Concordance evaluation of coronary risk scores: implications for cardiovascular risk screening

Objective: To assess the similarities and differences in predicted high-risk individuals identified by different cardiovascular risk calculation algorithms Research design and methods: A representative population of 10 000 individuals was modelled in a computer using baseline data from the National Health Survey for England. The effects of biological variation in each major model parameters were then applied to each hypothetical individual. The predictive capacities of 3 different risk identification systems based on computer calculation (the Framingham algorithm), or on tabular methods (the Sheffield tables and the General Rule to Enable Atheroma Treatment) were evaluated. Results: All three models predict that similar numbers would receive treatment with 2.9 and 10% receiving treatment at 30 and 15% 10 year risk thresholds, respectively. However, concordance is limited as 0.3 or 6.8% are positive on all three systems; 1.6 or 9.7% on any two calculators at the 30 and 15% thresholds, respectively. The risk groups identified by each calculator depend on the baseline assumptions in each model. Conclusion: Care needs to be taken with applying risk calculators to populations different from which they were derived. Any cardiovascular risk scoring system needs to be thoroughly evaluated against epidemiological data before it is introduced and also needs to be updated in line with changing trends in risk factors.