 Welcome to this tutorial on our package meta. I'm Gito Schwarzer. I'm the author of the package. The R script and the data set used can be found on Cnodo. There are two versions of the package, the official crown version and the development version on GitHub, where you can find bug fixes and so on. Let's start by looking at the help page of the package. So this provides a brief overview of the methods and some general hints on the package. So meta is a general package for meta analysis. And it also supports our book, Meta Analysis with R. For each outcome, there is a dedicated R function. For example, for binary outcomes, there's the meta BIM function. Or if you would like to do a meta analysis of correlations, then there is the meta core function. The standard plots are implemented for this final plot and some others as well. Some more advanced methods like free level models or generalized linear mixed models are also available. And these methods actually come from R package Meta4, which is used internally not only to estimate these methods, but also to conduct meta regression and also to estimate the between study variance. So let's have a closer look at the meta BIM function. Here you can see there is a very long list of arguments for this function. But the really essential are only the first four. So what we have to provide is the number of events in the experimental group as well as the sample size in this group and the same for the control group. All other arguments are optional. However, typically you would also use a data argument because then you can directly use here the names of the variables from the data set. OK, so let's have a look at an example. This is a meta analysis comparing Hallow Perry Dole and placebo in Schizophrenia. And here the outcome of interest is clinical improvement, which is a binary outcome. Let's have a look, load the data and have a look at some studies. So as you can see here, we have for each group three variables. So we have the number of responders, so patients with a clinical improvement. We have the number of failures in the group, H4 Hallow Perry Dole, P4 placebo. And we also have information on dropouts. So what we see here is that the number of dropouts is quite different in these studies. So the last three studies do not have dropouts at all. And especially this PSLA study has a large number of dropouts in the two groups. We only use this information here to do a subgroup analysis. But there are also other methods I will talk about later, just very briefly, that could be used in such a setting here in order to impute the missing information. So what we do here is we just add a new variable miss, which has information on whether there are dropouts or not. And as I said, this will be used in a subgroup analysis. Here is then now the meta bin command for the meta analysis with the binary outcome. The first two arguments are for the Hallow Perry Dole group. So a number of clinical improvements and total sample size. Dropouts are ignored here in this analysis. And the same here for the placebo group. The study labels here are just the combination of the first author name and the year of publication. If we run this analysis and print it out, we get the standard layout for the meta package, which is basically the same for any outcome type. First, we have information here on the number of studies that are combined here, 17. And then if this information is available on the number of observations and the number of events in total in these 17 studies, by default we get results for the common effect and the random effects model, as well as information on tau squared, i squared, and so on. And here at the bottom, we see that by default, the Mantle-Hensel method is used under the common effect model and the restricted maximum likelihood estimator is used to estimate tau squared in the random effects model. These are the default settings, but as I will show you later on, this can be changed quite easily by the user. Then we have here the summary command. The summary command is a more detailed print out. So here we see at the bottom the same print out as for the print function. But here for the summary function, we also get results for the individual studies. So we see here the estimated risk ratios, the weights under the common, and the random effects model. Typically, we would like to produce a forest plot for our data, and this can be done with the forest function. The first argument is the meta-analysis object. And here we use three more arguments with sort while we say we would like to sort the studies by the year of publication. And with label left and label right, we give some informative information here at the bottom of the forest plot, as you will see. When we produce the plot, so this is here the standard layout of this plot, which could be changed in quite many ways with the isn't the forest function. Again, we see here the studies here sorted by the year of publication and the basic information that we have already seen in the data set and so on. And at the bottom here, results for the common and the random effects model. Typically, what I do then next is what I do is to save forest plots that I would like to use in publication or to discuss with clinicians about. I save them as a PDF file. And here you have to tweak this or use a slightly different command than just the PDF command, because if we use the default settings, we get a PDF file with dimensions 7 by 7 inch. And this would then look like this. So here we see that on the right and left side information is missing. So what one has to do is one has to provide the width and the height of the PDF file. And if we do that here, then we see that this is then a nicely formatted PDF file that we could use either in a publication or to discuss. And what I do here is the only change is that I use here a different layout. So here I use the layout of Revment 5 for the forest plot and slightly wider PDF file. And this would then look like this here. And the same could be done also with SVG, for example, or PostScript, or JPEG, or PNG files. So there are all you would do here in a similar way. To conduct subgroup analysis, this is quite easy to do with the meta package. The only thing you have to do is you have to provide here the argument subgroup with the information on which variable you would like to use for the subgroup analysis. And what I also do here is I say I'm not interested in the common effect model. I just would like to see the results for the random effects model. And I use the update command. So I update my meta-analysis object. I do not show the common effect results. And I conduct sub-analysis. And then the print out here for this subgroup analysis would look like this. Again, at the top, we get the same results as before. And here are the results for our subgroups. Let me rerun this command of the print out here, OK, so that we get all results here in one line. What we can see is here we have 10 studies with missing data and seven studies of the 17 without missing data. They have quite different values here. So we have here a significant difference between the two groups of studies, which one then would, in an additional step, would like to further evaluate, especially as we have here the information on how many missing observations there are. And one possibility to do that would be to use the meta-miss function from the R-package meta sense. But here we do not talk about this in more detail. And as you can also see here, by default, we allow for a different tau squared estimate in the two subgroups. There is an argument tau common we could use. And then we would get a result where we assume a common tau squared value in the two groups. Some more information, for example, on the summary measures, are available on one help side. So here, for example, for the meta-bin function, we see that the available summary measures are observation risk ratio, risk difference, arc sign difference, and also here would be either vex sign efficacy or effectiveness. For continuous outcomes, the available summary measures are the mean difference, standardized mean difference, and the ratio of means, and so on here for the other methods. And also, let's now look here at the list elements of a meta-analysis object, because sometimes you are interested to extract some information from a meta object, and then it's important to know which information is available where. And this here is a rather long listing of all the information that is available in any meta-analysis object created with meta, starting with the study labels and so on. And then, for example, the level of confidence intervals for individual studies, lower upper confidence limits, and so on. So you could extract any of this information if you are interested in it. And for level, for example, if we go here back to the meta-bin function, we see here that by default, the value of GS level is used here as default. And this actually is here the 95% confidence level. And the confidence level for a meta-analysis estimate is saved here in level MA, which is also 95%. And all these two settings, as well as several other settings that are available in the meta package, can be printed with this command here. And this is really a very long list. I will not talk in detail about this, but just briefly show you here some of the more general arguments I already talked about, the first two here. Here is an argument for the common and the random effects model. So you could define whether you would like to use only the random effects, only the common effects or both. Default is both. Some information here on confidence intervals under the random effects model. The default estimate of what are squared is RML. This could also be changed, the question whether prediction intervals should be calculated and shown, and so on. And one thing you may have noticed or not in the printout is that I printed the risk ratios with four digits, which is quite a lot for a real risk ratio. So with this command here, we can define that estimates and confidence limits should be shown only with two digits. And if I now print again my meta-analysis object, I will see that in the printout, we only have here these two digits instead of four. What I could also do is I can change the layout of confidence intervals with this command here, and then, for example, it would look like this. And there are several other possibilities with the settings meta-function. And if you look here at this long printout and then go to the corresponding function, for example, for the forest meta-function, then you can see what is the actual meaning of the argument here. So that's it from me with a brief introduction to the meta-package. Thank you for your attention.