{smcl} {* April 15, 2008 @ 21:51:02}{...} {cmd:help metandi} {right: (Roger Harbord)} {right:also see: {help metandi postestimation}} {hline} {title:Title} {hi:metandi} {hline 2} Meta-analysis of diagnostic accuracy {title:Syntax} {p 8 17 2} {cmd:metandi} {it: tp fp fn tn} [{it:{help if}}] [{it:{help in}}] [{cmd:,} {cmdab:p:lot} {cmdab:g:llamm} {cmd:ip(g|m)} {cmdab:ni:p(}{it:#}{cmd:)} {cmdab:nob:ivariate} {cmdab:noh:sroc} {cmdab:nos:ummarypt} {cmdab:d:etail} {cmdab:l:evel(}{it:#}{cmd:)} {cmd:allc} {cmdab:tr:ace} {cmdab:nolo:g} {cmdab:do:ts}] {p 4 6 2} {cmd:by} may be used with {cmd:metandi}; see {helpb by}.{break} {p 4 4 2} See {help metandi postestimation} for features available after estimation, in particular the {cmd: predict} command. {helpb metandiplot} graphs the results from {cmd:metandi}. {title:Contents} {p 2 4 2}{help metandi##s_description:Description} - {help metandi##s_options:Options} - {help metandi##s_remarks:Remarks} - {help metandi##s_examples:Examples} - {help metandi##s_references:References} - {help metandi##s_acknowledgements:Acknowledgements} - {help metandi##s_author:Author} - {help metandi##s_citation:Citation of metandi} - {help metandi##s_also_see:Also see} {marker s_description}{title:Description} {p 4 4 2} {cmd:metandi} performs meta-analysis of diagnostic test accuracy studies in which both the index test under study and the reference test (gold standard) are dichotomous. It takes as input four variables {it:tp fp fn tn} giving the number of true positives, false positives, false negatives and true negatives within each study. It fits a two-level mixed logistic regression model, with independent binomial distributions for the true positives and true negatives conditional on the sensitivity and specificity in each study, and a bivariate normal model for the {help f_logit:logit} transforms of sensitivity and specificity between studies. {p 4 4 2} In Stata 10 {cmd:metandi} fits the model using the built-in command {helpb xtmelogit} by default. In Stata 8 or 9 it makes use of the user-written command {helpb gllamm}, which must therefore be installed. To ensure you have the most recent version of {cmd:gllamm}, type {cmd: ssc install gllamm, replace}. {p 4 4 2} {cmd: metandi} does not allow covariates to be fitted, i.e. meta-{it:regression} of diagnostic accuracy is not supported. {marker s_options}{title:Options} {dlgtab:Plot} {p 4 8 2} {cmd:plot} requests a plot of the results on a summary receiver operating characteristic (SROC) plot. This is a convenience option equivalent to executing the {helpb metandiplot} command after {cmd:metandi} with the same list of variables {it: tp fp fn tn} (and the same {it:if} and {it:in} qualifiers if specified). Greater control of the plot is available through the options of the {cmd:metandiplot} command when issued separately. {p 4 4 2} The remaining options below control details of the estimation algorithm and reporting of results. Typically you will not need to specify any of them. However, this model can be difficult or slow to fit to certain datasets, in which case they may prove useful. {dlgtab:Estimation} {p 4 8 2} {cmd:gllamm} specifies that the model be fitted using {helpb gllamm}. This is the default in Stata 8 and 9 so the option is only of use in Stata 10, in which the model is fitted using {helpb xtmelogit} by default. {p 4 8 2} {cmd:ip(g}{c |}{cmd:m)} specifies the quadrature method: {cmd:ip(g)}, the default, gives Cartesian product quadrature. {cmd:ip(m)} is only available when the model is fitted using {cmd:gllamm}. It gives spherical quadrature, which can be more efficient though its properties are less well known and it can sometimes cause adaptive quadrature to take longer to converge. See {help metandi##rhsp:Rabe-Hesketh, Skrondal & Pickles (2005)}. {p 4 8 2} {cmd:nip(#)} specifies the number of integration points used for quadrature. Higher values should result in greater accuracy, but typically at the expense of longer execution times. Specifying too small a value can lead to convergence problems, or even failure of adaptive quadrature - if {cmd:gllamm} reports the error "log-likelihood cannot be computed", try increasing {cmd:nip()}. For Cartesian product quadrature, {cmd:nip()} specifies the number of points for each of the two random effects; the default is {cmd:nip(5)}. For spherical quadrature, {cmd:nip()} specifies the degree {it:d} of the approximation; the default is {cmd:nip(9)}, and the only