The Coronary Artery Disease–Associated 9p21 Variant and Later Life 20-Year Survival to Cohort Extinction

Abstract

Background

Common variation at chromosome 9p21 (marked by rs10757278 or rs1333049) is associated with coronary artery disease (CAD) and peripheral vascular disease. A decreasing effect at older age was suggested, and effects on long-term mortality are unclear. We estimated 9p21 associations with CAD and all-cause mortality in a CAD diagnosis–free older population. We also estimated classification gains on adding the variant to the Framingham Risk Score (FRS) for CAD.

Methods and Results

DNA was from an Established Populations for Epidemiological Study of the Elderly–Iowa cohort from 1988 (participants >71 years), with death certificates obtained to 2008 for 92% of participants. Cox regression models were adjusted for confounders and CAD risk factors. Of 1095 CAD diagnosis–free participants, 52% were heterozygous (CG) and 22% were homozygous (CC) for the risk C allele rs1333049. Unadjusted CAD-attributed death rates in the CC group were 30 vs 22 per 1000 person-years for the GG group. The C allele was associated with all-cause (hazard ratio, 1.19; 95% CI, 1.08–1.30) and CAD (hazard ratio, 1.29; 95% CI, 1.08–1.56) mortality, independent of CAD risk factors. There was no association with stroke deaths. Variant associations with CAD mortality were attenuated after the age of 80 years (age-interaction term P=0.05). In age group 71 to 80 years, FRS classified as high risk 21% of respondents who died of CAD within 10 years; adding 9p21 identified 27% of respondents.

Conclusions

In 71- to 80-year-old subjects free of CAD diagnoses, 9p21 is associated with excess mortality, mainly attributed to CAD mortality. Adding 9p21 to the FRS may improve the targeting of CAD prevention in older people, but validation in independent samples is needed for confirmation.

Genome-wide association studies^1,2 have identified a common genetic variant on chromosome 9p21 (identified by single-nucleotide polymorphism [SNP] rs1333049 or rs10757278) to be strongly associated with coronary artery disease (CAD) and its major complication, myocardial infarction (MI),³ across races.⁴ The same variant is associated with other vascular diseases, including aneurysms at various arterial sites⁵ and peripheral arterial disease.⁶ The effect of the variant may be mediated by impaired response of vascular endothelial cells to inflammation⁷ and compromised repair capacities of vascular cells⁸ rather than the established CAD risk factors.²

Coronary artery disease is a major cause of mortality, yet few studies have estimated the impact of the 9p21 variant on CAD-related mortality.^9,10 Analyses thus far have included subjects mostly aged 50 to 70 years, because heritable factors are assumed to most strongly affect the risk of CAD in middle and early old age.¹¹ The risks associated with genetic variants may decrease with advancing age.¹² Thus, there has been little research on associations between the 9p21 variant and CAD mortality at older ages.¹³ In addition, there is little evidence on whether this variant is associated with mortality attributable to stroke.

Genome-wide studies generally use case-control designs,¹⁴ which are robust for discovering genetic associations with phenotypes. However, cohorts followed up provide more reliable estimates of survival hazards in populations.

The National Institute on Aging–sponsored Established Populations for Epidemiological Study of the Elderly (EPESE)–Iowa site recruited a large cohort of community-dwelling elderly people (≥65 years at baseline). The cohort was followed up for 20 years after DNA collection, by which time almost all cohort members had died (ie, to the virtual “extinction” of the cohort). The EPESE-Iowa, therefore, provides an exceptionally good opportunity to estimate the associations between 9p21 SNP rs1333049 and disease-specific mortality and its interaction with advancing age.

Methods

Cohort and Study Sample

The EPESE-Iowa¹⁵ site interviewed a large population representative sample of community living elderly (≥65 years), starting in 1981. Study participants were followed up with repeat interviews every year for 10 years. Six follow-up interviews were conducted between 1987 and 1989 (when participants were ≥71 years). Of the 1987–1988 interviewees, 1752 (76%) consented to blood sample collection. Of these individuals, DNA extraction for genotyping was successful for 1708 subjects.

