I have encountered two studies (observational) where the number of events is zero (0) for both disease (0 events out of 70 subjects) and control subjects (0 events out of 50 subjects). Meaning neither diseased nor healthy subjects exhibited the factor (for example antibodies). In this case RevMan says 'Non-estimable' in the table.
Would this result be alright? Shall I need to do anything else?
Your response would be highly appreciated.
Hi,
We cannot estimate the RR or OR in a study with 0 events in both the exposed and unexposed groups. Therefore, the result by RevMan ('non-estimable') is alright - no worries.
Such studies figure only in the qualitative analysis and not in the pooled analysis or meta-analysis.
Jeeva
I agree with Jeeva, no problem. This will be presented on the graph and won't affect your pooled estimates.
Great answer by Dr. Jeeva indeed. Very interestingly, I recently used the software R and this issue was resolved! Instead of the 'non-estimable' it shows a very minor change which affects very lightly the pooled analyses. I don't know how the software did, but it was interesting.
Dear Islam,
I understand what you mean.
RR cannot be calculated with '0' events because the incidence in both the groups is '0' - the RR will become 0/0.
Some of the statistical software programs try to circumvent the issue by a process called continuity correction. What they do is just to add 0.5 (or a similar small number) to each of the cells of the 2x2 table and calculate the RR. So we have the RR but the 95% CI will be too wide - the study will not have any weightage in the meta-analysis. It is almost like dropping the study altogether.
Jeeva
Many statistical softwares add 0.5 to cells with low observations by default.
- Badges
- Science topic
- Similar topics
- Social Psychology
- Attitude