Meta-Analysis of Dose-Response Relationships for Hydrochlorothiazide, Chlorthalidone, and Bendroflumethiazide on Blood Pressure, Serum Potassium, and Urate

Abstract

Thiazide and thiazide-like diuretics are widely used in the management of hypertension, but recently the equivalence of hydrochlorothiazide and chlorthalidone for blood pressure (BP) lowering and prevention of cardiovascular disease has been questioned. We performed a meta-analysis to characterize the dose-response relationships for 3 commonly prescribed thiazide diuretics, hydrochlorothiazide, chlorthalidone, and bendroflumethiazide, on BP, serum potassium, and urate. Randomized, double-blind, parallel placebo-controlled trials meeting the following criteria, ≥2 different monotherapy dose arms, follow-up duration ≥4 weeks, and baseline washout of medication ≥2 weeks, were identified using Embase (1980–2010 week 50), Medline (1950–2010 November week 3), metaRegister of Controlled Trials, and Cochrane Central. A total of 26 trials examined hydrochlorothiazide, 3 examined chlorthalidone, and 1 examined bendroflumethiazide. Studies included a total of 4683 subjects in >53 comparison arms. Meta-regression of the effect of thiazides on systolic BP showed a log-linear relationship with a potency series: bendroflumethiazide>chlorthalidone>hydrochlorothiazide. The estimated dose of each drug predicted to reduce systolic BP by 10 mm Hg was 1.4, 8.6, and 26.4 mg, respectively, and there was no evidence of a difference in maximum reduction of systolic BP by high doses of different thiazides. Potency series for diastolic BP, serum potassium, and urate were similar to those seen for systolic BP. Hydrochlorothiazide, chlorthalidone, and bendroflumethiazide have markedly different potency. This may account for differences in the antihypertensive effect between hydrochlorothiazide and chlorthalidone using standard dose ranges.

Introduction

Benzothiadiazines and thiazide-like diuretics (thiazides) are extensively used in the management of hypertension.¹ Recently the assumption of comparable efficacy of different thiazides, namely hydrochlorothiazide and chlorthalidone, with regard to hypertension and cardiovascular disease prevention has been questioned,^2,3 and a retrospective analysis of the Multiple Risk Factor Intervention Trial data reported that chlorthalidone reduced cardiovascular event rates more than hydrochlorothiazide.⁴ It has even been suggested that it might be inappropriate to consider hydrochlorothiazide and chlorthalidone as belonging to the same class of antihypertensive agents.⁵ However, comparisons of different drugs need to take account of both potency (ie, the location of the dose-response relationship with respect to concentration) and maximal efficacy (the maximum effect achievable by the drug). In addition, when comparing agents, their impacts on important adverse effects need to be taken into account. Thiazides and thiazide-like diuretics share an affinity for the NaCl cotransporter in the distal tubule,⁶ and inhibition of this transporter accounts for the natriuretic effects of these agents.⁷ The antihypertensive mechanism of action of thiazide and thiazide-like diuretics after acute administration of high doses is attributable to natriuresis and a reduction in plasma volume, but in the long-term ability they lower blood pressure (BP) through a reduction in peripheral resistance by mechanisms that remain poorly understood.⁷ Generally, the doses of thiazide required to induce acute 24 hour natriuresis⁸ are higher than those required for BP lowering,⁹ with higher doses associated with more frequent adverse effects, such as diabetes mellitus, hypokalemia, hyponatremia, and hyperuricemia. There is also some evidence that some thiazide-related adverse effects may compromise the benefits of thiazides at higher dose levels.^4,10,11 We, therefore, undertook a systematic review to examine the placebo-adjusted dose-response effect of thiazide and thiazide-like diuretic monotherapy on BP and relevant biochemistry.

