Neutral sentiment on patient's speech can predict the depressive symptom severity transdiagnostically

Area of Science:

• Clinical Psychology
• Computational Linguistics
• Psychiatry

Background:

• Sentiment analysis is increasingly applied to assess emotional states, particularly in depression.
• This study investigates the nuanced relationship between sentiment expression and depression severity in clinical interviews.

Purpose of the Study:

• To evaluate the correlation between positive, negative, and neutral sentiment scores from spoken language and depressive symptom severity.
• To explore sentiment analysis as a supplementary tool for understanding depression.

Main Methods:

• Transcribed interviews from 32 patients with depression were analyzed using the BERT (Bidirectional Encoder Representations from Transformers) model for sentiment classification.
• Depression severity was assessed using the Beck's Depression Inventory.
• Partial correlation analysis, controlling for age, sex, medication, and speech duration, examined sentiment-depression score relationships.

Main Results:

• Patients spoke an average of 29.9 sentences per 15-minute interview, with 71% negative, 19% neutral, and 11% positive sentiment.
• A higher proportion of neutral sentiment was weakly associated with greater depression severity (r=0.164, p=0.041).
• Negative sentiment showed a weak association with reduced severity (r=-0.120, p=0.028), while positive sentiment had no significant correlation.

Conclusions:

• Neutral sentiment may indicate emotional numbing in adult depression.
• Sentiment analysis offers a more comprehensive assessment of depressive symptomatology.
• Integrating sentiment analysis into clinical practice can enhance understanding of patient emotional states.