Ascertainment of Death

The survival status of each participant of the EPESE-Iowa was examined by obtaining death certificates and searching the National Death Index. The immediate and underlying causes of death were obtained using the International Classification of Diseases, 9^th Revision, Clinical Modification (ICD-9-CM) codes. For deaths that occurred after 1998, the International Classification of Diseases, 10^th Revision codes were used, which were converted to ICD-9-CM codes using general equivalence mappings from the US National Center for Health Statistics. In cases in which an equivalency was not provided, the nearest possible corresponding code between ICD-9-CM and International Classification of Diseases, 10^th Revision, was used. The underlying causes with codes 410, 4109, 413, 4140, 4141, 4148, and 4149 were used to ascertain the CAD-specific mortality. The ICD-9-CM codes 430 to 438 were used to ascertain deaths due to strokes.

Genotyping

DNA samples from the EPESE-Iowa were genotyped for rs1333049 using conventional Taqman PCR (probes and assays designed by Applied Biosystems; Foster City, CA). The SNP was in Hardy-Weinberg equilibrium (P>0.05); in duplicate samples tested, all were concordant. The C allele of the SNP rs1333049 is the risk allele, whereas the G allele is protective against CAD.

Covariates

Demographic characteristics, such as age and sex, were controlled for in the models. Genotype was included in the models as a risk allele count (classified as follows: 0, GG; 1, GC; and 2, CC). The 7 traditional risk factors of CAD included the following: smoking, body mass index, blood pressure (BP; both systolic and diastolic BP measures), blood lipid levels (serum triglyceride and cholesterol measures), and diabetes mellitus. These factors were controlled for in the analysis to estimate the independent effect of the SNP.

Smoking status was classified into 3 categories: never smokers, ex-smokers, and current smokers.

The body mass index was calculated in kilograms per meter squared from self-reported height and weight provided at the sixth follow-up and grouped into 4 ordered categories: <20 kg/m², underweight; 20 to 25 kg/m², normal weight; 25 to 30 kg/m², overweight; and >30 kg/m², obese. Systolic and diastolic BPs were measured during the sixth follow-up interview, and values were divided into ordered categories, per the international classification of hypertension. For systolic BP, the categories were as follows: <119, 120 to 139, 140 to 159, and >160 mm Hg. The categories for diastolic BP were as follows: <79, 80 to 89, and >90 mm Hg.

Blood lipid samples were collected during the sixth follow-up with the sample for DNA. The 2 relevant lipid measures included in the models were serum cholesterol and triglycerides, elevated levels of which are associated with a higher risk of CAD. These 2 measures were divided into quartiles without the extreme outliers (4 SDs from the mean). Physician-diagnosed reported history of diabetes was included in the model as a categorical variable (ever had, suspected of having, and never had).

Statistical Analysis

Initially, the baseline characteristics of the EPESE-Iowa participants included in the study were tabulated across the 3 genotypes of the SNP. Time to death and unadjusted disease-specific death rates were calculated for these 3 genotypes, and differences between them were tested by χ² tests.

Cox proportional hazard regression modeling was used to estimate the survival hazard associated with each risk (C) allele of rs1333049. The outcome variable in the analysis was time to death of the study participants as the result of all causes, plus disease-specific causes such as CAD and stroke. The initial models were controlled for age and sex. The traditional risk factors for CAD were then added to provide fully adjusted models. The interaction between rs1333049 and age was tested in all fully adjusted Cox models given the a priori hypothesis that age¹¹ is a 9p21 effect modifier. Age-stratified Kaplan–Meier survival estimation was initially performed to analyze the CAD mortality risks across 3 genotypes in age groups 71 to 80 and >80 years. The inequality in survival function across 3 geno-types in these age groups was tested by log-rank test. Subsequently, fully adjusted Cox regression analysis was performed in 3 age groups (71–75, 76–80, and >80 years) to estimate the age-related gradient in CAD mortality risk associated with the SNP.

Stata, version 10.1 (StataCorp; College Station, TX), was used to analyze the data.

Missing Values and Exclusion Criteria

The last death record was collected in September 2008. At that time, death certificates could not be obtained for 135 (8%) of the 1708 genotyped participants. There is no follow-up information available for these 135 participants after DNA sample collection in 1988. Consequently, they could not be included in the analysis as censored observations and had to be excluded from the analysis as truncated data.

