Secondary Logo

Share this article on:

Of headwind and tailwind, regression to the mean and Wilder's principle

Messerli, Franz H.a,b,c; Rexhaj, Emrusha

doi: 10.1097/HJH.0000000000002010
EDITORIAL COMMENTARIES

aDepartment of Cardiology, University Hospital Bern, and Department for Biomedical Research, University of Bern, Bern, Switzerland

bMount Sinai Medical Center, New York, New York, USA

cJagiellonian University, Krakow, Poland

Correspondence to Franz H. Messerli, MD, Department of Cardiology, University Hospital Bern, CH-3010 Bern, Switzerland; Department for Biomedical Research, University of Bern, Bern, Switzerland. E-mail: messerli.f@gmail.com

Two articles in this issue deal with what the authors call regression toward the mean (RTM). Moore et al. [1] show that high baseline ambulatory blood pressure (ABP) readings were substantially lower on long-term follow-up, and conversely low baseline readings tended to be higher. Thus, unexpectedly even ABP, similar to other physiological measures that exhibit intrinsic variability, is subject to RTM. Salam et al., in a meta-analysis of 86 trials with 349 488 participants, show that most mean blood pressure (BP) change was due to RTM rather than treatment. At high baseline BP levels, even after rigorous hypertension diagnosis, downward RTM caused much of the fall in BP and mutatis mutandis, at low baseline BP levels, upward RTM increased BP levels, even in groups exposed to antihypertensive therapy [2]. RTM is a statistical phenomenon that occurs when repeated measurements are made on the same subject or unit of observation. It was originally described by Sir Francis Galton, who called it ‘regression toward mediocrity’ [3]. If a variable is extreme on its first measurement, it will tend to be closer to the average on its second measurement – and conversely, if it is extreme on its second measurement, it will tend to have been closer to the average on its first [4]. If the first randomized controlled trial suggests that a new treatment is outperforming all previous treatments for a given disease or condition, it is more likely to be closer to the mean when it is tested the second time around. RTM happens because most often values are observed with random error. In contrast, systematic error, in which the observed values are consistently biased, is not the cause of RTM. As it is exceedingly rare to find data without random error, RTM is a common phenomenon.

RTM must be distinguished from Wilder's principle, the observation that pretreatment level determines to a large extent the change per se, that is, the principle or law of the initial value (German: Ausgangswertgesetz) [5]. Wilder proposed that the ‘direction of response of body function to any agent depends to a large degree on the initial valuable of that function’ [6]. RTM is a purely statistical phenomenon and the correction can be upward or downward. In contrast, Wilder's principle states that the mean response in a population depends on the mean pretreatment value. In other words, the higher is the mean BP in a patient population at baseline the greater will be the decrease in BP with identical antihypertensive therapy. Wilder's principle of initial value is an important concept, as it predicts that in the patients with the most severe abnormality or disease, that is, the ones with the highest BP, the fall in BP will be greater with the same medication than in those with less severe hypertension. If the effects were similar (i.e. if the fall in BP were independent of pretreatment level), we would encounter much more clinical complications related to hypotension with antihypertensive treatment.

Most illuminating in this context is the thorough meta-analysis by Mancia and Parati [7] showing that treatment-induced reduction in BP in general was smaller for the 24-h ABP than for the office BP. However, when we examine the stratum in which pretreatment BPs were similar, that is, 150–159 mmHg, the decreases in office and ambulatory BP monitoring (ABPM) were equal. This would indicate that the disparate decreases in office and ambulatory BP recordings were predominantly related to uneven pretreatment levels.

When looking at the studies of Moore et al. [1] and Salam et al. [2], it becomes clear that both, RTM and Wilder's principle are operative in the treated patient group. At a first glance, this is puzzling in the ABPM study by Moore et al. as multiple baseline measurements (which is exactly what ABPM does) is supposed to be a common way to mitigate RTM [8]. However, as documented here, this measure does not suffice to abolish RTM. The other option to reduce RTM is at the design stage and consists in including a randomly allocated placebo group. With two groups, placebo and treatment, the mean change in the placebo group allows to estimate the change caused by RTM (and placebo effect, if any). At the analysis stage, the best way to deal with RTM is by looking at the data with an analysis of covariance. To distinguish Wilder's principle from RTM we should remember that RTM is not a unidirectional phenomenon but takes place symmetrically from excessively high and excessively low values and occurs in both, the treated as well as the placebo group. In contrast, Wilder's principle is unidirectional and related to treatment only. Thus, in randomized controlled trials RTM is dealt with by design, and by the observed treatment effect in the placebo group, whereas the principle of initial value is not affected by study design.

