Systematic review of antihypertensive therapies

One of the main components of managing a patient with hypertension is deciding which drug to prescribe for first-line therapy. This decision should be made primarily on the basis of the best available evidence of effectiveness - that is, the drug's ability to prevent adverse health outcomes that are important to the patient. The available evidence has not been organized in a way that helps clinicians make such decisions. There have been a number of systematic reviews of the effectiveness of antihypertensive therapy, but most have focused on overall effectiveness [1, 2] or effectiveness in special groups such as elderly patients. [3–7]

When all drug therapies are included in one review, there is an underlying assumption that the benefits of lowering blood pressure are independent of the mechanism by which this is achieved. This assumption has not been proven, and it is likely that the mechanism by which a drug lowers blood pressure will have effects that are independent of the blood-pressure-lowering effect. In addition, all anti-hypertensive drugs have actions other than lowering blood pressure. These other actions, both known and unknown, could enhance or negate the effectiveness associated with the decrease in blood pressure.

Only 2 reviews have attempted to distinguish between the effectiveness of antihypertensive therapies used as first-line agents. [1, 8] Collins and associates [1] reviewed drug-drug comparison trials, but only 2 of the 3 trials included were appropriate to make that comparison. The review by Psaty and colleagues [8] had a number of deficiencies: drug-drug comparison trials were not included, trials were misclassified to drug therapy groups, and data on the effects of lowering blood pressure were not included.

The objectives for our systematic review were (a) to combine the evidence of effectiveness and efficacy from drug-drug comparison trials of first-line therapies; (b) to combine the evidence of effectiveness and efficacy from trials in which classes of drugs used as first-line therapy were compared with a placebo or an untreated control group; (c) to determine whether the dose of thiazide used affected outcomes; (d) to calculate from total cardiovascular events the best estimate of the overall absolute risk reduction and the number needed to treat for each drug class; and (e) to translate the evidence into clinical implications for first-line drug choices.

Methods

The following sources were searched: MEDLINE (1966-1997), the Cochrane Library (1998 CD-ROM, issue 2) and references from previous meta-analyses published from 1980 to 1997. In the case of incomplete reports, MEDLINE was searched for connected papers to retrieve missing information. Our search retrieved 9 trials not included in the review by Psaty and colleagues [8] and 6 trials that have not been included in previous reviews.

We included trials in which participants had hypertension defined as a systolic blood pressure of at least 160 mm Hg or a diastolic blood pressure of at least 90 mm Hg. We assumed that the effect on outcomes would be independent of whether the hypertension was defined in terms of systolic or diastolic pressure. All antihypertensive trials were included regardless of participants' comorbidities or baseline risk. We assumed that age and comorbidities would not affect the relative risk reduction associated with drug treatment. Trials using antihypertensive drugs for other indications (e.g., congestive heart failure) were excluded.

The drug classes of interest were thiazides, beta-adrenergic blockers, angiotensin-converting-enzyme (ACE) inhibitors, calcium-channel blockers, alpha-adrenergic blockers and angiotensin II receptor antagonists. Supplemental drugs could be used as combination therapy or as stepped therapy. We assumed that these supplemental drugs would not systematically interact to affect the occurrences of the end points studied.

The following were the outcome measures of interest: death, including all-cause mortality; stroke, fatal and nonfatal; coronary artery disease, including fatal and nonfatal myocardial infarction and sudden or rapid cardiac death; and total cardiovascular events, including stroke and coronary artery disease plus congestive heart failure and other significant vascular events such as ruptured aneurysm (not included were surgical or other procedures, angina and transient ischemic attacks).

When primary trials reported outcomes that did not explicitly fit the above definitions, we made a decision based on maximizing the inclusion of data and maintaining concordance with how the data were classified in previous reviews.

We included trials that met the following criteria: random allocation of participants; comparison between a first-line drug therapy and another first-line drug therapy or no treatment (including placebo); recording of group baseline characteristics; clearly defined mortality and morbidity end points; at least 1 year of follow-up; clearly defined first-line treatment in 1 of 6 defined categories; and the majority (more than 70%) of patients in the treatment group taking the drug class of interest after 1 year. We included trials in which patients received concurrent therapy with drugs not in the classes of interest as well as trials in which supplemental drugs from other drug classes of interest were given as long as they were not taken by more than 50% of patients.

Because of the number of trials in which thiazides were compared with no treatment or placebo, we were able to evaluate whether the dose of the thiazide affected the outcomes. We divided the thiazide trials into high or low dose based on the starting dose used in the trial. We selected hydrochlorothiazide at a starting dose of 50 mg and above to define the high-dose therapy. The doses of the other thiazides equivalent to that dose were defined based on diuretic effect in human studies as much as possible (Appendix 1 Table 6). We assumed that each thiazide would have a similar dose-response curve and that the usual range of prescription doses would represent a similar range on the dose-response curve. The average dose used in the predefined high- and low-dose groups was calculated as a weighted average from the trials in which the average dose was reported or could be estimated.

Data abstraction was done by 2 independent reviewers and compared when possible to data from previously published meta-analyses. The data represent only one morbid event per patient, because most studies usually recorded only the first morbid event. However, in some trials it was impossible to be certain, and so the category of total cardiovascular events could include a small proportion of patients counted more than once. This would occur similarly in both the treatment and the comparison groups.

Quantitative analyses of outcomes were based on intention-to-treat results. We used the relative risk (RR) ratio and a fixed-effects model to combine outcomes across trials. The weighting factor for each trial was the inverse of the within-study variance plus a between-study variance component (DerSimonian-Laird type random effects estimators [9]). The risk difference (RD), the absolute risk reduction (ARR = RD x 100) and the number needed to treat (NNT = 1/RD) were calculated only for the category of total cardiovascular events because this single value has the largest ARR and most meaningful NNT from a clinical standpoint.

Data for blood pressure reduction was combined using a weighted mean difference method, whereby the trials are weighted according to the number of subjects in the trial and the within-study variance. Some of the trials did not report a within-study variance for blood pressure decrease; in these studies an imputed standard deviation (SD) was assigned based on the highest SD for that group. In this way the weighting was based primarily on the numbers of subjects randomized to each group. Because of this limitation, we report 99% confidence intervals (CIs) for this data instead of 95% CIs, which were the standard for the other data.

We performed sensitivity analyses to test for the robustness of the data and to determine whether the decision to include certain studies had a significant effect on the final estimates of effect size.

Results

We retrieved 38 studies of first-line therapy for hypertension published between 1966 and 1997. We excluded 15 reports [10–32] because they did not meet our inclusion criteria (Appendix 2 Table 7). Table 1 summarizes the characteristics of the remaining 23 trials (representing 50 853 patients). [33–58] The first group of trials represents those that compared one first-line drug therapy with another. Two of the trials in this group [36, 37] also included a placebo group, so we included them in our analysis of drug-no treatment comparison trials as well.

Interpretation

Randomized trials comparing one drug with another offer the best opportunity to detect differences between 2 classes of drugs. In the trials comparing thiazides with beta-blockers [34–38] the number of patients who dropped out because of adverse effects was significantly lower in the thiazide group; the incidence of total cardiovascular events was lower in the thiazide group, although the difference was of borderline significance. This comparison also revealed a significant benefit in favour of thiazides for decreasing systolic blood pressure. These trials included 2 in which low-dose thiazide therapy was used [34, 36] and 3 in which high-dose thiazide therapy was used. [35, 37, 38]

One of the drug-drug comparisons we excluded deserves mention here. The Metoprolol Atherosclerosis Prevention in Hypertensives (MAPHY) study purported to show a benefit of metoprolol over thiazides. [20, 21] It started as part of the Heart Attack Primary Prevention in Hypertension (HAPPHY) trial, [35] and at the end of that trial it split off, and half the patients, those randomly assigned to receive metoprolol or thiazides, were followed for a further 14 months. The MAPHY trial has been criticized, [59, 60] and since these patients were already included in the HAPPHY trial data, it is clearly inappropriate to include them twice. The controversy over the MAPHY trial calls attention to the unexplained higher mortality in the atenolol group than in the thiazide group, which was seen in one of the Medical Research Council (MRC) trials [36] and in our combined analysis of the mortality results from the HAPPHY trial and this MRC trial. The relatively poor outcomes seen with the beta-blockers in our review are predominantly explained by the trials using atenolol.

In the only previous meta-analysis that included drug-drug comparison trials, Collins and associates [1] did not include the MAPHY trial either; however, they incorrectly included the International Prospective Primary Prevention Study in Hypertension (IPPPSH) trial. [19] In the IPPPSH trial thiazides were prescribed to 67% of patients in the beta-blocker group and to 82% of those in the non-beta-blocker group. The results of this trial cannot, therefore, be used to compare the effectiveness of beta-blockers and thiazides.

In the trials that compared a first-line therapy with no treatment or placebo, it was more difficult to make clear inferences about differences in effectiveness of the different drug classes. Only a few trials evaluated beta-blockers and calcium-channel blockers in this regard, and no trials evaluated ACE inhibitors or alpha-adrenergic blockers. It is important in this type of analysis not to include trials in which there was significant contamination by drugs from other classes. In our opinion, Psaty and colleagues [8] incorrectly included the trial by Coope and Warrender [12] and the STOP-Hypertension trial [27] in their beta-blocker group. In the trial by Coope and Warrender 67% of patients in the active treatment group received bendrofluazide and 70% atenolol; in the STOP-Hypertension trial more than 70% in the active treatment group received thiazides and more than 70% received beta-blockers.

