Information needed to decide about cardiovascular treatment in primary careBMJ 1997; 314 doi: https://doi.org/10.1136/bmj.314.7076.277 (Published 25 January 1997) Cite this as: BMJ 1997;314:277
- John Robson, senior lecturera
- a Academic Department of General Practice St Bartholomew's and the London Hospital Medical School Queen Mary and Westfield College London E1 4NS
- Accepted 9 January 1997
There is growing consensus that treatment of cardiovascular risks should be based on multiple rather than single factors and on absolute rather than relative risks. Thresholds for treatment should reflect the level of absolute risk at which the benefits and hazards of treating outweigh the benefits and hazards of not treating. Once a decision has been made to initiate a treatment programme, clinicians need to know the patient's absolute risk. At this level of risk, do the benefits of treatment outweigh the hazards? Given this information, which treatment option does the patient prefer? Using cardiovascular disease as an example, I review some measures that assist decision making in primary care. Practice guidelines should routinely include accessible presentation of treatment outcomes on benefit, hazard, and costs for a range of absolute risks. These measures enable patients and their doctors to weigh the pros and cons of treatment in their particular circumstances.
There are three main sources of information guiding decisions about cardiovascular treatment. Cohort studies provide information on multiple risk factors and subsequent cardiovascular events, so that a person's absolute risk may be estimated from a score.1 Risk might be considered high if the chance of death or disease from a major coronary event exceeded 6% a year (60% in 10 years). However, risk alone reveals nothing about the advisability of drug treatment, and, although cardiovascular risk scores have been available for more than 20 years, uncertainty about their role has hindered integration into routine clinical practice.2 3 The second source, based on clinical trials, yields at a particular level of risk a variety of treatment outcomes, including the “number needed to treat,”4 years of life gained, adverse events, and costs.5 The third source of information is the person to be treated, who, given an informed choice of treatment outcomes, can indicate his or her preference.6
Assessing cardiovascular risk
Cardiovascular risk scores were originally designed to facilitate behavioural change by communicating the degree of risk to clinicians and patients. More recently, they have been used to target treatment more effectively, because prediction of risk is improved when based on multiple rather than single risk factors. Relative risk scores such as the Dundee score, may have a continuing role in behavioural change, particularly among lower risk groups at younger ages.7 However, relative risk has been criticised as a method of allocating treatment, and absolute risk (from which relative risk may be derived) is preferable.2 8
The main coronary risk scores are based on data from the American Framingham study,9 the British regional heart study,10 and Scottish survey and British trial data (Dundee score).7 All were specifically devised to influence primary care doctors.11 12 The Framingham study has the advantages of longer follow up over a wider age range for both sexes and is expressed as absolute risk.1 In addition, it has shown that the ratio of serum total cholesterol concentration to high density lipoprotein cholesterol is a considerably better predictor than total cholesterol concentration alone. Five or 10 year risks may be estimated and scores tabulated by age so that the effects of multiple risk factors can be compared over time.13 It has the disadvantage that it is based on an American population. Scores are also available for the risk of stroke.12 14 12
Recent cardiovascular guidelines in Europe13 and New Zealand8 adapted the Framingham data and incorporated other risk factors such as obesity and family history. This has made the risk equations more user friendly and has taken some account of specialists in hypertension, who variably recommend treatment at diastolic thresholds of 90, 95, and 100 mm Hg16 even though data from both Framingham and the Medical Research Council indicate that systolic pressure is a considerably better predictor of stroke.14 17
Because prediction depends on the prevalence of the condition, none of the risk scores has a high predictive value. The scores overestimate risk, and only 60% of those subjects at highest risk (in the top 10% of the risk distribution) will actually have a coronary event within the next 10 years. Prediction is considerably worse at younger ages and lower levels of risk. Factors not included in the score, such as family history of cardiovascular disease (first degree relative aged under 55) or obesity, may mean that risks are higher than predicted. Social class and ethnic group may also be relevant. Nevertheless, for all their limitations, multiple risk scores are the best available tools for predicting risk and are superior to any one factor alone.
The adoption of absolute, rather than relative, risk as the criterion for treatment inevitably prioritises older age groups. On this basis, healthy men aged under 50 years with a serum cholesterol concentration of ≤9 mmol/l do not require drug treatment. Although their relative risk is high compared with their peers with a serum cholesterol concentration of 5 mmol/l, even if they smoke and have raised blood pressure they are unlikely to reach a coronary event rate of 3% a year, the threshold at which the Sheffield group considers treatment appropriate.18 19 20 Advice to stop smoking, exercise, and improve their diet remains the mainstay of intervention.
