Antihypertensive Treatment and Development of Heart Failure in Hypertension: A Bayesian Network Meta-analysis of Studies in Patients With Hypertension and High Cardiovascular Risk

Heart failure (HF) represents the final common pathway of the clinical history of different cardiac diseases, and it is viewed today as one of the major health care problems worldwide.¹ Arterial hypertension represents the most common cause of predisposition to the development of HF, which is also independent of the occurrence of coronary artery disease,^2,3 as highlighted by the Framingham Heart study⁴ and the Rotterdam study.⁵

Despite this evidence, over the past years HF has been often considered a “soft” end point in the main hypertension clinical trials, and greater attention has been given to myocardial infarction and stroke.⁶ A recent analysis from our group⁷ examining the main clinical trials in hypertension published in the past decade has substantially challenged this view by showing that the incidence of HF in hypertension is as frequent as that of stroke, being even more common in older patients, patients with diabetes mellitus, blacks, and patients at very high cardiovascular risk. The importance of HF is documented by the fact that it still represents one of the leading causes of hospitalization, disability, and mortality, and it imposes heavy medical and financial burdens on communities, which are expected to further rise in the next 2 decades.^8,9 Thus, it seems necessary to optimize strategies for preventing HF, especially in the presence of predisposing conditions, such as hypertension.

On the one hand, over the past years, numerous trials have provided evidence regarding the efficacy of different antihypertensive drugs in preventing HF development in patients with hypertension.^10-35 Placebo-controlled studies have consistently shown that all principal antihypertensive strategies reduce the incidence of HF, thus confirming that blood pressure reduction is a fundamental strategy to prevent HF in patients with hypertension. On the other hand, clinical studies based on active treatment comparisons suggest that calcium channel blockers (CCBs) are less effective than renin-angiotensin system (RAS) inhibitors, diuretics, and β-blockers in reducing HF development. No conclusive evidence about the optimal antihypertensive therapy, however, has been provided. This issue has not been resolved by recent hypertension meta-analyses, which are often based on a series of small meta-analyses and on limited comparisons between active treatments. We report herein the results of a network meta-analysis of recent trials conducted in patients with hypertension, or in patients at high cardiovascular risk with a predominant proportion with hypertension.^10-35 We compared antihypertensive strategies in HF prevention. Network meta-analysis is a meta-analytic technique that synthesizes the available evidence from clinical trials evaluating a wide range of drug comparisons in a single meta-analysis, in which direct evidence of different therapy comparisons are combined with indirect evidence that are constructed from studies that have treatments in common.³⁶

We systematically reviewed the medical literature to identify the clinical trials to be used for the analysis. These studies had to fulfill predefined, specific qualitative criteria and provide information on demography and the definition of HF diagnosis. Thus, the selected studies had to adhere to the following criteria: they had to be randomized, controlled design for clinical studies published in peer-reviewed journals indexed in medical databases, available by 1997; they had to include patients with hypertension or a population characterized as having a “high” cardiovascular risk profile and a predominance of patients with hypertension (>65%); the sample size had to have at least 200 patients; and information on the absolute incidence of HF and other major cardiovascular events had to be included. The computerized search was performed using the PubMed and EMBASE databases through December 2009. The specific keywords were “antihypertensive drugs” and “cardiovascular risk,” and the “limits” of the search were a date of publication between 1997 and 2009 and the selection of randomized controlled trials. We also checked in the references of a recently published meta-analysis for the presence of other trials matching our inclusion criteria.³⁷ Two investigators (S.S. and F.P.) independently searched the trials and verified the specific recruitment criteria. We decided to consider only trials published in the past decade because the clinical feature of included patients and the different antihypertensive strategies in these studies reflect more properly the present hypertension clinical practice. We arbitrarily chose 1997 as the first date for study selection because in the previous year a landmark analysis evaluating the prevention of HF in hypertension trials was published.³⁸ From a total of 824 screened clinical studies, 34 were considered to be eligible for the present analysis^10-35,39-46 (Figure 1). Seven studies were excluded because they did not report specific data on HF,^39-45 and 1 was excluded because it referred to patients with high-normal blood pressure levels.⁴⁶ Although the prevalence of patients with hypertension in the study population of the Heart Outcomes Prevention Evaluation Study (HOPE) study²¹ was lower than 65%, we considered this trial in the analysis anyway since the cutoff used for hypertension diagnosis was a systolic blood pressure higher than 160 mm Hg.⁴⁷ All of the included trials met at least 2 criteria defining the quality of clinical trials.⁴⁸ A list of the selected studies is provided in Table 1 and Table 2.

