Statistics from Altmetric.com
Editor,—Predicting the risk of coronary heart disease will always be prone to error. Haq et al compared four different risk functions1: the Framingham (USA),2 PROCAM (prospective cardiovascular Münster, German),3 Dundee (UK),4 and British regional heart study (UK) risk5 functions. These functions were applied to 206 male patients attending the Sheffield hypertension clinic. Haq et al used Bland–Altman difference plots to compare methods. Although they claim good agreement among the Framingham, PROCAM, and Dundee functions, close inspection shows that the difference in risk in the Framingham—PROCAM plot greatly increases above a mean coronary heart disease risk of 4% (fig 1B), and points in the Framingham—Dundee plot diverge above 0% mean coronary heart disease risk (fig 2B)—that is, there is poor agreement among the various methods. What is more, Haq et alseem to dismiss the British regional heart study function because its estimate of risk was fourfold lower than for the Framingham function yet the British study function was able to predict 59% of major ischaemic heart disease in subjects over the ensuing five years.5
Surely it would have been more informative to have applied each of the risk functions to subjects who attended the Sheffield hypertension clinic and who were followed up over five years and to see whether the predictions of risk were accurate. Risk analysis is a tricky business. We should use these functions and tables only if we are aware of their limitations.
This letter was shown to the authors, who reply as follows:
We agree that risk prediction is concerned with the probability of a future event and is not an exact science. We have shown reasonable—but by no means perfect—agreement between predictions by the Framingham, Dundee, and PROCAM equations. The important question is whether the agreement is close enough for clinical practice. The analysis shown in fig 5 addressed this point and showed that the Framingham risk function separates clearly groups at high and low risk as determined by the two other risk functions. The accuracy of targeting was acceptable and this supports the use of methods based on the Framingham equation in national and international guidelines.
The British regional heart study function predictedrelative risk well but seriously underestimated absolute risk compared with the other three risk functions. Possible reasons for this were discussed—for example, inclusion of people with established coronary heart disease, different definitions of risk variables, exclusion of HDL cholesterol, and the lower average risk of the population studied. The predictive value of the British regional heart study risk function that Johnston cites1-1 is for an internalvalidation, meaning that the risk function was tested in the population from which it was derived. Any systematic error would be common to the derivation and the testing of the risk function and would not therefore be detected. The British regional heart study risk function appears to have important inaccuracy for absolute risk in two external validations.1-2
It would of course be ideal to carry out a prospective cohort study, but the simpler analysis presented reassures us that use of the Framingham function is reasonable, at least in men. We agree that one must be aware of the limitations of risk functions. Coronary heart disease risk assessment methods based on Framingham are much more accurate than use of cholesterol or lipid thresholds, intuitive estimation of risk, or simple counting of risk factors.