American Journal of Epidemiology
ª The Author 2010. Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of
Public Health. All rights reserved. For permissions, please e-mail: [email protected].
Vol. 171, No. 5
DOI: 10.1093/aje/kwp428
Advance Access publication:
January 28, 2010
Original Contribution
Markers of Atherosclerosis and Inflammation for Prediction of Coronary Heart
Disease in Older Adults
Nicolas Rodondi*, Pedro Marques-Vidal, Javed Butler, Kim Sutton-Tyrrell, Jacques Cornuz,
Suzanne Satterfield, Tamara Harris, Douglas C. Bauer, Luigi Ferrucci, Eric Vittinghoff, and
Anne B. Newman for the Health, Aging, and Body Composition Study
* Correspondence to Dr. Nicolas Rodondi, Department of Ambulatory Care and Community Medicine, Faculty of Biology and
Medicine, University of Lausanne, Bugnon 44, 1011 Lausanne, Switzerland (e-mail: [email protected]).
Initially submitted September 24, 2009; accepted for publication December 4, 2009.
Although both inflammatory and atherosclerosis markers have been associated with coronary heart disease
(CHD) risk, data directly comparing their predictive value are limited. The authors compared the value of 2
atherosclerosis markers (ankle-arm index (AAI) and aortic pulse wave velocity (aPWV)) and 3 inflammatory
markers (C-reactive protein (CRP), interleukin-6 (IL-6), and tumor necrosis factor-a (TNF-a)) in predicting CHD
events. Among 2,191 adults aged 70–79 years at baseline (1997–1998) from the Health, Aging, and Body Composition Study cohort, the authors examined adjudicated incident myocardial infarction or CHD death (‘‘hard’’
events) and ‘‘hard’’ events plus hospitalization for angina or coronary revascularization (total CHD events). During
8 years of follow-up between 1997–1998 and June 2007, 351 participants developed total CHD events (197 ‘‘hard’’
events). IL-6 (highest quartile vs. lowest: hazard ratio ¼ 1.82, 95% confidence interval: 1.33, 2.49; P-trend < 0.001)
and AAI (AAI 0.9 vs. AAI 1.01–1.30: hazard ratio ¼ 1.57, 95% confidence interval: 1.14, 2.18) predicted CHD
events above traditional risk factors and modestly improved global measures of predictive accuracy. CRP, TNF-a,
and aPWV had weaker associations. IL-6 and AAI accurately reclassified 6.6% and 3.3% of participants, respectively (P ’s 0.05). Results were similar for ‘‘hard’’ CHD, with higher reclassification rates for AAI. IL-6 and AAI are
associated with future CHD events beyond traditional risk factors and modestly improve risk prediction in older
adults.
atherosclerosis; cohort studies; coronary disease; inflammation
Abbreviations: AAI, ankle-arm index; aPWV, aortic pulse wave velocity; CHD, coronary heart disease; CI, confidence interval;
CRP, C-reactive protein; IL-6, interleukin-6; TNF-a, tumor necrosis factor-a.
Several markers of inflammation and of atherosclerosis
have been shown to be predictors of coronary heart disease
(CHD) (1–3). However, several recent studies have found
that serum biomarkers, even when combined, added only
moderately to traditional risk factors (4, 5), and their incremental value over traditional risk factors is controversial (6,
7). Few investigators have performed direct comparisons of
markers of both inflammation and atherosclerosis in CHD
risk prediction (8), and none have used the novel statistical
method of net reclassification, which allows calculation of
the proportions of adults with more accurate risk stratification by novel markers as compared with traditional risk
factors (9, 10). We hypothesized that these markers, resulting from different mechanisms underlying CHD, might have
incremental value when used in combination with traditional risk factors in identifying high-risk older adults. In
a population-based cohort of older adults, we compared the
incremental value of 2 markers of atherosclerosis (anklearm index (AAI) and aortic pulse wave velocity (aPWV)—a
marker of arterial stiffness) and 3 markers of inflammation
(C-reactive protein (CRP), interleukin-6 (IL-6), and tumor
necrosis factor-a (TNF-a)) in predicting CHD events over
traditional risk factors. Actual risk prediction with traditional risk factors performs less well in the growing
540
Am J Epidemiol 2010;171:540–549
Atherosclerosis, Inflammation, and Coronary Disease Prediction
population of older persons (11, 12) compared with middleaged adults (4, 5), but it remains unclear which novel risk
markers should be further examined to improve CHD risk
prediction in clinical practice.
MATERIALS AND METHODS
Study population
Participants were part of the Health, Aging, and Body
Composition Study, a population-based cohort study of
3,075 community-dwelling men and women aged 70–79
years at entry in 1997–1998. Participants were identified
from a random sample of white and black Medicare-eligible
adults living in designated zip-code areas surrounding Pittsburgh, Pennsylvania, and Memphis, Tennessee. Details on
the eligibility criteria have been previously published (13).
All participants gave written informed consent, and relevant
institutional review boards approved the protocol.
Among the 3,075 participants, we excluded 841 who had
overt cardiovascular disease at baseline, defined as a diagnosis of CHD (angina, prior myocardial infarction, angioplasty
of coronary arteries, or coronary artery surgery), stroke or
transient ischemic attack, peripheral arterial revascularization, carotid artery disease, or heart failure or having a pacemaker (14), based on algorithms mirroring those of the
Cardiovascular Health Study. We also excluded 23 participants with missing data for all 5 markers of inflammation or
atherosclerosis and 20 participants with missing data on any
of the traditional risk factors. The final sample for our analyses comprised 2,191 participants.
541
were obtained from frozen plasma samples, collected after
an overnight fast (15). Cytokines were measured in duplicate using a high-sensitivity enzyme-linked immunosorbent
assay kit from R&D Systems (R&D Systems, Inc., Minneapolis, Minnesota) at the Health, Aging, and Body Composition Core Laboratory (University of Vermont). Blind
duplicate analyses (n ¼ 150) for IL-6, CRP, and TNF-a
showed interassay coefficients of variation of 10.3%,
8.0%, and 15.8%, respectively (15).
