Systematic review of randomised controlled trials of multiple risk factor interventions for preventing coronary heart diseaseBMJ 1997; 314 doi: http://dx.doi.org/10.1136/bmj.314.7095.1666 (Published 07 June 1997) Cite this as: BMJ 1997;314:1666
- Shah Ebrahim, professor of clinical epidemiologya,
- George Davey Smith, professor of clinical epidemiologyb
- a Department of Primary Care and Population Sciences, Royal Free Hospital School of Medicine, London NW3 2PF
- b Department of Social Medicine, University of Bristol, Bristol BS8 2PR
- Correspondence to: Professor Ebrahim
- Accepted 25 April 1997
Objective: To assess the effectiveness of multiple risk factor intervention in reducing cardiovascular risk factors, total mortality, and mortality from coronary heart disease among adults.
Design: Systematic review and meta-analysis of randomised controlled trials in workforces and in primary care in which subjects were randomly allocated to more than one of six interventions (stopping smoking, exercise, dietary advice, weight control, antihypertensive drugs, and cholesterol lowering drugs) and followed up for at least six months.
Subjects: Adults aged 17-73 years. 903 000 person years of observation were included in nine trials with clinical event outcomes and 303 000 person years in five trials with risk factor outcomes alone.
Main outcome measures: Changes in systolic and diastolic blood pressure, smoking rates, blood cholesterol concentrations, total mortality, and mortality from coronary heart disease.
Results: Net decreases in systolic and diastolic blood pressure, smoking prevalence, and blood cholesterol were 4.2 mm Hg (SE 0.19 mm Hg), 2.7 mm Hg (0.09 mm Hg), 4.2% (0.3%), and 0.14 mmol/l (0.01 mmol/l) respectively. In the nine trials with clinical event end points the pooled odds ratios for total and coronary heart disease mortality were 0.97 (95% confidence interval 0.92 to 1.02) and 0.96 (0.88 to 1.04) respectively. Statistical heterogeneity between the studies with respect to changes in mortality and risk factors was due to trials focusing on hypertensive participants and those using considerable amounts of drug treatment, with only these trials showing significant reductions in mortality.
Conclusions: The pooled effects of multiple risk factor intervention on mortality were insignificant and a small, but potentially important, benefit of treatment (about a 10% reduction in mortality) may have been missed. Changes in risk factors were modest, were related to the amount of pharmacological treatment used, and in some cases may have been overestimated because of regression to the mean, lack of intention to treat analyses, habituation to blood pressure measurement, and use of self reports of smoking. Interventions using personal or family counselling and education with or without pharmacological treatments seem to be more effective at reducing risk factors and therefore mortality in high risk hypertensive populations. The evidence suggests that such interventions implemented through standard health education methods have limited use in the general population. Health protection through fiscal and legislative measures may be more effective.
The effectiveness of health education approaches modifying lifestyle to prevent coronary heart disease is in doubt
Health promotion interventions result in only small changes in risk factors and mortality in the general population
In people with hypertension and in other high risk groups risk factor interventions have beneficial effects
Health protection by fiscal and legislative means deserves a higher priority
Primary prevention programmes in many countries attempt to reduce mortality and morbidity due to coronary heart disease through modifying risk factors.1 2 Randomised controlled trials of the efficacy of multiple risk factor intervention using counselling and education in addition to, or instead of, pharmacological treatments to modify major cardiovascular risk factors have been carried out in primary care and in the workplace. The findings of these trials have been equivocal; efficacy in reducing the incidence of disease seems to be associated with the degree of control achieved.3 4 Given the evidence from quasi-experimental studies, such as the North Karelia project5 6 and the Stanford heart disease prevention programme,7 8 9 multiple risk factor intervention using counselling and educational methods is widely believed to be efficacious and cost effective and worthy of expansion.
