Atherosclerosis 233 (2014) 441e446
Contents lists available at ScienceDirect
Atherosclerosis journal homepage: www.elsevier.com/locate/atherosclerosis
The effect of aging on the association between coronary heart disease risk factors and carotid intima media thickness: An analysis of the Atherosclerosis Risk in Communities (ARIC) cohort Seema Pursnani*, Marie Diener-West, A. Richey Sharrett Johns Hopkins Bloomberg School of Public Health, Room W6009B, 615 North Wolfe Street, Baltimore, MD 21205, USA
a r t i c l e i n f o
a b s t r a c t
Article history: Received 10 April 2012 Received in revised form 12 December 2013 Accepted 27 December 2013 Available online 28 January 2014
Objective: Aging decreases the strength of association between established coronary heart disease (CHD) and its risk factors. Carotid intima media thickness (IMT) is a widely used surrogate for coronary artery disease, which we hypothesized has a similar diminishing age-mediated strength of association with CHD occurrence and prevalence of its risk factors. Methods and results: Data from the Atherosclerosis Risk in Communities (ARIC) cohort of 14,562 individuals aged 45 to 64 (mean follow up nine years) was stratified into two age groups, 45e54 and 55 e64 years, within each of ARIC’s four examination visits (n ¼ 14,562; 13,622; 7869; 6628 for visits 1 to 4, respectively). Cross-sectional and longitudinal analyses with multiple linear and logistic regression modeling were used to compare the relationships between carotid IMT (the mean of six far wall sites from the right and left carotid bifurcation, common and internal carotid arteries with imputation of missing data) and the risk factors of smoking, hypertension, hypercholesterolemia, diabetes, and obesity with age. The strength of the associations between carotid IMT and most risk factors were qualitatively stronger across successive visits and within each visit, these associations were stronger in the older, as compared to the younger, age group. Conclusions: In a large cohort followed for nearly one decade, our hypothesis that age attenuates the association of CHD risk factors and carotid IMT was not supported by ARIC data. Rather, we found that associations between carotid IMT and CHD risk factors remained stable with advancing age, contrary to the relationship between risk factors and CHD outcomes with age. These findings suggest that there is efficacy to continued risk factor management in the elderly. Ó 2014 Elsevier Ireland Ltd. All rights reserved.
Keywords: Carotid intima media thickness Coronary heart disease Obesity Diabetes mellitus Hypercholesterolemia Hypertension Smoking
1. Background Despite advances in understanding and controlling risk factors, coronary heart disease (CHD) remains the single leading cause of death in the elderly in America [1]. With a growing aging population, it is important to determine if there is a crucial age window during which to focus preventive health efforts, and if risk factor modification is as beneficial in the later decades of life as it is during middle age. B-mode Doppler ultrasonography of the carotid arteries is used in large population studies because it is noninvasive, inexpensive and allows serial studies. Several studies [2e4] have shown that carotid intima media thickness (IMT) is strongly associated with
* Corresponding author. Tel.: þ1 415 600 5966; fax: þ1 415 563 5939. E-mail addresses: [email protected] (S. Pursnani), [email protected] (M. Diener-West), [email protected] (A.R. Sharrett). 0021-9150/$ e see front matter Ó 2014 Elsevier Ireland Ltd. All rights reserved. http://dx.doi.org/10.1016/j.atherosclerosis.2013.12.046
CHD incidence. A meta-analysis by Lorenz et al. [5] demonstrated a hazard ratio for CHD (adjusted for age, sex and CHD risk factors) of 1.10, with a 10% increase in CHD risk with each .1 mm difference in carotid IMT. While the risk factors for carotid atherosclerosis are shared with CHD [6e9], the trend in the strength of these risk factor associations with carotid atherosclerosis and CHD across increasing age groups may differ. The Multiple Risk Factor Intervention Trial (MRFIT) showed a marked decrease in strength of association between these risk factors and CHD and all-cause mortality with increasing age; for example, among nonsmoking men aged 35e45, the relative risk of CHD death for the highest to lowest diastolic blood pressure quintile was 8.3 whereas for men age 46e57, this same relative risk was 3.9 [10]. These trends were similar for other CHD risk factors, with the most striking being a relative risk of CHD death at the highest cholesterol quintile at 8.2 for the age group 35e39 falling to 2.4 for the age group 55e57 [10]. The Honolulu Heart Program [11] showed a similar decreasing association at
442