Cardiovascular events. We assessed incident CHD
events and mortality among participants without overt
cardiovascular disease at baseline (14). Using algorithms
mirroring those of the Cardiovascular Health Study (14),
diagnoses and causes of death were adjudicated until
2006–2007 on the basis of interviews, reviews of all hospital
records, death certificates, and other documents by a panel
of clinicians blinded to the results for subclinical CHD
markers. CHD events were classified as either nonfatal myocardial infarction or coronary death (defined as ‘‘hard’’
events according to the current Framingham Risk Score
(18)) and as ‘‘hard’’ events plus hospitalization for angina
or revascularization (coronary angioplasty or surgery) (total
CHD events).
Covariates. Participants reported their histories of smoking and were classified as never, current, or former smokers.
Hypertension was defined as self-reported hypertension and
the use of antihypertensive medication or as measured blood
pressure 140 mm Hg (systolic pressure) and/or 90 mm
Hg (diastolic pressure). Diabetes was defined as selfreported diagnosis of diabetes and/or the use of any hypoglycemic medication (14). Fasting total and high density
lipoprotein cholesterol were measured as previously described (15).
Measurements
Markers of atherosclerosis and of inflammation at
baseline. We examined all markers of atherosclerosis and
of inflammation for which data were available at baseline in
this cohort and that had reported associations with CHD (1,
3), including reported associations in this cohort (15–17).
For AAI, blood pressures were measured in the right arm
and both ankles (posterior tibial artery) with standard cuffs
and a pencil Doppler flow probe (Parks Medical Electronics,
Inc., Aloha, Oregon) (16). The systolic blood pressure of the
ankle was divided by the systolic blood pressure of the arm
to create the AAI. Measures were repeated, and the results
were averaged. The lower of the values between the 2 legs
was used, and AAI 0.90 was considered evidence of
occlusive disease. For assessment of aPWV, a marker of
arterial stiffness, simultaneous Doppler flow signals obtained from the right carotid and right femoral arteries with
transcutaneous Doppler flow probes were measured (16).
Three separate measurements were recorded for each participant, and results from all acceptable runs were averaged
for the final aPWV value. Stiffer vessels are identified by
a faster aPWV. Replicate measures of aPWV in 14 participants revealed intraclass correlations of 0.88 between sonographers and 0.84 between readers (16).
Measures of IL-6 and CRP levels were obtained from
frozen stored serum samples and measures of TNF-a levels
Am J Epidemiol 2010;171:540–549
Statistical analyses
We used Cox proportional hazards models to assess the
utility of adding subclinical CHD markers to traditional risk
factors for the prediction of future CHD events (4, 5). Primary analyses were adjusted for traditional risk factors included in the current Framingham Risk Score (18), as well
as diabetes, a strong independent CHD risk factor (19). We
also examined associations in a further adjusted model including other potential risk factors (i.e., creatinine levels) or
confounders. To allow for nonlinear effects, we estimated
the effects of the subclinical CHD measures by quartile and
for clinically defined categories of CRP (<1, 1–3, or >3
mg/L (20)) and AAI (<0.90, 0.91–1.00, 1.01–1.30, 1.31–
1.40, or >1.40 (3, 21)). For interactions, we hypothesized
that the relations between markers and CHD might differ by
race or gender. We verified the proportional hazards assumption using graphical methods and Schoenfeld tests (all P’s >
0.20, except P ¼ 0.14 for AAI, but graphical methods did
not suggest nonproportionality over time for AAI).
As was recently recommended for assessment of novel
markers (22), we examined several statistical measures. To
assess improvements in discrimination, we used Harrell’s C
index (23), an adaptation of the C statistic or area under the
receiver operating characteristic curve to the Cox model.
Similarly to previous studies (4, 5), we compared C indices
542 Rodondi et al.
Table 1. Characteristics of Participants at Baseline (n ¼ 2,191),
Health, Aging, and Body Composition Study, 1997–1998
Characteristic
No. or Mean (SD)
%
Sociodemographic factors
Age, years
73.5 (2.8)
Female gender
1,211
55.3
Black race
900
41.1
Memphis, Tennessee,
study site
1,124
51.3
Less than high school
532
24.3
High school graduation
734
33.6
Postsecondary education
920
42.1
Education
Smoking status
Never smoker
1,015
46.3
Former smoker
955
43.6
Current smoker
221
10.1
<1
1,534
70.3
1–7
481
22.1
>7
166
7.6
Alcohol consumption,
drinks/week
Physical activity, kcal/weeka
<500
1,148
500–1,500
596
52.4
27.2
1,500
447
20.4
Hypertensionb
1,002
46.1
Diabetes mellitus
292
13.3
Body mass indexc
27.4 (4.9)
Systolic blood pressure
136 (21)
Diastolic blood pressure
72 (12)
for traditional risk factors with and without each marker.
Because the sample included too few events for split-sample
validation, we instead adjusted the C index for optimism
using bootstrap resampling (24) with 1,000 replications
(23). To ensure that the comparisons across markers were
unconfounded by extraneous differences, we estimated all C
indices using the same subset of 1,515 participants with
complete data (mainly because of 422 missing aPWV
values).
As measures of overall model fit (10, 24), we examined
likelihood ratio test v2 statistics, Akaike’s Information Criterion (10), and the Bayes Information Criterion (24). To
assess model calibration, we used Parzen and Lipsitz’s adaptation (25) of the Hosmer-Lemeshow test to the Cox
model.
We computed net reclassification rates (9, 10) for markers
that both had strong relations with CHD events beyond
traditional risk factors and improved global measures of
predictive accuracy above traditional risk factors. To avoid
extrapolation beyond the range of our data, we used the Cox
models to estimate 7.5-year rather than 10-year risks. We
also assessed net reclassification in the intermediate risk
categories (10%–20% 10-year risk) of most clinical interest
(26)—that is, 7.5%–15% risk for a 7.5-year time frame. We
examined reclassification among the 1,985 participants with
no missing AAI and IL-6 data, to avoid confounding by
differences in the set of included observations.
Statistical analyses were conducted using Stata 9.2 (Stata
Corporation, College Station, Texas) and R (R Project for
Statistical Computing, Vienna, Austria).