values currently supported by {cmd:gllamm} are 5, 7, 9, 11 and 15. These defaults give approximately the same accuracy, as degree {it:d} for spherical quadrature corresponds in accuracy approximately to ({it:d} + 1)/2 points per random effect for Cartesian product quadrature. ({help metandi##rhs:Rabe-Hesketh & Skrondal 2005}, appendix B.) {dlgtab:Reporting} {p 4 8 2} {cmd:nobivariate}, {cmd:nohsroc} and {cmd:nosummarypt} suppress reporting of the bivariate parameter estimates, the HSROC parameter estimates or the summary point estimates respectively. {p 4 8 2} {cmd:detail} displays the default output of all {cmd:gllamm} or {cmd:xtmelogit} commands issued. If the model is fitted using {cmd:gllamm}, this includes the output of two univariate models used to give good starting values for the bivariate model. {p 4 8 2} {cmd:level(}{it:#}{cmd:)} specifies the confidence level, in percent, for confidence intervals of the coefficients; see help {help level}. {p 4 8 2} {cmd:nolog} suppresses display of the iteration log. {p 4 8 2} {cmd:trace} adds a display of the current parameter vector to the iteration log. {marker s_remarks}{title:Remarks} {dlgtab:Models} {p 4 4 2} Estimates are displayed for the parameters of both the bivariate model ({help metandi##reitsma:Reitsma et al. 2005}) and the Hierarchical Summary Receiver Operating Characteristic (HSROC) model ({help metandi##rutter:Rutter & Gatsonis 2001}). In the models without covariates fitted by {cmd:metandi}, these are different parameterisations of the same model ({help metandi##harbord:Harbord et al. 2007}, {help metandi##arends:Arends 2006}). {p 4 4 2} {cmd:metandi} fits the model using the approach {help metandi##chu:Chu & Cole (2006)} refer to as a generalized linear mixed model approach to bivariate meta-analysis of sensitivity and specificity. As they indicate, in the common situation of low cell counts this may be preferable to the approach originally used by {help metandi##reitsma:Reitsma et al. (2005)} involving empirical logit transforms. This generalized linear mixed model approach corresponds to the "empirical Bayes" approach to fitting the HSROC model introduced by {help metandi##macaskill:Macaskill (2004)}. {dlgtab:Output} {p 4 4 2} In the output of {cmd:metandi}, the bivariate parameters are denoted using E() for expectation (mean), Var() for variance, Corr() for correlation, logitSe for logit(sensitivity) and logitSp for logit(specificity). The HSROC parameters are denoted using the notation of {help metandi##rutter:Rutter & Gatsonis (2001)}, giving Latin names for Greek letters but abbreviating {it:sigma}^2_{it:alpha} and {it:sigma}^2_{it:theta} to s2alpha and s2theta. {p 4 4 2} The correlation rather than the covariance is displayed for the bivariate parameterisation as the correlation is more interpretable and its confidence interval can usually be computed. However, the correlation is often estimated imprecisely and may sometimes be estimated as -1 (or +1) ({help metandi##riley:Riley et al. 2007}). This corresponds to estimating s2alpha (or s2theta) to be very close to zero. When this occurs the estimated standard error and confidence interval for both are unreliable and will often be displayed as missing values. {p 4 4 2} The following estimates for the summary point are also displayed: sensitivity (Se), specificity (Sp), diagnostic odds ratio (DOR), positive likelihood ratio (LR+), negative likelihood ratio (LR-), inverse of the negative likelihood ratio (1/LR-). {p 4 4 2} Confidence intervals are computed assuming asymptotic normality after a log transformation for variance parameters and for DOR, LR+ and LR-, an {help atanh} (Fisher's {it:z}) transformation for the correlation, and a {help f_logit:logit} transformation for proportions. {* (This is implemented using {help _diparm}.) } {p 4 4 2} The covariance between the estimates of E(logitSe) & E(logitSp) is displayed at the end of the output as this is needed in addition to the other bivariate parameter estimates to construct confidence or prediction regions around the summary point. {dlgtab:GLLAMM} {p 4 4 2} For detailed information on the {cmd:gllamm} command, see the GLLAMM manual ({help metandi##gllammmanual:Rabe-Hesketh, Skrondal & Pickles 2004}) and {help metandi##rhs:Rabe-Hesketh & Skrondal (2005)}. {marker s_examples}{title:Examples} {p 4 8 2}{cmd:. metandi tp fp fn tn} {p 4 8 2}{cmd:. metandi tp fp fn tn, plot} {p 4 8 2}{cmd:. metandi a b c d, nobivariate