S. Pursnani et al. / Atherosclerosis 233 (2014) 441e446
older ages between CHD and hypertension, hypercholesterolemia, smoking, and BMI with increasing age, and the Framingham Heart Study showed a decreasing strength of association between CHD and hypercholesterolemia and smoking [12]. These results suggest that preventive efforts to control CHD risk factors are more effective when implemented at early ages, and is reflected in clinical risk scoring systems in which the presence of CHD risk factors at earlier ages are given increased weight [13,14]. Contrary to these findings, an analysis of baseline data combined from the ARIC and CHS cohorts showed a stable strength of association with age between these risk factors and carotid IMT, paradoxically suggesting that risk factor modification might be equally effective in older and younger individuals [15]. The purpose of this study is to clarify the age-dependency of the associations between CHD risk factors and carotid IMT using data from the ARIC cohort. With its 20 year age range of individuals at entry and nine year follow up with serial carotid IMT measurements, the cohort allows for two age comparisonsdcomparing younger and older age groups based upon age at each exam and comparing individuals within each of the two age groups at each successive visit. We studied smoking, hypertension, hypercholesterolemia, diabetes, and obesity. Given the known reduction in relative risk of CHD for these variables associated with aging, we hypothesized a similarly decreased strength of association between these risk factors and carotid IMT with aging. We hypothesized that at each visit, the associations between elevated levels of each of the risk factors and both outcomes, the difference in mean IMT and the odds ratio for highest quintile (as compared to the aggregate of the four remaining quintiles) carotid IMT, would be smaller in magnitude for the older than younger age groups, and that within each of the two age groups (age at entry 45e54 versus 55e64), these differences in mean IMT and odds ratio for highest quintile carotid IMT values would be reduced in magnitude with aging across the four visits. 2. Methods 2.1. Subjects and study design ARIC is a study of the natural history and risk factors for atherosclerosis and CHD initially (1987e89) enrolling 15,792 men and women age 45 to 64 from four diverse United States communities in North Carolina, Mississippi, Maryland, and Minnesota. After the first examination, surviving subjects were examined every three years for a total of four visits (visit 1: 1987e89, visit 2: 1990e92, visit 3: 1993e95, visit 4: 1996e98). The attrition rates, calculated as the proportion of visit 1 examinees alive at the time of the exam, were as follows: exam 2, 93%; exam 3, 86%; and exam 4, 80%. All were offered carotid ultrasound imaging at visits 1 and 2. African-American participants were invited for follow-up carotid imaging at both third and fourth visits, and the remaining subjects had one additional examination at either the third or fourth visit. The dataset used for this study was the de-identified Limited Access dataset available from the National Heart, Lung, and Blood Institute of the National Institute of Health. The data retrieved included carotid IMT, age, sex, race, LDL cholesterol, body mass index (BMI), blood pressure, smoking status, diabetes status, prevalent CHD, and medication use from each visit, and demographic variables from the initial visit. Details of cohort selection and variable measurements were described previously [16]. The carotid IMT measurements were conducted using B mode real-time ultrasound imaging, based upon a method validated by Pignoli et al., who showed their correspondence with measurements from pathologic common carotid and aortic wall specimens [17]. Carotid IMT far wall thickness was calculated as the mean of
six far wall sites 1 cm long taken from the right and left carotid bifurcation, common and internal carotid arteries. As described in prior ARIC publications, we used derived variables adjusted for measurement drift over the visit and systematic differences among readers [4]. Missing values in the dataset were previously imputed based on multivariable linear models, which related IMT measurements at each carotid site to each other, accounting for sex, race, age, body mass index, and imaging depth. We additionally conducted a sensitivity analysis using nonimputed data, where we related an average of available right and left common carotid measurements to the risk factors. For this study, risk factors were categorized as present or absent based on current clinical guidelines for treatment thresholds. Smoking was defined as self-report of current cigarette smoking. Hypertension was defined as a systolic blood pressure greater than or