# How to calculate the covariance of comparisons in meta analysis for two estimate from the same patients in a prepost design with two time of follow up

I am trying to do a meta-analysis on effect of a medication on a blood pressure. Some studies have longitudinal follow up with up to three measurements (pre treatment and two post treatment). The follow up timing is not the same in different studies). I want to test whether the effect of treatment remains the same as the follow up time increases. For this I need to calculate two effect sizes 1) first follow up compared to baseline and also 2) second follow up and the baseline. This gives me two estimates from each study with two different follow up times. I did a meta regression with ignoring the fact that I am including same patients in the analysis. Now I want to do a multilevel meta-analysis with including the within study variation in those studies that reported two follow ups. For this I need to include the covariance of the two estimates from each studies. There are available formulas for estimating the covariance in situations that same patients assessed by two different outcomes. However for this specific situation I didn't find any formula (same patient started on the treatment and two estimates are ES of first follow up vs. baseline and ES of second follow up vs. baseline). Any idea how I can calculate this covariance? Is it ok to assume that two estimates are comparison of cases and controls and just use formula for the within study covariance in the case that same patients evaluated based on two different outcomes?

• What is meta regression? I don't see how you can combine regressions from separate studies. – Michael Chernick Apr 3 '17 at 3:36
• @MichaelChernick See: en.wikipedia.org/wiki/Meta-regression – Wolfgang Apr 3 '17 at 9:11
• Wikipedia is looking for experts to edit each paragraph. Although I saw many references, I did not see any of the basic statistical work of Olkin and Hedges referenced. – Michael Chernick Apr 3 '17 at 11:51
• If you fit a random intercept for study in your multi-level model surely you are assuming the two estimates from that study are related? – mdewey Apr 3 '17 at 15:18