Framingham vs ASCVD: Which Risk Score Should You Use?

Framingham built the field with its 1998 point-based score, while the ASCVD Pooled Cohort Equations became the US guideline standard in 2013. This comparison covers their formulas, populations, accuracy, and the situations where each model fits.

The Framingham Risk Score and the ASCVD Pooled Cohort Equations are the two most widely used cardiovascular risk models, with different derivation cohorts, endpoints, and risk category thresholds.

Which risk score should be used? The answer depends on the setting. For primary-prevention adults aged 40 to 75, current US guidelines identify the ASCVD Pooled Cohort Equations as the preferred tool. The Framingham Risk Score remains the historical foundation of cardiovascular risk prediction and still appears in many calculators and clinical settings around the world.

The two models look similar on the surface — both combine age, blood pressure, cholesterol, smoking, and diabetes into a 10-year percentage. Their differences matter more: different derivation populations, different endpoints, different category thresholds, and different roles in guidelines. This article compares both models in detail, reviews how accurate each one is, and explains the situations where each fits.

What Is the Framingham Risk Score?

The Framingham Risk Score grew out of the Framingham Heart Study, a long-term research project that began in 1948 in Framingham, Massachusetts. The study enrolled 5,209 men and women aged 30 to 62 who had no history of cardiovascular disease. Researchers examined participants every two years, tracking blood pressure, cholesterol, smoking, diabetes, and cardiovascular events. Over decades, these observations revealed the major risk factors for heart disease.

In 1998, Wilson and colleagues published a landmark paper that converted Framingham data into a practical risk tool. The 1998 model assigned point values to risk factor categories — age, total cholesterol, HDL cholesterol, blood pressure, diabetes, and smoking — separately for men and women.³ Points summed to a 10-year probability of coronary heart disease (CHD), covering angina, myocardial infarction, and coronary death. Risk categories used familiar thresholds: below 10% was low risk, 10% to 20% was intermediate, and above 20% was high risk.

A decade later, D'Agostino and colleagues published a general cardiovascular risk model in 2008 that broadened the Framingham approach. The new model predicted a wider endpoint — coronary heart disease, cerebrovascular events, peripheral artery disease, and heart failure — making it useful for a broader set of clinical questions.² Many Framingham calculators in use today, including the Framingham Risk Score calculator on this site, follow the 2008 general model.

Framingham also shaped treatment guidelines. The National Cholesterol Education Program's Adult Treatment Panel III (ATP III), released in 2001 and updated in 2004, used Framingham 10-year CHD risk to set LDL cholesterol treatment goals. Patients with higher Framingham risk received more intensive cholesterol targets. This guideline role made Framingham the dominant risk tool in American practice for more than a decade.

The score's main limitation stems from its origins. The original cohort was almost entirely White and middle-class, drawn from a single New England town. Risk patterns observed there may not transfer exactly to other populations — a theme that later research would test repeatedly.

What Is the ASCVD Risk Estimator?

The ASCVD Risk Estimator arrived in 2013 as part of the ACC/AHA guideline on the assessment of cardiovascular risk. Goff and colleagues derived the Pooled Cohort Equations (PCE) by combining individual-level data from five major cohort studies: ARIC, CARDIA, the Cardiovascular Health Study (CHS), and the Framingham Original and Offspring cohorts.¹ Pooling gave the model a larger and more diverse derivation sample than any single study could provide.

The PCE predict a specific endpoint: hard ASCVD. Hard ASCVD includes nonfatal myocardial infarction, coronary heart disease death, and fatal or nonfatal stroke. This broader endpoint distinguishes the PCE from the original Framingham CHD score, which did not count stroke. By including stroke, the PCE capture more of the total atherosclerotic burden.

The equations are sex- and race-specific. Separate formulas exist for White men, White women, Black men, and Black women. Each equation combines age, total cholesterol, HDL cholesterol, systolic blood pressure, blood pressure treatment status, diabetes, and smoking into a 10-year risk percentage. The intended population is adults aged 40 to 75 without known cardiovascular disease — the primary-prevention group where treatment decisions are most uncertain.

