Follow us on Google News (click on ☆)
"This first major project is a great demonstration of the potential contribution of AI to prevention in mental health and addictions," says Christian Gagné, professor at the Faculty of Science and Engineering at Université Laval and director of the Institute for Intelligence and Data.
Fatemeh Gholi Zadeh Kharrat, a postdoctoral fellow at Université Laval, incorporated ecological data, related to demographics or the environment, and anonymized individual data collected between 2000 and 2019. She analyzed statistics related to the population, drug insurance, healthcare accessibility, and much more.
Understanding to better prevent
The initiative confirmed existing hypotheses on the subject and also uncovered new insights. "For example, we found that people who had received mental health follow-ups within the 60 days prior had a higher risk of suicide. The same goes for drug use. This is the kind of relationship we expected to see, but the machine learning analysis clearly showed their incidence," says Christian Gagné, who worked closely with Alain Lesage, a professor at the Faculty of Medicine at Université de Montréal.
AI models also showed that mental health and addiction disorders are key factors in predicting suicide. They highlighted the cumulative effect of risk factors, both related to the individual and the context in which they evolve. What happens at the individual level is also influenced by ecological factors, such as the regional budget for mental health and addictions.
The analyses also revealed that the suicide mortality rate among men is higher in regions where the addiction-related per capita budget is lower. "So, a clear link has been established between public investment levels in mental health and addictions and the risk of suicide. Conversely, if we increase the funding, there is a real effect in reducing this risk," professor Gagné emphasizes.
600 variables studied
This type of relationship was highlighted thanks to the consideration of a substantial number of variables. Socio-demographic situations, diagnoses and hospitalizations, physical or mental health history, regional mental health budgets—about 600 clinical or societal variables were taken into account. "We were able to see how rich the data provided by the INSPQ was!" says Fatemeh Kharrat.
Two AI models quickly emerged, defined by gender. "The differentiation of male and female risk factors is something already well understood by the clinical community. By developing gender-specific models, we were able to highlight other variables of interest, and even identify variables that might be gender-specific," notes Christian Gagné.
Over the course of simulations, the team measured the impact of various variables to identify the most influential factors. "If we adjust the social disadvantage of the person's neighborhood, what effect does that have on the risk factors?" illustrates Fatemeh Kharrat. "We were able to understand the relationships between the variables and their effect on risk levels."
For result interpretation, the researchers worked with specialists in the field. "They could verify whether the relationship exists or is likely from a clinical perspective," adds Christian Gagné.
This project, supported by funding from the New Frontiers in Research Fund from Canada's three research councils, has been the subject of scientific publications in the journals PLOS One and JMIR Public Health and Surveillance.