 My presentation is about network materializes with net metter. I would like to start with an example with a network of 14 treatments for acute mania. There are a total of 47 studies. Most of them are two arm studies comparing two of these 14 treatments. However, there are also 14 three arm studies comparing three of these 14 treatments. And here you can see the network graph for this network. And I do not have much time to talk about it. And so I would like to focus on the comparison of Ari Krippard-Soll worth as lithium. So you can see here there's just one study comparing it directly. But there are six studies comparing Ari Krippard-Soll with placebo and three studies comparing lithium with placebo. So the idea is in a network materializes that we can use this indirect information in order to get a more precise estimate of the comparison Ari Krippard-Soll was lithium. Yes, and this is the first aim typically of a network materializes. A second aim is often that we would like to rank treatments according to their efficacy. The assumptions that we have to make here is that studies are independent of each other and that the underlying effects are consistent. This means that for any two treatments that we look at that the treatment difference of the direct and the indirect evidence does not differ substantially. So the challenges that we have here are multi arm studies, as I've already shown you for the example, a problem of indirect comparisons and potentially inconsistency between direct and indirect evidence. Our package net meta is one possibility to conduct a network materializes and it's available on CRAN for almost eight years now. It's based on a frequentist method described in the paper by Gerta. And what's special about our implementation is that it's the handling of multi arm studies. So in net meta all pairwise comparisons of a multi arm study are included in the network materializes but they are weighted appropriately by inflating standard errors. This differs from the standard implementation which typically compares the treatments in a multi arm study with one reference treatment and only include those comparisons in the model. So here all pairwise comparisons are included and over the last six or seven years we implemented several methodological extensions and published their corresponding papers and more information on that is provided on the next slides. Regarding download statistics, you can see that net meta is among the most popular R packages for NMA and of these three R packages net meta is the only frequentist method and it's utilized in more than 100 publications so far. So here on the next two slides I will show you the most important functions from our package net meta. The first is pairwise which can be used to transform data sets into a format that is used in the network materializes functions and it also calculates the treatment estimates for all pairwise comparisons and the network materializes models are then actually fitted using net meta which is the standard frequentist NMA method net meta bin for rare binary outcomes and net comb and diskom for component NMA for connected or disconnected networks. The results can then be shown as network graphs, forest plots or in league tables. Treatment agreements can be ranked using net rank which implements peace cores which are a frequentist analog to sucra values which are used in a Bayesian setting and if you have more than one outcome you can order treatments using net poset which implements partial order of treatment rankings and the Hasse diagram. Evaluation of consistency can be done using net split which splits the network estimates into direct and indirect evidence and looks whether they are similar or not. The net heat plot is available and also a design-based decomposition of Cochrane's Q and finally comparison adjusted funnel plots are implemented in the test for funnel plot asymmetry. Here is the R code for the Acutemania network. As you can see here first we use pairwise in order to transform the dataset and to calculate pairwise odds ratio so here the summary measure is the odds ratio as we have binary outcomes we use here the events so the number of responders and the treatment sample sizes in the arms and we have also to provide the information on the treatments and the studies and then we conduct random effects network meta-analysis using net meta. Here we say we are not interested in the fixed effects results which would otherwise be also shown and we say here that the reference treatment should be placebo again this this will be then shown in the printouts and in the plots accordingly. Finally here we say that small values are bad because we have here responses we would like to see many responses by good fiends and so small values are bad here and this is the uses information in the network ranking. Here is the command for the network graph I've shown you before for the forest plot for the leak table and so on as you can see here in most of these commands you either just use this network meta-analysis object or some some additional arguments and I do not show you all these results just slightly extended or forest plot with more information so here you see all the network estimates comparing these treatments with placebo and also the number of studies providing a comparison of these treatments with placebo and the peace cost for the ranking and this forest plot is sorted by decreasing p-score and so the best treatments are here at the top and the worst at the bottom and what you can see here quite nicely is that most of the treatments are superior to placebo with regard to the response rate okay so this is what you can do with net meta at the moment what is planned for the future the first point is that the moment only the decimony layered estimator is implemented for the between study heterogeneity variants and our idea is to link net meta to metaphor and use rma mv to get remel and ml estimators for for the between study heterogeneity the problem here is that multi arm studies are handled differently in metaphor than in net meta if we can sort this out we think we can also provide methods for network meta regression and subgroup analysis again via a link to rma mv we would like to use resampling methods to describe uncertainty in treatment rankings the next point net contrary contribution of studies to network materializes is another presentation in this session by todores and finally we would like to link net meta to an r package for bayesian network materializes so our idea is that it's possible using an argument method that one can say that one is interested in the bayesian network materializes then afterwards one can use net uh two and all to print and plot the results and that's it from my side thank you for your attention