I2 Statistic as a Test for Selection Bias in Randomised Controlled Trials.

This technical report demonstrates that the use of the I statistic for testing selection bias in single randomised controlled trials (RCTs) has the potential to allow the prevention of false-positive test results, thereby allowing for high test specificity and a high positive predictive value. In addition, the I statistic provides utility for the in-depth identification of low-level selection bias in RCTs, thus assisting in the avoidance of false-negative test results and possibly for estimating the percentage of trial patients with biased allocation into RCT treatment groups. Future studies to this topic may investigate whether cases with I estimates above 0%, due to chance rather than selection bias, are possible and, if so, how to distinguish such cases from those with very low bias levels. Future studies may also test the null hypothesis that levels of selection bias are not associated with any over- or underestimation of the true effect estimates of RCTs.