Meta-analysis is a commonly used statistical approach for evidence synthesis by integrating results from independent studies and is considered to play an essential role in evidence-based medicine. Most applied implementations of meta-analysis are conducted under the Frequentist paradigm. However, it is often the case that using a Bayesian approach can be beneficial in the context of meta-analysis. The main advantages are that Bayesian methods allow to include a formal representation of prior belief in the model and uncertainties related to both parameters and model can be better accounted for. The

The observed data are uploaded at the 'Load data and model selection' tab. Once the user specifies a desired model and uploads the spreadsheet, the analysis will be automatically run.

An example of such a file can be obtained here. After specifying the desired meta-analytic model in the 'Load data and model selection' tab, this spreadsheet can be uploaded and the model will run bmeta in the background and create all the relevant output summaries. The output table will appear in the right panel and users can further specify the plots to visualise the outcomes. In addition, some diagnostic plots are available to assess the model fit and performance by clicking the 'Diagnostics' tab. The results of the Bayesian meta-analystic models performed using

Copyright: Tao Ding, Gianluca Baio

After specifying all these options, users can then upload the excel spreadsheet and bmeta will run in the background to perform the analysis.

Download model file

Download data (R format)

The tabs below are options for forest plot. Users can firstly select the scale of plot (only available for
binary and count data), for example, whether the study-specific estimates and summary estimate need to be
displayed on a log scale or a natural scale. The user can then specify the title of the plot and the label
of the x-axis. Users also have the full control over the size and color of the 'box' and the 'diamond',
representing the study-specific and summary estimates, respectively. The color of the line, each representing
the 95% credible interval of a study-specific estimate, can be changed easily as well. The lower and upper
limits for clipping credible intervals to arrows can be specified to make the plot look more decent. Notice
that the 'Two-line' plot option **only** works for random-effects models, which can be used to compare the differences between estimates obtained from
modelling and no-pooling effect models. When the two-line forest plot is selected, the user can adjust the line
margins.

The tabs below are options for posterior distribution plot of the summary estimate. Users can select the
scale to display the summary estimate (whether on log or natural scale) and then specify the title of the
plot and the lable of the x-axis. The range of the x-axis can also be changed easily.

The funnel plot examines the publication bias and it is a scatter plot of study effect
against some measure of study size. Here, the study effect on log scale against the
standard error of the study size is displayed. The tabs below are options for this
plot.

This posterior plot shows the between-study standard deviation and can be used to examine
the heterogeneity among all the studies. Notice that this plot is
**only**
available when a random-effects model is selected.

The diagnostic plot shows either the Gelman-Rubin statistic or the number of effective
sample size for all the parameters in the model.

The traceplot shows the sampled values versus the simulation index and is considered to
be effective for assessing Markov Chain convergence for model parameters. Users need to
specify the parameters for assessment.

The autocorrelation is often used to check dependency among Markov Chain samples as a
critical property of MCMC simulation is that the distribution of the current observation
always depends on that of the previous one. Users need to specify the parameters for
assessnent.

NB: generating the document can take some time.