How to combine covariate-adjusted and unadjusted standardised mean differences in meta-analysis? I have completed an meta-analysis of final scores (comparison of means at post test); however, i have noticed that despite being randomised, differences at baseline are present and are biasing the findings. Can I combine co-variate adjusted means (where reported) and final scores means in one meta-synthesis? 
 A: Whenever possible, you should be meta-analyzing unadjusted data rather than data adjusted in regression-based analyses (e.g. least-square means) whenever possible. There is no way to determine the exact effect of the adjustments without having access to the raw, unadjusted data. Additionally, different studies can adjust for different co-variables.
As for the differences in baseline characteristics, there is a misconception that proper randomization alone will create groups that are homogenous in all aspects. What is missing from that equation is the sample size... in other words only with an infinite population will randomization always create groups that are statistically homogenous in all known and unknown confounders. Small studies are more likely to have significant differences between groups at baseline just by 'chance' and as more patients are randomized there is a regression towards the mean. This is also a factor in stopping early for benefit or harm, whereby early studies often show exaggerated results, are stopped early, and published; while later trial show either an effect that is less eccentric.
If these are randomized trials, and you are using the Cochrane risk of bias tool to assess the risk of bias in the included trials, then you can consider the trials with 'significant' differences in baseline characteristics without any investigation of the effect or sensitivity analyses (e.g. adjusted analyses) as having a high risk of bias. Of course, I would contact the authors for possible explanations, a peek at their raw data or asking them to re-analyze their data for me.
Hope this helps.
Ahmed
