The tablet works even when there is nothing in it

Sixty patients who had stopped taking their statins because of muscle complaints were given, in the SAMSON trial, twelve monthly packs in random order: four with the drug, four with placebo, four empty. They could not tell drug from placebo, and they recorded daily how severe their complaints were. Mean complaint intensity came to 16.3 points in the statin months, 15.4 in the placebo months and 8.0 in the months without any tablet at all. Between drug and placebo no difference could be established.

The tablet worked although there was nothing in it. This piece is about that sentence and its limits, because the sentence is routinely stretched too far. The legend says: belief heals, a third of every effect is placebo, the sugar pill comes close to the drug. The research says something more precise, and the more precise version is the more interesting one.

What is left once the sums are done properly

First the bookkeeping. That patients in the placebo arm feel better after four weeks has many causes, and expectation is only one of them. Illnesses fluctuate, and anyone who enrols at a low point would have climbed again anyway — regression to the mean. Add the spontaneous course of the condition, the attention, the imposed daily structure and the wish to report improvement to friendly staff, a textbook case of demand characteristics.

The placebo effect proper is the remainder left over when sham treatment is compared not against nothing, but against no treatment at all. That is exactly what Asbjørn Hróbjartsson and Peter Gøtzsche did: they looked for trials carrying a placebo arm and an untreated arm — 114 trials in the first pass of 2001, 158 in the extended version of 2010. Their finding was sobering, and their abstract says so without ceremony:

A dependable effect showed up only on subjective measures reported by the patients themselves, pain above all. On the objective endpoints almost nothing was left. No placebo shrinks tumours, lowers viral loads or repairs joints.

Where the effect is biologically tangible

But there are domains in which the remainder is large, reproducible and biologically tangible: pain, nausea, some Parkinsonian symptoms, fatigue, anxiety — all of them quantities the brain itself helps regulate. For pain the chain is best mapped. Expecting relief releases the body’s own opioids, and the opioid blocker naloxone can abolish placebo analgesia again. That is the classic finding of Jon Levine, Newton Gordon and Howard Fields from 1978, and it carries weight because it makes something that looks like psychology testable by pharmacology.

Fabrizio Benedetti’s work showed the fine mechanics. His trick was the hidden dose: the same painkiller is given once visibly by a doctor and once unnoticed by a programmed infusion pump. A covertly given dose of painkiller works measurably less well than the same dose given visibly — the difference is the expectation share that runs along inside every open treatment.

And this archive’s Pavlov line makes a late appearance here. Placebo responses can be produced by classical conditioning. In 1975 Robert Ader and Nicholas Cohen paired an immunosuppressant with a taste stimulus in rats — and afterwards the taste alone dampened the immune response. Marion Goebel and colleagues demonstrated the same process in humans in 2002. Expectation and conditioning are the two motors, and they often work together.

How the staging plays its part

The packaging of expectation has been measured too, and the findings read like satire but are the state of the data. Injections and sham devices work better than tablets. More expensive shams work better than cheap ones — Rebecca Waber and colleagues showed this by presenting subjects with the same inert pill, once as an expensive preparation and once as a discounted one. Brand names work better than nameless pills. And colour joins in: red, yellow and orange count as stimulating, blue and green as calming.

The review of pill colour supplies the punchline along with the finding. In Italian men the effect of blue reverses — not in Italian women. The obvious explanation is the national team’s shirt. Expectation hangs on the meaning, not on the wavelength, and the meaning is local. None of this shifts the border to the objective endpoints. All of it shows that the staging of a treatment is part of its effect on experience. Medicine long treated that insight as embarrassing; placebo research treats it as an adjustable screw — if context works, it can be made deliberately good without deceiving anyone.

What nocebo costs

The dark sister reverses the sign: negative expectation produces and amplifies symptoms. Anyone who hears in the consent talk that the drug may cause headaches reports headaches more often — in the placebo group too, which is the clean proof that the information itself acts. Nocebo effects account for a substantial share of the side effects in the sham arms of clinical trials and for part of the drop-outs from real treatments. The SAMSON trial at the top of this text is the cleanest demonstration: the difference did not lie between drug and placebo, but between tablet and no tablet.

For practice, nocebo is therefore the more important half — because it is produced daily, by package inserts, internet forums and careless sentences in the consulting room. The choice of words is not a soft topic here but a measurable one. In a study of 140 women, nothing but the announcement sentence before the epidural was varied. One half heard that a sharp bee sting was coming and that this was the worst part of the procedure. The other heard that they were being given a local anaesthetic that would numb the site. Same procedure, same needle, different pain scores. Whoever phrases the consent talk is also setting the dose. Research therefore tests formats that inform without programming: the framing of frequencies, positive context sentences, graduated information on request.

Nocebo also has a mass form. In 1998, at a high school in McMinnville, Tennessee, large numbers of pupils and teachers fell ill with headache, dizziness and nausea; the school was evacuated and investigated. No toxic cause was found. What was found was a pattern of spread along the lines of sight — those affected were above all people who could see others affected. Expectation is contagious. That, too, is a conditioning story.

What this means for judging studies

For this archive, placebo is above all a question of method, and here the circle closes to the Beck file. Drug trials can blind; psychotherapy trials cannot — nobody sits unknowing in cognitive therapy. Their controls are called waiting list, attention control or treatment as usual, and depending on the choice the effect comes out larger or smaller. Part of the quarrel over how well these procedures work is in truth a quarrel over control conditions: against a waiting list almost anything works; against a credible sham treatment the differences shrink. The dodo debate, which this journal treats in a piece of its own, hangs on exactly this question.

Expectation travels on the practitioner’s side as well. The allegiance research of Lester Luborsky and colleagues shows that procedures do better in the hands of their adherents than in the hands of sceptics. That is a placebo relative in a lab coat — and one of the reasons independent replications are worth so much.

Frankl’s folder waves from the side at this point: that interpretation and stance change the experience of a situation was the clinical intuition of his doctrine of meaning. Placebo research turned the intuition into a measurand and drew its limits. Between “expectation modulates pain” and “mind heals body” lies the whole difference between a finding and a wish.

What remains

What is secure is a narrow but genuine effect. Expectation and conditioning measurably change how strongly a symptom is experienced, and they do it through demonstrable pathways in the body. Equally secure is the limit: on objective measures of disease a sham preparation achieves nothing. Two clarifications belong here, because the legend regularly blurs them. First, placebo effects do not prove that an illness was imagined — they prove that symptom regulation runs partly through the central nervous system; a pain that responds to expectation is as real as one that does not. Second, the research on open-label placebos is still young: the much-cited study by Ted Kaptchuk and colleagues covered 80 patients with irritable bowel syndrome over three weeks, all of whom knew they were taking a sham preparation. The direction is interesting, the samples are small, and the replication crisis has taught us to hold the verdict until the large tests arrive.

In everyday life the matter can be sorted by a simple test: is an endpoint reported or measured? With pain, nausea, fatigue and sleep the placebo share is real and the nocebo share is larger still. With lab values, imaging findings and mortality it is not. And the practical consequence lies less with the sugar pill than with the sentence that precedes it: whoever explains what is about to happen changes how it will feel — in both directions.