R Collins. S MacMahon

Transcription

1 British Medical Bulletin (1994) \ U 50, No 2, pp OThe British Council 1994 Blood pressure, antihypertensive drug treatment and the risks of stroke and of coronary heart disease R Collins MRC/ICRF Clinical Trial Service Unit, University of Oxford, UK S MacMahon Clinical Trials Research Unit, University of Auckland, New Zealand Most evidence about the effects of blood pressure on the risks of cardiovascular disease derives from two principal sources: prospective non-randomised observational studies of the associations between blood pressure and the incidence of stroke and of coronary heart disease, and randomised trials of antihypertensive drug therapy. The focus of the first part of this chapter concerns the evidence from observational studies, which - despite the possibility of confounding by other risk factors - may be more relevant to the eventual effects of prolonged blood pressure differences on stroke and coronary heart disease risk. 1 The focus of the second part concerns the evidence from randomised trials of antihypertensive drug treatment, which are more relevant to assessing how rapidly, and to what extent, the epidemiologically expected reductions in stroke or in coronary heart disease are produced by suddenly lowering blood pressure in middle and old age. 2

2 BLOOD PRESSURE AND ANTIHYPERTENSIVE DRUG TREATMENT EVIDENCE FROM PROSPECTIVE OBSERVATIONAL STUDIES Effects of prolonged blood pressure differences on stroke and on coronary heart disease The associations of blood pressure levels with the incidence of stroke and of coronary heart disease (CHD) have been investigated in a large number of prospective, observational studies involving populations from a variety of geographic regions. 3 " 11 Typically these associations have been estimated by relating the measured blood pressure of participants at the start of the study to the occurrence of stroke and CHD over subsequent years. In most studies, the participants were predominantly middle-aged or older when blood pressures were initially measured. Differences in blood pressure levels between participants at this age are likely, as a result of the phenomenon of blood pressure 'tracking', 12.' 3 to have existed for a decade or more prior to measurement and to have persisted during follow-up. Consequently, estimates of the differences in 'usual' blood pressure between participants in these studies can be made, and these can be used to provide information about the effects of prolonged 'usual' blood pressure differences on stroke and on CHD risk. 1 In general, however, analyses of most prospective observational studies have assessed not the effects of differences in 'usual' blood pressure, but rather those of 'baseline' blood pressures measured just at entry to the study. Such baseline blood pressure measurements are subject to substantial random fluctuations due both to the unreliability of the measurement process and to the frequent real, but transient, deviations of blood pressure from an individual's usual level. This largely random error in the estimation of the usual blood pressure of individuals results in systematic error (ie bias) in the estimation of the associations between usual blood pressure levels and the risks of stroke and of CHD. 14.' 5 Specifically, errors in the estimation of usual blood pressure levels result in substantial underestimation of the strength of these associations and, consequently, of the size of any difference in disease incidence associated with a given difference in blood pressure between individuals. The magnitude of this 'regression dilution' bias 1 can be assessed in several ways; however, it is perhaps most simply and clearly demonstrated by comparing mean blood pressure levels at the start of a study with mean levels observed during follow-up, for groups defined on the basis of their initial baseline blood pressure measurement. For example, among participants in the Framingham Study, 9 the average difference in diastolic blood pressure (DBP) between 5 consecutive groups categorised by baseline measurements was initially 11.4 mmhg whereas

3 274 HYPERTENSION after 4 years it was just 7.1 mmhg (Table 1). Similarly, the difference between the top and the bottom categories was initially 47.7 mmhg whereas after 4 years it was just 28.5 mmhg. These differences reflect the 'regression to the mean' of erroneously high or low initial blood pressure values. Analyses based on the baseline measurements alone would produce associations of disease risk with differences in blood pressure that were on average about 60% greater than the differences in blood pressure apparent after adjustment for regression to the mean. It is evident, therefore, that the true associations of risk with prolonged differences in usual DBP would be about 60% greater than those evident from analyses based only on crude initial blood pressure measurements. This can be graphically illustrated, using the blood pressure measurements and the relative risk of stroke in just the top and bottom categories of DBP in the Framingham Study (Fig. 1): it can be seen that the slope of the line joining these categories (and, as will be seen later, approximating the overall relationship of DBP with stroke) is about 60% steeper for the mean blood pressure values at 4 years compared to the mean values at baseline. The approximate size of this bias has been confirmed by the results of other studies 6-16 in which repeat measurements of blood pressure were made, and also affects estimates of the association of systolic blood pressure (SBP) with disease risks. These results suggest that the importance of blood pressure - both diastolic and systolic as a determinant of cardiovascular morbidity and mortality has been grossly underestimated in most previous analyses of prospective observational studies. The associations of usual levels of blood pressure (ie adjusted for regression to the mean) with the incidence of stroke and of CHD have been examined in a collaborative project 1 involving 9 prospective observational studies 3 -" and a total of individuals aged 25 years or older who were not known to have had a myocardial infarction, a stroke or treated diabetes prior to the study baseline examination (Table 2). The weighted average follow-up period for disease outcomes in these studies was 10 years. In all the studies, baseline blood pressure was measured at a single visit either from one reading or the average of two readings using a standard mercury sphygmomanometer. Participants in each study were arbitrarily divided into 5 groups according to their baseline DBP (<80, 80-89, 90-99, , >109 mmhg) and disease risks were calculated for each of these categories relative to the overall risk in the entire study population. For all studies combined, relative risks for each of the categories were calculated by logistic regression analyses with adjustment for study and gender. Average 'usual' (as opposed to 'baseline') blood pressure levels for each of the categories could not be calculated directly for all studies since follow-up

