CARDIOVASCULAR EPIDEMIOLOGY AND RISK PREDICTION
Cardiovascular epidemiology studies the determinants and distribution of cardiovascular disease (CVD). The overarching goal of CVD epidemiology is to reduce the incidence and prevalence of CVD within the population. Cardiovascular epidemiology has provided vital bidirectional connections between basic mechanistic science and clinical research. Through these types of investigations, our understanding of the extent of CVD and its natural history, mechanisms, and underlying pathophysiology is expanding greatly, which provides opportunities for individual-level therapeutic strategies, as well as population-level approaches to reduce the incidence and burden of CVD.
CVD remains the leading cause of death for men and women in the United
States, more than cancer and respiratory disease combined. Within the United
States, most individuals have an approximately 30% chance of dying of CVD and a
66% lifetime risk of CVD. Fortunately, over the past 40 years, there have been
significant and substantial reductions in CVD risk and age-adjusted mortality
in the United States by approximately 40%. However, the burden of CVD in the
United States and globally still remains massive.
Worldwide, CVD is the most common cause of death, accounting for an
estimated 31% of all deaths. Although 22% of reductions in the CVD death rate
have been observed, the total number of CVD deaths continues to increase
because of population growth and aging of the population. The death rate varies
widely across countries, with the highest death rates attributable to CVD in
Russia and the lowest rates seen in Western Europe, North America, and Central
Similarly, there are substantial differences in CVD incidence,
prevalence, and mortality rates seen across different geographic regions, race,
ethnic, and socioeconomic groups within the United States. For example, the
southeastern United States has a substantially higher incidence rate and
mortality associated with stroke. African Americans have long had substantially
higher age-adjusted rates of hypertension, stroke, heart failure, and coronary
heart disease than age-matched whites in the United States.
Epidemiological research has not only provided critical insights into the
prevalence of CVD, but has also identified risk factors and patient
characteristics that predict the presence and development of CVD. Several risk
factors like sex, race, and age are nonmodifiable and enhance our understanding of the risk of an individual, but these
factors are not useful as targets of therapy. In contrast, reducing the
incidence and optimizing levels of modifiable risk factors is the mainstay of
primary and secondary prevention efforts. The identification of modifiable risk factors led to multiple randomized
controlled clinical trials that demonstrated the primacy of risk factor
prevention and their management of CVD risk reduction.
The American Heart Association (AHA) defines ideal blood pressure as
<120/80 mm Hg. Observational cohort studies have consistently demonstrated
increased risk for stroke, heart attack, heart failure, and cardiovascular
mortality across all age groups at blood pressures above this level. On
average, every increase in the systolic blood pressure by 20 mm Hg or diastolic
blood pressure by 10 mm Hg is associated with a doubling in the risk of death
caused by stroke, coronary heart disease, or other vascular disease. Using JNC
7 guidelines, hypertension was defined as a systolic blood pressure greater
than 140 mm Hg, a diastolic blood pressure greater than 90 mm Hg, or the use of
blood pressure–lowering medications. In the fall of 2017, the AHA and ACC
redefined hypertension as blood pressure greater than 130/80 mm Hg.
Using statistics that reflect JNC 7 guidelines, there are an estimated 85.7 million adults in the United States
with hypertension. The prevalence of hypertension increases with age and varies
by race. African Americans have substantially higher rates of hypertension than
age- matched white Americans. Unfortunately, only 76% of those with
hypertension are on antihypertensive medication, and only 54% have their blood
pressure under adequate control.
Hyperlipidemia or Dyslipidemia
Higher levels of total cholesterol, low-density lipoprotein (LDL)
cholesterol, and non–high-density lipoprotein (HDL) cholesterol are associated
with increased risks for atherosclerotic cardiovascular disease (ASCVD). There
is approximately a 50% higher ASCVD risk for every 40 mg/dL increase in total
cholesterol. The association appears log-linear across higher levels of total
cholesterol. Thus, total cholesterol and the atherogenic fractions LDL
cholesterol and non-HDL cholesterol have become targets for therapy with
lifestyle modification and pharmaco- therapy. Conversely, across usual levels
of HDL cholesterol, there is an inverse association with ASCVD risk.
Hyperlipidemia is commonly defined as total cholesterol levels >230 mg/dL. Within the United States,
an estimated 28.5 million adults, or 11.9% of the population, are considered to
have hyperlipidemia. Over the last 14 years, the prevalence of hyperlipidemia
has decreased by approximately 7%. The most recent estimates of mean total
cholesterol, LDL cholesterol, HDL cholesterol, and triglyceride levels in the
United States are 196, 113, 53, and 103.5 mg/dL, respectively.
