Hierarchical, Interactive, and Dynamic Predictive Capacity of Current Biological, Psychological, Social, and Environmental Measurements in Depression, Anxiety, ADHD, and Social Quality across the Lifespan

TL;DR

This study explores how different biological, psychological, social, and environmental factors predict mental health outcomes like depression, anxiety, and ADHD across the lifespan.

Contribution

The study introduces a novel integrative ensemble stacking method to assess the hierarchical and interactive predictive power of biopsychosocial measures.

Findings

Subjective-report metrics explained significantly more variation in ADHD, depression, and anxiety than biophysical and experimental-task measurements.

There were robust bidirectional interactions between trait-like characteristics and dynamic features of psychopathology, somatic, and social factors.

Psychiatric and social-quality outcomes in adults showed similar hierarchical patterns but also revealed developmental differences.

Abstract

An extensive and perplexing plurality of psychometric assesments, experimental-tasks, and biophysical measurement modalities have evolved alongside increasingly biopsychosocial models of behavior and psychopathology. Yet, despite alarming recent increasing rates of mental health problems, cumulative progress regarding the validity and comparative utility of wide-ranging measures to predict, isolate, or explain hallmark features and interacting systems in depression, anxiety, and ADHD remains concerningly enigmatic. Utilizing adolescent (Age 9–14) and parent data when available (combined N=~23,760) across 5 years from the ABCD dataset - we parcellated over a thousand unique biopsychosocial measures from all time points into 30 theoretically relevant, commonly-used, or potentially-informative domains. We then assessed their hierarchical and interactive predictive power using a novel integrative ensemble stacking of 4 state-of-the-art ML models. Further, addressing prior tradeoffs between high-dimensional prediction and interpretability, several synergistic methods were applied to provide uncertainty estimation, central features, and possible causal developmental dynamics. Results reveal notable overlapping transdiagnostic features as well as pronounced hierarchical differences between measurement-domains in predicting psychiatric, social, and cognitive outcomes. Myriad subjective-report metrics accounted for 27–76 fold more variation in ADHD, depression, and anxiety than an extensive available range of biophysical and experimental-task measurements. Still, many domain-specific features were highly predictive of several clinically meaningful operationalizations of depression and psychopathology. Illustrated are robust, consistent, and often bidirectional interactions between specific trait-like characteristics and dynamic features of coextensive psychopathology, somatic, social, and family relationships - accounting for substantial variation in ADHD, anxiety, and depression (r2=.49–.56). Psychiatric and social-quality targets in adults using available parent data exhibited similar hierarchical predictive patterns but also reveal important developmental differences. Collectively, these findings underscore the comparative predictive utility of wide-ranging current biopsychosocial measurements across several cardinal explanatory targets in psychiatry, and illuminate the structure and dynamics of the most prevalent mental-health challenges across the lifespan in novel detail, with implications for prediction, explanation, and control.

Figures