Methods

Trial Inclusion Criteria

Included trials met the following criteria: (1) double-blind study of thiazide or thiazide-like therapy in people with hypertension (BP ≥160 mm Hg systolic or ≥90 mm Hg diastolic); (2) parallel design; (3) randomized allocation to ≥2 monotherapy thiazide fixed-dose arms or placebo; (4) duration of follow-up ≥4 weeks; (5) baseline washout of medication ≥2 weeks; (6) a placebo arm without other antihypertensive drugs; and (7) measurements of ≥1 of the following, systolic BP, diastolic BP, serum potassium, urate, sodium, cholesterol, glucose, plasma renin activity, or urinary electrolytes.

There were insufficient trials including thiazides or thiazide-like diuretics for meta-analysis with the exception of hydrochlorothiazide, chlorthalidone, and bendroflumethiazide, so only trials including ≥1 of these agents were included. Crossover trials were excluded because of the possibility of carryover effects. We also excluded trials where subjects were predefined as responders or nonresponders before the trial. If used, titration intervals in step-up protocols had to last ≥4 weeks and had to apply to all of the participants, regardless of BP response. Studies using potassium supplementation were included, but step-down and drug withdrawal protocols were ineligible. Trials were also ineligible if participants were <18 years old or had cirrhosis with ascites, nephrotic syndrome, renal insufficiency, documented serum creatinine level >1.5 times normal, cardiac failure, secondary hypertension, idiopathic hypercalciuria, hyperparathyroidism, hypoparathyroidism, or pseudohypoparathyroidism. Where resting BP measurements were available for >1 time during a 24-hour period, the trough measurement, defined as 22 to 26 hours after dose, was used. When BP was recorded in multiple positions, sitting BP was used, unless variance data were only given for another position, in which case that position was used.

Search Strategy

To identify eligible studies, searches of the databases Ovid Medline (1950–2010/November) and Embase (1980–2010/November), the metaRegister of Controlled Trials, and the Cochrane Central Register of Controlled Trials were carried out. In addition, references from relevant meta-analyses and reviews were searched, and authors of unpublished studies were contacted. Data from 1 as yet unpublished trial (clinicaltrials.gov identifier: NCT00153049) was also included.

Data Extraction and Statistical Analysis

Data were extracted using a standardized data extraction form and recorded in Microsoft Excel before transfer to Stata IC 11.2 (Stata Corp) for analysis. Trial quality was assessed on the basis of quality of reporting of methods, including double blinding, randomization, and dropouts; use of intention-to-treat analysis; and completeness of data. The mean and SE of the placebo-corrected difference between baseline and end of the trial for a given measure were calculated. If the SE was not reported, it was calculated from the SE of the change for each group or the relevant SDs and sample sizes. Further details are given in the online-only Data Supplement. Because each study had multiple comparisons with a common placebo group, to use the reported number of participants in the placebo in each comparison would be to “double count” participants in the placebo group. To overcome this unit-of-analysis error, while still allowing investigation of heterogeneity across intervention arms, we subdivided the number in the placebo group evenly into as many groups as there were active arms.¹²

Analysis was performed using random effects models and the I² statistic, an estimate of the proportion of the total observed variation that is attributed to between-study variance rather than within-study variance, calculated. Random effects metaregression models were then used to relate natural log dose of each active group to the effect size for each outcome measure. These data were used to calculate the equieffective doses of thiazide or thiazide drug for each outcome based on the regression equation. P values corrected for multiple testing were calculated using permutation tests,¹² and I² was calculated with and without natural log dose added as a covariate. Post hoc sensitivity analyses were also performed to assess the effect of excluding lower-quality trials.

Results

Characteristics of Included Trials

Of the 6477 abstracts screened, 26 trials met inclusion criteria (Figure 1). The studies included a total of 4683 subjects in >53 comparison arms. For further details of the included trials, please see Table S2 (in the online-only Data Supplement). Trials were broadly similar with regards to participant age (median, 53 years) and BP (mean, 153/100 mm Hg) at the end of the placebo run-in period. All of the trials bar 1 prescribed a single fixed dose per arm; the exception¹³ used forced escalation of therapy. Potassium supplements were mandated in 1 trial,¹⁴ but the dose of potassium varied depending on the thiazide dose.

Figure 1.

