We investigated whether adding a 9p21.3 variant, a KIF6 variant, or CRP to the FRS could improve MI risk prediction in CHS, a large prospective study of individuals aged 65 years or older from American communities. Adding individual risk markers had, at best, a modest effect on risk prediction. Adding two or more risk markers, specifically risk marker combinations that included the KIF6 719Arg genotype, resulted in a somewhat higher, though still modest improvement in risk prediction as measured among white males by either the AUC or by NRI, but not in the combined male and female population.
The appropriate methods for assessing improvement of risk prediction by risk markers have been widely debated [26
]. And given the limitations of the AUC measure, alternative assessment methods such as NRI have been proposed [24
]. Although the current study was not designed to compare the AUC and NRI, the results from these two measures were largely consistent.
The ability of CRP to improve risk stratification has been evaluated in multiple studies, which found that the addition of CRP to risk prediction models provided modest improvement in risk prediction (see Buckley et al.
] and Schnell-Inderst et al.
] for recent comprehensive evaluations). In CHS, CRP did not improve risk prediction by the FRS as assessed by the AUC or NRI measures, consistent with previously published observations that the association of a risk factor with MI after adjustment for the FRS variables does not necessarily mean that the risk factor will improve risk prediction beyond the FRS, as measured by the change in the AUC, partly because even risk factors with large odds ratio (up to 7) have distributions that overlap substantially between those with and without disease [28
]. However, the addition of CRP in combination with KIF6
719Arg to the FRS improved risk prediction by both the AUC and NRI measures among white men.
Although the association between CHD and SNPs in the 9p21.3 locus has been reported extensively [7
], Dehghan et al.
] reported that SNPs in this locus were not associated with risk of CHD in a prospective study of older Europeans. In contrast, we have observed in CHS, a population of individuals 65 years and older, that a SNP in the 9p21.3 locus is associated with MI among men, although this association did not reach statistical significance among women. We have also observed in CHS an association between MI and KIF6
719Arg carrier status among men but not among women. However, published results from the Women's Health Study [10
] suggest that both 9p21.3 and KIF6
719Arg are associated with MI among women. These inconsistent findings among women could be due to lack of power in CHS, or to different baseline risk of MI in CHS and the Women's Health Study. Additional analysis of the 9p21.3 and KIF6
variants in prospective studies of women would be required to understand this apparent inconsistency.
The addition of the 9p21.3 variant to the FRS did not improve risk prediction in CHS as measured by either the AUC or NRI among whites. Similarly, addition of a SNP in the 9p21.3 locus did not improve risk prediction by traditional risk factors in the Women's Health Study [10
]. However, in ARIC [8
] the addition of 9p21.3 to the FRS resulted in a modest but statistically significant improvement in risk prediction as measured by AUC or reclassification, and in the Northwick Park Heart Study II [9
] adding 9p21.3 improved reclassification but not AUC. These inconsistent results could be attributed to differences in baseline risk between the studies, (the older CHS population is likely to have greater baseline risk of MI), which could affect the power to detect the modest improvement in risk prediction contributed by the 9p21.3 variant.
Genetic and non-genetic biomarkers offer different benefits in the assessment of CHD risk. Non-genetic biomarkers could change over time, and therefore, repeat measurements may be necessary because of day-to-day variation in the level of these biomarkers. However, repeat measurements of non-genetic biomarkers may also provide an indication of successful medical therapy or life-style modification. Genetic biomarkers do not change and thus need only be measured once to obtain information about the lifelong exposure to that biomarker. The 9p21.3 and KIF6
gene variants were chosen for investigation because they have both been reported to be associated with CHD in multiple prospective studies and are common variants. For example, in the white population about 75% of carry at least one risk allele of 9p21.3 and about 65% carry at least one KIF6
719Arg risk allele. CRP was chosen because of the well-established association between CRP levels and risk of CHD and because of continuing interest in whether it should be added to risk prediction algorithms. Elevated CRP is also common. For example, although CRP was analyzed as a continuous variable in this study in order to increase the power of the study, others have reported that in CHS, 26% of the population have elevated CRP (> 3 mg/dL) [31
]. Thus, ~17% of the CHS white population have both elevated CRP and carry the 719Arg allele of KIF6
This study has several limitations. The AUC of the FRS model for white men (0.581) and white women (0.619) in this study of older individuals is lower than the AUC that has been reported for middle age populations (e.g., 0.75 and 0.83 among white men and among white women in ARIC [2
]), thus the markers we studied may only improve risk prediction in populations in which the ability of the FRS to predict CHD is modest. The difference observed between the AUC and NRI measures for the addition of KIF6
719Arg among men could be attributed to the overall modest improvement of risk prediction by single marker addition. Another limitation of this study is the limited number of genetic markers evaluated--a recent paper suggested that risk prediction models might be improved by incorporating large number of genetic markers into a genetic risk score [32
]. Lastly, the effect of these single and multiple marker additions on risk prediction was investigated in a population of individuals aged 65 or older at baseline and our observations may not be generalizable to younger populations.