Instruments for the assessment of suicide risk: A systematic review evaluating the certainty of the evidence

PLoS One. 2017 Jul 19;12(7):e0180292. doi: 10.1371/journal.pone.0180292. eCollection 2017.


Background: Instruments have been developed to facilitate suicide risk assessment. We aimed to evaluate the evidence for these instruments including assessment of risk of bias and diagnostic accuracy for suicide and suicide attempt.

Methods: PubMed (NLM), PsycInfo, Embase, Cinahl and the Cochrane Library databases were searched until December 2014. We assessed risk of bias with QUADAS-2. The average sensitivity and specificity of each instrument was estimated and the certainty of the evidence was assessed with GRADE. We considered instruments with a sensitivity > 80% and a specificity > 50% to have sufficient diagnostic accuracy.

Results: Thirty-five relevant studies were identified but 14 were considered to have high risk of bias, leaving 21 studies evaluating altogether 15 risk assessment instruments. We could carry out meta-analyses for five instruments. For the outcome suicide attempt SAD PERSONS Scale had a sensitivity of 15% (95% CI 8-24) and specificity of 97% (96-98), and the Manchester Self-Harm Rule (MSHR) a sensitivity of 97% (97-97) and a specificity of 20% (20-21). ReACT, which is a modification of MSHR, had a similar low specificity, as did the Sodersjukhuset Self Harm Rule. For the outcome suicide, the Beck Hopelessness Scale had a sensitivity of 89% (78-95) and specificity of 42% (40-43).

Conclusions: Most suicide risk assessment instruments were supported by too few studies to allow for evaluation of accuracy. Among those that could be evaluated, none fulfilled requirements for sufficient diagnostic accuracy.

Publication types

  • Review
  • Systematic Review

MeSH terms

  • Humans
  • Risk Assessment / methods
  • Self-Injurious Behavior / psychology*
  • Sensitivity and Specificity
  • Suicide / psychology*
  • Suicide, Attempted / psychology*

Grants and funding

The study was funded by and conducted within the framework of the Swedish Agency for Health Technology Assessment and Assessment of Social Services. Margda Waern had financial support by the Swedish Research Council (521-2013-2699).