Treatment thresholds and treatment outcomes
Once a person's absolute risk has been estimated how is the threshold at which treatment is indicated to be decided? More information is needed, and patients and their doctors require summaries of the hazards, benefits, and costs of treatment.
Risk scores describe the level, distribution, and relative importance of different factors but give no information on the outcomes resulting from treatment. The final choice of treatment is affected by the choice of outcome, cause of death or disease, deaths prevented or number needed to treat, years of life lost or gained, adverse events or indicators of quality of life, and cost. Should treatment of raised blood pressure depend on the risk of stroke or the risk of heart attack or both? Other outcomes may be decisive factors in treatment. The impact of diabetes on risk of stroke is modest, but the impact of raised blood pressure on diabetic renal disease is considerable.
Benefits of treatment
The manner in which benefits are expressed is particularly important. Expressing values as numbers needed to treat gives equal weight to the risk of dying from a heart attack at the age of 75 as at age 40. However, if years of life gained rather than numbers of deaths were used as the outcome, intervention would include more younger people and fewer older people.21 Both measures are useful, as each expresses a different component of outcome.22 An important gain in mean life expectancy would be about 60 days. This is not so small a gain as it seems, as it is averaged over all people receiving treatment.23 For example, if all cancer were eradicated the mean gain in life expectancy would be only one year, although for the people who would otherwise have developed cancer the gain would be measured in decades.
The format in which these outcomes are presented may influence choice of treatment. Thus, presenting results in terms of years of life gained may lead to different treatments being preferred than when results are given as numbers needed to treat,24 and relative risk may result in lower treatment thresholds than absolute risk.25
The number needed to treat to prevent one event or death is the reciprocal of the absolute difference in outcome between the treatment and control groups in a clinical trial26 and is a measure of the absolute efficacy of treatment. Assuming a 30% reduction in coronary heart disease as a result of treatment with “statins” (hydroxymethylglutaryl coenzyme A reductase inhibitors), 1332 people at low risk (with an annual coronary event rate of 0.05%) would require treatment for five years to prevent one coronary event. In contrast, 20 people would need treatment for five years to prevent one event if their annual risk was 3%. Table 1) shows the number needed to treat for different levels of risk.
Adverse effects and quality of life
There is most uncertainty about treating the many people at intermediate risk, rather than the smaller group at high risk who have most to gain from treatment. Small changes in the threshold of treatment in the intermediate range can turn large numbers of people into lifelong patients. Clinicians should require a high level of confidence in the extent of the benefits and hazards to convince them that treatment of such patients is worth while because small but serious risks applied to large numbers can transform gains in one area into a net loss.27 If recommendations are uncertain or contradictory, patients and doctors require more information, accessibly presented, about the hazards and benefits of treatment so that they can make their own judgments.
Quality of life may be valued more than years of life, and the adverse effects of treatment may be decisive in the choice of treatment.28 In the Medical Research Council trial of treatment of mild hypertension, for each cardiovascular event prevented, 33 men experienced adverse reactions (impotence and fatigue being prominent) and 20% stopped treatment as a result.17 Although adverse effects of treatment can be incorporated in the concept of years of quality adjusted life gained,21 the subjective perception of hazards can have profound effects on the choice of both patients and public.29 This is shown by the unpopularity of prostatic surgery for urinary symptoms: after the risks and benefits of different treatments had been explained, the degree of inconvenience caused by the treatment and concern about impotence were the most important predictors of choice, and only 20% of men with severe symptoms opted for surgery.28 Given an informed choice, patients often express preferences that are different from those of their doctors or peers.30 31
Decision analysis is able to concurrently assess the impact of multiple components, including quality of life, on treatment options,32 33 34 but many questions remain. What is a reasonable number needed to treat or cost per year of life gained? How sensitive is the analysis to alterations in parameters? Do the results need further qualification for population subgroups?