Lipoprotein level, mmol/L
Total cholesterol
5.31 (0.98)
High density lipoprotein
cholesterol
1.42 (0.44)
Low density lipoprotein
cholesterol
3.19 (0.88)
Triglyceridesd
1.31 (0.98–1.81)
10-year FRS, %e
RESULTS
16.6 (2.9)
Category of FRS
<5%
297
13.6
5–9.9%
523
23.9
10–19.9%
853
38.9
20%
518
23.6
Creatinine level, mmol/Ld
88.4 (79.6–97.2)
Medication use
Lipid-lowering drugs
229
10.5
Angiotensin-converting
enzyme inhibitors
273
12.5
Hormone replacement therapy
285
13.0
Aspirin
412
18.8
Abbreviations: FRS, Framingham Risk Score; SD, standard deviation.
a
Physical activity was assessed by questionnaire. Participants were asked
about all types of walking and exercise performed in the prior week (14).
b
Defined as self-reported hypertension and use of antihypertensive medication or measured blood pressure 140/90 mm Hg.
c
Weight (kg)/height (m)2.
d
Expressed as median (25th–75th percentiles). A Mann-Whitney rank-sum
test was performed.
e
10-year risk of coronary heart disease according to Adult Treatment Panel
III FRS (18).
At baseline, the mean age of study participants was 73.5
years; 55% were women, and 41% were black (Table 1).
The mean 10-year Framingham Risk Score was 16.6%,
and most participants were in the intermediate CHD risk
group. In gender-stratified and age-adjusted analyses, all
inflammatory markers were correlated (r ¼ 0.13–0.46; all
P’s < 0.001), AAI and aPWV were weakly correlated
(r ¼ 0.07 in men, r ¼ 0.14 in women), and no correlation was found between inflammatory and atherosclerotic
markers (except for aPWV with IL-6 (r ¼ 0.13) and TNF-a
(r ¼ 0.10) in women; both P’s < 0.01).
During a median follow-up period of 8.2 years (maximum, 10.2 years) between 1997–1998 and June 2007,
351 participants developed total CHD events (197 ‘‘hard’’
events). Event rates were more frequent at higher IL-6,
TNF-a, and aPWV levels and at both extremes of the AAI
distribution (Table 2). In multivariate analyses, IL-6, AAI,
and aPWV predicted CHD events beyond traditional risk
factors. There was borderline-significant evidence for increased CHD risk for CRP > 3.0 mg/L and increasing quartiles of TNF-a, given multiple comparisons. Estimates per
log unit increase in the continuous marker values or further
adjustment for other potential risk factors and confounders
yielded similar results. Participants in the highest quartile
of IL-6 combined with low (0.90; n ¼ 71) or high (>1.30;
n ¼ 28) AAI were at particularly increased risk, with hazard
Am J Epidemiol 2010;171:540–549
Atherosclerosis, Inflammation, and Coronary Disease Prediction
543
Table 2. Hazard Ratios for Incidence of Total Coronary Heart Disease Events (n ¼ 351 Events) in Older Adults During a Median Follow-up
Period of 8.2 Years (Maximum, 10.2 Years), According to Measures of Subclinical Cardiovascular Disease (n ¼ 2,191), Health, Aging and Body
Composition Study, 1997–2007
HR
Adjusted for
Traditional
CVD Risk
Factorsa
Range of
Values
(Median)
Incidence
Rate/1,000
PersonYears
No. of
Events
Quartile 1
0.21–1.18 (0.94)
15.40
66
1
Quartile 2
1.19–1.75 (1.46)
17.48
73
1.10
Quartile 3
1.76–2.64 (2.12)
22.62
91
1.57
1.14, 2.16
1.38
1.00, 1.90
1.32
0.95, 1.83
Quartile 4
2.65–15.96 (3.74)
29.18
108
2.10
1.54, 2.85
1.82
1.33, 2.49
1.72
1.24, 2.39
Measure of
Subclinical CVD
and Category
AgeAdjusted
HR
95% CI
95% CI
MultivariateAdjusted
HRb
95% CI
Interleukin-6, pg/mL
P for trendc
<0.001
Referent
0.79, 1.53
1
Referent
1.00
<0.001
0.71, 1.40
1
Referent
0.99
<0.001
0.71, 1.41
<0.001
C-reactive protein,
mg/L
Quartile 1
0.15–1.00 (0.78)
19.35
84
1
Quartile 2
1.01–1.64 (1.26)
21.81
91
1.15
0.86, 1.55
1.18
0.87, 1.59
1.12
0.82, 1.52
Quartile 3
1.65–3.07 (2.25)
20.92
89
1.14
0.84, 1.53
1.13
0.84, 1.53
1.14
0.83, 1.54
Quartile 4
3.08–85.18 (4.70)
21.60
87
1.23
0.91, 1.66
1.29
0.95, 1.76
1.28
P for trend
0.26
Referent
1
0.21
Referent
1
0.15
Referent
0.92, 1.78
0.15
C-reactive protein,
mg/L
Category 1
<1.0
19.21
82
Category 2
Category 3
1–3.0
21.46
179
1.18
0.91, 1.53
1.17
0.90, 1.52
1.15
0.88, 1.51
>3.0
21.48
90
1.24
0.92, 1.68
1.31
0.96, 1.78
1.31
0.95, 1.82
P for trend
1
0.20
Referent
1
0.16
Referent
1
0.09
Referent
0.11
Tumor necrosis
factor-a, pg/mL
Quartile 1
0.57–2.37 (1.93)
17.42
71
1
Quartile 2
2.38–3.08 (2.72)
16.73
67
0.95
0.68, 1.32
0.91
0.65, 1.28
0.93
0.66, 1.31
Quartile 3
3.09–3.94 (3.45)
19.88
80
1.08
0.78, 1.49
1.03
0.74, 1.42
1.02
0.72, 1.43
Quartile 4
3.95–29.55 (4.78)
29.03
107
1.63
1.21, 2.21
1.45
1.06, 1.99
1.54
P for trend
Referent
1
0.001
<0.001
Referent
1
Referent
1.09, 2.16
0.01
<0.02
Ankle-arm index
Category 1
0.9
32.60
49
1.90
1.40, 2.59
1.57
1.14, 2.18
1.66
1.20, 2.29
Category 2
0.91–1.00
19.65
36
1.04
0.73, 1.48
1.05
0.73, 1.49
1.10
0.77, 1.58
Category 3
1.01–1.30
18.44
218
Category 4
1.31–1.40
22.26
14
1.35
0.79, 2.32
1.29
0.75, 2.23
1.02
0.58, 1.77
Category 5
>1.4
43.08
9
3.34
1.71, 6.53
2.89
1.47, 5.68
2.60
1.32, 5.14
P for quadratic
patternd
1
<0.001
Referent
1
<0.001
Referent
1
<0.001
Referent
<0.001
Aortic pulse wave
velocity, cm/
second
Quartile 1
312–635 (546)
17.03
63
1
Quartile 2
636–795 (711)
19.06
67
1.45
1.02, 2.05
1.35
0.95, 1.93
1.34
0.93, 1.93
Quartile 3
796–1,042 (893)
21.13
72
1.73
1.22, 2.44
1.57
1.10, 2.23
1.47
1.01, 2.13
Quartile 4
1,045–2,916 (1,313)
23.00
77
1.96
1.39, 2.76
1.74
1.22, 2.49
1.68
P for trend
<0.001
Referent
<0.001
1
Referent
0.002
1
Referent
1.16, 2.44
0.006
Abbreviations: CI, confidence interval; CVD, cardiovascular disease; HR, hazard ratio.
