Differentiating Alcohol and Substance Use Disorders Using Multiclass Machine Learning Models Based on Routine...

Area of Science:

• Hematology
• Machine Learning
• Addiction Medicine

Background:

• Clinical evaluation is standard for diagnosing alcohol and substance use disorders.
• Objective biomarkers for these conditions are lacking, posing a diagnostic challenge.

Purpose of the Study:

• To investigate the utility of routine hemogram parameters in differentiating alcohol use disorder (AUD), substance use disorder (SUD), and healthy controls.
• To apply multiclass machine learning models for this classification task.

Main Methods:

• Retrospective case-control study involving 35 AUD patients, 61 SUD patients, and 132 controls.
• Analysis of routine hematological parameters using Random Forest, Support Vector Machine (SVM), and Artificial Neural Network (ANN) models.
• 10-fold cross-validation for performance evaluation using accuracy, sensitivity, precision, F1-score, and AUC.

Main Results:

• Significant differences in monocyte count, basophil count, and RDW-CV were observed between groups.
• The Random Forest model achieved the highest accuracy (81.6%) and AUC (0.93).
• Classification performance varied, with lower sensitivity for the alcohol use disorder group.

Conclusions:

• Routine hemogram parameters analyzed by machine learning offer a potential low-cost, accessible supportive tool for differentiating addiction-related conditions.
• Findings are exploratory due to study limitations (retrospective, single-center, small sample size, lack of external validation).
• Further research with larger datasets and explainable AI is necessary for clinical application.