Contributions of the Framingham Heart Study to the Epidemiology of Coronary Heart Disease

Abstract

Since its launch in 1948, the Framingham Heart Study has proved critical to shaping and enhancing our understanding of the history and root causes of coronary heart disease (CHD). A modern prototype for population-based studies, the Framingham Heart Study garnered widespread recognition in its early years for identifying risk factors for CHD and stroke and formulating CHD risk scores. Although the study remains iconic for its robust design and successes in uncovering risk factors for CHD, it has undergone transformations during the past 2 decades. The 21st century ushered in a new era in “molecular epidemiology” centered on cutting-edge genetic and “-omics technologies.” Framingham Heart Study investigators embraced these opportunities by pioneering genome-wide association studies at the population level and examining CHD through the lens of genetic variation, gene expression, and microRNA signatures. The Framingham Heart Study continues to evolve as it seeks to pinpoint new causes of disease with the hope of advancing personalized approaches to the treatment and prevention of CHD.

The Framingham Heart Study (FHS) has been at the forefront of cardiovascular disease (CVD) epidemiology since its inception in 1948.¹ Now a multigenerational study of participants spanning 3 generations within families,^2,3 the FHS is widely recognized for its contributions to identifying and illuminating the roles of multiple key risk factors for coronary heart disease (CHD) and generating risk scores for CHD and other diseases. Much of the FHS’s research efforts before the last decade focused on the identification and elucidation of risk factors for CHD based on extensive and often repeated measures over time of blood pressure, blood glucose levels, lipid levels, cigarette smoking, and left ventricular hypertrophy (LVH). With the advent of high-throughput array-based genotyping in the early 2000s, the FHS expanded its research scope to become one of the first population-based studies to incorporate genome-wide association studies (GWASs). The publication of a series of GWAS articles⁴ and the launch of the FHS SNP (single-nucleotide polymorphism) Health Association Resource (SHARe) project⁵ in 2007 marked a new era of genetics and genomics for the FHS, with an emphasis on studying genetic factors underlying CHD and many other traits and diseases.

To understand the evolution of FHS research activities it is instructive to review major initiatives focused on atherosclerotic CVD and CHD in particular. The progression of approaches to study CHD at the FHS serves as a microcosm of the evolution of the study itself, affirming that while the FHS’s continued success remains rooted in the careful and deep phenotyping and monitoring of its research participants and their commitment, current and future discoveries will be linked to genetic and genomic research as part of a new era of “molecular epidemiology.”

The Beginning: Identifying Risk Factors for CHD

The establishment of the FHS in 1948 by the US Public Health Service and its transfer shortly thereafter to the aegis of the National Heart Institute came at a critical time. The early 20th century witnessed the emergence of CHD and its epidemic growth. Despite the fact that CHD would account for more than 40% of all deaths in the United States by mid century, the etiologic factors of CHD remained poorly understood.^6–8 To this end, early FHS directors Gilcin Meadors, MD, MPH, and Thomas Royal Dawber, MD, MPH, laid the groundwork for demystifying CHD and its risk factors by initiating a prospective population-based study consisting of 5209 participants (Figure 1).⁶ The strength of the study was rooted in the fact that while the initial examination produced data mainly restricted to measurements of a variety of clinical traits, collection of blood samples, and results of an electrocardiogram and chest radiograph, a prospective approach incorporating data from follow-up visits and community surveillance would yield estimates of disease incidence and novel insights into CHD risk factors as more participants developed CHD with the passage of time.

Figure 1. Timeline of Major Framingham Heart Study Accomplishments Related to the Study of Coronary Heart Disease.

See The Beginning: Identifying Risk Factors for CHD section for details.

The first FHS article on new-onset CHD, published in 1957, summarized the findings from early follow-up of the original cohort.⁹ The 4-year follow-up report found CHD risk to be positively associated with hypertension, obesity, and serum cholesterol levels. In that report, Dawber et al⁹ highlighted one of the first pieces of evidence of a strong association between CHD and hypertension. In addition, the article included a rudimentary form of a risk score by analyzing different combinations of blood pressure, relative weight, and total cholesterol level (at low, medium, and high levels) in men aged 45 to 62 years. The study showed the compounding effect of hypertension and hypercholesterolemia on incidence of CHD, as participants with hypertension and high levels of cholesterol were at a higher risk for developing CHD than were participants with only one trait or normal blood pressure and cholesterol levels.

