Including random centre effects in design, analysis and presentation of multi-centre trials

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