Yeah, I understand your problem. I assume you are using RevMan, but for fixing this I recommend you to go with Meta-Essentials, in which you just simply need to enter the pre and post readings and Meta-Essentials will do all the calculations for you.

I hope it helps.

Thank you.

Dear Shruti Raghuraman,

it depends if you are doing meta-analysis from change from baseline measurements or only for final observations.

Considering that you are using final observations, I'll try a very simple example to display how to do:

Settings: randomized clinical trial, two arms, pre-post analysis with a follow up of 5 years.

Outcome: Depression and anxiety trough a scale.

Study #1 - Raghuraman et al, 2018.

Group A baseline score: 30 ± 10, sample size: 100

Group B baseline score: 31 ± 9, sample size: 101

Group A 1 year score: 29 ± 9, sample size: 90

Group B 1 year score: 20 ± 10, sample size: 80

Group A 5 year score: 31 ± 11, sample size: 50

Group B 5 year score: 19 ± 8, sample size: 45

Firstly, you need to decide what time range do you want to meta-analyze. This is quite a philosophical question - you may need to decide what is relevant to your analysis. But lets consider you want all the time lines.

Being a pairwise meta-analysis, you will need to compare a mean against another mean. To do so, you have two separate meta-analysis to do: pooled depression/anxiety scale in 1 year and pooled depression/anxiety in 5 years, and you will present those results separately in your meta-analysis (despite the fact you can present an overall diamond by treating timeframes as sub-groups, but this is another topic).

Then, you will insert only the final observed results for each treatment (just leave baseline values):

Meta-analysis #1 (1 year score):

Mean group A = 29, SD group A = 9, N = 90; Mean group B = 20, SD group B = 10, N = 80 and the same thing for study #2, and #3 and so on.

Meta-analysis #2 (5 year score):

Mean group A = 31, SD group A = 11, sample size = 50; Mean group B = 19, SD group B = 8, sample size = 45.

If this is your first experience with meta-analysis, I would recommend you the RevMan software (free, from the Cochrane Collaboration). The input will ask you exactly those values for each study.

Botton line: if you are dealing with psychometric variables, there is a possibility to arrive in different forms to measure anxiety/depression, for exemple. I mean, different instruments within they proper scales. In this scenario, you will do the same thing, but in the output please use the Standardized Mean Differences instead of Weighted Mean Differences (or usually just Mean Differences). You can combine different instruments with the SMD without losing sense in your pooled effect estimate (it would be odd to combine a scale from 0-50 with another being -50 to 100).

Please do not hesitate to ask me further questions.

Good luck!

LH

Dear Lucas Helal

Thank you so much for your extremely helpful response. This is exactly what I was trying to figure out, so you've answered my question perfectly. As the studies use highly variant follow-up points, I am simply going to pick post-test (that is, time 2) as my point of observation.

Can I ask, if I want to do a meta-analysis with change scores from baseline, can I still do this on RevMan? I can't find any tutorials on how to enter this type of data on RevMan. Would you know how?

I also have effect sizes (ES) from all my papers - I don't suppose this is a common input statistic for a meta-analysis? If so, different papers have presented different ES statistics (such as Cohen's d, Hedge's g etc. Is it possible to run a meta-analysis this way?

Many many thanks again for taking the time out to respond with such clarity.

Shruti

Thanks for the tip, Sai Krishna Gudi!

Dear Shruti Raghuraman, sorry for my delay.

I'm not a big user of RevMa (a shame for a Cochrane Collaboration's Member). I've came across the software and the book and they don't mention anything in regard to change-from-baseline scores. May I ask you why meta-analyze by deltas and not by only final observations?

As for d and g, I never needed to meta-analyze this statistics. However, I googled and found something that could be useful for you: ttp://psych.colorado.edu/~willcutt/pdfs/morris_2002.pdf

For instance, the paper states what I previously mentioned - you will need to find out how to estimate variance for each d or g, once its confidence interval is poorly reported among studies (you can derivate the SD by the confidence interval).

Sorry for this not so helpful answer, but both of them are out of my previous experiences. On the other hand, please do not hesitate to contact me in any doubt when interpreting this paper. In this I can help you.

Thanks,

LH

Hi, Lucas Helal - many thanks for your response once again! And thanks for sharing the paper - I will look into it. I was just trying to see if we could meta-analyze using d/g but I will stick to means and SDs to make it easier.

The issue I am facing with final observations is that studies report different follow-up time points (some 3/4 months, some 6/9 months). I am trying to understand if it is statistically viable to pool these estimates together, or if I will have to make judgments about what studies to pool together to avoid conceptual heterogeneity.

What software would you recommend if not RevMan? I've downloaded Meta Essentials - will give it a go!

Thank you once again for being so helpful!

Best,

Shruti