a
Adjusted for traditional CVD risk factors, including age, gender, total and high density lipoprotein cholesterol, systolic blood pressure, smoking, and diabetes.
b
Further adjusted for race, education, site, low density lipoprotein cholesterol, creatinine (as natural logarithm), alcohol consumption, physical activity, statins,
angiotensin-converting enzyme inhibitors, and estrogen use.
c
Linear and quadratic trends were assessed using orthogonal contrasts.
d
Because of the U-shaped relation, the middle category (category 3, 1.01–1.30) was used as the reference group, and categories were compared with an
orthogonal linear contrast to test for a quadratic pattern.
Am J Epidemiol 2010;171:540–549
544 Rodondi et al.
Table 3. Discrimination, Model Fit, and Calibration Estimates for Total Coronary Heart Disease Events (n ¼ 235
Events) in Older Adults During a Median Follow-up Period of 8.2 Years (Maximum, 10.2 Years), According to
Traditional Cardiovascular Disease Risk Factors, With or Without the Addition of Markers of Inflammation or of
Atherosclerosis (n ¼ 1,515), Health, Aging, and Body Composition Study, 1997–2007a
Discrimination
Model
Model Fit
Likelihood Ratio
Teste
C Indexc
Adjusted
C Indexd
x2 (df)
Traditional cardiovascular
disease risk factors
0.631
0.611
35.31 (8)
Measures Penalized for
Model Complexity
P Value
Akaike’s
Information
Criterionf
Bayes
Information
Criteriong
<0.001
3,102.6
3,145.2
Calibration:
Modified
HosmerLemeshow Testb
x2
P Value
4.23
0.90
0.52
Plus IL-6
0.650
0.629
58.71 (11) <0.001
3,085.3
3,143.8
8.19
Plus C-reactive protein
0.638
0.622
41.87 (11) <0.001
3,104.8
3,163.3
13.86
0.13
Plus tumor necrosis
factor-a
0.647
0.621
43.49 (11) <0.001
3,089.5
3,148.0
14.26
0.11
Plus AAI
0.650
0.624
45.80 (12) <0.001
3,100.2
3,164.1
15.01
0.09
Plus aortic pulse wave
velocity
0.629
0.605
41.42 (11) <0.001
3,102.6
3,161.1
3.76
0.93
Plus IL-6 þ AAI
0.662
0.634
68.58 (15) <0.001
3,083.5
3,163.3
13.01
0.16
Abbreviations: AAI, ankle-arm index; IL-6, interleukin-6.
All estimates reported are from multivariate analyses, adjusted for traditional cardiovascular disease risk factors,
with or without addition of each marker in quartiles or clinically defined categories (for AAI) (see Methods section in
text), among 1,515 participants with complete data on all markers (mainly because of 422 missing values for aortic
pulse wave velocity).
b
Parzen and Lipsitz’s adaptation (25), based on martingale residuals, of the Hosmer-Lemeshow test of goodness
of fit in the logistic model to the Cox model, comparing observed and expected failures within deciles of predicted risk.
Larger P values indicate better calibration (24).
c
Harrell’s C Index, an adaptation of the C statistic (equivalently, the area under the receiver operating characteristic curve) in logistic models. Higher values indicate better discrimination.
d
Harrell’s C index, corrected for optimism using bootstrap resampling (24).
e
Omnibus test of the overall statistical significance of all predictors in the model. A higher value indicates better fit.
f
Akaike’s Information Criterion (10) is a likelihood-based measure in which a simple measure of goodness of fit
(2 times the log-likelihood) is penalized for the number of predictors in the model. A lower value indicates better
prediction.
g
The Bayes Information Criterion (24) is similar to Akaike’s Information Criterion. This measure of predictiveness
imposes a more severe penalty for the number of predictors. A lower value indicates better prediction.
a
ratios of 2.32 (95% confidence interval (CI): 1.41, 3.79) and
2.67 (95% CI: 1.23, 5.79), respectively. The relation between different markers and CHD events did not differ by
race or gender (P’s for interaction > 0.05), except that
TNF-a was less strongly associated with CHD events
among blacks (P values for interaction were 0.03 for total
CHD events and 0.06 for ‘‘hard’’ CHD, to be interpreted
with caution in the context of multiple comparisons).
Results were similar for ‘‘hard’’ CHD events (see Web
Table 1, which is posted on the Journal’s Web site
(http://aje.oxfordjournals.org/)), except that associations
with aPWV were no longer statistically significant.
Table 3 summarizes comparisons of discrimination, model
fit, and calibration. In general, the 3 models adding IL-6, AAI,
or both to traditional risk factors scored highest in terms of
the C index and likelihood ratio v2 and near the minimum
in terms of Akaike’s Information Criterion. The AAI and
the IL-6 þ AAI models were more heavily penalized for
calibration and by the Bayes Information Criterion, reflecting
their additional parameters. Patterns were similar for hard
CHD events (Web Table 2 (http://aje.oxfordjournals.org/)).
