Automated speech and language markers of longitudinal changes in psychosis symptoms.
Level 3 - non-randomized controlled study
Prospective observational longitudinal cohort study
PubMed 40538505 · doi:10.1038/s44277-025-00034-z
What was done
Sixty-six participants hospitalized with psychotic disorders were evaluated across four longitudinal time points: shortly after inpatient admission, at discharge, and at 3- and 6-month follow-ups. Psychosis symptoms across three major domains (Thought Disorder, Negative Symptoms, and Positive Symptoms) were measured using semi-structured interviews and standardized rating scales. Speech was recorded during four tasks: paragraph reading, fluency, picture description, and open-ended speech. Linear mixed models were used to analyze longitudinal relationships between symptom scores and 357 automated speech and language features, evaluated both as a single component score and as individual features.
What was found
The single composite component score demonstrated significant longitudinal relationships with all three major symptom domains. Thought Disorder was associated with more subordinated syntactic constructions, less efficient identification of picture elements, and decreased semantic distance between sentences. Negative Symptoms were associated with decreased speech complexity. The composite Positive Symptoms domain did not retain significant individual feature associations after multiple-testing correction, though subdomain analyses indicated Suspiciousness was related to decreased noun use and Hallucinations to greater semantic distance. No exact effect sizes, coefficients, or p-values were reported in the abstract.
Why it matters
This study shows that automated natural language processing and acoustic features can objectively track longitudinal symptom fluctuations in psychotic disorders across inpatient and outpatient phases.
Limits
The sample size is relatively small (n = 66), and the abstract provides no exact numerical values, effect sizes, or confidence intervals. No individual features survived multiple-testing correction for the composite positive symptom domain. Speech was collected via structured clinical tasks rather than naturalistic continuous recording, and findings varied between acute and stable illness phases.
Cited by
- supports Automated analysis of syntactic completion and reading patterns can predict the likelihood of psychosis.