Study-centric data analyses vs custom-input analyses

Study-centric data is only available for the reviews that enter the pilot. If you are interested in learning more about the pilot and how to join, please contact revman@cochrane.org.

An overview

This page gives an overview of the differences between study-centric data analyses and custom input analyses. It also presents pros and cons of both approaches.

The main difference is:

  • In a custom-input analysis the study result data is entered and stored in the analysis data table. This is the traditional analysis type done in Review Manager 5.3.
  • In a study-centric data analysis the meta-analysis presented in the forest plot is pulled together from result data entered and stored in the included study. We call this structure 'study-centric data structure'.

Choosing to use study-centric data analyses in the review help authors think more systematically about their analysis. It also enables a future Covidence integration that supports updates. For sensitivity analyses where data is reused, study-centric data structure reduces risk of data entry errors. Long term, study-centric data structure will allow introduction of new statistical methods. Further information is available at: https://community.cochrane.org/help/tools-and-software/revman-web/updates-revman-web/study-centric-data-revman-web.

It is currently not possible to migrate all existing custom-input analyses to study-centric data analyses. Therefore a review can include both study-centric data analyses and custom-input analyses.


1. Where to enter study results?

In custom-input analyses, study results are entered in the data table of each analysis

Study-centric data analysis is done in RevMan Web. Study results are entered on the result data tab of an included study for each outcome, after adding the study arms of the study. In a future Covidence integration, the data extracted in Covidence will be automatically transferred to the studies.

2. How to set up an analysis?

In custom-input analysis, the study results are added for each outcome in a single data table.

In study-centric data structure, setting up an analysis is done after the interventions and outcomes of the review are set up in the PICO tab and the result data is added in the studies.  To set up an analysis, one needs to choose outcome, experimental intervention and control intervention, and a possible subgroup factor. RevMan Web then identifies which studies have result data that matches the synthesis PICO and pools the results for the analysis. 

3. How Review Manager 5.3 and RevMan Web tools handle the two analysis approaches?


Custom-input analyses

Study-centric data analyses

Create analysis in which tool?

RevMan 5 and RevMan Web

RevMan Web only.

View analysis in RevMan 5YesYes; it looks like a custom-input analysis.
Edit analysis in RevMan 5YesNo; it can be edited in RevMan 5 as a custom input analysis, but the edits will be discarded after check in.
Edit analysis in RevMan WebYesYes

4. Types of analyses available in the two approaches

This table presents the types of analyses available in two approaches and how they are set up.

Type of analyses

Custom input analyses

Study-centric data analyses

Analyses without subgroupsYesYes
Subgrouping by a property of the included studies
e.g by outcome measure - a study falls under one of the categories of the covariate
YesYes, set covariate categories for each study and choose subgroup factor in the synthesis PICO .
Subgroup by risk of biasYesNo, implementation planned. Use custom-input analysis for now.  
Combining armsYes, with the help of the calculator in RevMan 5.3.Yes, done automatically.
Splitting control arms
e.g. in analyses that subgroup by the type of intervention (e.g. low-dose vs high-dose),
sometimes the same control arm is used multiple times
Yes, compensate for this by "splitting" the control arm, so that the overall pooled estimate wouldn't be over-confident.Yes, the study will automatically be counted only once for the pooled estimate. But this is only relevant when subgrouping by intervention is implemented.
Subgroup by property of the intervention groups (arms)YesNo. Implementation planned, use custom-input analysis for now.  
Subgroup by variants of the outcomeYesNo. Implementation planned, use custom-input analysis for now.  

Subgroups within studies, e.g. subgrouping the participants of a study

YesNo, not planned. Use custom-input analysis.
Contrast-level data (the current GIV and "O-E and variance" analyses)YesNo. Implementation planned, use custom-input analysis for now.