Can AI Aid Urine Drug Test Sign-Outs?

Researchers reported that an artificial intelligence workflow maintained high interpretive accuracy while reducing urine drug test sign-out time in a supervised clinical laboratory setting.

By
Andrea Surnit
July 9, 2026
4 min

Conexiant

Objective:

To evaluate the accuracy and efficiency of an AI system in generating preliminary interpretations of urine drug tests.

Approach:

Study Design: The study analyzed 83,553 urine drug tests from 26,459 patients at the University of Washington Medical Center from January 2014 to February 2024.
AI Workflow Development: Large language models (LLM) were used to extract substance-use labels from historical interpretations, training machine learning models to predict substance use from test results.
Evaluation Metrics: Primary outcomes included accuracy of label extraction, substance-use prediction, and AI-generated interpretations, while secondary outcomes focused on clinician modifications and adoption rates.

Key Findings:

LLM achieved 99.9% accuracy in extracting substance-use labels, outperforming manual methods.
Machine learning models predicted substance use with an AUC greater than 0.99 for 23 of 26 substances, with accuracy exceeding 94%.
AI integration reduced average sign-out time by 28.5 seconds per case (23% efficiency gain) and by 65 seconds with an automated medication-list feature (51% efficiency gain).
70% of AI-generated interpretations required no substantive changes, while 4% prompted overrides due to incorrect classifications.