I'm conducting a Bayesian network meta-analysis with 7 studies. 5 studies were observational studies, 2 were randomized control trials. And now I have the following questions,
1. One reviewer said that a “meta-analytic” evaluation using forest plot including both RCTs and observational studies under the same analysis is methodological incorrect since the design of the studies considered are completely different.
I found that Efthimiou et al. in the "Combining randomized and non- randomized evidence in network meta-analysis" reported some methods to allow the inclusion of non-randomized studies. But the number of RCTs in my analysis is small, should I adopt the "the use of non-randomized evidence as prior information" model? or use observational studies in network meta-analysis as sensitivity analysis?
2. I only have 7 studies, should I use funnel plots to assess the publication bias (better when the number of studies > 10)? Are there any other methods to assess the publication bias for the small number of studies?
3. I also conducted pairwise meta-analysis, there only 2 studies in some outcomes, and the reviewer said I cannot perform a meta-analysis evaluating only two studies into a forest plot, so should I use narrative synthesis for these outcomes instead of pairwise meta-analysis?
- Badges
- Science topic
- Similar topics
- Psychology
- Mental Processes
- Learning