The validation of surrogate endpoints in meta-analyses of randomized experiments.