Predicting the 30-Year Risk of Cardiovascular Disease

The Framingham Heart Study

Michael J. Pencina, PhD; Ralph B. D’Agostino, Sr, PhD; Martin G. Larson, ScD;
Joseph M. Massaro, PhD; Ramachandran S. Vasan, MD

Background—Present cardiovascular disease (CVD) risk prediction algorithms were developed for a ⱕ10-year follow up
period. Clustering of risk factors at younger ages and increasing life expectancy suggest the need for longer-term risk
prediction tools.
Methods and Results—We prospectively followed 4506 participants (2333 women) of the Framingham Offspring cohort
aged 20 to 59 years and free of CVD and cancer at baseline examination in 1971–1974 for the development of “hard”
CVD events (coronary death, myocardial infarction, stroke). We used a modified Cox model that allows adjustment for
competing risk of noncardiovascular death to construct a prediction algorithm for 30-year risk of hard CVD.
Cross-validated survival C statistic and calibration ␹2 were used to assess model performance. The 30-year hard CVD
event rates adjusted for the competing risk of death were 7.6% for women and 18.3% for men. Standard risk factors
(male sex, systolic blood pressure, antihypertensive treatment, total and high-density lipoprotein cholesterol, smoking,
and diabetes mellitus), measured at baseline, were significantly related to the incidence of hard CVD and remained
significant when updated regularly on follow-up. Body mass index was associated positively with 30-year risk of hard
CVD only in models that did not update risk factors. Model performance was excellent as indicated by cross-validated
discrimination C⫽0.803 and calibration ␹2⫽4.25 (P⫽0.894). In contrast, 30-year risk predictions based on different
applications of 10-year functions proved inadequate.
Conclusions—Standard risk factors remain strong predictors of hard CVD over extended follow-up. Thirty-year risk prediction
functions offer additional risk burden information that complements that of 10-year functions. (Circulation. 2009;119:3078-
3084.)
Key Words: atherosclerosis 䡲 competing risk 䡲 lifetime risk 䡲 obesity 䡲 risk factors
Downloaded from [Link] by on August 22, 2020

I dentification of risk factors contributing to the incidence of

cardiovascular disease (CVD) is 1 of the major accom-
plishments of 20th century epidemiology. Going a step
the importance of risk factor levels in early adulthood on the
long-term risks of CVD as well as the substantial impact of
CVD risk factors on all-cause mortality. They also suggested
further, researchers were able to construct multivariable risk that 10-year functions may underestimate the true risk bur-
prediction algorithms intended to aid clinicians in risk assess- den, particularly in younger individuals and women. These
ment. The importance of these algorithms was underscored results underscore the need for long-term CVD risk predic-
by their incorporation into the treatment recommendations of tion models applicable to younger adults that account for the
the Third Adult Treatment Panel.1 Multiple risk scores have competing cause of non-CVD mortality. The need for long-
been proposed in the literature in the last 20 years.2–13 These term CVD risk prediction models was recently articulated by
have all been developed for risk assessment over a ⱕ10-year Blumenthal et al27 and Sniderman and Furberg.28 However, to
horizon. Some experts articulated the need to know the the best of our knowledge, no algorithm has been proposed to
longer-term risk to better understand the public health burden quantify 30-year risk of CVD as a direct function of risk
and the true need for intervention.14 As an answer to this factors (allowing risk assessment for any combination of risk
need, several reports presented the lifetime or long-term risks factors). This can be explained in part by the difficulty of
of CVD, coronary heart disease (CHD), and stroke and their finding a cohort with a sufficiently long and rigorous
risk factors.15–23 Some investigators attempted to calculate follow-up and also by the methodological complexities asso-
lifetime and long-term risks within the categories of specific ciated with incorporating the competing risk of death due to
risk factors or their clusters.15,24 –26 Their findings emphasized other causes into the multivariable risk estimation.

Received September 3, 2008; accepted April 6, 2009.

From the Department of Mathematics and Statistics (M.J.P., R.B.D., M.G.L., J.M.M.), School of Medicine (R.S.V.), and Department of Biostatistics
(M.J.P., R.B.D., J.M.M.), Boston University, Boston Mass; and Framingham Heart Study, Framingham, Mass (M.J.P., R.B.D., M.G.L., J.M.M., R.S.V.).
Guest Editor for this article was Mary Cushman, MD.
The online-only Data Supplement is available with this article at [Link]
Correspondence to Michael J. Pencina, PhD, Department of Biostatistics, Boston University, 111 Cummington St, Boston, MA 02215. E-mail
mpencina@[Link]
© 2009 American Heart Association, Inc.
Circulation is available at [Link] DOI: 10.1161/CIRCULATIONAHA.108.816694

3078
 Pencina et al Thirty-Year Risk of Cardiovascular Disease 3079

Clinical Perspective on p 3084 Î CVD 共30兲⫽ 冘

ti⬍30
ˆ␭ CVD 共ti兲Ŝ共ti⫺1兲.

