Jan A. Staessen, MD, PhD, Studiecoördinatiecentrum, Laboratorium Hypertensie, Campus Gasthuisberg, Gebouw Onderwijs en Navorsing, Herestraat 49, B-3000 Leuven, Belgium E-mail: firstname.lastname@example.org
In a quantitative overview of published trials, we investigated whether pharmacologic properties of antihypertensive drugs, as opposed to reduction in blood pressure, explain cardiovascular outcomes in hypertensive or high-risk patients. We used meta-regression to investigate the association between the odds ratios of outcome (experimental vs. reference treatment) and the corresponding blood pressure differences between study groups. Thus, we correlated odds ratios with between-group differences in systolic pressure. We then compared odds ratios of benefit observed in recent trials with those predicted by meta-regression on the basis of the differences in systolic pressure between randomized groups. Among nine actively-controlled trials in hypertension, significant differences in systolic pressure (follow-up minus baseline) between randomized groups (experimental minus reference) were observed in the ALLHAT, CAPPP, MIDAS, and NORDIL trials. Furthermore, the differences in achieved systolic and/or diastolic pressure between study groups were also significant in the hypertension trials and studies in high-risk patients, which involved untreated control patients. The differences between the observed odds ratios and those predicted by meta-regression did not reach statistical significance except for NORDIL and the single-drug therapy subgroup of the PROGRESS trial. In NORDIL, the risk of stroke was lower on diltiazem than on the older drug classes despite a 3.1 mm Hg higher systolic pressure on the calcium channel blocker. In PROGRESS, perindopril alone reduced blood pressure by 5/3 mm Hg, but did not affect the incidence of all cardiovascular events or the recurrence of stroke. In conclusion, the finding that in the reviewed trials blood pressure reduction largely accounted for outcome emphasizes the desirability of tight blood pressure control. The hypothesis that blood pressure-lowering medications might influence cardiovascular prognosis over and beyond their antihypertensive effect remains to a large extent unproved.
Placebo-controlled trials in hypertension have shown that antihypertensive drugs reduce cardiovascular morbidity and mortality.1,2 Recent outcome trials investigated the benefits associated with different levels of blood pressure control,3,4 or compared older classes of antihypertensive drugs such as diuretics or β blockers with newer agents such as calcium channel blockers,5–9 angiotensin-converting enzyme (ACE) inhibitors,8,10,11 or the α blocker doxazosin.12 Further-more, placebo-controlled studies explored whether calcium channel blockers13 or ACE inhibitors14–16 may confer additional benefit over and beyond blood pressure lowering in specific groups of patients with cardiovascular renal disorders. In the Heart Outcomes Prevention Evaluation (HOPE)14 and Antihypertensive and Lipid-Lowering Treatment to Prevent Heart Attack Trial (ALLHAT) studies,12 significant differences between the randomized groups in systolic and/or diastolic pressure were observed. This makes the interpretation of the results difficult. For this reason, we used meta-regression analysis to explore to what extent the morbidity and mortality results of recent outcome trials could be explained in terms of blood pressure reduction per se.17
More recently, four placebo-controlled outcome trials provided additional outcome results on treatment with antihypertensive drugs.17–20 Three studies tested the efficacy of angiotensin II receptor antagonists in patients with diabetic nephropathy.18–20 The Perindopril Protection Against Recurrent Stroke Study (PROGRESS)21 investigated whether an ACE inhibitor alone or in combination with a diuretic would reduce stroke recurrence in patients with a history of minor stroke or transient ischemic attack. In this review, we update our meta-regression analysis, which was previously published elsewhere.17
The inclusion and exclusion criteria for trials in the present analysis have been previously described.17 Briefly, we searched for outcome trials that tested drugs with blood pressure-lowering action in normotensive or hypertensive patients who did not have overt congestive heart failure at baseline. Other inclusion criteria were a randomized controlled parallel group design, publication in a peer reviewed journal, evaluation of blood pressure and cardiovascular events, follow-up of 2 years or longer, and a sample size of 100 or more. In the present analysis, we included 24 trials in hypertension3–12,22–36 and six placebo-controlled trials involving both normotensive and hypertensive subjects (Table I).14–16,19,21,37 Among these 30 trials, two had a single-blind design with alternate allocation of consecutive patients to placebo or active treatment.31,33
CV=cardiovascular; DM=diabetes mellitus; SBP/DBP=systolic/diastolic blood pressure; HCTZ=hydrochlorothiazide; HCTZ/A=hydrochlorothiazide/amiloride; SR=sustained release; ACEIs=angiotensin-converting enzyme inhibitors, DHPs=dihydropyridine calcium channel blockers (felodipine or isradipine); GITS=gastro-intestinal therapeutic system; NR=not reported UKPDS=United Kingdom Prosepective Diabetes Study; CAPPP=Captopril prevention project; INSIGHT=International Nifedipine GITS Study-Intervention as a Goal for Hypertensive Treatment *Other trial acronyms are explained in the text of the article; †a large proportion of patients were on antihypertensive drugs in ALLHAT, PART 2, SCAT, HOPE, RENAAL and PROGRESS; ‡mean or median follow-up
In keeping with our inclusion criteria, we excluded 13 trials,13,18,20,38–47 which included fewer than 100 subjects,42–44 covered less than 2 years of follow-up,39,41 or did not report systolic pressure13,40,45,46 or the number of cardiovascular events.18,20,38,47 We did not include a population-based trial48 and actively-controlled trials that compared old with old drugs,49–51 or new with new drugs,52,53 because only a few such studies were available, and because it was difficult to define the reference group for the calculation of the odds ratios.