Sensitivity analyses that did not exclude persons missing
any subclinical CHD data yielded similar results for
C indices. At baseline, the 1,515 participants with all subclinical CHD measures were generally similar to the remaining 676, except that those with complete data were slightly
older (73.6 years (standard deviation, 2.8) vs. 73.3 years (standard deviation, 2.9); P ¼ 0.02), more likely to be white (63%
vs. 37%; P < 0.001), and more likely to have higher education
(P < 0.001). Sensitivity analyses including the 416 participants missing aPWV values yielded similar results for the
other subclinical CHD markers.
The IL-6, AAI, and IL-6 þ AAI models, all 3 of which
had strong relations with CHD events beyond traditional
risk factors and improved global measures of predictive
accuracy, correctly reclassified 6.6% (95% CI: 1.2, 11.9),
3.3% (95% CI: 0.04, 6.5), and 5.0% (95% CI: 0.6, 10.6)
of participants, respectively, over traditional risk factors
(Table 4). IL-6, AAI, or both correctly reclassified 15.6%
(95% CI: 9.1, 22.1), 7.0% (95% CI: 2.9, 11.2), and 12.0%
(95% CI: 5.4, 18.6) of intermediate-risk participants (all
P’s < 0.001). For ‘‘hard’’ CHD events, reclassification rates
Am J Epidemiol 2010;171:540–549
Atherosclerosis, Inflammation, and Coronary Disease Prediction
545
Table 4. Predictive Risks and Reclassification of Total Coronary Heart Disease Events Using a Multivariate Risk Prediction Model With and
Without Inclusion of IL-6, AAI, or IL-6 þ AAI (n ¼ 1,985), Health, Aging, and Body Composition Study, 1997–2007a
Predicted 7.5-Year
Risk of a
CHD Event
No. of Participants, by
Predicted 7.5-Year Risk
<7.5%
7.5%–<15%
No. of Participants
Reclassifiedb
‡15%
Increased
Risk
Decreased
Risk
Proportion
Correctly
Reclassified, %c
95%
Confidence
Interval
P Value
Predicted 7.5-Year CHD Event Risk With IL-6
Predicted risk without IL-6 among
persons with events during
follow-up (n ¼ 313)
<7.5%
3
2
7.5%–<15%
9
83
25
22
169
15%
27
31
1.3
138
269
7.8
Predicted risk without IL-6 among
persons without events during
follow-up (n ¼ 1,672)
<7.5%
7.5%–<15%
39
19
136
648
119
133
578
15%
Net reclassification improvementd
6.6
1.2, 11.9
0.016
0.04, 6.5
0.047
0.6, 10.6
0.080
Predicted 7.5-Year CHD Event Risk With AAI
Predicted risk without AAI among
persons with events during
follow-up (n ¼ 313)
<7.5%
4
1
7.5%–<15%
1
103
13
6
185
15%
14
7
2.2
68
85
1.0
Predicted risk without AAI among
persons without events during
follow-up (n ¼ 1,672)
<7.5%
49
9
7.5%–<15%
30
814
59
55
656
15%
Net reclassification improvementd
3.3
Predicted 7.5-Year CHD Event Risk With IL-6 þ AAI
Predicted risk without IL-6 þ AAI
among persons with events
during follow-up (n ¼ 313)
<7.5%
7.5%–<15%
3
2
13
82
22
27
164
15%
24
40
5.1
143
312
10.1
Predicted risk without IL-6 þ AAI
among persons without events
during follow-up (n ¼ 1,672)
<7.5%
7.5%–<15%
15%
Net reclassification improvementd
37
19
2
161
620
122
151
560
5.0
Abbreviations: AAI, ankle-arm index; CHD, coronary heart disease; IL-6, interleukin-6.
Reclassification of CHD risk was evaluated among the 1,985 participants with no missing AAI and IL-6 data by comparing predicted risk
estimates based on multivariate models that included age, gender, total and high density lipoprotein cholesterol, systolic blood pressure, smoking,
and diabetes, with and without inclusion of IL-6, AAI, or both, for persons with and without CHD events during follow-up (10). Data shown are the
numbers of subjects cross-classified by their predicted 7.5-year risk using the models with and without inclusion of IL-6, AAI, or both. The 7.5-year
risk was divided into <7.5%, 7.5%–<15%, and 15%, which is close to the corresponding 10-year risk of <10%, 10%–<20%, and 20% used in
national guidelines (18).
b
Numbers of subjects who were reclassified upwards and downwards, respectively, when IL-6, AAI, or both were added to the model.
c
Proportion of all participants with CHD events who were ‘‘correctly’’ reclassified into a higher risk category, plus the proportion of all participants
remaining event-free who were reclassified into a lower category, minus the proportion of each group reclassified in the ‘‘wrong’’ direction.
d
Sum of the percentages of correctly reclassified subjects with and without CHD events. Significant P values indicate improved classification (9).
a
Am J Epidemiol 2010;171:540–549
546 Rodondi et al.
Table 5. Risk of Coronary Heart Disease Over a 7.5-Year Period
According to Framingham Risk Score and Quartile of Interleukin-6
Level (n ¼ 1,985), Health, Aging, and Body Composition Study,
1997–2007
FRS and IL-6
Quartile
No. at
Risk
No. of
Events
Predicted
CHD Risk, %
<5%
Table 6. Risk of Coronary Heart Disease Over a 7.5-Year Period
According to Framingham Risk Score and Category of AAI (n ¼
1,985), Health, Aging, and Body Composition Study, 1997–2007
FRS and AAI
Category
No. at
Risk
No. of
Events
Predicted
CHD Risk, %
<5%
0.90
15
1
9.72
1
102
6
4.12
0.91–1.00
37
4
5.79
2
73
2
4.09
1.01–1.30
218
13
4.90
3
50
7
6.27
1.31–1.40
5
0
6.38
4
54
3
7.86
1.41
4
0
10.72
1
145
18
4.87
0.90
38
8
10.64
2
99
15
4.85
0.91–1.00
54
8
6.40
3
119
13
6.84
1.01–1.30
370
47
5.71
4
114
23
9.60
1.31–1.40
12
3
7.70
3
3
14.40
5%–9.9%
5%–9.9%
10%–19.9%
1.41
1
175
26
7.29
2
207
25
6.90
0.90
89
21
13.86
3
196
37
9.98
0.91–1.00
85
11
8.60
4
177
33
13.53
1.01–1.30
534
82
8.30
1.31–1.40
36
6
11.94
1.41
11
1
20.51
20%
10%–19.9%
1
86
14
9.63
2
130
25
9.55
3
116
25
13.12
0.90
65
17
20.06
4
142
41
18.59
0.91–1.00
54
12
13.52
1.01–1.30
31
67
10.83
1.31–1.40
29
5
13.64
1.41
12
4
26.76
Abbreviations: CHD, coronary heart disease; FRS, Framingham
Risk Score; IL-6, interleukin-6.
