I haven’t seen a rstanarm or brms implementation of a parametric proportional hazards regression, and thus was wondering whether the one implemented by @sambrilleman in stan_jm
could be used completely separate from the longitudinal submodels (I know that completely misses the point of the function)?
I read that you can choose to use no association structure, thus making the event model independent from the longitudinal submodels. Would you still have to call stan_jm
with at least one longitudinal model or is there a way to omit it entirely?

If the latter holds, wouldn’t this potentially give us a rstanarm implementation of a parametric hazards regression, similar to
coxph
maybe even leading tostan_coxph
? 
If the former holds, would it make sense to call
stan_jm
with a dummy longitudinal model and data?