Preferred Reporting Items for Systematic Reviews and Meta-Analyses diagram for studies.

Dose-Response Relationships for Thiazides on BP, Serum Potassium, and Urate

Figure 2 shows dose-stratified forest plots relating hydrochlorothiazide, chlorthalidone, and bendroflumethiazide for placebo-corrected difference in systolic BP. Doses of hydrochlorothiazide <6.25 mg had no statistically significant effect on systolic BP, whereas there were insufficient data available to delineate the lower part of the dose-response relationship for chlorthalidone and bendroflumethiazide. For forest plots for the other measurements, see Figures S1 through S3. Figure 3A shows the dose-response relationship for placebo-corrected differences in systolic BP for each of the 3 drugs. The potency series was bendroflumethiazide>chlorthalidone>hydrochlorothiazide. For systolic BP, the estimated dose of each drug predicted to reduce systolic BP by 10 mm Hg was 1.4, 8.6, and 26.4 mg, respectively. A meta-analysis of systolic BP reductions in response to high doses of hydrochlorothiazide (>25 mg), chlorthalidone (>25 mg), and bendroflumethiazide (>5 mg) showed no heterogeneity in the reduction of systolic BP by drug class for what was assumed to be a near maximal response (pooled effect size [95% CI] for hydrochlorothiazide, –10.1 [–13.4 to –6.8]; chlorthalidone, –15.5 [–21.0 to –9.9]; bendroflumethiazide, –14.2 [–22.0 to –6.4]; I²=3.7; P=0.7 for heterogeneity). The dose-response relationship for diastolic BP is shown in Figure 3B and shows a similar potency series to systolic BP. The estimated doses of chlorthalidone and hydrochlorothiazide predicted to reduce diastolic BP by 4 mm Hg were 14.0 mg and 20.8 mg, respectively. It was not possible to estimate an equivalent dose of bendroflumethiazide, because reductions in diastolic BP exceeded 4 mm Hg at all of the dose levels.

Figure 2.

Dose-stratified forest plot for thiazides versus systolic blood pressure. A, Hydrochlorothiazide; B, chlorthalidone; C, bendroflumethiazide.

Figure 3.

Dose-response relationships for thiazides vs (A) systolic blood pressure; (B) diastolic blood pressure; (C) serum potassium; and (D) serum urate. The lines are the regression lines of log dose vs response for each drug and circle sizes depend on the precision of each estimate (the inverse of its within-study variance). Red line, hydrochlorothiazide; blue line, chlorthalidone; green line, bendroflumethiazide.

A meta-analysis of diastolic BP reductions in response to high doses of bendroflumethiazide (>5 mg), chlorthalidone (>25 mg), and hydrochlorothiazide (>25 mg) showed no heterogeneity in the reduction of diastolic BP by drug class for what was assumed to be a near maximal response (pooled effect size [95% CI] for bendroflumethiazide, –7.3 [–11.0 to –3.6]; chlorthalidone, –4.6 [–6.2 to –3.0]; hydrochlorothiazide, –3.7 [–4.1 to –3.2]; I²=0%; P=0.9).

The dose-response relationship for serum potassium is shown in Figure 3C. Again, the potency series was bendroflumethiazide>chlorthalidone>hydrochlorothiazide. The estimated doses of bendroflumethiazide, chlorthalidone, and hydrochlorothiazide predicted to reduce serum potassium by 0.4 mmol/L were 4.2, 11.9, and 40.5 mg, respectively.

The dose-response relationship for urate is shown in Figure 3D. The potency series was similar, but there appeared to be less difference between the dose-response relationships than for other outcomes. The dose of drug predicted to increase urate by 36 μmol/L was 2.1, 8.9, and 12.3 mg for bendroflumethiazide, chlorthalidone, and hydrochlorothiazide, respectively. Sensitivity analyses excluding lower-quality studies found no marked differences for any outcomes compared with all of the studies (data not shown).

