Statistical Issues in Randomized Controlled Trials: an editorial

Abstract

Randomization is the bedrock of randomized controlled trials, which ensures the elimination of selection bias and also to some extent the homogenous distribution of covariates between the intervention arms. Randomization does not always guarantee the baseline balance, and hence makes the statistical analysis more complex. Several published clinical trials have employed test of significance to compare baseline measures between the groups. However, such practice has been criticized by several authors and CONSORT statement also discourages it. This overview discusses various statistical designs that were employed in published trials. Post intervention data (follow up score) comparison between the arms was common practice in published RCTs. However, this approach fails to adjust baseline imbalance. Both Change score and Percentage change methods adjust the baseline imbalance. Both of the approaches give precise estimates when there is a high correlation between baseline and follow-up score. However, when correlation is low they both give biased and less precise estimates of treatment effect. Analysis of covariance (ANCOVA) is a regression method, which maintains high statistical power and gives less biased and more precise estimates of treatment effect regardless of correlation level. Understanding strengths and limitations of different statistical designs of RCTs will prevent statistical errors, which can yield an accurate estimate of treatment effect.