Adding multiple risk factors improves Framingham coronary heart disease risk scores.

Abstract

PurposeSince the introduction of the Framingham Risk Score (FRS), numerous versions of coronary heart disease (CHD) prediction models have claimed improvement over the FRS. Tzoulaki et al challenged the validity of these claims by illustrating methodology deficiencies among the studies. However, the question remains: Is it possible to create a new CHD model that is better than FRS while overcoming the noted deficiencies? To address this, a new CHD prediction model was developed by integrating additional risk factors, using a novel modeling process.MethodsUsing the National Health Nutritional Examination Survey III data set with CHD-specific mortality outcomes and the Atherosclerosis Risk in Communities data set with CHD incidence outcomes, two FRSs (FRSv1 from 1998 and FRSv2 from National Cholesterol Education Program Adult Treatment Panel III), along with an additional risk score in which the high density lipoprotein (HDL) component of FRSv1 was ignored (FRSHDL), were compared with a new CHD model (NEW-CHD). This new model contains seven elements: the original Framingham equation, FRSv1, and six additional risk factors. Discrimination, calibration, and reclassification improvements all were assessed among models.ResultsDiscrimination was improved for NEW-CHD in both cohorts when compared with FRSv1 and FRSv2 (P<0.05) and was similar in magnitude to the improvement of FRSv1 over FRSHDL. NEW-CHD had a similar calibration to FRSv2 and was improved over FRSv1. Net reclassification for NEW-CHD was substantially improved over both FRSv1 and FRSv2, for both cohorts, and was similar in magnitude to the improvement of FRSv1 over FRSHDL.ConclusionWhile overcoming several methodology deficiencies reported by earlier authors, the NEW-CHD model improved CHD risk assessment when compared with the FRSs, comparable to the improvement of adding HDL to the FRS.