Cigarette smoking increases ASCVD risk twofold to fourfold. Cessation of tobacco products is associated with
rapid changes in physiology and substantial
reductions in CVD risk. Therefore, tobacco use has been a major target of public health campaigns.
Because of these large efforts, tobacco use has been declining over the last 50
years from prevalence rates of 51% in men and 34% in women in 1965 to 16.7% in
men and 13.7% in women in 2017.
Although the overall trend in tobacco use is promising, certain minority
groups, including sexual and gender minorities, individuals with low
socioeconomic status, disabled persons, and individuals with psychiatric
illness, have not experienced the same decrease in prevalence rates that are
seen in the overall population. Furthermore, recent increases in electronic
cigarette use, particularly in adolescent populations, could result in increased
tobacco use in younger age groups.
The AHA d <100 mg/dL.
Currently, only 56% of adults in the United States meet this criterion.
Diabetes is associated with a twofold to threefold increase in risk for
coronary heart disease, stroke, peripheral artery disease, heart failure, and
atrial fibrillation. It is also associated with a 6- to 8-year shorter life
expectancy than is seen in nondiabetics.
The National Heart, Blood, and Lung Institute defines diabetes as a
fasting blood sugar >125 mg/dL, and it defines prediabetes as a fasting
blood sugar between 100 and 125 mg/dL. In a recent representative survey of the
United States population, 23.4 million adults have diabetes and 7.6 million are
not aware of having diabetes. Furthermore, 81.6 million have prediabetes. Of
the cases of diabetes identified, 90% to 95% are classified as type 2 diabetes.
In part because of the obesity epidemic, the prevalence of diabetes has been
increasing over the last 10 to 15 years. African Americans and Hispanic
Americans have substantially higher rates of diabetes than white Americans.
Obesity, Diet, and Physical Activity
The AHA defines an ideal body weight as a body mass index (BMI) of 18.5 to 25 kg/m2. Obesity is
defined as a BMI >30 kg/m2, and overweight, which also confers an
increased risk for CVD, is defined as a BMI between 25 and 30 kg/m2.
Obesity is a risk factor for CVD, including ASCVD, heart failure, stroke,
venous thromboembolism, and atrial fibrillation, and a risk factor for other
CVD risk factors, including dyslipidemia, hypertension, and diabetes. Obesity
rates have slowly increased over the past several decades, with most recent
prevalence rates reported in 2013 to 2014 as 37.7%. Women have a higher
prevalence of both obesity and class III obesity, defined as BMI >40 kg/m2,
with prevalence rates of 40.4% and 9.9%, respectively. There is substantial
regional variation in obesity prevalence rates, with higher levels observed in
the Midwest and southeastern United States.
A central determinant of the obesity epidemic is caloric excess and
physical inactivity. The AHA defines a healthy diet as one that is rich in
fresh fruits, vegetables, whole grains, low-fat dairy, seafood, legumes, and
nuts. This diet has consistently been found to reduce blood pressure, improve
lipid fractions, and reduce risks for heart attack and stroke. Conversely,
diets that are high in saturated fats and salt, and low in fruits and
vegetables are associated with adverse changes in blood cholesterol and blood
pressure, and likely increase ASCVD risk. Current estimates suggest that only
1.5% of adults consume an ideal healthy diet and that 678,000 deaths per year
are attributable to a sub- optimal diet.
Like obesity and dietary indiscretion, physical inactivity increases
risks for CVD and CVD risk factors. Currently, the AHA recommends that adults
perform at least 150 minutes of moderate intensity exercise per week or 75
minutes of vigorous activity a week plus 2 days of muscle strengthening.
Individuals who meet these recommendations have been found to have a 30% to 40%
lower risk for diabetes, a 30% to 40%
lower mortality risk, and a 20% to 30% lower risk for coronary heart disease.
Forty-four percent of adults meet the criteria specified by the AHA; however,
30% of adults do not engage in any physical activity. Women, older adults,
African Americans, and Hispanics meet these requirements less frequently than
other sex, race, and ethnic groups.
EFFORTS AT CARDIOVASCULAR DISEASE
Cardiovascular risk has a bell-shaped distribution within the population.
A high-risk approach targets individuals at the highest risk with aggressive
risk reduction. This type of approach is currently recommended to determine which
patients should receive statin therapy for primary prevention of ASCVD events.
Because the relative benefits of statin medications are consistent across
absolute levels of risk, the high-risk approach will result in the greatest
absolute benefit in individuals at highest absolute risk.
