Improving Prediction of Mental Disorders Using Multimodal Data

This research project is part of the international Accelerating Medicines Partnership – Schizophrenia (AMP SCZ) program, an initiative to accelerate the development of treatments and preventive approaches in mental health, particularly for schizophrenia and psychotic disorders.

When it comes to mental health, prevention and prediction are major challenges. Unlike many physical illnesses, there is no single blood test, medical imaging, or biomarker that can clearly assess the risk of developing a mental disorder such as schizophrenia. Diagnosis still relies heavily on observing symptoms, which can often be confused with those of other illnesses. This reality complicates both the understanding of individual trajectories and the early implementation of interventions.

Leveraging Multimodal Data for a Better Understanding of Trajectories

This is the context for a large-scale research project involving ÉTS Professor Sylvain Bouix. The goal is ambitious: to collect and analyze as much “multimodal” data as possible—cognitive tests, clinical questionnaires, brain imaging, biological signals, behavioural and language data—to better understand different mental health risk profiles. By combining these sources of information, researchers are seeking answers to crucial questions: Are the observed symptoms transient or persistent? Will they worsen over time? Will the person on a trajectory towards full-blown psychosis require hospitalization?

Beyond prediction, the ultimate goal is prevention and improved treatment. Today, prescribing medication is mainly based on observed symptoms. Ultimately, researchers hope to intervene earlier, offering treatments to prevent the development of certain mental illnesses. Access to detailed biological and cognitive data also paves the way for the development of new molecules targeting specific types of brain cells or receptors, primarily to reduce the cognitive deficits associated with these disorders.

Patient Classification: Towards Personalized Medicine

The project's goal is also to classify participants into well-defined subgroups according to their characteristics and trajectories. This approach, common in oncology through genetic testing, is part of the personalized medicine approach. However, in mental health, the lack of clear markers further complicates the process. Currently, the risk of psychosis is mainly assessed through two- to three-hour clinical interviews focusing on symptoms, life history, family background, and the prevalence of mental disorders in the family.

Participants in the study undergo an impressive series of tests. These include brain MRIs, electroencephalograms, blood tests for genetic analysis, saliva samples to measure different protein levels, and advanced cognitive tests. Open discussions are also recorded, during which participants describe their daily lives, social relationships, or activities.

These recordings are transcribed and then analyzed using tools that assess vocabulary richness, topics discussed, voice fluctuations, emotional expression, and nonverbal expressions. For some participants, monitoring goes even further with a mobile app that collects daily audio logs, mood data, and passive information such as activity level and location. A scientific watch equipped with a motion sensor completes the device.

Participants are monitored for two years as part of an international study involving approximately 40 sites across several continents. About 2,500 people were recruited, some of whom will, unfortunately, develop a full-blown psychosis. At that point, the objective will be to determine what sets them apart from others well before the most severe symptoms appear, based on the entire data collected.

Analyzing the data is done through a combination of statistical bioinformatics and artificial intelligence methods. It is one of the richest and most extensive datasets ever compiled for the study of high-risk psychotic disorders.

Data Structuring, Harmonization and Enhancement

As part of this extensive collaboration, Sylvain Bouix plays a pivotal role in the data infrastructure. He ensures that all collected data is stored in a centralized database, and that it is clean, harmonized and comparable, regardless of its origin. 

Rather than focusing on predictive models, he deals with data processing and preprocessing, applying the principles of software engineering, such as DevOps. The goal is to quickly detect technical problems, like a malfunctioning sensor, and provide immediate feedback to the site teams. This approach prevents the loss of valuable data, which is extremely costly to collect, and maximizes the scientific value of the study.

Note that the data is made available to the larger scientific community through the National Institute of Mental Health Data Archive (NDA).

Partners

The AMP SCZ program is managed by the Foundation for the National Institutes of Health (FNIH) and is based on a broad international partnership bringing together stakeholders from the public and private sectors.

Public Sector Partners

• European Medicines Agency (EMA)
• National Institute of Mental Health (NIMH)
• U.S. Food and Drug Administration (FDA)

Private Sector Partners

• AbbVie
• American Psychiatric Association Foundation
• Boehringer Ingelheim
• Johnson & Johnson Innovative Medicine
• National Alliance on Mental Illness
• One Mind
• Otsuka Pharmaceutical Development & Commercialization, Inc.
• Schizophrenia & Psychosis Action Alliance
• Wellcome

More detailed information about the AMP SCZ program is available on their web site.

Sylvain Bouix is a member of itechsanté, the ÉTS research institute for innovation in health technologies. To learn more about the institute, its mission, themes, flagship projects, and more, visit itechsanté.