Such beliefs have been associated with the development of health promotion as a supposed specialty,10 the routine collection of data on cardiovascular risk factors in British primary care, and primary prevention policy–for example, the health of the nation strategy in England.11 More recent trials examining changes in risk factors have cast considerable doubt on the effectiveness of these multiple risk factor interventions12 13 and even interventions against smoking,14 15 prompting a review of the reasons for the frequent failure of such community experiments.16
Two previous reviews of multiple risk factor intervention using counselling and education with or without pharmacological treatments were conducted before the publication of more recent trials, were not systematic in their coverage, and did not formally aggregate the findings through meta-analysis.17 18 Consequently, we carried out a formal systematic review and meta-analysis examining the published randomised controlled trials and obtaining supplementary information when possible.
Randomised controlled trials of coronary heart disease prevention by specific interventions and multiple risk factor intervention were identified by a systematic Medline search strategy from 1966 to April 1995. The strategy used a series of terms to identify randomised controlled trials and then used topic terms to define the diseases of interest: coronary heart disease and stroke. Text word searches using the terms “prevention” and “multiple risk factor” were applied, and specific interventions of interest were searched for, comprising smoking cessation, dietary change, exercise, weight loss, blood pressure control, and cholesterol lowering. (A full search strategy is available from us.) This was supplemented by examining the reference lists of each of the randomised controlled trials identified, consulting with experts in the subject, and checking citations. The chief investigator of each trial was written to requesting relevant data not included in the published reports. A database of references was compiled using reference manager software.19
Randomised controlled trials of primary prevention of coronary heart disease by means of multiple (more than one) risk factor interventions using counselling and education with or without pharmacological treatments in general populations, occupational groups, and in high risk groups were included in the systematic review. Studies of children or only adults under 40 were excluded. Trials of secondary prevention were excluded. As primary prevention requires sustained changes in behaviour studies with follow up of less than 26 weeks were also excluded.
Odds ratios were used to summarise clinical event treatment effects and logistic regression by egret20 was used to provide pooled estimates for groups of trials. Risk factor effects were assessed as net changes–that is, the difference in the intervention group minus that in the control group–as this makes allowance for secular trends, measurement habituation effects, and regression to the mean. Net changes for continuous variables were pooled by using the standard deviations and sample sizes of intervention and control groups as weights, as described by Fleiss.21 Changes in smoking were expressed as net changes in prevalence of smoking, and pooled estimates were calculated by weighting using the inverse of the variance for each study, with the pooled standard error calculated as the square root of the inverse of the sum of the weights.21 Fixed and random effects analyses were carried out. Random effect analysis tends to provide more conservative estimates of the precision of estimates of effect size.21
Sensitivity analyses were performed by inclusion and exclusion of trials as some characteristics of the interventions or of the participants varied between trials. The effect of intervention was related to initial risk of coronary heart disease using the event rates in the control group and also the combined control and intervention group.
Relations between baseline risk factors and the size of changes in risk factors in each of the trials were examined by using the statistical package for social sciences weighted least squares method with the trial sample sizes as weights.22
In trials where randomisation was by paired clusters–for example, those based in factories or general practices–rather than in individual people it is appropriate to take the variation between clusters into account in estimating the effects of intervention (S G Thompson, personal communication). However, the data available in published reports or from authors were insufficient to carry out such analyses and our estimates of treatment effects of these trials will tend to be overprecise, with their contribution to the pooled effect estimates being exaggerated.
Previously unpublished data on mortality, changes in risk factors, and use of pharmacological treatments were also obtained for some of the trials (see acknowledgements) and are available from us.
We found 14 trials of multiple risk factor intervention, of which nine reported both disease events and changes in risk factors as outcomes.23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 We excluded several trials of multiple risk factor interventions from consideration for the following reasons: changes in risk factors were not measured or reported,39 40 41 42 43 44 allocation to intervention and control groups was not random,45 46 47 48 49 50 randomisation was inadequate,51 52 there was no specific multiple risk factor intervention,53 the control group received some form of intervention,54 55 56 and follow up for six months was not reported.57 58 59 Some studies were set up under the auspices of the former Soviet Union,49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 but allocation to intervention and control groups seemed not to be random, although we have not been able to trace the investigators.
