External validation of clinical prediction models: simulation-based sample size calculations were more reliable than rules-of-thumb