were similar for IL-6 (for all participants/intermediate-risk
participants, 4.4%/11.0%) but higher for AAI (7.9%/19.3%)
and IL-6 þ AAI (7.7%/19.8%). For IL-6, net reclassification
of events was towards lower risk, with good calibration,
suggesting that the lower estimated risk fitted the data well.
The net reclassification index with IL-6 þ AAI was not
statistically different from simpler models with only IL-6
(P ¼ 0.43) or AAI (P ¼ 0.55). Table 5 and Table 6 show the
predicted 7.5-year CHD risks from Cox regression models
for categories of Framingham Risk Score stratified by IL-6
and AAI, respectively.
DISCUSSION
In this population-based study of older adults, IL-6 and
AAI were associated with increased risk of incident CHD
beyond traditional risk factors. IL-6 or AAI more accurately
reclassified 3%–7% of participants into persons who did and
did not develop an event as compared with traditional risk
factors (7%–16% for the intermediate-risk group). A similar
magnitude of net reclassification has been previously found
for CRP in middle-aged women (26). Elevated CRP, TNF-a,
and aPWV levels were associated with a pattern of increased
CHD risk but added less to risk prediction above traditional
20%
Abbreviations: AAI, ankle-arm index; CHD, coronary heart disease;
FRS, Framingham Risk Score.
risk factors than IL-6 and AAI. AAI was a stronger predictor
of CHD than was aPWV, particularly for ‘‘hard’’ CHD
events; to our knowledge, these 2 markers have not been
directly compared previously (27).
Our findings are consistent with previous studies on the
association between inflammatory markers and CHD in
older adults (15, 28) and on single markers of atherosclerosis (3, 17, 29), but our results add information on the direct
comparisons of both markers of inflammation and markers
of atherosclerosis. In older adults, we found improved
CHD risk prediction with IL-6 and AAI, in contrast to the
use of biomarkers in middle-aged adults (4, 5). In the Atherosclerosis Risk in Communities Study, Folsom et al. (5)
found that CRP and other novel risk factors did not significantly increase C index. In the Framingham Heart Study,
middle-aged adults with high multimarker risk scores had
increased risk of cardiovascular events (for the highest quintile vs. the lowest 2 quintiles, adjusted hazard ratio ¼ 1.84,
95% CI: 1.11, 3.05) but only a small improvement in C
index over traditional risk factors (4). Reclassification of
risk was not examined in those 2 studies. In contrast, a study
Am J Epidemiol 2010;171:540–549
Atherosclerosis, Inflammation, and Coronary Disease Prediction
in white men aged 69–74 years found that addition of 4
biomarkers to traditional risk factors significantly increased
C index values for cardiovascular death (0.748 vs. 0.688 in
661 participants without cardiovascular disease) (30), but
the investigators did not examine risk prediction for CHD
events. These potential differences might be related to the
age of the studied populations, since lower predictive value
(as measured by C index) and weaker associations of traditional risk factors with CHD have been reported in older
adults (11, 12) as compared with middle-aged adults (4, 5),
leading to a wider opportunity for improvement of risk prediction in older adults.
Few researchers have performed direct comparisons of
the predictive value of both markers of atherosclerosis and
markers of inflammation (8). In the Cardiovascular Health
Study, associations between AAI and cardiovascular events
remained after adjustment for CRP (3), as well as associations between CRP and CHD events after adjustment for
several subclinical cardiovascular disease measurements
(for CRP > 3 mg/L vs. CRP < 1 mg/L, the adjusted hazard
ratio was 1.37 (95% CI: 1.06, 1.78)) (31). However, the
authors did not directly compare the predictive values of
both of these sets of markers. In the same Cardiovascular
Health Study cohort, elevated CRP, elevated carotid intimamedia thickness, and plaques all increased the adjusted risk
for cardiovascular events, with modest improvement in C
index, but elevated CRP remained a predictor beyond traditional risk factors only in participants with carotid atherosclerosis on ultrasound (8).
The net risk reclassification with IL-6 and AAI over traditional risk factors was similar to that reported with CRP in
middle-aged women (5.7% overall, 15% for the intermediaterisk group) (26). The net increase in C index was modest, but
the use of the C index for assessment of novel markers is
controversial (9, 32, 33). The C index is particularly suited to
retrospective case-control studies (24), and it may have little
direct clinical relevance for assessing the role of novel
markers to better stratify risk and target preventive therapies
(32, 33). Other parameters of model fit were also improved by
the addition of IL-6 or AAI, with the exception of the Bayes
Information Criterion for AAI, which might be more sensitive to the increased complexity of these models.
Our study had several strengths and limitations. These
data were drawn from a well-characterized populationbased cohort of older adults and contained a high number
of CHD events over 8 years of follow-up, and the CHD
events were formally adjudicated. Blood sampling for cytokine concentrations was performed at a single point in time,
which is a limitation of most published observational cohort
studies (1, 4, 31). Use of a single blood sample for determination of cytokine concentration could be influenced by
circadian rhythm (15), which would tend to weaken the
observed associations because of added variability. Because
of the multiple testing, borderline-significant P values, particularly P values between 0.05 and 0.01 for the 5 markers
compared in Table 2, should be interpreted with caution and
in the light of other findings and biologic plausibility.
