On shrinkage and model extrapolation in the evaluation of clinical center performance