> Hi. I am having trouble using stata in meta-analysis.
> > I have used total 13 studies in my meta-analysis usig Random-effects model. Pulled estimate was 0.852 with 95% confidene interval (0.762 - 0.952) > using Egger's test, p value is 0.174 which suggests publication bias with little statistical significance. > however,although little publication bias in Egger;s test, funnel plot looked somewhat asymmetric so we decided to use Trim and Fill method to attain adjusted estimate. > > using commend as follows > > metatrim rr ll ul, ci eform reffect > > and result came out as follows > > Note: option 'ci' specified. > Meta-analysis > > | Pooled 95% CI Asymptotic No. of > Method | Est Lower Upper z_value p_value studies > -------+---------------------------------------------------- > Fixed | -0.102 -0.151 -0.053 -4.120 0.000 13 > Random | -0.160 -0.271 -0.050 -2.838 0.005 > > Test for heterogeneity: Q= 41.214 on 12 degrees of freedom (p= 0.000) > Moment-based estimate of between studies variance = 0.023 > > Trimming estimator: Linear > Meta-analysis type: Random-effects model > iteration | estimate Tn # to trim diff > ----------+-------------------------------------- > 1 | -0.160 41 0 91 > 2 | -0.160 41 0 0 > Note: no trimming performed; data unchanged > > Filled > Meta-analysis (exponential form) > | Pooled 95% CI Asymptotic No. of > Method | Est Lower Upper z_value p_value studies > -------+---------------------------------------------------- > Fixed | 0.903 0.860 0.948 -4.120 0.000 13 > Random | 0.852 0.762 0.952 -2.838 0.005 > Test for heterogeneity: Q= 41.214 on 12 degrees of freedom (p= 0.000) > Moment-based estimate of between studies variance = 0.023 > > ------------------------------------------------------------------------------------------------------------- > > I have used metatrim with other dataset before and it added some studies to reduce the asymmetric funnel plot. > with additional studies filled meta-analysis showed publication bias adjusted pulled estimate. > > However, using this dataset, to trimming performed and data unchanged. I think even egger;s test showes insignificant publication bias, there should be some cahnge in the estimate using trima nd fill method. so.. I believed there is something wrong in the way I used 'metatrim'. > > I would really appreciate some help by experts. Thank you very much for readling this long question. > > > > > * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/ |
Dear Chan,
If I got the big picture correctly, your results seem to be OK: the command is correct and the results are intuitive. Firstly, as the test for small-study effects indicates that there is no substantial bias, it is reasonable to expect some agreement between it and the 'trim and fill' approach. In other words, there is no reason to expect the trim and fill to adjust for publication bias if it in fact doesn't exist. If you have raw data (i.e. 2x2 tables), try to investigate small study effect using the Harbord's test as well (metabias a b c d, harbord) Secondly, I would be very concerned to use the trim and fill approach in results showing strong evidence for statistical heterogeneity. This seems to be the case in your data set of 13 studies: Q= 41.214 (12 df) and estimated tau^ = 0.023. To understand the problem, see: http://www.ncbi.nlm.nih.gov/pubmed/12820277 http://www.ncbi.nlm.nih.gov/pubmed/17476644 Thirdly, visual inspection of the funnel plot is highly subjective with 13 studies only. Indeed, the asymmetry in the funnel plot that you see may be either entirely subjective or a result of the statistical heterogeneity across study results. hope this helps. Tiago * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/ |
Free forum by Nabble | Edit this page |