Intended for healthcare professionals


Who benefits from medical interventions?

BMJ 1994; 308 doi: (Published 08 January 1994) Cite this as: BMJ 1994;308:72
  1. G D Smith,
  2. M Egger

    The results of clinical trials are often expressed in relative terms - for example, a particular treatment reduces the risk of an adverse outcome by 40%. Yet knowing that the treatment reduces the risk of such an outcome from 5% to 3% (an absolute reduction of 2%) may be more useful clinically.1

    The effects of antiplatelet treatment are a case in point. The meta-analysis of randomised controlled trials of such treatments by the Antiplatelet Trialists' Collaboration shows that the risk of dying from a vascular cause varies substantially according to patient group (p 81).2 It ranges from 10% over one month for patients with an acute myocardial infarction at entry to the trial, through 9% over two years for those who had survived an acute myocardial infarction, down to only 2% over five years in trials of primary prevention.

    With such varying risks of future vascular events it is unsurprising that the absolute effects of antiplatelet agents differ substantially by patient group. Reductions in relative risks may be similar - antiplatelet drugs reduce the risk of future non-fatal myocardial infarction by 30% in trials of both primary and secondary prevention. But when the results are presented as the number of patients who need to be treated for one non-fatal myocardial infarction to be avoided they look very different.

    In secondary prevention 50 patients need to be treated for two years, while in primary prevention 200 patients need to be treated for five years, for one non-fatal myocardial infarction to be prevented. In other words, it takes 100 patient-years of treatment in secondary prevention or 1000 patient-years of treatment in primary prevention to produce the same beneficial outcome of one fewer non-fatal myocardial infarction.

    The fact that the risk that patients are at before they begin treatment affects their absolute benefits from treatment has been much discussed with regard to the treatment of hypertension.*RF 3-5* In randomised controlled trials treatment has consistently reduced the risk of stroke by 40% (expressed in relative terms).6 The 10-fold variation in stroke rates in patients in these trials, however, leads to the number of patient-years required to prevent one stroke ranging from around 100 to 1000. To prevent one vascular death the equivalent figures are about double these. The baseline characteristics of patients - age, sex, blood pressure, other risk factors for cardiovascular disease, and end organ damage - can discriminate groups with widely differing risks of stroke and coronary heart disease. The balance between the benefits and side effects of treatment will be more favourable among patients at a higher risk of vascular disease.

    Consider, for example, the systolic hypertension in the elderly programme (SHEP), in which 28% of the treatment group reported what they considered to be an “ntolerable problem” during treatment, a significantly higher proportion than the 21% recording such an event in the placebo group.7 The trade off between reducing the risk of a stroke or vascular death and the increase - affecting 7 of every 100 patients - in the risk of experiencing an “intolerable problem” would clearly be weighed differently by patients at different levels of cardiovascular risk.

    This balance between the benefits and unwanted side effects of treatment is particularly stark when such side effects are serious. The balance will be more favourable for those patients with a higher risk of the condition that the treatment aims to prevent. For antiplatelet agents, some evidence exists that they may slightly increase the risk of haemorrhagic stroke. For participants in primary prevention trials who are at low risk this contributes to the fact that the reductions in vascular death in the treatment arms of these trials are small and non-significant, translating into a best estimate of 1 death avoided for each 1450 person-years of treatment. This compares unfavourably with one vascular death avoided for each 75 person-years of treatment in the survivors of a myocardial infarction and one death avoided for each 3 person-years of treatment in patients treated during an acute myocardial infarction.

    Similar considerations apply to the appropriate use of cholesterol lowering drugs.8 For patients at a high risk of dying from coronary heart disease, reducing blood cholesterol concentrations with drugs reduces mortality. For patients at a lower risk - for example, those with moderately high cholesterol concentrations but without symptomatic cardiovascular disease - cholesterol lowering drugs may have increased the risk of death. For such patients a small absolute increase in the risk of death attributable to treatment, of around 1-2 deaths per 1000 patient-years of treatment, is enough to cancel out any treatment related fall in the risk of coronary heart disease.

    Small risks may outweigh small benefits

    Commenting on this phenomenon, Geoffrey Rose pointed out that “in mass prevention, each individual has usually only a small expectation of benefit, and this small benefit can easily be outweighed by a small risk.” 9 This should serve to caution against the widespread use of drugs to prevent disease until overall benefits have been established by controlled trials.10

    A final example from cardiovascular medicine concerns angiotensin converting enzyme inhibitors for treating and preventing heart failure. While the use of these drugs in acute myocardial infarction apparently produces only small benefits,11 their long term use in patients with, or at risk of developing, heart failure has produced more impressive results. The results of five large trials,*RF 12-16* presented in the figure, show that treating patients at high risk produces considerably greater benefit than treating patients at lower risk. With the range in the number of patients who need to be treated for a year for one death to be prevented varying from 2 to 330, depending on their risk of death due to existing cardiac disease, evaluations of risk-benefit and cost-effectiveness will vary greatly among different clinical situations.


    Outside cardiovascular medicine, the same relation is evident - for example, when the benefit of treating HIV infection with zidovudine is assessed. The first big trial in patients with AIDS or AIDS related complex at high risk of death reported impressive effects: 42 fewer deaths per 100 person-years of treatment.17 Yet in patients with less advanced HIV infection entered into the large Concorde trial, early use of zidovudine was associated with a non-significant adverse effect on mortality.18 When the relation between the benefit of treatment and mortality among the controls in six published trials is examined,*RF 17-22* increasing risk of death with controls is associated with an increase in the numbers of deaths delayed by zidovudine treatment. For people with asymptomatic HIV infection the side effects of zidovudine together with the uncertain effects on mortality may make treatment an unattractive option. A recent formal demonstration that zidovudine treatment early in the course of disease is not cost effective supports this conclusion.23

    The greater benefit (and reduced likelihood of harm) experienced by patients at higher risk has several implications for clinical practice and public health policy. Firstly, how trial data are presented may influence doctors' perceptions of the advisability of treatment. Converting relative measures of benefit into measures such as the number of patients who need to be treated to prevent one outcome event can modify the response of doctors to the results of trials, arguably in the right direction.24 Both investigators and pharmaceutical companies should be encouraged to provide absolute as well as relative measures of benefit produced by treatment.

    Secondly, clinical guidelines should reflect the association between degree of benefit and level of risk. This does not necessarily occur, as the first set of guidelines from the United States national cholesterol education program shows.25 These recommended intensive treatment for some groups of patients who were at a considerably lower risk of a coronary event than other groups of patients who were deemed not to require any special attention.26 Encouragingly, both the recent revision of these guidelines27 and a new set of recommendations for treating raised blood pressure from New Zealand28 have taken a more risk based approach to decisions on treatment. Thirdly, meta-analyses should incorporate estimates of the risk in the included trials to examine whether systematic differences in benefit are seen in relation to the degree of risk of participants in the trials.8 Finally, since those who gain the most from medical treatment are those who have most to gain, the irrationality of current suggestions that population groups at increased risk of disease - such as smokers - should not be treated should be recognised.