Can machine learning be useful as a screening tool for depression in primary care?

Area of Science:

• Psychiatry
• Medical Informatics
• Computational Biology

Background:

• Depression is a prevalent condition with significant economic impact and complex pathophysiology.
• Underdiagnosis and undertreatment are major challenges, with only 50% diagnosed and 15% treated in primary care.
• Barriers include stigma, somatic symptoms, and recall bias, highlighting the need for effective screening tools.

Purpose of the Study:

• To evaluate the efficacy of Machine Learning (ML) algorithms in identifying patients with depression.
• To utilize clinical, laboratory, and sociodemographic data for depression detection.

Main Methods:

• The study employed ML algorithms to analyze data from the Brazilian National Network for Research on Cardiovascular Diseases (June 2016 - July 2018).
• Specific algorithms, including Random Forests, were assessed for their diagnostic performance.

Main Results:

• ML algorithms demonstrated promising results in detecting depressive patients.
• The Random Forests model achieved an accuracy of 0.89, sensitivity of 0.90, and an area under the receiver operating characteristic curve of 0.87.

Conclusions:

• ML algorithms show significant potential as diagnostic screening tools for depression.
• AI-driven approaches can aid in reducing the number of undiagnosed depression cases, improving patient outcomes.