In his editorial arguing that atypical antipsychotics can no longer be regarded as having advantages over conventional drugs, Kendall ^{Reference Kendall1} makes two statements which do not do justice to the available evidence.

First, he states: ‘With the exception of clozapine for treatment-resistant schizophrenia, the atypicals, as a group of antipsychotics, are no more efficacious for schizophrenia than the typicals, whether it is chronic or acute, for first or subsequent episodes, for the acute episode or for promoting recovery’. This is supported by a reference to the updated National Institute for health and Clinical Excellence (NICE) guideline for schizophrenia, ² which in turn based its conclusions on a series of meta-analyses carried out by the National Collaborating Centre for Mental Health (NCCMH; www.nccmh.org.uk). The problem here is that two other meta-analyses have reached different conclusions. In 2003, Davis et al ^{Reference Davis, Chen and Glick3} found that, apart from clozapine, three atypicals showed significant superiority over conventional antipsychotics: risperidone (22 studies, effect size (ES) 0.25), olanzapine (14 studies, ES = 0.21) and amisulpride (12 studies, ES = 0.29). Six years later, Leucht et al ^{Reference Leucht, Corves, Arbter, Engel, Li and Davis4} had closely similar findings for olanzapine (28 studies, ES = 0.28) and amisulpride (13 studies, ES = 0.31); the effect size had become smaller for risperidone, but it was still significant (34 studies, ES = 0.13).

One reason why the NICE/NCCMH meta-analysis may have reached negative conclusions concerning these three drugs is that it included fewer studies. The outcome measure used by Davis et al ^{Reference Davis, Chen and Glick3} and Leucht et al ^{Reference Leucht, Corves, Arbter, Engel, Li and Davis4} was reduction in total symptom scores, based on pooled data from the Positive and Negative Syndrome Scale (PANSS), Brief Psychiatric Rating Scale (BPRS) and CGI (Clinical Global Impression) scale, and from the PANSS and BPRS respectively. In the NICE/NCCMH meta-analyses there were only 17 studies of risperidone, 10 of olanzapine and 4 of amisulpride in which the drug was compared with a conventional antipsychotic on any of these outcome measures. Data were also pooled separately for studies carried out on patients with first-episode schizophrenia, on those with acute exacerbations or recurrences, and on those with treatment-resistant illness. As a result, the maximum number of studies included in any of the NICE/NCCMH meta-analyses of overall symptoms for these three atypicals was six, and several contained only one or two studies.

Later in the article, Kendall cites approvingly a meta-analysis by Geddes et al ^{Reference Geddes, Freemantle, Harrison and Bebbington5} in 2000, which found evidence that the superiority of atypicals (including clozapine) was an artefact of the high dose of the typical antipsychotic used as a comparator in some of the trials. These authors used meta-regression to examine the predictive value of haloperidol dose (23 studies) or chlorpromazine dose (7 studies) on the outcome of total symptom scores. In both cases, the findings were significant: an observed advantage in favour of atypicals disappeared as the dose of the comparator drug decreased. Davis et al ^{Reference Davis, Chen and Glick3} subsequently explored the effect of comparator dose in their meta-analysis. The results of several different analyses led them to conclude that there was no significant effect of haloperidol in a larger data-set of studies. Leucht et al ^{Reference Leucht, Wahlbeck, Hamann and Kissling6} also failed to find a significant effect of chlorpromazine comparator dose in another meta-analysis carried out at around the same time. Geddes et al ^{Reference Geddes, Harrison and Freemantle7} argued that a significant effect of comparator dose could be re-instated in this latter meta-analysis by using their own meta-regression technique; in their author reply, Leucht et al ^{Reference Geddes, Harrison and Freemantle7} countered that the effect was not significant when a variety of other statistical techniques were used, indicating that the finding was not robust.

Kendall states that the comparator drug effect has been ‘neither confirmed nor disproved by later meta-analyses’. An arguably more accurate conclusion is that it was an early finding which has not stood the test of time.