In this report, we develop a tool for estimating 30-year risk of

The quantities under summation denote the instantaneous hazard
hard CVD events among individuals free of the condition at of CVD at event time ti and survival rate from both CVD and
baseline. Our risk estimates allow for an adjustment for the noncardiovascular death past event time ti⫺1. Further statistical
competing risk of non-CVD death and utilize the standard risk details including estimation techniques are presented in the Techni-
factors that can be collected during a physician’s office visit. The cal Appendix in the online-only Data Supplement. The assumption
of linearity for all predictors was verified with cumulative sums of
tool is based on the Framingham Offspring cohort that has
martingale residuals as described by Lin et al.38 Given no significant
contributed to the creation of several successful risk score interactions with sex, the final model was sex-pooled but adjusted for
algorithms2–5,29 and offers over 35 years of rigorous surveillance sex. Likewise, no effect modification by baseline age was detected.
for CVD occurrence. Its performance is contrasted with methods We also verified that the choice of time scale (time on study versus
based on long-term applications of 10-year risks. age of onset39,40) did not affect the results. In addition to standard
factors (systolic BP [SBP] and antihypertensive treatment, total and
HDL cholesterol, smoking, and diabetes mellitus) used in CVD risk
Methods prediction (see D’Agostino et al12), we considered diastolic BP
The Framingham Heart Study started in 1948 with the enrollment of (DBP), triglycerides, and LDL cholesterol as a replacement for total
the “original” cohort of 5209 individuals. In 1971, some 5124 cholesterol (see Wilson et al3). Moreover, given recent reports
offspring of the original cohort and their spouses were enrolled into underscoring the usefulness of BMI in cardiovascular risk prediction
the Framingham Offspring Study.30 Constant monitoring of CVD models, we included it as a candidate risk factor (see References 12,
events and mortality has been performed and was available through 41, and 42). Continuous variables were log-transformed to decrease
the end of 2007 for this investigation. Attendees of the first offspring the impact of extreme observations.
examination were eligible for analysis if they were ⱖ20 and ⬍60 To assess model performance, we used the discrimination C
years of age (n⫽4828), were free of CVD (n⫽4758) and cancer at statistic, which takes into account the timing of events, as proposed
baseline (n⫽4723), were not lost to follow-up (n⫽4680), and had a by Harrell et al43 and Pencina and D’Agostino,44 and D’Agostino’s
complete risk factor profile, yielding a final sample of 4506 and Nam’s45 modification of the Hosmer-Lemeshow calibration ␹2
individuals (2333 women; mean age, 37 years). All participants gave with survival estimates adjusted for the competing risk of noncar-
written informed consent, and the study protocol was approved by diovascular death.37 Five-fold cross-validation46 was used to account
the institutional review board of Boston Medical Center. for the fact that we evaluated the model on the same data on which
A detailed physical examination, anthropometry, blood pressure it was developed; in this way, we were able to utilize all data
(BP) determination, and phlebotomy for vascular risk factors were available while correcting for potential overoptimism in the assess-
conducted at each Heart Study examination, as described by ment of model performance. Additionally, we performed internal
D’Agostino et al.12 Body mass index (BMI) was calculated as the validation by randomly splitting the sample 2:1 and developing the
weight in kilograms divided by the square of height in meters. function on the first two thirds and evaluating its performance on the
Downloaded from [Link] by on August 22, 2020

Antihypertensive medication use was ascertained by the physician remaining third. Net reclassification improvement, as proposed by
examiner at the Heart Study and based on self-report. Serum total Pencina and D’Agostino et al,47 was used to assess the clinical utility
and high-density lipoprotein (HDL) cholesterol and triglyceride of additional variables and different ways of estimating 30-year risk
levels were determined by standardized enzymatic methods. Low- (see below). Third Adult Treatment Panel– based1 cutoff points
density lipoprotein (LDL) cholesterol levels were calculated with the determining categories of low, intermediate, and high risk were
use of the Friedewald formula.31 Cigarette smoking in the year adjusted proportionally to the increased duration of follow-up and
preceding the examination was ascertained by self-report. Diabetes incidence (from 6% and 20% to 12% and 40%).
mellitus was defined as fasting glucose ⱖ126 mg/dL or use of insulin An Excel risk score calculator was constructed to facilitate applica-
or oral hypoglycemic medications. tion by clinicians and is available in the online-only Data Supplement.
Participants were followed for a maximum of 35 years (median, All probability values reported were 2-sided, and a conservative 0.01
32 years). We focused on “hard” CVD as the primary outcome of level of significance was adopted to avoid inclusion of weak effects.
interest and defined it as a composite of hard CHD (coronary death, SAS version 9.1 was used to perform all analyses.48
myocardial infarction) and stroke (fatal and nonfatal). Full CVD, as
defined as in D’Agostino et al12 (hard CHD plus coronary insuffi- Time-Dependent Analysis
ciency and angina pectoris, stroke plus transient ischemic attack, Given the long-term follow-up and the fact that all risk factors were
intermittent claudication, and congestive heart failure), was used as reassessed regularly approximately every 4 years between the 1970s
a secondary outcome. Medical histories, physical examinations at the and early 2000s, we performed an additional analysis updating all
study clinic, hospitalization records, and communication with per- variables as soon as the new values became available. This resulted
sonal physicians were all used to obtain information about CVD in a Cox regression with time-dependent covariates that corresponds
events on follow-up. to a short-term risk assessment. The results were contrasted with
those obtained for the 30-year model developed without updating.
Statistical Analysis
In our primary model, we assessed the effect of risk factors measured Comparison With Alternative Approaches for
at baseline on the long-term (30-year) risk of hard CVD using Cox 30-Year Risk Prediction
regression.32 In a secondary model, we used full CVD as outcome. The following approaches were considered:
Considering the extensive length of follow-up and the potential bias
due to the competing risk of noncardiovascular mortality in the 1. Naive. This method calculated the 30-year risk as 3 times the
prediction of long-term risk, we employed the model given by 10-year risk from the model that did not account for the competing
Andersen et al33–36 to adjust the risk estimates for the competing risk risks. It ignores aging as a key determinant of CVD risk, and
of non-CVD mortality. The standard Cox model, similar to the therefore we know a priori that it cannot be correct. However,
standard Kaplan–Meier estimator, may provide biased estimates of given its simplicity it might seem attractive, and we wanted to
absolute long-term risk because it fails to treat those who die of assess the amount of bias that it would introduce.
noncardiovascular causes as ineligible for development of CVD 2. Combined. This approach utilized an application of 10-year CVD
events.37 The competing risk model corrects this shortcoming by risk calculators. For fairness of comparison, we estimated 10-year
calculating the cumulative incidence of CVD in the following manner: probabilities of survival based on our data. Three probabilities
 3080 Circulation June 23, 2009

