Applying computer analysis to predict suicidal thoughts
While science has made significant progress in reducing leading causes of death such as HIV/AIDS, heart disease and cancer, progress in suicide reduction has essentially been non-existent. “Suicide is still among the leading causes of death in Europe and worldwide,” says the PS project’s Marie Curie global fellow Brian O’Shea. O’Shea is based at the University of Amsterdam in the Netherlands, and at Harvard University (United States of America). “It is the second leading cause of death – after accidents – for European adolescents.” In part this is because predicting suicide remains incredibly challenging. “Suicide has also been historically strongly stigmatised, and is still illegal in some countries. Perceived stigma can reduce the likelihood that people will disclose suicidal thoughts.”
Identifying suicidal patterns
The PS project, undertaken with the support of the Marie Skłodowska-Curie Actions programme, sought to address this challenge by analysing a specific set of computer-based Implicit Association Tests (IATs). These tests, called Death and Suicide (D/S)-IAT, involve sorting words related to Me (Self, My, I), Not Me (They, Them, Other), Life (Alive, Living) and Death/Suicide (Die, Dead). The test has been continually running online since 2012, and to date has been completed by over 12 000 volunteers. “Research has shown that the D/S-IAT can predict the recency and severity of self-harm,” says O’Shea. “Because I worked with Matthew Nock, a professor at Harvard University, and Bethany Teachman, a professor at the University of Virginia, who set up projectimplicithealth.com (Project Implicit Health), I had full access to the D/S-IAT data for this project.” O’Shea set about developing and validating a new method for analysing the IAT results. His aim was to achieve a better understanding of what differentiates suicide attempters from non-attempters. Forecasting techniques were also used to determine whether trends across years, seasonal changes, day of the week, etc. also play a role in self-harm and suicide. “My results showed that a weakened association between ‘Me = Life’ is more strongly predictive of having a history of suicide attempts,” adds O’Shea. “However, among those who previously attempted suicide, a strengthened association between ‘Me = Death’ is more strongly predictive of the recency (and frequency) of a suicide attempt.” Therefore, ‘Me = Death’ associations may be particularly useful when distinguishing between those with an imminent risk of suicide and those with other psychiatric conditions. O’Shea’s analysis was also able to show that suicidal thoughts are highest in December, preceding the peak of suicidal behaviour in late spring and early summer.
Suicide forecasting findings
With further advances in machine learning, O’Shea believes the methods pioneered in the PS project hold great potential for detecting those most at risk of suicide. “Although more work is still needed, testing is currently underway with Reinout Wiers, a professor at the University of Amsterdam, and the Salus Clinic Lindow (website in German),” he says. “Regarding our suicide forecasting findings, these will likely be relevant for policymakers when determining the availability of emergency or suicide support services.” O’Shea has recently secured a Japan Society for the Promotion of Science Fellowship to continue this line of research. “We will address cross-cultural variation in loneliness and its impact on suicide, at both the individual and the regional levels of analysis,” he remarks. “I also intend to lead more projects that tackle the terrible effects that suicide has on our communities.”
Keywords
PS, suicide, death, self-harm, stigma, suicidal, psychiatric