4 BLOOD PRESSURE AND ANTIHYPERTENSIVE DRUG TREATMENT 275 Table 1 Mean DBP (mmhg) at baseline and at examinations after 2 years and 4 years for 5 categories of baseline DBP in 3776 men and women in the Framingham Study Baseline DBP category: mm Hg No. of participants with repeat measurements Mean DBP in each category (and difference between adjacent categories) At baseline 2 years post-baseline 4 years post-baseline* 1: <79 2: : : : 110 Range of mean DBP Average difference between adjacent catego ries (12.9) (9.9) (9.9) (13.0) (7.3) (8.2) (8.0) (8.1) (7.7) (7-4) (7.2) (6.2) "Means of SBP 4 years post-baseline in baseline DBP categories 1-5 were 123.2, 136.0, 148.1, and mmhg, respectively. measurements of blood pressure were available for only a few of the study populations. For this reason, the 4 year post-baseline results of the Framingham Study (Table 1) were used as a guide to usual diastolic and SBP levels in the other studies. As noted above, data from two other studies with follow-up blood pressure measurements suggest that corrections for regression dilution bias based on the Framingham data are likely to be reasonably representative. Prolonged blood pressure differences and the primary incidence of stroke Data on death from stroke were available from 7 of the 9 studies and data on non-fatal stroke were available from two studies. In total, there were 843 strokes recorded in these studies (599 fatal strokes and 244 non-fatal strokes). The relative risks of stroke in the 5 categories of baseline DBP are shown in Figure 2 for the 7 studies in combination. While the 95% confidence limits for the estimates of relative risk in most of the individual studies were wide, 1 the confidence limits for the combined study data are much narrower and suggest an approximately constant relationship between the risk of stroke plotted on a logarithmic scale and blood pressure. This roughjy 'log-linear' relationship suggests that the proportional difference in stroke risk associated with a given

5 276 HYPERTENSION Baseline DBP Category 5 (i110mmhg) Relative Risk of Stroke Baseline DBP Category 1 (479 mmhg) Mean DBP at baseline Mean DBP after 4 years Average diastolic blood pressure 120 mmhg Fig. 1 The relative risks of stroke for individuals from the Framingham Study 9 in baseline DBP categories 1 and 5 plotted against average DBP at baseline and after 4 years (see Table 1). difference in blood pressure is similar at all levels of blood pressure within the range studied. Neither the individual study results nor the combined study results suggest that there is any threshold of DBP below which lower blood pressure levels are no longer associated with lower relative risks of stroke. However, so few strokes were observed in the lowest blood pressure category that it was not possible to assess reliably the relationship of blood pressure to stroke risk at levels of usual blood pressure below 76 mmhg diastolic (and 123 mmhg systolic). Nevertheless it is apparent that among most individuals who would typically be classified as 'normotensive', blood pressure levels were directly and steeply related to stroke risk. Indeed, the strength of the association in such individuals was not significantly different from that in those who would generally be classified as 'hypertensive' (ie 155/95 mmhg or greater). From these analyses (and from similar analyses adjusted for age, blood cholesterol and smoking 1 ), it is evident that small but prolonged differences in blood pressure have marked effects on the risk of stroke.

6 BLOOD PRESSURE AND ANTIHYPERTENSrVE DRUG TREATMENT 277 Table 2 Characteristics of the study populations of 9 prospective observational studies of stroke and CHD incidence Study No of individuals Age range at baseline % Male Follow-up duration (years) after DBP measurement MRFTT screenees Chicago Heart Whitehall Puerto Rico Honolulu LRC Prevalence Framingham* Western Electric People's Gas All studies The Framingham analyses are based on three successive 6 yr follow-up periods. People who, at the start of a particular period, did not then have a history of myocardial infarction or stroke were classified by the DBP at baseline in that period. (NB Many people contribute to more than 1 period, which more than doubles the apparent number of individuals in Framingham). A 5 mmhg lower DBP is associated with about a one-third lower risk of stroke whereas a 10 mmhg lower DBP is associated with more than a one-half lower risk of stroke. The strength of these associations was not clearly different in men and in women, although far fewer women were included in the studies. Nor were the associations with fatal and with non-fatal stroke clearly different, but only two studies provided data on non-fatal stroke. Prolonged blood pressure differences and the primary incidence of CHD Data on death from CHD were available from all 9 studies and data on non-fatal myocardial infarction were available from three of these studies. In total, there were 4856 coronary events recorded in these study populations (4260 coronary deaths and 596 non-fatal myocardial infarctions). The relative risks of CHD in the 5 blood pressure categories are shown in Figure 2 for all 9 studies combined. Once again, while the 95% confidence intervals for the estimates of relative risk were wide in many of the individual studies, 1 the confidence intervals for the combined study data are very narrow and the results suggest a constant, roughly log-linear, relationship of CHD risk with blood pressure. The slope of the association with CHD appeared to be about twothirds as steep as that with stroke. Once again, neither the individual study results nor the combined results indicate a threshold below which lower blood pressure levels are no longer associated with lower risks

