We searched the database through PubMed, EMBASE, and the Cochrane Central Register of Controlled Trials for randomized clinical trials (RCTs) from November 1977 to June 2017 using Medical Subject Heading ‘antihypertensive agents’ or ‘angiotensin II type 1 receptor blockers’ or ‘angiotensin-converting enzyme inhibitors’ and ‘heart failure’. Additionally, studies in the reference lists of the identified articles were also hand searched. The search was limited to RCTs, human subjects and English. The process was strict to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISRMA) statement [17].

Studies were deemed eligible if they: 1) were RCTs, targeting HF patients with reduced ejection fraction (HFrEF, left ventricular ejection fraction ≤45%), with a median or mean follow-up of more than 12 months; 2) compared ACEIs and ARBs treatment (any dose or type) with placebo treatment, no treatment, or other anti-HF drugs treatment; 3) reported cardiovascular or total mortality. When the outcomes obtained from the same population in different publications, only the latest report was included in the analysis.

Two independent investigators (Y. X. and D. X.) extracted data from these reports, and disagreements were resolved by consensus. After excluding the unrelated studies, the following data were extracted: study characteristics (author, publication year, sample size, follow-up period), population baseline characteristics (age, sex, cause of heart failure, risk factors) and end-points. Study quality was assessed using the Jadad score, which is a five-point quality scale, with low quality studies having a score of <2 and high-quality studies a score of ≥3 [18].

All-cause mortality and CV mortality were thought to be the main outcomes.

Dichotomous outcome data from individual trials was analyzed using the risk ratio (RR) measure and its 95% confidence interval (CI) [19]. Overall effect was estimated using the Mantel-Haenszel method for RRs [20, 21].

Heterogeneity was evaluated using χ² tests and I² statistics. Studies were considered statistically heterogeneous if I² > 50% and p ≤ 0.05. If heterogeneity between studies were identified, a random-effects model was applied. Otherwise, a fixed-effects model was taken instead [22]. Publication bias was assessed with funnel plots and the Begg regression test [22].

In sensitivity analysis, we removed anyone of the study at a time and repeated the meta-analysis to ensure that no single study would be responsible for the significance of any result separately [22].

Meta-regression was conducted to explore the potential heterogeneity related to the participants (age, cause of HF, left ventricular ejection fraction, and follow-up weeks), the agent used (different types). P < 0.05 was considered statistically significant [22].

Meta-analysis was performed by the Review Manager software (Version 5.0. Copenhagen: The Nordic Cochrane Centre, the Cochrane Collaboration) and the Stata software (version 12.0; Stata Corporation, College Station, TX).

Initial search identified 1002 reference articles, of these 107 relevant articles were selected and reviewed. Then, several studies were further excluded because they were publications from the same trial (n = 7) or reported of the end points other than cardiovascular events or death (n = 15) or used RAAS inhibitors simultaneously in both trial arms (n = 7) or were not relevant (n = 40). Finally, 38 RCTs assessing the association of cardiovascular outcome or cardiovascular or total mortality with ACEIs or ARBs were included in the meta-analysis [2‐14, 22‐47]. As shown in Fig. 1, literature research process was summarized by a chart flow. Baseline characteristics of all selected studies are detailed in Table 1. A total of 47,662 subjects were included with a mean/median follow-up ranged from 12 weeks to 4.5 years. Of all 38 studies, six (n = 8404) trials compared ARBs with placebo [13, 43‐47], while 32 trials (n = 39,254) compared ACEIs with various control therapies (13 arms (n = 10,134) compared ACEIs [2‐6, 23‐30] with placebo treatment; 10 arms (n = 8714) in which the comparator was active treatment [7, 8, 31‐38]; and 9 arms (n = 20,406) compared ACEIs with ARBs [9‐12, 14, 39‐42]). Two independent investigators (Y. X. and D. X.) assessed the quality of the studies included. There were 32 studies of good quality (Jadad score ≥ 3) with low risk of bias and six studies of low quality (Jadad score < 3) with high risk of bias.

