A new AI model that can accurately predict which children are at highest risk of a first suicide attempt during adolescence has been developed by Orygen researchers, and could enable earlier identification of, and intervention in, suicidal behaviour in this high-risk group.
The newly developed algorithm uses existing clinical, demographic and psychosocial data to assess 187 risk-factors, and a new study published in Child & Adolescent Psychiatry, which used data drawn from a longitudinal study of more than 600 suicidal children aged 9-10, has found it can address existing gaps in how risk is assessed.
Lead author and PhD candidate at Orygen, Josh Nguyen, said prediction models had the potential to greatly enhance current youth suicide prevention efforts by identifying those who may need more intensive interventions.
"We know that about one third of individuals who have suicidal thoughts go on to make an attempt, but currently our ability to predict those high-risk individuals is only slightly better than chance,” Nguyen said.
“Sadly, half of suicide deaths occur during the first attempt, and those who survive are at higher risk of subsequent attempts and severe mental ill-health.
"If we can identify this risk early, it allows us to intervene before the first attempt occurs, which could not only save lives but also reduce the huge social and economic costs associated with the first onset of suicidal behaviours in adolescence."
The study aimed to address the limitations of existing models and studies by testing a broad range of sociodemographic, clinical, neurocognitive, and imaging data.
“What is very promising about the potential utility of this model is that whilst we tested for over 180 diverse risk factors, we found that most key predictors are things we can measure in routine care, target and change," Nguyen said.
"This includes how severe a young person's suicidal thoughts are, if they experience anxiety or impulsivity, whether they have self-harmed, have access to means, or have lower parental income and a history of mental health treatment.
"Knowing the key predictors to look out for offers hope for more precise, early intervention strategies in adolescent suicide prevention."
Using these predictors, the model correctly classified approximately 7 out of each 10 young people in terms of whether or not they transitioned.
“Adolescence is a high-risk period for the first onset of suicidal behaviour, yet most current predictive models and frameworks for predicting those at highest risk are based on adults and are not necessarily applicable to young people – this new model changes that.”
Next steps will involve further external validation and feasibility studies before clinical implementation, however the new model offers a significant step forward in demonstrating that machine learning models can play a vital role in early intervention efforts.