equal to 140 mm Hg or diastolic blood pressure greater than or equal to 90 mm Hg or the use of an anti-hypertensive agent in the preceding two weeks. Hypercholesterolemia was defined as either an LDL measurement of greater than or equal to 130 mg/dl or the use of lipid lowering agents in the preceding two weeks. Diabetes was defined as a fasting blood glucose greater than or equal to 126 mg/dl or the use of an oral hypoglycemic agent or insulin the preceding two weeks. Obesity was defined as a BMI greater than or equal to 30 kg/m2. Prevalent CHD was defined as myocardial infarction (MI) adjudicated from the visit 1 electrocardiogram, a documented history of MI, prior revascularization with coronary artery bypass grafting or percutaneous coronary intervention. Medication use was defined as self reported use of lipid lowering or blood pressure lowering medication in the two weeks preceding examination. All risk factors were defined based on laboratory results or interviews at each of the four visits. 2.2. Statistical analysis Cross sectional analyses were performed using multiple linear regression models, calculating the regression coefficients for each of the five binary risk factors in relation to the outcome of carotid IMT, treated as a continuous variable: Average carotid IMT (in mm) ¼ b0 þ b1(gender) þ b2(race) þ b3(visit-age group) þ b4 (smoking) þ b5(hypertension) þ b6(hypercholesterolemia) þ b7(diabetes) þ b8(obesity). Age group at each visit was defined as follows: Visit 1: age 45e54 and 55e64, Visit 2: age 48e57 and 58e 67, Visit 3: age 51e60 and 61e70, Visit 4: age 54e63 and 64e73. Assumptions for the use of linear regression modeling were validated by means of the Shapiro Wilk test for normality of the residuals and assessment of influential points. Though residual estimates were found to be non-normal, regression analysis was deemed appropriate given the large size of the dataset and graphical demonstration of near normality. Collinearity was evaluated with calculation of variance inflation factors, which were all less than a value of 1.2. The lack of independence of repeated carotid IMT measurements was accounted for in the longitudinal analysis (see below). We also employed logistic regression analyses to determine the odds of a measure in the top carotid IMT quintile, based upon the distribution within each age group at each visit, in relation to the presence or absence of the five risk factors. The model was as follows: log odds (carotid IMT in top quintile) ¼ b0 þ b1(gender) þ b2(race) þ b3(visit-age group) þ b4(smoking) þ b5(hypertension) þ b6(hypercholesterolemia) þ b7(diabetes) þ b8(obesity). The HosmereLemeshow goodness of fit test was used to evaluate this model. To determine the statistical significance of age as an effect modifier, two models were developed. The first model addressed the interaction between the age group and each of the five risk
S. Pursnani et al. / Atherosclerosis 233 (2014) 441e446
factors using multiple linear regression with inclusion of interaction terms at each of the four visits. The interaction terms consist of age group as a binary variable and the visit-specific risk factor also coded as a binary variable. Therefore, for each of the four visits, the model was defined as follows: Average carotid IMT ¼ b0 þ b1(gender) þ b2(race) þ b3(age group) þ b4(smoking) þ b5(hypertension) þ b6(hypercholesterolemia) þ b7(diabetes) þ b8(obesity) þ b9(smoking*age group) þ b10(hypertension*age group) þ b11(hypercholesterolemia*age group) þ b12(diabetes*age group) þ b13(obesity*age group). The hypertension*age group interaction term therefore refers to the difference between the older and younger age groups in the change in average carotid IMT between those with versus without hypertension at the specified visit. The second model addressed the interaction within each age group across the four visits. Multiple linear regression models with robust variance estimation were used to account for individual clustering of carotid IMT measurements across visits. For each of the two age groups, the model was as follows: Average carotid IMT ¼ b0 þ b1(gender) þ b2(race) þ b3(visit) þ b4(smoking) þ b5(hypertension) þ b6(hypercholesterolemia) þ b7(diabetes) þ b8(obesity) þ b9(smoking*visit) þ b10(hypertension*visit) þ b11(hypercholesterolemia*visit) þ b12(diabetes*visit) þ b13(obesity*visit), where visit was coded as an indicator variable (with visit 1 omitted and the three other visits compared to the baseline visit) and each risk factor as the visitspecific binary variable, such that the interpretation of the hypertension*visit 4 variable, for example, reflects the difference between individuals at visit 4 and at visit 1 in the change in average carotid IMT between those with versus without hypertension for the specified age category. Adjustment for prevalent CHD and medication use were performed in sensitivity analyses. All statistical analyses and graphics were performed using StataIC 11 software. 