Opioid misuse detection from cognitive and physiological data with temporal fusion deep learning.

Area of Science:

• Digital health
• Machine learning applications
• Clinical informatics

Background:

• Opioid misuse presents significant risks, including overdose and opioid use disorder.
• Early detection of opioid misuse is crucial for preventing adverse outcomes.
• Machine learning (ML) offers potential for identifying at-risk individuals.

Purpose of the Study:

• To develop and validate an ML model for detecting opioid misuse.
• To assess the predictive value of cognitive task performance and physiological data.
• To compare ML-based detection with traditional assessment methods.

Main Methods:

• Collected data from 169 patients prescribed opioid analgesics, including on-body sensor data and performance on cognitive tasks (Dot Probe, Go/No-Go).
• Utilized a temporal fusion transformer ML model to predict opioid misuse status based on Current Opioid Misuse Measure (COMM) categorization.
• Employed Leave-One-Group-Out (LOGO) cross-validation for robust and unbiased model performance assessment.

Main Results:

• The ML model achieved good predictive performance for opioid misuse detection (AUC, 0.81).
• Behavioral responses during cognitive tasks demonstrated stronger predictive power than physiological signals (heart rate variability, respiration rate).
• Model sensitivity and specificity were 0.78 for both metrics.

Conclusions:

• ML models integrating cognitive task data and sensor data can detect opioid misuse with accuracy comparable to self-report measures.
• Wearable sensor data may offer limited additional predictive value beyond behavioral responses.
• Further research should benchmark ML models against objective measures of opioid misuse.