Can online assessment of speech predict psychosis-onset in people at clinical high risk of psychosis?
Dr Kelly Diederen
Lecturer in Psychosis Studies
Department of Psychosis Studies, Institute of Psychiatry, Psychology and Neuroscience (IoPPN), King’s College London
Dr Tom Spencer
Clinical Senior Lecturer in Early Intervention Honorary Consultant Psychiatrist OASIS Lewisham and OASIS Croydon Education co-Lead
Division of Academic Psychiatry Department of Psychosis Studies Institute of Psychiatry, Psychology & Neuroscience (IoPPN), King’s College London
Formal thought disorder (FTD), a marked disturbance in the organisation of thought expressed in patients’ speech, is a cardinal feature of psychosis and is predictive of poor clinical and social and occupational outcomes (Yalincetin et al., 2017). It is, therefore, crucial to assess FTD in early psychosis, to effectively tailor interventions to those most at risk of poor outcomes. It is unclear, however, whether FTD is already evident in the prodromal phase of illness, when individuals can be identified as being at clinical high risk for psychosis (CHR-P).
Recently, automated methods such as semantic and graph analysis have been developed to assess FTD based on speech samples (Corcoran et al., 2018; Mota et al., 2017). Using these novel methods, we found that people with first episode psychosis have disconnected speech compared with healthy controls and CHR-P groups. Importantly, lower scores in the CHR-P group were predictive of subsequent transition to psychosis, although this was in a small sample and thus requires large-scale replication.
Due to developments in online testing, speech samples can be acquired on a large-scale in people at CHR-P. This cost-effective and widely available approach means online assessment of speech is a promising marker to stratify this group, which may improve clinicians’ ability to target treatment to those who are most vulnerable.
The main objective of this PhD is to determine whether altered measures of speech predicts transition to psychosis within 2 years in 300 individuals at CHR-P. The secondary objective is to investigate whether these measures predict social and occupational outcomes. The student will analyse speech samples acquired online (via Gorilla.sc) as part of an ongoing multicentre study. Speech measures will be generated from state-of-the art sentence embedding to produce sentence-to-sentence semantic coherence using Python code written by the supervisors and software to calculate graph connectedness. Statistical analysis will use generalised linear models and structural equation modelling to test the prospective relationship between speech and clinical and functional outcomes. Machine learning will be used to estimate the risk of future psychosis in the CHR-P group at the level of the individual. There is also an opportunity to analyse speech data in people with established psychotic illness, and in people with subclinical psychotic symptoms in the general population.
Learning Python, R and automatic speech analysis, under supervision of Dr Diederen and Spencer who have already successfully employed these analyses
Training in general academic skills (e.g., conference presentations, literature review and writing scientific manuscripts) as well as specific analyses (e.g. prediction modelling, machine-learning)
Write the upgrade report and preparing for the upgrade viva
There is also the possibility to identify additional datasets and analytical techniques for the student to work on
Presenting at a conference (e.g., the annual meeting of the Schizophrenia International Research Society)
Preparing manuscripts for submission
Presenting at a conference (e.g., the meeting of the European Conference on Schizophrenia Research)
Exploring future career steps/applying for postdoctoral fellowships
ePREDICT – 300 individuals at clinical high risk of psychosis, 300 people at first episode psychosis, 100 healthy controls. People are followed up for a period of two years with speech assessed every six months. Clinical assessments and functional outcomes measured at baseline, 12 months and 24 months. Ethical approval obtained via REC.
ONASOS (ONline ASsessment Of Speech) study: Online assessment of speech (using a more extensive set of speech tasks than the ePREDICT study) in 450 people in the general population. Psychiatric symptom scores, with a focus on subclinical psychotic symptoms assessed using well validated questionnaires. Ethical approval obtained via REMAS.
Formal Thought Disorder, speech, psychosis, schizophrenia, online testing