Research questions: 1. Are “off-the-shelf” CV risk scores accurate, or is “local” customization needed to achieve good predictive accuracy? 2. Does the statistical model matter? Do more complex models yield more accurate risk scores than simpler ones? Conclusions: 1. “Off-the-shelf” FRS and PCS performed relatively well with EHD, but prediction accuracy can be improved by refitting models to “local” data. The age effect was stronger and the effect of lipid levels weaker in refitted models. 2. PCS performed slightly worse than FRS, indicating that increasing the complexity of the model used to compute the risk score may harm performance in new settings.