Introduction/Objectives: The criteria commonly used to assess the quality of randomised controlled trials (RCT) are concealed randomization, blinding and completeness of follow-up and intention-to-treat analysis. However, small studies meeting all these criteria may still have significantly biased estimate of effect because of confounding resulting from prognostic imbalance between the different treatment groups. The 'residual confounding' thus remaining even after concealed randomization should be considered as a criterion to assess the quality of randomized trials. Statistical methods commonly employed for meta-analysis do not correct for such bias. The objective of this paper is to determine how frequently this is a problem in meta-analyses.

Methods: 56 trials of various forms of treatment in acute stroke were examined for the presence of residual confounding. The direction of bias which this may cause in the estimate of treatment effect were also examined.

Results: The residual confounding, due to imbalance of known prognostic factors in the two arms of treatment was present in 22 (40%) of the trials.

Discussion: Residual confounding is not an uncommon problem in small, even though randomised trials. This may bias the estimate of effect from meta-analyses and hence be considered a criterion to assess the quality of RCTs.