Risk thresholds differ from Framingham. The 2018 cholesterol guideline defines four ASCVD categories: below 5% is low risk, 5% to 7.4% is borderline, 7.5% to 19.9% is intermediate, and 20% or higher is high risk.⁷ The intermediate band is wide on purpose. For scores in the 7.5% to 19.9% range, the guideline describes coronary artery calcium scoring as an option to refine the estimate when the treatment decision is unclear.

The PCE became the standard risk tool in American guidelines almost immediately. The 2018 cholesterol guideline and the 2019 primary prevention guideline both use PCE categories to structure discussions about statin therapy and other preventive measures.⁷ That guideline endorsement is the main reason the ASCVD risk calculator is now the default risk tool in US practice.

Framingham vs ASCVD: Side-by-Side Comparison

Placed side by side, the two models reveal how risk prediction evolved over 15 years of research. The table below compares their derivation, inputs, endpoints, and guideline roles.

The endpoint difference is the most important distinction. Framingham's 1998 score answered one question: how likely is coronary heart disease in the next 10 years? The PCE answer a broader question: how likely is any major atherosclerotic event, including stroke? A person with high stroke risk but moderate coronary risk could receive a higher ASCVD score than Framingham score.

The threshold difference follows from the endpoints. Because hard ASCVD covers more events, its percentages run lower. A 10% Framingham score and a 10% ASCVD score are not the same quantity — they sit in different categories (intermediate for Framingham, intermediate for ASCVD as well, though near different boundaries) and measure different outcomes. Direct comparison of the raw percentages is misleading.

The population difference affects generalizability. Framingham's single-town cohort limits how well it transports to other groups. The PCE's five pooled cohorts added geographic and racial diversity, though still within the United States. Neither model was derived from Asian, Hispanic, or South Asian populations, and both show calibration issues when applied outside their derivation groups.

Which Is More Accurate?

Accuracy in risk prediction has two parts: discrimination and calibration. Discrimination measures whether the model ranks people correctly — do those who later have events receive higher scores? Calibration measures whether the predicted percentages match observed event rates. A model can discriminate well while being poorly calibrated.

The Framingham CHD score showed strong discrimination in validation. D'Agostino and colleagues validated the Framingham scores in JAMA in 2001, finding that the model separated future cases from non-cases reliably in the derivation data.⁴ Discrimination, usually reported as a C-statistic, typically falls between 0.7 and 0.8 for both Framingham and the PCE. A C-statistic of 0.75 means that for a randomly chosen pair of people — one who had an event and one who did not — the model assigned the higher score to the person with the event 75% of the time.

Calibration proved more troublesome for both models. In 2003, Hense and colleagues tested the Framingham score in the German MONICA Augsburg cohort and the PROCAM study.⁵ Framingham overestimated coronary risk in these European populations, sometimes substantially. The finding illustrated a general principle: risk equations transport poorly across populations with different baseline event rates.

The PCE faced similar criticism. Soon after the 2013 release, researchers reported that the equations overestimated ASCVD risk in several modern cohorts, including MESA and REGARDS.⁷ Part of the gap may reflect genuine declines in cardiovascular event rates over time — models trained on older data expect more events than newer cohorts experience. Part may reflect differences in how risk factors were measured or how participants were selected.

The practical takeaway: neither model is a precise personal forecast. Both rank people reasonably well but can misstate absolute risk in populations unlike their derivation cohorts. Clinicians treat the numbers as starting estimates, adjusted by additional risk factors and clinical judgment. This is why guidelines pair the calculated score with risk-enhancing factors rather than using the number alone.

When to Use Each Model

Guidelines answer the "which to use" question directly. The 2018 ACC/AHA/Multisociety cholesterol guideline recommends the PCE for estimating 10-year ASCVD risk in adults aged 40 to 75 without established cardiovascular disease.⁷ For that population, the PCE is the preferred tool in US practice, and risk discussions about preventive therapy typically begin with the PCE result.

Framingham did not disappear. Many online calculators, research studies, and clinical tools outside the United States still use Framingham-based models. The 2008 D'Agostino general cardiovascular risk model remains a common choice when the question extends beyond ASCVD to include heart failure or peripheral artery disease.² Some settings also keep Framingham because of its long track record and simple point-based format.

For people outside the PCE's intended range — under 40, over 75, or with known cardiovascular disease — neither model is the right tool. Lifetime risk estimates, such as those from the Lloyd-Jones 2006 analysis, provide more useful context for younger adults whose 10-year risk is naturally low.¹¹ Secondary prevention follows different clinical pathways entirely.

