Starting to put together some material on matching-adjusted indirect comparisons, a method used for indirect comparisons between treatments when you have IPD of studies for one treatment but aggregate data for another. The most popular approach seems to be to use propensity score weighting matched on any effect modifiers, but I am wondering how this protects against spuriously large interaction effects (it doesn’t?).
At the end of the day, whether matching or weighting, my understanding is that the method should run into the same potential issues as if you ran an unregularized regression with all the treatment interaction terms of interest and then plugged the average treatment effects of the target trial into the results (i.e. your coefficient estimate is inflated). Is there something special about propensity score weighting that I’m missing that makes this not be a potential problem? Most of what I’ve pulled up for propensity scores and over fitting is focused on inflating standard error.