3. Results Table 1 shows characteristics of the ARIC cohort stratified by age at entry and the four visits. At the initial visit, carotid ultrasound data were available for 7662 and 6900 individuals in the younger and older age strata, respectively. As expected, carotid IMT values increased with agedas seen both within each age stratum across visits and between the two strata within visits. The coefficients of determination R2 were 21%, 18%, 14%, and 17% for the models at visit 1, 2, 3, and 4, respectively, consistent with previously published risk factor predictions of IMT variability. Partial R2 for the individual predictors were as follows: diabetes 1.1%, smoking .6%, hypercholesterolemia 1.2%, hypertension 2.7%, and obesity .01%. It should be noted that age itself contributes to much of the explained remaining variability in IMT prediction. 3.1. Smoking The percentage of current smokers was greater for the younger, as compared to the older, age group at each visit; and in each age group, it declined across visits. The regression coefficient for visit 1 indicates that average carotid IMT was .032 mm larger (95%CI: .026e.038) in current smokers than in non-smokers in the younger age group (Table 2). Similarly, average carotid IMT was .064 mm larger (95% CI: .052e.076) in current smokers than in non-smokers in the older age group. The association between IMT and current smoking was consistently stronger in the older, as compared to the younger, age group (p < 0.05 for smoking*age group interaction terms at visits 1 and 2, Table 2). Across visits, there was a relatively stable strength of association across the first three visits in each age group (Fig. 1), and in the younger group there was a statistically
443
Table 1 Cohort characteristics by age group and visit. Age at entry
Sex: number (%) male Race: number (%) black Total na
Mean carotid IMT in mm (sd)
Prevalent CHD
Visit Visit Visit Visit Visit Visit Visit Visit Visit Visit Visit Visit
Number with risk factor (% of total): Smoking Visit Visit Visit Visit Hypertension Visit Visit Visit Visit Hypercholesterolemia Visit Visit Visit Visit Diabetes Visit Visit Visit Visit Obesity Visit Visit Visit Visit Medication use in the last 2 weeks: Anti-hypertensive medication Visit Visit Visit Visit Lipid-lowering medication Visit Visit Visit Visit
45e54
55e64
1 2 3 4 1 2 3 4 1 2 3 4
3463 (41.8) 2454 (29.6) 7662 7290 4287 3621 .676 (.146) .695 (.150) .727 (.178) .755 (.203) 232 (3.0) 517 (7.1) 373 (8.7) 407 (11.2)
3592 (48.3) 1802 (24.2) 6900 6332 3582 3007 .782 (.207) .798 (.219) .832 (.252) .877 (.272) 533 (7.7) 438 (6.9) 337 (9.4) 324 (10.8)
1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4
2173 (28.4) 1784 (24.5) 930 (21.7) 640 (17.7) 2083 (27.2) 2114 (29.0) 1583 (36.9) 1419 (39.2) 3800 (49.6) 3599 (49.4) 1926 (44.9) 1682 (46.5) 662 (8.6) 877 (12.0) 612 (14.3) 517 (14.3) 2003 (26.1) 2127 (29.2) 1464 (34.1) 1266 (35.0)
1663 (24.1) 1276 (20.2) 591 (16.5) 377 (12.5) 2869 (41.6) 2672 (42.2) 1762 (49.2) 1645 (54.7) 4199 (60.9) 3583 (56.6) 1793 (50.1) 1524 (50.7) 966 (14.0) 1098 (17.3) 648 (18.1) 509 (16.9) 1786 (25.9) 1662 (26.2) 1081 (30.2) 923 (30.7)
1 2 3 4 1 2 3 4
1677 (21.9) 1854 (25.4) 1749 (40.8) 1649 (45.5) 452 (5.9) 438 (6.0) 450 (10.5) 394 (10.9)
2314 (33.5) 1668 (26.3) 1520 (42.4) 1420 (47.2) 615 (8.9) 580 (9.2) 566 (15.8) 536 (17.8)
Abbreviations: sd, standard deviation, IMT ¼ intima media thickness, CHD ¼ coronary heart disease. a Total n refers to number of individuals for whom carotid IMT data were available at each visit.
significant increased strength of association comparing the fourth to the first visits. The upper quintile logistic analysis shows similar trends in the odds ratios between age groups and across visits, i.e. stronger associations in the older than the younger group at visits 1 and 2, and stronger associations at visit 4 than visit 1 (Table 3). The age trends seen for smoking (Fig. 1) were generally similar to those seen for hypertension, hypercholesterolemia, and diabetes. 3.2. Hypertension The percentage of individuals with hypertension increased, as expected, with age both across visits and between age strata (Table 1). The difference in average IMT associated with hypertension was greater in the older, as compared to the younger, age group (p < 0.05 for visit 1 and 2, Table 2). Trends across visits within each age group were not strong, consistent or statistically significant. The logistic regression analyses (shown in the appendix) show slightly smaller odds of being in the top carotid IMT quintile for hypertensives in the younger age group than in the older group.
