Overall, 147 trial reports were included in the analysis: 108 were blood pressure difference trials and 46 drug comparison trials (seven trial reports with two treatment groups and a placebo group fell into both categories, treatment versus placebo and one treatment versus the other). Table 1 summarises the trials (see web extra tables 1i-iii and 2 for individual data from the trials). Forest plots of individual trial results are presented in 55 web extra figures (available at www.wolfson.qmul.ac.uk/bptrial/
) and the summary relative risk estimates and results for heterogeneity testing are shown in web extra table A. Results on CHD events and stroke are presented first, according to the five questions posed in the introduction, followed by results on heart failure and all cause mortality.
Table 1 Randomised trials of blood pressure lowering drugs according to category of trial (see web extra tables 1i-iii and 2 for details of individual trials)
Do β blockers have a special effect in preventing CHD events in people with a history of CHD?
Blood pressure difference trials
Figure 1 shows the reduction in CHD events in the 37 blood pressure difference trials of β blockers in people with a history of CHD, comparing β blockers with placebo (32 trials) or with an untreated control group (five trials). CHD events were, on average, reduced by 29% (relative risk 0.71, 95% confidence interval 0.66 to 0.78), significantly greater (P<0.001) than the 15% reduction in single drug trials of β blockers in people without a history of CHD and of other classes of drug in people with and without a history of CHD. The greater protective effect of β blockers in people with CHD was explained by a greater effect in the 27 trials that recruited participants at the time of an acute myocardial infarction (within a month in 25 trials and within four months in the other two). The risk reduction for recurrent CHD events in these 27 trials of people with an acute myocardial infarct was 31% (relative risk 0.69, 0.62 to 0.76); the duration of follow-up was short (77% of the events occurred in the first year and 94% in the first two years), so almost all the recurrent events occurred within one or two years of the infarct. Eleven trials remained (not 10 (37−27) because one trial recruited some participants with a recent infarct and some withoutw62); these recruited participants with a history of CHD but no recent infarct and in these the risk reduction was 13% (relative risk 0.87, 0.71 to 1.06; P=0.04 for the difference between the two groups of trials). In these 11 trials about 75% of the participants had had an infarct, but not within the last four months and typically several years before. The 13% risk reduction was similar to the 15% risk reductions in the other categories of single drug trials, whereas the 31% risk reduction after acute myocardial infarction was significantly greater (P<0.001). β blockers used for one or two years after an acute myocardial infarction were therefore about twice as effective as β blockers used in other circumstances and about twice as effective as other drugs used in any circumstances (see web figures 1a-e for forest plots of the individual trial results).
Fig 1Relative risk estimates of coronary heart disease events in single drug blood pressure difference trials according to drug (β blockers or other), presence of CHD, and for β blockers according to acute myocardial infarction (more ...) Drug comparison trials
The four drug comparison trials of β blockers compared with other drugs in people with CHD but no recent infarct (see web extra table 2) confirmed the absence of a special effect of β blockers in the absence of a recent infarct; the summary relative risk of CHD events was 0.99 (0.82 to 1.20), a relative risk of 1.0 indicating the same risk reduction from β blockers and other drugs.
In view of the special effect of β blockers, CHD events (but not stroke or heart failure) in all 37 blood pressure difference trials and all four drug comparison trials of β blockers in people with CHD were excluded from subsequent analyses according to the prior stipulation that we would do so if a special effect was observed, even though post hoc the special effect was limited to a subset (those with acute infarction).
Does the preventive effect of drugs differ in people with and without a history of cardiovascular disease?
The summary relative risk estimates of CHD events and stroke in the blood pressure difference trials, observed and standardised for reduction in blood pressure, were similar in the three categories of trials (no vascular disease, history of CHD, and history of stroke), showing no difference in effect in people with or without vascular disease (table 2, also see web extra figures 2a-f for forest plots of individual trial results). There was no heterogeneity across the trials (table 2) and no special effect of drugs other than β blockers after acute myocardial infarction.
