Background In large multicentre trials in diverse settings, there is uncertainty about the need to adjust for centre variation in design and analysis. A key distinction is the difference between variation in outcome (independent of treatment) and variation in treatment effect. Through re-analysis of the CRASH-2 trial (2010), this study clarifies when and how to use multi-level models for multicentre studies with binary outcomes. Methods CRASH-2 randomised 20,127 trauma patients across 271 centres and 40 countries to either single-dose tranexamic acid or identical placebo, with all-cause death at 4 weeks the primary outcome. The trial data had a hierarchical structure, with patients nested in hospitals which in turn are nested within countries. Reanalysis of CRASH-2 trial data assessed treatment effect and both patient and centre level baseline covariates as fixed effects in logistic regression models. Random effects were included to assess where there was variation between countries, and between centres within countries, both in underlying risk of death and in treatment effect. Results In CRASH-2, there was significant variation between countries and between centres in death at 4 weeks, but absolutely no differences between countries or centres in the effect of treatment. Average treatment effect was not altered after accounting for centre and country variation in this study. Conclusions It is important to distinguish between underlying variation in outcomes and variation in treatment effects; the former is common but the latter is not. Stratifying randomisation by centre overcomes many statistical problems and including random intercepts in analysis may increase power and decrease bias in mean and standard error estimates. Trial registration Current Controlled Trials ISRCTN86750102, ClinicalTrials.gov NCT00375258, and South African Clinical Trial Register DOH-27-0607-1919

Background The number of Phase III trials that include a biomarker in design and analysis has increased due to interest in personalised medicine. For genetic mutations and other predictive biomarkers, the trial sample comprises two subgroups, one of which, say [Formula: see text] is known or suspected to achieve a larger treatment effect than the other [Formula: see text]. Despite treatment effect heterogeneity, trials often draw patients from both subgroups, since the lower responding [Formula: see text] subgroup may also gain benefit from the intervention. In this case, regulators/commissioners must decide what constitutes sufficient evidence to approve the drug in the [Formula: see text] population. Methods and Results Assuming trial analysis can be completed using generalised linear models, we define and evaluate three frequentist decision rules for approval. For rule one, the significance of the average treatment effect in [Formula: see text] should exceed a pre-defined minimum value, say [Formula: see text]. For rule two, the data from the low-responding group [Formula: see text] should increase statistical significance. For rule three, the subgroup-treatment interaction should be non-significant, using type I error chosen to ensure that estimated difference between the two subgroup effects is acceptable. Rules are evaluated based on conditional power, given that there is an overall significant treatment effect. We show how different rules perform according to the distribution of patients across the two subgroups and when analyses include additional (stratification) covariates in the analysis, thereby conferring correlation between subgroup effects. Conclusions When additional conditions are required for approval of a new treatment in a lower response subgroup, easily applied rules based on minimum effect sizes and relaxed interaction tests are available. Choice of rule is influenced by the proportion of patients sampled from the two subgroups but less so by the correlation between subgroup effects.