Accounting for centre in the Early External Cephalic Version trials: An empirical comparison of statistical methods to account for centre in multicentre randomised controlled trials with binary outcomes

Abstract

<p><strong>Background </strong>External cephalic version (ECV) is an effective intervention to reduce breech presentation and the corresponding Caesarean section (CS) rate. The Early ECV (EECV) trials were international multicentre randomized controlled trials that compared the timing of ECV (early or delayed) on obstetric and neonatal outcomes. In consideration of current recommendations that multicentre trials should account for centre effects in their analysis, a secondary analysis of the EECV trials was undertaken.</p> <p><strong>Purpose</strong> To analyse the EECV Trial data using statistical methods that account for centre effect and compare the results to standard analysis.<strong></strong></p> <p><strong>Methods </strong>Fisher’s exact test was used to provide overall results unadjusted for centre effects. The outcomes of interest were CS, preterm birth, and non-cephalic presentation at birth.</p> <p>Seven statistical models that accounted for centre effects were applied to the data: i) Mantel-Haenzsel test, ii) fixed effects regression, iii) fixed effects regression with a treatment-by-centre interaction term (weighted and iv) un-weighted by centre size), v) random intercept model, vi) random intercept and random slope model, and vii) generalized estimating equations.</p> <p><strong>Results </strong>Accounting for centre effects showed strengthened statistical associations with point estimates moving away from the null value.</p> <p><strong>Conclusion </strong>Effect estimates and confidence intervals changed for the three selected outcomes after accounting for centre effects, but the overall conclusions of the trial did not change. For this application, the Mantel-Haenzsel test and the random effects regressions performed the best. This study provides empirical evidence to support recommendations that multicentre trials account for centre in both design and analysis.</p>