Because of the small number of events and similarity of trial design, in our analysis we combined three actively-controlled trials that tested a calcium channel blocker against a thiazide diuretic (the Multicenter Isradipine Diuretic Atherosclerosis Study [MIDAS],36 National Intervention Cooperative Study in Elderly Hypertensives [NICS],5 and Verapamil in Hypertension and Atherosclerosis Study [VHAS]9). For similar reasons, we also combined the Prevention of Atherosclerosis with Ramipril Trial (PART 2)15 and Simvastatin/Enalapril Coronary Atherosclerosis Trial (SCAT)16 that compared ACE inhibitors with placebo in patients at high cardiovascular risk.
We based our analysis on summary statistics reported in the literature.3–12,14–16,19,21,23–37,41,54–56 With the exception of fatal combined with nonfatal events in the European Working Party on High Blood Pressure in the Elderly (EWPHE) trial,26,54 all out-come results were reported on the basis of an intent-to-treat principle. Within each trial, the reference group consisted of patients left untreated34,35 or allocated placebo,14,15,19,21,23–34,37,41 or the patients randomized to older drug classes5–12,36 or to a treatment strategy leading to less blood pressure control.3,4 We had to accept the definitions of events used by study investigators. However, whenever possible, all cardiovascular events included stroke, myocardial infarction, congestive heart failure, and cardiovascular or sudden death.
We used the SAS statistical package (SAS Institute, Cary, NC), version 8.1, to correlate odds ratios of experimental vs. reference treatment with the corresponding blood pressure differences. For these calculations, odds ratios were logarithmically transformed. The regression lines were weighted by the inverse of the variance of the individual odds ratios.57 Net treatment effects on blood pressure were determined by subtracting the mean change in the experimental group (follow-up minus baseline) from the corresponding mean change in the reference group.
The 30 studies included in the meta-regression represent 149,407 patients (Table I). They comprised nine actively controlled trials5–12,36: the Hypertension Optimal Treatment (HOT)3 study that investigated different levels of blood pressure control, three placebo-controlled trials in isolated systolic hypertension (the Systolic Hypertension in the Elderly Program [SHEP],28 Systolic Hypertension in China [Syst-China] trial,33,58 and Systolic Hypertension in Europe [Syst-Eur] trial32,59), four placebo-controlled trials in normotensive or hypertensive patients at high cardiovascular or renal risk (HOPE,14 PART 2,15 SCAT,16 and Reduction of Endpoints in NIDDM with the Angiotensin II Antagonist Losartan Study [RENAAL]19), two trials in patients with a history of minor stroke or transient ischemic attack (Post-stroke Antihypertensive Treatment Study [PATS]37 and PROGRESS21), and 11 older trials testing the efficacy of antihypertensive drugs against no treatment (trial of hypertension in elderly patients in primary care [HEP]35 and Oslo Study on the Treatment of Mild Hypertension [OSLO]34) or placebo (Australian Trial in Mild Hypertension [ATMH],25 EWPHE,26 Hypertension-Stroke Cooperative Study [HSCS],23 Medical Research Council trial of treatment of mild hypertension [MRC-1],27 Medical Research Council trial of treatment of hypertension in older adults [MRC-2],30 Swedish Trial in Old Patients with hypertension [STOP-1],29 Shanghai Trial of Nifedipine in the Elderly [STONE],31 United States Public Health Service Hospitals Cooperative Study [USPHS]24 and Veterans Administration Cooperative Study in patients with dias-tolic pressure averaging 90–114 mm Hg [VACS]41).