Table 1. Baseline Characteristics and Incident Events

Clinical Features Women (n⫽2333) Men (n⫽2173)
Age, y 36.3⫾9.3 37.3⫾9.2
Age group, %
20–29 y 28.0 23.9
30–39 y 33.9 34.9
40–49 y 28.9 29.7
50–59 y 9.2 11.5
SBP, mm Hg 118⫾16 126⫾15 Figure 1. Thirty-year risk of hard CVD in women and men by age
group, adjusted for the competing risk of noncardiovascular death.
DBP, mm Hg 76⫾10 82⫾11
Antihypertensive treatment, % 2.7 3.1 previously from the Framingham data.23 This can be ex-
Total cholesterol, mg/dL 192⫾38 202⫾39 plained by the substantially younger ages of our cohort and
HDL cholesterol, mg/dL 57⫾15 44⫾12 the different definition of the end point of interest (only hard
LDL cholesterol, mg/dL 120⫾35 135⫾35 events were considered in our analysis). Event rates by sex
Triglycerides, mg/dL 77⫾73 115⫾97 and age decade are presented in Figure 1.
BMI, kg/m2 23.9⫾4.5 26.5⫾3.6 Standard CVD risk factors (male sex, age, SBP, antihyper-
Smoking, % 45.0 46.2 tensive treatment, total and HDL cholesterol, smoking, and
diabetes mellitus) were highly significant (0.01 level) in the
Diabetes mellitus, % 0.9 2.6
multivariable model. DBP and triglycerides were not statis-
Incident events, n
tically significant, and inclusion of LDL in place of total
Cardiovascular death 22 67
cholesterol did not improve model performance. BMI was
Myocardial infarction 111 281 weakly significant in the final model (P⫽0.04); it did not
Fatal and nonfatal stroke 86 104 increase the C statistic and had a nonsignificant net reclassi-
Values are mean⫾SD for continuous variables or percentages for categorical fication improvement of ⬍1%. Hence, we decided not to
variables. include it in the main risk prediction model. However,
following the example of D’Agostino et al,12 we constructed
were calculated for each person: the first using baseline age, a simplified office-based risk model in which BMI replaced
Downloaded from [Link] by on August 22, 2020

second using baseline age plus 10 years, and third using baseline the lipids. It was highly significant in the simple model along
age plus 20 years, with all other risk factors based on the baseline
values. The 30-year risk was calculated as the difference of 1
with all other risk factors (Pⱕ0.01). Hazard ratios with CIs
minus the product of these three 10-year probabilities. for both models are presented in Table 2. Corresponding
3. Unadjusted. Here we applied the standard Cox model to our data results for the secondary end point of full CVD are given in
with full follow-up, ignoring competing risk of death. the Table in the online-only Data Supplement.
4. Adjusted. This is our main approach following the aforementioned The 30-year risk model offered excellent discrimination
model.
(cross-validated C statistic⫽0.803; 95% CI, 0.786 to 0.820;
The aforementioned 4 methods were applied to individuals with internally validated C statistic⫽0.802; 95% CI, 0.772 to 0.832)
different combinations of risk factors for hard CVD events. and calibration (cross-validated Nam-D’Agostino ␹2⫽4.25;
The authors had full access to and take full responsibility for the P⫽0.894; Figure 2; internally validated ␹2⫽3.98; P⫽0.913). It
integrity of the data. All authors have read and agree to the is important to note that our model, which adjusted for the
manuscript as written.
competing risk of noncardiovascular death, improved the model
Results calibration compared with the model that ignored the competing
The sex-specific risk factor profile of our sample at baseline
as well as number and type of hard CVD events are given in Table 2. Hazard Ratios With 95% CIs for 30-Year Risk of
Hard CVD
Table 1. In our baseline cohort aged ⱖ20 but ⬍60 years, men
had numerically higher levels of all risk factors except HDL Variables Main Model Simple Model
cholesterol. The 3 younger age decades (20 to 29, 30 to 39, 40 Male sex 1.73 (1.45, 2.07) 2.08 (1.77, 2.46)
to 49 years) had similar representation between sexes, with Age 2.09 (1.88, 2.31) 2.22 (2.01, 2.45)
the fewest people in the oldest age group (50 to 59 years). SBP 1.29 (1.19, 1.39) 1.26 (1.16, 1.36)
Over a maximum of 35 years of follow-up, 671 participants
Antihypertensive treatment 1.48 (1.10, 2.00) 1.48 (1.09, 2.00)
(219 women) experienced a first hard CVD event, and 622
Smoking 2.01 (1.72, 2.35) 2.21 (1.90, 2.58)
(267 women) died of non-CVD causes. Of note, strokes
constituted almost 40% of CVD events experienced by women Diabetes mellitus 2.49 (1.82, 3.41) 2.82 (2.07, 3.84)
but ⬍25% of events experienced by men. The 30-year Total cholesterol 1.33 (1.23, 1.44) 䡠䡠䡠
Kaplan–Meier rate of hard CVD adjusted for the competing HDL cholesterol 0.78 (0.72, 0.84) 䡠䡠䡠
risk of non-CVD death (with the use of the adjustment of BMI 䡠䡠䡠 1.20 (1.10, 1.30)
Gaynor et al37) was 7.6% for women and 18.3% for men. Hazard ratios for continuous risk factors are given per 1-SD increase in the
These are lower than the long-term and lifetime risks reported natural logarithm. All Pⱕ0.01.
 Pencina et al Thirty-Year Risk of Cardiovascular Disease 3081