Our findings demonstrating a significant decrease in the incidence of coronary artery disease with high- and low-dose thiazide therapy are similar to those of Psaty and colleagues. [8] There were only minor differences in the trials reviewed by us and them. We classified the trial by Kuramoto and associates [47] as a low-dose thiazide trial, and they classified it as a high-dose trial. It fit into our low-dose category based on the starting dose of 1 mg of trichlormethiazide, although the article did not mention how many patients were maintained on the starting dose. Moving the trial from the low- to high-dose group did not change our findings. Unlike Psaty and colleagues, we excluded the Hypertension Detection and Follow-up Program trial [14–18] because it did not have an untreated control group and it studied the effect of lifestyle changes in addition to drug therapy, and we included one small trial, by Wolff and Lindeman, [58] which was also included in the meta-analyses by Collins and associates [1] and Gueyffier and colleagues. [2]

A potential limitation of attributing the benefit predominantly to thiazides in our review is the fact that in almost all of the trials supplemental drugs were added to various proportions of patients. We were able to do a sensitivity analysis of the impact of second-line central-nervous-system-active drugs, beta-blockers and potassium-sparing diuretics on the outcomes with thiazides. In all cases removing these trials had little or no effect on the final risk ratio, which demonstrated that the data for thiazides are robust and that the benefits achieved were most likely attributable to the thiazides. (The details of these sensitivity analyses are available upon request from the authors.)

Our systematic review demonstrates the importance of analysing efficacy as well as effectiveness, and of analysing the effect on systolic as well as diastolic blood pressure. [1] Thiazides were consistently better at lowering systolic blood pressure than the other drug classes. This may be somewhat surprising; however, in all trials other than those of therapy for isolated systolic hypertension, [50–52] titration was based on diastolic blood pressure. This explains why the drop in diastolic pressure was similar for the different drug classes. Since systolic pressure is a better indicator of risk than diastolic pressure, [7] this observation could be one reason for better outcome data with thiazides.

The ARR for low-dose thiazide therapy (5.5%, 95% CI 4.1%-7.0% over about 5 years) most likely underestimates the benefits seen in practice for a number of reasons: (a) patients in the trials were at lower risk of cardiovascular events than other patients of the same age; [7] (b) many of the trials did not report all morbidity outcomes; (c) one of the largest benefits in these trials (not included as an outcome) was the decrease in the number of patients withdrawn because of excessively elevated blood pressure (these patients in the control group received appropriate antihypertensive treatment and decreased the difference in outcomes between the groups).

We feel confident in concluding that the benefits seen in the thiazide trials are predominantly the result of the thiazide being used as first-line therapy. The evidence shows that starting with a low dose of thiazide and titrating up only if necessary significantly reduces the risk of coronary artery disease. In contrast, starting with a high dose (50 mg of hydrochlorothiazide or the equivalent) does not. Since the high-dose therapy was no more efficacious than the low-dose therapy in lowering blood pressure, we feel that there is no justification for using high doses or for comparing low- and high-dose thiazide therapies in a trial. The current standard recommendation for the use of thiazides in the management of hypertension is therefore justifiable based on the evidence presented here: start with a low dose (12.5 mg of hydrochlorothiazide or the equivalent), increase the dose only if necessary, and do not exceed a dose of 50 mg of hydrochlorothiazide or the equivalent.

There is growing evidence that the surrogate marker - the lowering of blood pressure - is inadequate in predicting health outcomes with antihypertensive therapy. The difference in the impact of low- and high-dose thiazide therapy on the incidence of coronary artery disease is one example. The results of a number of drug-drug comparison trials [11, 36, 39, 41] also suggest this conclusion.

More trials comparing drug therapy with a placebo or no treatment are unlikely to be forthcoming. Therefore, the data in Table 4 are unlikely to change. Further drug-drug comparison trials are under way. [61–64] Their results, when available, can be added to the data in Table 2. In the mean time, clinicians must choose a first-line drug therapy using the available evidence. At present low-dose thiazide therapy can be prescribed with confidence that the risk of death, coronary artery disease and stroke will be reduced. The same cannot be said for high-dose thiazide therapy or for any of the other classes of antihypertensive drugs.

This work was supported by a grant from the British Columbia Ministry of Health and by the University of British Columbia. The views expressed are entirely those of the authors.

Competing interests: None declared.

This article has been peer reviewed.

Reprint requests to: Dr. James M. Wright, Department of Pharmacology and Therapeutics, 2176 Health Sciences Mall, University of British Columbia, Vancouver BC V6T 1Z3; fax 604 822-0701; [email protected]