Translating promise into practice: a review of machine learning in suicide research and prevention
by Olivia J Kirtley, Kasper van Mens, Mark Hoogendoorn, Navneet Kapur, Derek de Beurs
Published March 2022
Use of machine learning in suicide research has the potential to improve prediction and prevention of suicidal thoughts and behaviours. But, while exciting and innovative, machine learning research often overlooks the practical and clinical implementation issues that might restrict its use in predicting and preventing suicide.
Research and findings
Researchers in Belgium, Netherlands, and the United Kingdom undertook a review of the literature to assess the benefit of machine learning in clinical practice. The paper discusses the factors that can impact the effectiveness and usefulness of machine learning in identifying suicide risk.
Electronic Health Record (EHR) data is a common starting point for applications of machine learning in healthcare. The effectiveness of machine learning is highly dependent on the comprehensiveness of the dataset it is modelled from. If EHR data is limited, then the usefulness of machine learning as a tool is reduced. Researchers also highlighted some ethical considerations of machine learning, particularly in the context of using private EHR data to inform the algorithms.
Although EHR may increase detection of suicide risk, the researchers note that the prevention of suicide is still highly dependent on the clinical and non-clinical intervention and support services available to the patient. The researchers state that even with machine learning, suicide prevention efforts will not improve if vulnerable individuals are left unsupported.
Machine learning as a clinical decision support tool has a potential role in suicide prevention, but evidence does not support the idea that machine learning algorithms are currently able to predict suicide. With further research and refinement, machine learning might develop into a tool that can complement and improve existing suicide prevention efforts by supporting clinicians and decision making when assessing risk. But, its effectiveness relies on the availability of treatments and interventions to meet patient needs.