From dry eye to depression: a machine learning-based framework for predicting adolescent mental health

Area of Science:

• Ophthalmology
• Psychiatry
• Public Health

Background:

• Adolescent depression is a significant public health issue.
• Current risk assessment tools for depression often overlook physical health indicators.
• The role of dry eye disease (DED) in predicting depressive symptoms in adolescents is under-explored.

Purpose of the Study:

• To investigate if incorporating dry eye disease (DED) into existing risk assessment models enhances the prediction of depressive symptoms in adolescents.
• To evaluate the effectiveness of machine-learning classifiers in predicting adolescent depression when ocular health is considered alongside psychosocial factors.

Main Methods:

• Analysis of 2,076 adolescent questionnaires encompassing ocular health, sleep patterns, electronic device usage, social support, and demographics.
• Utilized five machine-learning classifiers trained via cross-validation.
• Evaluated model performance based on discrimination (AUC) and calibration.

Main Results:

• Models incorporating DED demonstrated strong predictive discrimination with an AUC of approximately 0.84.
• The models showed good calibration, performing best in identifying no depression and severe depression.
• Prediction accuracy was lower for mild to moderate depression categories.

Conclusions:

• Integrating ocular health indicators, such as DED, with psychosocial factors improves the machine-learning prediction of adolescent depression.
• This approach holds potential for earlier, school-based identification and referral of at-risk adolescents.
• The low-cost, questionnaire-based method shows promise for population screening and targeted prevention, pending further validation.