Details of the 14 trials are summarised, with those reporting both changes in risk factors and disease event outcomes in table 1 and those reporting only changes in risk factors in table 2. In general, the trials compared an intervention comprising some form of counselling and information provision with control groups that either received nothing or usual care. The type of behavioural intervention used was seldom reported, but when it was, a “negotiated behaviour change” model was most often described.
The nine trials with clinical event outcomes comprise about 903 000 patient years of observation and the five trials with risk factor end points alone 303 000. The oldest subjects included in any of the trials were under 74, with most trials studying only middle aged people.
Changes in risk factors
The changes in blood pressures, smoking, and blood cholesterol concentrations observed in the trials are summarised in table 3 table 4 table 5. Not all trials reported or were able to provide data on each of these three risk factors. The Spanish arm of the World Health Organisation's factories study reported its risk factor changes separately and is considered as a separate trial. The Oslo Study Investigators26 could not provide data on changes in blood pressure but stated that there were no changes observed (personal communication).
Overall, the fixed effects analyses showed highly significant falls in systolic (net difference 4.2 mm Hg (SE 0.19 mm Hg)) and diastolic (net difference 2.7 mm Hg (0.09 mm Hg)) blood pressures. The multiple risk factor intervention trial,27 the hypertension detection and follow up programme,30 the Finnish businessmen study,29 and the family heart study37 stand out as showing particularly substantial falls in blood pressure, and all but the family heart study used antihypertensive drug treatment. The family heart study reported only one year follow up data and its design did not allow for measurement habituation effects.37 Exclusion of those trials in which a high proportion of participants were receiving pharmacological treatment (Gothenburg study,25 multiple risk factor intervention trial,27 Finnish businessmen study,28 29 hypertension detection and follow up programme,30 cost effectiveness of lipid lowering study32) resulted in much smaller net reductions in blood pressure: systolic pressure 2.3 mm Hg (0.3 mm Hg), diastolic 1.1 mm Hg (0.1 mm Hg).
Smoking rates, expressed as the changes in prevalence of smoking, showed an overall net reduction of 4.2% (0.3%). The rates fell particularly sharply in the multiple risk factor intervention trial27 and in the Oslo study,26 which both used individual smoking advice given by a doctor. Reported reductions in smoking rate in the multiple risk factor intervention trial27 were overestimated when compared with serum thiocyanate concentrations.64 The hypertension detection and follow up programme did not detect any changes in smoking rates,30 but published data are not available.
Blood cholesterol concentrations showed a small but highly significant fall (net fall 0.14 mmol/l (0.01mmol/l)). The hypertension detection and follow up program did not show any reduction in blood cholesterol concentrations,30 but published data are not available. Exclusion of the Finnish businessmen study28 29 on the grounds of use of cholesterol lowering drugs did not make a difference to the pooled fall observed (net difference -0.12 mmol/l (0.01 mmol/l)).
Blood pressure, smoking, and blood cholesterol outcomes were subject to substantial heterogeneity (see table 3 4 5). Random effects analyses were also conducted which showed similar pooled net effects, but as the variation between trials was taken into account (which was large), much larger standard errors were estimated.
Net changes in risk factors were strongly correlated with the initial diastolic blood pressure, smoking, and blood cholesterol concentration but not with systolic blood pressure. The sample size weighted correlation coefficients between initial value of and size of reduction in risk factor for diastolic blood pressure, smoking, and blood cholesterol concentration were 0.73 (P=0.006), 0.63 (P=0.01), and 0.74 (P=0.004) respectively. The studies with the highest baseline diastolic blood pressure, smoking prevalence, and blood cholesterol concentrations showed larger falls in these risk factors in association with the interventions.
Changes in mortality and rates of non-fatal myocardial infarction
Details of the nine trials reporting outcome in terms of total mortality and mortality from coronary heart disease are shown in table 6. The Oxcheck study was not designed to examine effects on mortality, but an intention to treat analysis was conducted in which the people randomised to health checks in years 1 to 3 were considered to be the intervention group and those randomised to health checks in year 4 were the control group.33 We compared deaths up to the start of year 4 using data provided by the investigators. Only the hypertension detection and follow up programme30 and the Johns Hopkins hypertension trial31 reported significant effects on total mortality; the Johns Hopkins trial also reported significant effects on mortality from coronary heart disease.
