First-onset major depression during the COVID-19 pandemic: A predictive machine learning model

Area of Science:

• Mental Health Research
• Epidemiology
• Machine Learning in Healthcare

Background:

• Longitudinal evaluation of first-onset major depression in Italian adults during the COVID-19 pandemic.
• Focus on individuals without prior clinician-diagnosed psychiatric disorders.
• Development of a predictive machine learning model (MLM) for independent sample evaluation.

Purpose of the Study:

• To determine the rates of first-onset major depression during the initial phases of the pandemic.
• To create and validate a machine learning model for predicting major depression.
• To identify key risk factors for developing depression during public health crises.

Main Methods:

• Online, self-reported survey conducted in two waves (May-June and Sept-Oct 2020).
• Provisional diagnoses of major depressive disorder (PMDD) determined using DSM criteria via Patient Health Questionnaire-9 algorithm.
• Machine learning model built using gradient-boosted decision trees and SHapley Additive exPlanations.

Main Results:

• First-onset PMDD observed in 7.4% (Wave 1) and 7.2% (Wave 2) of participants.
• The developed MLM achieved 76.5% sensitivity and 77.8% specificity when tested on an independent sample.
• Key predictors included low resilience, undergraduate student status, pandemic-related stress, poor sleep satisfaction, and low social support.

Conclusions:

• Significant rates of first-onset major depression were found in the Italian adult population during the pandemic.
• The predictive MLM demonstrated good performance, highlighting potential targets for depression prevention strategies.
• Findings suggest the utility of MLMs for identifying at-risk individuals during public health emergencies.