Translating clinical trials into practice

THE LANCET

Translating clinical trials into practice SIR—Your Jan 4 editorial1 addresses an important issue: how do we translate the results of clinical trials into practice? The ISIS studies were mega trials in cardiology, yet not all their results have become part of routine clinical practice. In ISIS-12 patients presenting within 5 hours of acute myocardial infarction (AMI) intravenous atenolol reduced mortality by 0·7%. The relative risk reduction was 15% but the number of patients needing to be treated to save one life was 200. In ISIS-23 placebo mortality was 13·2%, which was reduced to 8% after intravenous streptokinase and aspirin; an absolute reduction of 5·2% and a relative risk reduction of 42%—ie, one life saved after treatment of only 19 patients. In clinical practice in the UK, streptokinase and aspirin are now routine but intravenous atenolol is rarely used in the management of AMI. In ISIS-44 (oral captopril early after AMI), the absolute reduction in mortality was only 0·5% and yet the relative risk reduction was 7%. Nevertheless, 200 infarct patients must be treated with captopril to save one life. In the AIRE study5 angiotensin converting enzyme inhibitors (ACEI) were targeted to a population with clinical evidence of heart failure after AMI giving an absolute benefit of 6% and a relative risk reduction of 27%, thus only 17 patients need to be treated to save one life. Most cardiologists in the UK selectively use ACEI therapy for patients with overt heart failure or those who have had large anterior myocardial infarctions in whom left ventricular function is usually seriously impaired. Thus, not all results of the major clinical trials in cardiology have been translated into clinical practice. In AMI, treatments that show high risk reductions and reduce absolute mortality are quickly accepted by clinicians. Smaller benefits mean that treatments are less likely to be implemented. Why is this? First, statistical evaluations can be misleading. Relative risk reductions are always greater than absolute risk change—but it is the latter that concerns the individual patient. Small statistically highly significant effects can always be shown if a large number of patients is studied. If you can show a benefit by studying hundreds rather than tens of thousands of patients it is more likely to be an important effect. Second, if the size of the benefit is small far more patients will be exposed unnecessarily to side-effects. In ISIS-4, 199 patients were exposed to effects of early ACEI therapy without benefit, with significant doubling of the frequency of profound hypotension,

654

higher chance of cardiogenic shock, and excess chance of heart block. Thus, the chance of benefit in mortality was less than the chance of a serious side-effect. Third, trials are only relevant to the population studied. Patients who are enrolled in clinical trials are the minority, not the majority. The actual mortality of AMI in a major teaching hospital is 17% not 4·6% (ISIS-1), 13·2% (ISIS-2), or 7·7% (ISIS-4). Extrapolating the results of trials to the individual requires a great leap of faith. Commonsense is about balancing the risks and benefits of strategies, and good doctors have it. K S Channer Department of Cardiology, Royal Hallamshire Hospital, Sheffield S10 2JF, UK

1 2

3

4

5

Editorial. And now all this. Lancet 1997; 349: 1. ISIS-1 Collaborative Group. Randomised trial of intravenous atenolol among 16 027 cases of suspected acute myocardial infarction: ISIS-1. Lancet 1986; ii: 57–66. ISIS-2 Collaborative Group. Randomised trial of intravenous streptokinase, oral aspirin, both or neither among 17 187 cases of suspected acute myocardial infarction: ISIS-2. Lancet 1988; ii: 349–60. ISIS-4 Collaborative Group. ISIS-4: a randomised factorial trial assessing early oral captopril, oral mononitrate, and intravenous magnesium sulphate in 58 050 patients with suspected acute myocardial infarction. Lancet 1995; 345: 669–85. The Acute Infarction Ramipril Efficacy (AIRE) study investigators. Effect of ramipril on mortality and morbidity of survivors of acute myocardial infarction with clinical evidence of heart failure. Lancet 1993; 342: 821–28.

