Abstract:
|
In the presence of treatment effect heterogeneity, the treatment effect in a randomized controlled trial (RCT) may differ from the treatment effect in a target population (TP) of interest. If all effect moderators are observed in the RCT and TP, we can obtain an estimate for the TP treatment effect by adjusting for the difference in the distribution of the moderators between the two samples. We consider sensitivity analyses for when (1) we cannot adjust for a specific moderator V observed in the RCT because it is not observed in the TP; and (2) we are concerned that treatment may be moderated by factors not observed in the RCT, represented as a composite moderator U. For situation (1) we offer: (i) a bias-formula-based technique that involves specifying a range for the mean of V in the TP; (ii) a weighting-based technique that involves specifying a range for the distribution of V in the TP conditional on observed moderators Z (if any); and (iii) a hybrid of (i) and (ii). For situation (2), we offer a modified version of (i) for a U that is uncorrelated with observed moderators Z; and a modified version of (iii) for a U uncorrelated with Z and possibly other covariates X.
|