7 Relative Risk of 1 oo Stroke 1UU Stroke and Usual BP (in 5 categories defined by baseline DBP) 7 prospective observational studies: 843 events t w t Relative om by guest on July 14, 2016 Coronary Heart Disease and Usual BP (in 5 categories defined by baseline DBP) 9 prospective observational studies: 4856 events Baseline DBP category Usual SBP Usual DBP Approximate mean usual BP (estimated from tater remeasurements in the Framlngham study) mmhg Baseline DBP category Usual SBP Usual DBP Approximate mean usual BP (animated from later remeasuroments In the Framingtiam study) Fiy. 2 The relative risks of stroke and of coronary heart disease for 5 categories of blood pressure Irom the combined results of prospective observational studies Estimates ol usual systolic and diastolic blood pressure are taken from the average values 4 years after baseline in the Franiingham Sludy (.see Table I). Solid squares represent disease risks in each category relative to risk in the whole study population; sizes of squares are proportional lo number of events in each DBP category: and 95% confidence intervals for estimates of relative risk are denoted by vertical lines mmhg

8 BLOOD PRESSURE AND ANTWYPERTENSrVE DRUG TREATMENT of CHD. Moreover, in the combined analyses, when those in the lowest blood pressure category were further subdivided by baseline DBP (<70 mmhg or mmhg), the risk of CHD among those in the lower part was significantly less (2P <0.01) than that among those in the higher part. 17 Thus for CHD, as for stroke, blood pressure was directly and continuously related to risk throughout the range of usual blood pressure studied. Consequently, the strength of the association in individuals who would usually be classified as 'normotensive' was not clearly different in those who would usually be classified as 'hypertensive'. From these analyses (and from regression analyses adjusted for other risk factors for CHD 1 ) it would appear that small but prolonged differences in blood pressure have important effects on the risk of CHD, although diese are of somewhat smaller relative magnitude (but not necessarily smaller absolute magnitude) than the effects on stroke risk. A 5 mmhg lower DBP is associated with about a one-fifth lower risk of CHD whereas a 10 mmhg lower DBP is associated with more than a one-third lower risk of CHD. For CHD, as for stroke, the strength of these associations was not clearly different in men and in women, or for coronary death and for non-fatal myocardial infarction. Prolonged blood pressure differences and the secondary incidence of stroke and CHD The foregoing evidence of continuous, direct associations between usual blood pressure levels and the incidence of stroke and of CHD derives entirely from studies of individuals, most of whom did not have diagnosed cerebrovascular or CHD and all of whom had no recorded history of stroke or myocardial infarction prior to study entry. Few data are available from prospective observational studies on the relationship between blood pressure and recurrent stroke among individuals with a history of cerebrovascular disease. However, unpublished analyses from at least one study 18 suggest that after minor stroke or transient ischaemic attack, the risk of recurrent cerebrovascular events remains directly and continuously related to blood pressure levels throughout the entire distribution of blood pressure. In contrast, the observed association between blood pressure and recurrent coronary events after myocardial infarction appears to be 'J-shaped', with an increased risk among individuals with low blood pressure. 19 However, since myocardial infarction (and subsequent left ventricular dysfunction) can both lower blood pressure and, independently, increase the risk of further coronary events (particularly coronary death), this 'J-shaped' association may largely reflect 'confounding' rather than a causal relationship. 20 For even if there was a direct causal relationship between blood pressure and the secondary incidence of CHD, an artefactual inverse association,

9 280 HYPERTENSION particularly at the lower end of the blood pressure distribution, might be observed, reflecting the low blood pressure of those with poor ventricular function, whose risk of coronary death is high. This is supported by analyses among the 5440 men with a history of myocardial infarction screened for the Multiple Risk Factor Intervention Trial, 3 in whom a J-shaped association of coronary death with DBP was observed. Mortality from CHD among those with uncorrected baseline DBP in the lowest quintile (<76 mmhg) was about one-fifth greater than that among those with pressures in the second lowest quintile (76-80 mmhg). The direction of this association at these low levels of blood pressure was clearly different to that observed in the same study among the men without a history of myocardial infarction. However, recent analyses from this study (J Flack, personal communication) indicate that this difference is likely to be a consequence of the effects of CHD on blood pressure in survivors of myocardial infarction, rather than the effects of blood pressure on CHD. These analyses show that the J-shape relationship between blood pressure and coronary death persists for only a few years of follow-up; thereafter, a more linear relationship becomes apparent with lower coronary death rates experienced by those with the lowest blood pressure. The short-term nature of the increased risk among those with particularly low blood pressures after myocardial infarction strongly suggests that while low blood pressure may indicate a poor short-term prognosis, it is not itself likely to be a cause of the increased risk of coronary death. Systolic versus DBP Data from the Framingham Study (Table 1) and many others indicate that there is a high degree of co-linearity between systolic and DBP. Typically, the correlation between single measurements of SBP and DBP is about (This is about the same as the correlation between repeated measures of either SBP or DBP.) Inevitably the correlation between usual systolic and usual diastolic blood pressure will be even greater, since it will not be diluted by the random error that affects measurements of blood pressure made on a single occasion (see above). Within the range of usual blood pressures commonly observed in the Framingham Study (diastolic mmhg, systolic mmhg), there was on average a 1.86 mmhg increment in usual SBP for every 1 mmhg increment in usual DBP. Hence the difference in disease risk associated with a 5 mmhg difference in usual DBP was also associated with a 9 mmhg difference in usual SBP. Clearly both these blood pressure differences may have contributed to the difference in disease risk. However, the very high degree of co-linearity between systolic and diastolic blood pressure makes it very difficult to determine