Evidence of statistical heterogeneity was apparent in the pooled odds ratios for total mortality and to a lesser extent for mortality from coronary heart disease when all studies were included. Removal of the trials including hypertensive patients (hypertension detection and follow up programme30 and Johns Hopkins hypertension trial31) reduced the heterogeneity for total mortality but not for mortality from coronary heart disease. Including a term for interaction between treatment effect and baseline level of coronary heart disease risk calculated using either control group coronary heart disease risk or combined control and intervention group risk reduced heterogeneity between the trials to insignificant levels for total and coronary heart disease mortality.
Modelling the effects of age using the mean age of study participants and proportion of patients taking antihypertensive and cholesterol lowering drug treatment did not show any significant interactions between age, drug treatments, and outcome. The significant interaction between intervention and degree of risk of coronary heart disease (indexed either by event rate in the control group or by combined treatment and control group rate) indicated that trials recruiting participants at higher risk were more likely to show beneficial effects. This effect is explained by the inclusion of the two trials which studied hypertensive patients rather than members of the general population or of a workforce. It is impossible to separate this effect of baseline coronary heart disease risk from the benefits of pharmacological treatment of hypertension.
Non-fatal myocardial infarctions were reported in the WHO factory study,23 the Gothenburg study,25 the Oslo study,26 the multiple risk factor intervention trial,27 and the Finnish businessmen study.29 The pooled odds ratio for non-fatal myocardial infarction in these five trials was 1.0 (95% confidence interval 0.92 to 1.07).
The odds ratios for both fatal and non-fatal events were all close to 1. Clearly, any effect on mortality of multiple risk factor intervention in the follow up period was not large as the confidence intervals were tight; a difference as small as an 8% reduction in total deaths and an 11% reduction in deaths from coronary heart disease may be excluded by these data.
Systematic review and meta-analysis are powerful tools to aid policy and practice decisions in multiple risk factor intervention, in which received wisdom and current practice are at odds with the emerging scientific evidence.
Multiple risk factor interventions comprising counselling, education, and drug treatments were ineffective in achieving reductions in total mortality or mortality from cardiovascular disease when used in general or workforce populations of middle aged adults. The pooled effects of intervention were insignificant, but a potentially useful benefit of treatment (about a 10% reduction in mortality from coronary heart disease) may have been missed.
The changes in risk factors associated with interventions were modest but are probably optimistic estimates as changes could be measured only in those remaining in the trials. Habituation to blood pressure measurement, regression to the mean, and self reports of smoking will also tend to exaggerate the changes observed. It is, however, not possible to separate participants' degree of risk from the use of antihypertensive drugs in this set of trials because studies with participants at high risk tended to include participants with high rates of use of antihypertensive drugs. Furthermore, there are many problems in relating the outcome of a trial to a risk measure that is itself dependent on the outcome in meta-analysis.65 Therefore our conclusions on these issues can only be tentative.
Heterogeneity in the effects of intervention is apparent. This is caused by two factors: the participants included in the trials and the use of pharmacological treatments. People with hypertension, at highest risk, were more likely to benefit from counselling and education and effective drug treatment. These findings suggest that targeting of current health promotion activities to people at high risk would be valuable.
The benefits of drug treatments for lowering blood pressure and cholesterol are clear.66 67 68 69 However, those people at highest risk of disease either in needing control of hypertension70 or lowering of cholesterol concentration69 benefit most. Treatment of low risk populations may result in small treatment benefits being outweighed by small treatment risks,71 which may have occurred in subgroups within the multiple risk factor intervention trial27 and in the Finnish businessmen study.29 There were strong associations between baseline levels of risk factors and net falls experienced, suggesting that intervention may be more effective in populations with particularly adverse risk factor profiles.
