Predicting six-year depression trajectories in adulthood: A super learner ensemble approach using the Add Health...

Area of Science:

• Psychiatry and Mental Health
• Data Science and Machine Learning
• Public Health

Background:

• Depression is a significant global health issue, causing widespread disability.
• Identifying predictors of adult depression severity is crucial for public health interventions.
• This study explores biopsychosocial factors influencing depression using advanced machine learning.

Purpose of the Study:

• To identify the strongest predictors of future depression severity in adults.
• To utilize a Super Learner algorithm for comprehensive prediction modeling.
• To analyze a nationally representative longitudinal cohort of U.S. adults.

Main Methods:

• Employed the Super Learner (a stacked ensemble algorithm) on the Add Health dataset (n=3208).
• Utilized data from two waves (2016-2018 and 2022-2025) measuring depression severity.
• Applied Shapley Additive Explanations (SHAP) to interpret predictor importance on a held-out test set.

Main Results:

• The Super Learner did not significantly outperform individual base learners.
• Baseline depression score was the most influential predictor of future severity (mean |SHAP|=0.416).
• Other key predictors included perceived stress, prior depression diagnosis, everyday discrimination, and social gathering frequency.

Conclusions:

• A history of depression is the most robust predictor of long-term depression severity.
• Current predictive models explain less than 25% of the variance in depression severity.
• Individual-level prediction of depression severity has limitations, tempering expectations for highly accurate forecasts.