All data from each completed primary publication were tabulated into a computerized spreadsheet (Microsoft Excel; Microsoft Corp, Redmond, Washington). First, we performed a series of traditional meta-analyses of studies that directly compared different antihypertensive drugs as first-line agents.

Then we performed a Bayesian network meta-analysis to compare different antihypertensive drug-based therapies (angiotensin-converting enzyme [ACE] inhibitors, angiotensin II receptor blockers [ARBs], diuretics, CCBs, β-blockers, conventional treatment, and α-blockers) to placebo and one to each other.

We also performed additional network meta-analyses in 5 subgroups of studies: (1) a subgroup including only studies that did not enroll patients with a history of HF (186 559 patients)^{15,17,29,31-33}; (2 and 3) subgroups including trials that enrolled a greater proportion of males (153 447 patients)^{11-13,15,20-22,25,26,29,30,32,34,35} and a greater proportion of females (69 866 patients)^{10,14,16-19,23,24,27,28,31,33}; and (4 and 5) subgroups including trials with a mean age of the enrolled population younger than 67 years (151 832 patients)^{11-15,19,21-25,29,30,32,34,35} or 67 or older years (71 482 patients)^{10,16-18,20,26-28,31,33}. This value represents the median of the mean age of the populations of the trials included in our investigation.

Statistical analysis was performed by using SPSS software (version 16.0; SPSS Inc, Chicago, Illinois), SAS software (version 8.2; SAS Institute Inc, Cary, North Carolina), and WinBUGS packages (MRC Biostatistics Unit, Cambridge, England). Estimates of the effects of single direct comparisons between 2 strategies in HF prevention were calculated by using the χ² test. Meta-analytical overall estimates of the effect of direct comparisons were calculated according to the meta-analytical technique.^49,50 The assumption of homogeneity of treatment effect between different individual studies and subgroups of studies was tested using the χ² test for homogeneity. When appropriate, publication bias was tested as previously described.^51,52

Network meta-analysis was performed using the Bayesian hierarchical model proposed by Lu and Ades.³⁶The advantages of a Bayesian meta-analytic approach are represented by the fact that direct probability statements can be made, all evidence regarding a specific problem can be taken into account, and predictive statements can be easily made. The disadvantages are related to the use of prior beliefs that may undermine objectivity; to elicitation of priors, which is nontrivial with few guidelines; and to its computational complexity and the long period of time needed to perform it.

This model is the k-comparison version of the SST model of Smith et al⁵³ proposed by them to analyze the comparisons between treatments through adequate parameterizations. The estimates were obtained by using the Markow Chains Monte Carlo Method (MCMC). Specifically, 2 chains were generated with 5000 initial iterations (burn in), and 100 000 iterations were used for the estimations. The number of initial iterations (burn in) was determined on the basis of Gelman-Rubin approach. The accuracy of the posterior estimates was found by calculating the Monte Carlo error for each parameter. As a rule of thumb, the Monte Carlo error for each parameter of interest is less than 5% of the sample standard deviation. For the present analysis, following Lu and Ades,⁵⁴ among 5 models presented by those 2 authors (all incorporating the random multivariate treatment effects), we chose the SST-2-HOM model: random effects baselines with homogeneous treatment variance. We also used noninformative priors that represented complete lack of credible prior information. The WinBugs code for SST-2-HOM model is included in the current article.

All results for the mixed-treatment analysis are reported as posterior median with corresponding 95% credibility intervals (CrIs). To evaluate whether variability of results across different comparisons of the network may have affected the results, the inconsistency of the model was calculated as suggested by Lu and Ades.⁵⁴ When the confidence interval of effect estimates did not cross the unit, these were considered significant.

