Paracetamol

Why this poison is interesting

Paracetamol has almost no interesting pharmacology in overdose. It does not block a channel, occupy a receptor or uncouple anything. At ten times the therapeutic dose the parent molecule is still doing essentially nothing — which is why the patient sitting in front of you at four hours feels fine, and why they are entitled to feel fine.

What has changed is not what the drug does but which way it leaves. At therapeutic doses, 80–90% of a paracetamol dose is conjugated with glucuronic acid or sulfate and excreted by the kidney.² A small fraction takes a different exit: oxidation by cytochrome P450 to NAPQI, an electrophile so reactive that it is neutralised by glutathione within the same hepatocyte that made it. Nobody notices, because nothing survives long enough to be noticed.

In overdose the sulfation pathway saturates.² The oxidative route does not saturate, because it was never near its capacity. So a fixed minor percentage becomes a rising absolute quantity, the glutathione that used to mop it up is consumed, and NAPQI begins to bind the proteins around it. Nothing about the drug's mechanism of action is involved at any point.

A poison is a drug whose kinetics have escaped its pharmacology.

The toxic principle

The poison is N-acetyl-p-benzoquinone imine — NAPQI — a quinone imine formed by cytochrome P450 oxidation of paracetamol. It is a strong electrophile and oxidant, and it reacts readily with thiol groups: first with the free thiol of glutathione, and once that supply is short, with cysteine residues in cellular proteins.¹

Three properties follow from that, and between them they explain the entire clinical course.

• It never travels. NAPQI is too reactive to leave the cell that made it, which is why the injury maps onto the distribution of the enzyme that makes it rather than onto blood flow. Established
• It is consumed by a finite store. Glutathione is a buffer, not a catalyst — every molecule of NAPQI it neutralises costs a molecule of glutathione. A buffer that is being regenerated slower than it is spent will eventually run out, and the point at which it does is not visible from outside the patient. Established
• It preferentially adducts mitochondrial proteins. Total protein binding turns out to be a poor predictor of injury; binding to mitochondrial proteins specifically tracks it much better.¹ Inferred

Toxicokinetics

Metabolism and the metabolites

Paracetamol leaves the body by three routes, and the entire difference between a headache tablet and a liver transplant lies in their relative sizes.

Nothing in that diagram is switched on by the overdose. The oxidative branch is not induced, and no new enzyme appears. Sulfation simply runs out of capacity while oxidation does not, so the fraction going the dangerous way rises.² It is worth being precise about this, because it is often taught as though overdose activates a toxic pathway. It does not. It removes the competition. Established

What changes the size of the dangerous branch

• CYP2E1 is the principal activating isoform. Cyp2e1 knockout mice are less susceptible to paracetamol liver injury, and mice lacking β-catenin — which almost eliminates both Cyp2e1 and Cyp1a2 protein — are resistant.¹ Established
• CYP1A2, 2D6 and 3A4 can also activate paracetamol in various model systems — but Cyp1a2 single-knockout mice were not protected,¹ so the contribution of the non-2E1 isoforms in humans is much less secure than the textbook enzyme lists imply. Inferred
• Glucuronidation varies by genotype. UGT2B15 2/2 subjects show higher paracetamol–protein adduct concentrations than 1/2 and 1/1 individuals.² Established
• Glutathione supply is a nutritional variable, not a fixed constant — which is the honest version of the claim that malnutrition and chronic alcohol use increase risk. See the appraisal below for why UK practice stopped acting on it. Inferred

From NAPQI to a dead hepatocyte

The second half of the pathway is where modern understanding has moved furthest from the version most of us were taught, which stopped at covalent binding kills the cell.

Elimination and accumulation

Paracetamol accumulates in almost none of the ways this library will keep returning to. There is no ion trapping (a weak acid with pKa ≈ 9.5 is un-ionised at every physiological pH, so urinary alkalinisation does nothing¹), no enterohepatic recirculation worth managing, no rise in free fraction because binding was low to start with,³ and no rise in apparent volume of distribution. What accumulates is not the drug.

