The hazards of chasing subgroups in neutral stroke trials.

Abstract

Background: The majority of randomised controlled trials in acute stroke and many for prevention are neutral, i.e. they failed to reach statistical significance. However, many of these will find apparent benefit in a component of a subgroup, findings which may be 'chased' in a follow-up trial. The evidence to date is that these follow-on trials are very likely to be neutral.

Findings: We discuss the issue of chasing subgroups in neutral trials and illustrate the challenges in five pairs of exemplar acute stroke trials. Problems in the exemplar trials include failing to define the subgroup in advance or even changing its definition, failing to show that both the interaction test and the primary outcome in the component were statistically significant, failing to publish additional information on the positive subgroup component, having too many subgroups, failing to make the follow-on trial large enough and failing to report the findings of the follow-on trial.

Conclusion: When chasing a positive component in a subgroup, it is vital that the subgroup: should be plausible biologically, defined a priori and have a significant interaction test. Further the number of subgroups should be limited and the component of interest should be statistically significant. Explanations should be given as to why the component is positive and other components of the subgroup are negative. Other outcomes should also show potential benefit. Unless this guidance is followed, it is highly likely that follow-on trials will be neutral as has occurred previously.