Published online before print August 12, 2004, doi: 10.1161/01.STR.0000140886.65694.3a
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
(Stroke. 2004;35:2237.)
© 2004 American Heart Association, Inc.
Letters to the Editor |
Relationship Between Stroke Relative Risk and Change in Systolic Blood Pressure: The Misuse of Meta-Regression
Biostatistics Unit, Hospices Civils, Lyon, France
To the Editor:
In their recent review on the relationship between change in blood pressure and risk of stroke, Lawes et al1 reported that a 31% relative risk reduction (RRR) may be expected from a 10 mm Hg decrease in systolic blood pressure (SBP). This point estimate was derived from a meta-regression based on 7 meta-analyses of 47 controlled clinical trials, regressing the pooled RRR against the average difference in SBP change between treatment groups. Out of the 7 meta-analyses used in the meta-regression, 3 compared an active treatment versus placebo, 1 a more intensive versus less intensive treatment, and 3 compared 2 different classes of antihypertensive treatments. The authors emphasized that for a 10 mm Hg decrease in SBP, the 31% RRR estimated from clinical trial results is (nicely!) consistent with the 25% to 36% RRR observed in population-based epidemiological cohorts. The conclusion is that the effect obtained by pharmacological treatments to decrease SBP may directly translate into a decrease in stroke risk at a population level.
I would be cautious about these results.
First, the authors should have given a 95% CI around their 31% point estimate, which I calculated to be (12.2%–49.8%); ie, covering a quite wide spectrum of values.
Second, and more problematic, is the choice of the control groups in their meta-analysis. For example, in 1 of the 7 meta-analyses, ß-blockers and/or diuretics were compared with calcium antagonists, yielding a RR of 1.08 (95% CI, 0.99 to 1.16) for an average decrease in SBP of 1 mm Hg. As well, the authors could arbitrarily have chosen to compare calcium antagonists to ß-blockers/diuretics, yielding a RR of 0.92 (ie, 1/1.08) for an average increase in SBP of 1 mm Hg. Using the latter comparison in the meta-regression instead of the former provides an estimated relationship between RRR and SBP difference of 22% (95% CI, 14% to 30%), quite different from the published 31%.
Third, if instead of using pooled RR and average SBP effects from the 7 meta-analyses they had used the raw results observed in the 47 trials to perform the meta-regression, the estimated slope would have been equal to 22% (95% CI, 14% to 30%). If blocker/diuretic and calcium antagonist groups were reversed as above, the slope would have further decreased to 17% (95% CI, 7% to 21%).
Meta-regression is subject to several potential pitfalls. Appropriate modeling methods should be used,2 with cautious interpretation of the results.3 A fundamental difficulty with meta-regression may be the choice of the reference group such that all treatment comparisons have common meaningful effects across trials4 (eg, active treatment versus placebo, high dose treatment versus low dose treatment). As exemplified here, situations exist where no obvious reference group can be identified for some comparisons, leading to different results depending on which reference group was arbitrarily considered.
In short, the 31% reduction in stroke risk for 10 mm Hg SBP reduction published by Lawes et al is subject to important variations depending on the analysis performed. It should not be taken for truth in future medical papers.
References
- Lawes CM, Bennett DA, Feigin VL, Rodgers A. Blood pressure and stroke: an overview of published reviews. Stroke. 2004; 35: 1024.
[Abstract/Free Full Text] - Thompson SG, Sharp SJ. Explaining heterogeneity in meta-analysis: a comparison of methods. Stat Med. 1999; 18: 2693–2708.[CrossRef][Medline] [Order article via Infotrieve]
- Thompson SG, Higgins JP. How should meta-regression analyses be undertaken and interpreted? Stat Med. 2002; 21: 1559–1573.[CrossRef][Medline] [Order article via Infotrieve]
- Hughes MD. Evaluating surrogate endpoints. Control Clin Trials. 2002; 23: 703–707.[CrossRef][Medline] [Order article via Infotrieve]
Response
Clinical Trials Research Unit, Department of Medicine, University of Auckland, Auckland, New Zealand
We thank Drs Messerli, Fonier, and Boutitie for their comments. Drs Messerli and Fonier suggest that ß-blockers are less effective than other agents, especially for stroke prevention. The benefits of ß-blockers are difficult to assess precisely, since there are few direct randomized comparisons with other agents and most trials of ß-blockers also included other blood pressure lowering interventions (most commonly, diuretics). However, inference is not aided by focus on underpowered individual studies; eg, Drs Messerli and Fonier quote the TEST and Dutch TIA trials, but the blood pressure reductions achieved in these trials were only a few mm Hg and so, unsurprisingly, their results were equivocal. The evidence from direct randomized comparisons does not indicate a clear difference for ß-blockers compared with diuretic-based therapy1 and this is consistent with the dose-response we observed. Furthermore, the clear evidence of benefit from ß-blockers among patients with coronary disease2 and heart failure3 makes it extremely unlikely that there would be "complete inefficacy" in patients whose predominant diagnosis is hypertension. Hence we maintain that the totality of the evidence shows no detectable difference between the main drug classes mm Hg for mm Hg in their reductions in stroke and coronary heart disease risk.1,4 Since all agents lower blood pressure by about the same modest amount, and their effects are additive,5 the key issue seems to be which combinations of 2 or more drugs should be provided and how long-term adherence can be maximized.
We are grateful to Dr Boutitie for her thorough review of our article and calculating confidence intervals for meta-regression slope estimates. We agree that any meta-regression is subject to a number of limitations, and there are instances in which there is no obvious reference group and some arbitrary choices are required. However, we do feel that, ideally, overviews should be based on the individual participant results from trials rather than collated trial results. The other meta-regression estimates presented by Dr Boutitie are all consistent with the epidemiological associations (one quarter to one third lower stroke risk per 10 mm Hg), given an appropriate degree of uncertainty about the precision provided by this method.
References
- Collins R, Peto R, MacMahon S, Hebert P, Fiebach NH, Eberlein KA, Godwin J, Qizilbash N, Taylor JO, Hennekens CH. Blood pressure, stroke, and coronary heart disease. Part 2. Short-term reductions in blood pressure: overview of randomised drug trials in their epidemiological context. Lancet. 1990; 335: 827–838.[CrossRef][Medline] [Order article via Infotrieve]
- Yusuf S, Peto R, Lewis J, Collins R, Sleight P. Beta blockade during and after myocardial infarction: an overview of the randomized trials. Prog Cardiovasc Dis. 1985; 27: 335–371.[Medline] [Order article via Infotrieve]
- Doughty R, Rodgers A, Sharpe N, MacMahon S. Effects of beta-blocker therapy on mortality in patients with heart failure. A systematic overview of randomised controlled trials. Eur Heart J. 1997; 18: 560–565.
[Abstract/Free Full Text] - Blood Pressure Lowering Treatment Trialists’ Collaboration. Effects of different blood-pressure-lowering regimens on major cardiovascular events: results of prospectively-designed overviews of randomised trials. Lancet. 2003; 362: 1527–1535.[CrossRef][Medline] [Order article via Infotrieve]
- Law MR, Wald NJ, Morris JK, Jordan RE. Value of low dose combination treatment with blood pressure lowering drugs: analysis of 354 randomised trials. BMJ. 2003; 326: 1427–1431.
[Abstract/Free Full Text]
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||