For comments or questions on MetaInsight please e-mail .


1 Introduction


This guide contains step-by-step instructions to reproduce the example meta-analyses in the NICE Evidence Synthesis Technical Support Documents (https://www.sheffield.ac.uk/nice-dsu/tsds/evidence-synthesis), using the MetaInsight app (https://crsu.shinyapps.io/MetaInsight/). Six of the examples contained in TSD2 and TSD3 can be carried out in MetaInsight, though three of these are pairwise meta-analyses and are not included in this guide. Before starting, the reader should familiarise themselves with the example(s) they wish to reproduce and download the accompanying WinBUGS files.

There are eight examples in TSD2 and six in TSD3, of which there is one network meta-analysis in TSD2 and two network meta-regressions in TSD3 that can be reproduced in MetaInsight. These are:


TSD2 Examples

Example Model In MetaInsight
TSD2 Example 1, Blocker Binomial, logit (pairwise) Yes (not shown)
TSD2 Example 2, Dietary fat Binomial, cloglog No
TSD2 Example 3, Diabetes Poisson, log No
TSD2 Example 4, Schizophrenia Multinomial, log No
TSD2 Example 5 Parkinson’s Normal, identity Yes
TSD2 Example 6 Psoriasis Conditional Binomial, probit No
TSD2 Example 7 Parkinson’s difference Normal, identity, contrasts No
TSD2 Example 8 Parkinson’s shared Normal, identity, shared No

TSD3 Examples

Example Model In MetaInsight
TSD3 Example 1, Magnesium Binomial, logit, cross-validation (pairwise) No
TSD3 Example 2, Chemotherapy Binomial, logit, cross-validation No
TSD3 Example 3, Statins Binomial, logit, binary meta-regression (pairwise) Yes (not shown)
TSD3 Example 4, BCG vaccine Binomial, logit, continuous meta-regression (pairwise) Yes (not shown)
TSD3 Example 5, Certolizumab Binomial, logit, continuous meta-regression Yes
TSD3 Example 6, Certolizumab Binomial, logit, baseline risk Yes


2 TSD2 Example 5, Parkinsons


2.1 Create the dataset


The WinBUGS files are TSD2-5aRE_Normal_id.odc and TSD2-5bFE_Normal_id.odc. The data in the two files are the same, so only one needs to be opened. The file can be opened in WinBUGS (not OpenBUGS) or another text editor. Copy the data as shown below.



Paste the data into a spreadsheet, possibly using a paste special to maintain the column and row structure.



The data format corresponds to Wide format in MetaInsight.

Rename the first six columns to T.1, T.2, T.3, Mean.1, Mean.2, and Mean.3 respectively, delete the na[ ] column, and add a column called Study with distinct entries that consist of letters, numbers and underscores only. The Study and T columns should not consist entirely of numbers, so change the treatments to contain letters, for example change “1” to “Treat1”.

Unfortunately MetaInsight requires standard deviations and numbers of patients rather than standard errors, so the three se[ ] columns should be deleted and replaced with six columns named SD.1, SD.2, SD.3, N.1, N.2, and N.3. The standard devations and numbers of patients can be typed in manually from Table 8A on page 79 of TSD2-General-meta-analysis-corrected-2Sep2016v2.pdf, shown here:


The final dataset should look like this.



Depending on the type of spreadsheet used, the data may now be ready to upload to MetaInsight. If a problem occurs then saving the file as a CSV should fix it.


2.2 Analyse the data


Open MetaInsight and navigate to the Load Data tab.



Continuous outcome type should already be selected.



Upload the dataset, check that Treat1 is selected as the reference treatment, and view the data by selecting the View Data tab.



Navigate to the Data Analysis tab.



Select the required model options. This example uses Mean Difference, smaller outcome values are Desirable, and there is output from both a fixed effects and a random effects model, so either of these can be chosen.



Within the Data Analysis tab navigate to the Bayesian network meta-analysis tab.



Click the Click here to run the main analysis for all studies button.



2.3 Compare the results


The results from WinBUGS are shown in Table A9 on page 82 of the TSD2 pdf file:



The relative treatment effects \(d_{12}\), …, \(d_{15}\) and their 95% credible intervals are displayed in the forest plot in MetaInsight. The one below is from the random effects model.



The medians and standard deviations of the relative treatment effects are sometimes available in the Bayesian result details tab, as they are in this example. The underlying R package that creates the model, {gemtc}, sometimes parameterises the model in such a way that these statistics are not available for all relative treatment effects.



The absolute treatment effects \(T_1\), …, \(T_5\) are not provided by MetaInsight.

