Maximum and minimum activity in inpatient adolescents with Bipolar Disorders: Daily-Variability classification of...

Area of Science:

• Neuroscience
• Psychiatry
• Computer Science

Background:

• Adolescent bipolar disorder (BD) diagnosis can be challenging.
• Objective markers are needed to differentiate BD from attention deficit/hyperactivity disorder (ADHD).
• Actigraphy offers a potential objective measure of daily activity.

Purpose of the Study:

• To investigate the utility of actigraphy data and AI for classifying adolescent psychiatric diagnoses.
• To differentiate between BD without ADHD, BD with ADHD, ADHD without BD, and other diagnoses (OD).

Main Methods:

• Utilized chart-reviewed actigraphy data from 389 inpatient adolescents (2014-2023).
• Employed AI methods, including Random Forest and XGBoost models, with feature engineering on time-series activity data.
• Validated models on a dataset comprising 5193 days of activity data.

Main Results:

• The XGBoost model with feature selection achieved 91.5% accuracy in classifying diagnostic groups.
• The most influential feature was the engineered difference between peak daily activity hours.
• Actigraphy-derived activity patterns, combined with age, proved effective for diagnostic classification.

Conclusions:

• Actigraphy and machine learning provide a promising, objective approach for classifying diagnostic groups in adolescent psychiatric inpatients.
• Engineered features of hourly activity, particularly variability, may serve as objective markers to enhance diagnostic accuracy.
• Further validation in diverse cohorts and real-world settings is warranted to assess clinical implications.