Background: In randomized clinical trials, continuous baseline values could be imbalanced due to small sample size or imperfect randomization. Estimates from analysis of covariance (ANCOVA) have been shown to provide the least biased estimate under imbalance, however, this method is often not used. Mean differences between treatment groups at follow-up or mean differences of change score from baseline are usually used in meta-analysis based on the reported results of the included individual studies. The impact of this imbalance has not been widely investigated in systematic review.

Objectives: The overall objective of this study is to evaluate the influence of choice of effect measures under imbalance on the conclusion of meta-analysis. In particular, we evaluate the relative bias of effect measures and identify factors associated with reporting baseline imbalance.

Methods: All reviews in the library of systematic reviews from Agency of Healthcare research and Quality (AHRQ) are reviewed. A sample of 10 comparative effectiveness reviews (CER) with at least one meta-analysis for continuous outcomes was selected and data of continuous outcomes from included studies from each CER were abstracted. Results using mean differences at follow-up and of change score from baseline were compared to results using ANCOVA methods to evaluate relative bias and factors associated with reporting baseline imbalance were identified using logistic regression.

Results: There were 250 reviews in AHRQ library and 32% included a meta-analysis for continuous outcomes. None of these reviews examined the issue of baseline score imbalance in quantitative synthesis. Results on relative bias from choice of effect measures and factors associated with reporting baseline imbalance are in progress and will be available by the time of presentation.

Conclusions: Baseline score imbalance is a neglected issue in CERs and results of this study could help to inform the choice of effect measures for continuous outcomes in CERs.