Figure 2. Calibration by decile, adjusted for the competing risk

of noncardiovascular death.

Figure 4. Ten- vs 30-year risk of hard CVD for 25-year-old men

risk (cross-validated Nam-D’Agostino ␹2 for the model that with different risk profiles. No risk factors profile: total cholester-
ignored the competing risk⫽18.7; P⫽0.027). ol⫽150 mg/dL; HDL cholesterol⫽60 mg/dL; untreated
Figures 3 and 4 contrast the estimated 30-year risks of hard SBP⫽110 mm Hg; nonsmoker; nondiabetic. Adverse lipids: total
cholesterol⫽260 mg/dL; HDL cholesterol⫽35 mg/dL. Hyperten-
CVD adjusted for the competing risk of noncardiovascular sion: SBP⫽160 mm Hg, untreated.
death with 10-year risks, calculated for 25-year-old women
and men, respectively. The corresponding estimates for 45-
year-old subjects are given in Figures 5 and 6. These ages long-term, 30-year model (hazard ratio⫽1.10 per 1 SD;
were selected to illustrate CVD risk burden in young adults P⫽0.04) but lost its entire impact in the time-dependent
with varied combinations of risk factors as well as in model (hazard ratio⫽0.99; P⫽0.82). This finding illustrates
middle-aged adults. The results for 25-year-old subjects are how the effect of BMI is mediated through other risk factors:
striking, especially for women. The 10-year models suggest It is present in the 30-year risk model when the follow-up is
negligible risk levels (⬍2.5% in women and 5% in men), extended for a long period from the baseline, but then it
whereas the 30-year models give estimates that are almost 10 affects the individual risk factors, and after we control for this
times higher. For example, 10-year risk for a 25-year-old impact in time-updated models, BMI loses its significance.
smoking woman with adverse lipid profile and hypertension
is only 1.4%, but the corresponding 30-year risk reaches 12%. Comparison With Alternative Approaches for
30-Year Risk Prediction
The mean estimated 30-year risks based on our model (the
Downloaded from [Link] by on August 22, 2020

Results of Time-Dependent Analysis

In time-dependent analysis updating all variables approxi- “adjusted” approach) were 7.9% for women and 18.0% for
mately every 4 years, all standard risk factors remained men (as expected, very close to the incidence rates given
significantly related to the hard CVD outcome with hazard above). As expected, the “naive” approach consistently un-
ratios similar to those obtained in 30-year risk models (Table derestimated the true risk; the mean risks were 4.1% for
3). Smoking had the biggest numerical change: The hazard women and 13.3% for men. If we ignored the competing risk
ratio increased by approximately one third in the time- of noncardiovascular death (“unadjusted” approach), the
dependent model. This can be explained by the fact that mean risks increased to 8.6% and 20.4%, respectively. The
time-updated models focus on a shorter event follow-up and risks based on the “combined” approach averaged across our
can account for changes in risk factor levels (smoking cohort were even higher; however, the relationship varied
cessation in this case). across individuals with different levels of risk factors. When
The most interesting change occurred in the impact of BMI we applied these approaches to calculate 30-year risks for
on the risk of hard CVD. BMI was weakly significant in the individuals with different combinations of risk factors, the

Figure 3. Ten- vs 30-year risk of hard CVD for 25-year-old Figure 5. Ten- vs 30-year risk of hard CVD for 45-year-old
women with different risk profiles. No risk factors profile: total women with different risk profiles. No risk factors profile: total
cholesterol⫽150 mg/dL; HDL cholesterol⫽60 mg/dL; untreated cholesterol⫽150 mg/dL; HDL cholesterol⫽60 mg/dL; untreated
SBP⫽110 mm Hg; nonsmoker; nondiabetic. Adverse lipids: total SBP⫽110 mm Hg; nonsmoker; nondiabetic. Adverse lipids: total
cholesterol⫽260 mg/dL; HDL cholesterol⫽35 mg/dL. Hyperten- cholesterol⫽260 mg/dL; HDL cholesterol⫽35 mg/dL. Hyperten-
sion: SBP⫽160 mm Hg, untreated. sion: SBP⫽160 mm Hg, untreated.
 3082 Circulation June 23, 2009