Thirty-two studies [2‐12, 14, 23‐42] reported the effect of ACEIs on all-cause mortality in a total of 39,254 HF patients with moderate heterogeneity in overall analysis (I² = 44%, p = 0.005). ACEIs were associated with a statistically significant 11% reduction in all-cause mortality (RR: 0.89, 95% CI: 0.83–0.96, p = 0.001, Fig. 2). Similar findings were observed when ACEIs were compared with placebo treatment (p < 0.001, Fig. 2). There was no evidence of publication bias (p = 0.833).

Moreover, 15 studies [9‐14, 39‐47] reported the effect of ARBs on all-cause mortality in a total of 28,814 HF patients with no significant heterogeneity in overall analysis (I² = 26%, p = 0.17). ARBs were not associated with a reduction in all-cause mortality (RR: 1.03, 95% CI: 0.98–1.08, p = 0.28, Fig. 3). Similar findings were observed when comparing with placebo or ACEIs (p ≤ 0.60, Fig. 3). And there was no evidence of publication bias (p = 0.921).

Figure 4 showed the relation between the network of RCTs.

Seventeen studies [3‐6, 8‐11, 14, 24, 32, 35, 36, 38, 40‐42] reported the effectiveness of ACEIs for CV mortality in a total of 28,302 HF patients with moderate heterogeneity in overall analysis (I² = 51%, p = 0.009). ACEIs were associated with a statistically significant 14% reduction in CV mortality (RR: 0.86, 95% CI: 0.78–0.94, p = 0.001, Fig. 5). Similar findings were observed when ACEIs treatment was compared with placebo treatment (p < 0.001, Fig. 5). However, when ACEIs were compared with active treatment or ARBs, ACEIs did not significantly reduce CV mortality. There was no evidence of publication bias (p = 0.967). The SAVE [4], TRACE [6] and VALIANT [11] study were conducted in patients with HF after myocardial infarction. After exclusion of these three trials, heterogeneity among the trials was not significantly different (I² = 34%, p = 0.10, RR, 0.85, 95% CI: 0.76–0.95, p = 0.005).

Moreover, 11 studies [9‐11, 13, 14, 40‐42, 45‐47] reported the effectiveness of ARBs for CV mortality in a total of 27,991 HF patients with no significant heterogeneity in overall analysis (I² = 40%, p = 0.08). ARBs were associated with no reduction in CV mortality (RR: 1.01, 95% CI: 0.92–1.12, p = 0.78, Additional file 1: Figure S1). Similar findings were observed when ARBs were compared with placebo or ACEIs (p ≤ 0.50, Additional file 1: Figure S1). And there was no evidence of publication bias (p = 1.000).

Meta-regression was conducted in different ages (p = 0.97), causes of HF (p = 0.90), left ventricular ejection fractions (p = 0.09), follow-up weeks (p = 0.41) to observe effects of ACEIs treatment on all-cause mortality. The findings remained unaltered in these subgroup analyses. But, univariate meta-regression of ACEIs treatment on all-cause mortality varied by the types of ACEIs (p = 0.004). Captopril treatment reduced all-cause mortality by 9% (RR: 0.91, 95% CI: 0.85–0.98, p = 0.008) in HF patients as compared with control treatment. However, enalapril treatment did not reduce all-cause mortality in HF patients as compared with control treatment (RR: 0.93, 95% CI: 0.85–1.02, p = 0.13, Fig. 6). Of all these studies, one study compared ramipril with placebo and two studies compared lisinopril with placebo/active drugs. The results were shown in Fig. 5.

In this meta-analysis, we combined clinical trial data from 38 studies, which included 47,662 HF patients to assess the efficacy of RAAS inhibition on mortality. Overall, ACEIs reduce all-cause mortality by 11% and the corresponding value for CV mortality by 14%. However, ARBs have no significant effect on all-cause and CV mortality in HF patients. In head-to-head analysis, ACEIs are not superior to ARBs on all-cause and CV mortality. Thus, this meta-analysis provides compelling evidence that ACEIs are the most effective first-line treatment for preventing all-cause and CV mortality in HF patients.