• The conjugates are renally cleared. Paracetamol glucuronide, sulfate and the glutathione-derived conjugates are exported from the hepatocyte by transporters at the canalicular (Mrp2, Bcrp) and basolateral (Mrp3, Mrp4) membranes and then excreted.¹ Renal impairment retains conjugates, not the parent drug, and the conjugates are not toxic. Established
• Adducts accumulate, and can be measured. Paracetamol–protein adducts appear in human serum after therapeutic doses,¹ and adduct assays are an active research route to a biomarker that reflects the metabolite rather than the precursor. Not yet a bedside test in the UK. Established
• Half-life is a lagging indicator. It lengthens first because conjugation is saturated, and later because the liver making the conjugates is dying. Those are opposite prognoses reported by the same number. Inferred

Dialysability, and the EXTRIP position

The EXTRIP workgroup reviewed 24 articles, clinical data on 135 patients and toxicokinetic data on 54, and graded the overall quality of evidence as very low for all recommendations.³ Their general statement is that ECTR is suggested in severe paracetamol poisoning (2D) while being clear that "NAC is the only treatment required for the vast majority of patients".

• Recommended if the concentration exceeds 1000 mg/L (6620 µmol/L) and NAC has not been given (1D).³
• Recommended with altered mental status, metabolic acidosis and raised lactate if the concentration exceeds 700 mg/L (4630 µmol/L) and NAC has not been given (1D).³
• Recommended with those same features if the concentration exceeds 900 mg/L (5960 µmol/L) — even if NAC is being given (1D).³
• Not recommended on the basis of the reported ingested dose if NAC is being given (1D); not suggested on the reported dose alone even without NAC (2D), or on the concentration alone if NAC is given (2D).³
• Intermittent haemodialysis is preferred (1D), and acetylcysteine must be continued during dialysis at an increased rate (1D) — it is dialysable too.³

Target organs — and why those

Paracetamol does not injure the liver. It injures one third of the liver, reproducibly, and the reason it picks that third is the clearest example of organ selectivity as a mechanism in this whole library.

Timeline of effects

Paracetamol's clinical course is the standard four-phase description. What matters is not memorising the phases but noticing where the two tracks come apart — and they come apart immediately, for a full day.

What the mechanism predicts at the bedside

Why the nomogram starts at four hours

Because absorption of immediate-release paracetamol is generally complete by about four hours,³ and a treatment line is derived from post-absorptive elimination kinetics. A level taken at two hours is sampled on a rising limb and cannot be plotted against a line that assumes the peak has passed. The four-hour rule is not a safety margin; it is the point at which the graph's assumptions start being true. Established

Why a single early level can be reassuring and wrong

Gastric emptying is rate-limiting for absorption.¹ Anything that slows it — a large tablet load, a modified-release preparation, an opioid or anticholinergic co-ingestion — shifts the peak later without changing the total dose. The concentration curve is flatter and later, not smaller. This is the mechanistic reason that staggered ingestion, unknown timing and modified-release preparations all fall outside the nomogram rather than merely at its edge. Established

Three ingestion patterns, and why the nomogram fits only one

UK practice separates paracetamol ingestion into acute, staggered and therapeutic excess, and the separation is mechanistic rather than administrative. The treatment line assumes three things: that there was one ingestion at one identifiable time, that the sample is taken after absorption is complete, and that the resulting concentration therefore stands in for the total NAPQI still to be generated. Only the first pattern satisfies all three, and the other two fail differently.

