I apologize if the question is a bit unclear, please allow me to elaborate using specifics.
In regards to a surgical intervention (lets say an appendectomy), my collaborators and I would like to compare the outcomes of manoeuvre A and B in laparoscopic appendectomy. Unfortunately, there have not been any clinical trials yet that compare the outcomes manoeuvre A and B in laparoscopic appendectomy.
What we have found, however, is RCTs that have compared the outcomes of open versus laparoscopic appendectomy. In certain RCTs, manoeuvre A is used for the laparoscopic appendectomy, and in other studies, manoeuvre B is used. Again, in each study we have found, either A or B is used (not both), as the studies are really comparing laparoscopic versus open appendectomy.
My question is: Is it possible to gather the outcomes from one limb of multiple RCTs (i.e. the laparoscopic appendectomy limb)--some of which use manoeuvre A and some of which use B--and compare them in any sort of systematic or meaningful way?
I have not seen a study that has done this before, which may be somewhat telling, but my knowledge of statistics is limited and I would love another opinion. If a direct comparison is not possible, would the presentation of the results be meaningful, or would the bias be too great?
Thank you very much for your time.
I am afraid this method of indirect comparison will bring much more biases. Let's suppose a group of manoeuvre A in one RCT is well matched with another group. It is by means of randomization that eliminates the influence of some biases . So if the group of manoeuvre A is used for comparison with manoeuvre B group from another RCT, I think the advantage of randomization is gone and much more biased will be brought.
In terms of your purpose, I think indirect comparison meta-analysis and network meta analysis will be better choices.
Look forward you could find more professional solution.
Dear Lojan.
Short answer: No.
Explain: The groups in each trial is different. Therefore, you cannot compare them in a statistical way. If you suspect that manoeuvre A is better/worst or even the same than manouvre B and you want to check, you have to run a specific trial with just one sample and randomize. Another way can be an observational study (e.g. propensity score) but you need a lot of data and more complicated analysis.
Hope be useful.
Regards
Mauricio Gervasoni
This sounds like an ideal scenario to run a network meta-analysis with three potential treatment options (open appendicectomy, lap appendicectomy A, lap appendicectomy B); provided the patients in each of these RCTs are comparable. What I mean is that any patient in the network could have theortically been randomly allocated to open, lap A or lap B; and that the lap options A and B are not used selectively depending on a particular clinical or patient factor. This is known as the transitivity assumption for network meta-analysis, and if it cannot be met, then you should not really run an NMA.
You will then be able to use the indirect comparisons in the network to draw inferences about the effect estimate of 'lap A' vs 'lap B' using the common comparator treatment 'open'. Although you do not have any direct (lap A vs. lap B) trials, this should still provide a useful result, provided you keep in mind the known sources of clinical heterogeneity. Here are some useful links for this:
- http://www.bmj.com/content/331/7521/897 is a seminal paper by Caldwell and provides a good introduction to the theory and principles.
- http://www.mtm.uoi.gr/ which contains methodological step by step instructions, and a link to the paper by Chaimani et al with the Stata network meta package, and the graphical tools.
All the best.
- Badges
- Science topic
- Similar topics
- Academic Journals