The meta-regression line between the odds of an event and the differences in systolic pressure between the study groups was linear for cardiovascular mortality (Figure 1), i.e., the greater the difference the greater the benefit. It was curvilinear for all fatal and nonfatal cardiovascular events (Figure 2). The meta-regression line was also curvilinear for stroke and myocardial infarction including sudden death, if these two specific cerebrovascular and cardiac end points were analyzed separately. We did not specifically investigate the relationship between the odds of congestive heart failure and the between-group differences in systolic pressure because few studies reported the incidence of heart failure as a separate outcome in the analysis.
Significant differences between the randomized groups (experimental minus reference) in systolic and/or diastolic pressure were observed in four actively-controlled trials (Figure 2). As expected, the differences in achieved systolic and/or diastolic pressure between study groups were also significant in the hypertension trials which involved untreated control patients,23–35,41 as well as in other placebo-controlled trials.14–16,19,21,37
Further analyses compared the observed odds ratios with those predicted by the meta-regression lines for three actively-controlled trials7,10,12 (Table II) and five placebo-controlled studies (Table III).14–16,19,21 The authors of these reports suggested benefit or harm of the drugs tested over and beyond the effects on blood pressure. In contrast to their interpretation, the differences between the observed odds ratios and those predicted by the meta-regression lines (Figures 1 and 2) did not reach statistical significance except for the Nordic Diltiazem (NORDIL)7 and PROGRESS trials.21 In NORDIL, the risk of stroke was lower on diltiazem than on the older drug classes despite a 3.1 mm Hg higher systolic pressure. In the perindoprilonly subgroup of the PROGRESS trial,21 systolic pressure was reduced by 5 mm Hg, but monotherapy with the ACE inhibitor did not affect the risk of all cardiovascular events or stroke recurrence.
Table II. Observed Odds Ratios and Odds Ratios Predicted by the Difference in Systolic Pressure in Meta-Regression in Three Actively-Controlled Trials
OR=odds ratio; CV=cardiovascular; MI=myocardial infarction *OR reported in the published articles. Values smaller than unity indicate benefit of experimental treatment compared with reference group; †mean OR (95% confidence interval) predicted by the meta-regression lines (see Figures 1 and 2); ‡difference between predicted minus observed OR (95% confidence interval) expressed in percent of the predicted OR. Positive and negative values indicate benefit and risk associated with experimental treatment beyond blood pressure reduction, respectively; §significance of the difference between observed and predicted ORs
Table III. Observed Odds Ratios and Odds Ratios Predicted by the Difference in Systolic Blood Pressure in Meta-Regression in Four Placebo-Controlled Trials
OR=odds ratio; CV=cardiovascular; MI=myocardial infarction *OR reported in the published articles. Values smaller than unity indicate benefit of experimental treatment compared with reference group; †mean OR (95% confidence interval) predicted by the meta-regression lines (see Figures 1 and 2); ‡difference between predicted minus observed OR (95% confidence interval) expressed in percent of the predicted OR. Positive and negative values indicate benefit and risk associated with experimental treatment beyond blood pressure reduction, respectively; §significance of the difference between observed and predicted odds ratios
In this updated meta-regression analysis, three trials (PATS,37 RENAAL,19 and PROGRESS21) were added and the number of patients increased from 136,124 to 149,407. The present results corroborated our previous findings.17 Indeed, in trials in hypertensive patients and in studies including nor-motensive or hypertensive subjects at high cardiovascular risk, blood pressure gradients accounted for most of the differences in cardiovascular out-come. For fatal and nonfatal cardiovascular complications combined, most of the benefit of antihypertensive treatment was achieved by moderate differences in systolic pressure of approximately 15 mm Hg. In older patients with isolated systolic hypertension,1 lowering blood pressure by 10 mm Hg systolic and 4 mm Hg diastolic decreased the risk of stroke, and myocardial infarction by 26% and 23%.1 In patients with predominantly diastolic hypertension, the corresponding benefits produced by a 5–6 mm Hg decline in diastolic pressure were 38% and 16%, respectively.2
The HOPE trial14,56 was a placebo-controlled study and included approximately 90% of patients with previous cardiovascular complications. At baseline and probably also during follow-up, a large proportion of patients in both the ramipril and placebo groups were treated with calcium channel blockers, diuretics, or β blockers. Compared with the placebo group, treatment with ramipril reduced cardiovascular mortality (−27%) and the incidence of stroke (−32%), myocardial infarction (−21%) and congestive heart failure (−23%).14 There was a 3/1 mm Hg difference in blood pressure between the ramipril and placebo groups.14 The hypothesis has been put forward that the endothelial actions of ramipril56 might stabilize atherosclerotic plaques in large arteries.60 ACE inhibitors have shown proven benefits in patients with heart failure61 or dysfunction of the left ventricle.62 However, our results suggest that blood pressure may account for most, if not all, of the benefit observed in the ramipril group. In addition, the results of other large trials of ACE inhibitors in hypertensive patients,8,10 did not suggest a benefit of ACE inhibitors, which would not be explained by blood pressure.
