Accumulating Evidence of Benefits From Intensive Blood Pressure Lowering
Are We There Yet?
See related article, pp 642–653
In this issue of the journal, Verdecchia et al1 report an important analysis of the accumulation of evidence comparing more versus less intensive blood pressure (BP)–lowering strategies. Their key message is critical for guidelines and practice internationally: with the addition of SPRINT (Systolic Blood Pressure Intervention Trial),2 this set of trials now provides compelling evidence that more intensive BP-lowering reduces stroke and myocardial infarction (MI), and significant reductions are now also seen in cardiovascular death and heart failure. The authors have also assessed sequential monitoring boundaries for each outcome, making the analogy of a hypothetical Data and Safety Monitoring Board (DSMB) assessing the accumulation of evidence in the field.
Before examining the implications, it is worth reviewing the techniques and aims of the cumulative meta-analysis and trial sequential analysis methods used by Verdecchia et al.1 Cumulative meta-analysis orders trials by publication date and presents updated meta-analysis estimates with each additional trial, with the aim of establishing if and when a treatment effect reaches statistical significance. Cumulative meta-analysis has a long history of providing important insights. Most notably, a highly influential series of cumulative meta-analyses showed that traditional narrative reviews often failed to mention important advances, delayed recommending effective preventive measures, and continued to recommend some treatments long after accumulated evidence had shown them to be ineffective or harmful.3 As such the article was pivotal in the widespread adoption of systematic reviews in recent decades and indeed was one of the seminal articles in the evidence-based medicine movement. Trial sequential analyses add an important dimension and are motivated by the fact that as evidence emerges, chance extreme findings are common. It is generally not intuitive how likely, by chance alone, early data on effective treatments can appear neutral or even harmful; and early data on ineffective treatments can have positive trends.4 This has led to the development of numerous statistical approaches to guide DSMBs charged with advising whether individual trials should continue—these techniques basically require interim results to have much more extreme significance than convention P<0.05 thresholds. The same principle should be applied to meta-analyses—a few small trials can easily result in chance extreme results, from pure random error, even without publication bias. However, the statistical approaches to address this issue are more complicated and the assumptions are greater for meta-analyses, as evident here—it involves estimating a “required information size” (a sample size that should have been obtained for the assessment to have sufficient power) and a Z curve (a time-dependent P value threshold, which if passed, would provide extreme statistical evidence of effect) for each outcome—these estimates are dependent on the event rate, effect size, and between-trial heterogeneity for that outcome.
Verdecchia et al1 concluded that with the addition of SPRINT, the accumulated evidence on stroke, MI, heart failure, and vascular death crossed the nominal significance barrier and the effect on total mortality was borderline positive (P=0.09); however, only the effects on stroke and MI crossed the sequential monitoring boundary. They also noted that current hypertension guidelines require revision on the basis of these results, given that 7 trials targeted systolic blood pressure of <130 mm Hg, 7 targeted diastolic blood pressure of <80 mm Hg, and mean BP levels on more versus less intensive treatment were 129.4/75.9 mm Hg versus 137.9/80.8 mm Hg, respectively. Thus, the evidence now clearly shows that high-risk patients can benefit from further BP lowering <140 mm Hg; SPRINT showed this most clearly as an individual trial, but the totality of the evidence is even more convincing.
