Detection of Depression and Suicide Risk Based on Text From Clinical Interviews Using Machine Learning: Possibility...

Area of Science:

• Computational psychiatry
• Natural language processing in mental health
• Machine learning for clinical diagnosis

Background:

• Depression and suicide pose significant global health challenges.
• Objective diagnostic tools for these conditions are currently lacking.
• This study explores the potential of spoken language analysis for diagnosis.

Purpose of the Study:

• To develop a machine learning model for diagnosing depression using transcribed interview text.
• To assess the feasibility of identifying high-suicide-risk individuals based on their speech.
• To compare the efficacy of text-based analysis against demographic data for diagnosis.

Main Methods:

• Recruited 83 healthy and 83 depressed patients.
• Transcribed semi-structured interviews, extracting only participant speech.
• Employed Naive Bayes classifier for text-based machine learning models.

Main Results:

• The text-based depression diagnosis model achieved an Area Under the Curve (AUC) of 0.905, with 0.699 sensitivity and 0.964 specificity.
• Text-based classification significantly outperformed demographic models (AUC 0.761) for distinguishing patient groups (p=0.001).
• An ensemble model incorporating demographic variables improved suicide risk prediction (AUC 0.800).

Conclusions:

• Interview text, analyzed via machine learning, can objectively diagnose depression.
• Spoken language holds significant potential as a diagnostic marker for mental health conditions.
• Combining linguistic data with demographics enhances suicide risk assessment accuracy.