Muscle symptoms are the most-cited reason people stop or refuse a statin, so the tolerability decision is a distinct decision from the efficacy decision — it applies to everyone already on a statin (primary and secondary prevention), and the wrong call (stopping a drug that is preventing vascular events for a symptom the drug is not causing) is common and costly. The question is not do statins ever hurt muscles (rarely, they do) but when a specific patient reports muscle symptoms, what is the probability the statin caused them, and does that probability warrant stopping? Three sources answer it, and they compose rather than corroborate: a large randomised IPD meta-analysis sizes the causal excess and its timing (Reith/CTT), a blinded n-of-1 trial isolates the mechanism in the hardest stratum (SAMSON), and a guideline systematic review gives the general-adult null (USPSTF).
The size and timing of the real effect (Reith / CTT 2022)
The Cholesterol Treatment Trialists’ Collaboration pooled individual participant data on every recorded muscle adverse event from 19 double-blind statin-vs-placebo trials (n=123,940) and four more-intensive-vs-less-intensive trials (n=30,724) — ~155,000 people, weighted median follow-up 4.3 years. Restricting to double-blind randomised trials is the design choice that isolates the causal contribution of the drug from reporting bias.
- The all-years excess is tiny. 27.1% of statin-allocated vs 26.6% of placebo-allocated reported muscle pain or weakness: «rate ratio [RR] 1·03; 95% CI 1·01–1·06». (Reith et al., 2022) A 0.5-percentage-point absolute difference over 4.3 years.
- Almost all of it is in year 1. «During year 1, statin therapy produced a 7% relative increase in muscle pain or weakness (1·07; 1·04–1·10), corresponding to an absolute excess rate of 11 (6–16) events per 1000 person-years, which indicates that only one in 15 ([1·07–1·00]/1·07) of these muscle-related reports by participants allocated to statin therapy were actually due to the statin. After year 1, there was no significant excess in first reports of muscle pain or weakness (0·99; 0·96–1·02).» (Reith et al., 2022) The absolute excess after year 1 is 0 (-2 to 1) per 1000 person-years — an effectively flat second-year-onward curve. The heterogeneity test for year 1 vs all later years was p=0·0005.
- The intensity gradient. «more intensive statin regimens … yielded a higher RR than less intensive or moderate-intensity regimens (1·08 [1·04–1·13] vs 1·03 [1·00–1·05]) compared with placebo, and a small excess was present (1·05 [0·99–1·12]) for more intensive regimens after year 1.» (Reith et al., 2022) So the higher-dose harm is larger (year 1 ~11% vs ~6% relative) and slightly more persistent — but the population-attributable fraction is still small.
- No dose-response within a statin, no between-statin difference, generalisable across strata. The RR did not differ by statin type, run-in design, or hydrophilic-vs-lipophilic solubility, and was similar across clinical circumstances — «our results are likely to be widely generalisable». The one prognostic signal: «The RR for muscle pain or weakness was greater in women for less intensive and moderate- intensity statin regimens … (RRs 1·09, 99% CI 1·03–1·16 in women vs 1·00, 0·97–1·04 in men; heterogeneity p=0·0019)» — but this sex difference was not reproduced in the more-vs-less-intensive trials, so it is a weak, unreplicated subgroup signal, not a firm effect-modification claim (route-b bar unmet). (Reith et al., 2022)
Serious muscle injury is real, rare, and also front-loaded
Ordinary aches are not the dangerous end of the spectrum. «Myopathy … was reported by 0·08% of those assigned any statin regimen versus 0·04% of those assigned placebo (RR 1·74; 95% CI 1·11–2·74, p=0·016 …), which corresponded to an absolute excess of 0·08 (0·01–0·18) per 1000 person-years. The RR was 3·04 (1·43–6·47; p=0·0039) in year 1 and 1·28 (0·72–2·25; p=0·40) after year 1; thus, in year 1, statin therapy was the cause of approximately two-thirds ([3·04–1·00]/3·04) of myopathy cases reported by patients allocated to a statin.» (Reith et al., 2022) So myopathy — unlike ordinary myalgia — is mostly drug-caused when it occurs, but it is ~50x rarer than reported aches (0.08 vs 11 per 1000 py excess). The severe tail (rhabdomyolysis) was confined to 80 mg simvastatin (now withdrawn) -> the rhabdomyolysis magnitude is held on LDL Lowering and Cardiovascular Events.