Clinically, traditional risk factors have lower discrimination in older adults than in younger adults and poorly predict
CHD in older adults. In our data, the C index for traditional
Am J Epidemiol 2010;171:540–549
547
risk factors was 0.631 (Table 3). C indices for the Framingham Risk Score were 0.63 in men and 0.66 in women in the
Cardiovascular Health Study (11), as compared with 0.79
and 0.83, respectively, in the Framingham Heart Study (11).
Some traditional risk factors have weaker associations with
CHD risk; total and low density lipoprotein cholesterol were
strong cardiovascular risk factors in middle-aged adults but
not in older adults in our analysis and previous analyses
(34). The addition of IL-6 or AAI to traditional risk factors
modestly improved prediction of the true cardiovascular risk
in older adults, with 3%–7% of all participants and 7%–16%
of the intermediate-risk group being reclassified into more
accurate risk categories. It has been suggested that novel
markers could be particularly useful in refining risk prediction in intermediate-risk adults (26) to identify persons at
increased risk who would benefit from preventive therapies,
such as statins or aspirin. This might allow more efficient
allocation of resources for prevention, which should be confirmed by cost-effectiveness analyses. Combination of measures of inflammation and atherosclerosis might be useful
for identification of very-high-risk older adults, such as
those in the highest quartile of IL-6 with an AAI less than
or equal to 0.90 or an AAI greater than 1.30 (high AAI
values reflect decreased arterial compliance, usually attributed to arterial calcification (3)); however, these combined
categories had low numbers of participants (n ¼ 71 and n ¼
28, respectively), and net reclassification was significantly
improved in intermediate-risk participants but not in overall
participants. Currently, there are no standardized IL-6 assays available for clinical use. However, our data suggest
that IL-6 might be a better risk discriminator than CRP in
older adults. The hepatic synthesis of CRP is largely under
the regulation of IL-6; CRP might represent a ‘‘distal’’ secondary phenomenon to a more ‘‘proximal’’ primary excess
production of IL-6 in the pathophysiologic chain of events
(35). If our results for the association between IL-6 and
CHD events are confirmed in other studies, investigators
in future studies should validate standardized IL-6 assays
for use in clinical practice, similar to CRP (20).
In summary, our study suggests that IL-6 and AAI are
associated with future CHD events and help improve cardiovascular risk prediction beyond traditional risk factors in
the elderly, with modestly improved risk reclassification.
These findings are particularly important, since actual risk
prediction with traditional risk factors is less accurate in
older persons (11, 12) as compared with middle-aged adults
(4, 5). This study suggests that new risk scores that include
IL-6 or AAI might improve CHD prediction and risk stratification in the elderly.
ACKNOWLEDGMENTS
Author affiliations: Department of Ambulatory Care and
Community Medicine, Faculty of Biology and Medicine,
University of Lausanne, Lausanne, Switzerland (Nicolas
Rodondi, Jacques Cornuz); University Institute of Social
and Preventive Medicine, University of Lausanne,
Lausanne, Switzerland (Pedro Marques-Vidal); Center for
548 Rodondi et al.
Cardiovascular and Metabolic Research (CardioMet), University of Lausanne, Lausanne, Switzerland (Pedro
Marques-Vidal); Division of Cardiology, School of Medicine, Emory University, Atlanta, Georgia (Javed Butler);
Department of Epidemiology, Graduate School of Public
Health, University of Pittsburgh, Pittsburgh, Pennsylvania
(Kim Sutton-Tyrrell, Anne B. Newman); Department of
Preventive Medicine, College of Medicine, University of
Tennessee, Memphis, Tennessee (Suzanne Satterfield);
Laboratory of Epidemiology, Demography, and Biometry,
National Institute on Aging, Bethesda, Maryland (Tamara
Harris); Intramural Research Program, National Institute on
Aging, Baltimore, Maryland (Luigi Ferrucci); Department
of Epidemiology and Biostatistics, School of Medicine,
University of California, San Francisco, San Francisco,
California (Douglas C. Bauer, Eric Vittinghoff); and Division of General Internal Medicine, Department of Medicine,
School of Medicine, University of California, San
Francisco, San Francisco, California (Douglas C. Bauer).
This research was supported by grants N01-AG-6-2101,
N01-AG-6-2103, N01-AG-6-2106, N01-AG-6-2101, N01AG-6-2103, and N01-AG-6-2106 from the National Institute
on Aging, US National Institutes of Health, and in part by the
Intramural Research Program of the National Institute on
Aging. The National Institute on Aging funded the Health,
Aging, and Body Composition study, reviewed the manuscript, and approved its publication. This research was
partially supported by grant 3200B0-116097 to Dr. Nicolas
Rodondi from the Swiss National Science Foundation.
Conflict of interest: none declared.
8.
9.
10.
11.
12.
13.
14.
15.
16.
REFERENCES
1. Ridker PM, Hennekens CH, Buring JE, et al. C-reactive protein and other markers of inflammation in the prediction of
cardiovascular disease in women. N Engl J Med. 2000;
342(12):836–843.
2. Ridker PM, Rifai N, Stampfer MJ, et al. Plasma concentration
of interleukin-6 and the risk of future myocardial infarction
among apparently healthy men. Circulation. 2000;101(15):
1767–1772.
3. O’Hare AM, Katz R, Shlipak MG, et al. Mortality and cardiovascular risk across the ankle-arm index spectrum: results
from the Cardiovascular Health Study. Circulation. 2006;
113(3):388–393.
4. Wang TJ, Gona P, Larson MG, et al. Multiple biomarkers for
the prediction of first major cardiovascular events and death.
N Engl J Med. 2006;355(25):2631–2639.
5. Folsom AR, Chambless LE, Ballantyne CM, et al. An assessment of incremental coronary risk prediction using
C-reactive protein and other novel risk markers: the Atherosclerosis Risk in Communities Study. Arch Intern Med. 2006;
166(13):1368–1373.
6. Redberg RF, Vogel RA, Criqui MH, et al. 34th Bethesda
Conference: Task Force #3—what is the spectrum of current
and emerging techniques for the noninvasive measurement of
atherosclerosis? J Am Coll Cardiol. 2003;41(11):1886–1898.
7. Smith SC Jr, Greenland P, Grundy SM. Prevention conference
V. Beyond secondary prevention: identifying the high-risk
17.
18.
19.
20.
21.
22.
