Adjusting for random centre effects in large trials with a binary outcome: A case study using data from the international multi-centre WOMAN randomised controlled trial

BackgroundThe multicentre WOMAN trial showed that tranexamic acid reduces postpartum haemorrhage (PPH) deaths. Several studies have recommended adjusting for clustering at the country and centre level to improve power and reduce bias in the standard errors. We reanalysed data from the WOMAN trial, adjusting for these centre effects. MethodsThe WOMAN trial recruited 20,060 women with clinically diagnosed PPH from 193 centres in 21 countries. The intervention was intravenous tranexamic acid versus matching placebo and the outcome was death from bleeding within 42 days of randomisation. We reanalysed data for the 14,928 women treated within 3 hours of birth for whom tranexamic acid provided the most benefit. We used random effects logistic regression to calculate the effect of tranexamic acid taking into account variation in risk of death and treatment effectiveness by country and centre. We calculated intraclass correlations (ICCs) to quantify between country and between centre within country variation. Results216 (1.4%) women died from bleeding. Using a univariable analysis without adjusting for centre effects, we found tranexamic acid reduced the odds of death from bleeding by 31% (OR=0.69 95% CI: 0.52-0.90, p=0.007). Adjusting for baseline covariates (age, systolic blood pressure (SBP) and SBP2) but not country or centre yielded a 36% odds reduction (OR=0.64 95% CI: 0.48-0.85 p=0.002). Adjustment for baseline covariates, country and centre yielded a 37% odds reduction (OR=0.63 95% CI: 0.48-0.85 p=0.002). We found substantial between country and centre variation in outcomes but not treatment effectiveness. The ICC for outcome was 14% for country and 19% for centre within country. ConclusionsAdjusting for country and centre effects made negligible differences to the magnitude of the treatment effect estimate or its associated p-value. Consistent with other studies of large clinical trials for medicines with binary outcomes, we found considerable between country and centre variation in outcomes but not in relative treatment effectiveness. Despite substantial ICCs for the outcome, adjusting for country and centre effects had minimal impact on our results. Trial registrationclinicalTrials.gov:NCT00872469 (March 2009)