This text has been translated from Norwegian with the assistance of GPT UiO.
– When I noticed that suicidal behavior among children and adolescents is understood in light of theories of adult suicide, it was a red flag for me, says Haghish Fardzadeh.
The German Iranian researcher is a former post-doctoral fellow at the University of Oslo. In May of this year, he defended his dissertation at the Department of Psychology, focusing on adolescent suicidal and self-destructive behavior.
The project partially arose from his observation of a need to study adolescent suicidality in a holistic way. He aimed to develop a theory that not only explains suicidal tendencies, but also self-destructive behavior among youth. By doing so, he could include other life-threatening behaviors, such as risk-taking behaviors and eating disorders – and consider everything important to healthy development.
Haghish realized that he had a strong foundation for researching this topic, having a doctorate in applied statistics prior to this work.
Clusters of risk factors
In his new dissertation, entitled "Theory of Adolescent Suicidality and Self-Destruction. A Holistic Multi-Model Domain Analysis," he has employed AI models to review large datasets containing information about Norwegian youth. This has allowed him to study the topic in a much broader context than he perceives as typical research in this field.
– The most common practice is to model just a few risk factors that may cause a problem, so that we can try to prevent these factors, he says, pointing out the connection between smoking and cancer as an example from medical research.
– But one of the perhaps most surprising findings in my research is that suicidal behavior cannot be linked to a few risk factors but rather clusters of factors, or domains, which are essential for healthy development and well-being.
Using a large number of different AI models, he was able to pinpoint six such domains that can either help prevent or exacerbate the risk of suicide attempt, self-harm, and self-destructive behavior among youth. Everything depends on how adolescents orient themselves toward these six domains, which are:
Affect and self: How they think and feel about themselves.
Interpersonal connectedness: How strong and positive their relationships are to family, friends, teachers, and significant others, reflecting their interpersonal and social well-being.
Insecurity and vulnerability: How insecure they feel, both in life in general and when trying to achieve their goals, and how vulnerable they are to traumatic experiences.
Future, hopelessness, and optimism: What thoughts and expectations they have about what may happen in the future, and whether they will succeed in education, career, and personal life.
Autonomy and resilience: To what extent they feel capable and independent when facing problems in all these personal, interpersonal, environmental, and future
The body and the relationship to the body: The body undergoes rapid changes during adolescence. How young people perceive, understand, use, and present their bodies, and how they control and regulate their physical lives, including sleep, food, and physical well-being, is important. Both carelessness and excessive control are red flags that may indicate reduced physical well-being.
All or none
The findings are both discouraging and positive at the same time, believes Haghish.
– If we wish to find a pill that prevents suicide attempts and self-destructive behavior among youth, then this is a disappointing result, because my research shows that such a pill does not exist, and it likely never will. But in the long run, it may be advantageous to become aware of the pressures this age group undergoes, he says, explaining that this can lay the groundwork for future preventive measures:
– Regarding these six domains, it appears to be all or none. It is not enough to fix one of them. If we want to effectively reduce the risk, we must take a holistic approach. We need to improve the situation in all six domains, he says.
The six domains and the use of AI models to establish them may also represent a new way to assess the risk of suicide and self-destructive behavior among young people. Identifying who is at risk has always been very difficult, Haghish points out. This often involves having youths respond to very personal questions, with all the ethical challenges it brings, especially in research. Additionally, it can be challenging to know how well their answers reflect reality.
Could avoid sensitive questions
Haghish's dissertation shows that it is possible to assess suicide risk without using direct suicide-related information. The youths do not need to answer, for example, whether they have had or currently have suicidal thoughts, suicide plans, self-harming behavior, or previous suicide attempts. It is sufficient to ask them to reflect on how well they are doing within the six relevant domains.
– If we have access to information about all these domains, we can assess the risk, and these assessments have proven to be very accurate, he says, emphasizing that this does not mean we should not talk about suicidal thoughts and behaviors.
The data of the research is from the Norwegian Ungdata survey, collected between 2014 and 2019, with 173,664 respondents aged 13 to 18 years. According to Haghish, his analysis shows that the six domains can be used not only to assess suicide attempt risk but also, for instance, eating disorders. Suicidal behavior among youth is indeed closely linked to other self-destructive behaviors.
This does not, of course, mean that all self-destructive behavior can lead to suicide. In a recently published article related to the dissertation, he argues that the method he has employed can differentiate between suicidal and non-suicidal self-harm, to detect who is truly at risk. He cites this as proof of how precise the model is.
Suicide attempts and self-destructive behavior among youth do not necessarily result from a rational decision, as many philosophical texts about adult suicide assume, believes Haghish.
– My research shows that it involves a continuous process. Much of the self-destructive behavior may be unconscious, and yet it can precede suicidal behavior. Young people may not be able to explain why they engage in self-destructive and potentially suicidal behavior – and yet, by using a holistic approach with AI models, it may be possible to both predict and prevent future suicide attempts and life-threatening behaviors.
At the same time:
– This model suggests that if we want to reduce suicide and self-destructive behavior in this age group, the only way to do so is to enhance their well-being across all domains. But this is just an indication for future research, he says, emphasizing that much needs to be validated by later studies.
Many different AI models
He has used not just one but hundreds of AI models to analyze the data and reach his conclusions. The AI models made it possible to pinpoint the relevant indicators within the enormous dataset. And because he employed several different models, he could compare the results of the various AI models and ensure reliability and generalizability of the findings. What he discovered turned out to be among the most accurate suicide attempt risk assessment models in the scientific literature, according to Haghish.
– AI models are not necessarily identical and consistent; they can produce very different answers based on the same data. This is one of the challenges of this type of research that is not addressed much, he believes.
– Therefore, I used multiple models, so I could check if the findings were compatible and take such discrepancies between the AI models into account. I developed a new statistical method to account for such incompatibilities.
He believes there is a need for much more research in this field, but he hopes that his findings can serve as a basis for further studies, and? – if eventually confirmed by further research – maybe even as a basis for new preventive approaches.