Table 2 Summary relative risk estimates (95% confidence intervals) for coronary heart disease (CHD) events and stroke from randomised blood pressure difference trials observed and standardised to a blood pressure reduction of 10 mm Hg systolic and 5 (more ...)
Does blood pressure reduction alone explain the preventive effect of the drugs?
Blood pressure difference trials
Figure 2 shows the relative risk estimates of CHD events and stroke in the blood pressure difference trials, standardised to a blood pressure reduction of 10 mm Hg systolic and 5 mm Hg diastolic, together with the corresponding relative risk estimates derived from the meta-analysis of cohort studies (Prospective Studies Collaboration analysis25
), in the age group 60-69 years, the average age at the time of a cardiovascular event in the trials (table 1). The estimates from the trials meta-analysis were a 22% (95% confidence interval 17% to 27%) reduction in CHD events (relative risk 0.78) and a 41% (33% to 48%) reduction in stroke (relative risk 0.59). The cohort study meta-analysis showed a 25% decrease in CHD events (relative risk 0.75) and a 36% decrease in stroke (relative risk 0.64) for the same blood pressure difference of 10 mm Hg systolic, or 5 mm Hg diastolic (results from other cohort study meta-analyses were similar26 27
). Thus the reductions in disease events in the trials were similar to those expected from the cohort study results for the same reduction in blood pressure.
Fig 2Relative risk estimates of coronary heart disease events and stroke for a blood pressure reduction of 10 mm Hg systolic or 5 mm Hg diastolic in the blood pressure difference trials and in epidemiological cohort studies. (Total number of (more ...)
After only one year of follow-up (see web extra table 1) the reduction in CHD events was 20% (9% to 29%) and the reduction in stroke was 32% (18% to 44%) for a reduction of 10 mm Hg in systolic blood pressure and 5 mm Hg diastolic, similar to the long term trial results (22% and 41%) and similar to the results expected from the cohort studies (25% and 36%; see fig 2), indicating that the full potential effect of blood pressure reduction is achieved within a year.
Figure 3 shows the reductions in CHD events and stroke in the single drug trials comparing a specified drug with placebo (or with a control group not receiving the study drug in nine trials), separately for each of the five main classes of drug (the only drugs tested in single drug trials). The five classes of drug produced reductions in CHD events and stroke that were similar in magnitude. All the reductions were statistically significant but for angiotensin receptor blockers there were only four trials and hence insufficient statistical power to show an effect. Average differences between the treated and control groups in use of add-on drugs were small (0.03 drugs per participant) and similar for the different classes of drug (see web extra figures 3a-i for forest plots of the results of individual trials for each class of drug). There was no statistically significant heterogeneity for CHD events across trials of the five classes of drug (χ2=2.0, df=5, P=0.86), but the reduction in incidence of stroke was smaller in trials of β blockers (17%) than in single drug trials of the other four classes of drug combined (29%; P=0.03).
Fig 3Relative risk estimates of coronary heart disease events and stroke in single drug blood pressure difference trials according to class of drug (excluding CHD events in trials of β blockers in people with history of coronary heart (more ...) Drug comparison trials
Figure 4 for CHD and web figures 4a-e show the results of the drug comparison trials comparing each of the five main classes of drug with drugs from the other classes. The summary relative risk estimates for CHD events were close to 1.0, indicating no advantage of any one drug over others in the prevention of CHD. The differences between classes of drug in average blood pressure reductions were close to zero (fig 4), and the differences in use of add-on drugs were negligible (0.03 or fewer drugs per participant).The different classes of drug therefore reduced blood pressure to about the same extent and reduced CHD to about the same extent, providing evidence of a lack of preventive effects attributable to mechanisms other than lowering blood pressure.
Fig 4Relative risk estimates of coronary heart disease events and stroke in 46 drug comparison trials comparing each of the five classes of blood pressure lowering drug with any other class of drug (excluding CHD events in trials of β (more ...)
