We read with interest the recently published meta-analysis of suicide prevention strategies by Riblet et al. ^{Reference Riblet, Shiner, Young-Xu and Watts1} However, we have some concerns about the authors' conclusion that ‘unlike previous reviews, ^{Reference Mann, Apter, Bertolote, Beautrais, Currier and Haas2,Reference Zalsman, Hawton, Wasserman, Van Heeringen, Arensman and Sarchiapone3} we did not find that lithium significantly reduced suicide’. This statement is at odds with the finding from our own meta-analysis in 2013, which found that lithium was more effective than placebo in reducing the number of suicides. ^{Reference Cipriani, Pretty, Hawton and Geddes4} The difference between the two meta-analyses relies solely on the addition of data from a single non-blind pragmatic trial. ^{Reference Girlanda, Cipriani, Agrimi, Appino, Barichello and Beneduce5} Although the authors do state that ‘the results of the summary estimate for lithium became statistically significant after removing a more recent study [Girlanda et al ^{Reference Girlanda, Cipriani, Agrimi, Appino, Barichello and Beneduce5} ] with several methodological limitations’, they fail to point out two key issues with regard to the addition of this trial, on which one of us (A.C.) was co-investigator.

Riblet et al fail to highlight that this study was not placebo controlled, unlike all the other studies contributing data to their meta-analysis, and was reported as essentially a failed, underpowered study. ^{Reference Girlanda, Cipriani, Agrimi, Appino, Barichello and Beneduce5} Including this study is, at the very least, highly questionable. Just as the authors reasonably included only randomised controlled trials (RCTs) in their analysis, so we would argue that it is inappropriate to include a non-placebo-controlled trial in a meta-analysis aiming to estimate the efficacy of lithium.

Furthermore, the fact that the addition of data from a single RCT with 53 patients, and just one completed suicide, appears to materially change the estimate of effect serves to highlight the major point that Riblet et al fail to discuss. As we have previously noted, ^{Reference Cipriani, Pretty, Hawton and Geddes4} randomised data in this area are sparse and estimates of efficacy are therefore highly unstable. It simply is not yet possible to determine, on the basis of randomised evidence alone, whether lithium does or does not reduce – and this may be an enduring uncertainty, given the low event rate of suicide and the practical and feasibility challenges of conducting adequately powered trials.

Although acknowledging the limitations of the randomised evidence, it is important to note that there are several large-scale observational studies that also find a reduced incidence of completed suicide among those on lithium treatment that is of a size consistent with the randomised evidence. ^{Reference Song, Sjölander, Joas, Bergen, Runeson and Larsson6–Reference Silverman8} Taking the randomised and observational data together, and in view of the sensitivity of Riblet et al's results to the inclusion or exclusion of a single methodologically heterogeneous trial, we believe that the combined current evidence indicates that lithium probably has a substantial and clinically important anti-suicidal effect.