A web-based tool utilizing machine learning algorithms for predicting illicit drug use in emergency departments.

Area of Science:

• * Emergency Medicine
• * Clinical Toxicology
• * Data Science

Background:

• * Rapid identification of illicit drug use is crucial, especially with emerging psychoactive substances.
• * Traditional urine testing methods are time-consuming in emergency settings.
• * Developing predictive models can expedite the detection process for suspected cases.

Purpose of the Study:

• * To develop and validate a machine learning (ML) model for the early prediction of illicit drug use.
• * To identify key features from emergency department (ED) data for drug use prediction.
• * To create accessible tools for emergency physicians to aid in drug abuse surveillance.

Main Methods:

• * Utilized data from the Taiwan Emergency Department Drug Abuse Surveillance (TEDAS) database (2020-2023).
• * Included demographic, triage, referral, symptom, physical, and clinical data as features.
• * Trained and tested supervised ML algorithms (Random Forest, CatBoost, LightGBM) using chronological data splits and K-fold cross-validation, evaluating performance via AUC.

Main Results:

• * Analyzed 13,615 ED cases, with 23.4% testing positive for illicit drugs.
• * The CatBoost classifier achieved the highest performance with an Area Under the Curve (AUC) of 0.846 (95% CI: 0.831-0.859) in the test set.
• * A predictive web-based tool and mobile applications were developed and implemented.

Conclusions:

• * The developed ML model demonstrates high efficacy in predicting illicit drug use among ED patients.
• * The prediction tool is freely accessible, supporting emergency physicians in clinical decision-making.
• * Further research is recommended to evaluate post-implementation impact and international applicability.