The problem with combining change scores and final values in a meta-analysis using SMD is that you standardize using the SD. SDs from changes are often smaller than SDs from final values because of less variance in changes then final values. Combining them in an SMD you run the risk of giving relative more weight to the studies using change values than final values. 

If you need to use the SMD because your outcome is basically measuring the same underlying construct for example pain, but the studies use different scales, I would recommend either doing separate analysis on final values and change values, contacting the authors of the trials and ask for final or change values or as a last resort try to calculate the change. I would recommend that you ask a statistician for help with this.

The reason that many prefer final values over change is that there is a risk of selective reporting. The choice of whether to report the change or final values might depend on what gives the most 'favorable' result, introducing bias into your meta-analysis. Example: If small baseline imbalances favor the intervention group you opt for final if they favor the control you opt for change, to correct for them.

Hope this helps.

Cheers,

Britta 

Hi Valter

If you are referring to an RCT, then treat the change scores as if they were the mean difference of the final scores, but be careful that the SD of change score is used. 

Adrian

Hello Valter

I think it depends. If you only got the change scores or if the base-line values are significant different.

Dear Valter,

For studies with two group with before after data, you need to have the following data to be able to do meta-analysis:

1- Mean change of each group which can be calculated easily by subtraction of before after means 

2- number of patients in each group

3- the most tricky one is Standard Deviation of changes in each group. You can not get this one by subtraction of before and after SD. It should be reported by the authors. It is not enough to have before treatment SD and after treatment SD.  This is why you can not perform meta-analysis on many of these studies. 

If you have the above-mentioned data of each included RCT you can do meta-analysis. Unfortunately this is not the case for most reported RCTs.

Ramin

Hello guys. Thank you for their responses. However, I still have doubts.

Supposing for example that I need to apply SMD, e.g., all studies measure quality of life but they use different scales. I got mean and SD of change score for both groups (control and experimental). According to Cochrane Handbook (Section '9.4.5.2 Meta-analysis of change scores') "...final value and change scores should not be combined together as standardized mean differences, since the difference in standard deviation reflects not differences in measurement scale, but differences in the reliability of the measurements."

This statement is not clear to me. Can anyone explain this statement to me?

Many thanks.

http://handbook.cochrane.org/index.htm#chapter_9/9_4_5_2_meta_analysis_of_change_scores.htm

Hi Valter,

if you have change scores and SD of change scores for all studies then you ARE allowed to perform meta-analysis on change scores and compute a SMD.  What the Cochrane book is referring to in the specific passage that you quoted is that you are not allowed to mix final scores and change scores for computation of SMD.

Hi Valter again,

The sd of the change score reflects the variability of the change on that specific scale. If you can't get hold of the final scores by asking the authors, you should use the change scores and separate that study as a subgroup from the remaining studies. Computing of sd of change scores, if not available, is not that tricky and Cochrane suggests computing correlation coefficients.Use different ones (small, medium, large) or one of a similar study.

The problem with combining change scores and final values in a meta-analysis using SMD is that you standardize using the SD. SDs from changes are often smaller than SDs from final values because of less variance in changes then final values. Combining them in an SMD you run the risk of giving relative more weight to the studies using change values than final values. 

If you need to use the SMD because your outcome is basically measuring the same underlying construct for example pain, but the studies use different scales, I would recommend either doing separate analysis on final values and change values, contacting the authors of the trials and ask for final or change values or as a last resort try to calculate the change. I would recommend that you ask a statistician for help with this.

The reason that many prefer final values over change is that there is a risk of selective reporting. The choice of whether to report the change or final values might depend on what gives the most 'favorable' result, introducing bias into your meta-analysis. Example: If small baseline imbalances favor the intervention group you opt for final if they favor the control you opt for change, to correct for them.

Hope this helps.

Cheers,

Britta 

Dear Dr Britta,

Thank you for your valuable explanation.

Best regards,

Valter

Hi,

I have 14 studies, 11 of which provide final scores. For the other 3 studies, it simply provides one score which suggest they are the change scores (2 of the 3 state pre and post measurements, the other 1 does not state anything). Am I allowed to use SMD, or not?

Dear Azeem Rathore,

You do not need to use SMD. It is applyed when different studies usaged different outcome measureament tools.

For the data you described apply mean difference. It is not a problem combining data of post intervention and change from baseline in a meta-analysis.

Please, take a look in:

Silva V, et al. Statistical simulation to assess results of meta-analyses using post-intervention, change from baseline and mixed methods. In: Better Knowledge for Better Health | Un meilleur savoir pour une meilleure santé. Abstracts of the 21st Cochrane Colloquium; 2013 19-23 Sep; Québec City, Canada. John Wiley & Sons; 2013. https://abstracts.cochrane.org/2013-québec-city/statistical-simulation-assess-results-meta-analyses-using-post-intervention-change

Can anyone recommend publications on this topic?

@anna Here using metafor package in R

http://www.metafor-project.org/doku.php/tips:assembling_data_smd