Blood pressure variability predicts clinical outcomes. Now what?

The high prevalence of hypertension means that even small refinements to treatment can have a significant impact on population health.¹ However, hypertension treatment appears to have much more benefit for some patients than others, largely because of differences in baseline risk. Thus, two patients who achieve similar relative risk reduction can have much different absolute risk reductions, based on their baseline level of risk. Table 1 demonstrates this for stroke, using 4 studies which I admittedly chose to make this point.^2–5 Age and baseline systolic blood pressure (BP) were the most obvious risk factors generating the widely disparate event rates in these studies. Current guidelines use these and other established risk factors to classify patients as low, medium or high risk, but with minimally different recommended approaches to treatment.^1,6 Researchers have identified additional potential risk factors, more recently including variability in blood pressure.⁷

Table 1.

Progressive increase in absolute risk reduction (ARR) with increasing baseline risk and stable relative risk reduction (RRR) in a series of clinical trials

Trial Name	Baseline values		Strokes per 1000 pt-yrs		RRR	ARR (per 1000 pt-yr)
	Age (yrs)	Mean SBP (mmHg)	Control	Treated
STOP4	76	195	31.3	16.8	46%	14.5
Syst-Eur5	70	174	13.7	7.9	42%	5.8
AUST3	51	157	4.5	2.4	47%	2.1
MRC I2	50	158	2.6	1.4	46%	1.2

In this issue of Hypertension, Juhanoja and colleagues present research with several methodological strengths to argue that variability over several days in home BP measurements is associated with cardiovascular risk.⁸ Previous studies have demonstrated the association of adverse cardiovascular outcomes with variability among BP measurements separated by months, and variability that occurs over the course of a single day.⁷ In particular, their study population combines four population based, geographically diverse cohorts with excellent and long term (mean 9.3 years) follow up using validated outcome ascertainment methods. They are able to adjust for the baseline risk factors included in commonly used risk calculators (population averages had be used to estimate total and HDL cholesterol for approximately 10% of the cohort), as well as baseline office BP. The authors also attend to two common methodological issues regarding measurement of BP variability. First, they examined several commonly used measures of variability (coefficient of variation, average real variability and variation independent of the mean) and found similar results with each. Second, they demonstrate that BP variability was associated with adverse cardiovascular outcomes regardless of whether it was determined using 3, 7 or 3–7 daily measurements over a one week period.

Thus, this analysis provides incremental new evidence supporting the prognostic significance of home BP variability, although similar analyses of some of the individual cohorts included in the present work have been published previously. In addition, several factors temper enthusiasm for the cutpoint approach they advocate. First, they identify their cutpoint empirically in this cohort but do not verify in a second cohort. Future attempts to identify cutpoints should use split sample validation, or prospectively test cutpoints that have been proposed in other studies. Second, while they advocate for a cutpoint, their analysis did not find reason to reject their null hypothesis that the increase in risk was linear. A more convincing analysis would demonstrate that using a cutpoint improves predictive power compared to other approaches. Third, they treat deciles as if all values in the decile were insignificantly different – however, the tenth decile of BP variability, where the effect appears, included a wide range of variability, with the coefficient of variation ranging from 11 to 37.7. They do not show why we should believe risk begins to increase at the bottom of this range.

While there are reasons to question the specific cutpoint they identify or even using a cutpoint at all, Juhanojo et al do highlight the important issue of whether and how BP variability should be used in clinical hypertension management. First, researchers should identify what measure of BP variability is most useful for risk stratification in the target population – I would argue this is persons with hypertension, where knowledge of risk is most likely to affect management decisions. Thus, researchers should compare the predictive value of BP variability measured over hours, days or months, in clinical populations. Other details may also be important – the measures used by Juhanoja were the first one measured each day by trained participants using an automated oscillometric cuff in their own home between 5 AM and 12 AM. More or less well-trained individuals, or measures taken at more times throughout the day, might provide a measure of BP variability that is more or less predictive.

Populations are also important. Much of the work documenting the predictive value of BP variability comes from clinical trial populations, though relatively few allow comparison of methods to measure BP variability.^9,10 These results appear to demonstrate a more substantial impact of BP variability when measured across visits than when measured by an ambulatory BP monitor. They also seems to demonstrate greater predictive power than is seen in the present work by Juhanoja et al. This may be because these clinical populations had higher event rates than the 1.38 events/100 patient years seen in the present study, or perhaps because this is just a better way to analyze variability. Future similar analyses of data from the large SPRINT¹¹ and HOPE-3¹² studies may provide further clarification. Use of BP variability based on such “in-clinic” measurements taken over months or years has become increasingly feasible with the proliferation of electronic health records, although it is not yet clear if values obtained in routine practice will function as well as these more careful BP measures. Analyses of some clinical trial data may also provide access to measures of preclinical cardiovascular disease endpoints that could allow researchers to determine whether BP variability is also important early in the pathophysiologic processes triggered by hypertension.⁹

In addition to identifying the optimal method to measure BP variability, we must also consider how much BP variability adds to currently available risk scores. When Juhanojo et al tested the incremental information provided by their measure of risk they found a very low impact; while the c-statistic was significant, it was very small and there was no significant movement between the 10 year cardiovascular risk categories they selected (<5%, 5–10%, 10–20%, >20%). Future research around the prognostic importance of BP variability should examine how it might add to existing risk scores as well as how it compares to other measures of risk that are not included in the most commonly used risk scores, such as coronary artery calcium or c-reactive protein.

Finally, researchers must consider whether improved risk estimates are likely to add value to hypertension management. Hypertension treatment guidelines generally describe risk estimation as a key first step in clinical practice, but beyond suggesting that low risk people might be treated non-pharmacologically for a few months, almost all would treat individuals with BP persistently above 140/90 mmHg, regardless of absolute risk and therefore potential benefit. Previous suggestions that diabetes or prior cardiovascular events justified treatment at lower levels of BP¹³ were largely dropped after randomized trials did not support this differentiation. One risk factor that continues to be used to alter treatment thresholds is age; paradoxically, the tenor of recommendations in this population is that people at higher risk (those who are older) should be treated less aggressively.¹ This “risk independent” approach reflects randomized trial data, which have in fact shown benefit with treatment even in very low risk populations.^2,3 Because trials have often excluded very old individuals, there is less data to justify their treatment. Moreover, no randomized trials have examined whether a strategy tailoring hypertension treatment goals based on risk would improve outcomes.

As future studies determine which measures of BP variability are most predictive of risk and most practical for clinical use, it will be important to consider whether they identify a clinical population that should be managed with a different approach. It is possible that BP variability identifies populations at higher risk for adverse effects of hypertension treatment, as well as at higher risk for adverse consequences of hypertension, which could mean this high risk group should be treated less aggressively. Changes in management based on risk, whether or not this risk estimate includes BP variability, will need to be justified by randomized trials demonstrating that risk estimation can identify one or more groups that benefit more from more (or less) intensive management.