nohsroc plot} {p 4 8 2}{cmd:. metandi} (Replays default output) {p 4 8 2}{cmd:. metandi a b c d if test==2, detail} {marker s_references}{title:References} {p 4 8 2}{marker arends} Arends LR (2006). Multivariate meta-analysis: modelling the heterogeneity. Doctoral Thesis, Erasmus University Rotterdam. {browse "http://hdl.handle.net/1765/7845":http://hdl.handle.net/1765/7845} {p 4 8 2}{marker chu} Chu H, Cole SR (2006). Bivariate meta-analysis of sensitivity and specificity with sparse data: a generalized linear mixed model approach. Journal of Clinical Epidemiology 59:1331-1332. {browse "http://dx.doi.org/10.1016/j.jclinepi.2006.06.011":doi:10.1016/j.jclinepi.2006.06.011} {p 4 8 2}{marker harbord} Harbord RM, Deeks JJ, Egger M, Whiting P, Sterne JA (2007). A unification of models for meta-analysis of diagnostic accuracy studies. {it:Biostatistics} 8:239-251. {browse "http://dx.doi.org/10.1093/biostatistics/kxl004":doi:10.1093/biostatistics/kxl004} {p 4 8 2}{marker macaskill} Macaskill P (2004). Empirical Bayes estimates generated in a hierarchical summary ROC analysis agreed closely with those of a full Bayesian analysis. Journal of Clinical Epidemiology 57:925-932. {browse "http://dx.doi.org/10.1016/j.jclinepi.2003.12.019":doi:10.1016/j.jclinepi.2003.12.019} {p 4 8 2}{marker rhs} Rabe-Hesketh S, Skrondal A (2005). {it:Multilevel and Longitudinal Modeling Using Stata.} College Station, TX: Stata Press. {browse "http://www.stata.com/bookstore/mlmus.html":http://www.stata.com/bookstore/mlmus.html} {p 4 8 2}{marker gllammmanual} Rabe-Hesketh S, Skrondal A, Pickles A (2004). GLLAMM Manual. {it:U.C. Berkeley Division of Biostatistics Working Paper Series.} Working Paper 160. {browse "http://www.bepress.com/ucbbiostat/paper160":http://www.bepress.com/ucbbiostat/paper160} {p 4 8 2}{marker rhsp} Rabe-Hesketh S, Skrondal A, Pickles A (2005). Maximum likelihood estimation of limited and discrete dependent variable models with nested random effects. {it:Journal of Econometrics} 128:301-323. {browse "http://dx.doi.org/10.1016/j.jeconom.2004.08.017":doi:10.1016/j.jeconom.2004.08.017} {p 4 8 2}{marker reitsma} Reitsma JB, Glas AS, Rutjes AWS, Scholten RJPM, Bossuyt PM, Zwinderman AH (2005). Bivariate analysis of sensitivity and specificity produces informative summary measures in diagnostic reviews. {it:Journal of Clinical Epidemiology} 58:982-990. {browse "http://dx.doi.org/10.1016/j.jclinepi.2005.02.022":doi:10.1016/j.jclinepi.2005.02.022} {p 4 8 2}{marker riley} Riley RD, Abrams KR, Sutton AJ, Lambert PC, Thompson JR (2007). Bivariate random-effects meta-analysis and the estimation of between-study correlation. {it:BMC Medical Research Methodology} 7:3. {browse "http://dx.doi.org/10.1186/1471-2288-7-3":doi:10.1186/1471-2288-7-3} {p 4 8 2}{marker rutter} Rutter CM, Gatsonis CA (2001). A hierarchical regression approach to meta-analysis of diagnostic test accuracy evaluations. {it:Statistics in Medicine} 20:2865-2884. {browse "http://dx.doi.org/10.1002/sim.942":doi:10.1002/sim.942} {marker s_acknowledgements}{title:Acknowledgements} {p 4 4 2} Joseph Coveney worked out how to fit the bivariate model using {cmd:gllamm} and posted the syntax on Statalist. I also thank him for generous and helpful email correspondence. {cmd:metandi} is essentially a wrapper for {cmd:gllamm} and I thank the authors of {cmd:gllamm} for their work, and Sophia-Rabe-Hesketh in particular for helpful email correspondence. Thanks also to Susan Mallett for pointing out a bug in a previous version. {marker s_author}{title:Author} {p 4 4 2} Roger Harbord, Department of Social Medicine, University of Bristol, UK. {browse "mailto:roger.harbord@bristol.ac.uk":roger.harbord@bristol.ac.uk} {marker s_citation}{title:Citation of metandi} {* based on that in help ivreg2}{...} {p 4 4 2} {cmd:metandi} is not an official Stata command. It is a free contribution to the research community, like a paper. Please cite it as such: {p 4 8 2} Harbord, RM (2008). metandi: Stata module for meta-analysis of diagnostic accuracy. Statistical Software Components, Boston College Department of Economics. Revised 15 Apr 2008. {p 4 4 2} {cmd:metandi} is based on the ideas and relies on the results in {help metandi##harbord:Harbord et al. (2006)}, which should therefore be cited as well. {marker s_also_see}{title:Also see} {p 4 13 2} Online: {helpb metandiplot}, {helpb xtmelogit} (in Stata 10 or above), {helpb gllamm} (if installed) {helpb metan} (if installed).