Using deep learning to identify adolescents most in need of mental health support

The personal but global struggle with mental health is perhaps more visible than ever. Yet many people still struggle to access the support they need.

In Japan, suicide is unfortunately the leading cause of death among young people. Researchers, including from the University of Tokyo, conducted a six-year study to better understand the myriad factors that can impact adolescent mental health. After surveying 2,344 adolescents and their caregivers, and using computer-based deep learning to process the results, they were able to identify five categories into which young people could be grouped.

Nearly 40% of those involved were classified as groups experiencing some problems. Among them, almost 10% were living with mental health problems that had not been identified by their caregivers. This group was most at risk of self-harm and suicidal ideation. Identifying the factors that can lead young people to suicide and determining who is most at risk is essential to supporting prevention and early intervention efforts.

Last year in Japan, 514 youth and children aged 18 or younger tragically lost their lives to suicide. This is the highest figure for this age group since records began in 1978. Suicide is the leading cause of death among people aged 15 to 34, according to data from the Japanese Ministry of Health , Labor and Social Protection. While suicide rates among adults have generally declined over the past 10 to 15 years, the opposite has been observed among adolescents. Officials believe school-related issues, difficult personal and family relationships and the lingering impacts of the pandemic may have contributed to the high number of deaths.

The World Health Organization (WHO) identifies suicide as a major global public health problem, but also says it can be prevented through evidence-based interventions and by addressing the factors that can lead to poor mental health. Researchers from the University of Tokyo and the Tokyo Metropolitan Institute of Medical Sciences are analyzing data on various adolescent problems that were assessed by both themselves and their caregivers, resulting in identify young people likely to be at risk of suicide.

“We recently found that adolescents considered to have no problems by their caregivers actually had the highest risk of suicide,” said Daiki Nagaoka, a doctoral student in the Department of Neuropsychiatry at the University of Tokyo and a hospital psychiatrist. “It is therefore important that society as a whole, rather than relying solely on carers, plays an active role in recognizing and supporting adolescents who have difficulty seeking help and whose distress is often overlooked .”

The team interviewed adolescents and their caregivers in Tokyo over a six-year period. Participants completed self-report questionnaires, answering questions about psychological and behavioral problems such as depression, anxiety, self-harm and inattention, as well as their feelings about family and school life. The team also took note of factors such as maternal health during pregnancy, involvement in harassment and the psychological state of caregivers.

Now published in The Lancet Regional Health – Western Pacificthe study began when the children were 10 years old and was repeated with them at 12, 14 and 16 years old. A total of 3,171 adolescents participated, with 2,344 adolescent couples and their guardians participating throughout the full study.

“Psychiatry faces challenges in understanding adolescent psychopathology, which is diverse and dynamic. Previous studies typically classified adolescent psychopathological development based on the trajectories of only two or three indicators. In contrast, “Our approach enabled the classification of adolescents based on a number of symptom trajectories simultaneously by employing deep learning techniques that facilitated a more comprehensive understanding,” Nagaoka explained.

Deep learning, a computer program that mimics our brain’s learning process, allowed the team to analyze the large amounts of data collected to find patterns in the responses. By grouping together the trajectories of psychological and behavioral problems identified in the survey, they were able to classify adolescents into five groups, which they named based on their key characteristic: unaffected, internalizing, discordant, externalizing, and severe.

The largest group, 60.5% of the 2,344 adolescents, consisted of young people classified as “unaffected” by suicidal behavior.

The remaining 40% were affected in one way or another. The “internalizing” group (16.2%) persistently internalized their problems and experienced depressive symptoms, anxiety, and withdrawal. The “divergent” group (9.9%) presented with depressive symptoms and “psychotic-like experiences,” but were not recognized as having such problems by their caregivers. The “externalizing” group (9.6%) presented with hyperactivity, inattention and/or behavioral problems but few other problems.

Finally, the smallest group was classified as “severe” (3.9%) and faced chronic difficulties of which their caregivers were aware, including psychotic-like experiences and obsessive-compulsive behaviors.

Of all groups, youth in the “divergent” category were most at risk of self-harm and suicidal thoughts. The researchers found that they could significantly predict who would be included in this group based on whether the child avoided seeking help for depression and whether their caregiver also had a mental health problem.

Researchers suggest that the mental state of the caregiver could impact the adolescent’s mental health due to genetic factors and the parental environment, such as the caregiver’s ability to pay attention to the difficulties facing a child. teenager could face. Although this research has several limitations, it nevertheless allowed the team to identify a number of risk factors that could be used to predict which groups adolescents might belong to.

“In my daily practice as a psychiatrist, I have observed that existing diagnostic criteria often do not adequately address the diverse and fluid difficulties faced by adolescents,” Nagaoka said. “We sought to better understand these difficulties so that appropriate support can be provided. Next, we want to better understand how adolescents’ psychopathological problems interact and evolve with the people and environment around them. Recognizing that many adolescents face serious challenges and problems but are hesitant to seek help, we need to establish support systems and structures as a society.