Why the reported burden is mostly not the drug — the nocebo mechanism (SAMSON)
Reith establishes that >90% of reported muscle symptoms on a statin are not caused by it, but a population RR cannot show what produces the other 90%. SAMSON supplies that within-person, in exactly the stratum that matters — people who had «previously discontinued statins because of side effects that occurred within 2 weeks after the initiation of treatment». Each of 60 patients rotated month-by-month through atorvastatin 20 mg, placebo, and no-tablet, scoring symptoms daily.
- «In patients who had discontinued statin therapy because of side effects, 90% of the symptom burden elicited by a statin challenge was also elicited by placebo.» (Wood et al., 2020)
- Mean symptom intensity 8.0 no-tablet, «15.4 during placebo months … and 16.3 during statin months … P = 0.39 for the com-parison with placebo months». (Wood et al., 2020) Taking a tablet at all roughly doubled symptoms; statin vs placebo was indistinguishable. The symptoms are real; they are not statin-specific. Full extraction on the source page.
The three sources compose — they are NOT independent backings
It is tempting to read three sources reaching most muscle symptoms are not the statin as a robust type-E convergence. It is not: they measure different quantities in different populations, and two of the three share their evidence base.
| Source | Quantity measured | Population | Design |
|---|---|---|---|
| Reith / CTT 2022 | causal excess vs placebo (RR 1.03; year-1 excess 11/1000 py) | general statin-trial adults | IPD MA of parallel double-blind RCTs |
| SAMSON (Wood) 2020 | nocebo ratio 0.90 (within-person) | symptomatic-intolerant (already stopped for side effects) | blinded n-of-1 crossover |
| USPSTF 2022 | myalgia RR vs placebo (no increase) | general adult primary-prevention | systematic review of statin RCTs |
- Reith and USPSTF are not independent — both pool largely the same double-blind statin RCTs, so their agreement is one evidence base read twice, not two.
- Reith and CTT-2010 are the same body (the CTT Collaboration, CTSU-Oxford), so Reith’s benefit side — «statins might cause 11 (generally mild) episodes of muscle pain or weakness, but prevent 50 major vascular events in those with pre-existing vascular disease … and 25 major vascular events in those without» (Reith et al., 2022) — is CTT’s own prior efficacy work restated (cited to ref 2), not an independent confirmation of the benefit magnitude -> LDL Lowering and Cardiovascular Events.
- Reith and SAMSON differ in design, population, and quantity, so they compose (population magnitude
- within-person mechanism) rather than independently confirming one number. The composite — a small causal excess confined to year 1, with the rest of the reported burden reproducible by an inert pill — is what no single source states. That is the beyond-summary move here, and it is type-A/F, not type-E.
The nocebo-isolation design generalizes across drug classes [E-independent]
The cross-class phenomenon is owned at concept altitude by The Nocebo Component of Drug Side-Effects; this section is the statin instance of it.
Within the statin evidence the most symptoms are not the drug finding is type-A/F (above). But the
design that isolates it — an inert comparator separating a self-reported drug symptom from its nocebo
component — recurs in a wholly separate drug class, and there it is genuine independent backing. Henssler’s
2024 SR+MA of antidepressant discontinuation symptoms finds «approximately half of antidepressant
discontinuation symptoms could be attributable to expectation or non-specific symptoms» — corroborated
by Henssler 2024 (chunk 01) [E-independent]: same placebo-arm logic, a
different drug, a different symptom (withdrawal rather than an ongoing ache), a different design, and a
non-overlapping author group (Imperial cardiology vs a German psychiatry team — the author-diff
independence check passes). Independent backing, kept a body line and not a sources: pad.