Table 3. Hazard Ratios With 95% CIs for 30-Year Risk of Hard
CVD vs Model With Time-Dependent Risk Factors Updated
Every 4 Years
Variables 30-Year Risk Model Time-Updated Model
Male sex 1.72 (1.44, 2.05) 2.05 (1.72, 2.44)
Age 2.08 (1.88, 2.31) 2.18 (1.97, 2.42)
SBP 1.26 (1.16, 1.37) 1.28 (1.19, 1.39)
Antihypertensive treatment 1.48 (1.10, 2.00) 1.36 (1.14, 1.62)
Smoking 2.04 (1.74, 2.38) 2.74 (2.32, 3.24)
Diabetes mellitus 2.42 (1.77, 3.31) 2.30 (1.89, 2.81)
Figure 6. Ten- vs 30-year risk of hard CVD for 45-year-old men Total cholesterol 1.32 (1.22, 1.43) 1.23 (1.14, 1.33)
with different risk profiles. No risk factors profile: total cholester-
ol⫽150 mg/dL; HDL cholesterol⫽60 mg/dL; untreated HDL cholesterol 0.80 (0.73, 0.87) 0.75 (0.68, 0.81)
SBP⫽110 mm Hg; nonsmoker; nondiabetic. Adverse lipids: total BMI 1.10* (1.00, 1.20) 0.99† (0.91, 1.08)
cholesterol⫽260 mg/dL; HDL cholesterol⫽35 mg/dL. Hyperten-
sion: SBP⫽160 mm Hg, untreated. Hazard ratios for continuous risk factors are given per 1-SD increase in the
natural logarithm. All Pⱕ0.01 except as noted.
*P⫽0.04.
unadjusted approach consistently overestimated the correct †P⫽0.82.
predictions based on the adjusted model. The combined
approach underestimated the true risk for people with lower
30-year risk despite young age, an effect that is entirely
risk (younger and with fewer risk factors) and overestimated
missed by the 10-year risk model. Moreover, most contem-
the risk in people with higher risk (older with several risk
porary cohorts are confounded by treatment effects that
factors). Differences were more pronounced for higher risk
significantly influence short-term prediction. The use of
levels (⬎20%). There was a 10% (95% CI, 6% to 14%) net
30-year instruments might partially overcome this problem
reclassification improvement resulting from using the ad-
with its long-term focus. Effective risk communication is
justed 30-year risk estimates over the tripled 10-year risks
(naive approach) but no improvement when compared with another reason why 30-year risk might be helpful. Individuals
the unadjusted or combined approaches. might be more likely to adopt necessary lifestyle changes on
hearing that their 30-year risk of CVD is 1 in 8 (75th percentile
of the 30-year risk in men aged ⬍40 years) than when they are
Downloaded from [Link] by on August 22, 2020

Discussion
In this report, we present a simple way to estimate 30-year risk told it is 1 in 50 in 10 years (75th percentile of the 10-year risk
of hard CVD based on risk factors routinely measured during an in men aged ⬍40 years). Moreover, the potent impact of
office visit. The results are based on ⬎30 years of rigorous accumulation of risk factors as presented in Figures 3 through 6
follow-up and ascertainment of CVD incidence and death. Our may serve as an effective risk communication tool.
algorithm allows for risk assessment for individuals with any We have shown that established CVD risk factors that are
combination of continuous and categorical risk factors. It also significant in models based on shorter follow-up duration3,4,12
accounts for the competing risk of noncardiovascular death. are also significantly related to hard CVD incidence in 30
Our approach is based on advanced statistical techniques years. The impact of risk factors measured only at baseline is
that allow avoiding bias in the assessment of true absolute similar to that of risk factors updated regularly at follow-up.
risk. Ignoring the competing risk of death inflates the esti- The same is not true for BMI, which loses its independent
mates by an average of 1% to 2% on the absolute scale (or impact when other covariates are time updated. No significant
10% on the relative scale), which leads to inferior calibration effect modifications by sex were detected despite differences
as demonstrated in the ␹2 statistics. On the other hand, in hard CVD composition, with strokes comprising almost
simpler approaches that try to make 30-year inferences on the 40% of all first events in women and ⬍25% in men.
basis of a 10-year risk model are inadequate and may lead to As indicated earlier, this is the first report to our knowledge
underestimation or overestimation of the true risk burden. that presents a risk score for incidence of hard CVD in the
The need for long-term risk prediction tools has been 30-year horizon. In their recent publications, researchers from
articulated for many years14 as a complement for the shorter- the Chicago Heart Association Detection Project in Industry
term calculators. There are several reasons why it is neces- calculate the remaining lifetime risk until age 85 years adjusted
sary. As Sniderman and Furberg28 point out, studies with for the competing risk of death for participants aged 40 to 59
shorter follow-up miss cases that would be found if the with 0 to 5 elevated CVD risk factors24 and quantify the effect of
duration was extended and thus “restrict our appreciation of these standard risk factors on the 30-year risk of CVD, coronary,
the true importance of the modifiable factors that cause and all-cause mortality in women aged 18 to 39 years.25
vascular disease.” As seen here, 30-year risk cannot be Although their reports offer valuable insights into the effect of
adequately replaced by different combinations of 10-year risk factors on the long-term risk of CVD, they were not designed
risks. Blumenthal et al27 raise the issue of high lifetime risk of for individual-specific risk prediction in a clinical setting.
CVD in women to underscore the need for long-term risk The analysis of changes in impact on the risk of CHD and
prediction algorithms. This point finds a striking illustration death of baseline risk factors during long-term follow-up under-
in our data, in which adverse risk factor profile leads to high taken by the Chicago Heart Association authors24 –26 as well as
 Pencina et al Thirty-Year Risk of Cardiovascular Disease 3083