Why a staggered overdose is worse than the same total dose taken at once

The mechanism predicts this before any data are consulted. Glutathione is a finite buffer that regenerates over hours, not a catalyst. A single large ingestion depletes it once and steeply, and the liver then rebuilds it. Ingestions spread across a day deplete it, allow partial recovery, and deplete it again — so each successive fraction of the dose meets a smaller reserve than the one before. The variable that matters is not how much was swallowed but how much buffer was standing when each part of it arrived, and no single plasma concentration reports that. Inferred

Why late presentation is its own risk factor

In the same cohort, of the 396 single-time-point overdoses with reliable timings, 178 (44.9%) presented to medical services more than 24 hours after the overdose — and delayed presentation beyond 24 hours was independently associated with death or liver transplantation (OR 2.25, 95% CI 1.23–4.12, P = 0.009).¹⁰ The mechanism is the one already described: acetylcysteine acts upstream of adduct formation, so every hour before it starts is an hour of unopposed NAPQI generation, and by 24 hours the mitochondrial cascade is downstream of anything the antidote can do. The label puts the same point plainly — the maximum protective effect is obtained up to 8 hours post-ingestion, and the effectiveness of the antidote declines sharply after this time.⁹ Established

Why therapeutic excess is the pattern that hides

Repeated supratherapeutic ingestion is taken for pain rather than for self-harm, which changes who presents, when, and with what history. There is no ingestion time to work from, the concentration may be low or undetectable while injury is already established, and — critically — the label's own risk factors act on precisely the variables that repeated dosing probes: enzyme-inducing drugs, regular excess alcohol, and likely glutathione depletion through eating disorder, cystic fibrosis, HIV infection, starvation or cachexia.⁹ A single acute overdose overwhelms a normal buffer; therapeutic excess finds a buffer that was already small. Inferred

Why the UK treats on one line

In September 2012 the Commission on Human Medicines recommended a single 100 mg/L nomogram treatment line, an end to risk-factor assessment, treatment of all staggered and uncertain ingestions, and a longer initial acetylcysteine infusion (15 → 60 minutes).⁴ The mechanistic logic is defensible: the risk factors being stratified on — enzyme induction, glutathione supply — are real variables that nobody can measure at the bedside, and a threshold that pretends otherwise is a false precision.

Why activated charcoal is not a one-hour drug here

The universal teaching that activated charcoal is useful only within one hour of ingestion is traditional teaching where paracetamol is concerned. Traditional teaching In the ATOM-2 cohort of 200 massive (≥40 g) paracetamol overdoses, charcoal was given at a median of 2 hours, and those receiving it within 4 hours had a substantially lower paracetamol ratio than those who did not — 1.4 (IQR 1.1–1.6) versus 2.2 (IQR 1.5–3.0), p < 0.0001 — with lower rates of hepatotoxicity (unadjusted OR 0.12, 95% CI <0.001–0.91).⁵

Why the antidote works where it works

Acetylcysteine restores the substrate for the reaction that removes NAPQI. That places it upstream of adduct formation and does nothing about anything downstream of it. So its effect should be excellent before glutathione is exhausted, good while NAPQI is still being generated, and progressively weaker once the mitochondrial cascade is self-sustaining — which is exactly the time-dependence observed. Inferred

Two consequences that follow from the mechanism rather than from a protocol: waiting for a level costs NAPQI-generating time where the history suggests a large, late ingestion — every hour spent waiting is an hour of unopposed NAPQI generation, and TOXBASE is where that trade-off is actually set. And acetylcysteine retains benefit in established hepatic failure, well beyond the window in which it prevents injury — which the antioxidant and microcirculatory effects, rather than the glutathione effect, are usually invoked to explain. Inferred

The antidote, from the poison's side

Read from the poison's side, acetylcysteine is an almost exact answer to the lesion:

• It supplies cysteine, the rate-limiting substrate for glutathione synthesis — restoring the buffer that NAPQI consumed. This is the primary mechanism and it addresses the actual failure. Established
• It substitutes for glutathione directly at high concentrations, conjugating NAPQI as a thiol donor in its own right. Inferred
• It increases sulfation, diverting paracetamol away from the oxidative branch — attacking the fork itself rather than its product. Inferred
• It has non-specific antioxidant and microcirculatory effects, which are the usual explanation for benefit in established liver failure, long after there is any NAPQI left to scavenge. Inferred

