Extension using post intercurrent event data to impute missing data for binary endpoints

[statasaurus]

We were wondering if you are planning to extend this package for binary endpoints?

[gowerc]

Hey @statasaurus ,

If memory serves me right I think at present we limit to to continuous only because the conditional mean imputation approach is only theoretically valid for linear models. (@wolbersm / @nociale / @gravesti - please correct me if I've stated that incorrectly).

I don't think we currently have any plans to implement binary endpoint support.

That being said I believe the package does already support binary endpoints that are constructed by discretising a continuous endpoint (e.g. impute the continuous endpoint and then use a custom analysis function + rubin's rules on the derived binary endpoint).

[wolbersm]

Thanks for your interest in rbmi @statasaurus.
I agree with @gowerc that we currently don't have plans to extend the rbmi package to other types of endpoints such as binary endpoints. Having said that, a recent paper on Reference-Based Multiple Imputation for Longitudinal Binary Data presents "Joint Modeling With Multivariate Normal Model and an Adaptive Rounding Algorithm" as one possible (but somewhat crude) approach. It might be possible to implement this approach with the current version of rbmi (while selecting one of the imputation algorithms other than conditional mean imputation) and then adding the rounding step.

[statasaurus]

Thank you! It looks like this paper doesn't handle intercurrent events. Thank you so much for this though