Predicting adolescent depression and anxiety from multi-wave longitudinal data using machine learning.

Area of Science:

• Developmental psychology
• Machine learning applications
• Adolescent mental health

Background:

• Predicting adolescent depression and anxiety is crucial for early intervention.
• Longitudinal data across multiple developmental stages offers rich insights into risk factors.
• Machine learning models can integrate complex, multi-dimensional datasets.

Purpose of the Study:

• To evaluate the predictive power of information from various developmental stages for mid-adolescence depression and anxiety.
• To determine the earliest developmental stage at which risk can be reliably identified.
• To assess the added value of developmental data over demographics and prior disorder status.

Main Methods:

• Utilized a community sample with tri-annual assessments from ages 3-15.
• Employed canonical correlation analysis (CCA) to reduce high-dimensional risk factor data.
• Conducted ablation analysis to quantify the predictive contribution of different developmental periods.

Main Results:

• Canonical correlation analysis (CCA) components predicted age 15 disorder status better than chance from ages 3-12 for anxiety and ages 9-12 for depression.
• Information from age 12 significantly improved depression prediction, while age 9 and 12 data improved anxiety prediction over baseline.
• Predictive accuracy surpassed that of standard demographics and prior disorder history.

Conclusions:

• Screening for adolescent depression risk is feasible as early as age 9.
• Screening for adolescent anxiety risk can be initiated as early as age 3.
• Integrating developmental risk factors enhances prediction of adolescent mental health outcomes.