[R-meta] diagnostic meta-analysis of studies with multiple readers

Philipp Doebler doebler at statistik.tu-dortmund.de
Sat Feb 3 14:20:28 CET 2018

Dear Mario,

your are correct to note that dependencies among readers must be respected.
I would also recommend you try a bivariate approach.

As a consequence I suggest you work with a logit-link function and aim to
estimate two random effects:

1. a bivariate within-study (reader-level) effect that captures the
deviations of the reader from the study mean, and
2. a between study effect, similar to the Reitsma et al. (2005) bivariate
model, which you are probably aware of.

Estimating 1. is a bit tricky, as some studies will report only two
readers, which is clearly suboptimal.  There are two ways to go:

(i) it could pay to "borrow strength" (in the Bayesian sense) by specifying
an (Inverse-Wishart) prior distribution for the within study effect, that
is maybe scaled by a study specific factor. This amounts to the assumption,
that the covariance matrix of the deviations from the study mean is similar
in all studies. I am aware of one paper with a similar idea by Verde (
http://onlinelibrary.wiley.com/doi/10.1002/sim.4055/abstract) which also
has WinBUGS code.
(ii) Try and use your favourite GLMM software (say glmer or SAS PROC
MIXED). It could be tricky to take into account that the within study
covariance matrix is (probably) not the same for all studies, so you will
probably have to assume it is the same.

I hope this rough sketch is helpful. Let me know if you need more precise

Kind wishes,

On Fri, Feb 2, 2018 at 7:53 PM, Mario Petretta <petretta at unina.it> wrote:

> Thanks for your kindly replay.
> At this time I have 14 studies and 4 of them give information useful to
> obtain the number of true-positive, true-negative, false-positive and false
> negative for more than one reader (two in two studies and three in the
> other
> two); three  studies  include reproducibility information (Bland Altman
> plots or ICC).
> For all the four studies with multiple readers, the readers evaluated the
> same imaging examination of all study patients, i.e. there was a unique
> study for patient but the imaging studies were read independently and
> blinded. Thus, sensitivity and specificity vary across the readers.
> Among the various possibilities I thought to report the same study several
> times in the analysis with the different 2 x 2 tables, but the problem is
> how do you take into account that it is for some studies of the same
> population (a sort of cluster). I also hypothesize to do, thereafter, a
> sensitivity analysis including only the results of the best or worst
> readers
> or the average of the different readers.
> Thanks for the attention.
> Mario
> __________________________________________________
> Date: Fri, 2 Feb 2018 18:57:33 +0100
> From: Philipp Doebler <doebler at statistik.tu-dortmund.de>
> To: Mario Petretta <petretta at unina.it>
> Cc: r-sig-meta-analysis at r-project.org
> Subject: Re: [R-meta] diagnostic meta-analysis of studies with
>         multiple        reader
> Message-ID:
>         <CAMU7UxFGYpLYesqKc2+2xpphqkQWhzsn=Py9M9C08VL6k9BoD
> w at mail.gmail.com>
> Content-Type: text/plain; charset="UTF-8"
> Dear Mario,
> could you give us some more details? Is it realistic to obtain a 2x2-table
> for each reader?
> Best,
>   Philipp
> On Fri, Feb 2, 2018 at 6:18 PM, Mario Petretta <petretta at unina.it> wrote:
> > Dear all.
> > I'm planning a diagnostic meta-analysis with an imaging test and a
> > binary outcome (yes/no) .
> > Some (not all) studies contain multiple readers, which means that more
> > than one physician interprets each examination.
> >
> > At present, it appears that there are no recommendations for which
> > strategy is optimal (see Eur J Radiol. 2017;93:59-64 Systematic Reviews
> 2017;6:194).
> >
> > I would appreciate very much suggestions on this topic.
> > Thanks for the attention.
> > Sincerely
> > Mario Petretta
> > ___________________________________________
> > Mario Petretta, MD, FAHA
> > Associate Professor of Internal Medicine
> > Department of Translational Medical Sciences
> > Naples University "Federico II" - Italy
> > _______________________________________________
> > R-sig-meta-analysis mailing list
> > R-sig-meta-analysis at r-project.org
> > https://stat.ethz.ch/mailman/listinfo/r-sig-meta-analysis
> >
> --
> Prof. Dr. Philipp Doebler
> Technische Universit?t Dortmund
> Fakult?t Statistik
> Vogelpothsweg 87
> 44227 Dortmund
> Tel.: +49 231-755 8259
> Fax: +49 231-755 3918
> doebler at statistik.tu-dortmund.de
> www.statistik.tu-dortmund.de/1261.html

Prof. Dr. Philipp Doebler
Technische Universität Dortmund
Fakultät Statistik
Vogelpothsweg 87
44227 Dortmund

Tel.: +49 231-755 8259
Fax: +49 231-755 3918
doebler at statistik.tu-dortmund.de

Wichtiger Hinweis: Die Information in dieser E-Mail ist vertraulich. Sie
ist ausschließlich für den Adressaten bestimmt. Sollten Sie nicht der
für diese E-Mail bestimmte Adressat sein, unterrichten Sie bitte den
Absender und vernichten Sie diese Mail. Vielen Dank.
Unbeschadet der Korrespondenz per E-Mail, sind unsere Erklärungen
ausschließlich final rechtsverbindlich, wenn sie in herkömmlicher
Schriftform (mit eigenhändiger Unterschrift) oder durch Übermittlung
eines solchen Schriftstücks per Telefax erfolgen.

Important note: The information included in this e-mail is confidential.
It is solely intended for the recipient. If you are not the intended
recipient of this e-mail please contact the sender and delete this
message. Thank you.
Without prejudice of e-mail correspondence, our statements are only
legally binding when they are made in the conventional written form
(with personal signature) or when such documents are sent by fax.

	[[alternative HTML version deleted]]

More information about the R-sig-meta-analysis mailing list