The 4-year⁹ and 6-year¹⁰ follow-up reports by Dawber et al would serve as the cornerstone for one of the most recognized articles in 20th-century CVD epidemiology, “Factors of Risk in the Development of Coronary Heart Disease—Six-Year Follow-up Experience: The Framingham Heart Study” led by William B. Kannel, MD, MPH (Figure 2), in 1961.¹¹ That article (Figure 3) is widely credited for coining the term risk factor. The 1961 report was a significant advancement compared with its predecessors because it assessed quantitative estimates of the magnitude of the increased risk of CHD associated with individual risk factors. In particular, hypertension was associated with a 2.6-fold increase in risk of CHD in men aged 40 to 59 years and a 6-fold increase in women of the same age group. Increased cholesterol level was associated with more than a 3-fold increase in risk of CHD in men and a 1.6-fold increase in women. Although these insights from the article by Kannel et al¹¹ were significant in their own right, more important, they motivated subsequent clinical trials that translated discovery of modifiable risk factors into novel approaches to treatment and prevention of CHD. Clinical trials of reducing blood pressure and low-density lipoprotein cholesterol levels would soon lead to new strategies for reducing risk of CHD.^12,13 The lessons learned from clinical trials of interventions for these risk factors would in turn inform clinical guidelines for the treatment of hypertension¹⁴ and dyslipidemia¹⁵ that have transformed CVD into a largely preventable condition.

Figure 2. William Kannel, MD, MPH.

Framingham Heart Study director from 1966 to 1979.

Figure 3. The 1961 Article on Risk Factors in Coronary Heart Disease.

One of the most influential articles in epidemiologic research, “Factors of Risk in the Development of Coronary Heart Disease—Six-Year Follow-up Experience: The Framingham Heart Study,” by Kannel et al,¹¹ was the first to assess quantitative estimates of the increased risk for coronary heart disease associated with individual risk factors. Adapted with permission from The Annals of Internal Medicine.

Application of Clinical Data: Developing and Fine-tuning a CHD Risk Score

The FHS developed an algorithm in 1976 to predict risk of CHD, constructing a multivariable risk function or score that included age, sex, smoking status, systolic blood pressure, total serum cholesterol level, electrocardiographic evidence of LVH, and glucose intolerance.¹⁶ At the time of its creation, the FHS risk score was particularly useful for identifying the group at high risk for CHD, with individuals in the top 10% of risk accounting for 22% of events during 8 years of follow-up.

As the FHS entered the later decades of the 20th century, its researchers and statisticians continued to fine-tune the CHD risk score. In 1971, the recruitment of 5124 new participants into the offspring cohort² would allow for an improved risk score with an expanded age range of 30 to 74 years.¹⁷ The updated risk equations included the ratio of total cholesterol to high-density lipoprotein cholesterol levels as a predictor in place of total cholesterol levels alone; they also differentiated systolic blood pressure from diastolic blood pressure as a predictor of risk of CHD. In 1998, Wilson et al¹⁸ simplified the CHD prediction algorithm further by omitting LVH as a variable because of its association with hypertension and its diminishing prevalence in an era of improved treatment and control of hypertension. To simplify risk assessment, the investigators also provided a point-scoring strategy using categories of blood pressure, total cholesterol, and high-density lipoprotein cholesterol levels based on blood pressure levels recommended by the Joint National Committee¹⁹ and cholesterol levels recommended by the National Cholesterol Education Program.²⁰

Simplification of the FHS risk score in this manner, however, did not always equate to improvement in predicting risk of CHD. Efforts by the FHS in the early 2000s centered on making nuanced adjustments to enhance the accuracy of predicting risk of CHD for certain subgroups of the general population. D’Agostino et al²¹ demonstrated that the risk score could be recalibrated and applied to other populations in which risk of CHD was poorly estimated by FHS equations. The need for more robust risk assessment tools featuring race- and sex-specific models led the FHS to pool its data as part of the 2014 Risk Assessment Work Group, which introduced the pooled cohort equations that provided more accurate prediction of the risk of CVD for diverse populations.²² Another major limitation of the FHS CHD risk score is that it was designed to estimate 10-year risk and thus was not effective at estimating lifetime risk for CHD, particularly in younger men in whom low short-term risk may differ from high long-term risk.²³ To better capture the effect of risk factor levels on long-term risk of CVD, the FHS developed the first 30-year CVD risk calculator that improved assessment of long-term risk.²⁴