| Parameter | Statin (Wood/SAMSON) | Antidepressant (Henssler) | Same quantity? |
|---|---|---|---|
| what is isolated | nocebo part of a reported ongoing muscle symptom | nocebo part of a reported discontinuation symptom | pattern YES, phenomenon NO |
| nocebo share | «90% of the symptom burden» also on placebo (within-person) | «approximately half» attributable to expectation (placebo-arm proportion) | NO — different metric + design |
| inert comparator | blinded n-of-1 (statin / placebo / no-tablet) | placebo arm of discontinuation RCTs | related, not identical |
What is independently confirmed is the qualitative claim — a large fraction of a self-reported drug side-effect is nocebo, and only an inert comparator reveals it — not a shared number (90% of a within-person burden and ~50% of a discontinuation-symptom incidence are different metrics on different phenomena). The design travels across drug classes; the magnitudes do not. Shared decision consequence: do not discontinue a tolerated drug on reported symptoms alone; use a blinded rechallenge or a placebo-controlled taper to separate real pharmacological harm from the nocebo component -> Antidepressants for Depression, The Observational-Trial Discordance.
Decision relevance — the deprescribing rule
- A symptom in someone who has tolerated a statin for a year or more is very unlikely the statin. The excess is confined to year 1, so the prior probability that a new symptom after that is drug-caused is near baseline. «This finding is particularly true if the treatment has been well tolerated for a year or more before developing symptoms; but, even during the first year of a moderate-intensity statin regimen, it is likely to be the cause in only approximately one in 15 patients who report muscle symptoms, rising to approximately one in 10 in those who are taking a more intensive regimen.» (Reith et al., 2022)
- Continue, don’t stop reflexively. Reith’s own management implication: «for patients who report mild muscle symptoms when taking a statin … it is most likely that the symptoms are not due to the statin, and statin therapy should continue until other potential causes have been explored.» (Reith et al., 2022) (Serious myopathy — pain with markedly raised creatine kinase, not an ordinary ache — is the exception and does warrant stopping; it is rare and usually front-loaded.)
- The benefit-harm trade runs strongly toward continuing. The muscle harm is front-loaded and mild; the cardiovascular benefit grows with duration («9% [99% CI 3–15%] in year 1 vs 24% [95% CI 21–26%] in years 1 to ≥5»). (Reith et al., 2022) So the longer a person stays on, the more the ledger favours staying on.
- A dose de-escalation is a lever short of stopping. Because the excess is intensity-graded, a person with genuine year-1 intolerance to a high-intensity regimen can often move to a moderate-intensity one (smaller excess, most of the LDL benefit) rather than abandoning the drug — a substitution, not a binary.
- This sizes the rock (pharmacotherapy taper). The statin muscle-symptom scare is a large driver of non-adherence, and the evidence says the scare is mostly nocebo — so the tolerability limitation that appears to shrink the statin lever mostly does not. This is the standard-drug LIMITATION appraisal the taper calls for: appraised, the limitation is small, which keeps the statin’s marginal rank high.
- The statin’s OTHER harm — new-onset diabetes — differs from this one in KIND. Muscle symptoms are >90% nocebo (a perceived harm the drug mostly does not cause); the glycaemic effect is a small but real pharmacological harm, concentrated in people already near the diabetes threshold. The two harms therefore call for opposite handling — reassure-and-continue for muscle aches, versus a genuine (still-outweighed) metabolic cost to weigh -> Statins and New-Onset Diabetes. (Both endpoints come from the same CTT body, so their appraisal is one lab’s work, not independent confirmation.) (inferred from Reith et al., 2024)
Limits
- Population is trial-enrolled and largely older/higher-risk (mean age 63, ~48% with vascular disease); most trials completed enrolment before statins were generic and did not screen out today’s “statin-intolerant” phenotype, so the estimate is for the general treated population, not for a pre-selected intolerant person (that is SAMSON’s stratum).
- Ascertainment was heterogeneous (first-year reporting rates ranged 1.0%–60.5% across trials); the RR was stable across that range, but symptom capture was not designed as a primary endpoint in the source trials.
- The open loop (R1). No source here randomises a deprescribing strategy (continue-and-explain vs stop) against patient-important outcomes — the continue until other causes explored rule is inferred from the causal-attribution evidence, not tested as an intervention. SAMSON’s 50% naturalistic restart rate is an observation, not a tested rescue.
(inferred from Reith et al., 2022; US Preventive Services Task Force, 2022; Wood et al., 2020)