23.
patient for primary prevention: executive summary. Circulation. 2000;101(1):111–116.
Cao JJ, Arnold AM, Manolio TA, et al. Association of carotid
artery intima-media thickness, plaques, and C-reactive protein
with future cardiovascular disease and all-cause mortality:
the Cardiovascular Health Study. Circulation. 2007;116(1):
32–38.
Pencina MJ, D’Agostino RB Sr, D’Agostino RB Jr, et al.
Evaluating the added predictive ability of a new marker: from
area under the ROC curve to reclassification and beyond. Stat
Med. 2008;27(2):157–172.
Ingelsson E, Schaefer EJ, Contois JH, et al. Clinical utility of
different lipid measures for prediction of coronary heart disease in men and women. JAMA. 2007;298(7):776–785.
D’Agostino RB Sr, Grundy S, Sullivan LM, et al. Validation of
the Framingham coronary heart disease prediction scores: results of a multiple ethnic groups investigation. JAMA. 2001;
286(2):180–187.
Greenland P, LaBree L, Azen SP, et al. Coronary artery calcium score combined with Framingham score for risk prediction in asymptomatic individuals. JAMA. 2004;291(2):
210–215.
Yaffe K, Kanaya A, Lindquist K, et al. The metabolic syndrome, inflammation, and risk of cognitive decline. JAMA.
2004;292(18):2237–2242.
Rodondi N, Newman AB, Vittinghoff E, et al. Subclinical
hypothyroidism and the risk of heart failure, other cardiovascular events, and death. Arch Intern Med. 2005;165(21):
2460–2466.
Cesari M, Penninx BW, Newman AB, et al. Inflammatory
markers and onset of cardiovascular events: results from
the Health ABC Study. Circulation. 2003;108(19):2317–
2322.
Sutton-Tyrrell K, Najjar SS, Boudreau RM, et al. Elevated
aortic pulse wave velocity, a marker of arterial stiffness, predicts cardiovascular events in well-functioning older adults.
Circulation. 2005;111(25):3384–3390.
Sutton-Tyrrell K, Venkitachalam L, Kanaya AM, et al. Relationship of ankle blood pressures to cardiovascular events in
older adults. Stroke. 2008;39(3):863–869.
National Cholesterol Education Program. Executive summary
of the third report of the National Cholesterol Education Program (NCEP) Expert Panel on Detection, Evaluation, and
Treatment of High Blood Cholesterol in Adults (Adult Treatment Panel III). JAMA. 2001;285(19):2486–2497.
Lee CD, Folsom AR, Pankow JS, et al. Cardiovascular
events in diabetic and nondiabetic adults with or without
history of myocardial infarction. Circulation. 2004;109(7):
855–860.
Pearson TA, Mensah GA, Alexander RW, et al. Markers of
inflammation and cardiovascular disease: application to clinical and public health practice: a statement for healthcare
professionals from the Centers for Disease Control and Prevention and the American Heart Association. Circulation.
2003;107(3):499–511.
Resnick HE, Lindsay RS, McDermott MM, et al. Relationship
of high and low ankle brachial index to all-cause and cardiovascular disease mortality: the Strong Heart Study. Circulation. 2004;109(6):733–739.
Hlatky MA, Greenland P, Arnett DK, et al. Criteria for evaluation of novel markers of cardiovascular risk: a scientific
statement from the American Heart Association. Circulation.
2009;119(17):2408–2416.
Harrell FE Jr, Lee KL, Mark DB. Multivariable prognostic
models: issues in developing models, evaluating assumptions
Am J Epidemiol 2010;171:540–549
Atherosclerosis, Inflammation, and Coronary Disease Prediction
24.
25.
26.
27.
28.
29.
and adequacy, and measuring and reducing errors. Stat Med.
1996;15(4):361–387.
Cook NR, Buring JE, Ridker PM. The effect of including
C-reactive protein in cardiovascular risk prediction models
for women. Ann Intern Med. 2006;145(1):21–29.
Parzen M, Lipsitz SR. A global goodness-of-fit statistic for
Cox regression models. Biometrics. 1999;55(2):580–584.
Cook NR. Comments on ‘Evaluating the added predictive
ability of a new marker: from area under the ROC curve to
reclassification and beyond’ by M. J. Pencina et al., Statistics
in Medicine (DOI: 10.1002/sim.2929). Stat Med. 2008;27(2):
191–195.
Simon A, Mijiti W, Gariepy J, et al. Current possibilities for
detecting high risk of cardiovascular disease. Int J Cardiol.
2006;110(2):146–152.
Koukkunen H, Penttilä K, Kemppainen A, et al. C-reactive
protein, fibrinogen, interleukin-6 and tumour necrosis factoralpha in the prognostic classification of unstable angina pectoris. Ann Med. 2001;33(1):37–47.
Fowkes FG, Murray GD, Butcher I, et al. Ankle brachial index
combined with Framingham Risk Score to predict cardiovas-
Am J Epidemiol 2010;171:540–549
30.
31.
32.
33.
34.
35.
549
cular events and mortality: a meta-analysis. JAMA. 2008;
300(2):197–208.
Zethelius B, Berglund L, Sundström J, et al. Use of multiple
biomarkers to improve the prediction of death from cardiovascular causes. N Engl J Med. 2008;358(20):2107–2116.
Cushman M, Arnold AM, Psaty BM, et al. C-reactive protein
and the 10-year incidence of coronary heart disease in older
men and women: the Cardiovascular Health Study. Circulation. 2005;112(1):25–31.
Janes H, Pepe MS, Gu W. Assessing the value of risk predictions by using risk stratification tables. Ann Intern Med. 2008;
149(10):751–760.
Cook NR. Use and misuse of the receiver operating characteristic curve in risk prediction. Circulation. 2007;115(7):928–935.
Psaty BM, Anderson M, Kronmal RA, et al. The association
between lipid levels and the risks of incident myocardial infarction, stroke, and total mortality: the Cardiovascular Health
Study. J Am Geriatr Soc. 2004;52(10):1639–1647.
Yudkin JS, Kumari M, Humphries SE, et al. Inflammation,
obesity, stress and coronary heart disease: is interleukin-6 the
link? Atherosclerosis. 2000;148(2):209–214.