of individuals shows promising performance in detecting ingestion of lorazepam.

Keywords: benzodiazepines; eye convergence; pupillometry; smartphone; substance use disorder.

Figure 1

Illustration of the measured ability to converge one’s eyes (left eye, red, right eye blue) in a sober state (a) and under the influence of 2 mg of orally administered lorazepam. (b) The subject was video-filmed for 8 s, during which audio guidance told the subject to cross their eyes (at ∼1.5 s) and later to look straight ahead (∼5 s). GDFN is the gaze distance from the nose. NCdiff values from each measurement made during the study (first visit sober baseline, measurements at home, second visit baseline, and finally measurements under the influence of the BZD drug lorazepam) for subject B is shown in (c).

Figure 2

NCdiff results from all 12 subjects pre (BZD = 0) and post (BZD = 1) lorazepam administration. The red line indicates the classification limit. Open blue circles denote true negative results, filled blue circles denote false positive results, open red diamonds denote true positive results, and filled red diamonds represent false negative results.

Figure 3

Receiver operator characteristic (ROC) graphs for classifiers indicating BZD intake based on (a) NCdiff and (b) NCdiffInd. The intersection of the yellow line and the black ROC curve defines the threshold for the classifier.

Figure 4

All the NCdiffInd results from all study subjects in the study are plotted in order of the performed tests. The red line indicates the classification limit. The open symbols represent true positives and negatives, while the filled symbols indicate false positives and negatives. V1Base and V2Base refer to the tests performed during the two hospital visits before drug administration. The Home-true negative (TN) / false positive (FP) tests represent all tests conducted at home between these visits. The BZD1-5h-false negative (FN) / true positive (TP) tests refer to those performed 1–5 h after the administration of 2 mg of lorazepam at visit 2.