In the drug comparison trials the overall risk reduction in CHD events with thiazides was similar to that of other classes of drug (fig 4). There was, however, an increased risk of sudden cardiac death from using thiazides in very high dose, concealed in the summary results because few of the thiazide trials used very high doses and because sudden cardiac deaths were a small proportion of all CHD events. In trials in which the thiazide dose was high (≥4 times the standard dose33
) there were 33 sudden cardiac deaths in participants allocated thiazides and 16 in those allocated other drugs (with similar numbers of participants in each group): relative risk 2.1 (P=0.01).w125 w163
In trials using doses between standard and twice standard, 57 and 40 sudden cardiac deaths occurred, respectively (relative risk 1.4),w15 w128 w130 w134
and in trials using around half standard dose there were 16 and 19 sudden cardiac deaths (relative risk 0.8;w127 w131 w133
; P for trend 0.058). These results indicate that higher doses of thiazides probably cause sudden death, and prospective observational studies of people taking and not taking thiazides show the same (attributable to lower serum potassium levels causing ventricular arrhythmias).34 35 36 37 38 39 40 41
Figure 4 for stroke and web extra figures 4f-j show the corresponding drug comparison trial results on stroke. The summary relative risk estimates for stroke in the drug comparison trials were close to 1.0, with two exceptions. Figure 4 suggests a greater preventive effect of calcium channel blockers than other drugs, and a lesser effect of β blockers. The greater preventive effect of calcium channel blockers (relative risk 0.91, 95% confidence interval 0.84 to 0.98; P=0.01) was not materially altered after adjustment for the small difference in blood pressure reduction between the two groups (relative risk 0.92, 0.85 to 0.98), and is equivalent to a reduction in risk of stroke of 33% rather than 27% (the overall summary estimate), since 92% of 0.73 (the average relative risk in the single drug trials, fig 3) is 0.67 and 1.0−0.67 is 0.33. The lesser effect of β blockers in preventing stroke (fig 4; relative risk 1.18, 1.03 to 1.36; P=0.02) was not materially altered by adjusting for the small average difference in blood pressure reduction between the randomised groups and is equivalent to a 19% reduction in risk of stroke rather than 27% (since 1.11×0.73 (average relative risk from fig 3) is 0.81 and 1.00−0.81 is 0.19). The observed lesser effect of β blockers, however, rested on trials comparing calcium channel blockers with β blockers.w136-w140 Exclusion of the results of these trials weakened the evidence favouring a disadvantage of β blockers over the three other classes (relative risk 1.11, 0.86 to 1.44; P=0.40) but had little effect on the strength of evidence favouring an advantage of calcium channel blockers over the three other classes of drug (relative risk 0.93, 0.86 to 1.01; P=0.07).
The drug comparison trial results in figure 4 were similar and not significantly different when subdivided into the three prespecified groups (no vascular disease on entry (34 trials), history of CHD (10), and history of stroke (2)), but there were too few trial data to identify any but quite large differences in this respect.
Should the use of blood pressure lowering drugs be limited to people with “high” blood pressure?
The relative risk estimates of CHD events and stroke in the blood pressure difference trials were similar across all levels of blood pressure before treatment down to 110 mm Hg systolic and 70 mm Hg diastolic, below which there were too few data (fig 5). At each blood pressure level the relative risk reductions were statistically significant and consistent with the summary relative risk estimates for all the trials of 0.84 for CHD events and 0.70 for stroke (table 2, also see web extra figures 5a-l and 6a-m for forest plots of the individual trial results). A meta-regression analysis showed no significant trend in proportional disease reduction with lower blood pressures before treatment indicating a constant proportional effect. The trial results mirror those in cohort studies,25 26 27
which show a proportional reduction in risk that is constant over all measured levels of blood pressure—that is, the same in people with lower and higher blood pressures.
