Software for meta-analysis of DTA studies

Researchers have prepared macros or modules for meta-analysis of diagnostic test accuracy studies in different statistical analysis software programs. Below we provide brief information and links to the tutorials we have developed to aid meta-analysts in fitting hierarchical models for test accuracy meta-analysis in SAS, R and Stata. Additional information and other analyses, including meta-analysis within a Bayesian framework can be found in Chapter 11 Undertaking meta-analysis of the Cochrane Handbook for Systematic Reviews of Diagnostic Test Accuracy (Version 2).

 SAS

MetaDAS: A SAS macro for meta-analysis of diagnostic accuracy studies, contains both the bivariate and the HSROC model. Please find the required documents hereunder:

  • User guide version 1.3 (2012). (PDF 2.7MB, opens in new window)
  • Quick reference and worked example (2012). (PDF 2.6MB, opens in new window)
  • The SAS macro itself: METADAS v1.3. This is provided as a text-file and opens in a new window. 

R

There are several user-written packages for conducting meta-analysis of diagnostic test accuracy (DTA) studies in R. This tutorial includes a summary of their characteristics and provides step-by-step instructions on carrying out the bivariate binomial method by fitting a generalized linear mixed model (GLMM) using the glmer function in the R package lme4. In August 2021 we updated version 1 of the tutorial to version 2. A .R file, “Bivariate binomial meta-analysis of diagnostic test accuracy studies v2.0.R” and example dataset based on a review by Schuetz et al. 2010, are included with the tutorial in the zipped folder.

STATA

METANDI: A Stata user-written package for meta-analysis of diagnostic accuracy studies (Harbord and Whiting 2009; Harbord 2008). metandi performs bivariate meta-analysis of sensitivity and specificity using a generalized linear mixed model approach (Reitsma et al 2005; Chu & Cole 2006). In Stata 10 and above, metandi fits the model using the built in command xtmelogit by default. In Stata 8 or 9, metandi uses the user written function gllamm (Rabe-Hesketh et al 2004). metandi can be found from within Stata by typing 'findit metandi' or installed from within Stata by typing 'ssc install metandi' (while connected to the internet). You may also need to install gllamm.

METANDI cannot be used to formally investigate heterogeneity or to compare the accuracy of two or more tests because it does not have an option for including a covariate in the bivariate model. However, such analyses are possible by using xtmelogit or megrlogit directly. In August 2021 we updated the previous tutorial to the current version 2. This practical tutorial guides both novice and experienced Stata users on how to fit the bivariate model using metandi or megrlogit. The syntax for xtmelogit is the same as the syntax for megrlogit so you can substitute directly in the code if you are using Stata 10, 11 or 12. The example dataset used in the tutorial, schuetz.csv, is based on a published diagnostic test accuracy review (Schuetz et al. 2010). A do-file, "Meta-analysis of test accuracy studies in Stata v2.0.do", accompanies this tutorial.

References

Chu H, Cole SR. Bivariate meta-analysis for sensitivity and specificity with sparse data: a generalized linear mixed model approach (letter to the Editor). J Clin Epidemiol. 2006;59(12):1331-2.

Harbord RM, Whiting P. metandi: Meta-analysis of diagnostic accuracy using hierarchical logistic regression. Stata Journal. 2009;9(2):211-29.

Harbord, R. metandi: Stata module for meta-analysis of diagnostic accuracy. Statistical Software Components, Boston College Department of Economics. Revised 15 Apr 2008.

Rabe-Hesketh S, Skrondal A, Pickles A. "GLLAMM Manual" (October 2004). U.C. Berkeley Division of Biostatistics Working Paper Series. Working Paper 160. Accessed at http://biostats.bepress.com/ucbbiostat/paper160 on 31 October 2013.

Reitsma JB, Glas AS, Rutjes AW, Scholten RJ, Bossuyt PM, Zwinderman AH. Bivariate analysis of sensitivity and specificity produces informative summary measures in diagnostic reviews. J Clin Epidemiol. 2005;58(10):982-90.

Schuetz GM, Zacharopoulou NM, Schlattmann P, Dewey M. 2010. Meta-analysis: noninvasive coronary angiography using computed tomography versus magnetic resonance imaging. Ann Intern Med. 2010;152(3):167-77.