10 BLOOD PRESSURE AND ANT1HYPERTENSIVE DRUG TREATMENT 281 the independent contribution of each of these components of blood pressure with disease risk. Previous efforts to delineate the independent contributions of SBP and of DBP have focussed entirely on crude unadjusted baseline blood pressure measurements from prospective observational studies. 3 ' 21 In general, these analyses have suggested that SBP measured on a single occasion confers additional prognostic information to that conferred by DBP measured on a single occasion. Unfortunately, however, it is not possible to determine whether measurements of SBP would confer important additional prognostic information to that conferred by a more accurate estimate of usual DBP (which would itself increase by 60% the prognostic value of the blood pressure measurement). It is of course possible that among individuals with the same usual DBP, the level of usual SBP could be an independent determinant of the risks of stroke and of CHD. Similarly the level of usual DBP could well be a determinant of risk among those with the same usual SBP. However, to determine reliably the independent and additional contributions, if any, of these components of blood pressure would require very large scale prospective, observational studies (ie involving several hundred thousand individuals) with repeated measurements of blood pressure for all participants. Summary and Implications of the observational study results: benefits of lower blood pressure for 'hypertensive' and 'normotensive' individuals The available data from prospective observational studies of individuals without a history of serious cardiovascular disease indicate that usual levels of blood pressure are directly and continuously related to the risks of stroke and CHD. The strength of these associations has been substantially underestimated by many previous analyses that have not taken account of the unreliability with which blood pressure levels are typically assessed in such studies. Correction for this regression dilution bias increases by about 60% the strength of the associations between levels of DBP and the risks of stroke and of CHD. After appropriate adjustment for this bias, the data from prospective observational studies indicate that a prolonged difference in usual DBP of just 5 mmhg is associated with about a one-third difference in stroke risk and a one-fifth difference in CHD risk. For neither stroke nor CHD was any threshold of blood pressure identified below which cardiovascular disease rates did not continue to decline, and the associations between blood pressure and the risks of stroke and CHD were of similar strength in individuals who might be classified as 'hypertensive' and in those who would usually be consid-

11 282 HYPERTENSION ered 'normotensive'. In Western populations, about three-quarters of all cardiovascular events occur in 'normotensive' individuals (since they make up the large majority of these populations), and so the implications of these analyses are that the burden of blood pressure-related risk is actually greater in 'normotensives' than 'hypertensives'. While very low blood pressures appear to be predictive of reduced short-term survival in individuals with a history of myocardial infarction, there is increasing evidence that this reflects the independent effects of serious CHD both on blood pressure and on survival, rather than any direct adverse effect of low blood pressure on outcome. The strength and consistency of the associations between blood pressure and the primary incidence of stroke and of CHD are suggestive of a direct causal relationship between the level of blood pressure and the occurrence of these diseases. This is further supported by the apparent independence of these associations from the effects of potential confounding factors such as age, blood cholesterol and smoking. 1 While it is possible that some other factor such as obesity 22 or diabetes 23 could predispose both to blood pressure elevation and to stroke and CHD, it is unlikely that such confounding could explain much of the observed relationships between blood pressure and these cardiovascular diseases. There is a strong biological rationale for expecting blood pressure levels to have a direct influence on the development of both cerebrovascular 24 and CHD. 25 ' 26 Moreover strong evidence of causality in the relationships between blood pressure and cardiovascular disease is provided by the results of randomised trials of blood pressure reduction (see below). The results of the prospective observational studies suggest, therefore, that prolonged changes in the blood pressure of populations or individuals are likely to alter the incidence or risks of cardiovascular disease. For large populations in which cardiovascular diseases are common, it is apparent that a relatively small change in the average adult blood pressure could have large effects in terms of the number of serious cardiovascular events prevented. In China, for example, a decline of 5 mmhg in the average adult blood pressure could be expected to prevent about a million deaths from stroke and CHD each year. For individuals, the benefits of blood pressure reduction are likely to be greatest in those at highest risk of serious potentially preventable events, such as those with a history of cardiovascular disease. Since the available data suggest that blood pressure is similarly related to cardiovascular outcome in 'normotensives' as well as 'hypertensives', it is possible that the absolute benefits of a given reduction in blood pressure would be greater for 'normotensive' individuals with a history of cerebrovascular or CHD than for individuals with uncomplicated hypertension (see below).