Several alternative models exist for specific contexts. The Reynolds Risk Score, published by Ridker and colleagues in 2007, added high-sensitivity C-reactive protein and family history of premature heart disease to traditional risk factors.⁶ PROCAM, developed by Assmann and colleagues in 2002 from a German occupational cohort, offered a European alternative to Framingham.¹⁰ QRISK serves a similar role in the United Kingdom. Each model reflects the population and era of its derivation.

Heart age tools on this site translate the same underlying risk math into an intuitive format, expressing a person's estimated risk as the age of an average person with that risk level. The choice between Framingham and ASCVD is therefore a guideline and context question rather than a contest. US guidelines point to the PCE for primary prevention in midlife adults. Framingham persists where its endpoints, history, or simplicity fit the need better.

The Evolution: Framingham → PCE → PREVENT

The story of cardiovascular risk prediction runs through a single Massachusetts town and then spreads outward.

1948: The Framingham Heart Study enrolls 5,209 residents. The concept of "risk factors" does not yet exist — researchers are still searching for single causes of heart disease.

1961: Framingham investigators publish the insight that coronary risk rises with the number of risk factors present, laying the groundwork for multivariable prediction.

1991: Anderson and colleagues publish Framingham risk profiles that express risk as probability statements, an early form of the risk calculator.⁹

1998: Wilson et al. publish the point-based CHD risk score, giving clinicians a worksheet-style tool with risk factor categories.³

2001: ATP III adopts Framingham 10-year CHD risk for cholesterol treatment decisions, cementing the score in US practice for the next decade.

2007: Ridker et al. publish the Reynolds Risk Score, adding inflammation (hsCRP) and family history to the traditional risk factors.⁶

2008: D'Agostino et al. publish the general cardiovascular risk model, expanding Framingham's endpoint beyond coronary disease to include stroke, peripheral artery disease, and heart failure.²

2013: The ACC/AHA Pooled Cohort Equations replace Framingham as the guideline standard in the United States, bringing pooled derivation data and a stroke-inclusive endpoint.¹

2018: The cholesterol guideline refines PCE use with borderline and intermediate categories, plus coronary artery calcium scoring as a tiebreaker for uncertain cases.⁷

2023: Khan et al. publish the PREVENT equations, adding kidney function (eGFR), urine albumin, HbA1c, and a social deprivation index.⁸ PREVENT estimates 10-year and 30-year risk for both ASCVD and heart failure — the broadest risk tool yet.

Each generation kept the previous one's core inputs — age, blood pressure, cholesterol, smoking, diabetes — and added context the earlier models missed. The trajectory moves toward longer time horizons, broader endpoints, and more diverse derivation populations.

Limitations of Both Models

Both models share structural limits that are worth understanding before interpreting any score.

Population generalizability comes first. Framingham's original cohort was largely White and middle-class. The PCE added Black participants but still derived from five US cohorts, mostly enrolled decades ago. Risk patterns shift over time and differ across ethnic groups, so equations may miscalibrate elsewhere — the European overestimation findings illustrate the problem.⁵

Missing risk factors come second. Neither model includes family history of premature cardiovascular disease, lipoprotein(a), or coronary artery calcium. These factors carry independent risk information. The 2018 guideline lists several "risk-enhancing factors" — including family history, elevated Lp(a), and chronic inflammatory conditions — that clinicians consider alongside the calculated score.⁷ A favorable PCE result does not cancel out a strong family history.

The 10-year window is the third limit. A 35-year-old with multiple risk factors can have a reassuringly low 10-year score simply because events are rare at that age. Lifetime risk tells a different story. The Lloyd-Jones 2006 analysis showed that cardiovascular lifetime risk remains high even for people with optimal midlife risk factor profiles.¹¹ For younger adults, the 10-year number can understate the case for long-term attention to risk factors.

Fourth, the models assume risk factors stay put. Blood pressure, cholesterol, smoking, and diabetes status change over years. A score reflects the values entered at one moment. Guidelines suggest reassessing every 4 to 6 years, and sooner when values change.

Finally, both models estimate group-level probability, not personal destiny. A 10% score means roughly 10 of 100 similar people would have an event in 10 years — the model cannot say which ones. Risk scores open the conversation; they do not close it.