Differentiating Pediatric Bipolar Disorder, Attention-Deficit/Hyperactivity Disorder, and Other Psychopathologies Using Self-Reported Mood and Energy Data and Actigraphy Findings: Correlation and Machine Learning-Based Prediction of Mood Severity
Paediatrics
Differentiating Pediatric Bipolar Disorder, Attention-Deficit/Hyperactivity Disorder, and Other Psychopathologies Using Self-Reported Mood and Energy Data and Actigraphy Findings: Correlation and Machine Learning-Based Prediction of Mood Severity
Paediatrics

Differentiating Pediatric Bipolar Disorder, Attention-Deficit/Hyperactivity Disorder, and Other Psychopathologies Using Self-Reported Mood and Energy Data and Actigraphy Findings: Correlation and Machine Learning-Based Prediction of Mood Severity

JMIR Ment Health. 2025 Dec 4;12:e78163. doi: 10.2196/78163.

ABSTRACT

BACKGROUND: Distinguishing pediatric bipolar disorder (BD) from attention-deficit/hyperactivity disorder (ADHD) is challenging due to overlapping fluctuations in mood, energy, and activity. Combining objective actigraphy with self-reported mood and energy data may aid differential diagnosis and risk monitoring.

OBJECTIVE: This study aimed to test same-day associations between actigraphy-derived activity extremes and self-reported mood and energy, and to evaluate whether these measures predict same-day and next-day severe mood in adolescents with BD, ADHD, and other diagnoses.

METHODS: We analyzed 209 inpatients (2148 patient-days) across 4 groups (ADHD without BD: n=54; BD with ADHD: n=42; BD without ADHD: n=34; other diagnoses: n=79). Actigraphy data (Philips Actiwatch 2) were summarized into daily maximum and minimum quartiles (Max1-Max4 and Min1-Min4). Mood and Energy Thermometer (-10 to +10) ratings were categorized as follows: OK (<3), mild (3-4), moderate (5-6), and severe (>6). Group differences used Kruskal-Wallis and Mann-Whitney U tests with Bonferroni correction (P<.004). Associations used chi-square tests with Cramér V. Leak-safe machine learning (patient-wise GroupKFold) classified SevereDay (same day) and SevereTomorrow (next day) using actigraphy, sleep, energy, and demographic data.

RESULTS: BD without ADHD showed the tightest coupling of extreme activity with negative mood and energy (Cramér V of up to 0.24; P<.004). ADHD without BD showed stronger links between activity and positive energy. Machine learning achieved a receiver operating characteristic area under the curve (ROC-AUC) of 0.85, an accuracy of 0.79, and an F1-score of 0.67 for SevereDay. SevereTomorrow performance was moderate (ROC-AUC=0.80; accuracy=0.79; F1-score=0.60). Energy variability and actigraphy averages/peaks were the top predictors.

CONCLUSIONS: Integrating actigraphy, sleep, and daily energy ratings identifies severe mood days and provides early next-day risk signals in hospitalized adolescents. The findings support wearable-based phenotyping for precision monitoring, with external validation needed in outpatients.

PMID:41343774 | DOI:10.2196/78163