We thank Roberts and colleagues for their thoughtful critique of our meta-analysis. ^{Reference Riblet, Shiner, Young-Xu and Watts1} They question our decision to include Girlanda et al ^{Reference Girlanda, Cipriani, Agrimi, Appino, Barichello and Beneduce2} in our meta-analysis of trials of lithium for the prevention of death by suicide. Roberts et al aptly highlight that the Girlanda et al study had several methodological limitations. Although the study was described as a randomised assessor-masked trial, the comparison arm consisted of usual care; in addition, the study did not achieve the target sample size.

Since our meta-analysis evaluated randomised trials of behavioural and pharmacological interventions, we included trials that used usual care, placebo or waiting-list control conditions. Although there are many benefits to using a placebo control condition, a number of legitimate counter-arguments have also been raised, even in the case of pharmacological trials. ^{Reference Avins, Cherkin, Sherman, Goldberg and Pressman3} In fact, some authors have suggested that, if a trial is of pragmatic design, a usual-care control may be more appropriate than placebo. ^{Reference Avins, Cherkin, Sherman, Goldberg and Pressman3} We had no specific inclusion criteria involving study size. In fact, one advantage of meta-analysis is the ability to pool multiple underpowered studies; consequently, we feel that the size of the individual studies is less relevant. In our original manuscript, we did perform a sensitivity analysis by removing the Girlanda et al trial from our analysis because of its multiple methodological limitations. We agree, however, with Roberts et al that we should have made it clear to the reader that the Girlanda et al trial used usual care, rather than placebo, as the control condition. ^{Reference Girlanda, Cipriani, Agrimi, Appino, Barichello and Beneduce2}

Consistent with the salient points made by Girlanda et al in the discussion section of their paper, ^{Reference Girlanda, Cipriani, Agrimi, Appino, Barichello and Beneduce2} we agree that it is important that readers are aware of the results of all randomised trials evaluating lithium for suicide prevention, regardless of the findings or the power of the individual study. In fact, Girlanda et al highlighted that it would be important for their results to be ‘incorporated into future meta-analytical reviews’. ^{Reference Girlanda, Cipriani, Agrimi, Appino, Barichello and Beneduce2} A co-author of Roberts et al's letter to the editor (Cipriani) was also a co-author on this publication by Girlanda et al.

Ultimately, although we agree that the results of observational studies certainly support a role for lithium in suicide prevention, we feel that there is a clear need for more randomised trials evaluating its efficacy in preventing death by suicide. The substantial effect of a single trial highlights the tenuousness of findings regarding lithium in RCTs. Fortunately, a brief search of clinicaltrials.gov suggests that there is a large trial of lithium for suicide prevention underway (NCT01928446) and another trial that was recently completed (NCT01134731). Notably, a third trial was prematurely terminated (NCT00520026).