Meta-analysis of diagnostic test using MCMC

Hi. Is there some good hands-on tutorial about running MCMC analysis for the meta-analysis of diagnostic tests? Is it inbuilt STATA Bayesian analysis is sufficient for this task or is necessary to install some additional patch?



Omega Contributor
Interesting stuff. I am not familiar enough with it to be of much help. They seem to run multilevel modeling to control for study heterogeneity, but then incorporate Bayesian priors along with using a MCMC approach. I have not used WinBUGS, but I would wonder if their software documentation may be a place to start. You may be better off not focusing on the Meta-analysis part but more on the multilevel models (hierarchical reg). It should be the same approach, so if you can figure it out, then you should be able to apply it to the meta-analysis setting.

Thanks and good luck.
Thanks. From what I know they use true positive, true negative, false positive and false negative results from diagnostic tests as priors (random variables) and then they apply the function to those which as outcomes give sensitivity and specificity. After that, they repeat this using MCMC in 10 000 -15 000 iterations which on the end results in probabilistic results of sensitivity and specificity with credible intervals.
I already have finished modeling with hierarchical SROC (HSROC) model of Rutter and Gatsonis but I want to have the double methodological approach and overcome some pitfalls of Rutter and Gatsonis models with this MCMC approach, especially in the subgroup analysis.
Anyway, I will try with WinBugs technical documentation and try to better understand this approach.
Last edited:


Omega Contributor
Not to reactivate this thread, but I used the following recently:

Metadas v1.3 beta Macro by Takwoingi in SAS 9.4, Cary NC

It appeared to be endorsed by Cochrane Collaborative and provides HSROC. I also seem to remember reading that there are some types of issues with trying to calculate accuracy of diagnostics using meta-analyses and you should focus more on SEN and SPEC, but I can't recall the source or full context (which method it was related to).