Dear all,
Currently I'm working on a systematic review. In many studies I just have information on the odds ratios and not the raw data (events, totals). In RevMan I want to make a forest plot with the OR, but I can't figure out how to work only with the OR because the program asks for also the raw data (events and totals).
Is there a way how to work only with the odds ratio, in case of missing the raw data?
Thank you.
Unfortunately, I don’t think you can work with odds ratios alone in RevMan. Your best bet will be to contact the corresponding author so he/she can provide information that will help you to reconstruct the relevant contingency table.
Another option will be to “reconstruct” the said table for the outcome in a particular study on your own (i.e., without input from the corresponding author) by imputation, to reach as close as possible the reported odds ratio, while being explicit on the underlying assumptions you will utilize (e.g., in terms of the row and column totals, in terms of the distribution of the number of events, etc.). However, I think that this option is extremely risky even with expert assistance.
@Aedrian Abrilla You cannot impute raw data from odds ratios even semi accurately, even if you know the number of patients in each group, because many iterations will give the same odds ratio but with varying precision which can drastically affect the results. This should never be done. The reason RevMan does not impute them is because it is not accurate.
Odds ratios are often combined that are not derived from contingency tables, such as when pooling ordinal shift analyses - there is no need to impute the raw data.
@Richard, to work with the raw odds ratios in RevMan, log transformed odds ratios can be combined using generic inverse variance meta-analysis. You will also need the standard error, which you can calculate from the confidence interval if it isnt provided in the paper.
Jack Henry , hence my last statement.
My point is that the first-line strategy is to contact the corresponding author, if possible. You are also assuming details that may or may not be relevant to the context of the question.
Your suggestion also involves utilization of the standard error/confidence interval, which theoretically can also be used in reconstructing/imputing the contingency table. You don't just stop and settle when the point estimate of the OR from your reconstructed table matches the one reported, you should also consider if the dispersion measure/precision/confidence interval of the resulting OR is close or similar.
The odds ratio is calculated by dividing the odds of the first group by the odds in the second group. In the case of the worked example, it is the ratio of the odds of lung cancer in smokers divided by the odds of lung cancer in non-smokers: (647/622)/(2/27)=14.04.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size