Prevalent cases of CAD were those who reported a physician-diagnosed or suspected episode of MI and/or angina (identified from the Rose Questionnaire on angina). We excluded prevalent cases of CAD to remove potential “response” bias at analysis baseline, from healthier survivors of MI and/or unstable angina who were more likely to be recruited into the study. There were 211 prevalent MI cases with or without a history of angina and 267 participants who reported angina without MI. These 478 observations were excluded from the analysis. All the remaining 1095 study participants were included in the analysis.

The missing values and extreme outliers for all the covariates were coded as separate “missing” categories and were included in analyses.

Predicting Risk of CAD Attributable Death by Framingham Score Plus 9p21 Variation

We performed a post hoc CAD risk classification analysis with people aged 71 to 80 years at enrollment. The study sample was classified into those dying of CAD within 10 years of follow-up or not. The classic Framingham Risk Score (FRS) for CAD within 10 years, comprising sex-specific weighted scores for age, smoking status, cholesterol, high-density lipoprotein cholesterol, and systolic BP, was calculated for each participant.¹⁶ Then, the FRS was entered into a Cox regression model with CAD deaths as the outcome event. The linear predictor (xβ) from the FRS-only Cox model was calculated for every participant and divided into tertiles (low, intermediate, and high risk). Then, the binary variable (dying of CAD within 10 years or not) was cross tabulated against the tertiles of the linear predictor from the FRS-only model.

The Cox modeling was repeated with the 9p21 SNP allele count added to the FRS (FRS+SNP Cox model). The linear predictor from the FRS+SNP Cox model was then calculated for every participant, and risk predictors were then divided into 3 groups of similar participant numbers to the FRS-only model. Then, the cross tabulation with CAD deaths within 10 years was repeated with the 3 groups of linear predictors from the FRS+SNP model.

Results

Of the included 1095 MI or angina-free study participants of EPESE-Iowa at baseline, 570 (52%) were heterozygous (GC) and 247 (23%) were homozygous (CC) for the deleterious C allele of rs1333049. These subjects were followed up for 10 218 person-years. A total of 259 deaths were attributed to CAD, and 117 deaths were attributable to stroke. Table 1 shows the distribution of the baseline demographic and other traditional CAD risk factors in the genotyped groups. The distribution of the demographic characteristics, such as age and sex, and traditional risk factors of CAD, such as smoking status, history of diabetes, systolic and diastolic BP levels, body mass index, and lipid levels, were similar across the 3 genotypes. There was no significant association of the variants with any of these risk factors. The C allele group experienced significantly higher unadjusted CAD mortality rates (22/1000 person-years in the GG group vs 30/1000 person-years in the CC group; P=0.01), but no differences in mortality rates specific to stroke were observed across the genotypes (Table 1).

Table 1. Baseline Characteristics Across 3 Genotypes of rs1333049.

Characteristic	GG (n = 278)	GC (n = 570)	CC (n = 247)
Age, median (interquartile range), y	79 (9)	80 (9)	79 (8)
Sex
Male	97 (35)	191 (34)	76 (31)
Female	181 (65)	379 (66)	171 (69)
Smoking status
Current smoker	24 (9)	41 (7)	22 (9)
Ex-smoker	49 (18)	95 (17)	34 (14)
Never smoker	203 (73)	432 (76)	190 (77)
Diabetes history
Ever had	21 (8)	43 (8)	18 (7)
Suspected	10 (4)	7 (1)	2 (1)
Never had	247 (89)	519 (91)	227 (92)
Systolic BP, mm Hg
<120	35 (13)	53 (10)	27 (12)
120–139	89 (34)	210 (40)	79 (34)
140–159	95 (36)	168 (32)	89 (38)
>160	44 (17)	90 (17)	37 (16)
Diastolic BP, mm Hg
<80	185 (70)	369 (71)	167 (72)
80–89	53 (20)	117 (22)	50 (22)
90–99	25 (10)	35 (7)	15 (6)
Body mass index, kg/m2
<20	20 (8)	40 (8)	15 (7)
20–24.9	91 (36)	206 (41)	88 (39)
25–29.9	95 (38)	174 (35)	84 (37)
>30	44 (18)	84 (17)	37 (17)
Cholesterol quartile
1	55 (20)	163 (29)	66 (27)
2	66 (24)	129 (23)	67 (27)
3	84 (30)	141 (25)	49 (20)
4	71 (26)	136 (24)	63 (26)
Triglycerides quartile
1	68 (24)	157 (28)	69 (28)
2	73 (27)	143 (25)	63 (26)
3	67 (24)	130 (23)	55 (23)
4	66 (24)	134 (24)	56 (23)
CAD-specific death rate/1000 person-years*	22	25	30
Stroke-specific death rate/1000 person-years	13	11	12

