Citation

Sewall CJR, Wright AGC. Crisis 2021; 42(6): 405-410.

Copyright

(Copyright © 2021, International Association for Suicide Prevention, Publisher Hogrefe Publishing)

DOI

10.1027/0227-5910/a000834

PMID

34783589

Abstract

What makes people suicidal? Unfortunately, despite a century of suicide research and theorizing, the field of suicidology has yet to provide a sufficient answer to this foundational question. We have managed to identify a multitude of empirically and theoretically derived risk factors - spanning everything from social forces at the societal level (e.g., Durkheim, 1897/1951) to biological mechanisms at the microscopic level (e.g., Mann, 2013; Pedersen et al., 2012), and everywhere in between (see Turecki et al., 2019) - but "there is no evidence that any known risk factors - broad or specific - approach what many might define as clinical significance" (Franklin et al., 2016, p. 215). The inability of single or small sets of risk factors to adequately predict suicide risk has reinforced the highly complex nature of suicidal thoughts and behaviors (STBs) and prompted the use of machine-learning methods, which are better suited to model the types of complex dynamics that may be necessary to predict risk (Franklin et al., 2016; Ribeiro et al., 2016).

Initial attempts at applying machine learning to suicide risk prediction have appeared to support this assertion, as machine-learning approaches have improved predictive accuracy and identified novel risk factors (Burke et al., 2019). However, despite these advances, our ability to predict short-term risk for suicide remains poor (Belsher et al., 2019; Burke et al., 2019) - which is a crucial impediment to effective suicide intervention - and there is evidence that, due to methodological oversights, the predictive performance of machine-learning algorithms in recent suicide research is not as impressive as advertised (Jacobucci et al., 2021). While we share in the enthusiasm for machine learning and the promise it holds for capturing the complexity inherent in STBs and agree with calls to improve and advance machine-learning applications in suicide research (Burke et al., 2019; Jacobucci et al., 2021), we believe that efforts at suicide prediction will continue to fall short because of a fundamental misalignment between how STBs are typically studied and how they inherently emerge...

Language: en