Peluso et al report on the differential effect of first-generation antipsychotics (FGAs) v. second-generation antipsychotics (SGAs) in ameliorating or exacerbating extrapyramidal side-effects (EPS) in a secondary analysis of the CUtLASS-1 trial data. ^{Reference Peluso, Lewis, Barnes and Jones1} They report their findings as ‘essentially null’ and mention that there is weak evidence for clinically significant differences in emergent or relieved EPS between FGAs and SGAs. These findings, although based on a secondary analysis, pose interesting and important challenges for the focus of future research, but also raise some questions about the interpretation of negative study findings.

The majority of participants (49%) in the FGA group were prescribed sulpiride, a substituted benzamide that has been demonstrated in a meta-analysis to have a significantly lower propensity to cause EPS than other FGAs. ^{Reference Soares, Fenton and Chue2} It could be argued that it would not be unusual to find little difference between the two groups, as the FGA group was biased towards sulpiride selection.

A priori odds ratios of 2 and 0.5 were selected as clinically relevant, but no reason is given for this choice. The choice of this cut-off seems arbitrary. The authors conclude that their results are ‘essentially null’ and that these two classes of drugs could be used with equivalence in EPS. Although equivalence is possible, failure to reject the null hypothesis does not imply that the null hypothesis is true or that treatments are equal. ^{Reference Nickerson3} Failure to reject the null at this effect size means that the null would not be surprising at this particular value. ^{Reference Aberson4} However, given a power of 78%, this implies a relatively high chance (22%) of a type 2 error. In some cases, even a reduction of 20% in EPS occurrence can be clinically meaningful. The CUtLASS study would be underpowered even if a true effect existed at this effect size. Confidence limits around the EPS outcomes also appear to be wide at a number of time points. Although negative findings in superiority trials are important to report, it should be noted that some may argue that meaningful scientific evidence centres on replicated falsification.

In turn, the dichotomisation of EPS outcome measures, instead of using changes in continuous EPS scores over multiple time points in a longitudinal design and analysis strategy, could potentially underestimate any treatment effect.

Nevertheless, these findings raise important points for the design of superiority trials. Given the lack of superior efficacy in symptom relief of most SGAs, if the presence of EPS has become the sine qua non for treatment switches to SGAs, would this not highlight the importance of adequately powered trials where the primary outcome would be EPS? Inaddition, intrialswhere EPS isonly a secondary outcome, as is commonplace, is it not necessary that this outcome be adequately powered at well-motivated, pre-agreed effect sizes? Although of global importance in the current economic climate, this would be particularly important for low- and middle-income countries where funding authorities meticulously scrutinise the benefits of more expensive treatments.