AI has been taught to assess suicide and self-harm risk in adolescents
Researchers have used machine learning to identify factors that influence the risk of suicide and intentional body harm (self-harm) in adolescents.Researchers from the University of New South Wales in Sydney have developed AI to assess the risk of suicide and self-harm in adolescents. Unlike existing approaches, which rely only on recorded previous attempts at dangerous behavior, the model uses more than 4,000 factors.
Researchers used long-term follow-up data from the LSAC Australian Children's Study. The researchers tracked the children's development and their environment since 2004. The database collected information about health, family, social, economic and cultural environment. In addition, children, parents or guardians and educators were regularly surveyed.
To train the neural network, the researchers selected 2,809 study participants, who were divided into two age groups: 14-15 and 16-17 year olds. 10.5% reported intentionally self-injuring (self-harm) and 5.2% reported attempting suicide at least once in the past 12 months.
The AI identified more than 4,000 potential risk factors related to mental and physical health, relationships with others, school and home environments. The researchers used random forest (a machine learning algorithm) to determine which traits at age 14-15 best predicted suicide attempts and self-harm at age 16-17.
Key risk factors were depressed feelings, emotional and behavioral difficulties, self-perception problems, and school and family dynamics. Meanwhile, low self-efficacy - a lack of belief in one's own future and the effectiveness of one's actions - influenced suicide risk, and impaired emotional regulation led to self-harm.
The researchers also noted that the presence of a history of previous suicide attempts or self-harm, which are often used to distinguish risk groups, were not key factors. That said, the environment at school and at home influenced much more than previously thought. This feature could be used for prevention, the researchers said.
To implement predictive models in clinical practice, more research is needed, the authors of the paper note. It is necessary to check whether the model will work on a smaller amount of data in electronic medical records of patients.