12 BLOOD PRESSURE AND ANT1HYPERTENSIVE DRUG TREATMENT EVIDENCE FROM RANDOMISED CONTROLLED TRIALS OF ANTIHYPERTENSIVE DRUG THERAPY Effects of short-term blood pressure reductions on stroke and on CHD Appropriate analysis of prospective observational studies, making due allowance for the substantial and systematic extent to which the true relation is diluted by purely random fluctuations in the 'baseline' measurements (the 'regression dilution' bias), suggests that a prolonged difference of only 5 mmhg in usual DBP is associated with avoidance of at least one-third of the risk of stroke and at least one-fifth of the risk of CHD. 1 However, even a causal relation between blood pressure and risk does not imply that a blood pressure reduction that is maintained for only a few years would confer the full protective effect of a long-term difference in usual blood pressure. For example, chronic hypertension might accelerate the rate of development of some chronic processes, such as atherosclerosis. If so, then the association between the risk of vascular disease and the usual blood pressure may, at least to some extent, reflect the cumulative effects of blood pressure differences that have persisted for many years. Over the past few decades, numerous studies of the treatment of hypertension have been conducted to determine whether blood pressure reduction in middle age reduces the risk of stroke and of CHD. In early uncontrolled studies in malignant hypertension, the benefits of antihypertensive drugs appeared to be so great that randomised trials were not undertaken, and the value of pharmacological therapy for this particular condition gained rapid, widespread acceptance. The benefits of treating severe but non-malignant hypertension were somewhat less rapidly established, but after some randomised, (and some alternate-allocation 35 ) trials in the 1960s it became generally accepted that the treatment of severe hypertension could at least prevent stroke. However, despite several subsequent randomised studies, there has been controversy as to whether the treatment of severe hypertension can also prevent CHD, 36 while for the treatment of mild-to-moderate hypertension questions have even been asked 37 about the effects on stroke. This continuing uncertainty about the benefits of lowering blood pressure has been due at least partly to chance differences between the results of particular trials. Several reviewers have tried to explain the differences between different trial results in terms of design features and study populations, but perhaps those differences chiefly reflect the inability of individual trials, even those with several hundred CHD events, to detect moderate CHD reductions reliably, rather than any important heterogeneity of the real effects of treatment. The mean DBP

13 284 HYPERTENSION difference between treatment and control groups in the trials was only about 5-6 mmhg, and the epidemiological evidence suggests that a long-term difference of this magnitude is associated with about 20-25% less CHD. Even if such trial treatments would eventually, after many years, produce 20 25% less CHD, the effects seen within the 2 or 3 years that is available on average between randomisation and death in a 5-year trial might well be somewhat smaller (eg 10-15%). Considered separately, none of the trials recorded enough CHD events (or enough vascular deaths) for statistically reliable assessment of 10-15% risk reductions. Overview of results of trials of antihypertensive drug therapy Reliable detection or refutation of treatment effects that are only moderate in size requires reliable exclusion both of moderate biases and of moderate random errors. Both requirements are reasons for seeking a proper, systematic overview of the results of all the 'unconfounded' randomised trials 42 " 44 (ie one that includes all randomised trials except those in which any effects of lowering blood pressure are deliberately confounded by differences between the treatment and control groups in some other risk factor interventions). First, without a systematic search for all relevant trials, an unrepresentative selection of the trial results may be reviewed. The resultant selection biases might not matter much if treatment had a large effect, but these and other selection biases (especially those that can be produced by unduly data-dependent emphasis on particular subgroups) do matter if moderate treatment effects are to be assessed reliably. Second, unless comparisons of the effects of treatment are based on trials that together include several hundred (or, preferably, a few thousand) CHD events, the random play of chance can produce favourable or unfavourable random errors that are comparable in size with the sort of moderate effects that might well need to be detected or refuted. Such large numbers of CHD events are difficult to achieve in individual trials but may at least be approached in an overview of the results of many different trials. The present overview updates an earlier report. 2 It includes all available data from all unconfounded randomised trials of antihypertensive therapy and involves about individuals, with an average of about 5 years of follow-up. By comparing the unconfounded trial results with the epidemiological evidence, it may be possible to estimate the extent to which the eventual effects of a lower blood pressure on disease incidence rates can be achieved within just a few years. Consideration of overviews may help not only in interpreting the previous trials but also in planning the size and duration of any future antihypertensive trials, since the epidemiological evidence provides ap-

14 BLOOD PRESSURE AND ANTIHYPERTENSIVE DRUG TREATMENT proximate upper limits to the risk reductions that could plausibly be expected in the later years of such trials. Characteristics of the trials and their participants 4 large and 13 small unconfounded randomised trials of pharmacological treatment, reported between 1965 and 1992, are included in this overview (Table 3a & Table 3b: see Ref. 2 for further details). Of these, 4 included only people with 'mild' hypertension (DBP <110 mmhg) at presentation ' and a further 5 included only people with 'mild-to-moderate' hypertension (DBP ^115 mmhg).33, ,58,64,65 The HDFP study was reported in such a way that people with DBP <110 mmhg and mmhg could be examined separately from those with DBP >115 mmhg, but in the remaining trials this was not possible. One study included only patients with isolated systolic hypertension. 63 The sample sizes in these studies ranged from under 100 to over 17000, with a total of individuals studied overall. Table 3A Population and trial design in recent and previous unconfounded randomised trials of at least 1 year of antihypertensive drug treatment (see also Table 1 of Ref. 2) Trial (ref) HDFP (48-52) MRC younger adults (56) SHEP (63) MRC older adults (64) STOP-H (65) No. of patients Entry DBP and/or SBP DBP 290 DBP of DBP <90 with SBP of DBP<115 and SBP of DBP of , or DBP of with SBP of Mean age (years) Male (%) smaller trials 8614 DBP Total: all trials Several different antihypertensive drug regimens were tested. In general, a 'stepped care' approach was used - increasing doses of the first-line drug, and introduction of second and third line drugs if DBP

