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

Background: Sentiment analysis is increasingly used to assess emotional states in clinical settings, particularly in depression. This study investigates the relationship between sentiment and depression severity in psychiatric interviews. Aims: We aimed to evaluate whether sentiment scores derived from patients' spoken words, specifically positive, negative, and neutral sentiments, correlate with the severity of depressive symptoms. Method: We analyzed transcribed interviews from 32 patients with depressive symptoms receiving treatment. The BERT (Bidirectional Encoder Representations from Transformers) model was used for sentiment analysis, which classified the sentiment of each sentence spoken by the patients. Depression severity was measured by the Beck's Depression Inventory. To examine relationships between sentiment and depression scores, descriptive statistics and partial correlation analysis were conducted including the covariates age, sex, medication, and number of sentences spoken. Results: Interviews lasted an average of 15 min and 32 s, with patients speaking a mean of 29.9 sentences. On average, 71 % of sentences were negative, 19 % neutral, and 11 % positive. A higher proportion of negative sentiment was weakly associated with reduced depression severity (r = −0.120, p = .028), while a higher proportion of neutral sentiment was associated with greater depression severity (r = 0.164, p = .041). Positive sentiment showed no significant association with depression severity. Conclusion: These results suggest that neutral sentiment may serve as a valuable indicator of emotional numbing in adult depression. Incorporating sentiment analysis into clinical practice could provide a more comprehensive view of depressive symptomatology, beyond traditional assessments focusing mainly on negative emotions.