Fig 5Relative risk estimates of coronary heart disease events and stroke in blood pressure difference trials according to pretreatment diastolic and systolic blood pressures (taken as average in placebo group over course of trial). (Totals are (more ...)
There was no heterogeneity across the relative risk estimates for CHD according to pretreatment diastolic blood pressure (χ2=3.9, df=6, P=0.69; fig 5). There was heterogeneity for stroke (χ2=19, df=6, P=0.004), owing to a greater risk reduction in trials with the highest blood pressure before treatment (≥95 mm Hg), which arose because of more intensive treatment in these trials (average difference between treated and placebo groups of 1.7 blood pressure lowering drugs per participant, compared with 1.0 drugs per participant in the remaining trials with lower pretreatment blood pressure). The same applied to the analysis based on systolic blood pressure (CHD, χ2=3.7, df=7, P=0.82 and stroke χ2=12.24, df=6, P=0.06; fig 5).
What is the quantitative effect of one or more blood pressure lowering drugs in lowering blood pressure and in preventing CHD events and stroke?
The effect of taking blood pressure lowering drugs in reducing the incidence of CHD and stroke according to number of drugs used, dose of drugs, and age cannot be estimated from the blood pressure differences trials alone. This is because a quarter of treated participants stopped taking their allocated drugs, individual trials used varying doses of drugs, trial data on combination drug therapy were limited, and the age range was relatively narrow. All this can be overcome by doing a two stage analysis, which is set out in figure 6 (based on diastolic blood pressure) and figure 7 (based on systolic blood pressure). The first stage (upper portion of figures) was to estimate the effect of one drug at standard dose in reducing blood pressure according to the level of blood pressure before treatment, which was done using results from a meta-analysis of short term (a few weeks) placebo controlled trials of the drugs in fixed dose.32
Because of their short duration these trial results have the advantage of not underestimating the effects of blood pressure reduction through treated participants dropping out or controls receiving treatment. The second stage (lower portion of figures) was to estimate the effect of these blood pressure reductions in preventing CHD and stroke. This was done using the results of the meta-analysis of cohort studies,25
rather than those from our meta-analysis of the trial results, because the cohort studies quantified the effects on disease reduction across a wide range of blood pressure reduction and age (which the trials cannot do), and the evidence presented here shows that the cohort study results reliably predict the results of randomised trials over the ranges of age and blood pressure reduction observed in the trials, so validating the use of the cohort study data in this way.
Fig 6Reduction in incidence of coronary heart disease (CHD) events and stroke in relation to reduction in diastolic blood pressure according to drug dose, number of drugs, pretreatment diastolic blood pressure, and age. *Blood pressure reductions (more ...)
Fig 7Reduction in incidence of coronary heart disease (CHD) events and stroke in relation to reduction in systolic blood pressure according to dose and combination of drugs, pretreatment systolic blood pressure, and age. *Blood pressure reductions (more ...)
Figures 6 and 7 show estimates of the effects of three drugs in combination at half standard dose. Tables 3 and 4 show the estimates of the effects of one, two, and three drugs at standard dose and one, two, and three drugs at half standard dose in preventing CHD and stroke according to pretreatment blood pressure and age.
Table 3 Estimates of preventive effect of taking one or more blood pressure lowering drugs on coronary heart disease (CHD) events and stroke according to pretreatment systolic blood pressure, age, number of drugs, and dose (as multiple of standard33 (more ...)
Table 4 Estimates of preventive effect of taking one or more blood pressure lowering drugs on coronary heart disease (CHD) events and stroke according to pretreatment diastolic blood pressure, age, number of drugs, and dose (as multiple of standard33 (more ...)