Where does this leave the practicing physician who is treating patients with high BP?

We should remember that the fall in BP after initiating antihypertensive therapy is influenced by at least four different mechanisms: first, Wilder's principle; second, RTM; third, placebo effect; and last but not least fourth, by the effect of the antihypertensive drug. A large BP drop after initiating antihypertensive treatment could lead to the erroneous conclusion that patient responded exceedingly well to that drug. As stated by Salam et al. [2] such a ‘tailwind’ effect is most pronounced if initial BP is high. In fact, as the astute clinicians know, the higher the pretreatment BP the more pronounced the antihypertensive response. Clearly, attributing all BP reduction after initiating hypertension treatment to the drug alone is prone to grossly overestimate efficacy. Conversely, when initial BP is below average, ‘headwind’ effects occur because now the effect of some of the above four mechanisms are less synergistic and may even become antagonistic. Because of this we should now not erroneously conclude that treatment was less effective.

Little surprise that pharma has used these and other data for marketing purposes. The simple fact that response to a drug is more marked when pretreatment levels are high and gradually diminishes as pretreatment levels become lower seems to attest to its safety and efficacy. Such a drug has been touted to be ‘intelligent’ in that it lowers the endpoint, that is, BP progressively better the more it has been out of the normal range. As discussed above, this is due to being exposed to ‘headwind’ only and certainly not a question of the intelligence of a drug.

Of more concern is the current narrow focus on pretreatment BP levels, and treating only those who fulfill the current definition of hypertension, that is, have a BP above a set level such as 140/90 mmHg. As we have shown [9], in current guidelines, despite identical evidence, three panels of experts arrived at setting apart by as much as 20 mmHg in SBP the definition of what is said to be hypertension and what are optimal on-treatment BP levels.

Instead of rhapsodizing on what consists hypertension and getting lost in the schism of guidelines, it would be far more important for those who preach, those who teach and those who treat [10] to jointly focus on prompt initiation of antihypertensive therapy and its long-term maintenance in all patients with high BP and/or high cardiovascular risk.

Back to Top | Article Outline

ACKNOWLEDGEMENTS

Conflicts of interest

There are no conflicts of interest.

Back to Top | Article Outline

REFERENCES

1. Moore MN, Atkins E, Salam A, Callisaya ML, Hare JL, Marwick TH, et al. Regression to the mean of repeated ambulatory blood pressure monitoring in five studies. J Hypertens 2019; 37:24–29.
2. Salam A, Atkins E, Sundström J, Hirakawa Y, Ettehad D, Emdin C, et al. Effects of blood pressure lowering on cardiovascular events, in the context of regression to the mean: a systemic review of randomized trials. J Hypertens 2019; 37:16–23.
3. Galton F. Natural inheritance. London: Macmillan; 1989.
4. In: Wikipedia, editor. Regression toward the mean. https://en.wikipedia.org/wiki/Regression_toward_the_mean. [Accessed 5 October 2018]
5. Messerli FH, Bangalore S, Schmieder RE. Wilder's principle: pretreatment value determines posttreatment response. Eur Heart J 2015; 36:576–579.
6. Wilder J. Basimetric approach (law of initial value) to biological rhythms. Ann N Y Acad Sci 1962; 98:1211–1220.
7. Mancia G, Parati G. Office compared with ambulatory blood pressure in assessing response to antihypertensive treatment: a meta-analysis. J Hypertens 2004; 22:435–445.
8. Barnett AG, van der Pols JC, Dobson AJ. Regression to the mean: what it is and how to deal with it. Int J Epidemiol 2005; 34:215–220.
9. Messerli FH, Bangalore S. The blood pressure landscape: schism among guidelines, confusion among physicians, and anxiety among patients. J Am Coll Cardiol 2018; 72:1313–1316.
10. Messerli FH, Bangalore S, Messerli AW. Age, blood pressure targets, and guidelines: rift between those who preach, those who teach, and those who treat? Circulation 2018; 138:128–130.
Copyright © 2019 Wolters Kluwer Health, Inc. All rights reserved.