Active antihypertensive treatment in the placebo-controlled PROGRESS study21 was also based on an ACE inhibitor (perindopril). This double-blind trial demonstrated that among patients with a history of stroke or transient ischemic attack, blood pressure-lowering treatment with perindopril alone or in combination with indapamide reduced the risk of recurrent stroke by 28% and that of stroke, heart attack, or death from cardiovascular disease by 26%. However, a prespecified subgroup analysis revealed striking heterogeneity of treatment effect sizes for the risks of all cardiovascular events and stroke between participants who received perindopril plus indapamide and those who received perindopril alone. Combination therapy reduced blood pressure by 12/5 mm Hg, and decreased stroke and all cardiovascular events by 43% and 40%, respectively. Treatment with perindopril alone lowered blood pressure by 5/3 mm Hg, but did not affect stroke recurrence or the incidence of all cardiovascular events. The relative risk reductions observed in PROGRESS for the perindopril-only group deviated significantly from those predicted by the meta-regression, but those for combination therapy did not.
In the ALLHAT trial,12 the patients randomized to doxazosin experienced higher rates of stroke and congestive heart failure than the patients given chlorthalidone. The ALLHAT investigators suggested that the observed blood pressure differences were sufficient to explain the higher incidence of stroke on doxazosin, but only a 10%–20% increase in the occurrence of heart failure, not a doubling of the rate.12 However, for cardiovascular mortality, stroke, and myocardial infarction as well as all cardiovascular events including congestive heart failure, we did not find significant differences between the observed odds ratios and those expected on the basis of the blood pressure difference between the two treatment groups.
Three trials on angiotensin II receptor antagonists consistently demonstrated that this class of agents reduced the incidence of renal dysfunction in patients with diabetic nephropathy.18–20 However, active treatment also reduced blood pressure, especially systolic pressure by 2–4 mm Hg. The RENAAL trial19 reported the incidence of all cardiovascular events and myocardial infarction and hence was included in our meta-regression. The observed insignificant reduction in cardiovascular outcome in the RENAAL trial was not different from that predicted by the modest blood pressure difference between losartan and placebo.
Our overview should be interpreted within the context of its limitations. As in all meta-analyses that start from published summary statistics, we achieved less standardization than is attainable in quantitative overviews based on individual patient data. We did not correct for regression dilution bias.1,63 However, clinicians cannot leave hypertensive patients untreated for prolonged periods to determine their usual1,63 blood pressure. In fact, the current guidelines for the treatment of hypertension64,65 find their justification in trials done with conventional blood pressure. Our analysis does not indicate to what extent blood pressure should be lowered. The latter issue remains unsettled, because the published studies did not account for systolic pressure or pulse pressure3 or had blood pressure targets that were too high.4 Finally, as a meta-regression analysis, our study tests whether and to what extent outcome results of trials depend on blood pressure differences, but does not specifically deal with superiority or inferiority of one drug to another.66,67
In the trials in hypertensive patients or in normotensive or hypertensive patients at high cardiovascular risk, blood pressure gradients largely accounted for most—if not all—of the differences in cardiovascular outcome. The hypothesis that in patients with uncomplicated hypertension new drug classes, such as ACE inhibitors, angiotensin II receptor antago-nists, or α blockers, might influence cardiovascular outcome over and beyond their blood pressure-low-ering effects remains unproved.
Acknowledgments: Dr. Wang was supported by the bilateral scientific and technical collaboration between the People's Republic of China and Flanders (contract number BIL98/15).