Discussion

A dose-stratified meta-analysis and metaregression has been used to characterize the dose-response relationships for 3 commonly prescribed thiazide/thiazide-like diuretics, hydrochlorothiazide, chlorthalidone, and bendroflumethiazide. The potency of the 3 diuretics differed quite markedly, with bendroflumethiazide the most potent and hydrochlorothiazide the least, both for BP and for biochemical outcomes, such as serum potassium and urate. The similarity of the potency series is consistent with a common mechanism of action for all of these agents.² There was no convincing evidence that “near-maximum” reductions in systolic BP differed between individual thiazides, particularly taking into account the differences in potency of the drugs. We show that dose-related responses for BP, serum K, and urate at the summary level are well described by a log-linear relationship over the dose ranges studied, although there was a suggestion of departure from linearity at the highest dose (>50 mg) of hydrochlorothiazide for systolic and diastolic BPs.

The findings of this study should be set in context of previous meta-analyses. Law et al, in 2003,¹⁵ performed a meta-analysis including data on 7 thiazides in trials lasting ≥2 weeks. Thiazides were grouped on the basis of a standard daily dose (defined as hydrochlorothiazide 25.0 mg, chlorthalidone 25.0 mg, and bendroflumethiazide 2.5 mg), and the standard dose of thiazide was found to reduce systolic BP by 8.8 (8.3–9.4)/4.4 (4.0–4.8) mm Hg; to reduce potassium by 0.38 mmol/L; and to increase urate by 48 μmol/L. These data are compatible with our findings, although our analysis shows that the “standard” doses of the 3 thiazides do not represent equipotent doses, and we believe that pooling data in this way is likely to be misleading with regard to efficacy. Wright et al¹⁵ analyzed 14 randomized, controlled trials of thiazide versus no treatment lasting ≥1 year. In each trial, different proportions of patients received supplementary nonthiazide drugs, and doses were mostly titrated on the basis of diastolic BP-lowering efficacy. “High-dose” therapy, defined as starting doses of hydrochlorothiazide ≥50 mg, chlorthalidone ≥50 mg, and bendroflumethiazide ≥5 mg (with weighted mean “equivalent” to hydrochlorothiazide 90 mg) reduced systolic/diastolic BP by 14.9/7.3 mm Hg; low-dose therapy (“equivalent” to hydrochlorothiazide 25 mg) reduced BP by 15.6/6.0 mm Hg. The effects of high-dose thiazide seen by Wright et al¹⁶ are, therefore, similar to the “near-maximum” effect of thiazides seen in our study. Baguet et al¹⁶ performed a systematic review based on trials lasting 8 to 12 weeks, without discriminating between parallel and crossover designs. Meta-analysis of 4 trials of hydrochlorothiazide ≤25 mg/d (none fitting this article's inclusion criteria, because we required dose-response information) gave a reduction in systolic BP of −18.9/−11.0 mm Hg. This report is similar to our observations, but comparisons are complicated by inclusion of studies using combinations of drugs and/or multiple doses by Baguet et al.¹⁷ More recently, Ernst et al¹⁷ conducted a meta-analysis to compare hydrochlorothiazide monotherapy with chlorthalidone monotherapy and concluded that hydrochlorothiazide and chlorthalidone did not result in equivalent reductions in systolic BP over the dose range (12.5–25.0 mg). This study differs from ours in that Ernst et al¹⁸ grouped drug doses into 3 groups (12.5–25.0, 25.0–37.5, and 37.5–50.0 mg); did not adjust for placebo effect sizes; and used fixed-effects meta-analysis and equivalence testing within the 12.5- to 25.0-mg dose range. Nevertheless, our data are in agreement with their conclusion that hydrochlorothiazide lowers systolic BP less than chlorthalidone in the dose range (12.5–25.0 mg); however, we show that this is explained by differences in potency rather than efficacy.

