Side Effects: Real, Imagined, and How to Tell

TL;DR. Most side effects are the drug doing exactly what it does, in a tissue where you did not want it. Some are genuinely caused by the drug; some are the illness, the ageing, or life happening at the same time; and a substantial share, especially for subjective symptoms, are nocebo: caused by expecting them. The statin muscle-pain literature is the clearest demonstration, where blinded trials found nearly all reported symptoms occurred equally on placebo. None of this means the symptoms are imaginary. It means "is this drug causing this?" is a question that can often be answered, and the answer changes what you should do.

1. Where side effects come from

Type A reactions are the ones you can reason about, and understanding the mechanism usually tells you what to expect (Chapter 62).

2. Allergy versus intolerance

These are routinely conflated on medical records with real consequences.

Nausea from an antibiotic is not an allergy, and recording it as one is how people end up denied the best drug for a serious infection for the rest of their lives. Over 90 percent of people labelled penicillin-allergic are not, and the label causes worse outcomes (Chapter 72).

The severe delayed reactions that are absolute contraindications: Stevens-Johnson syndrome and toxic epidermal necrolysis (widespread blistering, mucosal involvement, skin peeling), DRESS (rash with fever, swollen lymph nodes, and organ involvement), and acute generalised exanthematous pustulosis. Any rash with mucosal involvement, blistering, or systemic illness needs immediate medical assessment.

3. Nocebo: the mirror of placebo

Placebo is benefit from expectation. Nocebo is harm from expectation, and it is just as real, just as measurable, and considerably less discussed.

How it works: expectation activates the same neurobiology that produces symptoms. Anticipating pain increases activity in pain-processing regions; expecting nausea raises the likelihood of feeling it. The symptoms are genuinely experienced. This is not malingering and it is not imagination in the dismissive sense.

The evidence:

• In placebo arms of clinical trials, participants report side effects at high rates. A meta-analysis of placebo arms in migraine trials found participants reported the specific side effects of the active drug they thought they might be receiving.
• Being told about a side effect increases its incidence. In a trial of beta blockers, patients informed of the possibility of erectile dysfunction reported it at three times the rate of those not informed.

The statin case is the clearest demonstration in medicine:

• In clinical practice, 10 to 25 percent of statin users report muscle symptoms.
• In blinded randomised trials, muscle symptoms occur at nearly identical rates on statin and placebo.
• The SAMSON trial (2020) took 60 patients who had stopped statins because of side effects and gave them, in random order, four months of statin, four of placebo, and four of no tablet. The symptom burden during placebo months was 90 percent of that during statin months. Symptoms during no-tablet months were much lower.
• StatinWISE ran similar n-of-1 trials with the same result.
• After SAMSON, half the participants successfully restarted statins, having seen their own data.

What this does not mean: that statin muscle symptoms never occur (rhabdomyolysis is real and rare), or that patients are lying. What it does mean: that "I got muscle pain on a statin" and "the statin caused my muscle pain" are different statements, and the second can be tested.

Nocebo also explains a good deal of reported reactions to non-drug exposures: blinded provocation studies of self-reported gluten sensitivity, MSG sensitivity, and electromagnetic hypersensitivity have generally failed to reproduce symptoms when the participant does not know what they are exposed to.

4. Working out whether the drug did it

Clinicians use structured criteria (the Naranjo algorithm is the best known). The questions underlying them are usable by anyone:

1. Timing. Did it start after beginning the drug, and within a plausible interval? Some effects are immediate; some take weeks (agranulocytosis, liver injury); some appear after years.
2. Dose relationship. Does it get worse with a higher dose and better with a lower one?
3. Dechallenge. Does it improve on stopping, and over a timescale that fits the drug's half-life?
4. Rechallenge. Does it return on restarting? This is the strongest evidence, and it is only appropriate for non-serious effects and never for suspected allergy.
5. Alternative explanations. Could it be the illness, another drug, ageing, or coincidence? Age and comorbidity generate symptoms continuously, and they do not stop just because a new drug was started.
6. Is it a known effect? Check the leaflet and its frequency category.

The n-of-1 approach, as used in SAMSON, is available in ordinary practice for some drugs and under-used: try on, off, on again, ideally with a symptom diary, and see whether the pattern tracks.

5. Reading a leaflet without being frightened by it

A leaflet lists everything ever reported during trials and post-marketing that could plausibly be related. It is a legal document, not a prediction.

Use the frequency words, which have exact meanings (Chapter 64):

Leaflet-induced non-adherence is a real clinical problem. People read the list, become alarmed, and either do not start or stop early. The countermeasure is to read the "stop taking and seek medical help" section carefully, note the frequencies, and treat the rest as background.

6. What to do when you get a side effect

Do not just stop, unless it is serious. Several drugs are dangerous to stop abruptly (Chapter 62): beta blockers, steroids, benzodiazepines, antidepressants, opioids, antiepileptics, and clonidine.

