Image: Mpelletier1, CC BY-SA 3.0, via Wikimedia CommonsThe science of the placebo
In 1955 the anaesthetist Henry Beecher published "The Powerful Placebo", which claimed on the basis of 15 trials that about 35% of patients were satisfactorily relieved by an inert treatment. The figure was quoted for fifty years. It was also wrong, because Beecher had compared the placebo groups with nothing. Patients get better on their own, symptoms fluctuate and are measured when they are at their worst, and people in trials want to please. To find out what the placebo does you need a third group that gets no treatment at all, and the trials that have one tell a more careful story.
How large it is
Asbjørn Hróbjartsson and Peter Gøtzsche collected those trials for the Cochrane Collaboration: 202 trials in 60 conditions in which patients were randomized to placebo or to no treatment. Their 2010 review is at the Cochrane Library. Across all conditions and outcomes the placebo effect was small and often absent. For outcomes that the patient reports, such as pain and nausea, it was real and moderate, with an average benefit in pain of about a third of a standard deviation, which for a typical pain scale is close to a point out of ten. For outcomes that a doctor measures with an instrument, such as blood pressure or a tumour, there was nothing. The placebo changes what people feel and report; it does not shrink tumours, and anyone who tells you otherwise is quoting Beecher.
Within its domain the effect is not trivial. In trials of antidepressants, people on placebo improve by about 80% as much as people on the drug, and the difference between the two, averaged over trials, is a Cohen's d of about 0.3, which Irving Kirsch and colleagues showed in 2008 in a free paper in PLOS Medicine that started a debate the masterclass takes up in its article on antidepressants. Most of the placebo improvement in depression is probably the natural course of the illness plus the attention of the trial, but a part of it is the pill.
Where the relief comes from
The evidence that a placebo does something in the body rather than only in the report comes from pain. In 1978 Jon Levine, Newton Gordon, and Howard Fields gave patients recovering from dental surgery a placebo, which relieved pain in some of them, and then naloxone, a drug that blocks the body's opioid receptors. The relief disappeared. The placebo had been working through the same endorphin system that morphine uses. Later studies with brain imaging showed the opioid release directly, and a 2001 study in Parkinson's disease, in which patients expected an injection of a dopamine drug and received salt water, measured a release of dopamine in the striatum comparable to that from amphetamine. The review by Tor Wager and Lauren Atlas in Nature Reviews Neuroscience in 2015, at the journal, collects the evidence.
Two mechanisms are at work. One is expectation: a person who believes that relief is coming shows activity in the prefrontal cortex that engages the brain's own pain-control pathways before the stimulus arrives. The other is conditioning: a person who has taken a real painkiller several times in a particular capsule responds to the capsule alone, in the way Pavlov's dogs responded to the metronome, and this component works even for responses the person cannot consciously influence, such as the immune response. Both are forms of learning, and both can be studied with the methods of Unit 4.
The nocebo
The same machinery runs in reverse. Tell a trial participant that a drug may cause headaches and a proportion of those on placebo report headaches. In statin trials, muscle pain is reported by nearly as many people on placebo as on the drug, and a trial in 2020 that gave patients alternating months of statin, placebo, and nothing found that about 90% of the symptom burden attributed to the drug occurred on placebo as well. The nocebo effect is the reason a doctor's description of side effects is itself a medical intervention, and the reason that a warning on a packet changes what the packet does.
When the patient knows
The ethical objection to placebos in practice has always been that they require deception. In 2010 Ted Kaptchuk and colleagues tested whether the deception was necessary. Eighty patients with irritable bowel syndrome were randomized to no treatment or to pills in a bottle labelled "placebo pills", with the explanation that placebo pills have been shown in studies to produce improvement through mind-body processes. After three weeks the open-label placebo group had improved significantly more than the no-treatment group, by an amount comparable to that in trials of the drugs used for the condition. The paper is free at PLOS ONE, and the result has been repeated in chronic back pain and in cancer-related fatigue. The effect is smaller than with a concealed placebo, and it is not clear how long it lasts, but it exists, which is hard to explain if the effect requires belief that the pill is a drug.
Expectation as prediction
The explanation that has emerged over the past decade fits the placebo into a wider picture of what the brain does. On the view set out in Unit 15 of the masterclass, perception, including the perception of pain, is an inference that combines the signal from the body with a prediction of what the signal should be. A strong prediction of relief, from a confident doctor, a familiar ritual, or a bottle with a label, shifts the inference, and what the person feels is the result. The signal from the tooth has not changed; the weight given to it has. The same account explains why the colour and price of a placebo matter, why injections work better than pills and surgery better than injections, and why the effect is largest for symptoms, which are inferences, and absent for tumours, which are not.
Seen that way, the placebo is a measurement of how much of a symptom is prediction, and that fraction, for pain, is large enough to matter and to be used.