The FHS risk scores relied predominantly on baseline values of risk factors rather than repeated measures, with a few important exceptions. For a few CVD risk factors, Cupples et al²⁵ compared parameter estimates derived from a pooling of repeated observations vs those derived from analysis of only baseline measures. Applying the pooling of repeated observations method to its studies, in 1987, the FHS reported a 30-year follow-up of CVD incidence and death,²⁶ and later published risk models of CHD based on repeated observations.²⁷

The Role of Medical Technology: Electrocardiography and Echocardiography

Medical technology has played an underappreciated but crucial role in the scientific contributions of the FHS. Electrocardiography (ECG), for example, enabled the inclusion of LVH among the original CHD risk factors highlighted in the 1961 study by Kannel et al¹¹ on risk factors. The utility of ECG-LVH as a prognostic tool for CHD morbidity and mortality had certain limitations early on in the FHS. The initial ECG-LVH classification system took a crude form—categories of definite (meeting both voltage and repolarization criteria), possible (meeting voltage criterion only), and no ECG-LVH classified each FHS participant without taking age and sex into consideration.^28,29 Yet, Kannel and other FHS investigators were also quick to realize that ECG-LVH alone still revealed important insights into CHD risk. For instance, definite ECG-LVH was associated with a 3-fold increase in CHD risk even after accounting for blood pressure levels, and definite and possible ECG-LVH had preceded 44% of total CVD deaths.^28,29

As ECG-LVH became increasingly recognized as a risk factor for CVD, FHS investigators began to explore the intricacies of the ECG features of LVH in a process that resembled the method by which they had developed and fine-tuned the risk score. In 1983, Kannel³⁰ showed that different ECG-LVH criteria were accompanied by different risks; for example, ECG-LVH with voltage and repolarization abnormalities showed a more robust association with risk of CVD than did ECG-LVH that met only the voltage criteria. The FHS continued to study ECG-LVH in depth in the following decades, and a reciprocal relationship of sorts formed between the FHS and ECG by the 1990s. Electrocardiography had aided the FHS in identifying LVH as one of the earliest risk factors for CHD and other CVD traits during the 1960s, and the FHS a few decades later would publish studies that in turn contributed to the field by improving ECG diagnostic strategies. One such study published by Levy et al³¹ in 1990 tested the sensitivity and specificity of ECG-LVH in a population-based setting, concluding that the sensitivity of ECG could be enhanced by taking into account noncardiac factors such as age, sex, smoking status, and obesity, while maintaining a high level of specificity.

Amid ongoing studies of ECG-LVH, other medical technologies were introduced into the FHS by the 1980s. Echocardiography rapidly drew interest from clinicians in the late 1970s owing to the key advantage it offered vs ECG by providing more detailed anatomical information on cardiac structure and function, including assessment of valve disease.³² The FHS investigators likewise were quick to incorporate echocardiography measurements and diagnoses into their CVD studies. By 1987, the FHS had established echocardiographic criteria for LVH (based on the calculation of left ventricular mass) using original and offspring cohort data.³³ Levy et al^34,35 then conducted a 4-year prospective study of 3220 FHS participants older than 40 years to analyze associations of echocardiographically determined left ventricular mass with CVD incidence, CVD mortality, and all-cause mortality. The findings provided evidence for the prognostic value of echocardiography; left ventricular mass was significantly associated with all 3 outcome events even after adjusting for age, systolic blood pressure, pulse pressure, use of antihypertensive medication, smoking status, diabetes, obesity, ratio of levels of total cholesterol to high-density lipoprotein cholesterol, and ECG-LVH.³⁵

Start of a New Era: Examining the Genetic Factors Underlying CHD

Four decades after its inception, the FHS had made invaluable contributions to the epidemiology of CHD, which included identifying CHD risk factors, developing a CHD risk score, and incorporating cardiovascular imaging technology. As large as these contributions were, however, they failed to explain several questions, including the molecular underpinnings of CHD and interindividual variation in CHD risk factors and susceptibility. Genetics, FHS investigators quickly realized, would be key to answering these questions. By the late 1980s, the FHS began collecting DNA from study participants in preparation for its pursuit of genetic factors contributing to risk of CHD. With this step, the FHS transitioned from elucidating CHD risk factors to exploring genetic signatures of disease; doing so marked a turning point in the history of the FHS.