The main clinical and methodological characteristics of each trial are shown in Table 1 and Table 2. The studies selected were performed in patients enrolled on the basis of a diagnosis of hypertension (186 378 [83.5%])^{10-20,23-31,33} or in a population at high cardiovascular risk with a predominant presence of hypertensive patients (36 935 [16.5%]).^{21,22,32,34,35} The beginning year of recruitment ranged from 1987 to 2004. All studies were published from 1997 through 2009.

A total of 223 313 patients were enrolled in the selected studies. Among these individuals, 24 009 (10.8%) were randomized to receive a conventional treatment. In the Captopril Prevention Project (CAPPP),¹⁵ Controlled Onset Verapamil Investigation of Cardiovascular End Points (CONVINCE),²⁴ Swedish Trial in Old Patients with Hypertension–2 (STOP-2),¹⁷ and Nordic Diltiazem (NORDIL)¹⁹ studies, patients receiving conventional treatment were treated with β-blockers or diuretics or both, whereas in the Jikei Heart Study³² and Efficacy of Candesartan on Outcome in Saitama Trial (E-COST)³⁰ the conventional treatment included any agent different from the active drug (ARBs). These agents were mostly CCBs, whereas only a negligible portion of these patients were receiving diuretic-based treatment (<4%).

A total of 40 516 patients (18.1%) were specifically assigned to a diuretic-based therapy, 9067 (4.1%) to an α-blocker–based strategy, and 14 564 (6.5%) to a β-blocker–based therapy. The remaining patients were treated with an antihypertensive strategy based on the newer antihypertensive drug classes, including CCBs (55 805 [25%]), ACE inhibitors (33 651 [15.1%]), and ARBs (27 095 [12.1%]). In addition, 18 606 patients (8.3%) were randomized to receive placebo.

In the pooled studies, a total of 8554 cases of HF were recorded during the follow-up (3.8% of patients).

We performed a series of conventional meta-analyses to summarize the results of trials directly comparing the same classes of drugs. These results are reported in Figure 2, in which we constructed a diagram of the network of clinical trials comparing specific antihypertensive strategies. According to these direct comparisons, ACE inhibitors, ARBs, CCBs, and diuretics were superior to placebo (Figure 2). Diuretic-based therapies were more effective than those based on α-blockers and CCBs, and slightly better than those based on ACE inhibitors. However, conventional treatment, ACE inhibitors, and ARBs showed higher efficacy than CCBs in preventing HF (Figure 2). No significant heterogeneity was detected in any of these meta-analyses (P > .05), with the exception of that comparing ARBs vs placebo (P = .02). No significant publication bias was detected (P > .10).

As shown in Figure 3 and Table 3, network meta-analysis confirmed the trends observed in the direct comparisons, since all the active treatments, with the exception of α-blockers, were more effective than placebo in HF prevention. In the case of β-blockers, however, the difference was not statistically significant. Antihypertensive therapies based on diuretics, ACE inhibitors, and ARBs were the most effective therapeutic strategy in HF prevention. Among these treatments, diuretics were significantly more effective than RAS inhibitors. Calcium channel blockers and β-blockers seemed to be less effective as first-line antihypertensive classes, since they were significantly inferior to diuretics, and they tended to be inferior to RAS inhibitors. The overall inconsistency of the model was low (σ²_W = 0.06), as also represented by the posterior probability that σ²_W > σ² (posterior mean of between-trials variance) (where “w” indicates the weighting factor, which was equal to 0.42).

Similar results were obtained in all the subgroup analyses (Table 4) since diuretics, closely followed by RAS inhibitors, seemed to be superior to CCBs and β-blockers in HF prevention. Interestingly, in studies with greater proportions of women and elderly patients, the difference between diuretics and ACE inhibitors was smaller. Moreover, in studies with greater percentages of men and younger subjects, the difference between CCBs and RAS inhibitors was less evident with respect to the results of the other analyses. However, these interesting results should be interpreted taking into account that the sample size, and therefore the power of these subanalyses, was lower than that of the main analysis.

To our knowledge, our study is the largest network meta-analysis ever performed in essential hypertension, and it was aimed specifically at investigating the efficacy of different antihypertensive strategies in the prevention of HF. The main finding of our analysis is that all antihypertensive strategies are better than placebo in HF prevention, with the exception of α-blockers.Diuretics and RAS inhibitors (ACE inhibitors, ARBs) are the most effective class of drugs, a diuretic-based treatment being the more effective first-line antihypertensive intervention for preventing HF. Our analysis also shows that CCBs and β-blockers are significantly less effective than diuretics as first-line agents, and they tend to be inferior to RAS inhibitors. These results do not seem to be significantly influenced by age and sex, as highlighted by our subanalyses (Table 4).

