In other words, if I have 5 studies and 2 only have continuous variables that are post-test values and the other 3 studies actually have the posttest minus prettest values can I add all them together or do they need to be uniform (i.e for the 3 studies with change, if they have posttest values that is)?
EDIT: And the values are measured using different instruments.
Ref for help: http://handbook.cochrane.org/chapter_16/16_1_3_2_imputing_standard_deviations_for_changes_from_baseline.htm
Hi Azeem Rathore,
I understand that you have 5 studies and there are two problems with pooling these studies together:
Question (1): Three studies reported the mean change from baseline while 2 studies reported the post-test value only.
RE: There is no consensus regarding whether to consider the post-test value or the mean change from baseline while calculating the effect estimate for the meta-analysis. However, some report suggest that the mean change from baseline is more accurate. This opinion is advocated by the randomization of participants to the study groups making the groups nearly balanced and any differences are usually due to chance. However, from my experience in multiple work, the change from baseline alone might be misleading especially if the scale of your outcome is large (e.g. a score of 100 points). Therefore, if the studies have balanced baseline variables with proper randomization, you can proceed with the post-test values only. If not, I advise you to calculate the mean change from baseline manually or contact the study authors and get the unpublished data to calculate the mean change from baseline.
Question (2): The studies reported the outcome on different scales.
RE: This type of meta-analysis is possible if you considered standardization of the effect estimate. This is usually done by dividing the mean difference by the standard deviation, to get the standardized mean difference. If you are using the Review Manager software, you will find it as the option of (Continuous data > SMD). It is also called (Cohen's d) and can be calculated as (MD/SD). Therefore, you will transform all raw differences into standardized differences. Note that if the scores are in different direction (e.g. score A increases with the severity of disease but score B decreases with severity of the disease), you will need to make adjustment for the score direction through multiplying the effect estimate by (-1) to the different scores to make all estimates in the same direction.
I hope this helps.
Best Wishes,
Ahmed Negida
Thank you for a very informative reply. A follow-up question, according to Cochrane it says as follows: "There is no statistical reason why studies with change-from-baseline outcomes should not be combined in a meta-analysis with studies with final measurement outcomes when using the (unstandardized) mean difference method in RevMan." I am confused by the bolded--does this mean when I conduct my analysis I should not use SMD as the function?
Source: http://handbook.cochrane.org/chapter_9/9_4_5_2_meta_analysis_of_change_scores.htm
EDIT: This was also stated in the next paragraph: "However, 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." I suppose I am more confused on a conceptual level than anything really....
I also do not know how to calculate the SD of a change score if the pre and post means and SDs are provided--is it as easy as subtracting the SDs as well (I don't think it would be?) And how to account for differences in sample size....
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis