Results of the search

Up to November 2017, the search strategy identified 15,145 citations (Figure 1). After excluding all the studies that did not meet the inclusion criteria or those we have included before, we performed full‐text assessment of five potentially eligible studies and identified three new trials (NESTOR 2015; SILVHIA 2001; Xiao 2016) (four citations) that we included in the review update. This update includes 45 RCTs with 87 citations, i.e. the three new RCTs and the 42 RCTs (83 citations) in the first publication of this review (Xue 2015).

Included studies

The 45 included studies involved 66,625 participants with a mean age of 66 years. Participants in nine studies were under 50 years old (Buus 2004; Buus 2007; Dahlöf 1993; Pedersen 1997; Schiffrin 1994; Sørensen 1998; Tarnow 1999; Xiao 2016; Zeltner 2008); in 22 studies participants were between 50 and 59 years old (Ariff 2006; Dahlöf 2002; Dalla 2004; Derosa 2004; Derosa 2005; Derosa 2014; Esnault 2008; Estacio 1998; Gottdiener 1998; Hauf‐Zachariou 1993; Hughes 2008; IDNT 2001; Malmqvist 2002; Parrinello 2009; Petersen 2001; Roman 1998; Schmieder 2009; Schneider 2004; Seedat 1998; SILVHIA 2001; Tedesco 1999; TOHMS 1993); and over 60 years old in the remaining 14 studies (ALLHAT 2002; BENEDICT 2004; Devereux 2001; Fogari 2012; Gerritsen 1998; Hajjar 2013; Hayoz 2012; Himmelmann 1996; LIFE 2002; NESTOR 2015; Ostman 1998; Schram 2005; Terpstra 2004; VALUE 2004). The mean duration of therapy was 1.9 years, ranging from 0.5 to 5.6 years. The number of participants who received RAS inhibitors was 25,421, while 5,525 received beta‐blockers, 19,040 CCBs, 16,316 thiazides, 240 alpha‐blockers, and 83 CNS active drugs. Three studies contained multiple different drug groups: Gottdiener 1998 contained six, TOHMS 1993 contained five, and ALLHAT 2002 contained three, so the numbers of studies comparing RAS inhibitors with other drug classes were 17 for beta‐blockers, 22 for CCBs ‐ within which there were two studies that used non‐dihydropyridine (BENEDICT 2004; Gottdiener 1998), and 20 studies that used dihydropyridine, 10 for thiazides, 3 for alpha‐blockers, and 1 for CNS active drugs.

Most of the included studies were industry‐sponsored (28/45). Participants with diabetes were involved in 14 studies, while one study included participants with impaired fasting glucose; participants with decreased renal function in seven studies, and seven studies contained participants with at least one risk factor for CV diseases. Three studies recruited only men (Dahlöf 1993; Gottdiener 1998; Schiffrin 1994). One study included only women, as it focused on postmenopausal women (Hayoz 2012). All 87 included citations were published in English with publication years ranging from 1993 to 2016.

Most participants (30 studies) were recruited from European countries; seven studies included participants from North America; two studies included participants from North America, Europe, and Asia (Dahlöf 2002; VALUE 2004); one study included participants from North America, South America, Europe, Asia and Australia (IDNT 2001); NESTOR 2015 included participants from North America, South America, Europe and Asia; one study included participants from North America and Europe (LIFE 2002); Devereux 2001 included participants from Europe and Asia; Seedat 1998 included participants from South Africa; and Xiao 2016 included participants from Asia. Fifteen of the 45 included studies reported ethnicity; the percentages of white, Hispanic, Asian, Black and other race participants were 71.0%, 0.3%, 1.7%, 23.7% and 3.3%, respectively.

In terms of baseline comorbidities, nine studies stated that they would not include people with a history of prior MI or stroke; 14 studies allowed participants with a history of prior MI or stroke if this had not occurred within the previous three or six months); the other 22 studies made no clear statement, but in general the proportion of participants without cardiac‐cerebral vascular comorbidities was high. Overall, this review represents treatment effects for primary prevention.

A stepwise therapeutic regimen was used in 34 studies, in which add‐on drugs were allowed to achieve BP goals. The second‐line drugs included open‐label, non‐study agents such as CCBs, thiazides, or beta‐blockers. The remaining eleven studies restricted the therapeutic regimens to monotherapy (Dahlöf 1993; Derosa 2004; Derosa 2014; Gottdiener 1998; Himmelmann 1996; Hauf‐Zachariou 1993; Sørensen 1998; Tedesco 1999; Terpstra 2004; TOHMS 1993; Xiao 2016).

With regard to the clinical classification of hypertension (see ESH/ESC 2013), we classified mean blood pressure of participants at baseline into two groups: 30 studies included Grade 1 hypertensive participants (SBP: 140 mmHg to 159 mmHg); 15 studies included Grade 2 hypertensive participants (SBP: 160 mmHg to 179 mmHg). Baseline untreated mean BP was 156/89 mmHg (SBP/DBP) for RAS inhibitors; 151/86 mmHg for CCBs; 172/98 mmHg for beta‐blockers; 146/85 mmHg for thiazides; 150/96 mmHg for alpha‐blockers; 152/99 mmHg for CNS active drugs.

For details, see Characteristics of included studies.

Excluded studies

Full‐text screening according to the prespecified inclusion criteria led to us excluding three of the seven citations during the update, in addition to the four citations excluded in the previous version of the review. In total, we excluded seven citations of six studies in the updated review. The reasons for each study's exclusion are described in Characteristics of excluded studies.

Allocation

We assessed seven of the 45 studies as being at low risk of bias for reporting the method for generation of random sequence and one study as being at high risk (Roman 1998); in the remaining 37 studies the risk of bias for this domain was unclear. We assessed three of the 45 studies as being at low risk for allocation concealment, one study as being at high risk, and 41 studies as being at unclear risk. The three studies at low risk used either a central office allocation (ALLHAT 2002; VALUE 2004), or were strictly confidential until unblinding time (Schmieder 2009); one study reported using alternate allocation, which is a high risk method (Seedat 1998); two studies at unclear risk of selection bias reported the allocation concealment by using an envelope to maintain the random number (Derosa 2004; Derosa 2005); however, it was not clear whether the envelope was transparent or opaque.

Blinding

All the 45 included studies were at low risk of performance bias as they were all double‐blinded and met the inclusion criteria. In terms of detection bias, we judged only six studies to be at low risk due to the use of blinding for outcome assessment for BP or HR, which was critical for the control of detection bias; the risk of bias for this domain was unclear for the remaining 39 studies. We thought that unblinded assessment of outcomes like MI, stroke, HF, CV events, all‐cause death, and ESRF was not as critical as it would be for BP and HR.

Incomplete outcome data

We judged the risk of attrition bias in 40 of the 45 studies included in the review to be low because missing data were unlikely to have an impact because of low or equal numbers of dropouts between arms. In one study this risk was unclear (Petersen 2001); and in the other four studies we judged it to be high. Among these four studies with a high risk of attrition bias, Gottdiener 1998 only included participants with a left atrial dimension measurement (a small proportion of all participants) in the analysis. Pedersen 1997 and Tarnow 1999 had many participants lost to follow‐up at the end of study and Hayoz 2012 reported inconsistent numbers of participants in Figure 1 and Table 2.

Selective reporting

In this review, we judged 43 included studies to have a low risk of reporting bias; we judged that two studies had a high risk of selective reporting as they did not report HR, which was a prespecified outcome in their 'Methods' sections (Derosa 2004; Terpstra 2004).

Other potential sources of bias

Seven studies had a high risk of other potential sources of bias. Pedersen 1997 and Roman 1998 had unbalanced baseline characteristics. VALUE 2004 had an unbalanced proportion of monotherapy and highest dose between the two groups (including HCTZ and other non‐study add‐on drugs); Estacio 1998 had an unbalanced proportion of monotherapy. LIFE 2002 was evaluated as being at high risk as it was funded and conducted by the pharmaceutical company Merck. Similarly, many of the authors of IDNT 2001 had received research grants from Bristol‐Myers Squibb. Numbers of participants reported for different outcomes were not consistent in SILVHIA 2001.

First‐line RAS inhibitors versus first‐line CCBs

Compared with CCBs, RAS inhibitors decreased HF (5 RCTs, 35,143 participants, RR 0.83, 95% CI 0.77 to 0.90; Analysis 1.3), and increased stroke (4 RCTs, 34,673 participants, RR 1.19, 95% CI 1.08 to 1.32; Analysis 1.5), but were not significantly different for all‐cause death (5 RCTs, 35,226 participants, RR 1.03, 95% CI 0.98 to 1.09; Analysis 1.1), total CV events (6 RCTs, 35,223 participants, RR 0.98, 95% CI 0.93 to 1.02; Analysis 1.2), total MI (5 RCTs, 35,043 participants, RR 1.01, 95% CI 0.93 to 1.09; Analysis 1.4), and ESRF (4 RCTs, 19,551 participants, RR 0.88, 95% CI 0.74 to 1.05; Analysis 1.6). CCBs lowered SBP and DBP to a greater degree than RAS inhibitors (SBP: 20 RCTs, 36,437 participants, MD 1.23, 95% CI 0.90 to 1.56; Analysis 1.7; DBP: 20 RCTs, 36,437 participants, MD 0.98, 95% CI 0.79 to 1.18; Analysis 1.8). There was no difference in the effect of CCBs and RAS inhibitors on HR (5 RCTs, 540 participants, MD 0.30, 95% CI ‐1.63 to 2.22; Analysis 1.9).

First‐line RAS inhibitors versus first‐line thiazides

Compared with thiazides, RAS inhibitors increased HF (1 RCT, 24,309 participants, RR 1.19, 95% CI 1.07 to 1.31; Analysis 2.3), and increased stroke (1 RCT, 24,309 participants, RR 1.14, 95% CI 1.02 to 1.28; Analysis 2.5), but were not significantly different for all‐cause death (1 RCT, 24,309 participants, RR 1.00, 95% CI 0.94 to 1.07; Analysis 2.1), total CV events (2 RCTs, 24,379 participants, RR 1.05, 95% CI 1.00 to 1.11; Analysis 2.2), total MI (2 RCTs, 24,379 participants, RR 0.93, 95% CI 0.86 to 1.01; Analysis 2.4), and ESRF (1 RCT, 24,309 participants, RR 1.10, 95% CI 0.88 to 1.37; Analysis 2.6). Thiazides lowered SBP to a greater degree than RAS inhibitors (10 RCTs, 26,382 participants, MD 1.60, 95% CI 1.20 to 1.99; Analysis 2.7), but had a similar effect to RAS inhibitors on DBP (9 RCTs, 26,335 participants, MD ‐0.12, 95% CI ‐0.36 to 0.13; Analysis 2.8). There was also no difference in the effect on HR, but only two small trials reported this outcome (2 RCTs, 84 participants, MD 0.66, 95% CI ‐2.87 to 4.19; Analysis 2.9).

First‐line RAS inhibitors versus first‐line beta‐blockers

Compared with beta‐blockers, RAS inhibitors decreased total CV events (2 RCTs, 9,239 participants, RR 0.88, 95% CI 0.80 to 0.98; Analysis 3.2) and decreased stroke (1 RCT, 9,193 participants, RR 0.75, 95% CI 0.63 to 0.88; Analysis 3.5). Beta‐blockers and RAS inhibitors were not significantly different for all‐cause death (1 RCT, 9,193 participants, RR 0.89, 95% CI 0.78 to 1.01; Analysis 3.1), HF (1 RCT, 9,193 participants, RR 0.95, 95% CI 0.76 to 1.18; Analysis 3.3), or MI (2 RCTs, 9.239 participants, RR 1.05, 95% CI 0.86 to 1.27; Analysis 3.4). The effect on ESRF could not be assessed because there was only one small trial that examined this outcome (1 RCT, 46 participants, RR 0.33, 95% CI 0.01 to 7.78; Analysis 3.6). Beta‐blockers lowered DBP and HR more than RAS inhibitors (DBP: 16 RCTs, 10,905 participants, MD 0.48, 95% CI 0.14 to 0.83; Analysis 3.8; HR: 10 RCTs, 9,979 participants, MD 6.05, 95% CI 5.59 to 6.50; Analysis 3.9). The effect on SBP did not differ between the two classes of drug (16 RCTs, 10,905 participants, MD ‐0.55, 95% CI ‐1.22 to 0.11; Analysis 3.7).

First‐line RAS inhibitors versus first‐line alpha‐blockers

RAS inhibitors lowered SBP more than alpha‐blockers did (3 small RCTs, 380 participants, MD ‐2.38, 95% CI ‐3.98 to ‐0.78; Analysis 4.1), but did not differ in their effect on DBP and HR (DBP 3 small RCTs, 380 participants, MD ‐0.12, 95% CI ‐1.09 to 0.85; Analysis 4.2; HR: 1 small RCT, 44 participants, MD 3.10, 95% CI ‐2.41 to 8.61; Analysis 4.3).

First‐line RAS inhibitors versus first‐line CNS active drugs

When compared with CNS active drugs in one small trial, RAS inhibitors did not differ in their effect on SBP (1 RCT, 56 participants, MD 1.30, 95% CI ‐6.01 to 8.61; Analysis 5.1), DBP (1 RCT, 56 participants, MD ‐0.30, 95% CI ‐1.85 to 1.25; Analysis 5.2), or HR (1 RCT, 56 participants, MD 1.50, 95% CI ‐4.13 to 7.13; Analysis 5.3).

Subgroup analysis and investigation of heterogeneity

In this review, when the result was significant and the value of I² was greater than 50%, we tested whether the result was still significant using the random‐effects model. However, in presenting the data we use the fixed‐effect model, as it weights the contributing trials more appropriately.

In an attempt to explore the heterogeneity of RAS inhibitors versus CCBs on HF (I² of 68%) we analyzed the trials according to whether or not the participants had decreased renal function. In those with decreased renal function the RR was 0.55, 95% CI 0.43 to 0.70, without heterogeneity (Dalla 2004; IDNT 2001); while in those without decreased renal function the RR was 0.87, 95% CI 0.80 to 0.95, without heterogeneity (ALLHAT 2002; Estacio 1998; VALUE 2004). Subgroup analysis thus provided a possible explanation for the variation of effect sizes across the studies. The magnitude of the effect of RAS inhibitors for decreasing HF, when compared to CCBs, was greater in hypertensive participants with kidney dysfunction than in those with normal renal function.

The key results on the clinically important outcomes and grading of the evidence quality are presented in the 'Summary of findings' tables, which we created by using the software GRADEpro 3.6. (Atkins 2004) (summary of findings Table for the main comparison; summary of findings Table 2; summary of findings Table 3) These tables provide the absolute effects as well as the relative effects.