15 286 HYPERTENSION Table 3B Trial (ref) Mean follow-up (years) Blinding Main drugs Mean DBP difference (mm Hg) in attenders* HDFP (48-52) 5.0 None CD 5 MRC younger adults (56) 5.0 Single BFor PR 6 SHEP (63) 4.5 Double CD+AT (or RE) 4 MRC older adults (64) STOP-H (65) 12 smaller trials Total: all trials Single Double - - AT or HZ AT or ME or PI or HZ+AM DI or BB *The difference in mean DBP per person-year of follow-up, based on data from those who attended follow-up for blood pressure measurement, was 6 mmhg. The difference between all those allocated treatment (irrespective of compliance) and all those allocated control is likely to have been somewhat smaller - eg 5-6 mmhg. AM=Amiloride, AT=Atenolol, BB=Beta-blocker-based regimens, BF=Bendrofluazide, CD=Chlorthalidone, DI=Diuretic-based regimens, HZ=Hydrochlorothiazide, ME=Metoprolol, PI=Pindolol, PR=Propranolol, RE=Reserpine. had not come down to pre-specified levels. The first-line drug was generally a diuretic, the only exceptions being the beta-blocker parts of the MRC trials and the STOP-Hypertension trial, 65 the atenolol trial in the elderly, 61 and a small trial of reserpine. 34 The usual aim was to achieve a DBP of 90 mmhg or less in those allocated antihypertensive treatment. Advice on the use of non-drug antihypertensive treatment (for example, salt restriction) was given to both groups in the ANBPS 55 and HSCSG 57 but to the active treatment group only in the HDFP Placebo was used in the control groups of 11 of the studies, while in the others the control groups received usual care. About one-quarter of all patients in the control groups received antihypertensive drug treatment at some stage during follow-up. Follow-up generally continued for 4 6 years, although the MRC trial in older adults and the STOP-H trial had longer follow-up, while 4 small trials had shorter follow-up: the overall mean was 4.9 years (so, people who died in these trials would have been on treatment for an average of only 2-3 years) *

16 BLOOD PRESSURE AND ANTIHYPERTENSIVE DRUG TREATMENT Effects of treatment on blood pressure For each trial (or trial stratum) the average DBP difference between those allocated treatment and those allocated control has been estimated from the published reports (Tables 3A & B). Among those attending follow-up for blood pressure measurement, these differences varied widely, with an average (weighted to reflect the person-years in each trial) of about 6 mmhg. This estimate excludes those who lapsed from follow-up before the end of the trial (but from whom endpoints were still included). If it is assumed that among these non-compliers there was little difference in DBP between the treatment and control allocated individuals, then the average difference in DBP between all those allocated treatment (irrespective of compliance) and all those allocated control is likely to have been somewhat less - perhaps 5 6 mmhg. Information on SBP during follow-up was available from some trials but not from others. Where available, it suggests that the absolute difference in SBP between those individuals allocated treatment and those allocated control is likely to have been about twice as great as the absolute difference in DBP. In many of the smaller studies no direct evidence was available about any effects of treatment on other risk factors, such as blood cholesterol (but see below). Effects of blood pressure reduction on stroke The overview of the results of all trials indicates a highly significant reduction in stroke of 38% SD 4 (95% confidence interval: 31 45%; IP < : Table 4 and Fig. 3), with similar reductions in fatal and in non-fatal strokes (each IP < : Table 4 and Fig. 4). Overall, 525 treatment allocated patients suffered a stroke as against 835 control patients. Significant reductions in stroke were observed in each of the four large trials and in the 13 combined trials (Fig. 3). Although the effect in the MRC older adults study was slightly smaller than average, a test of heterogeneity between the 5 results of HDFP, MRC younger adults, SHEP, MRC older adults and an aggregate of all the smaller trials yields a completely non-significant result (chi-square = 4.2 on 4 degrees of freedom; NS). It has been suggested that differences in stroke mortality and morbidity between treatment groups in those studies that did not use placebo control may have been due at least partly to differences in medical care other than the randomly allocated treatments. Any such biases would, however, appear to be small in comparison with the effects being measured, for even after exclusion of such studies the effect on stroke is of similar magnitude, and remains highly significant.

