Machine learning (ML) approaches are a promising venue for identifying vocal markers of neuropsychiatric disorders, such as schizophrenia. While recent studies have shown that voice-based ML models can reliably predict diagnosis and clinical symptoms of schizophrenia, it is unclear to what extent such ML markers generalize to new speech samples collected using a different task or in a different language: the assessment of generalization performance is however crucial for testing their clinical applicability.

In this research, we systematically assessed the generalizability of ML models across contexts and languages relying on a large cross-linguistic dataset of audio recordings of patients with schizophrenia and controls.

We trained ML models of vocal markers of schizophrenia on a large cross-linguistic dataset of audio recordings of 231 patients with schizophrenia and 238 matched controls (>4.000 recordings in Danish, German, Mandarin and Japanese). We developed a rigorous pipeline to minimize overfitting, including cross-validated training set and Mixture of Experts (MoE) models. We tested the generalizability of the ML models on: (i) different participants, speaking the same language (hold-out test set); (ii) different participants, speaking a different language. Finally, we compared the predictive performance of: (i) models trained on a single language (e.g., Danish) (ii) MoE models, i.e., ensemble of models (experts) trained on a single language whose predictions are combined using a weighted sum (iii) multi-language models trained on multiple languages (e.g., Danish and German).

Model performance was comparable to state-of-the art findings (F1: 70%-80%) when trained and tested on participants speaking the same language (out-of-sample performance). Crucially, however, the ML models did not generalize well - showing a substantial decrease of performance (close to chance) - when trained in a language and tested on new languages (e.g., trained on Danish and tested on German). MoE and multi-language models showed a better increase of performance (F1: 55%-60%), but still far from those requested for achieving clinical applicability.

Our results show that the cross-linguistic generalizability of ML models of vocal markers of schizophrenia is very limited. This is an issue if our first goal is to translate these vocal markers into effective clinical applications. We argue that more emphasis needs to be placed on collecting large open datasets to test the generalizability of voice-based ML models, for example, across different speech tasks or across the heterogeneous clinical profiles that characterize schizophrenia spectrum disorder.

None Declared

Non-suicidal self-injury (NSSI) is associated with emotional distress and mental disorders. In clinical samples NSSI is reported by 21% to 60% of all psychiatric patients. Developed NSSI instruments are not suitable for clinical settings because they are too time-consuming or lack validation across psychiatric diagnoses.

The Transdiagnostic Self-Injury Interview (TSI) is semi-structured interview that accesses onset, frequency, methods, and severity of NSSI. It is transdiagnostic and developed for clinical settings.

The purpose of the study is to evaluate the feasibility of a TSI validation study. The study will also provide preliminary validation of the instrument.

The feasibility study will recruit participants at in- and outpatient units from a university hospital. Participants can be included in the study if they are 18 years old and admitted to a psychiatric in- or outpatient unit.

Instruments: The Deliberate Self-Harm Inventory will be used to test concurrent validity. Convergent validity will be tested with the Columbia Suicidality Severity Rating Scale, the Personal and Social Performance scale, the Affective Lability Scale-short, and the Brief Trauma Questionnaire. Interrater reliability will be evaluated in groups of medical doctors, psychologist, and other clinical professionals.

Feasibility are measured by inclusion of participants per week, the time each participant takes to complete the study instruments, and number of dropouts.

Recruitment of participants will start in the fall of 2021. We aim to recruit 50 participants.

When TSI has been validated, it can be used to assess prevalence and severity of NSSI and clarify the need for treatment and supervision.

No significant relationships.

Neurocognitive and social cognitive impairments are central characteristics of schizophrenia and, to a lesser extent, of bipolar disorder. Birth cohorts and familial high risk studies have described cognitive impairments in subjects before onset of diagnosis as well as in children with increased genetic risk for development of the disorders.

To our knowledge, this is the first study to investigate the correlations between neurocogntion and social cognition in parents and offspring simultaneously and with the same methodology. We will divide the parents into subgroups (cognitive impairment and good cognitive functioning) and use these subgroups to describe correlations with their offspring. Identifying associations between parents and offspring can add important clues to risk factors for schizophrenia and bipolar disorder and, on the long-term, help the development of more effective and potentially preventive treatments.

This study is part of the Danish high risk and resilience study–VIA7. The VIA7 cohort consists of 522 children age 7 with zero, 1 or 2 parents diagnosed with schizophrenia or bipolar disorder and both of their biological parents. We assessed neurocognition and social cognition with a comprehensive test battery including: intelligence (RIST), executive functions (WAIS-IV, D-KEFS, CANTAB), verbal memory (TOMAL2), attention, emotion recognition, decision making and response control (CANTAB), theory of mind (animated triangles) and social perception (TASIT). Parental subgroups were based on the 95% CI of the controls (cognitive impairment < 95%CI and good cognitive functioning > 95% CI).

Data analysis is ongoing and results will be presented at the conference.

The authors have not supplied their declaration of competing interest.