Data are given as number (percentage) of each group unless otherwise indicated. Of the 1095 participants, the GG group composed 25%; GC group, 52%; and CC group, 23%. The following numbers of observations had missing values or extreme outlying values: systolic BP (n=79), diastolic BP (n=79), body mass index (n=17), cholesterol (n=5), triglycerides (n=14), smoking status (n=5), and diabetes (n=1). These observations were categorized as “missing” for the respective variables and included in the analysis.

^*Statistically significant, P<0.01.

The fully adjusted Cox model estimated a hazard ratio (HR) of 1.19 per risk (C) allele (95% CI, 1.08–1.30; P<0.01) for deaths from all causes. The HR was 1.29 per C allele (95% CI, 1.08–1.56; P<0.01) for CAD-specific mortality (Table 2). The risk allele showed no significant association with deaths due to stroke (HR, 1.07; 95% CI, 0.81–1.41; P=0.63). An interaction term for SNP status and age was statistically significant for CAD deaths (P=0.05), indicating a need for a further age-stratified survival analysis (71–80 vs >80 years). There was no significant interaction detected between age and the genetic variant in models with all-cause or stroke mortality.

Table 2. The HRs Associated With the C Allele of rs1333049 for 1095 Participants.

	All-Cause Mortality		CAD Mortality
Risk Factors	HR (95% CI)	P Value	HR (95% CI)	P Value
SNP rs1333049	1.19 (1.08 –1.30)	<0.01	1.29 (1.08 –1.56)	<0.01
Sex
Male	1		1
Female	0.75 (0.56–0.93)	<0.001	0.72 (0.52–0.98)	0.04
Smoking
Never smoker	1		1
Current smoker	1.60 (1.32–1.95)	<0.001	1.43 (0.98–2.46)	0.20
Ex-smoker	1.34 (1.16–1.55)	<0.001	1.42 (0.98–2.06)	0.06
Diabetes mellitus
Diagnosed	1		1
Suspected	0.74 (0.49–1.11)	0.14	0.46 (0.16–1.34)	0.16
None	0.76 (0.63–0.93)	<0.01	0.48 (0.32–0.72)	<0.01
Systolic BP, mm Hg
<120	1		1
120–139	0.99 (0.85–1.15)	0.87	0.93 (0.62–1.40)	0.73
140–159	1.16 (0.98–1.38)	0.09	1.24 (0.80–1.92)	0.34
>160	1.56 (1.25–1.95)	<0.01	1.64 (0.96–2.82)	0.07
Diastolic BP, mm Hg
<80	1		1
80–89	0.88 (0.77–1.02)	0.09	0.97 (0.54–1.77)	0.92
>90	1.06 (0.84–1.33)	0.61	1.23 (0.88–1.71)	0.22
Body mass index, kg/m2
20–24.9	1		1
<20	0.96 (0.73–1.26)	0.78	0.90 (0.43–1.88)	0.79
25–29.9	1.1 (0.97–1.24)	0.14	1.16 (0.85–1.57)	0.35
>30	1.11 (0.93–1.32)	0.26	1.56 (0.99–2.46)	0.06
Cholesterol quartile
1	1		1
2	0.90 (0.78–1.04)	0.15	1.08 (0.75–1.56)	0.66
3	0.86 (0.74–1.00)	0.05	1.02 (0.70–1.49)	0.91
4	0.82 (0.70–0.95)	0.01	1.02 (0.69–1.51)	0.91
Triglycerides quartile
1	1		1
2	0.92 (0.80–1.06)	0.24	1.04 (0.73–1.48)	0.83
3	0.99 (0.86–1.15)	0.93	1.14 (0.79–1.66)	0.48
4	1.15 (0.99–1.33)	0.06	1.55 (1.08–2.22)	0.02

