Which measurements are the most appropriate?

EDITOR—I would like to predict the risk score for all patients in my practice aged 35 years or older as so clearly described in the article by Pocock et al.1 My enthusiasm invariably turns to disappointment when using risk scores because of the following six factors.

Age—Most of these studies stop at the age of 74. As a significant proportion of my patients are aged 75 and over, I feel cheated. Is there any possibility of extrapolating the risk scores to an older age?

Smoking—Should I include former cigarette smokers who have only recently stopped smoking; recent being, for example, during the past year?

Blood pressure—Which measurement do I use? The most recent measurement, the average of the last three measurements, the highest recorded measurement, the average of the three highest recorded measurements, or a measurement with the patient seated, standing, or ambulatory, or the patient's own measurement?

Cholesterol concentration—Which measurement is the most appropriate? The most recent measurement, the average of the last three measurements, the highest recorded measurement, or the average of the three highest recorded measurements?

Creatinine concentration—Which measurement is the most appropriate? The most recent measurement, the average of the last three measurements, the highest recorded measurement, or the average of the three highest recorded measurements?

Left ventricular hypertrophy—How should it be defined? By evidence from echocardiography only, or by evidence from electrocardiography?

Ideally I would like to be able to discuss with a patient their current risk score and be able to compare with their previous risk scores, since when they have stopped smoking, blood pressure has been reduced, and their hypercholesterolaemia has been treated.

Likewise on a practice basis, I would like to be able to produce a “before and after” risk score to assess the effectiveness of the various treatments, which my practice has implemented often at great expense, time, and energy.

I am disappointed that the above points are not specifically addressed in this and other similar risk score assessments as to which are the most appropriate measurements of blood pressure, cholesterol, and creatinine concentrations to use to assess risk score.

Outcome prediction is hampered by methodological problems

EDITOR—The paper of Pocock et al reports the detection of several risk factors of death in adults with raised blood pressure by the Cox proportional hazards model.1 The relation of the created new risk score to the probability of dying was clearly shown in the development set.

The paper is, however, flawed by the lack of validation in an independent sample of patients—a basic methodological principle of any predictive model.2 It would therefore be of utmost importance for Pocock et al to test the discrimination by receiver operator characteristic analysis and calibration of this suggested new risk model, especially when their calculations are based on older study populations, as pointed out by Vesti-Nielsen (electronic response published as letter below).3 An assessment of the usefulness of the model of Pocock et al will be possible only after this mandatory validation process.

Individual cardiovascular risk cannot be predicted

EDITOR—The article by Pocock et al is an interesting experiment in mathematics, but the authors cannot be serious when they suggest using this experiment for giving advice in the real world.1 If the rates for heart disease had been constant it would be useful in the real world instead of in the closed system of the article only, but they are not. Pocock et al mention the differing rates in different countries and admit that it would be difficult to predict a risk in, for example, France, on the basis of—as the calculations are—mainly old American and English studies.

But Pocock et al forget the even larger variations over time in many parts of the world. Cardiovascular mortality has been falling constantly at least since 1980, in most countries (except in the former Eastern bloc), by 2-7% yearly (for example, in the United Kingdom: Glasgow 2.6%, Belfast 6%; and in Sweden 3.8-8% in different centres per year).2

Two thirds of this reduction results from decreasing incidences of coronary heart disease.2 People are getting healthier. The calculations in the article have a built in blemish in that they are based on older studies in populations with substantially higher basic rates for cardiovascular disease. The logic in itself is flawless, but the arguments do not hold. It is impossible, on the basis of old studies, to predict in figures the absolute risk of cardiovascular death for a person living today, since the rates in the future are unknown, and whatever drives the changing rates is unknown as well. Perhaps the relative risks are the same, which I seriously doubt, but the absolute risks, necessary to advise a patient, have changed substantially.

Authors' reply

EDITOR—In reply to White, extrapolating risk scores to ages over 75 is desirable, but our data (like those of others) include few such elderly subjects. The benefits of stopping smoking are not immediate so recent former smokers may be counted as current smokers. Showing the reduction in risk, however, if counted as a non-smoker may help motivation.

For blood pressure, cholesterol and creatinine concentrations we would encourage use of averages of repeat recent measurements if available since this reduces within-subject variation. Our study used at least three seated blood pressures. Evidence of left ventricular hypertrophy was provided by electrocardiogram..

Patients' changes in risk score are valuable, especially if efforts to reduce risk are happening. Do, however, emphasise comparison with others of the same age and sex (our figure 3), since the inevitable risk increase with age may otherwise disappoint.

We appreciate Pettilä's desire for independent validation, but in this instance it is unrealistic. One would need a new current large cohort of subjects, screened for all risk factors and followed up for five years. From which populations would they come, what inclusion criteria would apply, and how long can we wait for such validation?

It cannot realistically be done in time or with sufficient representativeness. As Vesti-Nielsen points out, any risk score for predicting five year cardiovascular mortality is by definition based on old data. We could adopt the defeatist stance that we cannot predict the future, and a person's true risk is dependent on time and place (and countless other features). This negative attitude is not shared by thousands of people accessing our website (http://www.riskscore.org.uk/) to quantify individual risk. We argue that our risk score is based on a large international database of high quality. The established nature of the risk factors we included and the large number of deaths observed mean we have a precise estimation of individual risk.

It would be naive to propose that the risk score and its consequent probability of cardiovascular death within five years give the true exact future risk for each individual. But that is not really the point. Our risk score should be seen as a motivational tool that encourages assessment of a patient's overall risk rather than focusing on a single factor—for example, blood pressure. This has been achieved in a style that many primary care physicians and other health professionals are finding useful. Many people might prefer to have their own copy of our risk score program rather than accessing our website. If so, please email Mike Bennett (mike.bennett{at}lshtm.ac.uk), whom we thank for writing this program.