Acetaminophen (Paracetamol)

Tertiary referral centres see the worst cases, so 42% is acetaminophen's share of the liver failures that reach those units and never a risk to anyone taking it. Doses in the accidental group were reconstructed from what patients recalled afterwards. The three routes overlap, so one person can sit in all of them and the percentages do not add to a hundred. The narcotic-combination finding is US-specific, since those products are not sold everywhere.

Single doses, in adults, for pain after an operation. That is the setting this drug performs best in, and it is not the setting most people take it in. Adverse-event reporting across the pooled trials was patchy, so the reassuring safety picture is thinner than the efficacy picture. The review dates from 2008 and has not been updated since.

Fourteen days in hospital with every meal provided is not ordinary home use, and 145 people is a small sample for a safety question. The 28-day study used 3 g rather than 4 g, so the two rates are not measuring the same exposure. Its placebo group had an 11% rate of its own, a reminder that this enzyme moves without any drug at all. The 4 g trial ran inside a hydrocodone product development programme, with several industry-affiliated authors.

A retracted trial leaves an absence of evidence, not evidence that the drug fails, and only the first of those is true here. Cochrane's inclusion rules were strict, which is why three trials survived a search of a much larger literature. This covers low back pain with no identified cause, not back pain traced to a known injury or disease.

No drug was tested against another here. These are numbers from separate trials in separate people, lined up side by side. The authors flag 13 of the drug and dose pairs as possibly affected by publication bias. All of it is a single dose for pain after an operation, which says nothing about taking either drug for weeks. And the arithmetic that makes the pair work is the same arithmetic that makes an unnoticed second acetaminophen product dangerous.

Short-term outcomes only, out to twelve weeks, so this says nothing about months of use. A group average does not rule out individuals who feel better on it. A separate network of 98 trials covering 69 medicines for the same condition rated confidence in every analgesic tested as low or very low, which is a different verdict from a confident null and worth holding alongside it. Four authors, lead and senior among them, had GlaxoSmithKline support for one of the pooled trials, a conflict that runs against the finding rather than towards it.

The nine-point line comes from a companion review by the same authors, not from this analysis, and other groups draw it elsewhere. The 192-trial network excluded any trial with fewer than 100 people per group and anything published outside English. Averages hide people, and some individuals in these trials did improve. Patients on paracetamol were also nearly four times as likely to show an abnormal liver function test, a laboratory result the authors say they cannot interpret clinically.

An umbrella review inherits every weakness of what it reviews and adds no data of its own, so it can say the evidence base is poor without settling the question underneath it. Failing to demonstrate a link is not the same as demonstrating there is none, and the authors word their conclusion as the former. It covers autism and attention-deficit disorder only. Its search closed in September 2025, before the sibling-matched work published in 2026.

This is frequent episodic tension-type headache of moderate or severe intensity, not migraine, which the trials generally excluded, and not the chronic form. Ask an easier question and the number improves: for pain-free or only mild pain at two hours it is 10, and for avoiding a second painkiller it is 7.8. No trial was at low risk of bias on every count, though the reviewers put most of that down to thin reporting rather than bad conduct. The senior author declares grants and fees from analgesic manufacturers.

This was an open treatment series with partly historical comparisons, because withholding an antidote to build a control group was not available. It dates from 1988 and the regimens have changed since, most notably to a shorter two-bag drip. The time bands rest on when people said they took the tablets, which is unreliable after an overdose and missing altogether when doses were spread across days. What survives all of that is the steepness of the gradient, which later guidelines reproduce.

One specialist transplant unit between 1992 and 2008, so these are the most severe cases and the mortality figures are not overdose mortality generally. Ingestion times and patterns were reconstructed from records after the fact. The repeat-dose group did die more often, 37.3% against 27.8%, but that gap did not hold up once age and alcohol misuse were accounted for, so it travels with those rather than clearly causing the worse outcome. Treatment of both patterns has changed since.

One urban public hospital clinic where 78% of parents had limited health literacy, so these are not general-population error rates. The benefit was concentrated in that group and was not detectable among parents with adequate literacy. It was a supervised measurement task rather than several days of dosing at home, and no child was harmed in it. The product tested was the older concentrated infant drops with a dropper, a format the United States has since changed.

Neither number marks a point where injury begins, and reading either as one is the most common mistake made with this drug. This is one product from one manufacturer in one country: other acetaminophen products, Tylenol's own included, print different strengths and different ceilings, and labels outside the United States differ again. The two figures also count differently, one per product and one across everything taken, which is exactly what makes them easy to swap.

None of these three figures is the dose at which injury begins, and the 10,000 mg line is set deliberately below where harm is expected so that clinicians act early. That threshold is also 200 mg per kilogram of body weight wherever that comes to less, so it is lower for a smaller person. It is Australian and New Zealand guidance, and the United Kingdom and United States set their own numbers. The 24,000 mg median describes people who had already reached liver failure, not a dose at which anyone should expect to.

Sibling comparison removes what families share but keeps only the families whose pregnancies differed, which is an unusual subset by construction, and it magnifies any error in recording who was exposed. Exposure came from antenatal notes and dispensing records in countries where paracetamol is sold freely, so a large share of real use is invisible: the Hong Kong figure of 43% exposed sits well below what surveys find. A null across these three diagnoses is not a clean bill of health for outcomes nobody measured. One author of each study declares payments from law firms or consulting in an area under active litigation.

Whether paracetamol in pregnancy raises autism or attention-deficit risk is still open. Comparing siblings best separates the drug from the family and finds no link; measuring it in cord blood best shows who was exposed and finds one.

People were showing what they would do, not observed doing it at home, so this measures intent and understanding rather than behaviour. A public hospital clinic and a community practice were among the four sites, and the sample leans towards lower health literacy, which was itself one of two independent predictors of going over. The threshold scored against is the 4,000 mg liver warning rather than the lower per-product ceiling many labels now print, so against that figure the rates would be higher still.

Eleven people took acetaminophen in this trial, out of 36 across three arms, and one muscle group was measured. Healthy older adults on a supervised programme are not a general population and not athletes. The dose was 4 g a day, the labelled maximum, which carries liver considerations this trial was neither designed nor powered to detect. The drug did nothing to the untrained hamstrings, so whatever is happening depends on the training rather than on the drug alone.

There is no sibling or family-based control here, which is the design that dissolved associations of exactly this shape. Cord blood at delivery captures the hours and days before birth, so a gradient in late exposure is standing in for exposure across a whole pregnancy. Fever, infection and pain are themselves linked to these outcomes and are the reason the drug gets taken, which stratifying by recorded indication does not fix. This cohort is low-income, urban and enriched for preterm birth, and 26% of it had an attention-deficit diagnosis against a few percent in the general population.

Measuring paracetamol in cord blood is a better way to know who was exposed, and here it found a link. What it cannot do is separate the drug from what leads a family to take it; sibling studies can, and find none. The question stays open.