Abstract:
|
Network meta-analysis pools the results from randomized controlled trials comparing different treatments. It has been noted in the literature that the choice of treatment comparison made at the study-design stage is often dependent on the chosen/available population to be recruited into each study. Therefore, pooling results over studies without adjustment might lead to confounded estimates of the effect of interest and lack of clarity as to the population on which the effects are defined. Current proposed methods for network meta-analysis often place a model on the contrast (e.g. risk ratio) estimated in the studies, the results of which are heavily reliant on both parametric modeling and contrast choice. Making explicit parallels to the causal inference framework and using counterfactual principles, we instead focus on the mean outcome when assigning a specific treatment to a study arm. We seek to 1) define the population of interest in the network meta-analysis, 2) nonparametrically define the target parameter, 3) determine the requirements for identifiability of this parameter, and 4) propose modeling strategies to consistently estimate the target parameter.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.