The options, most of which people do not know exist:

1. Wait. Many effects fade within one to two weeks: SSRI nausea, GLP-1 nausea, metformin GI upset, opioid drowsiness.
2. Adjust timing. Take a sedating drug at night, a stimulating one in the morning, a stomach-irritant with food.
3. Reduce the dose.
4. Change formulation. Modified-release metformin resolves GI upset for many people. Liquid formulations enable finer dose steps.
5. Switch within the class. An ACE inhibitor cough resolves on an ARB; sexual dysfunction on an SSRI may resolve on bupropion or mirtazapine.
6. Switch class.
7. Add something to manage it. A PPI with an NSAID; a laxative with an opioid.
8. Accept it, if the benefit is large and the effect is tolerable. This is a legitimate choice and should be an informed one.
9. Stop the drug, if the benefit does not justify it. Also legitimate, and should be a decision rather than a silent default.

Seek urgent help for: any rash with blistering, mucosal involvement, or fever; swelling of face, lips, or tongue; difficulty breathing; sore throat, mouth ulcers, or unexplained fever on a drug that can cause agranulocytosis (carbimazole, clozapine, some others); yellowing of the eyes or skin; dark urine with muscle pain; unusual bruising or bleeding; severe abdominal pain; sudden severe headache; suicidal thoughts.

7. Reporting

Yellow Card (UK), VAERS and MedWatch (US), EudraVigilance (EU). Anyone can report, including patients directly.

Why it matters: clinical trials involve thousands of people for months. Rare effects and delayed effects only appear when millions take a drug for years. Spontaneous reporting is how those signals are found. Practolol's oculomucocutaneous syndrome, the rofecoxib cardiovascular signal, and the ranitidine NDMA contamination all emerged post-marketing.

Report even if you are unsure it was the drug. The system is designed to detect patterns across many uncertain reports; certainty in any individual case is not the requirement, and under-reporting is the system's main weakness.

8. The other side: undertreatment

A chapter on side effects should say plainly that fear of them causes its own harm.

• Statin non-adherence is associated with more heart attacks and deaths.
• Steroid phobia causes undertreated eczema and asthma (Chapter 78).
• Opioid-phobia in cancer and palliative care causes avoidable suffering (Chapter 68).
• Stopping antidepressants abruptly because of withdrawal symptoms misread as the drug being harmful.
• Refusing vaccines on the basis of misunderstood rare risks (Chapter 74).
• Stopping medication in pregnancy unilaterally (Chapter 83).

The right comparison is never "drug versus nothing." It is "drug versus the untreated condition." That comparison is the one most often missing.

9. The bottom line

• Most side effects are the drug's own pharmacology in a tissue you did not intend. Understanding the mechanism usually predicts them.
• Allergy and intolerance are different. Nausea is not an allergy, and mislabelling it costs people the best drug for the job for decades afterwards.
• Nocebo is real and large, especially for subjective symptoms. The statin trials found nearly all reported muscle symptoms occurred equally on placebo, and half of participants restarted after seeing their own data.
• The symptoms are genuinely experienced either way. "Did the drug cause this?" is a question that can often be answered by timing, dose, dechallenge, and rechallenge.
• You have more options than "carry on" or "stop": wait, change timing, lower the dose, change formulation, switch within or between classes, or treat the side effect.
• Report it. Rare effects are only ever found after marketing, and under-reporting is the main limitation.
• Fear of side effects causes real harm through undertreatment. The comparison is always against the untreated illness, not against nothing.

Sources and notes

The Type A to E adverse reaction classification follows Rawlins and Thompson and its later extensions. Allergy versus intolerance distinctions and severe delayed reaction recognition (SJS/TEN, DRESS, AGEP) follow BSACI and EAACI guidance. Penicillin allergy de-labelling and its outcome consequences follow Blumenthal et al., The Lancet, 2019. Nocebo neurobiology follows Benedetti's work and Colloca and Barsky, NEJM, 2020. The beta blocker erectile dysfunction disclosure experiment is Cocco, Heart, 2009. Statin nocebo evidence is SAMSON, Wood et al., NEJM, 2020, and StatinWISE, Herrett et al., BMJ, 2021, plus the ASCOT-LLA blinded versus unblinded phase analysis by Gupta et al., The Lancet, 2017. Blinded provocation failures in self-reported non-coeliac gluten sensitivity follow Biesiekierski et al., Gastroenterology, 2013; in electromagnetic hypersensitivity, Rubin et al.'s systematic review. Causality assessment follows the Naranjo algorithm and WHO-UMC criteria. Leaflet frequency categories follow the European Commission's guideline on the readability of labelling. Leaflet-induced non-adherence follows Berry et al.'s work on risk communication. Yellow Card, VAERS, and EudraVigilance limitations, including that reports are not confirmed causal links, are stated by those systems. Practolol and rofecoxib as post-marketing detections follow the pharmacovigilance literature.

Open questions. Distinguishing nocebo from genuine drug effect in an individual patient is only possible with an n-of-1 rechallenge, which is rarely done. Under-reporting to spontaneous surveillance systems is substantial and its size is unknown.

👉 Next: storing, expiry, disposal, and fakes.