Menotti and Lanti49 revealed that single-occasion measurement Framingham Heart Study research is supported by National Institutes
risk factors remain strong predictors even in the long-term of Health/National Heart, Lung, and Blood Institute contract No.
models. This finding has been confirmed in our setting. N01-HC-25195. Dr Vasan is supported in part by 2K24 HL04334
(National Heart, Lung, and Blood Institute).
A few limitations of our study need to be acknowledged.
First, our results and the risk score were derived on the basis
Disclosures
of a white cohort, which potentially limits the generalizability Dr D’Agostino has served as a consultant or on the advisory board
to other ethnic groups. Appropriate recalibration (see for Sanofi and Pfizer. The other authors have nothing to report
D’Agostino et al29) may correct differences in baseline beyond the funding sources listed above.
survival between ethnicities, but further investigation is
warranted to determine the impact of relative risks for CVD References
and the competing risk of non-CVD mortality differing 1. Executive Summary of the Third Report of the National Cholesterol
Education Program (NCEP) Expert Panel on Detection, Evaluation, and
between ethnicities. Second, in the assessment of model
Treatment of High Blood Cholesterol in Adults (Adult Treatment Panel
performance, we accounted for overoptimism introduced by III). JAMA. 2001;285:2486 –2497.
evaluating the model on the same data on which it was 2. Anderson KM, Odell PM, Wilson PWF, Kannel WB. Cardiovascular
developed using 5-fold cross-validation and internal valida- disease risk profiles. Am Heart J. 1991;121:293–298.
3. Wilson PWF, D’Agostino RB, Levy D, Belanger A, Silbershatz H,
tion. Although these techniques are well suited for this
Kannel WB. Prediction of coronary heart disease using risk factor cate-
purpose, they cannot be equated with the preferred method of gories. Circulation. 1998;97:1837–1847.
validation in a different cohort. Third, we considered only 4. D’Agostino RB, Russell MW, Huse DM, Ellison RC, Silbershatz H,
standard risk factors and did not investigate the effect of Wilson PFW, Hartz SC. Primary and subsequent coronary risk appraisal:
new results from the Framingham Study. Am Heart J. 2000;139:272–281.
novel biomarkers on the risk of hard CVD because they were
5. D’Agostino RB, Wolf PA, Belanger A, Kannel WB. Stroke risk profile:
not available at the baseline examination in the early 1970s. adjustment for antihypertensive medication. Stroke. 1994;25:40 – 43.
Wilson et al50 have shown recently that C-reactive protein 6. Assmann G, Cullen P, Schulte H. Simple scoring scheme for calculating
measured in the Framingham cohort between the late 1970s the risk of acute coronary events based on the 10-year follow-up of the
Prospective Cardiovascular Munster (PROCAM) Study. Circulation.
and early 1980s might contribute to improvement in risk
2002;105:310 –315.
reclassification, a finding postulated before by Ridker et 7. Conroy RM, Pyorala K, Fitzgerald AP, Sans S, Menotti A, DeBacker G,
al10,13 on the basis of other cohorts. It is plausible that the use DeBacquer D, Ducimetiere P, Jousilahti P, Keil U, Njolstad I, Oganov RG,
of novel biomarkers could help to reduce the amount of risk Thomsen T, Tunstall-Pedoe H, Tverdal A, Wedel H, Whincup P, Wil-
helmsen L, Graham IM. Estimation of ten-year risk of fatal cardiovascular
underestimation with 10-year models compared with 30-year disease in Europe: the SCORE project. Eur Heart J. 2003;24:987–1003.
models. Further research is needed to investigate this issue. 8. Hippisley-Cox J, Coupland C, Vinogradova Y, Robson J, May M, Brindle
Downloaded from [Link] by on August 22, 2020

