Orbit logo

Outcome Reporting Bias in Trials


Copas Adjustment Method

Upload a file

The Copas adjustment method allows users to perform a sensitivity analysis in order to estimate the impact of outcome reporting bias on their review meta-analyses. The method can be applied to both benefit and harm outcomes and is the current method of adjustment that is recommended by the ORBIT team. The method is currently implementable for binary outcome data but the study team are working on other data types for the future.

To implement the method, follow the 10 easy steps below. The functions require a specific .csv input file - an example template for a review of Epilepsy can be downloaded which can be easily amended to apply to other examples. The Epilepsy example had 12 studies included in the analysis and 14 outcomes (2 benefits outcomes and 12 harms). The number of studies (rows) and outcomes (columns) can be added or deleted depending on the specific example. Each outcome has two columns (the number of events for the treatment group and the number of events for the comparator group)

To download the example template please click here






Steps for implementing the COPAS adjustment method

  1. Download the example template
  2. In the first two columns, the sample sizes (denominators) to be used in the meta-analysis should be added for each study (column 1: treatment, column 2: comparator)

    [NOTE: this information is required and should be available for all studies]

    The top row of the file is required. It is an identifier to denote whether the outcome being considered is a benefit or harm outcome
  3. Enter either ‘ben’ for a benefit outcome or ‘harm’ for a harm outcome. ‘ben’ or ‘harm’ needs to be entered for both the treatment and comparator groups.

    [NOTE: it is important that benefit and harm outcomes are specified correctly as the Copas adjustment applied is different dependent on the outcome type]

    The second row of the file is required. It is an identifier to denote whether 95% of 99% confidence limits are required
  4. Enter either ‘95’ or ‘99’ for 95 or 99% confidence limits respectively. ‘95’ or ‘99’ needs to be entered for both the treatment and comparator groups. Different limits can be specified for different outcomes.
  5. Change the outcome names accordingly (row 3)
  6. Enter the number of events for the treatment group and comparator (in that order) for each outcome

    Some studies will have missing outcome data, in which case there will be no numeric values to enter in 6). Missing outcome data will have been identified by completing the 'Outcome Matrix'. An ORBIT assessment of each outcome with missing data is required using the a separate ORBIT classification system for the benefit outcomes and the harm outcomes. Based on the judgment from the classifications, the reason for missing outcome data is assigned either ‘Low’ or ‘High’ risk of bias.
  7. Enter either ‘Low’ or High’ according to the ORBIT assessment when the number of events is missing for a particular outcome. ‘Low’ or ‘High’ needs to be entered for both the treatment and comparator groups.

    [NOTE: the COPAS method does no currently account for ‘No Risk’ classifications although our empirical research found these to be rare. Please contact us through 'Contact Us' if you have ‘No-Risk’ classifications in your assessments for further advice]
  8. Save the .csv file
  9. Select the method of estimation: Mantel-Haenszel and Maximum Likelihood are currently available
  10. Search for the saved data file using the browse facility on the webpage and click on 'upload'

The website will then display the unadjusted and adjusted estimates for each outcome, separated by whether they were declared as a benefit or harm outcome. As a check, unadjusted estimates will typically match those produced by statistical packages or RevMan provided the method of estimation selected is comparable.