Predicting PANSS symptoms in schizophrenia spectrum disorders using speech only: an international, multi-centre, computational secondary analysis across multiple languages
Abstract
Background: Speech carries cues to variation in mental state in schizophrenia, typically indexed with clinician-rated scales such as Positive and Negative Syndrome Scale (PANSS). In this study, we aggregate multi-centre recordings to assemble a large corpus and assess symptom-prediction models at scale, to enable rater-independent, efficient, and reproducible assessments for early detection of relapse-related signals from speech.
Methods: In this computational secondary analysis, we compiled data from 453 patients with schizophrenia spectrum disorders, recruited between 2005 and 2026 from ten global sites in the Czech Republic, United States, Spain, Chile, France, Switzerland, the Netherlands, Türkiye, Germany, and China, and clipped their speech recordings into 6664 segments. Across three featuresets, acoustic-prosodic profile, pretrained multilingual embeddings, and their concatenation, we compared 16 algorithms in separate item-specific models to predict eight relapse-related PANSS items, including three positive (P1, P2, P3), three negative (N1, N4, N6), and two general (G5, G9) items, on speaker-disjoint splits. Performance was assessed by root-mean-squared-error (RMSE) at both segment and participant (median-aggregation) levels. Best model per item underwent bias checks for age, sex, education, and symptom severity.
Findings: Best-performing models predicted symptoms with prediction errors of 1.5 or lower: P1 1.544/1.527, P2 1.318/1.107, P3 1.487/1.542, N1 1.071/1.030, N4 1.521/1.430, N6 0.869/0.855, G5 0.951/0.882, G9 1.298/1.282 (segment/participant). Performance of pretrained embeddings surpassed acoustic-prosodic features and concatenation. Results were comparable in languages or language variants with fewer available speech and language computational resources. We found no bias by age, sex, or education, aside from reduced N4 accuracy in males; but performance degraded with higher symptom severity.
Interpretation: Speech can support automatic assessment of schizophrenia symptoms using pretrained embeddings, even without transcripts. Such models show promise as clinically meaningful, efficient, and low-burden tools for real-time monitoring of symptom trajectories. Future work should validate these models prospectively in higher-severity samples and improve clinical interpretability.
Funding: European Union Horizon.
Related
- Deficient Monetary Reward Processing in Schizophrenia: A Meta-Analysis and Computational Modeling of a Monetary Incentive Delay Task
- Ultra-rare coding variants in schizophrenia risk genes are associated with longitudinal phenotypic differences across psychotic disorders
- Cost-effectiveness of adding enzalutamide vs standard non-steroidal antiandrogen drugs for patients with metastatic hormone-sensitive prostate cancer undergoing standard first-line therapy: a within-trial, secondary end-point analysis of the ENZAMET phase 3, international randomized trial (ANZUP 1304)
- Prevalence, severity, and correlates of post-traumatic stress disorder in 205 outpatients with schizophrenia: results from a cross-sectional, observational study assessing diagnoses via the Mini-International Neuropsychiatric Interview
- Pleiotropic genetic architecture linking schizophrenia and substance use disorders
BibTeX
@article{he2026predicting,
title = {{Predicting PANSS symptoms in schizophrenia spectrum disorders using speech only: an international, multi-centre, computational secondary analysis across multiple languages}},
author = {Rui He and Maryia Kirdun and Claudio Palominos and Lara Navarrete Orejudo and Sophie Barthelemy and Simran Bhola and Silvia Ciampelli and Amandine Decker and Cemal Demirlek and Riccardo Fusaroli and José Tomás García-Molina and Guy Gimenez and Roya Hüppi and Katja Koelkebeck and Amandine Lecomte and Renxiang Qiu and Arndis Simonsen and Vincent Tourneur and Burcu Verim and Huiling Wang and Berna Yalincetin and Sandy Yin and Yuan Zhou and Maxime Amblard and Rosa Ayesa Arriola and Emre Bora and Janna de Boer and Alicia I. Figueroa-Barra and Sanne Koops and Michel Musiol and Lena Palaniyappan and Alberto Parola and Filip Spaniel and Sunny X. Tang and Iris E. Sommer and Philipp Homan and Wolfram Hinzen},
journal = {eClinicalMedicine},
year = {2026},
month = oct,
doi = {10.1016/j.eclinm.2026.104265},
url = {https://linkinghub.elsevier.com/retrieve/pii/S2589537026005183},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Predicting PANSS symptoms in schizophrenia spectrum disorders using speech only: an international, multi-centre, computational secondary analysis across multiple languages
AU - Rui He
AU - Maryia Kirdun
AU - Claudio Palominos
AU - Lara Navarrete Orejudo
AU - Sophie Barthelemy
AU - Simran Bhola
AU - Silvia Ciampelli
AU - Amandine Decker
AU - Cemal Demirlek
AU - Riccardo Fusaroli
AU - José Tomás García-Molina
AU - Guy Gimenez
AU - Roya Hüppi
AU - Katja Koelkebeck
AU - Amandine Lecomte
AU - Renxiang Qiu
AU - Arndis Simonsen
AU - Vincent Tourneur
AU - Burcu Verim
AU - Huiling Wang
AU - Berna Yalincetin
AU - Sandy Yin
AU - Yuan Zhou
AU - Maxime Amblard
AU - Rosa Ayesa Arriola
AU - Emre Bora
AU - Janna de Boer
AU - Alicia I. Figueroa-Barra
AU - Sanne Koops
AU - Michel Musiol
AU - Lena Palaniyappan
AU - Alberto Parola
AU - Filip Spaniel
AU - Sunny X. Tang
AU - Iris E. Sommer
AU - Philipp Homan
AU - Wolfram Hinzen
PY - 2026
DA - 2026/10/28
JO - eClinicalMedicine
DO - 10.1016/j.eclinm.2026.104265
UR - https://linkinghub.elsevier.com/retrieve/pii/S2589537026005183
AB - Background: Speech carries cues to variation in mental state in schizophrenia, typically indexed with clinician-rated scales such as Positive and Negative Syndrome Scale (PANSS). In this study, we aggregate multi-centre recordings to assemble a large corpus and assess symptom-prediction models at scale, to enable rater-independent, efficient, and reproducible assessments for early detection of relapse-related signals from speech. Methods: In this computational secondary analysis, we compiled data from 453 patients with schizophrenia spectrum disorders, recruited between 2005 and 2026 from ten global sites in the Czech Republic, United States, Spain, Chile, France, Switzerland, the Netherlands, Türkiye, Germany, and China, and clipped their speech recordings into 6664 segments. Across three featuresets, acoustic-prosodic profile, pretrained multilingual embeddings, and their concatenation, we compared 16 algorithms in separate item-specific models to predict eight relapse-related PANSS items, including three positive (P1, P2, P3), three negative (N1, N4, N6), and two general (G5, G9) items, on speaker-disjoint splits. Performance was assessed by root-mean-squared-error (RMSE) at both segment and participant (median-aggregation) levels. Best model per item underwent bias checks for age, sex, education, and symptom severity. Findings: Best-performing models predicted symptoms with prediction errors of 1.5 or lower: P1 1.544/1.527, P2 1.318/1.107, P3 1.487/1.542, N1 1.071/1.030, N4 1.521/1.430, N6 0.869/0.855, G5 0.951/0.882, G9 1.298/1.282 (segment/participant). Performance of pretrained embeddings surpassed acoustic-prosodic features and concatenation. Results were comparable in languages or language variants with fewer available speech and language computational resources. We found no bias by age, sex, or education, aside from reduced N4 accuracy in males; but performance degraded with higher symptom severity. Interpretation: Speech can support automatic assessment of schizophrenia symptoms using pretrained embeddings, even without transcripts. Such models show promise as clinically meaningful, efficient, and low-burden tools for real-time monitoring of symptom trajectories. Future work should validate these models prospectively in higher-severity samples and improve clinical interpretability. Funding: European Union Horizon.
SN - 2589-5370
ER -