Validation of two Framingham cardiovascular risk prediction algorithms in an Australian population: the ‘old’ versus the ‘new’ Framingham equation

Background: Multivariable risk prediction equations attempt to quantify an individual’s cardiovascular risk. Those borne from the Framingham Heart Study remain the most well-established and widely used. In February 2008, a new Framingham risk equation was published. We sought to determine the differences between the most commonly used Framingham equation from 1991 and the 2008 version through their application to a contemporary Australian population.

Methods and results: The two risk equations were applied to 7329 individuals from the Australian Diabetes, Obesity and Lifestyle study. All individuals were aged 30–74 years and free of cardiovascular disease. Differences in median risk scores were analyzed through the Wilcoxon’s signed rank test. Compared with the 1991 equation, median cardiovascular risk scores derived from the 2008 equation increased by 7 and 24% over 5 years, among males and females, respectively. The differences were statistically significant across all age-groups for both males and females, P value of less than 0.001. The performance of the equations in predicting cardiovascular outcomes were compared using event rates. The discriminative ability was increased using the 2008 equation; however the difference was non-significant [area under the receiver operating characteristic curve: 1991 equation 0.74 (0.69–0.80); 2008 equation 0.76 (0.71–0.81)].

Conclusion: Earlier Framingham equations have been suggested to over-predict cardiovascular risk in low-risk populations and under-predict risk in high-risk groups. This is the first comparative validation of the previous 1991 and most recent 2008 equations. This study highlights the need to validate and calibrate cardiovascular risk prediction equations using the population-specific outcome data.