Abstract:
|
When we are interested in the effect of a treatment will have in a specific target population, it is unlikely that we will have access to a randomized clinical trial (RCT) whose participants are a random sample from that population. Because RCTs often lack the ability to naturally generalize to populations outside the population of study participants, obtaining causally interpretable estimates of the treatment effect in our target population can be challenging. When multiple RCTs on a treatment are available, we may be able to obtain a causally interpretable estimate of the treatment effect in the target population by using weights to synthesize participant data across the trials. Existing weighting methods construct weights by first pooling across studies, treating the data as if it came from one large study instead of several smaller ones. An alternative weighting method accounts for study-membership through the use of participant-level and study-level weights. We examine the performance of the different weighting methods using simulations, highlighting settings where different methods provide the most favorable statistical properties.
|