Background: The Framingham risk models and Pooled Cohort Equations (PCE) are widely used for predicting the 10-year risk of developing coronary heart disease (CHD) and cardiovascular disease (CVD), respectively. Over the past few decades, these models have been extensively validated across different settings and populations. However, no efforts have yet been made to formally synthesise the evidence on the predictive performance of these models and to assess their potential generalisability across different subgroups and geographical regions.

Objectives: To systematically review and summarise the predictive performance of 3 common cardiovascular risk prediction models (Framingham Wilson 1998, Framingham ATP III 2002 and PCE 2013), and to determine sources of heterogeneity.

Methods: A search was performed in December 2015, to identify studies investigating the predictive performance of the aforementioned models. Studies were eligible for inclusion if they validated the original prediction model separately for men or women, to predict its respective clinical outcome in the general population. Performance estimates (observed expected (OE) ratio and c-statistic) were summarised using random effects models and sources of heterogeneity were explored using meta-regression.

Results: The search identified 820 references, of which 29 were included, describing 82 validations. Results indicate that, on average, all models overestimated the risk of CHD and CVD (Figure). Overestimation was most pronounced for high-risk individuals and for European populations. Discriminative performance was better in women for all models. There was considerable heterogeneity in the c-statistic (range pooled estimates 0.64-0.73), likely due to differences in patient spectrum across studies.

Conclusions: The Framingham Wilson, ATP III and PCE have adequate discriminative ability but all overestimate the risk of developing CHD or CVD, especially in European and high-risk populations. Future research should focus on facilitating strategies to tailor these prediction models to specific populations, rather than developing more models that are based on the same or similar predictors.