More intensive interventions might be expected to produce better effects, although those used in many of the trials would far exceed what is feasible in routine practice. A recent meta-analysis of dietary modifications found that increasing intensity of dietary intervention was associated with greater falls in blood cholesterol concentrations in participants at high risk.72 In the Minnesota heart health programme, a non-randomised community trial of intensive health promotion, changes in risk factors and mortality showed virtually no difference between intervention and control communities.73 74 The limited impact on the practice of health promotion of the essential failure of these community intervention trials is curious, especially given the huge resources that have been put into them.
Latency of effects
Benefits may not be detected in the early stages of an intervention, but they may emerge over time. Longer term follow up of participants in the multiple risk factor intervention trial has shown increased divergence in mortality between the control and intervention groups75; this has also been found in the Tromsø family trial (S Knutson, personal communication). However, evidence from pharmacological trials suggests that benefits from reducing blood pressure and blood cholesterol concentration are observed within two to four years.66 76 The effects of giving up smoking vary depending on the clinical outcome considered: the risk of stroke falls rapidly after stopping,77 but the risk of coronary heart disease may be less reversible.78 79
Quality of trials
The quality of the trials examined deserves comment. Few of the published reports provided sufficient detail to replicate the intervention used, and in several trials the intervention varied between sites and over time. The quality of the intervention, in terms of intensity and frequency, person carrying out activities, and the theoretical framework of behavioural change used will probably determine the impact of intervention. Losses to follow up were a particular problem as changes in risk factors cannot be assessed in an intention to treat analysis. Validation of smoking outcomes using biochemical assay of thiocyanate was only used in one trial.27
Evidence of benefit
The quasi-experimental North Karelia study has been highly influential in supporting multiple risk factor intervention. Examination of the trends in both risk factors80 81 and mortality from coronary heart disease82 observed in North Karelia and comparison regions show similar patterns occurring at the same time, suggesting that the interventions in North Karelia were not instrumental in causing the improvements observed. Indeed, the North Karelia and similar projects may be viewed as effects, or epiphenomena, of the high mortality from coronary heart disease experienced in many countries in the 1960s.
In secondary prevention after myocardial infarction and angina, trials of multiple and single risk factor interventions have shown substantial benefits.69 83848586 Intervention aimed at modifying lifestyle after myocardial infarction is probably effective because participants are much more likely to change their behaviours.
Limitations of randomised controlled trials
The interventions reviewed were essentially individual or family approaches. Randomised controlled trials impose limitations on the nature of interventions that may be tested and are of more value in examining high risk rather than population and social approaches to prevention.87
Health protection through national fiscal and legislative changes that aim at reducing smoking and dietary consumption of fats and so called hidden salt and calories and at increasing facilities and opportunities for exercise should have a higher priority than health promotion interventions applied to general and workforce populations. The current concepts and practices of multiple risk factor intervention, primarily through individual risk factor counselling, must not be exported to poorer countries as the best policy option for dealing with existing and projected burdens of cardiovascular disease, as is currently happening.88 Health protection should be promoted as the mainstay of preventing chronic diseases in poorer countries.
We thank the following investigators for providing us with data: M Shipley (WHO factories study), L Wilhelmsen (Gothenburg study), I Hjermann (Oslo study), J Shaten (multiple risk factor intervention study), T Miettinen (Finnish businessmen study), J Muir and T Lancaster (Oxcheck study), J Baron (Abingdon trial), S Pyke (family heart study), S Boles (Take Heart study), T Ekbom (cost effectiveness of lipid lowering study). The following investigators replied to our request but were unable to provide us with further data for various reasons: G Payne (hypertension detection and follow up programme), D Morisky (Johns Hopkins study), R Stamler (primary prevention of hypertension study), S Knutson (Tromsø family trial).
We would be grateful for any information on randomised controlled trials that we may have overlooked in this review. We would be grateful for any information on randomised controlled trials that we may have overlooked in this review.
Funding: The work reported was commissioned by the York Centre for Reviews and Dissemination and funded by the Health Education Authority, England.
Conflict of interest: None.