How you are calculating the changes in each group (i.e post- pre)? I am facing same problem. CMA will ask for pre-post correlation I think.

ANCOVA is not appropriate when participants are not randomly allocated to group. See: http://www.ncbi.nlm.nih.gov/pubmed/11261398  

Propensity score matching is the preferred approach to deal with confounding factors in a non-randomised between-group design, though this is obviously difficult without individual subject data.

ِDear Mikhail,

I do not think ANCOVA is a viable option for your meta-analysis. Pre-post SD and means will give you an effect size which is more and intuitive. Then if you have other variables you can used subgroup analysis or meta-regression to check their effects on your meta-analysis.

and dear Souvik, you are right working with pre-post data is pain in the neck but I think the following link would help you a lot for calculating the correlation you need.

http://handbook.cochrane.org/chapter_16/16_1_3_2_imputing_standard_deviations_for_changes_from_baseline.htm

Best wishes,

Ramin

Thank you. I was just thinking what Cochrane Handbook says: "The preferred statistical approach to accounting for baseline measurements of the outcome variable is to include the baseline outcome measurements as a covariate in a regression model or analysis of covariance (ANCOVA)." (Chapter 9.4.5.2) And they have their point here.

E.g., I have an RCT with unsuccessful randomization, like means are 3.0 and 3.9 at baseline. Sure, I can just put them into CMA an it calculates me effect size based on change difference. But what should I do if I'd like to use ANCOVA and final values, not the change difference? 

Actually, it is a real problem, as it is what peer-reviewer has suggested. And, on the other hand, I'm interested of this too.I think, this is my about 10th meta-analysis, and the process got quite automatic and a little boring. This would be a new approach.

Dear Souvik. Would it be possible to get the full text of Miller paper you suggested?

I think the Miller and Chapman paper is problematic. I've read it and I think to interpret is implying ANCOVA can only be used for randomised studies is incorrect (though this seems to be a common view). Weirdly the same journals that dislike using ANCOVA publish many studies using multiple regression for non-randomised designs.

See also the discussion here:

http://andrewgelman.com/2011/08/18/misunderstandin_1/

Any meta-regression package should allow you to model the post test means with pre test as a predictor.

Thank you Tom. I think, meta-regression is not what Cochrane Handbook means when talking about ANCOVA. I did a little search on Internet and, hopefully, found the answer. It seems, that you have to calculate (manually) an effect size for each included study using ANCOVA, and then to put them into meta-analysis (any software, CMA is just fine). There is a presentation on Cochrane site by Jo McKenzie.

Andrew, propensity score is hardly of use in meta-analysis, except maybe for IPD. PS needs huge numbers. I did my PhD using propensity score on sample of 50 000. I believe, there is no way you can use propensity score in a meta-analysis on aggregate scores. 

Hi Mikhail, 

I was to ask the same question for RG community. I was trying some Google search and I found your well-explained question. I am now exactly on the same situation. Do you think I should calculate the ANCOVA effect estimate and the baseline and followup correlation manually? It would be helpful if you add the link to the presentation by John McKenzie. Thank you.

Hi Mikhail Saltychev is this "It seems, that you have to calculate (manually) an effect size for each included study using ANCOVA, and then to put them into meta-analysis (any software, CMA is just fine)" possible to do using RevMan meta analysis software afterwards? Also, for ANCOVA are the std. dev. and means (pre and post) for control and intervention enough to complete it?

It's probably too late for yours Mihiretu M. Kebede but for anyone else who needs it, I think this is the presentation:

https://methods.cochrane.org/statistics/teachingtraining-workshops/smg_training_course_2010/session1a_mckenzie