Fourth, the effect and interplay of risk factors might be P. Derivation and validation of QRISK, a new cardiovascular disease risk
different now than it was 30 years ago. However, no other score for the United Kingdom: prospective open cohort study. BMJ.
2007;335:136.
way of constructing a 30-year risk score is possible given the 9. Menotti A, Lanti M, Gabiti-Rosei E, Carratelli L, Cavera G, Dormi A,
length of the time horizon. Fifth, the nature of our design did Gaddi A, Mancini M, Motolese M, Muiesan ML, Muntoni S, Notar-
not account for changes in risk factor levels that can take bartolo A, Prati P, Remiddi S, Zanchetti A. Riskard 2005: new tools for
place during the course of follow-up. For example, smoking prediction of cardiovascular disease risk derived from Italian population
studies. Nutr Metab Cardiovasc Dis. 2005;15:426 – 40.
cessation between the early 1970s and the present time period 10. Ridker PM, Buring JE, Rifai N, Cook RN. Development and validation of
might have reduced the true absolute risk, leading to an improved algorithms for the assessment of global cardiovascular risk in
underestimation of the effect of continued smoking. How- women: the Reynolds risk score. JAMA. 2007;297:611– 619.
ever, we have shown that regardless of the absence or 11. Woodward M, Brindle P, Tunstall-Pedoe, for the SIGN Group on Risk
Estimation. Adding social deprivation and family history to cardiovascu-
presence of risk factor adjustment on follow-up, they remain lar risk assessment: the ASSIGN score from the Scottish Heart Health
strong independent predictors of hard CVD. Finally, we did Extended Cohort (SHHEC). Heart. 2007;93:172–176.
not attempt to predict the 30-year risk of death from noncar- 12. D’Agostino RB, Vasan RS, Pencina MJ, Wolf PA, Cobain MR, Massaro
diovascular causes; however, this risk was implicitly factored JM, Kannel WB. General cardiovascular risk profile for use in primary
care. Circulation. 2008;117:743–753.
into the calculations as the competing cause. Further research 13. Ridker PM, Paynter NP, Rifai N, Gaziano JM, Cook NR. The Reynolds
is needed to provide estimators of long-term risks of all-cause risk score for men. Circulation. 2008;118:2243–2251.
and noncardiovascular mortality that would complement the 14. Grundy SM, Pasternak R, Greenland P, Smith S, Fuster V. Assessment of
predictions available in this report and allow for patient- cardiovascular risk by use of multiple-risk-factor assessment equations.
Circulation. 1999;100:1481–1492.
specific treatment strategies. 15. Lloyd-Jones DM, Wilson PFW, Larson MG, Leip E, Beiser A,
We hope that the simple way of quantifying 30-year risk of D’Agostino RB, Cleeman JI, Levy D. Lifetime risk of coronary heart
hard CVD based on a combination of standard risk factors disease by cholesterol levels at selected ages. Arch Intern Med. 2003;
and additional insights into the nature of their effect presented 163:1966 –1972.
16. Lloyd-Jones DM, Larson MG, Leip EP, Beiser A, D’Agostino RB,
in this report will complement the currently available 10-year Kannel WB, Murabito JM, Vasan RS, Benjamin EJ, Levy D. Lifetime
risk algorithms and serve as a useful tool in the clinical and risk for developing congestive heart failure. Circulation. 2002;106:
public health settings and provide a useful framework for 3068 –3072.
future research. 17. Seshadri S, Beiser A, Kelly-Hayes M, Kase CS, Au R, Kennel WB, Wolf
PA. The lifetime risk of stroke: estimates from the Framingham Study.
Stroke. 2006;37:345–350.
Sources of Funding 18. Vasan RS, Beiser A, Seshadri S, Larson MG, Kannel WB, D’Agostino
This study was supported by the National Heart, Lung, and Blood RB, Levy D. Residual lifetime risk of developing hypertension in
Institute’s Framingham Heart Study, National Institutes of Health. middle-aged women and men. JAMA. 2002;287:1003–1010.
 3084 Circulation June 23, 2009

