st: RE: FW: "no trimming performed" with publication bias evidence

[Timothy Mak <tshmak@hku.hk>]

Hi Nassib, 

I hope you understand that it is difficult to say whether there is publication bias or not based on statistics alone. You'd be better off thinking hard about possible ways some studies might have been missed rather than rely on the Egger statistics. As for trim-and-fill, that is again a very "ad hoc" way of addressing publication bias, bordering on being ridiculous. Saying whether there is or is not publication bias based on these statistics is probably even worse than saying whether there is an effect or not based on whether the p-value is < 0.05. 

Tim

-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Nassib Bueno
Sent: 04 July 2013 06:19
To: statalist@hsphsun2.harvard.edu
Subject: st: FW: "no trimming performed" with publication bias evidence

Dear Statalisters,

I am performing a meta-analysis of eleven randomized controlled trials using Stata 9.1 for windows. 
When I used the -metabias- command, the Egger's regression test showed  evidence of publication bias (Slope coefficient: 0.18; std error: 0.34; P = 0.61; Bias coefficient: -1.54; Std error: 0.65; P = 0.042).
Syntax used: -metabias _ES _seES-

Typing -which metabias- shows me:
. which metabias
*! version 1.2.0  TJS 17feb00   Version 6 and gweight option (STB-58: sbe19.3)
*! version 4.1.0  26sep97 TJS
*! version 1.2.1  19nov98

Then, I used the -metatrim- command to perform a trim and fill analysis. I am working on a continuous variable. 

Typing -which metatrim- shows me:
. which metatrim
*! version 1.0.5  22jul2003 TJS  added flip, mixed, logical switches

Syntax used: -metatrim _ES _seES, reffect funnel-
The output come as follows:

"Note: default data input format (theta, se_theta) assumed.

Meta-analysis 

       |  Pooled      95% CI         Asymptotic      No. of
Method |     Est   Lower   Upper  z_value  p_value   studies
-------+----------------------------------------------------
Fixed  |  -0.555  -0.870  -0.240   -3.452    0.001     11
Random |  -0.695  -1.102  -0.287   -3.337    0.001

Test for heterogeneity: Q= 13.667 on 10 degrees of freedom (p= 0.189)
Moment-based estimate of between studies variance =  0.117

Trimming estimator: Linear
Meta-analysis type: Random-effects model

iteration |  estimate    Tn    # to trim     diff
----------+--------------------------------------
    1     |   -0.695     24         0          66
    2     |   -0.695     24         0           0

Note: no trimming performed; data unchanged

Filled 
Meta-analysis 

       |  Pooled      95% CI         Asymptotic      No. of
Method |     Est   Lower   Upper  z_value  p_value   studies
-------+----------------------------------------------------
Fixed  |  -0.555  -0.870  -0.240   -3.452    0.001     11
Random |  -0.695  -1.102  -0.287   -3.337    0.001

Test for heterogeneity: Q= 13.667 on 10 degrees of freedom (p= 0.189)
Moment-based estimate of between studies variance =  0.117"

Based on the absence of trimming, can I conclude that, even in the presence of a significant P value for the Egger's test, indicating publication bias, this bias is not likely to affect the results? Or something is wrong and Stata should have performed trimming in this case? Or maybe, due to the reduced number of studies, I should not be concerned in testing for publication bias and/or trim-and-fill analysis?

Thanks for your consideration,

Nassib Bueno 		 	   		  
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/