17 288 HYPERTENSION STROKE AND CHD IN HDFP, MRC, SHEP AND 13 SMALLER ANTIHYPERTENSIVE TRIALS (mean DBP difference 5-6 mm Hg for 5 years) TRIAL (or group of trials) NUMBERS OF EVENTS TREAT: CONTROL ODDS RATIOS & 95% confidence Omit* (TREAT : CONTROL) TREATMENT - BETTER TREATMENT WORSE HDFP trial MRC trial SHEP MRC trial 13 others ALL TRIALS (HeterogeneityX* -4Z NS) HDFP trial MRC trial SHEP MRC trial 13 others (II) CHD EVENTS 275: : : : :226 ALL TRIALS 934:1104 (Heterogeneity X* NS) 38% SO 4 reduction 2P < %SD4 reduction 2P = Difference In risk associated epldemlologlcalry with a LONG-TERM drfference of 5-8 mm Hg DBP: 0-5 STROKE 35-40% 10 CHD 20-25% Fig. 3 Reductions in the odds of stroke and of CHD in the HDFP trial, SHEP trial, MRC trials, and in all 13 other smaller unconfounded randomised trials (including STOP-H) of antihypertensive drug treatment (mean DBP difference of 5-6 mmhg for 5 years). Solid squares represent the odds ratios (treatment:control) for the 4 larger trials and the properly stratified odds ratio for the combination of the 13 smaller trials. The sizes of the squares are proportional to the amount of 'information' contributed by the study, 2 and 95% confidence intervals are denoted by honzontal lines (for individual large trials or the combined small trials) and by diamonds (for overviews of all trials).

18 BLOOD PRESSURE AND ANTIHYPERTENSIVE DRUG TREATMENT SB = fatal events T = treatment C = control 1104 ALL AVAILABLE EVIDENCE FROM RANDOMISED ANTI- HYPERTENSIVE DRUG TRIALS (mean DBP difference 5-6 mm Hg for 5 years) ! o " (5 % reduction In odds: No. of SD: 2P-vaJue: STROKE 38%SD4 8-7 < i: j- CHD %SD REMAINING ALL ALL VASCULAR VASCULAR OTHER DEATHS * DEATHS * DEATHS 4-8 < * Indudat my owtittmm unknown rtintt Fig. 4 Crudely summated results of the unconfounded randomised trials of antihypertensive drug treatment All were evenly randomised: total individuals, mean DBP difference during follow-up of 5-6 mmhg, mean time from entry to vascular event 2-3 years. Effects of blood pressure reduction on CHD The overview of all available trials also indicates a highly significant reduction in CHD. There was a 16% SD 4 reduction (95% confidence interval: 8-23%; IP = : Table 4 and Fig. 3) among those allocated antihypertensive therapy, with similar reductions in fatal CHD and in non-fatal myocardial infarction (each 2P <0.01: Table 4 and Fig. 4). Overall, 934 treatment-allocated patients suffered an episode of CHD as against 1104 control patients. As is the case for stroke, there is no significant heterogeneity between the results of the 4 individual large

19 Table 4 Outcome data in recent and previous unconfounded randomised trials of antihypertensive drug treatment (see also Tables II-IV of Ref. 2) Trial No. randomized Stroke (fatal + not) CHD (fatal + not) Death (vascular + not) Active Control Active Control Active Control Active Control HDFP MRC younger adults SHEP MRC older adults* rom by guest on July 14, O STOP-H smaller trials Total: all trials Proportional reduction 95% confiden ce interval 2P-value ( two-sided) 38% SD 4 31% to 45% < % SD4 8% to 23% = % SD 4 (vascular) 13% to 28% < *ln the MRC trial among older adults, 25% of people were lost to follow-up for non-fatal events (but not for fatal events, which were determined from central government records) during the 5.5 years of the study. See also footnote to Tables 2 and 3 of Ref. 2.

20 BLOOD PRESSURE AND ANTIHYPERTENSIVE DRUG TREATMENT 291 trials and the results from an aggregate of the 13 small trials about the overall 16% reduction (chi-square = 4.3 on 4 degrees of freedom; NS). The proportional reductions in CHD were smaller than those in stroke. Even so, since CHD was more common than stroke, the absolute reduction in CHD (crude total: 934 treatment vs 1104 control; difference of 170) was more than half as big as the absolute reduction in stroke (525 vs 835; difference of 310: Table 4 and Fig. 4). Effects of blood pressure reduction on mortality Most deaths were vascular, and most vascular deaths involved stroke or CHD. The proportional reduction in vascular mortality (21% SD 4: IP <0.0001) was, therefore, intermediate between the proportional reductions in stroke and in CHD mortality (Table 4 and Fig. 4). Non-vascular deaths were evenly distributed between the treatment and the control groups, so total mortality was also reduced significantly (2P = ). The epidemiological context: implications for stroke, CHD, and cause-specific mortality analyses The mean difference in DBP between treatment and control groups in these trials was only about 5-6 mmhg and, for those dying or experiencing a non-fatal stroke or myocardial infarction during the course of the trials, this difference had persisted for an average of only 2-3 years before the event (ie for about half the typical trial duration). Appropriate analysis 1 (see above) of prospective observational studies indicates that a prolonged difference of 5-6 mmhg in usual DBP would eventually be associated with differences of about 35-40% in the risk of stroke and of about 20-25% in the risk of CHD. If treatment could quickly produce effects of this magnitude then such benefits should be detected very reliably by an overview of total stroke incidence (involving over 800 strokes among the controls) and by an overview of total CHD incidence (involving over 1100 CHD events among the controls) in these 5-year trials. But if chronic processes are involved - either for stroke or for CHD - so that the eventual effects of a 5-6 mmhg difference in DBP are not fully seen during just the few years of treatment duration for those dying in the trials, then the power of even an overview of all the trials might be inadequate to detect benefit. Emphasis just on fatal strokes or on fatal CHD, rather than on all events (fatal or not), would substantially limit the ability even of an overview to detect such effects of treatment, while analyses of total mortality would be particularly unreliable and could well yield misleading conclusions. The potential unreliability of analyses of total mortality is troublesome, because an important aim of such trials is to discover reliably whether treatment can reduce total mortality. Paradoxically, however, direct assessment

