We agree with Samara and Leucht that clinicians can feel overwhelmed by the vast quantity of published meta-analyses and variety of methodologies. It is therefore important that protocols for potential meta-analyses are published on open access repositories such as PROSPERO to reduce the risk of duplication. Unfortunately, we were unaware of the 2016 Samara et al ^{Reference Samara, Dold, Gianatsi, Nikolakopoulou, Helfer and Salanti1} meta-analysis at the time of conducting ours, as they did not register their protocol on PROSPERO, only including a protocol as a supplementary document at the time of publication.

The results of the two meta-analyses were broadly similar. ^{Reference Samara, Dold, Gianatsi, Nikolakopoulou, Helfer and Salanti1,Reference Siskind, McCartney, Goldschlager and Kisely2} Our primary outcome, difference in total psychotic symptoms over the short and long term, showed that clozapine was not superior to other antipsychotics in long-term studies, which corresponds to the results of Samara et al. We did find that clozapine was superior to other antipsychotics for positive symptoms in the short and long term, an important finding for clinicians, patients and their carers.

There are key differences between the meta-analyses. First, Samara et al did not divide by study duration. We separated studies that reported data before 3 months from those that reported data after 3 months. We feel that it is inappropriate to include results from a 6-week study with those from a 78-week study. Second, unlike Samara et al, we conducted sensitivity analyses on the effects of pharmaceutical funding and found that studies without such funding favoured clozapine more strongly.

There remains debate as to the validity of network meta-analyses. They are at higher risk of bias, and require an underlying assumption that all included interventions should be jointly randomisable. ^{Reference Li, Puhan, Vedula, Singh and Dickersin3} This is clearly not the case for people with treatment-refractory schizophrenia, as some will have previously been on the same antipsychotics that are the intervention arm of other trials.

We identified and excluded four of the five papers listed as ‘missed’ by Samara and Leucht as they did not have usable data. The other, Honigfeld (1984), provided 4-week data for total psychotic symptoms, which, when included, did not alter the short-term results. We note that Samara et al did not include Honigfeld (1984) in their analysis. Similarly, excluding McEvoy's partially blinded study made little difference.

Samara and Leucht were inaccurate regarding what was included in our meta-analysis. Although we included studies from different age groups, children were excluded on sensitivity analyses, making no difference to the results. We also reported sensitivity analyses of comparisons with first- and second-generation antipsychotics, as well as specific antipsychotics, in our original article.

Samara and Leucht are dismissive of Chinese data, even though their original protocol stated they would be included in their meta-analysis. Furthermore, there are increasing calls to include such data. ^{Reference Cohen, Korevaar, Wang, Spijker and Bossuyt4} In a review of PubMed English-language articles published in 2000 and 2010, 11% of retractions were from China, while 33% were from the USA. Both Chinese studies in our meta-analysis were identified in the Cochrane database and analysed for risk of bias.