444
S. Pursnani et al. / Atherosclerosis 233 (2014) 441e446
Table 2 Risk factor associations with average carotid IMT: Adjusted regression coefficients and standard errors. Smoking Younger Visit Visit Visit Visit
1 .032 2 .036 3 .038 4 .065*
(.003) (.004) (.006) (.008)
Hypertension Older .064 .071 .051 .091
P value Younger
(.006)
Contents lists available at ScienceDirect
Atherosclerosis journal homepage: www.elsevier.com/locate/atherosclerosis
The effect of aging on the association between coronary heart disease risk factors and carotid intima media thickness: An analysis of the Atherosclerosis Risk in Communities (ARIC) cohort Seema Pursnani*, Marie Diener-West, A. Richey Sharrett Johns Hopkins Bloomberg School of Public Health, Room W6009B, 615 North Wolfe Street, Baltimore, MD 21205, USA
a r t i c l e i n f o
a b s t r a c t
Article history: Received 10 April 2012 Received in revised form 12 December 2013 Accepted 27 December 2013 Available online 28 January 2014
Objective: Aging decreases the strength of association between established coronary heart disease (CHD) and its risk factors. Carotid intima media thickness (IMT) is a widely used surrogate for coronary artery disease, which we hypothesized has a similar diminishing age-mediated strength of association with CHD occurrence and prevalence of its risk factors. Methods and results: Data from the Atherosclerosis Risk in Communities (ARIC) cohort of 14,562 individuals aged 45 to 64 (mean follow up nine years) was stratified into two age groups, 45e54 and 55 e64 years, within each of ARIC’s four examination visits (n ¼ 14,562; 13,622; 7869; 6628 for visits 1 to 4, respectively). Cross-sectional and longitudinal analyses with multiple linear and logistic regression modeling were used to compare the relationships between carotid IMT (the mean of six far wall sites from the right and left carotid bifurcation, common and internal carotid arteries with imputation of missing data) and the risk factors of smoking, hypertension, hypercholesterolemia, diabetes, and obesity with age. The strength of the associations between carotid IMT and most risk factors were qualitatively stronger across successive visits and within each visit, these associations were stronger in the older, as compared to the younger, age group. Conclusions: In a large cohort followed for nearly one decade, our hypothesis that age attenuates the association of CHD risk factors and carotid IMT was not supported by ARIC data. Rather, we found that associations between carotid IMT and CHD risk factors remained stable with advancing age, contrary to the relationship between risk factors and CHD outcomes with age. These findings suggest that there is efficacy to continued risk factor management in the elderly. Ó 2014 Elsevier Ireland Ltd. All rights reserved.
Keywords: Carotid intima media thickness Coronary heart disease Obesity Diabetes mellitus Hypercholesterolemia Hypertension Smoking
1. Background Despite advances in understanding and controlling risk factors, coronary heart disease (CHD) remains the single leading cause of death in the elderly in America [1]. With a growing aging population, it is important to determine if there is a crucial age window during which to focus preventive health efforts, and if risk factor modification is as beneficial in the later decades of life as it is during middle age. B-mode Doppler ultrasonography of the carotid arteries is used in large population studies because it is noninvasive, inexpensive and allows serial studies. Several studies [2e4] have shown that carotid intima media thickness (IMT) is strongly associated with
* Corresponding author. Tel.: þ1 415 600 5966; fax: þ1 415 563 5939. E-mail addresses: [email protected] (S. Pursnani), [email protected] (M. Diener-West), [email protected] (A.R. Sharrett). 0021-9150/$ e see front matter Ó 2014 Elsevier Ireland Ltd. All rights reserved. http://dx.doi.org/10.1016/j.atherosclerosis.2013.12.046
CHD incidence. A meta-analysis by Lorenz et al. [5] demonstrated a hazard ratio for CHD (adjusted for age, sex and CHD risk factors) of 1.10, with a 10% increase in CHD risk with each .1 mm difference in carotid IMT. While the risk factors for carotid atherosclerosis are shared with CHD [6e9], the trend in the strength of these risk factor associations with carotid atherosclerosis and CHD across increasing age groups may differ. The Multiple Risk Factor Intervention Trial (MRFIT) showed a marked decrease in strength of association between these risk factors and CHD and all-cause mortality with increasing age; for example, among nonsmoking men aged 35e45, the relative risk of CHD death for the highest to lowest diastolic blood pressure quintile was 8.3 whereas for men age 46e57, this same relative risk was 3.9 [10]. These trends were similar for other CHD risk factors, with the most striking being a relative risk of CHD death at the highest cholesterol quintile at 8.2 for the age group 35e39 falling to 2.4 for the age group 55e57 [10]. The Honolulu Heart Program [11] showed a similar decreasing association at