21 292 HYPERTENSION of the effect of treatment on total mortality may be less reliably informative than indirect assessment of that effect, based on analyses of the proportional effects of treatment on total stroke and on total CHD, with confirmatory analyses of vascular and of non-vascular mortality (see below). The proportional reductions in stroke or in CHD may be somewhat different in different categories of patient, but such differences are difficult to assess directly, since the play of chance can produce surprisingly large irregularities in subgroup analyses Hence, data-dependent hypotheses, such as those in reports of the MRC study in younger adults^6-68 about large sex-specific, smoking-specific differences in the effects of particular blood pressure reductions, may well be misleading. Misleading patterns would be particularly likely to emerge if any of the less extreme results (such as those for CHD) were to be examined by data-dependent subgroup analyses to help decide which patients benefit most, and so no such analyses have been attempted in the present review. Stroke reduction: 31-45% in trials, and 35-40% in epidemiology For stroke, the overview provides direct and highly significant evidence that both fatal and non-fatal strokes are prevented within just a few years of blood pressure lowering (even if DBP is below 110 mmhg at the start of treatment 2 ). Even the lower confidence limit for the effects of treatment on stroke is similar to the difference of about 35 40% that corresponds in observational studies to a prolonged difference in usual DBP of 5-6 mmhg. This suggests that most or all of the stroke avoidance associated with a prolonged difference in usual DBP appears soon after the lowering of blood pressure. CHD reduction: 8-23% in trials, and 20-25% in epidemiology In contrast with the evidence for stroke, the significant reduction in CHD seen in the trials (16% SD 4; 95% confidence interval of 8-23%; IP = ) falls somewhat short of the difference of about 20 25% suggested by the observational epidemiological evidence for a prolonged 5 6 mmhg difference in usual DBP. This apparent shortfall might, of course, be due chiefly to the play of chance, for the upper 95% confidence limit for the significant CHD reduction seen in these trials is 23%. If the shortfall is not just a chance effect then it may indicate that some chronic processes (such as atherosclerotic changes accelerated by raised blood pressure) are important in mediating the effects of blood pressure on CHD, and/or that some treatments had important cardiotoxic side-effects that limited, but did not abolish, the CHD reduction. But, the trial results cannot be used as evidence for

22 BLOOD PRESSURE AND ANTIHYPERTENSIVE DRUG TREATMENT any such toxic effects, or even as evidence for any delay in the effects of blood pressure reduction on myocardial infarction. For, although this overview indicates that the CHD reduction produced within a few years of lowering blood pressure is only about two-thirds of the full effect to be expected eventually from the blood pressure reductions that were studied, the confidence interval for this fraction runs from a lower limit of one-third to an upper limit of one. Despite this quantitative uncertainty, however, it has been determined that the CHD reduction is substantial and real (2/> = ). Mortality reduction: vascular and non-vascular deaths This overview provides strong direct evidence of a reduction in vascular mortality {IP <0.0001) and in total mortality (2P = ), there being no apparent effect on non-vascular mortality. The size of the absolute reduction in vascular mortality (or in total mortality) depends not only on the sizes of the proportional reductions in stroke and in CHD, which might well carry over approximately to other quite different populations, but also on the absolute stroke and CHD death rates, which might well be very different in different populations. So, for example, in many Asian populations, where stroke rates are high but CHD rates are low, a halving of stroke rates could be of substantial importance for both vascular and total mortality, even if little reduction in CHD was achieved. Conversely, among people at high risk of CHD (such as MI survivors) even a modest reduction in CHD from antihypertensive therapy might be of substantial importance. Hence, it is perhaps not the overall conclusions for vascular (or total) mortality but the diseasespecific conclusions for stroke and for CHD that should be adopted from the present overview; these are summarised in Figures 3 and 4. Implications of the trial results: treatment of high-risk individuals From an affected patient's viewpoint, mild or moderate hypertension is less important than severe or malignant hypertension. In public health terms, however, the former may be more important, for it is much more common and so accounts for a greater proportion of the deaths and serious non-fatal vascular events associated with hypertension. 69 The overall results of the trials indicate that a 5 6 mmhg reduction in DBP maintained for a few years typically produced reductions of about 38% SD 4 in stroke and of about 16% SD 4 in CHD, with similar sized proportional reductions among trials in mild, moderate and severe hypertension. 2 Likewise, in the prospective observational studies (see above), both for stroke and for CHD, the proportional differences in risk associated with a given difference in DBP were similar among 'hypertensive' and 'normotensive' individuals. Actual compliance with the