Meta analysis of observational studies

[kapil_sas]

Hello sir/madam,

           My name iskapil, I am a post graduate in pharmacy. I am unable to perform meta analysisof observational studies using sas 9.1.3. I have cohort and case-controlstudies with me where i have to pool up using fixed and random effects model,identifying publication bias and heterogeneity. Please can any one assist me ............

[Doc_Duke]

SAS is not really designed for doing meta-analysis, whether of clinical trials or observational studies, unless you have access to the original data from each study (extremely rare).  You are much better off with some software dedicated to meta-analysis.  Here is a reasonable list

http://www.uky.edu/ComputingCenter/SSTARS/www/documentation/MetaAnalysis.htm

The Cochrane Collaborative ( http://www.cochrane.org/ ) has really good resources as well.

Doc Muhlbaier

Duke

PS.  For the benefit of the readers, please just post to one forum for each question.  It helps organize the reponse threads.

[lvm]

It appears that many of your questions are not SAS questions, but meta-analysis questions. That is, there are many things to do in meta-analysis before the actual mixed model analysis. I assume you have access to the standard excellent books on the subject. I highly recommend the new book Introduction to Meta-Analysis, by Borenstein et al. You must decide on the effect size for each study, which becomes the response variable in the meta-analysis. You also need the estimted standard error for the estimated effect size from each study. The square of the SE is known as the "sampling variance" with meta-analysts. To use standard software (not specifically written for meta-analysis), you must fix the sampling variance as a separate residual for each study (each observation in the meta-analysis). This can easily be done with proc mixed or proc glimmix. There are tricks to do this quite easily (or relatively easily). The on-line supplement that goes with the new article: Meta-Analysis for evidence synthesis in plant pathology: An overview (journal: Phytopathology, vol. 101, pp. 16-30 [2011]; authors: L.V. Madden and P.A. Paul) gives great details on the use of SAS for meta-analysis. This article also goves over publication bias and power in some detail.

Good luck.

[lvm]

Forgot to mention, for a slightly older article, read: van Houwelingen et al. [2002]. Advanced methods in meta-analysis: multivariate approach and meta-regression. Statistics in Medicine, vol 21, pp. 589-624. This is a great tutorial article that is based entirely on sas (proc mixed) for meta-analysis. But note, the recent on-line supplement I mentioned in the previous posting gives a couple of easier ways of fitting the identical mixed model to the effect-size data.

[kapil_sas]

Thanks alot @ Doc@Duke, lvm. As i am new to the forms i dont have idea, that is why i repeated my questions in different forms. I wont repeat this again.

[kapil_sas]

Hello sir/madam,

                         I followed the instructions from above comments/replies, but i am unable to execute a SAS program. Can any one support me in writing program. Data available with me is

1. Effect estimate (odds ratio/ Risk ratio)

2. 95% Confidence interval

3. Standard error

I want to calculate:

1. Heterogeniety in studies (cochrane Q test and I2 test)

2. Mantel Hanszel test (to pool the effect estimate)

3. To plot forest plot

4.  To check publication bias using funnel plot and Beggs test/Eggers test

Example:

study                   Effect estimate                          95%CI                                      

kasper et al           0.89                                           0.5826-1.3595                                             

lin et al                 1.1                                              0.83-1.45

Vieneis et al          1.3                                              0.702-2.40       

I have data like this, where i can calculate the standard error. i want to pool the effect estimates......

Please help me.........

[lvm]

The articles I mentioned give the details on doing most of what you are requesting.  I can't post the new article (or supplemental sas instructins) because they are copyrighted by the journal, and open access starts a year after publication. If you give me your email, I will send you the information. Some of the plots will have to be done with GPLOT or the SG graphics procedures. In addition to the previously mentioned articles, the book by Whitehead (Meta-Analysis of Controlled Clinical Trials [2003]) is very useful, and gives many more details on the use of SAS (MIXED) for meta-analysis. Note, as suggested by Doc@Duke, you might want to look into one of the customized meta-analysis programs, if you are not a regular user of SAS (especially of MIXED). SAS does not do an "automatic" meta-analysis, but with the right options, one can do a full meta-analysis with MIXED (or GLIMMIX).

[kapil_sas]

Thanks alot sir  @ IVM. I have read those articles which you have mentioned, but those articles have mentioned about clinical trials but not about observational studies where there is no intervention given. I am not a regular user of MIXED procedure that is why i am unable to write code using proc mixed for observational study. My mail id is "kapil.gudala@gmail.com". IF i want to publish my data, i have to provide licensce number of the particular software. I cannot use trial version of them. I have access to SAS from my university.

Thanks again for your support.

[lvm]

You can conduct a meta-analysis on the results from controlled studies or from observational studies. The Borenstein et al. book I previously mentioned discusses this in detail (see pages 359-361, for example). There are important things to watch out for with observational studies (read those pages and other parts of the book related to the subject), but it is certainly common to do this type of analysis. The meta-analysis is done in the same way. An important thing to keep in mind: whether the original studies are controlled (randomized, replicated, etc.) or are observational, the actual meta-analysis is of the observational type. One simply starts with "observations" of estimated effect sizes and their standard errors for each study, which is a set of observations. The precautions about the interpretation of individual observational studies always apply to the results from a meta-analysis.