SIR—That patients and the public are seeking definitive, authoritative statements about diagnosis, causes, and treatment of diseases is understandable, and it is only natural that doctors and public health practitioners should satisfy that need. But when scientists extend that practice into biomedical science they invite shortcuts to the truth. I have taken the liberty of questioning some of your assertions in your Jan 4 editorial.1 You say “we take the view. . . that the public should be told about uncertainty when data with public-health implications are preliminary or inconclusive”. I take this to imply that in other instances data with public health implications can be conclusive. Useful discussions of the so-called conclusiveness of empirical or statistical tests of scientific theories can be found in the books of Karl Popper, Joseph Agassi, and Michael Oakes; in a book chapter by Miller;2 and in a collection of essays about causal inference.3 You go on to state that “while clinical and epidemiological research marches on, individuals are failing to benefit from proven remedies”. I think that Senn4 has managed to muddy the

prevailing notion that clinical trials are capable of proving statements such as “treatment A will always benefit patients who meet the eligibility criteria of the relevant trial”. Senn provides a logical interpretation of the relation between medical theories, clinical trials, and medical practice, which need not subscribe to illogical concepts such as proven remedies and generalisability of the results. I agree that “epidemiology and clinical trials are generally uninformative at the level of the individual”, if the emphasis is on the word generally. However, there is at least one research method that is informative at this level whenever the condition in question is chronic and fairly stable (eg, osteoarthritis): it is the single-patient randomised trial. The debate between Bayesians and Popperians is hardly a settled issue that can be dismissed by your assertion that “Bayesian thinking, which is intimately linked to the individual, is long overdue more formal use”. Suffice it to say that much of Bayesianism is intimately linked to the psychology of scientists: what scientists ought to believe, given previous beliefs and the data at hand.5 Whether the business of science is quantifying beliefs in scientific theories is an open question. I would hardly expect research clinicians or statisticians, as you suggest, to solve the time-honoured problem of induction—namely, how to draw inference from the observed to the unobserved—eg, from the average response to aspirin observed under experimental conditions to the future response of one particular patient. No one has any secret to share because the logical answer to this question is straightforward: we cannot draw such inference. It is impossible to construct a system of inductive logic.2 What we can do is: (1) put our theories of medical regularities to severe tests, and (2) keep in mind that when a doctor prescribes aspirin to any number of survivors of a first heart attack, it is possible that no patient will benefit from that well rationalised practice. I conclude with a reminder of Sir Karl Popper to those of us who, from time to time, cannot resist the temptation to make authoritative statements about human knowledge: “there are all kinds of sources of our knowledge; but none has authority”. E Shahar Division of Epidemiology, University of Mnnesota, Minneapolis, MN 55454, USA

1 2

Editorial. And now all this. Lancet 1997; 349: 1. Miller D. Conjectural knowledge: Popper’s solution of the problem of induction. In: In Pursuit of Truth, Humanities Press and Harvester Press, 1982.

Vol 349 • March 1, 1997

THE LANCET 3

4 5

Rothman KJ (ed): Causal Inference. Epidemiology Resources Inc., Chestnut Hill, MA, 1988. Senn SJ. Falsificationism and clinical trials. Stat Med 1991; 10: 1679–92. Poole C. Feelings and frequencies: two kinds of probability in public health research. Am J Publ Health 1988; 78: 1531–33.

SIR—I was surprised to read,1 that research clinicians are suspected of having a secret method of translating their trial results into a language easily understood by their patients. The problem was solved with the introduction of the National Lottery in the UK, but the solution was so obvious that it did not seem worth publishing. The average layman knows that his chances of winning the lottery are small, but he also knows that “it could be you”, and he plays the lottery on that understanding. He or she can therefore easily understand, for example, that although the risk of death during coronary artery bypass is 1–2%, “it could be you”. I always now explain risks and benefits of surgery or other treatment on that basis, and it works very well without adding much to the consultation time. John Hampton Department of Cardiovascular Medicine, Queen’s Medical Centre, Nottingham NG7 2HU, UK

1

Editorial. And now all this. Lancet 1997; 349: 1.