19. Vasan RS, Pencina MJ, Cobain MR, Freiberg MS, D’Agostino RB. competing risks survival data. Comput Meth Programs Biomed. 2004;74:
Estimated risks for developing obesity in the Framingham Heart Study. 69 –75.
Ann Intern Med. 2005;143:473– 480. 35. Putter H, Fiocco M, Geskus RB. Tutorial in biostatistics: competing risks
20. Cobain MR, Pencina MJ, D’Agostino RB, Vasan RS. Lifetime risk for and multi-state models. Stat Med. 2007;26:2389 –2430.
developing dyslipidemia: the Framingham Offspring Study. Am J Med. 36. Kalbfleisch JD, Prentice RL. The Statistical Analysis of Failure Time
2007;120:623– 630. Data. 2nd ed. Hoboken, NJ: Wiley; 2002.
21. Narayan KMV, Boyle JP, Thompson TJ, Sorensen SW, Williamson DF. 37. Gaynor JJ, Feuer EJ, Tan CC, Wu DH, Little CR, Straus DJ, Clarkson
Lifetime risk for diabetes mellitus in the United States. JAMA. 2003;290: BD, Brennan MF. On the use of cause-specific failure and conditional
1884 –1890. failure probabilities: examples from clinical oncology data. J Am Stat
22. Pekkanen J, Nissinen A, Puska P, Punsar S, Karvonen M. Risk factors and Assoc. 1993;88:400 – 409.
25 year risk of coronary heart disease in a male population with a high 38. Lin D, Wei L, Ying Z. Checking the Cox model with cumulative sums of
incidence of the disease: the Finnish cohorts of the Seven Countries martingale-based residuals. Biometrika. 1993;80:557–572.
Study. BMJ. 1989;299:81– 85. 39. Korn EL, Graubard BI, Midhune D. Time-to-event analysis of longi-
23. Lloyd-Jones DM, Leip EP, Larson MG, D’Agostino RB, Beiser A, tudinal follow-up of a survey: choice of time scale. Am J Epidemiol.
Wilson PFW, Wolf PA, Levy D. Prediction of lifetime risk of cardiovas- 1997;145:72– 80.
cular disease by risk factor burden at 50 years of age. Circulation. 40. Pencina MJ, Larson MG, D’Agostino RB. Choice of timescale and its
2006;113:791–798. effect on significance of predictors in longitudinal studies. Stat Med.
24. Lloyd-Jones DM, Dyer AR, Wang K, Daviglus ML, Greenland P. Risk 2007;26:1346 –1359.
factor burden in middle age and lifetime risks for cardiovascular and 41. Gaziano TA, Young CR, Fitzmaurice G, Atwood S, Gaziano JM.
non-cardiovascular death (Chicago Heart Association Detection Project Laboratory-based versus non-laboratory-based method for assessment of
in Industry). Am J Cardiol. 2007;99:535–540. cardiovascular disease risk: the NHANES I Follow-up Study cohort.
25. Daviglus ML, Stampler J, Pirzada A, Yan LL, Garside DB, Liu K, Wang Lancet. 2008;371:923–931.
K, Dyer AR, Lloyd-Jones DM, Greenland P. Favorable cardiovascular 42. Wilson PFW, Bozeman SR, Burton TM, Hoaglin DC, Ben-Joseph R,
risk profile in young women and long-term risk of cardiovascular and Pashos CL. Prediction of first events of coronary heart disease and stroke
all-cause mortality. JAMA. 2004;292:1588 –1592. with consideration of adiposity. Circulation. 2008;118:124 –130.
26. Greenland P, Knoll MD, Stamler J, Neaton JD, Dyer AR, Garside DB, 43. Harrell FE, Lee KL, Mark DB. Tutorial in biostatistics: multivariable prog-
Wilson PFW. Major risk factors as antecedents of fatal and nonfatal nostic models: issues in developing models, evaluating assumptions and
coronary heart disease events. JAMA. 2003;290:891– 897. adequacy, and measuring and reducing errors. Stat Med. 1996;15:361–387.
27. Blumenthal RS, Michos ED, Nasir K. Further improvements in CHD risk 44. Pencina MJ, D’Agostino RB. Overall C as a measure of discrimination in
prediction for women. JAMA. 2007;297:641– 643. survival analysis: model specific population value and confidence interval
28. Sniderman AD, Furberg CD. Age as a modifiable risk factor for cardio- estimation. Stat Med. 2004;23:2109 –2123.
vascular disease. Lancet. 2008;371:1547–1549. 45. D’Agostino RB, Nam BH. Evaluation of the performance of survival
29. D’Agostino RB, Grundy S, Sullivan LM, Wilson PWF. Validation of the analysis models: discrimination and calibration measures. Handbook of
Framingham coronary heart disease prediction scores. JAMA. 2001;286: Statistics. New York, NY: Elsevier; 2004:1–25.
180–187. 46. Hastie T, Tibshirani RFJ. The Elements of Statistical Learning. New
30. Kannel WB, Feinleib M, McNamara PM, Garrison RJ, Castelli WP. An York, NY: Springer; 2001.
investigation of coronary heart disease in families: the Framingham 47. Pencina MJ, D’Agostino RB Sr, D’Agostino RB Jr, Vasan RS. Evaluating
Downloaded from [Link] by on August 22, 2020

Offspring Study. Am J Epidemiol. 1979;110:281–290. the added predictive ability of a new marker: from area under the ROC
31. Friedewald WT, Levy RI, Fredrickson DS. Estimation of the concen- curve to reclassification and beyond. Stat Med. 2008;27:157–172.
tration of low-density lipoprotein cholesterol in plasma, without the use of 48. SAS/STAT Software, Version 9.1, of the SAS System for Windows. Cary,
the preparative ultracentrifuge. Clin Chem. 1972;18:499 –502. NC: SAS Institute Inc; 2002–2003.
32. Cox DR. Regression models and life tables. J Roy Stat Soc B. 1972;34: 49. Menotti A, Lanti M. Coronary risk factors predicting early and late
187–220. coronary deaths. Heart. 2003;89:19 –24.
33. Andersen PK, Borgan O, Gill RD, Keiding N. Statistical Models Based 50. Wilson PFW, Pencina MJ, Jacques P, Selhub J, D’Agostino RB,
on Counting Processes. New York, NY: Springer; 1993. O’Donnell CJ. C-reactive protein and reclassification of cardiovascular
34. Rosthøj S, Andersen PK, Abildstrom SZ. SAS macros for estimation of risk in the Framingham Heart Study. Circ Cardiovasc Qual Outcomes.
the cumulative incidence functions based on a Cox regression model for 2008;1:92–97.

CLINICAL PERSPECTIVE
The impact of standard risk factors (male sex, age, systolic blood pressure, antihypertensive treatment, cholesterol levels,
smoking, and diabetes mellitus) on 10-year risk of coronary or cardiovascular disease (CVD) has been studied extensively,
and reliable algorithms exist for risk prediction. In the present investigation, we evaluated the impact of standard risk
factors on CVD incidence on long-term follow-up (ie, over 30 years). Our observations suggest that standard risk factors
remain highly predictive of CVD risk over a 30-year follow-up period and that their impact is substantial even if levels are
not updated. We also quantified the impact of body mass index on 30-year risk of CVD and observed that its association
with increased CVD risk is mediated partly through promoting higher levels of standard risk factors over a long-term
follow-up. Additionally, we have formulated a 30-year CVD risk prediction algorithm that adjusts for the competing risk
of death on long-term follow-up. Our observations suggest that different applications of 10-year risk prediction functions
for estimating 30-year CVD risk may be suboptimal, especially when applied to younger women and men who have an
adverse risk factor profile. We also have constructed a simple calculator for estimating 30-year risk of CVD that is based
on standard risk factors (with and without knowledge of laboratory values) and that could be implemented easily in primary
care settings.