[Cardiovascular risk assessment for informed decision making. Validity of prediction tools]

Patient involvement in health care decisions is increasingly requested. The authors investigated whether currently available assessment tools for prediction of cardiovascular risk can be used for individual risk prediction as a basis of informed decision making.

The authors searched for risk assessment tools and respective validation studies in Medline (until August 16, 2004) and the Cochrane Library (issue 2/2004). The following criteria were used for evaluation of prognostic studies: (1) discrimination between risk groups; (2) predictive values; (3) prognostic agreement; (4) transferability across populations.

A total of twelve assessment tools were identified. The Framingham function, Sheffield Tables, Canadian Tables, Framingham Categorial, New Zealand, Joint British, and European Charts (1994 and 1998) are based on the Framingham Study; PROCAM Risk Score, UKPDS Risk Engine, and SCORE Risk Charts use different source data. Framingham-based instruments overestimate cardiovascular risk of Central-European populations by at least 30%, with substantial regional variation even within a country (between 30% and 100%, British Regional Heart Study). Therefore, prior to application the assessment tools would need recalibration using regional data of cardiovascular mortality and adjustment for social class differences. Published sensitivity, specificity, and C-statistics for external validation (area under the curve [AUC] approximately 0.6) are clearly inferior to internal validation (AUC approximately 0.8). Agreement between instruments beyond chance is moderate (kappa approximately 0.5). No studies on external validation could be identified for the new European SCORE Risk Charts and UKPDS Risk Engine.

Validation of currently available assessment tools for cardiovascular risk prediction is inadequate. Uncritical use may lead to substantial under- or overestimation of individual cardiovascular risk and inappropriate treatment decisions.