— this page is the wiki’s own synthesis. It owns, at concept altitude, a pattern that two held drug-class instances demonstrate and route to it; the verbatim figures and their provenance live on those instance pages, not re-extracted here.
A large fraction of what people report as a drug’s side-effect is not the drug — it is expectation. The claim is not that reported symptoms are imagined; they are real symptoms, but a symptom felt while taking a pill is caused partly by the pharmacology and partly by the act of taking a pill you believe may harm you. Self-report cannot separate the two, and the separation changes what you should do. This is the side-effect-side mirror of The Observational-Trial Discordance: unblinded observation inflates a drug’s apparent adverse-event rate the same way it inflates its apparent benefit, and only blinding strips the expectation component out.
You need an inert comparator — self-report alone cannot reveal the nocebo share
The nocebo component is invisible without a comparator the person cannot distinguish from the real drug. Three designs isolate it, each answering a slightly different question:
- Blinded n-of-1 rechallenge. The same person cycles through drug, identical placebo, and no-tablet, scoring symptoms daily. Isolates the nocebo share within one person, in the stratum that matters most — someone who already stopped the drug for side-effects -> Statin Muscle Symptoms and the Nocebo Effect (SAMSON).
- Placebo arm of parallel double-blind RCTs. The between-group randomised estimate: what fraction of the reported-symptom rate appears in people randomised to an inert pill. Answers the population question that the n-of-1 cannot -> Statin Muscle Symptoms and the Nocebo Effect (Reith / CTT).
- Placebo-controlled taper / placebo-discontinuation arm. For a withdrawal symptom, the inert-pill arm of a discontinuation trial isolates how much of the discontinuation syndrome is expectation rather than pharmacological rebound -> Antidepressants for Depression (Henssler).
The common element is an inert comparator the participant cannot tell from the drug. A dose-timing association, a within-person symptom diary, or a plausible mechanism does none of this work — without the blind, the expectation component rides along undetected.
The pattern holds across drug classes — a verified independent instance [E-independent]
Two instances the fabric holds reach the same qualitative claim by genuinely independent routes:
- Statin muscle symptoms. In people who had stopped a statin for side-effects, roughly 90% of the symptom burden a statin challenge produced was also produced by an identical placebo (within-person, blinded n-of-1) -> Statin Muscle Symptoms and the Nocebo Effect.
- Antidepressant discontinuation. About half of antidepressant discontinuation symptoms are attributable to expectation or non-specific effects rather than the drug (placebo-discontinuation arm) -> Antidepressants for Depression.
The independence is real and checked: a different drug, a different symptom (an ongoing ache versus a
withdrawal syndrome), a different design, and a non-overlapping author group — Imperial cardiology versus a
German psychiatry team, so the author-list-diff independence test passes. What is independently backed is
the qualitative pattern — a large share of a self-reported drug side-effect is nocebo, and only an
inert comparator reveals it. The magnitudes are not commensurable (90% of a within-person symptom
burden and ~50% of a discontinuation-symptom incidence are different metrics on different phenomena), so
the pattern is [E-independent], not a pooled number. The design travels across drug classes; the
magnitude does not.
Decision relevance — do not deprescribe on reported symptoms alone
- The rule. Do not discontinue a tolerated drug on reported symptoms alone. Where the question matters, use a blinded rechallenge or a placebo-controlled taper to separate real pharmacological harm from the nocebo component before deciding.
- Time since starting is diagnostic. A drug’s genuine early-onset harm is usually front-loaded, so a symptom appearing after long, uneventful tolerance carries a low prior of being the drug — the statin case makes this concrete (excess muscle events confined to year 1) -> Statin Muscle Symptoms and the Nocebo Effect.
- The stakes are asymmetric. Stopping an effective drug on a nocebo symptom forfeits a real benefit to avoid a symptom the drug is largely not causing. This is the loss-function move from The Estimate-to-Action Gap: weigh the reported symptom against the benefit given up, not in isolation.
- The serious exception still stops the drug. True pharmacological harm exists and some of it is dangerous (rare severe myopathy, not an ordinary ache); the rule is do not stop reflexively on an unverified symptom, not ignore side-effects. Objective signs and a real mechanism override the nocebo prior.