In contrast, a population-level approach attempts to reduce risk factor
levels in the population as a whole through optimization of population mean
risk factor levels. Although this may seem counter- intuitive, this approach
often results in greater numbers of prevented events because there are much
larger numbers of individuals around the mean value in a bell-shaped
distribution. Therefore, most events occur within this portion of the
population. To continue with the example of impact of cholesterol on ASCVD
events, the risk of an individual for ASCVD with mean cholesterol levels is low
because of the large
denominator of people close to the mean. However, the risk for this segment of
the population is still high. Therefore, even a modest decrease in the mean
cholesterol concentrations would translate into a large number of prevented CVD
events (Fig. 5.1).
High-risk and population-level approaches are not mutually exclusive: in
fact, they are complimentary. Thus, CVD epidemiological efforts are aimed at
reducing population-level burdens of risk factors and at identifying
individuals at higher risk for aggressive primary prevention interventions.
INTRODUCTION TO RISK ESTIMATION
Individual risk factors are probabilistic and typically do not capture
the complete ASCVD risk of an individual. Risk estimator equations have been
developed to help clinicians determine which patients are in the highest risk
groups and, therefore, who would benefit most from primary prevention interventions.
The currently recommended use of 10-year and lifetime risk estimator for the
management of dyslipidemia with statin medications provides an excellent
example of the use of risk scores in clinical practice.
It is important to understand the term risk is a construct and not a
reality. For example, when risk factors of a patient confer a 50% 10-year risk
of developing an ASCVD event, the patient in reality will either have the
clinical event or not have the clinical event. Because of the inherent
limitations in predicting the future, risk estimation is only a starting point
when making decisions about initiation of primary prevention measures.
Clinicians may need to consider other factors, including family history of CVD,
subclinical CVD, and comorbid conditions, among others, to individualize risk
estimation for patients and make decisions
about the initiation of statin medications.
Currently recommended CVD risk assessment tools generate absolute risk
estimates. Absolute risk represents the likelihood of an event occur- ring over
a given unit of time. In contrast, relative risk is a metric of how much higher
or lower the risk of an individual is compared to some referent level of risk
(e.g., relative risk, hazard ratio, and odds ratio).
Absolute risk is chosen over relative risk for clinical risk estimation
for multiple reasons. First, estimates of relative risk are dependent on the
risk in the referent group, which can be inherently misleading. Second, studies
have demonstrated that both patients and clinicians are poor at predicting true
risk. Third, in terms of guideline-based recommendations for statin therapy in
patients with dyslipidemia, the calculation of absolute risk for ASCVD events
allows clinicians to directly compare the average expected benefits of statin
therapy to the average expected harms of statin therapy. Therefore, quantifying
absolute risk allows for a quantitative risk–benefit decision and improves
communication between clinicians and patients.
An alternative approach to a risk-based approach, albeit a cumber-some
and impractical approach, would be for a clinician to gather all of the
clinical trials that examined the effect of statins as primary prevention for
ASCVD and identify which trial identified their specific patient by its
inclusion criteria. However, because of the nature of randomized clinical
trials, which typically have strict inclusion and exclusion criteria, these
studies are unlikely to ever include the complete spectrum of patients
encountered in clinical practice.
To apply risk scores appropriately in clinical practice, it is important
to have an understanding of their development. Typically, risk scores are
derived from cohort study datasets. When developing and using risk estimation
equations, it is important that the individuals in the cohort(s) from which the
score is derived are similar to the patients to whom the risk score is applied.
If the baseline rates of risk factors or disease are substantially higher or
lower than the population in which the risk score is applied, the risk score
may underestimate or overestimate risk in such a population.
The accuracy and clinical usefulness of absolute risk estimates are
sensitive to the time horizon over which they are intended to predict. Most CVD
risk estimation scores use 10-year time horizons, because it is a convenient time interval that is
easy to remember and understand. Furthermore,
it is also easier to develop robust estimates of risk over this time interval. Although 10-year time horizons are
clinically convenient and provide robust estimates of risk, they may lead to an
incomplete understanding of CVD risk burden in some patients. This is best
exemplified in a young woman with multiple risk factors for ASCVD. Although a
risk score predicts a low 10-year risk of ASCVD events, this patient has a high
lifetime risk for CVD. Thus, if only a 10-year risk estimate were communicated
to such a patient, she may overestimate her cardiovascular health and not
institute much needed lifestyle change. However, if she is informed of how her
modifiable risk factors are increasing her risk well beyond the 10-year time
horizon, she may improve adherence to healthy lifestyle modifications. The 2013
AHA risk calculator not only includes 10-year risk estimates of ASCVD events,
but estimates lifetime risk for individuals younger than 64 years old, as well.
