Explainable illicit drug abuse prediction using hematological differences.

Area of Science:

• Biomedical Informatics
• Clinical Chemistry
• Machine Learning in Healthcare

Background:

• Illicit drug use (IDU) presents significant public health challenges.
• Accurate and early identification of IDU is crucial for timely intervention and treatment.
• Hematological parameters offer potential biomarkers for predicting IDU, but require robust analytical models.

Purpose of the Study:

• To develop and validate a reliable and explainable machine learning (ML) model for predicting illicit drug use (IDU).
• To identify key hematological features that contribute to the prediction of IDU.
• To assess the clinical utility of the predictive model for preliminary IDU screening.

Main Methods:

• Utilized hematological data from 286 illicit drug users (IDUr) and 302 non-users (n-IDUr).
• Compared the performance of eight ML algorithms to predict IDU.
• Developed an explainable Light Gradient Boosting Machine (LGB) model using 13 selected features.
• Employed SHapley Additive exPlanations (SHAP) for model interpretability.

Main Results:

• The explainable LGB model achieved high prediction accuracy, with an area under the curve (AUC) of 0.925 in internal validation and 0.915 in external validation.
• Key predictive features identified include chloride (Cl), β-hydroxybutyrate (BHB), and anion gap (AG).
• These features are associated with kidney, liver, and thyroid function, indicating potential organ dysfunction in IDUr.

Conclusions:

• An explainable ML model effectively predicts illicit drug use based on hematological parameters.
• Chloride, BHB, and anion gap are significant indicators for IDU prediction.
• The model holds potential for clinical application in preliminary IDU screening and guiding further organ-specific examinations.