Voice-Based Biomarkers: a Novel Approach to Monitoring and Predicting Schizophrenia Relapses

Schizophrenia is a highly complex disease of unknown origin. It is characterized by a heterogeneous etiology and variable clinical manifestations. Numerous studies have investigated genetic, biochemical or neurodevelopmental factors of the pathology, without providing a definitive answer. The environmental factor is also analyzed, especially urban density and pollution. Finally, the immune pathway, in particular neuroinflammation, has been studied. Certain parasites, such as Toxoplasma, may be involved in the development of schizophrenia in combination with genetic factors. All these studies tend to show that the parasite/gene/environment association would influence the development of schizophrenia, but the origins remain unclear and a more complete knowledge of the pathophysiology is needed to improve diagnosis and patient management through new therapeutic targets.

More than 80% of patients with schizophrenia have language abnormalities. These abnormalities are manifested in syntax, semantics and phonology. The most common include monotonous intonation, poverty of speech, increased pauses, lack of spontaneity, and disruption of speech coherence. As speech is an important factor in social interaction, patients have great difficulty in maintaining their social relationships.

A major problem in schizophrenia is the discontinuation or misuse of antipsychotic treatment, which leads to relapse and additional hospital costs. According to a 2013 study, 50% of patients discontinue treatment after six months, often leading to decompensations.To avoid relapses, clinicans can either administer hetero-questionnaire to monitor the patients' symptoms or monitor treatment adherence.

To monitor symptoms, clinicians have at their disposal various standardized questionnaires such as the PAANS. However, those tests are time-intensive.

To monitor treatment adherence, clinicians can use blood drug concentrations as an evidence of compliance, although this method is invasive, and requires costly administration coordination between healthcare profesionnals and patients. There are also several standardized tests to monitor adherence, such as the clinician-administered BARS questionnaire and the self-administered BEMIB. These measures have often been criticized due to factors such as recall bias and poor self-perception, which limit the accuracy of patient reports and overestimate adherence.

It is therefore essential to develop new tools to objectively measure evolution of symptoms and treatment effects to detect onset of relapses, without increasing the burden on patients' daily lives.

Speech voice markers stand out because they have characteristics that make it easy to use in clinical practice and can be easily integrated into patients' daily lives. Advances in signal processing and machine learning algorithms now make it possible to measure the different components of speech: phonatory skills, articulation, the different linguistic levels (semantics, syntax, morphology, pragmatics) as well as the specific disfluencies of spontaneous speech.

These different speech markers have been validated in different neurological and psychiatric pathologies: in particular, Parkinson's disease, Huntington's disease, depression, suicidal risk, and schizophrenia. These markers of speech in psychiatry are now generalizing across languages and are also being taken into patients' homes to measure changes in patient states.

Distinctive voice characteristics have been a feature of schizophrenia since it was first defined. They are often associated with negative symptoms, such as the inability to show emotion, and with observed social impairments. It has been quantitatively observed that people with schizophrenia have poorer speech, more pauses, distinctive tones, and differences in voice intensity.

Studies have shown antipsychotics also affect language, since patients are dopamine-deficient, and blockade of these receptors by antipsychotics would exacerbate language impairments. In addition, blockade of the striatal dopamine receptor leads to extrapyramidal side effects, causing tremors, muscle rigidity, and tics that interfere with the joint movements required for speech.

Therefore, voice and language are very good candidates as biomarkers in monitoring both symptoms and treatment effects in schizophrenia to detect onsets of relapses.

The aim of this study is to determine whether voice biomarkers can be used to objectively monitor symptoms and treatment effects in schizophrenia in order to detect onset of relapses.

Inclusion criteria

• Affiliated to a social security or other social protection scheme
• Diagnosed with schizophrenia according to the DSM-5 (code F20)
• Stable state without duration criteria
• Main treatment with risperidone or paliperidone, aripiprazole, olanzapine in oral or long-acting injectable form
• Complementary treatment with an antipsychotic for anxiolytic or sedative purposes: chlorpromazine, loxapine or cyamemazine
• Able to speak and read French
• Able to perform speech evaluations
• Able to answer questionnaires on smartphone
• May be under curatorship or guardianship
• Agreeing to participate in the study and with informed consent signed by the subject, as well as by the legal representative in the case of a person under curatorship or guardianship

Exclusion criteria

• Suffering from a pathology that impairs French
• Suffering from a neurological pathology: multiple sclerosis, Parkinson's disease, Huntington's disease or neurodegenerative disease
• Treatment with lithium salts
• Anti-epileptic treatment
• Participant in another study involving medication