\(\overline{D}_{res}\), \(p_D\), and \(DIC\) are all provided underneath the forest plot. The mean of \(\tau\) and its 95% credible interval can also be found there for the random effects model. The median and standard deviation of \(\tau\) can be found in the Bayesian result details tab in the sd.d rows (above).



3 TSD3 Example 5, Certolizumab: continuous covariate


3.1 Create the dataset


The WinBUGS files are TSD3-4aRE_continuousCZP.odc and TSD3-4bFE_continuousCZP. The data in the two files are the same, so only one needs to be opened. The file can be opened in WinBUGS (not OpenBUGS) or another text editor. Copy the data as shown below.



Paste the data into a spreadsheet, possibly using a paste special to maintain the column and row structure.



The data format corresponds to Wide format in MetaInsight.

Delete the na[], # and ID columns and rename the remaining columns respectively to T.1, T.2, N.1, N.2, R.1, R.2, covar.x, and Study. Add letters to the T columns (e.g. change “1” to “Treat1”) and delete any characters from Study that are not letters, numbers or underscores.


The final dataset should look like this.



Depending on the type of spreadsheet used, the data may now be ready to upload to MetaInsight. If a problem occurs then saving the file as a CSV should fix it.


3.2 Analyse the data


Open MetaInsight, navigate to the Load Data tab and select Binary outcome type.



Upload the dataset, check that Treat1 is selected as the reference treatment, and view the data by selecting the View Data tab.



Navigate to the Data Analysis tab.



Select the required model options. This example uses Odds Ratio, smaller outcome values are Undesirable, and there is output from both a fixed effects and a random effects model, so either of these can be chosen.



Within the Data Analysis tab navigate to the Meta-regression tab and then the Covariate Analysis tab.



Leave the regression coefficient at the default shared and click the Click here to run the main analysis for all studies button.



3.3 Compare the results


The results from WinBUGS are shown in Table 7 on page 41 of the TSD3 pdf file:



The relative treatment effects \(d_{12}\), …, \(d_{17}\) and their 95% credible intervals (on the odds ratio scale) are displayed in the forest plot in MetaInsight. The one below is from the random effects model.



The medians and standard deviations of the relative treatment effects are in the Bayesian result details tab.

Also in this tab are statistics on the shared regression parameter \(b\). This parameter is not the same as the parameter \(B\) in MetaInsight, and so their statistics cannot be directly compared. In MetaInsight the covariate values are scaled to have standard deviation 0.5, whereas in WinBUGS they are not. The scaling factor in this example is approximately 6.5, and it can be seen that the medians differ by approximately this factor.



\(resdev\) (\(Dbar\) in MetaInsight) , \(p_D\), and \(DIC\) are all provided underneath the forest plot. The mean of \(\sigma\) and its 95% credible interval can also be found there for the random effects model. The median and standard deviation of \(\sigma\) can be found in the Bayesian result details tab in the sd.d rows (above).



A plot showing the regression lines is on page 39 of TSD3:



This is the equivalent plot in MetaInsight, which also displays study contributions and confidence regions:



4 TSD3 Example 5, Certolizumab: baseline risk


4.1 Create the dataset


Follow the steps in section 3.1.


4.2 Analyse the data


Follow the steps in section 3.2 up to the stage of selecting the Meta-regression tab. Within the Meta-regression tab, select the Baseline Risk Analysis tab. Leave the type of regression coefficient at the default shared and click the Click here to run the main analysis for all studies button.



4.3 Compare the results


The results from WinBUGS are shown in Table 8 on page 45 of the TSD3 pdf file:



The relative treatment effects \(d_{12}\), …, \(d_{17}\) and their 95% credible intervals (on the odds ratio scale) are displayed in the forest plot in MetaInsight. The one below is from the random effects model.



The medians and standard deviations of the relative treatment effects are available in the Bayesian result details tab, which also displays statistics on the shared covariate parameter \(b\). Note that in contrast to the covariate meta-regression there is no scaling in MetaInsight in the baseline risk model, so this parameter is the same as the MetaInsight parameter \(b\_bl\).



\(resdev\) (\(Dbar\) in MetaInsight), \(p_D\), and \(DIC\) are all provided underneath the forest plot. The mean of \(\sigma\) and its 95% credible interval can also be found there for the random effects model. The median and standard deviation of \(\sigma\) can be found in the Bayesian result details tab in the sd rows (above).



A plot showing the regression lines is on page 39 of TSD3:



This is the equivalent plot in MetaInsight, which also displays study contributions and confidence regions: