MetaInsight

MetaInsight v6.0.1

Selection of continuous or binary outcomes has moved to the Load Data tab

Latest Updates:

Patch (12th July 2024 v6.0.1):

  • Hot fix for colouring in the regression plot.

Major update (10 July 2024 v6.0.0):

  • Meta-regression has been added. One covariate is allowed, which can be a new continuous or binary variable, or baseline risk. Two new graphs are available for meta-regression. The first displays the covariate values grouped by treatment and study. The second plots the covariate against relative treatment effects, with confidence regions and study-level contributions.

Patch (16th April 2024 v5.2.1):

  • Troubleshooting tab now links to our troubleshooting wiki page.
  • Added logo to home page.

The full update history for MetaInsight is available on GitHub

The code for MetaInsight is available on GitHub



Naomi Bradbury, Ryan Field, Tom Morris, Clareece Nevill, Janion Nevill, Yiqiao Xin, Rhiannon K Owen, Nicola Cooper, and Alex Sutton

For feedback/questions about this app please email the CRSU team at apps@crsu.org.uk. If you encounter any errors with using the app, please check the trouble shooting page first before contacting us.


If you use the app please cite it as:

Owen, RK, Bradbury, N, Xin, Y, Cooper, N, Sutton, A. MetaInsight: An interactive web-based tool for analyzing, interrogating, and visualizing network meta-analyses using R-shiny and netmeta. Res Syn Meth. 2019; 10: 569-581.


App powered by R and Shiny. All frequentist statistical calculations are performed using R package netmeta (Gerta Rücker, Guido Schwarzer, Ulrike Krahn and Jochem König 2017). netmeta: Network Meta-Analysis using Frequentist Methods. R package version 0.9-8. All Bayesian statistical calculations are performed using R package gemtc (Gert van Valkenhoef, Joel Kuiper 2016) gemtc: Network Meta-Analysis Using Bayesian Methods R package version 0.8-2. and R package BUGSNET (Audrey Beliveau, Devon J. Boyne, Justin Slater, Darren Brenner & Paul Arora) BUGSnet: Bayesian inference Using Gibbs Sampling to conduct NETwork meta-analysis version 1.0.3.

For users wishing to analyse large treatment networks or fit complex network meta-analysis models, please seek advice from technical experts.


THE SOFTWARE IS PROVIDED AS IS, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

Funding and Support Acknowledgement:

MetaInsight is part of the Complex Reviews Synthesis Unit (CRSU) suite of evidence synthesis apps. The development of these apps are currently funded (majority) and overseen by the Evidence Synthesis Group @ CRSU (NIHR153934). Further details of other funders and support, current and past, can be found on our GitHub page . The views expressed are those of the author(s) and not necessarily those of the NIHR or the Department of Health and Social Care.

More information about the UK NIHR Complex Reviews Synthesis Unit (CRSU) can be found on our website

Instructions for uploading long format data


MetaInsight allows data in either long format, or wide format. This tab provides instructions for long format data, where each row contains one treatment arm. Instructions are as below. Please note that MetaInsight is not compatible with studies containing multiple arms of the same treatment.

Step 1:

The long format data file should contain the following columns:

  • Study contains the name (e.g., author,year) of the study. The study name must be unique for each study.
  • T contains the name of the treatment used in each arm of the study. Treatment names should only contain letters, numbers and underscores.
  • N contains the number of participants in each arm of the study.
  • Mean contains the mean value of the outcome in each arm of the study.
  • SD contains the standard deviation of the outcome in each arm of the study.

The long format data file should contain the following columns:

  • Study contains the name (e.g., author,year) of the study. The study name must be unique for each study.
  • T contains the name of the treatment used in each arm of the study. Treatment names should only contain letters, numbers and underscores.
  • R contains the number of participants with the outcome of interest in each arm of the study.
  • N contains the number of participants in each arm of the study.

N.B. Continuity corrections will need to be applied to cells containing 0 values

The long format data file may also contain the following column:

  • covar.<COVARIATE_NAME> contains the study-level covariate value, where <COVARIATE_NAME> is replaced by the name of the covariate. This must be identical for each arm of the study. The name of the covariate will be extracted and used in the analysis output.

The maximum number of arms for each trial allowed in the MetaInsight app is 6.

An example of this structure can be seen in the 'View Data' tab.

The csv file that is used to produce the example dataset can be downloaded from here:

Download the example dataset in long format

Step 2:

Select the reference treatment in the drop-down box. By default, MetaInsight will select a treatment which matches common names for a standard reference treatment.


This default dataset for continuous outcome data is a reduced version of the data from Brett Doleman, Ole Mathiesen, Alex J Sutton, Nicola J Cooper, Jon N Lund, John P Williams (2023), Non-opioid analgesics for the prevention of chronic postsurgical pain: a systematic review and network meta-analysis Br J Anaesth 2023 Jun;130(6):719-728. doi: 10.1016/j.bja.2023.02.041. The outcome is pain on a scale of 0 to 10 and the covariate is the mean age of the participants.

This default dataset for binary outcome data is from S. Dias, A.J. Sutton, N.J. Welton, and A.E. Ades (2013b), Heterogeneity - Subgroups, Meta-Regression, Bias, and Bias-Adjustment , Medical Decision Making 33(5):618-640. The outcome is ACR-50, a reduction of at least 50% in the American College of Rhematology score, and the covariate is the mean disease duration of the participants.


Note: The default dataset, pre-loaded on the 'View Data' tab, will be used for analysis if no file is selected. The 'View Data' tab will automatically update once a file is successfully loaded.

Instructions for uploading wide format data


MetaInsight allows data in either long format, or wide format. This tab provides instructions for wide format data, where each row contains all the treatment arms from one study. Instructions are as below. Please note that MetaInsight is not compatible with studies containing multiple arms of the same treatment.

Step 1:

The wide format data file should contain the following columns:

  • Study contains name (e.g., author,year) of the study. The study name must be unique for each study.

  • T.1, T.2, ..., up to T.6 contains name of the treatment given for study arm 1, 2, ..., up to 6, respectively. Treatment names should only contain letters, numbers and underscores.

  • N.1, N.2, ..., up to N.6 contains number of participants in study arm 1, 2, ..., up to 6, respectively

  • Mean.1, Mean.2, ..., up to Mean.6 contains the mean value of the outcome in study arm 1, 2, ..., up to 6, respectively

  • SD.1, SD.2, ..., up to SD.6 contains standard deviation of the outcome in study arm 1, 2, ..., up to 6, respectively

  • R.1, R.2, ..., up to R.6 contains number of participants with the outcome of interest in study arm 1, 2, ..., up to 6, respectively

  • N.1, N.2, ..., up to N.6 contains number of participants in study arm 1, 2, ..., up to 6, respectively

The wide format data file may also contain the following column:

  • covar.<COVARIATE_NAME> contains the study-level covariate value, where <COVARIATE_NAME> is replaced by the name of the covariate. The name of the covariate will be extracted and used in the analysis output.

The maximum number of arms for each trial allowed in the MetaInsight app is 6.

An example of this structure can be seen in the 'View Data' tab.

The csv file that is used to produce the example dataset can be downloaded from here:

Download the example dataset in wide format

Step 2:

Select the reference treatment in the drop-down box. By default, MetaInsight will select a treatment which matches common names for a standard reference treatment.


This default dataset for continuous outcome data is a reduced version of the data from Brett Doleman, Ole Mathiesen, Alex J Sutton, Nicola J Cooper, Jon N Lund, John P Williams (2023), Non-opioid analgesics for the prevention of chronic postsurgical pain: a systematic review and network meta-analysis Br J Anaesth 2023 Jun;130(6):719-728. doi: 10.1016/j.bja.2023.02.041. The outcome is pain on a scale of 0 to 10 and the covariate is the mean age of the participants.

This default dataset for binary outcome data is from S. Dias, A.J. Sutton, N.J. Welton, and A.E. Ades (2013b), Heterogeneity - Subgroups, Meta-Regression, Bias, and Bias-Adjustment , Medical Decision Making 33(5):618-640. The outcome is ACR-50, a reduction of at least 50% in the American College of Rhematology score, and the covariate is the mean disease duration of the participants.


Note: The default dataset, pre-loaded on the 'View Data' tab, will be used for analysis if no file is selected. The 'View Data' tab will automatically update once a file is successfully loaded.

Users can use the filter box under each column of heading to select studies to exclude in the sensitivity analysis.

This tab shows a summary of study characteristics.

Characteristics table of all studies

Characteristics table with selected studies excluded

If the formatting of the text for this plot needs adjusting please see options at the bottom.

If the formatting of the text in the above plot needs adjusting (for on screen or download) please use the following options:
The download text size alters the text sizing for all elements of the plot and is integer point sizes. The title and group header options are an additional text sizing option in terms of percentage (e.g. 0.5 indicates half the specified size).

Please note: the formatting is not guaranteed to be identical between what is shown on screen and what is downloaded.

Download

Numbers on the line indicate the number of trials conducted for the comparison. The shaded areas indicate there exist multi-arm trials between the comparisons.

The size of the nodes and thickness of edges represent the number of studies that examined a treatment and compared two given treatments respectively.

Download



                                    

Numbers on the line indicate the number of trials conducted for the comparison. The shaded areas indicate there exist multi-arm trials between the comparisons.

The size of the nodes and thickness of edges represent the number of studies that examined a treatment and compared two given treatments respectively.

Download



                                    

Options to change limits of the x-axis:

Download

Options to change limits of the x-axis:


Download
Treatments are ranked from best to worst along the leading diagonal. Above the leading diagonal are estimates from pairwise meta-analyses, below the leading diagonal are estimates from network meta-analyses Relative treatment effects in ranked order for all studies
Download Relative treatment effects in ranked order with selected studies excluded
Download
Assessment of inconsistency for all studies
Download Assessment of inconsistency with selected studies excluded
Download
Baysesian result using the gemtc package.
Heterogeneity prior: standard deviation ~ U(0,X), where X represents a very large difference in the analysis' outcome scale and is determined from the data.
Please note the outcome for continuous data has to be mean difference for the Bayesian analysis. Standardised mean difference cannot be analysed.
Please note the outcome for binary data has to be Odds Ratio or Risk Ratio for the Bayesian analysis. Risk difference cannot be analysed. Please note each simulation may take 20 seconds.

Results for all studies

Please click the button below to run Bayesian analysis for all studies, and after each time when you change the radiobutton selections.

Results with selected studies excluded

Please click the button below to run each time after you finish the selection of studies, or change the radiobutton selections.

Loading...

Options to change limits of the x-axis:

Model fit:



Download
Loading...

Options to change limits of the x-axis:

Model fit:



Download
Please note: if you change the selections on the sidebar, you will need to re-run the primary and/or sensitivity analysis from the 'Forest Plot' page.

In contrast to the 'comparison of all treatment pairs' tab in the frequentist NMA results, this table only contains the estimates from the network meta analysis, i.e. does not contain estimates from pairwise meta-analysis which only contains direct evidence. If you would like to obtain the pairwise meta-analysis results, please run 3d. Nodesplit model


Treatment effects for all studies: comparison of all treatment pairs.

Download

Treatment effects with selected studies excluded: comparison of all treatment pairs.

Download
Please note: if you change the selections on the sidebar, you will need to re-run the primary and/or sensitivity analysis from the 'Forest Plot' page.
Please note it may take up to 5 minutes to load the results.
IMPORTANT: If you export and include the Litmus Rank-O-Gram or the Radial SUCRA plot in your work, please cite it as: Nevill CR, Cooper NJ, Sutton AJ, A multifaceted graphical display, including treatment ranking, was developed to aid interpretation of network meta-analysis, Journal of Clinical Epidemiology (2023)

Ranking panel for all studies

Relative Effects

Loading...
Download

Ranking Results

Loading...

Litmus Rank-O-Gram: Higher SUCRA (Surface Under the Cumulative Ranking Curve) values and cumulative ranking curves nearer the top left indicate better performance

Loading...
Loading...

Radial SUCRA plot: Higher SUCRA values indicate better treatments; size of nodes represent number of participants and thickness of lines indicate number of trials conducted

Summary of evidence

Loading...

Numbers on the line indicate number of trials conducted for the comparison. Any shaded areas indicate existence of multi-arm trials between the comparisons.

The size of the nodes and thickness of edges represent the number of studies that examined a treatment and compared two given treatments respectively.

Download

Ranking panel with selected studies excluded

Relative Effects

Loading...
Download

Ranking Results

Loading...

Litmus Rank-O-Gram: Higher SUCRA (Surface Under the Cumulative Ranking Curve) values and cumulative ranking curves nearer the top left indicate better performance

Loading...
Loading...

Radial SUCRA plot: Higher SUCRA values indicate better treatments; size of nodes represent number of participants and thickness of lines indicate number of trials conducted

Summary of evidence

Loading...

Numbers on the line indicate number of trials conducted for the comparison. Any shaded areas indicate existence of multi-arm trials between the comparisons.

The size of the nodes and thickness of edges represent the number of studies that examined a treatment and compared two given treatments respectively.

Download
Please note: if you change the selections on the sidebar, you will need to re-run the primary and/or sensitivity analysis from the 'Forest Plot' page.

Please note: This may take more than 10 minutes depending on the number of treatment options. The node splitting option for the Bayesian analysis is highly numerically intensive and using it on the app can cause the app to disconnect in some circumstances. We have produced a guide to running MetaInsight locally through RStudio on the user's own machine if they want to make use of this function.

Inconsistency test with notesplitting model for all studies

Download

Inconsistency test with notesplitting model with studies excluded

Download
Please note: if you change the selections on the sidebar, you will need to re-run the analysis from the 'Forest Plot' page.

Results details for all studies

Loading...

                                            

Gelman convergence assessment plot for all studies

Loading...

Results details for the sensitivity analysis

Loading...

                                            

Gelman convergence assessment plot for the sensitivity analysis

Loading...
Please note: if you change the selections on the sidebar, you will need to re-run the primary and/or sensitivity analysis from the 'Forest Plot' page.

PLEASE NOTE: the package does not currently include unrelated-mean-effects meta-regression models, therefore the consistency vs UME graph that is displayed in the deviance report tab under Bayesian network meta-analysis is not available here.

Deviance report for all studies and the sensitivity analysis

Residual deviance from NMA model and UME inconsistency model for all studies

Residual deviance from NMA model and UME inconsistency model for the sensitivity analysis

This plot represents each data point's contribution to the residual deviance for the NMA with consistency (horizontal axis) and the unrelated mean effect (ume) inconsistency models (vertical axis) along with the line of equality. The points on the equality line means there is no improvement in model fit when using the inconsistency model, suggesting that there is no evidence of inconsistency. Points above the equality line means they have a smaller residual deviance for the consistency model indicating a better fit in the NMA consistency model and points below the equality line means they have a better fit in the ume inconsistency model. Please note that the unrelated mean effects model may not handle multi-arm trials correctly. (Further reading: Dias S, Ades AE, Welton NJ, Jansen JP, Sutton AJ. Network meta-anlaysis for decision-making. Chapter 3 Model fit, model comparison and outlier detection. @2018 John Wiley & Sons Ltd.)




Per-arm residual deviance for all studies

Per-arm residual deviance for the sensitivity analysis


This stem plot represents the posterior residual deviance per study arm. The total number of stems equals the total number of data points in the network meta analysis. Going from left to right, the alternating symbols on the stems indicate the different studies. Each stem corresponds to the residual deviance ($dev.ab) associated with each arm in each study. The smaller residual deviance (the shorter stem), dev.ab, the better model fit for each data point. You can identify which stem corresponds to which study arm by hovering on the stem symbols. (Further reading: Dias S, Ades AE, Welton NJ, Jansen JP, Sutton AJ. Network meta-anlaysis for decision-making. Chapter 3 Model fit, model comparison and outlier detection. @2018 John Wiley & Sons Ltd.)




Leverage plot for all studies

Leverage plot for the sensitivity analysis


This leverage plot shows the average leverage across the arms for each study ({sum($lev.ab)}/{number of arms} for each study) versus the square root of the average residual deviance across the arms for each study (sqrt({sum($dev.ab)}/{number of arms}) for each study). The leverage for each data point, is calculated as the posterior mean of the residual deviance, minus the deviance at the posterior mean of the fitted values. The leverage plot may be used to identify influential and/or poorly fitting studies and can be used to check how each study is affecting the overall model fit and DIC. Curves of the form x2 + y = c, c = 1, 2, 3, ., where x represents square root of residual deviance, and y represents the leverage, are marked on the plot. Points lying on such parabolas each contribute an amount c to the DIC (Spiegelhalter et al., 2002). Points that lie outside the line with c = 3 can generally be identified as contributing to the model's poor fit. Points with a high leverage are influential, which means that they have a strong influence on the model parameters that generate their fitted values. (Further reading: Dias S, Ades AE, Welton NJ, Jansen JP, Sutton AJ. Network meta-anlaysis for decision-making. Chapter 3 Model fit, model comparison and outlier detection. @2018 John Wiley & Sons Ltd. Spiegelhalter et al. (2002) Bayesian measures of model complexity and fit. J. R. Statist. Soc.B 64, Part4, pp.583-639)



Please note: if you change the selections on the sidebar, you will need to re-run the primary and/or sensitivity analysis from the 'Forest Plot' page.

Model codes for analysis of all studies

Download

                                          

Deviance data for all studies

NMA (consistency) model


                                                

Deviance data for the sensitivity analysis

NMA (consistency) model


                                                

UME (inconsistency) model


                                                  

UME (inconsistency) model


                                                  

Summary Characteristic Plot

Download

Covariate value:

Baysesian covariate meta-regression using the package.
Heterogeneity prior: standard deviation
Please note each simulation may take 60 seconds.

Results for all studies

Please choose your regressor type, then click the button below to run meta-regression analysis (and each time you subsequently change any options).

Loading...
Download

This graph was adapted from Graphs of study contributions and covariate distributions for network meta-regression , Sarah Donegan, Sofia Dias, Catrin Tudur-Smith, Valeria Marinho, Nicky J Welton, Res Syn Meth , 2018; 9 :243-260. DOI: 10.1002/jrsm.1292

Loading...

Options to change limits of the x-axis:

Model fit:



Download
Please note: if you change the selections on the sidebar, you will need to re-run the analysis from the 'Forest Plot' page.

This table only contains the estimates from the network meta analysis, i.e. does not contain estimates from pairwise meta-analysis which only contains direct evidence. If you would like to obtain the pairwise meta-analysis results, please run 4c-4. Nodesplit model


Treatment effects for all studies: comparison of all treatment pairs.

Download
Please note: if you change the selections on the sidebar, you will need to re-run the primary and/or sensitivity analysis from the 'Forest Plot' page.
Please note it may take up to 5 minutes to load the results.
IMPORTANT: If you export and include the Litmus Rank-O-Gram or the Radial SUCRA plot in your work, please cite it as: Nevill CR, Cooper NJ, Sutton AJ, A multifaceted graphical display, including treatment ranking, was developed to aid interpretation of network meta-analysis, Journal of Clinical Epidemiology (2023)

Ranking panel for all studies

Relative Effects

Loading...
Download

Ranking Results

Loading...

Litmus Rank-O-Gram: Higher SUCRA (Surface Under the Cumulative Ranking Curve) values and cumulative ranking curves nearer the top left indicate better performance

Loading...
Loading...

Radial SUCRA plot: Higher SUCRA values indicate better treatments; size of nodes represent number of participants and thickness of lines indicate number of trials conducted

Summary of evidence

Loading...

Numbers on the line indicate number of trials conducted for the comparison. Any shaded areas indicate existence of multi-arm trials between the comparisons.

The size of the nodes and thickness of edges represent the number of studies that examined a treatment and compared two given treatments respectively.

Download

Due to limitations with the underlying R package {bnma} it is not currently possible to run a regression nodesplit model within MetaInsight.

When this functionality becomes available, it will be added here.

Please note: if you change the selections on the sidebar, you will need to re-run the analysis from the 'Forest Plot' page.

Results details for all studies

Loading...

                                                          

Gelman convergence assessment plot for all studies

Loading...
Please note: if you change the selections on the sidebar, you will need to re-run the primary and/or sensitivity analysis from the 'Forest Plot' page.

PLEASE NOTE: the package does not currently include unrelated-mean-effects meta-regression models, therefore the consistency vs UME graph that is displayed in the deviance report tab under Bayesian network meta-analysis is not available here.

Deviance report for all studies and the sensitivity analysis

Residual deviance from NMA model and UME inconsistency model for all studies

This plot represents each data point's contribution to the residual deviance for the NMA with consistency (horizontal axis) and the unrelated mean effect (ume) inconsistency models (vertical axis) along with the line of equality. The points on the equality line means there is no improvement in model fit when using the inconsistency model, suggesting that there is no evidence of inconsistency. Points above the equality line means they have a smaller residual deviance for the consistency model indicating a better fit in the NMA consistency model and points below the equality line means they have a better fit in the ume inconsistency model. Please note that the unrelated mean effects model may not handle multi-arm trials correctly. (Further reading: Dias S, Ades AE, Welton NJ, Jansen JP, Sutton AJ. Network meta-anlaysis for decision-making. Chapter 3 Model fit, model comparison and outlier detection. @2018 John Wiley & Sons Ltd.)




Per-arm residual deviance for all studies


This stem plot represents the posterior residual deviance per study arm. The total number of stems equals the total number of data points in the network meta analysis. Going from left to right, the alternating symbols on the stems indicate the different studies. Each stem corresponds to the residual deviance ($dev.ab) associated with each arm in each study. The smaller residual deviance (the shorter stem), dev.ab, the better model fit for each data point. You can identify which stem corresponds to which study arm by hovering on the stem symbols. (Further reading: Dias S, Ades AE, Welton NJ, Jansen JP, Sutton AJ. Network meta-anlaysis for decision-making. Chapter 3 Model fit, model comparison and outlier detection. @2018 John Wiley & Sons Ltd.)




Leverage plot for all studies


This leverage plot shows the average leverage across the arms for each study ({sum($lev.ab)}/{number of arms} for each study) versus the square root of the average residual deviance across the arms for each study (sqrt({sum($dev.ab)}/{number of arms}) for each study). The leverage for each data point, is calculated as the posterior mean of the residual deviance, minus the deviance at the posterior mean of the fitted values. The leverage plot may be used to identify influential and/or poorly fitting studies and can be used to check how each study is affecting the overall model fit and DIC. Curves of the form x2 + y = c, c = 1, 2, 3, ., where x represents square root of residual deviance, and y represents the leverage, are marked on the plot. Points lying on such parabolas each contribute an amount c to the DIC (Spiegelhalter et al., 2002). Points that lie outside the line with c = 3 can generally be identified as contributing to the model's poor fit. Points with a high leverage are influential, which means that they have a strong influence on the model parameters that generate their fitted values. (Further reading: Dias S, Ades AE, Welton NJ, Jansen JP, Sutton AJ. Network meta-anlaysis for decision-making. Chapter 3 Model fit, model comparison and outlier detection. @2018 John Wiley & Sons Ltd. Spiegelhalter et al. (2002) Bayesian measures of model complexity and fit. J. R. Statist. Soc.B 64, Part4, pp.583-639)



Please note: if you change the selections on the sidebar, you will need to re-run the primary and/or sensitivity analysis from the 'Forest Plot' page.

Covariate value:

Covariate value outside data range
0
1
Baysesian covariate meta-regression using the package.
Heterogeneity prior: standard deviation
Please note each simulation may take 60 seconds.

Results for all studies

Please choose your regressor type, then click the button below to run meta-regression analysis (and each time you subsequently change any options).

Loading...
Download

This graph was adapted from Graphs of study contributions and covariate distributions for network meta-regression , Sarah Donegan, Sofia Dias, Catrin Tudur-Smith, Valeria Marinho, Nicky J Welton, Res Syn Meth , 2018; 9 :243-260. DOI: 10.1002/jrsm.1292

Loading...

Options to change limits of the x-axis:

Model fit:



Download
Please note: if you change the selections on the sidebar, you will need to re-run the analysis from the 'Forest Plot' page.

This table only contains the estimates from the network meta analysis, i.e. does not contain estimates from pairwise meta-analysis which only contains direct evidence. If you would like to obtain the pairwise meta-analysis results, please run 4c-4. Nodesplit model


Treatment effects for all studies: comparison of all treatment pairs.

Download
Please note: if you change the selections on the sidebar, you will need to re-run the primary and/or sensitivity analysis from the 'Forest Plot' page.
Please note it may take up to 5 minutes to load the results.
IMPORTANT: If you export and include the Litmus Rank-O-Gram or the Radial SUCRA plot in your work, please cite it as: Nevill CR, Cooper NJ, Sutton AJ, A multifaceted graphical display, including treatment ranking, was developed to aid interpretation of network meta-analysis, Journal of Clinical Epidemiology (2023)

Ranking panel for all studies

Relative Effects

Loading...
Download

Ranking Results

Loading...

Litmus Rank-O-Gram: Higher SUCRA (Surface Under the Cumulative Ranking Curve) values and cumulative ranking curves nearer the top left indicate better performance

Loading...
Loading...

Radial SUCRA plot: Higher SUCRA values indicate better treatments; size of nodes represent number of participants and thickness of lines indicate number of trials conducted

Summary of evidence

Loading...

Numbers on the line indicate number of trials conducted for the comparison. Any shaded areas indicate existence of multi-arm trials between the comparisons.

The size of the nodes and thickness of edges represent the number of studies that examined a treatment and compared two given treatments respectively.

Download

Due to limitations with the underlying R package {gemtc} it is not currently possible to run a regression nodesplit model within MetaInsight.

When this functionality becomes available, it will be added here.

Please note: if you change the selections on the sidebar, you will need to re-run the analysis from the 'Forest Plot' page.

Results details for all studies

Loading...

                                                                  

Gelman convergence assessment plot for all studies

Loading...
Please note: if you change the selections on the sidebar, you will need to re-run the primary and/or sensitivity analysis from the 'Forest Plot' page.

PLEASE NOTE: the package does not currently include unrelated-mean-effects meta-regression models, therefore the consistency vs UME graph that is displayed in the deviance report tab under Bayesian network meta-analysis is not available here.

Deviance report for all studies and the sensitivity analysis

Residual deviance from NMA model and UME inconsistency model for all studies

This plot represents each data point's contribution to the residual deviance for the NMA with consistency (horizontal axis) and the unrelated mean effect (ume) inconsistency models (vertical axis) along with the line of equality. The points on the equality line means there is no improvement in model fit when using the inconsistency model, suggesting that there is no evidence of inconsistency. Points above the equality line means they have a smaller residual deviance for the consistency model indicating a better fit in the NMA consistency model and points below the equality line means they have a better fit in the ume inconsistency model. Please note that the unrelated mean effects model may not handle multi-arm trials correctly. (Further reading: Dias S, Ades AE, Welton NJ, Jansen JP, Sutton AJ. Network meta-anlaysis for decision-making. Chapter 3 Model fit, model comparison and outlier detection. @2018 John Wiley & Sons Ltd.)




Per-arm residual deviance for all studies


This stem plot represents the posterior residual deviance per study arm. The total number of stems equals the total number of data points in the network meta analysis. Going from left to right, the alternating symbols on the stems indicate the different studies. Each stem corresponds to the residual deviance ($dev.ab) associated with each arm in each study. The smaller residual deviance (the shorter stem), dev.ab, the better model fit for each data point. You can identify which stem corresponds to which study arm by hovering on the stem symbols. (Further reading: Dias S, Ades AE, Welton NJ, Jansen JP, Sutton AJ. Network meta-anlaysis for decision-making. Chapter 3 Model fit, model comparison and outlier detection. @2018 John Wiley & Sons Ltd.)




Leverage plot for all studies


This leverage plot shows the average leverage across the arms for each study ({sum($lev.ab)}/{number of arms} for each study) versus the square root of the average residual deviance across the arms for each study (sqrt({sum($dev.ab)}/{number of arms}) for each study). The leverage for each data point, is calculated as the posterior mean of the residual deviance, minus the deviance at the posterior mean of the fitted values. The leverage plot may be used to identify influential and/or poorly fitting studies and can be used to check how each study is affecting the overall model fit and DIC. Curves of the form x2 + y = c, c = 1, 2, 3, ., where x represents square root of residual deviance, and y represents the leverage, are marked on the plot. Points lying on such parabolas each contribute an amount c to the DIC (Spiegelhalter et al., 2002). Points that lie outside the line with c = 3 can generally be identified as contributing to the model's poor fit. Points with a high leverage are influential, which means that they have a strong influence on the model parameters that generate their fitted values. (Further reading: Dias S, Ades AE, Welton NJ, Jansen JP, Sutton AJ. Network meta-anlaysis for decision-making. Chapter 3 Model fit, model comparison and outlier detection. @2018 John Wiley & Sons Ltd. Spiegelhalter et al. (2002) Bayesian measures of model complexity and fit. J. R. Statist. Soc.B 64, Part4, pp.583-639)



Please note: if you change the selections on the sidebar, you will need to re-run the primary and/or sensitivity analysis from the 'Forest Plot' page.

User Guide

User Guide

Click the button below to download a pdf copy of the MetaInsight User Guide.

Please note: the user guide is based on version 3 of MetaInsight. Some elements of the app have been changed since the guide was originally produced.

Download User Guide

NICE Technical Support Documents for Evidence Synthesis

The University of Sheffield host a series of NICE Technical Support Documents on evidence synthesis: https://www.sheffield.ac.uk/nice-dsu/tsds/evidence-synthesis

Some of the models described in TSD2 and TSD3 can be carried out in MetaInsight. A guide has been written with instructions to reproduce the relevant analyses, which can be downloaded here. The guide is based on MetaInsight version 6.0.0.

Download NICE TSD instructions

ESMARConf 2023 Tutorial

MetaInsight - an R Shiny web-app for conducting network meta-analysis

A tutorial for MetaInsight v4.0.0 produced for ESMARConf 2023



Treatment Ranking Demo

A short demo video of how to use the Bayesian analysis ranking panel in tab 3c



Cochrane Training Webinar

MetaInsight: Background, introduction, demonstration, limitations, and future plans

These videos were recorded live in 2019 as part of the Cochrane Training network meta-analysis learning live webinar series. They are intended for people who are interested in undertaking a network meta-analysis using MetaInsight. Please note: these videos refer to a past version of MetaInsight and elements of the app have changed since the videos were originally recorded.