Arm level vs contrast level data

---

Study results come in two main forms: 

1. Arm level data: capture the outcome for each arm separately, for example the number of participants in each arm that experienced an event of interest, or the mean change from baseline in a symptom rating scale for participants in each arm.
2. Contrast level data: capture the differences between arms as an effect size, such as mean difference, (adjusted) odds ratio, hazard ratios. 

Either or both types of data can be included in an analysis. RevMan will automatically calculate effect sizes from arm level data when needed but will generally only include contrast level data if the study result has used the same effect measure as the analysis (so, for example, a risk ratio study result won’t be included in an odds ratio analysis). 

With automatic analyses (under ‘Data source’), you can choose to include:

1. 1. Only arm-level data
   2. Only contrast-level data
   3. Contrast- and arm-level data (preferring arm-level data where both exist)
   4. Contrast- and arm-level data (preferring contrast-level data where both exist) 

The compatible study results will then be included in the analysis according to your preference. 

Section 5.3.6 of the Cochrane Handbook for Systematic Reviews of Interventions provides more guidance on the different types of data and in what circumstances either type might be preferred. 

Specific concerns for multi-arm trials

Multi-arm trials are trials that include more than two arms and therefore more than one comparison. When including contrast level data on multi-arm trials, there are two specific scenarios you need to be aware of: 

1. If multiple arms report on the same intervention (from the perspective of the review or the analysis), then multiple comparisons from that study can contribute information to the analysis. However, as these comparisons are both to the same reference arm (i.e. B vs A and C vs A), they are not statistically independent.
2. If the study reported results against one comparator (i.e. B vs A and C vs A), but in the analysis we need results against another comparator (e.g. C vs B). Here, we need to calculate the third comparison from the other two.
3. The previous two scenarios may apply simultaneously for the same study, including for studies with more than three arms. 

The transformations needed to handle these scenarios are applied automatically if contrast level data are included in the analysis (details are covered in the Statistical methods). However, the covariance between the comparisons in the study is required for this. The covariance can be calculated or approximated in various ways (see Covariance calculator). 

Examples of both types of data

Here is an example of arm level data: 

Here is an example of contrast level data: