Mistaken conclusions

Tom Marshall, Senior Lecturer in Public Health,
June 16, 2011

To whom it may concern,

The meta-analysis reported by Lasserson et al provides evidence that half of the effect of antihypertensive treatment on blood pressure takes place within the first week of treatment. [1] There is no reason to doubt these findings. However the analysis does not support the conclusion that "estimation of maximal effect could be made between 1 and 2 weeks after initiation of antihypertensive therapy". Nor does it support the view that "this knowledge will guide practitioners in deciding when a newly started antihypertensive agent can be judged to be ineffective".

First, it is not possible to determine whether blood pressure has been reduced by treatment within a week. The authors reach their conclusion after meta-analysis of measurements in 4168 individual patients. But clinicians must decide whether a newly started antihypertensive agent is ineffective on the basis of measurements in one patient. There is an important difference.

The mean reduction in systolic blood pressure on maximum treatment is reported as 14.7 mm Hg. The reduction at one week is therefore 7.4 mm Hg. The aim of measurement is therefore to distinguish between a reduction of 7.4 mm Hg and no reduction in blood pressure.

The short term within-individual variation in office measured blood pressure has a coefficient of variation of 7.3% (for both systolic and diastolic blood pressure).[2] This means that when multiple office measurements are taken over a series of days in an individual with a mean systolic blood pressure of 150 mm Hg the measurements will have a standard deviation of 11 mm Hg [11 = 7.3% x 150].

For an individual with a mean untreated blood pressure of 150 mm Hg and a mean treated blood pressure of 142.6 mm Hg the standard error of the change in blood pressure is 15.1 mm Hg [15.1 = Sqrt(11.0^2 + 10.4^2)]. This means that as a result of chance variation in measured blood pressure before and after treatment, there is a probability of 0.31 that the difference will be less than zero (there is no apparent reduction in blood pressure). In plain language, even if the treatment is effective, on one third of occasions it will appear not to be. The clinician has no way of knowing whether the patient has failed to respond to treatment or the on treatment measurement is higher by chance.

The problem is reduced but not solved by taking the average of three office measurements (meaning three separate visits) before treatment and the mean of three measurements on treatment. This reduces the standard error of a blood pressure of 150 mm Hg to 6.3 mm Hg {6.3 = 11/Sqrt(3)} and it reduces the standard error of the difference to 8.7 mm Hg. The probability that treated blood pressure measurements will be higher than untreated is now 0.20. If the treatment is effective, on one in five occasions the clinician will believe it to be ineffective.

Nor is the problem solved by the use of daytime average 24-hour ambulatory blood pressure before and after treatment. For 24-hour ambulatory blood pressure the within-individual coefficients of variation for systolic and diastolic blood pressure are 5.5% and 4.9%.[3] The probability that a treated daytime average 24-hour ambulatory blood pressure will be higher than an untreated daytime average 24-hour ambulatory blood pressure is 0.26.

With an expected effect size of 7.4 mm Hg in order to be reasonably certain (probability <0.05) that no reduction in measured blood pressure indicates non response, a clinician must take the average of 12 office measurements before treatment and 12 after treatment. Alternatively they may take the average of five daytime average 24-hour ambulatory blood pressures before and after treatment. This is of course impossible within a week.

Second we need to ask how likely it is whether it is likely that the patient will not respond. It is well documented that very little (3%/1% for systolic / diastolic) of the apparent variation between individuals in blood pressure response to treatment is due to genuine differences in treatment response. [4] Most of the apparent variation in treatment response is the result of chance variation in measured blood pressure.

It is not possible to determine the extent to which individual patients respond to antihypertensive treatment. It is essentially unknowable. Fortunately we do not need to determine the extent to which individual patients respond to antihypertensive treatment, because we know that almost all respond similarly.

Yours sincerely

Tom Marshall

Senior Lecturer in Public Health

^ = to the power of (ie: S^2 = S squared) Sqrt = Square root of

REFERENCES

1. Lasserson DS, Buclin T, Glasziou P. How quickly should we titrate antihypertensive medication? Systematic review modelling blood pressure response from trial data Heart 2011;Published Online First: 17 May 2011 doi:10.1136/hrt.2010.221473 2. Keenan K, Hayen A, Neal BC, Irwig L. Long term monitoring in patients receiving treatment to lower blood pressure: analysis of data from placebo controlled randomised controlled trial. British Medical Journal 2009; 338: b1492. 3. Warren RE, Marshall T, Padfield PL, Chrubasik S. Variability of Office, 24-hour Ambulatory and Self-Monitored Blood Pressure Measurements British Journal of General Practice 2010 Sep;60(578):675-80. 4. Bell KJL, Hayen A, Macaskill P, Craig JC, Neal BC, Fox KM, Remme WJ, Asselbergs FW, van Gilst WH, MacMahon S, Remuzzi G, Ruggenenti P, Teo KK, Irwig L. Monitoring Initial Response to Angiotensin-Converting Enzyme Inhibitor-Based Regimens: An Individual Patient Data Meta-Analysis From Randomized, Placebo-Controlled Trials Hypertension Sep 2010; 56: 533 - 539.

Conflict of Interest:

None declared

Conflict of Interest

None declared