Identifying depression in the United States veterans using deep learning algorithms, NHANES 2005-2018

Area of Science:

• Mental Health Research
• Artificial Intelligence in Healthcare
• Veterans Affairs

Background:

• Depression is a significant mental health concern for veterans, characterized by high mortality rates.
• Current methods for predicting and identifying depression risk factors in veterans remain limited.
• This study addresses the need for improved identification and understanding of depression in this population.

Purpose of the Study:

• To utilize deep learning algorithms for identifying depression in veterans.
• To pinpoint clinical manifestations and risk factors associated with veteran depression.
• To compare the efficacy of deep learning against traditional machine learning methods.

Main Methods:

• A dataset of 2,546 veterans was analyzed using data from the National Health and Nutrition Examination Survey (2005-2018).
• Deep learning and five traditional machine learning algorithms were employed with 10-fold cross-validation.
• Model performance was evaluated using metrics including AUC, accuracy, recall, specificity, precision, and F1 score.

Main Results:

• Deep learning demonstrated superior performance with the highest AUC (0.891) and specificity (0.906) in identifying veteran depression.
• Deep learning models achieved high AUC values in middle-aged (0.929) and older age groups (0.924).
• Identified risk factors included sleep difficulties, memory impairment, work incapacity, income, BMI, chronic diseases, vitamins E and C, and palmitic acid.

Conclusions:

• Deep learning algorithms significantly outperformed traditional machine learning methods in identifying depression among veterans.
• The study highlights deep learning's potential for accurately identifying depression and its associated risk factors in the veteran population.
• This advanced approach offers a promising tool for improving mental health care strategies for veterans.