9.2 GOLOMB B. A. (1995). Paradox of placebo effect. Nature 375(6532): 530
Paradox of placebo effect
SIR - When
testing for the effectiveness of a drug in ameliorating a disease, it is common practice
to compare the treated group to a control population, matched in all essential respects,
to whom is administered a placebo containing substances that are presumed to be
inert. The placebo controlled trial is an integral part of
evidence-based medicine. Although the nonspecific effects of placebos are widely studied,
the possibility that the chemicals used as placebos may have specific effects has received
virtually no attention.
The US Food and
Drug Administration sets no regulations on the constituents of placebos, and any
guidelines are at best informal. Astonishingly, no systematic efforts are made to ensure
the inertness of placebos: there is nothing validating
the placebo standard against which other agents arc measured. Further, the drug companies
funding the trials control the placebo ingredients.
The identity of
the placebo and fillers used with the experimental drug are rarely stated inscientific
studies. In one exception to this practice, several early papers exploring the use of
cholesterol-lowering agents to curb heart disease did in fact name the placebos used:
olive oil in one case 1 , and corn oil in another
2 . Mono- and poly-unsaturates such as olive oil and corn oil are now
widely known to decrease low-density lipoproteins 3 , so
that with hindsight these agents may not have been inert with respect to the outcome
studied. Indeed, it was noted in one such study 2 that
the rate of cardiac mortality was lower in the placebo group than expected.
How can we be
sure that placebos are free of specific effects ? Few if any agents are truly inert, and
placebos are given systematically over prolonged periods. Even substances that are not
absorbed, such as methylcellulose (which reduces cholesterol), can have significant
across trials may differ: such differences are among many factors that may underlie
outcome differences in otherwise similar trials. Placebos used across trials may, on the
other hand, be similar, or have similar effects: thus
we cannot rely on meta-analvses to cancel differing specific effects of placebos. In
addition, it is felt by many that the correct solution to the thorny problem of
irreconcilable smaller studies is a single very large trial rather than a meta-analysis.
Indeed, in some fields the evidential basis of a population-based treatment regimen rests
on the outcome of a single significant trial. The Scandinavian Simvastatin Survival Study
4 is the lone cholesterol-lowering trial, among many, that showed
improved overall mortality with cholesterol reduction during the time of treatment, even
in the secondary prevention population; it is cited as proof of the benefits of
cholesterol reduction therapy. Clearly, small effects by a placebo may be pivotal when
clinical practice is determined by the results of a single very large trial, whose large
size is necessitated by small effect size of the treatment.
the possibility of small effects of placebos accruing over long periods may not have been
of serious concern because most trials sought large effects in small subject populations
studied over relatively short times. Large-scale trials (or aggregate analyses of trials)
involving thousands of individuals studied over many years and seeking small effects (see,
for example, ref. 5) are a modern phenomenon. It is in this setting that small beneficial
or harmful effects of placebos could be significant. An apparent positive, negative or
null effect of a drug may instead be the consequence of a negative, positive or same-direction
effect of the placebo.
What should be
done ? First, it is essential that all studies should state the composition of the
placebos as well as all fillers included with the drugs. Studies should be carried out to
test for possible specific effects of placebo agents, even though this process will be
fraught with difficulty. Meanwhile, in interpreting the results of prior trials, we should
be aware that possible specific effects of placebos represent a potential confounding
factor that may be vital to interpreting the study
of evidence-based medicine is undermined by the absence of evidence that placebos are
inert. It is paradoxical that there is no standard of evidence to support the standard of
Department of Medicine,
of California, Los Angeles, Los Angeles,
California 90024, USA
Acheson,J & Hutchinson, E. C. Atherosclerosis 15,177-183 (1972).
Research Committee of the Scottish Society of Physicians, Br Med. J. 4,755-784
Grundy, S.M. & Denke,M A J Lipid Res 31,1149-1172 (1990).
Scandinavian Simvastatin Survival Study Group, Lancet 344,1383-1389 (1994)
Lau,J et al New Engl. J. Med.327,248-254(1992).