The unadjusted Kaplan–Meier survival estimation graphs (Figure 1) showed significantly lower survival probability in the CC and GC genotypes compared with GG. This association was present only in the age group 71 to 80 years (log-rank test P=0.01) and was attenuated in the subjects aged ≥81 years. The age-stratified Cox models (Figure 2) estimated significant association between rs1333049 and CAD mortality for the age groups 71 to 75 years (HR, 1.59; 95% CI, 1.07–2.36; P=0.02) and 76 to 80 years (HR, 1.52; 95% CI, 1.09–2.12; P=0.01). The effect of the SNP for people aged >80 years was attenuated (HR, 1.02; 95% CI, 0.76–1.36; P=0.91).

Figure 1.

Kaplan–Meier survival estimation (for coronary artery disease [CAD]–specific mortality) across genotypes of rs1333049 in different age groups.

Figure 2.

Age group–specific hazard ratio (HR) per C allele of rs1333049.

Risk Classification

We compared the capacity of the 9p21 SNP and the FRS for CAD within 10 years to classify older people (aged 71–80 years at enrollment) for the risk of dying of CAD. Table 3 shows that the linear predictor from the FRS-only model classified as high risk 17 of 81 participants who died of CAD within 10 years of follow-up. The linear predictors with the 9p21 SNP allele count added to the FRS (FRS+SNP model) identified an additional 6% (n=5) of the 81 CAD deaths within 10 years of follow-up in the high-risk group (Table 3).

Table 3. Reclassification of CAD Death Risk.

Variable	Low Risk	Intermediate Risk	High Risk
FRS	30	34	17
FRS+SNP	29	30	22

Data are for 81 CAD deaths within 10 years of follow-up. Of the 81, 5 (6.2%) were reclassified into high risk.

Post Hoc Analyses

As an additional analysis, we examined associations between risk C allele count and mortality in the 478 participants who had evidence of CAD at baseline. The per C allele HR was 1.10 (95% CI, 0.96–1.25; P=0.16) for all-cause mortality and 1.08 (95% CI, 0.84–1.37; P=0.56) for CAD-attributed mortality. To explore the association with overall mortality, we estimated 9p21 association with deaths attributed to any cancer (n=105): no associations were found either with (HR, 1.24; 95% CI, 0.93–1.66; P=0.14) or without (HR, 1.22; 95% CI, 0.93–1.62; P=0.149) adjustment for coronary artery risk factors.

Discussion

It is well established that the genetic variant in chromosome 9p21 is strongly associated with CAD,^14,17,18 especially in midlife. Although this disease association has been intensively studied, there have been few data on the effect of this variant on mortality, especially at older ages. In our analysis, we showed, for the first time to our knowledge, that the 9p21 variant is associated with increased risk of CAD-attributed mortality in later life. We found substantially increased CAD-attributable mortality risks in respondents aged 71 to 80 years and CAD free at enrollment.

Ellis¹⁹ and Meng¹² and colleagues have reported the 9p21 locus to be associated with an earlier onset of CAD. The same study by Ellis et al also showed no increase in short-term mortality associated with rs1333049 in 2 coronary disease cohorts. Our analysis shows that this variant can predict mortality in old age in those free of CAD at baseline. Interestingly, when we examined 9p21 associations with mortality in the 478 people who had CAD diagnoses or symptoms at baseline, there was only a modest nonsignificant trend, suggesting that these participants represent a “healthy survivor” CAD group. Case-control studies of the 9p21 variant often included such survivors, which might explain the mixed findings on survival from other studies. We did find that the predictive capacity of 9p21 was attenuated in late life (aged >80 years). This finding was somewhat similar to observations made by Zdravkovic et al,¹¹ who suggested that heritable factors for CAD are in operation throughout an individual’s lifespan, but with evidence that the effect decreases with advancing age.

The CAD mortality risk associated with rs1333049 in old age in the EPESE-Iowa cohort members is independent of the traditional risk factors of CAD, such as smoking, obesity, BP, blood lipids, and history of diabetes. This independence has been shown for disease presence in younger cohorts,^3,18 but our study demonstrates that a similar independent effect applies to mortality at older ages. A recent study⁷ has shown that genetic variants in 9p21 interfere with the interferon-γ signaling pathway and affect the repair response of vascular endothelial cells to inflammation, thus accelerating the pathological process of atherosclerosis leading to CAD. Ye et al¹⁸ observed an association of the variant with incident CAD disease events in a population-representative prospective cohort, which led them to conclude that this SNP affects both development and progress of atherosclerosis.

