Artificial intelligence may improve suicide prevention in the future

The loss of any life can be devastating, but the loss of life through suicide is especially tragic.

Every day around nine Australians take their own life and it is the leading cause of death for Australians aged 15-44. Suicide attempts are more common, with some estimates 30 times more common than fatalities.

“Suicide has serious consequences when it occurs. It affects many people and has far-reaching consequences for family, friends and society,” says Karen Kusuma, PhD, UNSW Sydney. candidate in psychiatry at the Black Dog Institute, researching adolescent suicide prevention.

Ms. Kusuma and a team of researchers from the Black Dog Institute and the Center for Big Data Research in Health recently explored the evidence base for machine learning models and their ability to predict future suicidal behavior and thoughts. They evaluated the performance of 54 machine learning algorithms previously developed by researchers to predict the suicide-related outcomes of ideation, attempts, and death.

A meta-analysis published in Journal of Psychiatric Researchfound that machine learning models outperform traditional risk prediction models in predicting suicide-related outcomes, which traditionally perform poorly.

“Overall, the results show that there is a preliminary but compelling evidence base that machine learning can be used to predict future suicide outcomes with very good performance,” says Ms. Kusuma.

Traditional models of suicide risk assessment

Identifying people at risk of suicide is important to prevent and control suicidal behavior. However, predicting risks is difficult.

In emergency departments, clinicians routinely use risk assessment tools such as questionnaires and rating scales to identify patients at increased risk for suicide. However, evidence suggests that they are ineffective for accurate prediction risk of suicide in practice.

“While there are some common factors associated with suicide attempts, what the risks look like for one person may look very different for another,” says Ms. Kusuma. “But suicide is complex, with many dynamic factors that make it difficult to assess the risk profile using this assessment process.”

A post-mortem analysis of people who took their own lives in Queensland showed that of those who received an official suicide risk assessment, 75% were classified as low risk and none were classified as high risk. Previous studies looking at quantitative prediction models for suicide risk over the past 50 years have also found that they are only slightly better than chance at predicting future suicide risk.

“Suicide is the leading cause of years of life lost in many parts of the world, including Australia. But the way we assess suicide risk hasn’t evolved recently, and we haven’t seen a significant reduction in suicide rates. After a few years, we saw an increase,” says Ms. Kusuma.

Despite the lack of evidence in favor of traditional suicide risk assessment, their use remains standard practice in health care settings to determine the level of care and support for the patient. Those identified as high-risk generally receive the highest level of care, while those identified as low-risk are discharged.

“Using this approach, unfortunately, high-level interventions are not being delivered to people who really need help. Therefore, we need to reform the process and explore ways to improve suicide prevention,” says Ms. Kusuma.

Machine learning suicide screening

Ms. Kusuma says more innovation in suicidology and a reevaluation of standard models for predicting suicide risk are needed. Efforts to improve risk prediction have led to the use of her research artificial intelligence (AI) to develop suicide risk algorithms.

“Having artificial intelligence that can take in a lot more data than a clinician can better recognize patterns associated with suicide risk,” says Ms. Kusuma.

In a meta-analysis study, machine learning models outperformed benchmarks previously established by traditional clinical, theoretical, and statistical models for predicting suicide risk. They correctly predicted 66% of people who would commit suicide and correctly predicted 87% of people who would not commit suicide.

“Machine learning models can predict suicide mortality well compared to traditional prediction models and can be an effective and efficient alternative to conventional risk assessments“, says Ms. Kusuma.

The strict assumptions of traditional statistical models do not bind machine learning models. Instead, they can be flexibly applied to large datasets to model the complex relationships between multiple risk factors and suicidal outcomes. They can also incorporate adaptive data sources, including social media, to identify peaks in suicide risk and highlight times when intervention is most needed.

“Over time, machine learning models could be configured to handle more complex and larger data to better identify patterns associated with suicide risk,” says Ms. Kusuma.

The use of machine learning algorithms to predict suicide-related outcomes is still an emerging area of ​​research, with 80% of the identified studies published in the last five years. Ms Kusuma says future research will also help address the risk of aggregation bias found in algorithmic models to date.

“Further research is needed to improve and validate these algorithms, which will then help advance applications machine learning in suicidology,” says Ms. Kusuma. “Although we are still a long way from implementation in clinical settings, research shows that this is a promising way to improve the accuracy of suicide risk screening in the future.”