It is interesting to note that while all the participants in the debate on clozapine v. typical neuroleptics referred to Kristian Wahlbeck's meta-analysis (Reference Wahlbeck, Cheine and EssaliWahlbeck et al, 2000), none of them referred to her sub-analysis of her own earlier meta-analysis on this topic (Reference Wahlbeck and AdamsWahlbeck & Adams, 1999). In this subanalysis, all randomised trials comparing clozapine with typical neuroleptic medication for schizophrenia were divided into sponsored (reporting some kind of connection with manufacturers of clozapine) and non-sponsored trials. Odds ratios and 95% confidence intervals were calculated for the primary outcomes of relapse, clinical improvement, and leaving the study early, separately for sponsored and non-sponsored studies. Odds of relapsing were significantly in favour of clozapine in the sponsored trials (OR=0.5, 95% CI 0.3–0.7). Non-sponsored studies reported equivocal findings (OR=0.4, 95% CI 0.1–1.4). Similarly, sponsored studies showed a significant difference in favour of clozapine on the outcome measure of leaving the study early (OR=0.5, 95% CI 0.4–0.7). Non-sponsored studies showed a non-significant difference (OR=0.6, CI 0.3–1.2). Only on the outcome measure of improvement did both sponsored and non-sponsored studies show a significant benefit of clozapine over older antipsychotics. Wahlbeck suggested that those undertaking meta-analysis of drug treatment should investigate for sponsorship bias by using sensitivity analysis.

Outside of psychiatry, similar associations between sponsorship and outcome of trials has been demonstrated in randomised controlled trials (RCTs) published in five general medical journals (Reference DavidsonDavidson, 1986; Reference Yaphe, Richard and KnishkowyYaphe et al, 2001), RCTs of nonsteroidal anti-inflammatory drugs in the treatment of arthritis (Reference Rochon, Gurwitz and SimmsRochon et al, 1994) and RCTs published in the BMJ over 4½ years (Reference Kjaergard and Als-NielsenKjaergard & Als-Nielsen, 2002).

Although RCTs and meta-analyses have contributed greatly to increasing our knowledge base about which treatments work and which do not, maybe it is time we began to consider other factors that might explain the observed difference between two treatments in RCTs and meta-analyses, beyond the standard critical appraisal questions. Maybe we need to ask not only how the efficacy of clozapine (or any other drug for that matter) has been analysed but also who has analysed it.