Cost and workload
The monetary cost of different treatments may be compared through cost-effectiveness analysis. Because gains are often fairly small, the conclusions of such analyses can be highly sensitive to minor changes in parameters, including drug costs.35 While this may not be a problem when comparing two treatments under similar conditions or the incremental gains of extending a programme, it is a major issue when working out the basic costs of a programme for treating mild hypertension.36 In addition, equity is omitted as a consideration in most analyses,37 although there is no reason in principle why it should not be incorporated.38
Policy decisions may be influenced by both workload and the overall cost of a programme. These are determined by the size of the target population, and any new programme identifying treatment for more than 2% of the population has major implications for workload in primary care.2 At a policy level, consideration should be given to whether greater benefit may be obtained at the same cost from some other intervention.
Accessible summaries of treatment outcomes
These aids to decision making offer no tablets of stone. Instead, they provide estimates of the limits of uncertainty and a means to compare the relative value of different options. The presentation of a range of outcomes may not solve the problem, but it does establish a common currency in which options are explicit and decisions may be shared.23 I discuss the use of such measures in the current debates on the treatment of raised blood pressure and cholesterol.
While the New Zealand and European cardiovascular guidelines usefully summarise cardiovascular risk, neither provide an accessible summary of the numbers needed to treat, years of life gained, adverse event rates, costs, or workload at each level of absolute risk. Doubts remain about treating mild hypertension at younger ages. For example, 400 men aged 52 with systolic blood pressure <150 mm Hg would require treatment for five years to prevent one stroke. This number falls to 250 for systolic pressure of 150-169 mm Hg and to 100 for systolic pressure of 170-199 mm Hg.39 40 Lifetime treatment is estimated to add an average of 30 to 64 days to life expectancy; at 1995 prices, the cost ranged from £14 000 to £82 000 for men (greatest in the youngest patients) and from £28 000 to £252 000 for women for each quality adjusted life year gained.41
Patients and doctors need easy access to such data to inform their decisions. Table 2) shows a summary of outcome measures that might be included in future guidelines. It is not until annual cardiovascular risk is 1.5% or more that the benefits of treating raised blood pressure begin to become clear.8 While a national strategy to reduce dietary salt and obesity would have considerably more impact than drug treatment, and at considerably less cost,1 42 patients and their doctors must, for now, rely on evidence of this kind to inform their choice of treatment.
Managing serum cholesterol
The management of serum cholesterol presents greater uncertainty, although this is now changing for high risk patients. For people with heart disease and serum cholesterol above 6.5 mmol/l, and possibly at lower levels, treatment with “statins” to reduce cholesterol confers benefit.19 43 For people without heart disease at lower absolute risk, recommendations for drug treatment based solely on their relative risk due to raised cholesterol have been controversial. The European Atherosclerosis Group13 and a Sheffield group18 20 have attempted to resolve the dilemma by recommending that treatment be based on absolute risk calculated with data from the Framingham study. Even with these more conservative guidelines, the number of people for whom treatment is recommended is substantially higher than reviews of previous trials seem to warrant. For people without pre-existing heart disease, adverse effects of treatment may be decisive, and further large scale trials are necessary to establish lower limits for treatment.27 44
The west of Scotland study indicates that, among healthy men with an average serum cholesterol concentration of 7 mmol/l and annual coronary event risk of 1.5%, 200 people would require treatment for five years to prevent one coronary death and 40 would require treatment for this period to prevent one coronary event.45 Table 3) compares the results of this study with those of the Scandinavian simvastatin survival study.19 Again, this simple summary of treatment outcomes would enhance future guidelines, although gains in life years were not available at the time, nor were costs in the Scandinavian study. Even at the low levels of adverse reaction reported, for every one person gaining from prevention of coronary death, two or three people would experience adverse effects from treatment. It is in these areas of uncertainty that there is the greatest need for accessible presentation of available data.
Absolute multiple risk scores usefully summarise risk and are better predictors than any single factor. However, risk is only half the story. Accessible presentation of numbers needed to treat, years of life gained, adverse events, cost, and workload is also needed for shared decision making, enabling patients and their clinicians to weigh the pros and cons of treatment in their particular circumstances. These measures should be routinely included in management guidelines. Decision tools may only approximate to the truth, but presenting a range of treatment outcomes establishes a common currency for informed choice. Placing finite limits on uncertainty is the best guarantee that treatment will be optimal and appropriate.47
I thank Carol Dezateux, Kambiz Boomla, Gene Feder, Rod Jackson, Anna Livingstone, Pat McBride, and the referees who commented on this article.
Funding: I wrote this article while on study leave funded by the Department of Health, the Royal College of General Practitioners, and the King's Fund.
Conflict of interest: None.