¹Department of Public Health and Epidemiology, University of Birmingham, Edgbaston, Birmingham, UK

Correspondence: Tom P. Marshall, (T.P.Marshall@bham.ac.uk)

Measured blood pressure is intrinsically variable because every cardiac cycle produces a different blood pressure. We can think of this variability mathematically. Each measurement is a sample from the population of all possible blood pressures: an estimate of the population mean. If measurements follow a normal distribution, knowing the mean and standard deviation allows us to determine the probability of obtaining a measurement above a given threshold. We can therefore determine the probabliity of misdiagnosis.

How much does measured blood pressure vary? When measured blood pressure is from the average of two or three measurements at each clinic visit, the within-individual coefficient of variation between clinic visits is 9.9% for systolic blood pressure (J.M. Wright and V.J. Musini, personal communication).¹ Thus in a patient whose true mean systolic blood pressure is 130 mm Hg, the standard deviation of measured blood pressure is 13 mm Hg (130 mm Hg 9.9%). From the characteristics of a normal distribution we can determine that with this degree of variability, at 22% of clinic visits, her measured systolic blood pressure will exceed 140 mm Hg. Taking the average of N measurements reduces the coefficient of variation by the inverse of the square root of N. For our patient, the standard deviation of the average of measurements at two successive visits is 10 mm Hg (130 mm Hg 9.9%/). Therefore in 14% of occasions, the average of measurements at two clinic visits will exceed 140 mm Hg. According to the recommendations of the Joint National Committee, this makes her eligible for treatment.^2,3 In other words, after every two clinic visits, she has a 14% probability of being judged eligible for treatment. After ten clinic visits, her probability of being misdiagnosed as hypertensive is 64%.

The fundamental variability of measured blood pressure has a number of implications. Two papers in this journal model the effects of blood pressure variability on the findings of the TROPHY study.^4,5,6 Using slightly different assumptions, they demonstrate that a proportion of patients will be misclassified as hypertensive simply because blood pressure is measured repeatedly (18 times) throughout the study. Because mean blood pressure on treatment is lower, the rate of misclassification is substantially lower in treated patients. But the rate of misclassification is the same when treatment is stopped. Misdiagnosis of hypertension is therefore deferred for two years, leading to a lower apparent cumulative incidence of hypertension. Both papers may underestimate within-individual blood pressure variability and may underestimate misclassification. In the largest study of diastolic blood pressure variability, the coefficient of variation between clinic visits was 11.4–16.6%.⁷

There are further implications. In a low prevalence population, a small probability of being misclassified as hypertensive can exceed the true prevalence of hypertension. In healthy adults <35 years, the probability of being misclassified as hypertensive exceeds the probability of being correctly diagnosed.⁸ This remains true even if 24-h ambulatory blood pressure measurements are used for diagnosis.

The most striking implications of blood pressure variation concern monitoring of patients on treatment. We cannot know the true effect of treatment on blood pressure. We estimate the effect of treatment from the change in blood pressure measured before and after treatment. The standard deviation of the change in measured blood pressure is the geometric sum of the standard deviations of pre and post-treatment measures. Consider a patient whose true mean systolic blood pressure is 140 mm Hg before treatment and 130 mm Hg after treatment. His measured pretreatment blood pressure (from two clinic visits) has a mean of 140 mm Hg with a standard deviation of 10 mm Hg (10 = 140 9.9%/). His measured post-treatment blood pressure (from two clinic visits) has a mean of 130 mm Hg with a standard deviation of 9 mm Hg. The standard deviation of the change in measured blood pressure is therefore 13 mm Hg

The change in measured blood pressure will therefore have a mean 10 mm Hg and a standard deviation of 13 mm Hg. On 5% of occasions, the change in measured blood pressure will be >2 standard deviations different to the mean change in measured blood pressure. This means that on 2.5% of occasions, the change in measured blood pressure will be 36 mm Hg (two standard deviations greater than the mean) and on 2.5% of occasions it will be 16 mm Hg (2 standard deviations less than the mean). On 22% of occasions, the change in measured blood pressure will be 0.77 standard deviations (10 mm Hg) less than the mean. In other words, 22% of measured post-treatment blood pressures will exceed measured pretreatment blood pressures. It is practically impossible to distinguish nonresponse from measurement variation in an individual patient.⁹ It is therefore practically impossible for a clinician to know whether he is changing a drug or dose in response to chance variation in blood pressure or true changes in the underlying mean blood pressure. We know that on average antihypertensive treatment works.¹⁰ We cannot know the precise effect in any individual.

Where do we go from here? First, we need informed scepticism. Advocating treatment of systolic blood pressures >130 mm Hg greatly expands the market for antihypertensive treatment. Every new diagnosis of hypertension is a new potential customer. Every nonresponder is a marketing opportunity for a new drug. All of these aims are assisted by frequent measurement and naïve interpretation of variation. Our patients' interests are different from this. Rather than control blood pressure as an end in itself, patients' interests are in prevention of cardiovascular disease. Indeed the focus on a single risk factor measurement is surely the antithesis of holistic care. From the patient's perspective, the decision to take drugs is best informed by the potential benefits of treatment: the absolute reduction in risk of cardiovascular disease. Reduction in risk of cardiovascular disease is proportionate to cardiovascular risk. The decision to offer drugs should be therefore informed by the patient's pretreatment cardiovascular risk, and not by blood pressure alone. This challenges the current recommendation that a blood pressure of 140/90 mm Hg should be treated irrespective of cardiovascular risk.²

Second, this is not a problem with measurement. It has been demonstrated that even 24-h ambulatory blood pressure measurements do not provide a technical fix.⁹ This is a problem with interpreting measurements. We need a better understanding on how clinicians interpret variation in order to make clinical decisions. This is a new area of research. We also need tools to help clinicians recognize variation in clinical measurement, to interpret that variation correctly and use this interpretation to make better decisions. This is the real challenge of variation.