 |
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: st: Questions on a meta-regression problem
I think that you need to use the method of Greenland and Longnecker
(Greenland S, Longnecker MP. Methods for trend estimation from summarized
dose-response data, with applications to meta-analysis. Am J Epidemiol
1992; 135(11):1301-1309.). This is implemented in the glst command, which
is described in the Orsini, Bellocco and Greenland Generalized least
squares for trend estimation of summarized dose–response data. Stata
Journal (2006)
6, Number 1, pp. 40–57 203. The article is reprinted in the recently
published collection "Meta-analysis in Stata".
You might also be interested in the following paper, which describes some
of the problems that can be encountered in a meta-analysis such as the one
you describe. Bekkering, G.E., Harris, R.J., Thomas, S., et al. (2008) How
Much of the Data Published in Observational Studies of the Association
between Diet and Prostate or Bladder Cancer Is Usable for Meta-Analysis?
American Journal of Epidemiology 167: 1017-1026.
Best wishes
Jonathan Sterne
--On 24 April 2009 02:33 -0400 statalist-digest
<[email protected]> wrote:
Date: Thu, 23 Apr 2009 21:38:22 +0100
From: "G Livesey" <[email protected]>
Subject: st: Questions on a meta-regression problem
Dear Statalisters
I am meta-analysing prospective studies relating health outcomes to
nutrient dose, and would gladly welcome comments and help with solutions
where possible.
Rather than meta-analyse the slopes from each study to obtain a mean, tau
and se for the slopes (such as with the GLST module) , I would like the
primary unit of study to be the quantile (irater than the study level) and
use dose values (as continuous) rather than quantile levels for dose
(categories). To achieve this involves some prior calculations to bring a
common metric to the dose and a common referent dose, which otherwise
differ between studies. The broader range of nutrient dose than found in
individual studies (>2 fold) would help with fitting an overall slope and
viewing the wider picture. Such advantage seems not to be without other,
possible disadvantage - unless statlisters have a solution:
The problem is that quantiles within studies may not be truly independent
(e.g. Studies might not have identical conditions or may be biased). The
question is. therefore, is there (or can there be) a solution to this
problem? No ready made solution is obvious (to me) in a search of
meta-analysis commands or net searches. If not, what post-analyses might
be done to limit any error incurred? For example, metareg P-values for the
coefficients might be recalculated to a lower number of degrees of
freedom, likewise for tau, but how should one define the degrees of
freedom for the coefficients and for tau, and how should one perform an
appropriate recalculation of Tau (the estimate of Tau within studies is
close to zero, possibly averaging near zero, though may not always be
so)?
The problem does not seems extensive with this dataset, however, I would
appreciate viewpoints or potential solutions on this issue . An example
dataset with meta-regression commands is described below. But does anyone
know of another module that could offer a better approach to this problem.
With thanks, and in admiration of the many solutions statalisters release
each day on the statalist.
Geoff. Livesey.
*
* 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/