The initial forays into genetics conducted by FHS researchers centered on studying familial patterns associated with CHD, largely motivated by previous observations that risk factors for CHD such as blood pressure, cholesterol level, and smoking status clustered in families.^36–38 Schildkraut et al³⁹ reported in 1989 that parental history of CHD death was a significant predictor of CHD occurrence in offspring. Individuals with a parental history of CHD death had a 30% increased risk for developing CHD, with the risk being higher for early-onset CHD (adjusted relative risk of 1.5) than for late-onset CHD (adjusted relative risk of 1.2). Furthermore, the effect of a positive parental CHD history on CHD risk was not mediated by other CHD risk factors. Taken together, these results pointed to a separate genetic contribution to CHD.

Although a series of FHS articles examining familial CHD patterns provided findings that suggested a prominent genetic component, these studies could not shed light on what this genetic component exactly entailed. Efforts to understand genetic factors associated with CHD and other CVD-related phenotypes seemed to meet a roadblock. The missing critical piece was the technology required to harness the information contained in the participants’ DNA to study genetic influences on a large scale.

Within this context, the emergence of high-throughput array-based genotyping in the early 2000s proved to be timely for the FHS. Marked advances in genotyping technology enabled FHS researchers to distill the relatively vague concept of genetic influences on CHD and its risk factors by using GWASs to identify genetic loci associated with risk of CHD. In 2007, the FHS published its first GWAS results in a series of articles (referred to as the FHS 100K GWAS articles) based on genotyping 100 000 common SNPs in 1345 FHS participants from the original and offspring cohorts. The FHS researchers examined associations of SNPs with interindividual variation in 987 clinical phenotypes collected during 56 years of follow-up. At the time of their publication, the FHS 100K articles was the largest series of GWAS publications ever conducted in terms of the number of phenotypes analyzed, covering CVD risk factors and biomarkers, clinical and subclinical CVD, longevity and aging traits, and cancer.⁴ In addition, the FHS created a web-based, open-access database of all GWAS results that served as a valuable resource for investigators seeking to perform replication studies. Among the 17 FHS 100K articles published in the series, 1 focused specifically on genome-wide associations for 4 major CVD outcomes that included major atherosclerotic CVD, CHD, atrial fibrillation, and heart failure.⁴⁰ The findings from this early foray into GWASs served as an important building block for future GWASs of CVD phenotypes.

In 2007, the FHS also launched the SHARe project.⁵ Funded by the National Heart, Lung, and Blood Institute, SHARe aimed to analyze associations between common genetic polymorphisms and clinical traits, including CVD risk factors, after genotyping 550 000 SNPs (using the Affymetrix 500K mapping array and Affymetrix 50K supplementary array [Thermo Fisher Scientific Inc]) in more than 9300 individuals from the original, offspring, and newly recruited third-generation FHS cohorts. One of the key lessons that FHS researchers took away from early GWAS projects was the need for larger sample sizes across multiple cohorts to provide stronger evidence for genetic associations with disease phenotypes. With the goal of creating more opportunities for GWAS meta-analyses, the FHS and 4 other population-based studies founded the Cohorts for Heart and Aging Research in Genetic Epidemiology Consortium in 2008.⁴¹ This consortium, which quickly grew to 16 cohort studies, has produced more than 400 publications to date.

In recent years, the FHS has also participated in a key fashion in GWAS consortia meta-analyses of CVD risk factors involving even larger sample sizes. For instance, in 2011, the FHS contributed to a GWAS meta-analysis comprising 200 000 individuals of European descent, which created a genetic risk score associated with risk of CHD and other CVD traits based on 29 genome-wide allelic variants associated with blood pressure.⁴² The FHS was also a member of the Coronary Artery Disease Genome-Wide Replication and Meta-Analysis Consortium study that identified 13 novel loci associated with CHD at a genome-wide significant level by meta-analyzing GWAS results from 14 cohort studies totaling 22 233 patients with CHD and 64 672 controls.⁴³ This study also validated 10 previously reported CHD loci including the 9p21 and 1p13 loci. Although the translational potential of GWASs and other “-omics technologies” works remains largely unrealized, the progression of studies of the 1p13 locus—from the initial genomic discovery to the later functional studies that elucidated a novel regulatory pathway involving low-density lipoprotein cholesterol regulation via expression of the SORT1 gene (GenBank NC_000001)⁴⁴—exemplifies how GWAS findings may hold important therapeutic implications.