Critical appraisal

1. The 70% glutathione-depletion threshold is a paradigm, not a finding. Traditional teaching The teaching that liver glutathione must fall by about 70% before protein binding begins is described by McGill and Jaeschke as "a paradigm in the field" which "several observations have challenged": the non-hepatotoxic isomer AMAP binds protein despite a much smaller effect on glutathione; binding is detectable in human HepaRG cells within one hour, before glutathione depletion; and paracetamol–protein adducts can be measured in human serum after therapeutic doses.¹ The relationship between activation and glutathione is real and inverse — the threshold is the part that does not hold.
2. Covalent binding alone does not explain the injury, and the mitochondrial refinement is correlative. Inferred The search for a single adducted protein that explains toxicity did not succeed; most adducted proteins were enzymes whose activity was barely affected, and none had vital functions.¹ The current account — that binding to mitochondrial proteins specifically is what matters — is supported by several independent lines of evidence, but the authors state the limitation themselves: "all of these data are correlative. There is no direct evidence that selectively preventing mitochondrial protein binding can eliminate injury after APAP overdose."¹
3. Zone 3 selectivity is secure; the relative weight of its two explanations is not. Inferred That centrilobular hepatocytes have the highest CYP2E1 and the lowest oxygen tension is not in dispute.³ How much of the zonation is enzyme distribution and how much is redox environment has not been separated quantitatively in humans, and the two are usually stated together precisely because they cannot be told apart.
4. The non-2E1 cytochromes are on firmer ground in textbooks than in mice. Inferred CYP1A2, 2D6 and 3A4 activate paracetamol in model systems, but Cyp1a2 knockout mice were not protected.¹ Enzyme lists in review articles tend to accumulate isoforms that were shown to be capable of a reaction rather than shown to matter.
5. The UK and Australian treatment lines differ, and cohort figures do not transfer between them. UK practice since 2012 uses a single 100 mg/L at 4 h line;⁴ the Australian studies quoted on this page calculate their "paracetamol ratio" against 150 mg/L at 4 h.⁵ Both figures are correct for their own system. A ratio of 2 means different things on each side, which is a real trap when reading the toxicology literature — most of the best recent cohort work on massive overdose is Australian.
6. The 2012 UK guidance has a published critique from the people who evaluated it. More patients admitted, half of all presentations now treated, no reduction in adverse reactions from the longer infusion, and an estimated £8.3 million a year — with the authors calling for a safety and cost-benefit review.⁴ That is worth knowing about a guideline that is otherwise followed without comment, and it is not the same thing as a reason to deviate from it.
7. All of the EXTRIP recommendations rest on very low quality evidence — 24 articles, one randomised controlled trial, 135 patients with clinical data.³ The recommendations are sensible and the grading is honest; the numbers in them should be read as thresholds a Delphi panel could agree on, not as measured cut-points.
8. The UK product labels and UK clinical practice have diverged, and both are primary sources. Three current paracetamol SmPCs from different manufacturers were fetched and read separately, and they agree with each other: liver damage is possible at 10 g or more, and at 5 g or more if the patient has risk factors — enzyme-inducing drugs, regular excess alcohol, or likely glutathione depletion.⁹ UK clinical practice abandoned exactly that stratification in September 2012, when the CHM recommended a single treatment line and the ceasing of risk assessment.⁴ Each document is an accurate description of itself; they are not compatible as instructions, and TOXBASE is what UK practice follows. The mechanistic irony is that the label's risk factors are the right variables — CYP induction and glutathione reserve, exactly what this page's metabolism section describes — and the CHM's judgement was not that they are unreal but that they cannot be measured well enough at the bedside to stratify on.
9. The same labels still print the one-hour charcoal window,⁹ which is the traditional teaching flagged earlier on this page.⁵ A document can be simultaneously a primary source and a repository of superseded practice, and this one is both in the space of two paragraphs.
10. None of the three labels mentions staggered ingestion or therapeutic excess at all.⁹ That is the single clearest demonstration on this site of why an SmPC is the wrong document to risk-assess a poisoning from: it describes what the medicine does when taken as a medicine, and two of the three ingestion patterns that bring people to hospital are outside its frame.
11. Craig's staggered-overdose cohort predates the guidance it is often used to justify. Inferred The 663 patients were admitted between 1992 and 2008,¹⁰ entirely before the September 2012 changes that made treatment of all staggered and uncertain ingestions standard.⁴ The mortality difference and the buffer mechanism behind it are convincing; the figures nonetheless describe a management era in which many of those patients would not have been treated, and they should not be quoted as the outcome of staggered overdose today. It is also a single tertiary liver-unit cohort, so every patient in it already had severe liver injury — the denominator is not staggered overdoses, it is staggered overdoses that reached a transplant centre.
12. The therapeutic-excess data are a systematic review of reported cases and carry every bias that implies — the authors say so explicitly: low patient numbers, publication bias, and reliance on the accuracy of histories in reported cases.¹¹ The 93% severe-liver-damage figure is a property of which cases get written up, not an incidence, and it must not be read as a risk. What survives those limitations is the negative finding, and even that is weakened by availability: both a concentration and a transaminase were on record in only 79 (40%) of the 199 cases.¹¹

