Just ask the computer

Article in Review: D’Hotman, D. & Loh, E. (2020). AI enabled suicide prediction tools: a qualitative narrative review. BMJ Health Care Inform, doi:10.1136/bmjhci-2020-100175

Summary: This article is the first of a four part series examining the article by D’Hotman & Loh (2020) on artificial intelligence (AI) and seeking to better understand the capability of using such tools for suicide prediction – from a medical and a social perspective. The article summaries will consider some of the challenges facing ‘suicide prediction’, as defined by D’Hotman & Loh, before diving into medical prediction tools. Parts 2 – 4 of this series extend on their narrative review, reflecting on their use of data from non-medical sources and what it means to use social media content, before looking at specific health/mental health conditions and Facebook specifically as a ‘suicide prevention tool’. It’s a new perspective on suicide prevention that we believe everyone should become more familiar with, as technology continues to advance at a startling rate.

___________________________________________________________________

It might be strange to think of computers, or more specifically artificial intelligence (AI) as being better able to identify when a person is at risk of suicide than a trained medical professional. Some researchers are now of the opinion that we should shift our efforts from assessing patient suicide risk factors to the machine learning and data approach. Results indicate that AI consistently outperforms doctors at predicting suicide attempts and completions. An article by D’Hotman and Loh (2020) has got us thinking about the possibilities.

The challenge of suicide risk assessment

As the study featured in our last blog indicated, there are limitations in the tools we currently use to assess the risk of suicide. Many people who are at risk of suicide do not seek medical assistance, or in some cases, do not identify as being at risk. Doctors and clinicians often have difficulty identifying suicide risk due to the complex and varying factors contributing to a person become suicidal. And while there are several assessment tools in use, none of these instruments has been proven to have effectiveness. Despite fifty years of research in this field, we have not improved from an only slightly better than chance rate of suicide prediction.

Improvements in AI and data analytics

In the last few years, there have been significant recent improvements in the capacity of AI technology in a whole gamut of fields. As a result, there is increasing interest in the use of AI, data science and other analytical techniques to improve suicide prediction and risk identification. D’Hotman and Loh detail many research studies that reference both AI (and associated terms) and suicide prediction, ideation or risk. Citing many studies of interest, their article intends not to stress the use of AI in clinical applications, but rather to show the scope of opportunities that exist in this area and offer guidance about promising findings.

Medical AI suicide prediction tools

Medical AI suicide prevention tools involve doctors and researchers using AI techniques such as natural language processing and machine learning to assess electronic medical records, hospital records and other data. The information is processed to identify patterns that could indicate increased suicide risk. AI can be used as a real-time decision support tool to assist clinicians in assessing potentially at-risk patients.

Relevant examples

• A 2017 study to predict suicide completion among military veterans within 26 weeks following an outpatient visit, had improved capabilities at predicting suicide within the first five weeks
• A 2017 study where machine learning was used to assist in the prediction of a suicide attempt within the next two years and the next week, respectively and which found that depression with psychosis, schizophrenia and prior suicide attempts were significant predictors
• A 2018 study to assess suicide risk in patients attending a health service for any reason, which was able to match control patients and suicide cases with an accuracy rate of over 70%
• A 2019 study which used machine learning to predict suicide attempts among people with suicide ideation, with an accuracy of 88.9%

Using AI in real-world programs

And it’s not just theoretical- there are examples of AI being put to work in this area already. The American Department of Affairs in the USA has developed the Recovery Engagement and Coordination for Health- Veterans Enhanced Treatment (REACH VET) program. This program identifies veterans most at risk of suicide through AI, which examines millions of records related to:

• Medications
• Treatment
• Traumatic events
• Overall health

Those classified at the top 0.1% were 81 times more likely to attempt and 15 times more likely to complete suicide within the year after the assessment. Although it is early days, in the first year since implementing the program, there were 250 less suicides (a 4% reduction) in the number of suicide to be expected.

Opportunities for the future

Of course, once suicide risk is identified, there is still a need to determine what can be done about it. Individuals need to be supported with the most appropriate support and treatments, from the range of what is available. Identifying which types of treatments should be used for which patients is one immediate area of opportunity. It will be fascinating to see what the rapid developments in technological capability present us over the coming years.

PART II – Part 2 – Learning from machine learning