High sensitivity C-reactive protein (hsCRP) and family history independently associate with future cardiovascular events and have been incorporated into risk prediction models for women (the Reynolds Risk Score for women). However, no cardiovascular risk prediction algorithm incorporating these variables currently exists for men.
Methods and Results
Among 10,724 initially healthy American non-diabetic men who were followed prospectively over a median period of 10.8 years, we compared the test characteristics of global model fit, discrimination, calibration, and reclassification in two prediction models for incident cardiovascular events, one based on age, blood pressure, smoking status, total cholesterol, and high-density lipoprotein cholesterol (traditional model), and the other based on these risk factors as well as hsCRP and parental history of myocardial infarction before age 60 years (Reynolds Risk Score for men). 1,294 cardiovascular events accrued during study follow-up. Compared to the traditional model, the Reynolds Risk Score had better global fit (likelihood ratio test P<0.001), a superior (lower) Bayes Information Criterion, and a larger C-index (P<0.001). For the endpoint of all cardiovascular events, the Reynolds Risk Score for men reclassified 17.8 percent [1,904/10,724] of the study population (and 20.2 percent [1,342/6,884] of those at 5 to 20 percent 10-year risk) into higher- or lower-risk categories with markedly improved accuracy among those reclassified. For this model comparison, the net reclassification index (NRI) was 5.3 percent and the clinical net reclassification index (CNRI) was 14.2 percent (both P-values<0.001). In models based on the ATP-III preferred endpoint of coronary heart disease and limited to men not taking lipid-lowering therapy, 16.7 percent of the study population (and 20.1 percent of those at 5 to 20 percent 10-year risk) were reclassified to higher- or lower-risk groups, again with significantly improved global fit, larger C-index (P<0.001), and markedly improved accuracy among those reclassified. For this model, NRI was 8.4 percent and CNRI 15.8 percent (both P-values <0.001).
As previously shown in women, a prediction model in men that incorporates hsCRP and parental history significantly improves global cardiovascular risk prediction.