It is worth considering the implications of this review in a similar fashion to the remit of a DSMB. A DSMB is generally charged with advising whether the accumulated trial evidence, along with existing evidence, provides proof beyond reasonable doubt that treatment is on balance either definitely harmful or definitely favorable for all, or for a particular category of patients. Several issues should be considered. First, a DSMB would not require every single outcome under consideration to cross the sequential monitoring boundary nor even to reach statistical significance—if overall benefit or harm is unequivocal, then findings for individual outcomes are of less importance. Second, a DSMB should consider all relevant evidence, not just the evidence accumulating in the trial it is monitoring. It would be remiss not to consider a treatment side effect that becomes apparent in another set of trials, and a similar approach should be adopted for treatment benefits. There are clearly a large number of other BP-lowering trials. As well as the 53 000 patients in more versus less BP-lowering trials reviewed by Verdecchia et al,1 there have been over 60 000 patients in treatment versus control trials among hypertensive patients and >200 000 patients in treatment versus control trials in other patient groups.5 Collectively, these trials show conclusively that BP lowering reduces stroke, CHD, heart failure and total mortality.5 There are compelling reasons to consider evidence from more versus less and some versus none trials alongside each other because they are all BP difference trials. Prospective studies show that each absolute difference in BP is associated with a constant proportional difference in CHD and stroke risk.6 Indeed, across many different areas in biology, a given absolute delta in exposure is associated with a constant proportional difference in outcome.7 Thus, there is no particular reason to expect a 6 mm Hg systolic blood pressure difference to act differently if it occurs in a trial of more versus less intensive BP lowering or in a trial of some BP lowering versus none. Hence, it seems difficult to justify considering more versus less trials in complete isolation—from a Bayesian perspective, one’s prior probability should not be that more intensive BP lowering has no benefit. Although, of course, this has to be evaluated directly and the effects of BP differences in different patient populations and at different BP levels should be quantified, a priori one expects them to be comparable. The trial evidence confirms this, as shown in the Figure that compares the more versus less trials reviewed by Verdecchia et al1 with trials comparing active BP lowering with control.5 Furthermore, there is almost exact consistency with the size of the reduction in MI and stroke seen in these trials with that predicted by prospective observational studies.6
Finally, a DSMB should consider if the benefits are clearly greater than harms, overall or in a subgroup. Although data on treatment side effects are reported much less consistently than for cardiovascular events, there is as yet no trial evidence that BP lowering increases any fatal or life-threatening event, ie, events of comparable severity to MI or stroke. However, an excess of less severe adverse events is apparent—on average in previous trials, there were about three side effects sufficient to withdraw treatment for each major vascular event prevented.8 The excess of side effects increases with the degree of BP reduction, as does the extent of the cardiovascular risk reduction.8 It is clear, therefore, that the margin of safety for benefits exceeding side effects will be greatest in patients at highest risk. Here, again a DSMB would appropriately assess heterogeneity rather than require separately significant evidence for each patient subgroup. SPRINT played a major role in extending our knowledge on the patient subgroups that can benefit from intensive BP lowering—the previous trials reviewed mostly involved patients with one or more of hypertension, diabetes mellitus, and renal disease, whereas SPRINT extended evidence of benefit to the elderly, patients with chronic renal failure, and those at high global cardiovascular risk. There is no evidence of lack of effect in any high-risk patient group studied to date.5,9,10
What is the relevance of these findings for research? One consideration is whether further trial evidence is required in some high-risk patient groups or among other patient groups—in large part, this rests on judgments about generalizability of existing data versus need for direct evidence in each group. Verdecchia et al1 suggest that their results support the decision to continue the ongoing ESH-CHL-SHOT (Optimal BP and Cholesterol Targets for Preventing Recurrent Stroke in Hypertensives), among hypertensive patients with a history of stroke or transient ischemic attack being randomized to three systolic BP targets: 135 to 145, 125 to 135, and <125 mm Hg. We would suggest a key question now is how lower BP targets can be achieved consistently, reliably, and with minimum adverse effects.
In conclusion, Verdecchia et al1 present an important new analysis showing the accumulation of evidence, indicating lower BP targets bring important benefits for high-risk individuals. Together with previous relevant evidence, the rationale for lowering the threshold for initiating BP treatment and target levels in high risk patients is now compelling. The big question now is how—progress in reaching existing BP goals has been steady but not spectacular. With much lower targets, system-wide efforts will clearly need to be redoubled.
Sources of Funding
A. Rodgers was funded by a research fellowship from the National Health and Medical Research Council of Australia.
George Health Enterprises, the social enterprise arm of The George Institute for Global Health, has received investment to develop combination products containing aspirin, statin, and BP-lowering drugs.
The opinions expressed in this article are not necessarily those of the editors or of the American Heart Association.
- © 2016 American Heart Association, Inc.
- Verdecchia P,
- Angeli F,
- Gentile G,
- Reboldi G
- Law MR,
- Wald NJ
- Thomopoulos C,
- Parati G,
- Zanchetti A