Our results extend the evidence provided by previous meta-analyses. In 2003, Psaty et al⁵⁵ published the results of a network meta-analysis that demonstrated that low-dose diuretics were the most effective class of drugs in the prevention of total cardiovascular mortality and morbidity, particularly HF, compared with other classes of antihypertensive agents, especially CCBs. Our current analysis is based on a substantially larger number of studies, it is more specifically oriented to HF outcomes, and it is up to date, including all the most recent trials published over the past 6 years. The recent trials (eg, the Anglo-Scandinavian Cardiac Outcomes Trial–Blood Pressure Lowering Arm [ASCOT-BPLA]³¹ and the Ongoing Telmisartan Alone and in Combination with Ramipril Global Endpoint Trial [ONTARGET]³³/Telmisartan Randomised AssessmeNt Study in ACE iNtolerant subjects with cardiovascular Disease [TRANSCEND]³⁵ trials) mostly evaluated “newer” antihypertensive drugs as CCBs, ACE inhibitors, and ARBs, which are extensively used today in hypertension treatment. Based on these reasons, our results significantly update and consolidate previous data and may provide evidence that can be more appropriately translated within our current clinical contexts.

With respect to a landmark meta-analysis of the Blood Pressure Lowering Treatment Trialists' Collaboration (BPLTTC) study,⁵⁶ also published in 2003, our data confirm that diuretics are more effective than CCBs in HF prevention in patients with hypertension, but also extend substantially the analysis of ACE inhibitors and ARBs. In keeping with our results, it has been recently demonstrated in a meta–regression analysis conducted in patients with hypertension or in patients with a high cardiovascular risk profile that RAS inhibitors are more effective than CCBs in HF prevention, although in that study CCBs were also shown to be significantly more effective than placebo.³⁷

On the one hand, a major advantage of our current meta-analytical approach is obviously the larger size compared with individual trials, in which HF usually represents a secondary end point. The approach used in our current study differs from traditional meta-analyses, which are characterized by a series of smaller meta-analyses of different active comparisons, which provide less robust information. Although specific comparisons between some classes of antihypertensive drugs have been investigated in multiple studies, other comparisons have been performed only in a single trial, and to our knowledge some other comparisons have never been performed. In the network meta-analysis it is possible to combine the results of direct comparisons to indirect comparisons extrapolated by trials that have treatments in common. Previous empirical evidence has validated indirect comparisons in many conditions and interventions.⁵⁷

On the other hand, some important issues cannot be resolved by our study and require a comment. First of all, our methodological approach did not analyze the influence on HF prevention of the differences in blood pressure reductions achieved with the different therapies. Therefore, the existence of a relationship between blood pressure reduction differences and HF incidence cannot be excluded. According to the blood pressure differences observed in the different studies, however, this aspect is likely to play a role in the comparisons between different active treatments and placebo. On the contrary, the small differences between blood pressure reductions induced by the various antihypertensive agents do not seem to account for the different effects on HF prevention. This is in line with the analysis by Psaty et al,⁵⁵ in which the various treatment strategies were associated with slightly different degrees of blood pressure reductions. Moreover, in the analysis by the BPLTTC study⁵⁶ the reduction of HF incidence associated with different antihypertensive strategies was independent of the extent of blood pressure reduction. Finally, the recent analysis by Verdecchia et al³⁷ demonstrated that a RAS inhibitor–based therapy is more effective than CCB-based therapy in HF prevention, independently of blood pressure reductions.