442
S. Pursnani et al. / Atherosclerosis 233 (2014) 441e446
older ages between CHD and hypertension, hypercholesterolemia, smoking, and BMI with increasing age, and the Framingham Heart Study showed a decreasing strength of association between CHD and hypercholesterolemia and smoking [12]. These results suggest that preventive efforts to control CHD risk factors are more effective when implemented at early ages, and is reflected in clinical risk scoring systems in which the presence of CHD risk factors at earlier ages are given increased weight [13,14]. Contrary to these findings, an analysis of baseline data combined from the ARIC and CHS cohorts showed a stable strength of association with age between these risk factors and carotid IMT, paradoxically suggesting that risk factor modification might be equally effective in older and younger individuals [15]. The purpose of this study is to clarify the age-dependency of the associations between CHD risk factors and carotid IMT using data from the ARIC cohort. With its 20 year age range of individuals at entry and nine year follow up with serial carotid IMT measurements, the cohort allows for two age comparisonsdcomparing younger and older age groups based upon age at each exam and comparing individuals within each of the two age groups at each successive visit. We studied smoking, hypertension, hypercholesterolemia, diabetes, and obesity. Given the known reduction in relative risk of CHD for these variables associated with aging, we hypothesized a similarly decreased strength of association between these risk factors and carotid IMT with aging. We hypothesized that at each visit, the associations between elevated levels of each of the risk factors and both outcomes, the difference in mean IMT and the odds ratio for highest quintile (as compared to the aggregate of the four remaining quintiles) carotid IMT, would be smaller in magnitude for the older than younger age groups, and that within each of the two age groups (age at entry 45e54 versus 55e64), these differences in mean IMT and odds ratio for highest quintile carotid IMT values would be reduced in magnitude with aging across the four visits. 2. Methods 2.1. Subjects and study design ARIC is a study of the natural history and risk factors for atherosclerosis and CHD initially (1987e89) enrolling 15,792 men and women age 45 to 64 from four diverse United States communities in North Carolina, Mississippi, Maryland, and Minnesota. After the first examination, surviving subjects were examined every three years for a total of four visits (visit 1: 1987e89, visit 2: 1990e92, visit 3: 1993e95, visit 4: 1996e98). The attrition rates, calculated as the proportion of visit 1 examinees alive at the time of the exam, were as follows: exam 2, 93%; exam 3, 86%; and exam 4, 80%. All were offered carotid ultrasound imaging at visits 1 and 2. African-American participants were invited for follow-up carotid imaging at both third and fourth visits, and the remaining subjects had one additional examination at either the third or fourth visit. The dataset used for this study was the de-identified Limited Access dataset available from the National Heart, Lung, and Blood Institute of the National Institute of Health. The data retrieved included carotid IMT, age, sex, race, LDL cholesterol, body mass index (BMI), blood pressure, smoking status, diabetes status, prevalent CHD, and medication use from each visit, and demographic variables from the initial visit. Details of cohort selection and variable measurements were described previously [16]. The carotid IMT measurements were conducted using B mode real-time ultrasound imaging, based upon a method validated by Pignoli et al., who showed their correspondence with measurements from pathologic common carotid and aortic wall specimens [17]. Carotid IMT far wall thickness was calculated as the mean of
six far wall sites 1 cm long taken from the right and left carotid bifurcation, common and internal carotid arteries. As described in prior ARIC publications, we used derived variables adjusted for measurement drift over the visit and systematic differences among readers [4]. Missing values in the dataset were previously imputed based on multivariable linear models, which related IMT measurements at each carotid site to each other, accounting for sex, race, age, body mass index, and imaging depth. We additionally conducted a sensitivity analysis using nonimputed data, where we related an average of available right and left common carotid measurements to the risk factors. For this study, risk factors were categorized as present or absent based on current clinical guidelines for treatment thresholds. Smoking was defined as self-report of current cigarette smoking. Hypertension was defined as a systolic blood pressure greater than or equal to 140 mm Hg