Bayesian predictive projection for variable selection

That is the model approximate approach I used in RMS. What is not described above is how you get posterior distributions for the second stage model taking into account uncertainties from both the first stage and from decisions made from the first stage regarding which features to discard.