Meta-analysis of paired data

[Athanasios Kordalis]

Hello everybody,
I am conducting a meta-analysis trying to estimate the summary effect of an intervention on a continuous variable before and a given period after treatment.

The studies are reporting:

1. pre-treatment mean
2. pre-treatment standard deviation
3. post-treatment mean
4. post-treatment standard deviation
5. p value (not always the exact value)
6. sample size

Having the above data I am not able to estimate the SD of the mean difference.
Is there any option in metan command or any other way to let my handle this kind of data?
Thank you in advance,
Athanasios

[Tiago Pereira]

You have to assume a correlation between the pre- and post-treatment mean. That estimate can be obtained from previous studies or "guestimated". Then, you can calculate the mean difference with its appropriate standard error.

[Athanasios Kordalis]

Thank you for your reply, do you know if there is any statistical tool or module in STATA to handle this "guestimation"?

[Marcos Almeida]

You may use the difference, instead. To estimate the pooled SD, shall you have the r of Pearson, there is a formula you may use, according to this article. Otherwise, as also stated in the article, you may "impute" a given r value, logically on account of an "educated guess".

[Athanasios Kordalis]

Thank you very much for your valuable post.
The reference was very instructive regarding the handling of missing data and the empirical assessment of within-arm correlation.

Best regards.
Athanasios

[Marcos Almeida]

Hello Athanasios,

thank you for informing you consider the thread reached the solution satisfactorily.