After having established that the 9p21 variant is associated with a mortality hazard independent of traditional risk factors, we next explored the effect of adding this variant to the FRS. In this study, the SNP reclassified 6% of CAD deaths in the age group 71 to 80 years from intermediate- to high-risk groups when added to the widely used FRS for CAD. This apparently modest risk reclassification capacity of 9p21 genetic variants was also observed by others,^20–22 when applied to incident episodes of CAD, but its capacity to classify risk for CAD deaths, especially among elderly individuals, has not been previously studied. The traditional risk factors for CAD included in the FRS have a reduced predictive capacity²³ in older populations, thus perhaps increasing the importance of the 9p21 genetic variant. Given the high CAD death rates among older people in the general population, a 6% risk reclassification capacity could identify many individuals who might benefit from more aggressive interventions, such as intensive therapy with statins,²⁴ that have been efficacious, even in old age.^25,26 Clearly, the CAD mortality risk predictive ability of 9p21 in our study needs to be validated in larger cohorts and further characterized before predictive genetic testing for CAD-related mortality could be suggested as a routine procedure.²⁷

The lack of association of rs1333049 with stroke deaths among EPESE-Iowa participants contradicts some previous findings^28,29 of increased risk of stroke and validates others³⁰ who have found no association. This inconsistency in association between 9p21 genetic variants and ischemic stroke across studies was identified by Anderson et al.³¹

In evaluating these results, it should be noted that this study was conducted on a relatively homogeneous population of rural Americans and almost all were of white European ancestry.

The blood collection concentrated on fitter subjects able to attend the study center, resulting in a 24% nonresponse: this may reduce the representativeness of the results, although our focus on hazards in those free of prevalent CAD will have minimized this effect.

Some misclassification of CAD deaths in death certificates of EPESE-Iowa is expected because similar overestimation of CAD deaths to a variable degree was also observed in the Atherosclerosis Risk in Community Study³² when the death data were collected from death certificates: the overestimation showed a decline with advancing age. Because the misclassification is expected to affect all the groups nondifferentially, irrespective of their SNP status, such misclassification is likely to underestimate the actual effect size³³ in our study. Many of the measures of health used in the study, including prevalent medical conditions, weight, and height, were self-reported by the participants, thus introducing some misclassification into the measures.

Conclusion

The MI genetic variant at 9p21 is independently associated with survival in an older population free of prevalent CAD at baseline and followed up for 20 years, to virtual extinction. This excess mortality was predominantly the result of CAD-attributed deaths, and the CAD mortality risk associated with 9p21 was present up to the age of 80 years. Adding 9p21 variation to the FRS may provide small, but potentially useful, gains in predicting those at high 10-year CAD mortality risk, in older populations provided the results of this study can be validated in an independent and larger cohort of elderly people.

CLINICAL PERSPECTIVE.

The single-nucleotide polymorphisms (SNPs) at chromosome 9p21 emerged as the inherited genetic variant with the largest effect that is associated with coronary artery disease (CAD) and myocardial infarction in middle-aged groups. Some have speculated that the effect dwindles in later life. We examined the SNP effect on cardiovascular mortality in a population-based cohort aged >70 years with no prior cardiovascular disease diagnoses and followed up for 20 years, by which time nearly 95% had died. Those with the 9p21 risk variant were significantly more likely to die with a cardiovascular diagnosis: the hazard ratio was ≈50% higher for the 25% of the population with 2 risk alleles compared with those with 0 risk alleles. We found that this effect was reduced for those aged >80 years. Adding the 9p21 variant to the risk prediction formula in middle age has little impact because conventional risk factors provide efficient prediction. In older groups, several traditional risk factors for CAD lose their predictive capacity, so could the 9p21 variant help? Modeling 9p21 into the Framingham Risk Score for CAD resulted in 6% more of those who died within 10 years being classified to the highest-risk group, mainly from the middle-risk band. This improved prediction warrants further evaluation given the high incidence rates and effectiveness of intervention in later life.