These estimates of the preventive effect of blood pressure lowering drugs were calculated independently of the trials of blood pressure lowering drugs, so the trial results could be used to validate them. Table 5 shows the predictions of the reductions in CHD events and stroke in the blood pressure difference trials of single drug therapy (1.0 drug per participant on average) and of combination drug therapy (2.0 drugs per participant), calculated taking into account blood pressure before treatment, dose as a multiple of standard dose, and age and adjusted for the fact that a quarter of participants did not take their allocated treatment. Table 5 shows that the observed reductions in CHD events and stroke in the trials are similar to the predicted values, so the trial results validate the estimates.
Table 5 Observed percentage reductions in coronary heart disease (CHD) events and stroke in single and combination drug treatment blood pressure difference trials compared with predicted reductions according to number of drugs, dose, pretreatment blood (more ...)
Figures 6 and 7 show that one drug at standard dose reduced the incidence of CHD by about 24% and of stroke by 33% in people aged 60-69 with a systolic blood pressure of 150 mm Hg and a diastolic blood pressure of 90 mm Hg. Three drugs at half standard dose about doubled this effect, reducing the incidence of CHD by about 45% and of stroke by 60%. At higher blood pressure (180/105 mm Hg) and at lower blood pressure (120/75 mm Hg) the effect of one drug at standard dose was about 7-9 percentage points greater and smaller, respectively, and of three drugs at half standard dose about 12-14 percentage points greater and smaller. The proportional effect of age was relatively small; in people 10 years older the effect of one drug at standard dose was only 3 percentage points lower on average, and of three drugs at half standard dose 5 percentage points lower. Because mortality from CHD and stroke approximately trebles with each 10 year increase in age, the absolute gain from blood pressure reduction was greater at older ages.
If the drug treatment included a calcium channel blocker, and the greater effect of calcium channel blockers in preventing stroke (fig 4) was real and causal, the relative risk estimates for stroke in figures 6 and 7 and in tables 3 and 4 should be reduced by 8% (multiplied by 0.92).
Table 6 shows the results on heart failure (17
872 episodes), recorded in 64 blood pressure difference trials and 31 drug comparison trials. Heterogeneity existed within the results of the trials of β blockers and heart failure (P=0.008), explained by the observation that β blockers without cardioselective or α blocking (vasodilatory) properties (propranolol, oxprenolol, pindolol, and sotalol) lacked a preventive effect on heart failure (relative risk 1.01, 95% confidence interval 0.76 to1.35), but β blockers with one or other of these properties had a preventive effect (relative risk 0.77, 0.69 to 0.87; P=0.01 for difference). Data from trials of the first category of β blockers (seven trials, 385 episodes) were therefore excluded from table 6.
Table 6 Results on heart failure in 64 blood pressure difference trials and 31 drug comparison trials of blood pressure lowering drugs, separately for calcium channel blockers and other drugs
Calcium channel blockers reduced heart failure in the blood pressure difference trials by 19% (P=0.007), although the drug comparison trials showed that they were statistically significantly less effective in doing so than the other four classes of drugs (relative risk 1.22, 1.10 to 1.35; P<0.001). Each of the other four classes of drug significantly reduced the incidence of heart failure in the blood pressure difference trials (P<0.001) by 24% on average, with no significant differences in effect between them either in the blood pressure difference trials or the drug comparison trials. The effect of calcium channel blockers in reducing heart failure in the blood pressure difference trials (19%) was therefore not much less than that of the other classes of drug (24%). The effect of the drugs was similar in primary and secondary prevention (preventing new diagnoses of heart failure and preventing deterioration (hospital admission or death) in people with heart failure).
Non-vascular mortality and all cause mortality
In the blood pressure difference trials there was no increase in cancer mortality (relative risk 0.96, 0.85 to 1.09) or in non-vascular mortality (relative risk 1.00, 0.94 to 1.06). There were statistically significant reductions in all cause mortality in all the blood pressure difference trials (relative risk 0.87, 0.84 to 0.90; P=0.001) and in trials of people with no vascular disease on entry (0.89, 0.84 to 0.95) and with a history of CHD (0.86, 0.81 to 0.90) and stroke (0.91, 0.83 to 1.01).