Unlocking the potential of wearable technology: Fitbit-derived measures for predicting ADHD in adolescents

Area of Science:

• Neuroscience
• Computational Psychiatry
• Digital Health

Background:

• Attention-deficit/hyperactivity disorder (ADHD) is a prevalent neurodevelopmental disorder with intricate causes.
• Current ADHD diagnostic methods are often lengthy and subjective.
• Machine learning (ML) offers novel approaches to improve ADHD diagnosis through diverse data integration.

Purpose of the Study:

• To investigate the utility of Fitbit-derived physical activity data in enhancing ADHD diagnosis.
• To explore correlations between specific Fitbit metrics and ADHD diagnosis.
• To evaluate the predictive power of wearable data for ADHD using ML models.

Main Methods:

• Analysis of 450 participants from the Adolescent Brain Cognitive Development (ABCD) study.
• Correlation analyses examining associations between ADHD diagnosis and Fitbit data (sedentary time, resting heart rate, energy expenditure).
• Multivariable logistic regression and ML classifiers (including Random Forest) for ADHD prediction and classification.

Main Results:

• Significant correlations were found between ADHD diagnosis and Fitbit-derived physical activity data.
• Specific Fitbit measurements significantly predicted ADHD diagnosis in logistic regression models.
• The Random Forest classifier achieved high performance: 0.89 cross-validation accuracy, 0.95 AUC, 0.88 precision, 0.90 recall, 0.89 F1-score, and 0.88 test accuracy.

Conclusions:

• Fitbit data holds potential for predicting ADHD diagnosis.
• ML classifiers, especially Random Forest, demonstrate strong predictive capabilities using wearable data.
• Wearable sensor data may facilitate more objective and efficient ADHD identification, improving clinical practices.