Framingham Risk Score for hard coronary heart disease

Clinical background

The Framingham Risk Score for hard coronary heart disease was developed to combine established risk factors into a single sex-specific scoring system and thereby to estimate the absolute 10-year risk of myocardial infarction or coronary death. The tool meets a need in primary prevention: the decision to start lipid-lowering or blood pressure-lowering treatment in a patient without known cardiovascular disease rests on a trade-off between absolute risk and treatment benefit, and that trade-off is hard to make intuitively when several risk factors interact.

The scoring system was the historical basis for the NCEP ATP III risk categories and came to dominate primary preventive practice in North America during the 2000s. In Europe it has largely been replaced by SCORE and, in North American guidelines, by the Pooled Cohort Equations, but it remains a reference instrument that many clinicians recognise and that features in several older but still-used guidelines.

Calculating the Framingham Risk Score

The score rests on sex-specific point tables in which each risk factor contributes a number of points depending on sex and, for total cholesterol and smoking, on age group as well. The variables included are:

• Sex: male or female (separate point tables)
• Age: 20 to 79 years
• Total cholesterol: age-stratified points
• HDL cholesterol: points by sex
• Systolic blood pressure: separate points for untreated and treated blood pressure
• Current antihypertensive treatment: yes or no
• Current smoker: yes or no

The total score is obtained as:

$$\text{Total score} = P_{\text{age}} + P_{\text{total cholesterol}} + P_{\text{HDL}} + P_{\text{systolic blood pressure}} + P_{\text{smoking}}$$

where $P_{\text{systolic blood pressure}}$ is taken from the treated or the untreated column depending on whether the patient is on antihypertensive treatment. The total score is then converted into an estimated 10-year risk of hard coronary heart disease via a sex-specific table.

The derivation cohort consisted of 2,489 men and 2,856 women aged 30 to 74 years, recruited from the Framingham Heart Study [1]. Follow-up was 12 years, during which 383 men and 227 women developed coronary heart disease, defined as myocardial infarction, coronary death or unstable angina. The cohort was predominantly White and middle-aged. The authors found that the categorical scoring method (using the JNC-V blood pressure categories and the NCEP cholesterol categories) performed comparably to a model based on continuous variables, which justified the table-based format [1].

Interpretation in practice

The predicted 10-year risk translates into clinical action according to the applicable guidelines, not according to a fixed rule within the instrument itself. The thresholds that have historically been used, and on which NCEP ATP III was built, can be summarised as follows:

It is important to note that these bands come from NCEP ATP III and are not an intrinsic property of the Framingham score. Modern guidelines, particularly the North American guidelines from 2018, have instead moved to the Pooled Cohort Equations and a statin treatment threshold of a 7.5% 10-year risk of atherosclerotic cardiovascular disease, which is not directly applicable to the Framingham score for hard coronary heart disease.

Validation and performance

The Framingham Risk Score is one of the most externally validated risk instruments in clinical medicine. A systematic review from 2024 identified 98 studies that had evaluated the Wilson rule, of which 40 were external validation studies [2]. Of these, 67.5% judged that the instrument performed inadequately in the population studied, but only four studies attempted to update the model by recalibration or re-estimation of parameters [2].

An earlier systematic review from 2007, based on 25 validation cohorts with a total of approximately 128,000 participants, found that calibration varied widely between populations [3]. For cohorts from the USA, Australia and New Zealand, calibration was good, with predicted/observed ratios between 0.87 and 1.08 for the five largest cohorts. For 18 European cohorts, by contrast, there was a systematic overestimation of absolute risk, with a regression coefficient of 0.58 (95% CI 0.39 to 0.77), meaning that the score predicted approximately twice as many events as were observed [3].

A large Canadian study from 2020, comprising 84,617 individuals in Ontario without known cardiovascular disease, confirmed the pattern [4]. At 5 years of follow-up the observed event rate was 2.6%, while the Framingham score predicted 5.78%, an overestimation of approximately 101%. The c-statistic was 0.74, comparable to the Pooled Cohort Equations (0.73). The overestimation varied with age and ethnicity [4].

An individual participant data meta-analysis from the Emerging Risk Factors Collaboration, with data on 360,737 participants in 86 prospective studies from 22 countries, compared the Framingham score with SCORE, the Pooled Cohort Equations and the Reynolds Risk Score [5]. Before recalibration, the Framingham score overestimated risk by an average of 10%, less than SCORE (52%) and the PCE (41%). Discrimination was similar for all four algorithms. After simple recalibration to the target population's risk factor profile and event incidence, the differences in performance were almost eliminated, and the proportion of individuals classified as high risk fell from 29 to 39% to 22 to 24% for all the algorithms [5].

In summary: discrimination is moderate and stable across populations, with a c-statistic in the range 0.69 to 0.74 in the larger external validations [2, 4, 5]. Calibration, by contrast, is population-dependent and requires recalibration to give correct absolute risk estimates outside the original Framingham cohort.

Limitations

The instrument predicts hard coronary heart disease only, defined as myocardial infarction and coronary death. It does not cover stroke, heart failure or peripheral arterial disease. The later Framingham score for general cardiovascular risk (D'Agostino et al. 2008) has a broader outcome but is not the one this calculator computes.

The derivation cohort was predominantly White and middle-aged, recruited in an American suburban community during a period when the incidence of coronary heart disease was higher than it is today. This explains part of the systematic overestimation seen in European and modern North American populations [3, 4, 5]. Diabetes was included in the original Wilson model but is not a field in this calculator, which can lead to underestimation of risk in patients with diabetes.

The commonest misuses are: applying the score to patients with already known cardiovascular disease, for whom it is not intended; interpreting the absolute risk percentage as population-independent without recalibration; and using it to make decisions on stroke prevention, for which it has no predictive coverage.

References

1. Wilson PWF, D'Agostino RB, Levy D, et al. Prediction of coronary heart disease using risk factor categories. Circulation. 1998;97(18):1837-47. PMID: 9603539
2. Ban JW, Abel L, Stevens R, et al. Research inefficiencies in external validation studies of the Framingham Wilson coronary heart disease risk rule: A systematic review. PLoS One. 2024;19(9):e0310321. PMID: 39269949
3. Eichler K, Puhan MA, Steurer J, et al. Prediction of first coronary events with the Framingham score: a systematic review. Am Heart J. 2007;153(5):722-31. PMID: 17452145
4. Ko DT, Sivaswamy A, Sud M, et al. Calibration and discrimination of the Framingham Risk Score and the Pooled Cohort Equations. CMAJ. 2020;192(17):E442-E449. PMID: 32392491
5. Pennells L, Kaptoge S, Wood A, et al. Equalization of four cardiovascular risk algorithms after systematic recalibration: individual-participant meta-analysis of 86 prospective studies. Eur Heart J. 2019;40(7):621-631. PMID: 30476079