References

1. 1
   McGill MR, Jaeschke H. Metabolism and disposition of acetaminophen: recent advances in relation to hepatotoxicity and diagnosis. Pharm Res 2013;30(9):2174–87. PMC3709007 Source of the glucuronide (50–70%) and sulfate (25–35%) fractions, the pKa and duodenal-absorption point, the 1.5–3 h therapeutic half-life, the UGT/SULT/CYP isoform evidence, the conjugate transporters, and — quoted directly above — the challenge to the 70% glutathione-depletion paradigm and the statement that the mitochondrial-binding data are correlative. Verified 31 Aug 2026.
2. 2
   Ramachandran A, Jaeschke H. Acetaminophen toxicity: novel insights into mechanisms and future perspectives. Gene Expr 2018;18(1):19–30. PMC5885144 Source of the 80–90% conjugation figure, the statement that the sulfation pathway saturates in overdose, the complex II 47% inhibition figure, and the full NAPQI → mitochondrial adduct → peroxynitrite → JNK/Sab → permeability transition → programmed necrosis cascade described above. Verified 31 Aug 2026.
3. 3
   Gosselin S, Juurlink DN, Kielstein JT, Ghannoum M, Lavergne V, Nolin TD, Hoffman RS, on behalf of the EXTRIP workgroup. Extracorporeal treatment for acetaminophen poisoning: recommendations from the EXTRIP workgroup. Clin Toxicol (Phila) 2014;52(8):856–67. PubMed 25133498 Source of every dialysability threshold and recommendation grade quoted above, the 10–30% protein binding and 0.9–1.0 L/kg Vd, the "absorption generally complete by 4 h" statement, the centrilobular CYP2E1/oxygen quotation, and the mitochondrial-dysfunction phenotype in massive ingestion. All recommendations graded very low quality of evidence. Verified 31 Aug 2026 from the full text.
4. 4
   Bateman DN, Carroll R, Pettie J, et al. Effect of the UK's revised paracetamol poisoning management guidelines on admissions, adverse reactions and costs of treatment. Br J Clin Pharmacol 2014;78(3):610–18. PMC4243911 Source of the description of the September 2012 CHM changes and of every percentage, confidence interval and cost figure quoted for them. Verified 31 Aug 2026.
5. 5
   Chiew AL, Isbister GK, Kirby KA, Page CB, Chan BSH, Buckley NA. Massive paracetamol overdose: an observational study of the effect of activated charcoal and increased acetylcysteine dose (ATOM-2). Clin Toxicol (Phila) 2017;55(10):1055–65. PubMed 28644687 Source of the charcoal-within-4-hours paracetamol ratios and odds ratios. Australian cohort; ratios are calculated against the 150 mg/L at 4 h line, not the UK 100 mg/L line. Verified 31 Aug 2026.
6. 6
   Mitchell JR, Jollow DJ, Potter WZ, Gillette JR, Brodie BB. Acetaminophen-induced hepatic necrosis. IV. Protective role of glutathione. J Pharmacol Exp Ther 1973;187(1):211–17. PubMed 4746329 — with Jollow DJ, Mitchell JR, Potter WZ, Davis DC, Gillette JR, Brodie BB. Acetaminophen-induced hepatic necrosis. II. Role of covalent binding in vivo. J Pharmacol Exp Ther 1973;187(1):195–202. The original series establishing the protective role of glutathione and the role of covalent binding. Cited here for those findings only; the numeric depletion threshold often attributed to this work is discussed under reference 1 rather than quoted here. Citations verified 31 Aug 2026; full texts not retrieved.
7. 7
