Predicting Risk of Suicide Attempts Over Time Through Machine Learning

Abstract

Suicide attempts are a major public health problem, with an estimated 25 million nonfatal suicide attempts occurring each year worldwide (Centers for Disease Control and Prevention [CDC], 2016; World Health Organization, 2016). Beyond considerable economic and societal burdens associated with nonfatal attempts (Shepard, Gurewich, Lwin, Reed, & Silverman, 2016), nonfatal suicide attempts are among the strongest predictors of suicide death-a leading cause of death worldwide (Ribeiro et al., 2016a). The scope and seriousness of the problem have prompted substantial research attention (Franklin et al., 2017). Yet our ability to predict nonfatal attempts remains marginally above chance levels (Bentley et al., 2016; Chang et al., 2016; Franklin et al., 2017; Ribeiro et al., 2016a). Rates of nonfatal suicide attempts remain intractable, with recent estimates suggesting that nonfatal attempts may be on the rise (CDC, 2016). The purpose of the present study was to evaluate the accuracy and temporal variation of a potentially scalable suicide attempt risk detection strategy: machine learning applied to electronic health records (EHRs). Recent meta-analyses have shown that the ability to predict suicide attempts has been near chance for decades (Franklin et al., 2017). A major reason for this poor prediction is that the majority of studies tested predictors in isolation (e.g., a depression diagnosis), and even the best isolated predictors are inaccurate (Franklin et al., 2017; Ribeiro et al., 2016a). Accurate suicide attempt prediction may require complex combinations of hundreds of risk factors. Traditional statistical techniques are not ideal for such analyses; fortunately, machine learning (ML) techniques are well suited for such problems. These techniques can test a wide range of complex associations among large numbers of potential factors to produce algorithms that optimize prediction. Retrospective ML studies suggest that this approach may be promising 691560C PXXXX10.

Medicine and the rise of the robots: a qualitative review of recent advances of artificial intelligence in health

“…In other words, when it comes to predicting suicide attempts, AI appears to be better than human beings, although the clinical applicability in the real world remains unproven 23. In another study, researchers used machine-learning algorithms to identify individuals at risk of suicide with high (91%) accuracy, based on their altered fMRI neural signatures of death-related and life-related concepts 24.…”

Determinants for Sustained Use of an Activity Tracker: Observational Study

“…Furthermore, Random Forest is capable of detecting nonlinear relations between independent and dependent variables. Random Forest analysis methods have recently been applied successfully in genetics, clinical medicine, bioinformatics, and the social sciences (see [56,67] for examples).…”

Annual Research Review: Suicide among youth – epidemiology, (potential) etiology, and treatment

“…machine learning), as some researchers have already begun to do (e.g. Kessler et al., ; Pestian et al., ; Walsh, Ribeiro, & Franklin, ). Machine learning may be particularly helpful with meaningfully integrating the many small to modest effects from risk factors and correlates observed in the field (Franklin et al., ).…”