The clinical heterogeneity of the trials included in our analysis might also have implications for the variations in HF incidence observed in the studies, particularly when considering the variability in the diagnostic criteria for HF assessment and the different doses of active treatments used in the various trials. In a previous published analysis,⁷ we demonstrated that in hypertension trials, a diagnosis of HF based on less rigorous criteria was not associated with an overestimation of HF diagnosis. Previous analyses may suggest that HF clinical diagnosis in clinical trials is in agreement with other different diagnostic criteria for HF.⁵⁸ Moreover, in our study no significant heterogeneity in traditionally performed meta-analyses was detected, and the inconsistency of the network meta-analysis was low. This implies that differences among studies did not have a substantial impact on the results, which therefore seem to be consistent and reliable. Of course, this aspect does not represent a bias within each of the trials used for the analysis because the same diagnostic criteria of HF were applied to treatment arms.

Our analysis did not allow us to specifically test the efficacy of different antihypertensive strategies in patients with and without diabetes mellitus, nephropathy, and a history of myocardial infarction. However, with regard to diabetes mellitus, a substudy of the Antihypertensive and Lipid-Lowering Treatment to Prevent Heart Attack Trial (ALLHAT)⁵⁹ demonstrated the superiority of diuretics compared with ACE inhibitors and CCBs. Because heterogeneity and inconsistency among trials were low, it is likely that our results do not significantly differ among these categories of patients. Further specific studies may be required to address this issue.

Another possible limitation of our analysis is the different sample size among patients prescribed different antihypertensive therapies and a relative low overall incidence of HF. Moreover, the multiple antihypertensive treatment regimens in the different trials could not be specifically considered in the analysis, so only first-line agents’ effects were generally evaluated. Finally, the aim of our study was to evaluate specifically the impact of different antihypertensive treatments in the prevention of HF, while cardiovascular mortality, myocardial infarction, and stroke were not addressed. Therefore, it is important to note that our results are limited to the impact of antihypertensive treatment in the prevention of HF, which continues to be a major issue in hypertension treatment.⁷ Indeed, none of the antihypertensive drugs were found to be clearly superior to others with respect to cardiovascular mortality beside the blood pressure–lowering effect.⁶⁰ The different efficacy of antihypertensive classes in HF prevention reported in our study cannot be extended to prevention of other cardiovascular outcomes. In fact, with regard to prevention of important outcomes, such as nonfatal myocardial infarction and stroke, there is evidence that several classes are equally effective. For instance, CCBs, which were found to be less effective in HF prevention in our analysis, are effective in stroke prevention.⁶¹

Altogether, our results seem to support the use of diuretics and ACE inhibitors (or ARBs) as first-line therapy for HF prevention in hypertension. Therefore, especially in patients at risk of developing HF, diuretics alone or in combination with RAS-inhibiting drugs could be preferred. Further studies comparing different combination therapies are required to assess this latter hypothesis. Unfortunately, the only trial addressing the effects of combination therapies on outcomes (ACCOMPLISH⁴⁴) did not provide data on HF incidence. Based on our current results, it would have been interesting to compare associations based on a diuretic and ACE inhibitor vs combinations of CCBs with ACE inhibitors in the prevention of HF in patients with hypertension.⁴⁴

On the other hand, based on our results and the results of other recently published analyses,^37,62 CCBs and β-blockers should not be preferred as first-line agents in patients with hypertension at higher risk to develop HF.

In conclusion, our results seem to support the use of diuretics and RAS-inhibiting drugs alone or in combination as first-line therapies for HF prevention in hypertension. These classes of drugs should be preferred to CCBs and β-blockers in patients with hypertension at high risk of developing HF. Further studies in these clinical subgroups of hypertensive patients, as well as studies comparing different combination therapies, may be helpful to corroborate our findings.

Correspondence: Massimo Volpe, MD, Department of Cardiology, Second Faculty of Medicine, University of Rome “La Sapienza,” Via di Grottarossa 1039, Rome, Italy (massimo.volpe@uniroma1.it).

Accepted for Publication: June 9, 2010.

Published Online: November 8, 2010. doi:10.1001/archinternmed.2010.427

Author Contributions: Dr Volpe had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Sciarretta, Palano, and Volpe. Acquisition of data: Sciarretta and Palano. Analysis and interpretation of data: Tocci and Baldini. Drafting of the manuscript: Sciarretta, Palano, Tocci, Baldini, and Volpe. Critical revision of the manuscript for important intellectual content: Volpe. Statistical analysis: Sciarretta and Baldini. Study supervision: Sciarretta, Palano, Tocci, and Volpe.

Financial Disclosure: None reported.