   TOXBASE — paracetamol. National Poisons Information Service. toxbase.org (NHS login required. NPIS: 0344 892 0111. The authoritative UK source for treatment thresholds, staggered ingestion, the weight-based risk-assessment threshold and the acetylcysteine regimen. Login-gated, so not quoted here.)
8. 8
   Medicines and Healthcare products Regulatory Agency / Commission on Human Medicines. Treatment of paracetamol overdose: simplified new guidance, September 2012. Drug Safety Update. gov.uk The guidance itself; its content as implemented is described and evaluated in reference 4.
9. 9
   Paracetamol 500mg Tablets — Summary of Product Characteristics. emc product 5164 — and Paracetamol 500 mg Tablet, emc product 100828 — and Paracetamol 500 mg Tablets PL43461/0006, Flamingo Pharma, emc product 13494. §4.9 of all three products was fetched and read separately, and they agree. Source of the 10 g and 5 g-with-risk-factors thresholds, the risk-factor list (enzyme-inducing drugs; regular excess ethanol; likely glutathione depletion in eating disorders, cystic fibrosis, HIV infection, starvation, cachexia), the one-hour charcoal statement, the "measure at 4 hours or later, earlier concentrations are unreliable" instruction, and the statement that maximum protective effect is obtained up to 8 hours (present in all three), with effectiveness declining sharply thereafter — a clause carried by 5164 and 13494 but absent from 100828. None of the three mentions staggered ingestion or therapeutic excess, and all three describe risk stratification that UK practice ceased in 2012 — see the appraisal. Verified 1 Sep 2026.
10. 10
    Craig DG, Bates CM, Davidson JS, Martin KG, Hayes PC, Simpson KJ. Staggered overdose pattern and delay to hospital presentation are associated with adverse outcomes following paracetamol-induced hepatotoxicity. Br J Clin Pharmacol 2012;73(2):285–94. PMC3269587 Single-centre cohort, Scottish Liver Transplantation Unit, 663 patients admitted 1992–2008. Source of the 161 (24.3%) staggered proportion, the "despite lower total ingested doses and lower admission ALT" comparison, the 37.3% versus 27.8% mortality with P = 0.025, the King's College criteria sensitivity of 77.6% (95% CI 70.8–81.5), the 58.2% pain rationale, the 178/396 (44.9%) late-presentation figure and the delayed-presentation odds ratio of 2.25 (95% CI 1.23–4.12, P = 0.009). Cohort predates the 2012 guidance change — see the appraisal. Verified 1 Sep 2026.
11. 11
    Acheampong P, Thomas SHL. Determinants of hepatotoxicity after repeated supratherapeutic paracetamol ingestion: systematic review of reported cases. Br J Clin Pharmacol 2016;82(4):923–31. PMC5137817 From the NPIS Newcastle unit. Systematic review with individual-level data from 199 reported cases. Source of the 186/199 (93%) severe liver damage and 77/78 (99%) in children aged 6 or under, the 127 (64%) liver failure and 49 (39%) deaths among those, the 71 (36%) not meeting US–Australasian intervention thresholds with 35 (49%) liver failure and 10 (14%) deaths, and the negative finding that no case developing liver damage had both a paracetamol concentration <20 mg/L and normal ALT/AST at initial assessment — with both values available in only 79 (40%) of cases. A review of reported cases; the authors state the publication-bias limitation themselves. Verified 1 Sep 2026.