Does attrition introduce bias in randomised trials: a methodological review

In undertaking randomised controlled trials (RCTs) attrition can be a problem, as it may introduce bias post randomisation if the characteristics of those lost to follow-up are materially different between groups. Attrition bias in RCTs is based on theory and there is relatively little empirical evidence. One approach to assess whether attrition is introducing bias is to undertake a meta-analysis of measured covariates comparing group values as randomised (no attrition) versus group values as analysed (with attrition). In the first analysis, because of randomisation, any baseline differences are simply due to chance and a meta-analysis should show zero imbalances across treatment groups and zero heterogeneity. In contrast, in the presence of bias, the second analysis would show baseline differences and/or baseline heterogeneity. In this study, we took a sample of 37 completed RCTs funded by the Education Endowment Foundation (EEF) in the UK, and we compared the baseline values of as randomised and as analysed samples. There was no heterogeneity observed for pre-test for as randomised or for as analysed (I2 = 0.00%). Further work is required using a larger sample of trials.