New statistical methods for the evaluation of cardiovascular risk markers: what the clinician should know

Calculation of odds ratios or hazard ratios by multivariate logistic regression or Cox regression analyses have traditionally been used to show that candidate risk markers provide prognostic information independently of conventional risk markers, but it has become increasingly clear that a statistically significant increase in risk in multivariate models may not represent a clinically meaningful improvement in overall prediction. This observation has prompted the development of more clinically relevant statistical methods, including tests of the ability to provide incremental discrimination compared with traditional risk markers by calculation of the C-statistic (corresponding to the area under the receiver operating curve) of the model and methods for evaluating improvement in risk classification by the use of event-specific re-classification tables or the use of the integrated discrimination improvement. In the present issue of Clinical Science, Khan and co-workers have evaluated the prognostic value of the GRACE (Global Registry of Acute Coronary Events) risk score, the cardiac biomarker NT-proBNP (N-terminal pro-B-type natriuretic peptide), and their combination, for the prediction of mortality after acute myocardial infarction, using a combination of statistical methods.