The therapeutic dose ranges of hydrochlorothiazide, chlorthalidone, and bendroflumethiazide monotherapy are stated to be 12.0 to 50.0, 12.0 to 25.0, and 2.5 mg, respectively.^19,20 From Figure 3A and 3B, it can be seen that these dose ranges are not equieffective for BP reduction; for systolic BP the equivalent dose of hydrochlorothiazide 25.0 mg is ≈8.0 mg of chlorthalidone and 1.5 mg of bendroflumethiazide. A common reference source²¹ states that hydrochlorothiazide and chlorthalidone are equipotent, although in vitro studies suggest that the affinity of chlorthalidone for the NaCl cotransporter is ≈2-fold greater than hydrochlorothiazide and that bendroflumethiazide has a 20-fold greater affinity for the NaCl cotransporter than hydrochlorothiazide.⁶ In addition, the half-life of chlorthalidone is considerably longer than hydrochlorothiazide.²² Other studies (reviewed in Reference 21⇓) suggest that chlorthalidone is ≈1.5-fold more potent than hydrochlorothiazide; however, these estimates are based on comparisons of high doses of thiazide close to the therapeutic ceiling and may underestimate the relative potency. This meta-analysis allowed comparisons to be made over the log-linear portion of the dose-response relationship and indicates that the chlorthalidone is ≈3-fold more potent than hydrochlorothiazide and that bendroflumethiazide is ≈18-fold more potent than hydrochlorothiazide. These differences in potency allied to the differences in pharmacokinetics²² are likely to account to a large extent for the reported differences in effectiveness in some studies comparing hydrochlorothiazide with chlorthalidone^4,23 or other antihypertensive classes.³

This meta-analysis has a number of limitations. Our goal was to quantify dose relationships for individual thiazide or thiazide-like agents in the dose range used in the treatment of hypertension, so stricter criteria were imposed than in some previous meta-analyses. It should be noted that we did not examine dose relationships for diuretic effects, and this is known to differ from the dose relationship for antihypertensive effects.^8,9 Also, similar results cannot be assumed when these different thiazides are used in combinations with other drugs. For chlorthalidone and bendroflumethiazide, there was a dearth of high-quality trials; for example, there were only 3 trials of chlorthalidone; furthermore, there was scant data examining the lower parts of the dose-response relations, and our conclusions regarding the effect of lower doses of these agents must be considered tentative. However, the paucity of literature on lower doses of chlorthalidone and bendroflumethiazide as monotherapy suggest that relaxation of inclusion criteria would not have provided more information and would have compromised quality. Many of the studies were conducted >25 years ago, and differences in inclusion and exclusion criteria, BP measurement techniques, and drug formulation will contribute to heterogeneity between studies. We included studies with a treatment duration >4 weeks. The BP response to diuretic monotherapy may take 12 to 14 weeks to achieve full effect,⁹ but most of the response is present by 4 weeks, and it seems unlikely that this will have resulted in a major underestimation of effect size. We also imputed SEs in some studies where they were not available from the published data, but sensitivity analyses excluding these studies produced very similar results. We attempted to overcome bias related to unit-of-analysis error by dividing the sample size of the placebo group by the number of active arms in each trial. Although this approach reduces the sample size for each comparison appropriately, it does not ensure that each comparison within the same trial is completely independent.

In summary, we performed a dose-stratified meta-analysis for the 3 most commonly used antihypertensive thiazide/thiazide-like drugs. We observed considerable differences in potency among hydrochlorothiazide, chlorthalidone, and bendroflumethiazide, and we suggest that differences in potency rather than differences in efficacy probably explain most of the difference in BP lowering by a particular thiazide or thiazide-like drug at a given dose.

Perspectives.

Thiazide and thiazide diuretics are widely used in the management of hypertension. Recently, questions have been raised regarding the equivalence of hydrochlorothiazide with chlorthalidone for BP lowering and prevention of cardiovascular disease. A meta-analysis of the dose-response relationships for 3 commonly prescribed thiazide/thiazide-like diuretics, hydrochlorothiazide, chlorthalidone, and bendroflumethiazide, suggests that differences in thiazide effect are largely explained by differences in potency and that 25 mg of hydrochlorothiazide should not be regarded as equivalent to 25 mg of chlorthalidone.