Understanding the Molecular Underpinnings of CHD

More recent FHS projects have gone beyond identifying genetic variants associated with CHD and have pursued the molecular mechanisms underlying CHD via analysis of messenger RNA (mRNA) expression, microRNA (miRNA) expression, and proteomic profiles in a systems biology framework. For example, in 2013, FHS researchers used a CHD case-control study consisting of 188 patients with CHD and 188 age- and sex-matched controls to examine the genomic architecture of CHD.⁴⁵ Based on whole blood gene expression data, the study identified 24 differential CHD coexpression modules enriched for genes implicated in B-cell activation, immune response, and ion transport. The differential modules were integrated with SNPs previously identified in GWASs of risk of CHD to identify putatively causal gene modules. Gene ontology and cluster analysis revealed that CHD was associated with perturbed innate immune activity pathways.⁴⁶

The FHS also characterized a miRNA signature of CHD, finding 15 miRNAs differentially expressed in patients with CHD vs controls. Integrative analysis found a substantial difference in coexpressed miRNA-mRNA pairs between patients with CHD vs controls.⁴⁷ Some of the coexpressed miRNA-mRNA pairs were enriched for genetic variants affecting miRNA and mRNA levels associated with CHD, suggesting a causal link between these coexpression modules and CHD.

Several recent studies published by the FHS used a network-driven approach coupled with bioinformatics analysis. One such investigation constructed a CVD network linking 1512 SNPs associated with 21 CVD traits in previous GWASs.⁴⁸ By integrating these SNPs with whole blood gene expression in 5257 FHS participants, the investigators identified 370 cis-expression quantitative trait loci (cis-eQTLs [expression quantitative trait loci], or SNPs affecting expression of nearby genes) and 44 trans-eQTLs (SNPs affecting expression of remote genes) forming an eQTL network with 13 CVD-related traits. The expression of a subset of eQTL genes was significantly associated with CVD-related phenotypes or risk factors, and mediation analysis showed that some SNPs influenced these CVD-related phenotypes by altering levels of eQTL gene expression. The FHS investigators speculated that this integrative genomic approach could aid in identifying novel therapeutic targets for the treatment and prevention of CVD and its risk factors.

Conclusions

With their sights set on the future, FHS investigators are currently coleading several working groups within the National Heart, Lung, and Blood Institute’s Trans-Omics for Precision Medicine program that is conducting whole-genome sequencing of approximately 60 000 individuals, including 4200 FHS participants. This new initiative will allow FHS investigators to integrate whole-genome sequencing data with multipleomic components, encompassing gene expression, DNA methylation, and metabolomics, to uncover novel biological mechanisms underlying risk of disease and develop more personalized approaches for the treatment and prevention of heart, lung, and blood disorders. To make the vast genetic, genomic, and phenotype data resources of the FHS accessible to the broader scientific community at no cost, these databases have been deposited in the Databases of Genotypes and Phenotypes (http://www.ncbi.nlm.nih.gov/gap) and the Biologic Specimen and Data Repository Information Coordinating Center (https://biolincc.nhlbi.nih.gov/search/). Investigators are able to access these resources to conduct research either in collaboration with FHS scientists or working independently. The FHS is also expanding its research scope into electronic device capture of health-related measures. In 2015, the FHS conducted a pilot study as part of the larger Health eHeart Study⁴⁹ that uses smartphones, wearable digital technology, and online questionnaires for data collection integral to investigating CVD and its risk factors. Moving forward, an electronic study component will be included in future participant examinations.

The FHS’s journey to uncover risk factors for CHD has been long, both in terms of its duration and breadth. To an outsider, the evolving scientific pursuits of the FHS may seem spontaneous and sudden, but these changes did not occur without considerable challenges and pauses. Nor has the FHS been alone in shedding light on the etiologic factors and treatment of CHD. The transformation of the FHS was a consequence of extensive planning by FHS investigators to take their research in new directions determined by the most pressing scientific questions. To adapt to a changing world of scientific opportunities, the FHS introduced cutting-edge technologies and applied them to the field of population science. This was the approach used when the FHS introduced new cardiovascular imaging tests, such as echocardiography, magnetic resonance imaging, and computed tomographic scanning, to answer important questions about subclinical structural and functional changes to the heart in response to elevated blood pressure levels. It was also the approach used when the FHS helped pioneer the introduction of genome-wide association analysis at the population level, and more recently, when the FHS positioned itself at the cutting edge of whole-genome sequencing. It should come as no surprise that the FHS is preparing for the next chapter of its journey to pinpoint the molecular basis of CVD and to aid in developing personalized approaches to its treatment and prevention.