or diastolic blood pressure greater than or equal to 90 mm Hg or the use of an anti-hypertensive agent in the preceding two weeks. Hypercholesterolemia was defined as either an LDL measurement of greater than or equal to 130 mg/dl or the use of lipid lowering agents in the preceding two weeks. Diabetes was defined as a fasting blood glucose greater than or equal to 126 mg/dl or the use of an oral hypoglycemic agent or insulin the preceding two weeks. Obesity was defined as a BMI greater than or equal to 30 kg/m2. Prevalent CHD was defined as myocardial infarction (MI) adjudicated from the visit 1 electrocardiogram, a documented history of MI, prior revascularization with coronary artery bypass grafting or percutaneous coronary intervention. Medication use was defined as self reported use of lipid lowering or blood pressure lowering medication in the two weeks preceding examination. All risk factors were defined based on laboratory results or interviews at each of the four visits. 2.2. Statistical analysis Cross sectional analyses were performed using multiple linear regression models, calculating the regression coefficients for each of the five binary risk factors in relation to the outcome of carotid IMT, treated as a continuous variable: Average carotid IMT (in mm) ¼ b0 þ b1(gender) þ b2(race) þ b3(visit-age group) þ b4 (smoking) þ b5(hypertension) þ b6(hypercholesterolemia) þ b7(diabetes) þ b8(obesity). Age group at each visit was defined as follows: Visit 1: age 45e54 and 55e64, Visit 2: age 48e57 and 58e 67, Visit 3: age 51e60 and 61e70, Visit 4: age 54e63 and 64e73. Assumptions for the use of linear regression modeling were validated by means of the Shapiro Wilk test for normality of the residuals and assessment of influential points. Though residual estimates were found to be non-normal, regression analysis was deemed appropriate given the large size of the dataset and graphical demonstration of near normality. Collinearity was evaluated with calculation of variance inflation factors, which were all less than a value of 1.2. The lack of independence of repeated carotid IMT measurements was accounted for in the longitudinal analysis (see below). We also employed logistic regression analyses to determine the odds of a measure in the top carotid IMT quintile, based upon the distribution within each age group at each visit, in relation to the presence or absence of the five risk factors. The model was as follows: log odds (carotid IMT in top quintile) ¼ b0 þ b1(gender) þ b2(race) þ b3(visit-age group) þ b4(smoking) þ b5(hypertension) þ b6(hypercholesterolemia) þ b7(diabetes) þ b8(obesity). The HosmereLemeshow goodness of fit test was used to evaluate this model. To determine the statistical significance of age as an effect modifier, two models were developed. The first model addressed the interaction between the age group and each of the five risk
S. Pursnani et al. / Atherosclerosis 233 (2014) 441e446
factors using multiple linear regression with inclusion of interaction terms at each of the four visits. The interaction terms consist of age group as a binary variable and the visit-specific risk factor also coded as a binary variable. Therefore, for each of the four visits, the model was defined as follows: Average carotid IMT ¼ b0 þ b1(gender) þ b2(race) þ b3(age group) þ b4(smoking) þ b5(hypertension) þ b6(hypercholesterolemia) þ b7(diabetes) þ b8(obesity) þ b9(smoking*age group) þ b10(hypertension*age group) þ b11(hypercholesterolemia*age group) þ b12(diabetes*age group) þ b13(obesity*age group). The hypertension*age group interaction term therefore refers to the difference between the older and younger age groups in the change in average carotid IMT between those with versus without hypertension at the specified visit. The second model addressed the interaction within each age group across the four visits. Multiple linear regression models with robust variance estimation were used to account for individual clustering of carotid IMT measurements across visits. For each of the two age groups, the model was as follows: Average carotid IMT ¼ b0 þ b1(gender) þ b2(race) þ b3(visit) þ b4(smoking) þ b5(hypertension) þ b6(hypercholesterolemia) þ b7(diabetes) þ b8(obesity) þ b9(smoking*visit) þ b10(hypertension*visit) þ b11(hypercholesterolemia*visit) þ b12(diabetes*visit) þ b13(obesity*visit), where visit was coded as an indicator variable (with visit 1 omitted and the three other visits compared to the baseline visit) and each risk factor as the visitspecific binary variable, such that the interpretation of the hypertension*visit 4 variable, for example, reflects the difference between individuals at visit 4 and at visit 1 in the change in average carotid IMT between those with versus without hypertension for the specified age category. Adjustment for prevalent CHD and medication use were performed in sensitivity analyses. All statistical analyses and graphics were performed using StataIC 11 software. 