RAAS inhibition has long been identified as a preferred first-line treatment for heart failure. However, previous studies indicated that there were different outcomes between AECIs and ARBs for heart failure. As early as 1987, CONSENSUS study [3] was conducted to evaluate the efficiency of enalapril in patients with HF. Six-month mortality in the enalapril group was 26% compared with 44% in the placebo group, giving a relative risk-reduction of 40% (p = 0.002) and at 1 year, these proportions were 36% and 52% (p = 0.001). After that, several studies demonstrated that ACEIs reduced all-cause and CV mortality in HF patients, particularly after myocardial infarction [4, 6]. However, most of ARBs are not proved to be effective on these crucial outcomes in HF patients. For example, the ELITE II study [10] found that losartan is not superior to captopril, although it has been suggested that the dose of losartan (50 mg) tested is not adequate. And in the CHARM-Alternative trial [13], candesartan did not reduce all-cause mortality in HF patients, but reduced the risk of CV death or HF hospitalization by 23% (p = 0.0004). The Val-Heft study46 showed the same results. These may due to the negative effect of ARBs on heart failure, which could be mediated through a vasoconstrictor-induced increase in blood pressure or a direct effect on cardiac and vascular tissues. So, more related studies are expected to conducted in this area. Besides, in some recent meta-analysis, Vark et al. [15] and Cheng et al. [16] presented that in patients with hypertension and diabetes, treatment with an ACEI resulted in a significant further reduction in all-cause and CV mortality, whereas ARBs had no benefit on these outcomes. These results are in agreement with our meta-analysis.

From a pharmacological viewpoint, ACEIs can reduce the negative effects caused by binding of angiotensin II and its receptor by inhibiting the conversion of angiotensin I to angiotensin II. In addition, by restraining the degradation of angiotensin (1–7) and promoting its combination with Mas receptor, ACEIs may have effect on dilating blood vessels, anti-inflammatory and anti-fibrosis. Moreover, ACEIs can also reduce degradation of bradykinin and promote its role in β2 receptor, which contributes to dilation of blood vessels, anti-proliferation, endothelial protection and other positive effects [48, 49]. In contrast, RAAS blockade with ARBs is achieved by inhibiting the binding of angiotensin II to the angiotensin II type one receptor, which is believed to mediate the harmful cardiovascular effects of angiotensin II due to the elevated level of angiotensin II by compensatory mechanism. These different pharmacological mechanisms may explain that ACEIs may be superior to ARBs in reducing CV events. Therefore, an ACEI agent may be a superior ARB antagonist in hypertension and heart failure.

There was low to moderate heterogeneity of analysis on the effect of ACEIs on all-cause and CV mortality. Meta-regression, sensitivity and subgroup analysis were conducted to estimate the influence of each study. Firstly, no evidence shows that the observed effects varied by age, causes of HF, left ventricular ejection fractions and follow-up weeks by meta-regression. However, different types of ACEIs may influence the effect on all-cause mortality, which means that captopril may be superior to enalapril in reducing all-cause mortality in HF patients. Secondly, the SAVE [4], TRACE [6] and VALIANT [11] study were conducted in patients with heart failure after myocardial infarction. After exclusion of these three trials, heterogeneity among the trials exploring the effect of ACEIs on CV mortality was not significantly different (I² = 34%, p = 0.10, RR, 0.85, 95% CI: 0.76–0.95, p = 0.005). So, the significant heterogeneity was attributable to the different control treatment. There was no evidence of publication bias (p = 0.721).

Strengths of the present study are no other than the large sample size with a mean/median follow-up ranging from 12 weeks to 4.5 years and a high representativeness.

It has been acknowledged that there are some limitations to this study. Firstly, this analysis used aggregate data as reported or calculated in published articles, rather than data of individual patients. Secondly, there were a great deal of variations between the studied populations. For example, causes of heart failure differed from each other. In addition, these trials used different ACEIs or ARBs at a different dosage. It is likely that different ACEIs and ARBs may have a total different effect on the cardiac mortality. Moreover, the present study is unable to address whether the efficacy may be varied in HF patients with different ethnic backgrounds.