+ Reply to Thread
Results 1 to 6 of 6

Thread: Meta-analysis of cross-sectional studies with both one and two study groups

  1. #1
    Points: 59, Level: 1
    Level completed: 18%, Points required for next Level: 41

    Posts
    3
    Thanks
    2
    Thanked 0 Times in 0 Posts

    Meta-analysis of cross-sectional studies with both one and two study groups




    Hello,

    I have a question about a meta-analysis I would like to run, comparing outcome measure Y in 'patients' groups (patients) and 'healthy' groups.

    I included cross-sectional studies in which Y is measured in either a patient group, a control group (healthy population), or both. That means some studies report a comparison, but some studies just report mean data for only one group. For the studies with a comparison, I also have the group means. So basically I have a bunch of 'patient' group means and a bunch of 'control' group means. See attached table for what my data looks like.

    In RevMan or Stata, using a forest plot, it is only possible to compare studies that have a comparison in it. However, since there is a lot of data available in studies reporting only one group, I would like to include these studies in the meta-analysis.

    I hope someone can help me finding an answer how to do this.

    Thanks,
    Rutger
    Attached Images  

  2. #2
    Phineas Packard
    Points: 16,013, Level: 81
    Level completed: 33%, Points required for next Level: 337
    Lazar's Avatar
    Location
    Sydney
    Posts
    1,159
    Thanks
    198
    Thanked 336 Times in 299 Posts

    Re: Meta-analysis of cross-sectional studies with both one and two study groups

    You could try something like the metasem package in R which, I believe uses FILM for missing cell information. So this might be worth trying.
    "I have done things to data. Dirty things. Things I am not proud of."

  3. The Following User Says Thank You to Lazar For This Useful Post:

    RutgerAussie (06-16-2015)

  4. #3
    Points: 59, Level: 1
    Level completed: 18%, Points required for next Level: 41

    Posts
    3
    Thanks
    2
    Thanked 0 Times in 0 Posts

    Re: Meta-analysis of cross-sectional studies with both one and two study groups

    Hello Lazar,

    Thank you for your reply.
    I'm new to R but have tried metaSEM. I managed to do a simple meta-analysis on data without missing values. However, when I import my data as above, the output for a univariate meta-analysis looks like this:

    Code: 
    Call:
    meta(y = case, v = control, data = test)
    
    95% confidence intervals: z statistic approximation
    Coefficients:
               Estimate Std.Error   lbound   ubound z value  Pr(>|z|)    
    Intercept1  22.8852    3.8543  15.3310  30.4394  5.9376 2.892e-09 ***
    Tau2_1_1    14.7161   33.2467 -50.4462  79.8784  0.4426     0.658    
    ---
    Signif. codes:  0 *** 0.001 ** 0.01 * 0.05 . 0.1   1
    
    Q statistic on the homogeneity of effect sizes: 4.396531
    Degrees of freedom of the Q statistic: 2
    P value of the Q statistic: 0.1109955
    
    Heterogeneity indices (based on the estimated Tau2):
                                 Estimate
    Intercept1: I2 (Q statistic)   0.3296
    
    Number of studies (or clusters): 8
    Number of observed statistics: 3
    Number of estimated parameters: 2
    Degrees of freedom: 1
    -2 log likelihood: 19.63753 
    OpenMx status1: 0 ("0" or "1": The optimization is considered fine.
    Other values may indicate problems.)
    It only uses the studies with a comparison, and that is the problem I am trying to solve.
    Can you (or someone else) help me any further?

    Cheers

  5. #4
    Pirate
    Points: 15,159, Level: 79
    Level completed: 62%, Points required for next Level: 191
    victorxstc's Avatar
    Posts
    875
    Thanks
    229
    Thanked 332 Times in 297 Posts

    Re: Meta-analysis of cross-sectional studies with both one and two study groups

    It is not recommended to mix studies of such hugely different qualities into one single meta-sample. Most meta-analyses simply discard the weaker studies, aiming for more reliable final conclusions.

    However, if you still want to include the weaker ones too (those with a single case or control group), you should conduct a different meta-analysis on each of them and report the three meta-analyses at once: the main meta-analysis will compare the good studies having both groups. The additional meta-analysis 1 will compare studies on patients having the disease (cases). The additional meta-analysis 2 will compare studies on people without the disease (controls).

    Plus, I am not sure if filling the missing cells in a meta-analysis is a very good idea. It is not a good idea even in single studies; but in meta-analyses, it would become less relevant because they deal with different populations and thus larger variations (and not simple inter-individual variations we face in a single study).

    Moreover, meta-analyses are very sensitive reports and should be very reliable (this is why people usually exclude weaker evidence). So if you don't want to exclude the weaker evidence, at least don't mix it with the better evidence, nor attempt to "generate" new evidence out of nothing. Simply analyze it separately but discuss them (the results from the three meta-analyses) together.
    "victor is the reviewer from hell" -Jake
    "victor is a machine! a publication machine!" -Vinux

  6. The Following User Says Thank You to victorxstc For This Useful Post:

    RutgerAussie (06-17-2015)

  7. #5
    Points: 59, Level: 1
    Level completed: 18%, Points required for next Level: 41

    Posts
    3
    Thanks
    2
    Thanked 0 Times in 0 Posts

    Re: Meta-analysis of cross-sectional studies with both one and two study groups

    Quote Originally Posted by victorxstc View Post
    It is not recommended to mix studies of such hugely different qualities into one single meta-sample. Most meta-analyses simply discard the weaker studies, aiming for more reliable final conclusions.

    However, if you still want to include the weaker ones too (those with a single case or control group), you should conduct a different meta-analysis on each of them and report the three meta-analyses at once: the main meta-analysis will compare the good studies having both groups. The additional meta-analysis 1 will compare studies on patients having the disease (cases). The additional meta-analysis 2 will compare studies on people without the disease (controls).

    Plus, I am not sure if filling the missing cells in a meta-analysis is a very good idea. It is not a good idea even in single studies; but in meta-analyses, it would become less relevant because they deal with different populations and thus larger variations (and not simple inter-individual variations we face in a single study).

    Moreover, meta-analyses are very sensitive reports and should be very reliable (this is why people usually exclude weaker evidence). So if you don't want to exclude the weaker evidence, at least don't mix it with the better evidence, nor attempt to "generate" new evidence out of nothing. Simply analyze it separately but discuss them (the results from the three meta-analyses) together.
    Thank you for your reply. I agree with most of you comments, but I do not know if a study with one study arm per definition is a much weaker study in the case of a cross-sectional study?

    You mention to do meta-analyses separately for studies with a comparison and for both conditions with only one group. I do not see how you would do this, what would you 'compare' in the meta-analysis of only 1 group studies? Or do you suggest to calculate an average mean and compare this for both groups?

    Many thanks for thinking with me!

  8. #6
    Pirate
    Points: 15,159, Level: 79
    Level completed: 62%, Points required for next Level: 191
    victorxstc's Avatar
    Posts
    875
    Thanks
    229
    Thanked 332 Times in 297 Posts

    Re: Meta-analysis of cross-sectional studies with both one and two study groups


    You are most welcome!

    You mention to do meta-analyses separately for studies with a comparison and for both conditions with only one group. I do not see how you would do this, what would you 'compare' in the meta-analysis of only 1 group studies? Or do you suggest to calculate an average mean and compare this for both groups?
    Exactly, I would suggest to calculate and report descriptive statistics for each type and then compare them.
    "victor is the reviewer from hell" -Jake
    "victor is a machine! a publication machine!" -Vinux

+ Reply to Thread

           




Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts






Advertise on Talk Stats