Impact of sample size on the stability of risk scores from clinical prediction models: a case study in cardiovascular disease