To the Editor Belsher et al1 reported a systematic review and simulation to evaluate the diagnostic accuracy of models in predicting suicide and suicide attempts. The authors found that positive predictive value was low even in the higher percentiles in studies predicting suicide. This is a common scenario when assessing outcomes with extremely rare events. We acknowledge the limitations in the scientific literature presented in the article; however, a major drawback that was not discussed was the follow-up period of the included studies. It is expected that models with rare events have a very low positive predictive value when dealing with a short follow-up period, as deaths by suicide could still be ahead for people considered as false positives. It is worth mentioning that only 1 of the included studies had a follow-up of more than 2 years. On the other hand, less rare events, such as suicide attempts, are more easily handled, and some of the current models present a good positive predictive value, as shown in the article.1 Therefore, we recommend that future studies should have longer follow-up periods and address specific populations with higher rates for suicide attempts, such as bipolar disorder or schizophrenia, thus mitigating limitations that are showcased in the article by Belsher et al.1 We also think that suicide is highly associated with method lethality—a variable that cannot be included in predictive models for clinical purposes. Lastly, the argument that individuals identified as being at risk of suicide could experience psychosocial prejudice is indeed a major concern; however, this issue can be handled by fostering medical confidentiality, which is possible provided that these individuals are not at acute risk of suicide.
Passos IC, Ballester P. Positive Predictive Values and Potential Success of Suicide Prediction Models. JAMA Psychiatry. Published online June 26, 2019. doi:10.1001/jamapsychiatry.2019.1507
Customize your JAMA Network experience by selecting one or more topics from the list below.
Create a personal account or sign in to: