Using the Framingham model to predict heart disease in the United Kingdom: retrospective studyBMJ 2000; 320 doi: https://doi.org/10.1136/bmj.320.7236.676 (Published 11 March 2000) Cite this as: BMJ 2000;320:676
- S Ramachandran, senior registrar, chemical pathologya,
- J M French, research associatec,
- M P J Vanderpump, consultant endocrinologistd,
- P Croft, professorb,
- R H Neary, consultant, chemical pathology ()a
- a Department of Clinical Biochemistry, North Staffordshire Hospital, Stoke on Trent ST4 7PA
- b Department of Epidemiology, North Staffordshire Hospital
- c Department of Statistics, University of Newcastle, Newcastle upon Tyne NE1 7RU
- d Department of Endocrinology, Royal Free Hospital, London NW3 2QG
- Correspondence to: R H Neary
- Accepted 19 July 1999
Editorial by Jackson
Guidelines on the use of drugs to lower serum concentrations of lipids to prevent coronary heart disease target treatment to patients who have a high absolute risk of the disease. Although a patient's absolute risk of heart disease can be derived using risk tables1 —for example, the Sheffield table—these are based on the Framingham model which may not be applicable to the population in the United Kingdom.2 We aimed to determine whether the Framingham model accurately predicts the risk of coronary heart disease among white men and women in the United Kingdom.
Participants, methods, and results
A cross section of the population of Whickham, north east England, was enrolled in a study of ischaemic heart disease between 1972 and 1974 and followed up 20 years later.3 At baseline, data was collected on body mass index, family history of coronary heart disease, fasting glucose concentrations, and triglyceride concentrations. Standardised WHO questionnaires on chest pain were administered, and the information necessary to complete the Framingham model (age, sex, systolic blood pressure, ratio of total cholesterol to high density lipoprotein cholesterol, presence of left ventricular hypertrophy, presence of diabetes, and smoking habits4 was also collected, with the exception of concentrations of high density lipoprotein cholesterol for which values of 1.15 mmol/l were used for men and 1.4 mmol/l for women.1
Altogether, 77 (2.8%) of the 2779 adults initially enrolled were lost to follow up. Of the remaining 2702, a total of 1877 were still alive at follow up, of whom 1802 (96%)participated. A total of 927 participants were excluded from the analysis for one or more of the following reasons: if they had had heart disease at baseline (172), were aged younger than 30 or older than 75(702) years, or if they had previously been smokers (371); those who had previously been smokers were excluded because the length of time since quitting was unknown.
Evidence of heart disease occurring in those who had died was identified using death certificates, records from postmortem examinations, hospital notes, or the general practitioner's notes. Coronary morbidity was determined in participants by identifying a history of myocardial infarction or angina, evaluating answers to the WHO questionnaire, and by examining the results of repeat electrocardiography which were classed according to the Minnesota Code. The predicted 20 year risk of heart disease was calculated for each participant using baseline measurements and the Framingham model. Participants were ranked in groups according to predicted risk (for example, 0-4.99%, 5-9.99%, etc), and the percentage of participants in each group who actually had had an event during follow up was determined. Differences between patients with and without heart disease and the goodness of fit between actual and predicted coronary events were tested using the Student's t test and χ2 analysis.
Of the 1700 participants remaining, 529 (31.1%) had developed heart disease. A higher proportion of men than women had developed heart disease (257/751 (34.3%) men v 272/949 (28.7%) women; P=0.015), as had a higher proportion of smokers than non-smokers (344/1017 (33.8%) v 185/683 (27.1%); P=0.003); and 8 (57%) of 14 participants with diabetes had developed heart disease. Those participants who had developed heart disease were older (mean age 54.7 years v 48.1 years, P<.0001), had higher serum cholesterol concentrations (6.32 mmol/l v 6.05 mmol/l, P<.0001), and higher systolic blood pressure (151.2 mm Hg v 138.9 mm Hg, P<0.0001). In terms of the Framingham risk score, those who had developed heart disease had a mean 20 year risk of 30.5% (95% confidence interval 29.2% to 31.8%) compared with those who did not (20 year risk 20.5%, 19.7% to 21.4%; P<0.0001). When individual variables were subjected to logistic regression, male sex, age, blood pressure, smoking status, and cholesterol concentrations were all significant predictors of heart disease but when corrected for the Framingham risk score no single factor remained predictive on its own.
The figure shows the number of coronary events predicted by the Framingham model and the number observed during follow up. The agreement is good at a predicted event rate above 30% (1.5% per year), with no significant difference between the observed and expected event rates (P=0.85). However, at lower event rates the predictive model significantly underestimates the number of observed events (P<0.01). The wide confidence intervals indicate that there is significant overlap between risk scores in those participants who developed heart disease and those who did not.
These results confirm that the Framingham model reliably predicts the absolute risk of heart disease in white men and women in the United Kingdom when the annual risk is above 1.5%, but the model underestimates the risk when the absolute risk is lower. This is consistent with studies that have shown that the model is inaccurate when applied to low risk populations.5 We might have achieved a closer fit with the model by measuring concentrations of high density lipoprotein cholesterol and using a 4 to 12 year follow up period similar to that from which the model was derived. Nevertheless, the recommended threshold for treatment with lipid lowering drugs is based on an annual risk of 3% per year, 1 so the Framingham model can be used in clinical practice in the UK population.
We would like to thank Drs M Tunbridge and D Appleton for providing access to the data from the Whickham study and Professor Gilbert MacKenzie for helpful comments on the manuscript.
Contributors: SR and RHN developed the computer database and used it to analyse the data, as suggested by MPJV. MPJV and JMF provided the data. PC provided advice on data analysis and presentation. The paper was written by all of the authors. RHN is guarantor for the study.
Funding The Department of Health and Newcastle District Research Committee provided financial support for this study.
Competing interests None declared