3. Results Table 1 shows characteristics of the ARIC cohort stratified by age at entry and the four visits. At the initial visit, carotid ultrasound data were available for 7662 and 6900 individuals in the younger and older age strata, respectively. As expected, carotid IMT values increased with agedas seen both within each age stratum across visits and between the two strata within visits. The coefficients of determination R2 were 21%, 18%, 14%, and 17% for the models at visit 1, 2, 3, and 4, respectively, consistent with previously published risk factor predictions of IMT variability. Partial R2 for the individual predictors were as follows: diabetes 1.1%, smoking .6%, hypercholesterolemia 1.2%, hypertension 2.7%, and obesity .01%. It should be noted that age itself contributes to much of the explained remaining variability in IMT prediction. 3.1. Smoking The percentage of current smokers was greater for the younger, as compared to the older, age group at each visit; and in each age group, it declined across visits. The regression coefficient for visit 1 indicates that average carotid IMT was .032 mm larger (95%CI: .026e.038) in current smokers than in non-smokers in the younger age group (Table 2). Similarly, average carotid IMT was .064 mm larger (95% CI: .052e.076) in current smokers than in non-smokers in the older age group. The association between IMT and current smoking was consistently stronger in the older, as compared to the younger, age group (p < 0.05 for smoking*age group interaction terms at visits 1 and 2, Table 2). Across visits, there was a relatively stable strength of association across the first three visits in each age group (Fig. 1), and in the younger group there was a statistically
443
Table 1 Cohort characteristics by age group and visit. Age at entry
Sex: number (%) male Race: number (%) black Total na
Mean carotid IMT in mm (sd)
Prevalent CHD
Visit Visit Visit Visit Visit Visit Visit Visit Visit Visit Visit Visit
Number with risk factor (% of total): Smoking Visit Visit Visit Visit Hypertension Visit Visit Visit Visit Hypercholesterolemia Visit Visit Visit Visit Diabetes Visit Visit Visit Visit Obesity Visit Visit Visit Visit Medication use in the last 2 weeks: Anti-hypertensive medication Visit Visit Visit Visit Lipid-lowering medication Visit Visit Visit Visit
45e54
55e64
1 2 3 4 1 2 3 4 1 2 3 4
3463 (41.8) 2454 (29.6) 7662 7290 4287 3621 .676 (.146) .695 (.150) .727 (.178) .755 (.203) 232 (3.0) 517 (7.1) 373 (8.7) 407 (11.2)
3592 (48.3) 1802 (24.2) 6900 6332 3582 3007 .782 (.207) .798 (.219) .832 (.252) .877 (.272) 533 (7.7) 438 (6.9) 337 (9.4) 324 (10.8)
1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4
2173 (28.4) 1784 (24.5) 930 (21.7) 640 (17.7) 2083 (27.2) 2114 (29.0) 1583 (36.9) 1419 (39.2) 3800 (49.6) 3599 (49.4) 1926 (44.9) 1682 (46.5) 662 (8.6) 877 (12.0) 612 (14.3) 517 (14.3) 2003 (26.1) 2127 (29.2) 1464 (34.1) 1266 (35.0)
1663 (24.1) 1276 (20.2) 591 (16.5) 377 (12.5) 2869 (41.6) 2672 (42.2) 1762 (49.2) 1645 (54.7) 4199 (60.9) 3583 (56.6) 1793 (50.1) 1524 (50.7) 966 (14.0) 1098 (17.3) 648 (18.1) 509 (16.9) 1786 (25.9) 1662 (26.2) 1081 (30.2) 923 (30.7)
1 2 3 4 1 2 3 4
1677 (21.9) 1854 (25.4) 1749 (40.8) 1649 (45.5) 452 (5.9) 438 (6.0) 450 (10.5) 394 (10.9)
2314 (33.5) 1668 (26.3) 1520 (42.4) 1420 (47.2) 615 (8.9) 580 (9.2) 566 (15.8) 536 (17.8)
Abbreviations: sd, standard deviation, IMT ¼ intima media thickness, CHD ¼ coronary heart disease. a Total n refers to number of individuals for whom carotid IMT data were available at each visit.
significant increased strength of association comparing the fourth to the first visits. The upper quintile logistic analysis shows similar trends in the odds ratios between age groups and across visits, i.e. stronger associations in the older than the younger group at visits 1 and 2, and stronger associations at visit 4 than visit 1 (Table 3). The age trends seen for smoking (Fig. 1) were generally similar to those seen for hypertension, hypercholesterolemia, and diabetes. 3.2. Hypertension The percentage of individuals with hypertension increased, as expected, with age both across visits and between age strata (Table 1). The difference in average IMT associated with hypertension was greater in the older, as compared to the younger, age group (p < 0.05 for visit 1 and 2, Table 2). Trends across visits within each age group were not strong, consistent or statistically significant. The logistic regression analyses (shown in the appendix) show slightly smaller odds of being in the top carotid IMT quintile for hypertensives in the younger age group than in the older group.
444
S. Pursnani et al. / Atherosclerosis 233 (2014) 441e446
Table 2 Risk factor associations with average carotid IMT: Adjusted regression coefficients and standard errors. Smoking Younger Visit Visit Visit Visit
1 .032 2 .036 3 .038 4 .065*
(.003) (.004) (.006) (.008)
Hypertension Older .064 .071 .051 .091
P value Younger
(.006)
