We have shown that apparent validation is limited in value because of optimism in the performance estimates.

Internal validation should be attempted, which is preferably done by bootstrapping.

External validation is a stronger test of model performance, and should be determined in other populations, which are (subjectively) considered ‘plausibly related’.