Field of distinction

No of physicians

Poetry Fiction, general literature Scholarship, bibliography, scientific collecting Invention Crime Art, music The church Piracy Governors, administrators Philosophy Law, politics Explorers Sports, games Drama Philanthropy Aeronautics Ambassadors Saints The stage Soldiers Rosicrucianism

72 63 16 16 14 13 12 12 11 10 10 7 7 7 4 4 4 3 2 2 1

Doctors’ fields of distinction outside medicine

that medicine and poetry have a special affinity, although this may also, of course, reflect the greater ease of maintaining dual careers in the two than in, say, medicine and piracy or medicine and sainthood; and it may also be that the editors of the Dictionary of national biography (Monro’s principal source) are more sympathetic to including, for example, doctors who are also minor poets than doctors who are also minor criminals. Chris McManus Academic Department of Psychiatry, Paterson Centre for Mental Health, Imperial College School of Medicine at St Mary’s, London W2 1PD, UK

Physician-poets SIR—Do doctors have an exceptional interest or talent for poetry (Jan 25, p 275),1 or is it instead merely inevitable that by chance alone some good poets will also be physicians? Perhaps, even, there is some antipathy between the callings, so that the physician-poet is a peculiarly rare beast? Hard statistical information there may not be but an interesting resource is Thomas Kirkpatrick Monro’s The physician: as man of letters science and action,2 first published in 1933 with a second edition in 1951. The work provides classified biographies of 395 doctors who distinguished themselves in nonmedical fields, excluding 20 from America, 106 from Europe and elsewhere, and 17 “students of medicine who never qualified” who are not classified. Ignoring the 105 doctors also distinguished as scientists, since that almost seems part of the job description, and with the use of Monro’s own somewhat eccentric classifications, the table shows the various categories in descending order of frequency. That poetry is at the top of the list perhaps gives some support to the idea

Vol 349 • March 1, 1997

1 2

Jones AH. Literature and medicine: physician-poets. Lancet 1997; 349: 275–78. Monro TK. The physician: as man of letters science and action (2nd ed). Edinburgh: E&S Livingstone, 1951.

SIR—Although not intended to be exhaustive in coverage, it is regrettable that Jones’ article omitted Erasmus Darwin (1731–1802)—grandfather of Charles Robert—and arguably the greatest of the physician-poets.1 Through his The botanic garden (1789–91) and The temple of nature (1803), Darwin—the most eminent physician of his age, and a noted polymath (philosopher, botanist, inventor, evolutionist, and founder of the Lunar Society)—had a profound influence on the great romantic poets of the early nineteenth century,2,3 in particular, Wordsworth, Coleridge, and Shelley. As King-Hele3 has pointed out, they were especially influenced by: The loves of the plants (1789), The economy of vegetation (1791), and The loves of the triangles (1798). G C Cook Hospital for Tropical Diseases, London NW1 0PE, UK

1 2 3

Jones AH. Literature and medicine: physician-poets. Lancet 1997; 349: 275–78. Krause E. Erasmus Darwin. London: John Murray, 1879; p 216. King-Hele D. Erasmus Darwin 1731–1802. London: MacMillan & Co, 1963; p 183.

Psychiatric diagnosis in addiction treatment experiments SIR—The public health emergency of AIDS underlines the growing need for more medical treatment of substanceabuse-related disorders. Across-culture treatment of drug abuse has been found consistently to be a principal determinant in the prevention of the spread of AIDS. Furthermore, a galling limitation of AIDS prevention interventions is the prevalence of psychopathology in the target population, which curbs the effectiveness of services. Substanceabuse disorders have grown in medical importance and are now a focus for progress in psychiatry. However, progress in diagnosis is not reflected in treatment protocols. Treatment is often triaged to professions without the training to make a psychiatric diagnosis and effective clinical decisions:1 sympathy is scant for a person with a perceived self-inflicted disease. It turns to aversion if that person refused to conform to the patient role and comes from socially excluded populations. Substance abuse is life-threatening, but it has favourable recovery rates after treatment. Without intervention a progressive disease is often fatal. Patients drop out of treatment for many reasons. The lack of adequate psychiatric services to help the patient to cope with an underlying psychopathology needs to be addressed. In the Netherlands, clinical management of patients with a dual diagnosis of substance-related disorder and another psychiatric illness has become a priority because proposals for new treatments such as heroin maintenance are being planned. The addiction severity index (ASI), first validated in Europe on a Dutch addict population, showed high levels of psychopathology.2 Estimates by the Dutch National Council of Public Health indicate that 40–70% of the drug-using population show evidence of psychopathology in addition to substance-abuse diagnosis. In studies in home and clinical settings that use intensive time-sampling techniques, the daily lives of such people are characterised by hidden, undiagnosed psychopathology, detectable in mood disorders and craving behaviour.3 The

655