It is vital that risk calculators use outcomes that are important to both
patients and clinicians. The ATP-III risk calculator estimated the risk of
nonfatal and fatal myocardial infarction. Unquestionably, these are important
outcomes. However, contemporary CVD epidemiological data have demonstrated that
women, particularly young African Ameri- can women, are at substantial risks
for acute stroke, which was not accounted for by earlier versions of the risk
calculator. In response, the 2013 AHA risk calculator was designed to estimate
the risk for both fatal and nonfatal stroke, as well as myocardial infarction.
When choosing inputs for risk estimation calculators, researchers choose
risk factors that are readily available in clinical practice and that enhance
the ability of the statistical model to accurately predict risk. Because of the
mathematics of statistical models, variables that are highly correlated tend to
be redundant and rarely add substantially to the ability of a statistical model
to accurately predict risk. However, it is critical to understand that a
variable excluded from the risk estimation model may be of significant clinical
importance. For example, obesity is not included in many ASCVD risk calculators
despite its central role in CVD risk. Obesity is a major risk factor for the
development of hypertension, diabetes, and dyslipidemia, and it is strongly
associated with poor diet and a lack of exercise. Because of these
associations, it does not add to risk prediction above and beyond blood
pressure, the presence of diabetes, levels of HDL cholesterol and total
cholesterol. However, when counseling a patient about ways to reduce their CVD
risk, a clinician would be remiss to not discuss the importance of optimal body
weight with a patient.
Discrimination and calibration are used to evaluate risk score
performance. A rudimentary understanding of these concepts allows clinicians to
identify risk scores that are most appropriate for their clinical practice and
to integrate risk scores into the care of individual patients.
Discrimination refers to a risk estimator’s ability to rank order
individuals who will develop disease at a higher level of risk than those who
will not develop disease. It is important to note that discrimination does not
capture the accuracy of the absolute risk estimate. Statistically, the degree
of correct discrimination is described by the c-statistic. The c-statistic is
the area under the receiver-operating characteristic curve, or the plot of the
sensitivity and 1-specificity of the outputs of the statistical model.
Generally, c-statistics of <0.7 are considered poor discrimination, whereas
c-statistics between 0.7 and 0.8 are considered adequate discrimination, and
c-statistics >0.8 are excellent discrimination (Fig.
5.2). To describe discrimination in common language, a c-statistic of 0.8
means that 80% of the time the risk model ranked individuals who were more
likely to have an event as being at higher risk than those who did not have an
event and were therefore at lower risk. This metric partially describes the
performance of the risk estimation equation when used as a population-wide
screening test. It is not sensitive to risks that may be present in certain
niche groups within the population (e.g., individuals with HIV, chronic
inflammatory conditions, or those in the extreme distributions of CVD risk
Risk estimation is dependent on accurate absolute risk estimates, not
just rank ordering. Calibration compares the estimated absolute risk for an
event with the observed event rate in a population sample. When risk estimators are applied to populations
with different event rates, the risk
calculators often require recalibration to avoid overestimating or
underestimating risk. Thus, it is quite possible that a risk estimator may have
a high c-statistic but is poorly calibrated for a given population. In many
instances, risk estimation equations can be recalibrated to different
populations, although this is not routinely done in clinical practice (Fig. 5.3).
When patients have an intermediate estimate of CVD risk, or a 10-year CVD
risk of 4% to 7.5%, clinicians commonly use testing above and beyond what is
included in the risk estimation equation to individualize the risk of CVD in
the patient. For example, high coronary artery calcium scores identify
individuals with high atherosclerotic burden. When high coronary artery calcium
scores are added to an estimated intermediate 10-year risk of CVD, patients can
appropriately be reclassified to a higher level of risk. Therefore, this test
can help clinicians
determine which patients may benefit from statin therapy and those who may not
benefit from statin therapy in the near term.
Although great advances in CVD epidemiology have been achieved over the
last 70 years, significant challenges remain. The burden of CVD and CVD risk
factors remains high in the US population and worldwide. Thus, despite having
identified the determinants of most CVD events, further
improvements need to be made to reduce the prevalence of these risk factors.
This will require advancements in population-level interventions (changes in
food policy, smoking bans, and so on) and continued efforts to understand the
underlying physiology that contributes to the development of incident risk
factors. Collaborative efforts that compile large cohort, genomic, proteomic,
and metabolomic data with the aim of understanding the physiological
underpinnings of CVD are currently underway and will certainly be a continued focus of epidemiological
research in the future. Similarly, the development of digital and remote
sensing technologies and wear-able devices will allow researchers opportunities
for surveillance of physiological characteristics and health-related behaviors
that may contribute to CVD risk.