Upper GI bleeding is common and can be fatal. The single strongest predictor of death is not the volume vomited but age and comorbidity — an 80-year-old with ischaemic heart disease may decompensate from a bleed a 30-year-old would tolerate. Melaena deserves particular emphasis because patients do not volunteer it: the specific question is "has your stool been black, tarry and unusually smelly?" Iron and bismuth turn stool black but not tarry or offensive, and do not cause the classic melaena smell. NICE CG141 recommends that all patients with acute upper GI bleeding are risk-assessed and that unstable patients receive endoscopy immediately after resuscitation, with other admitted patients scoped within 24 hours (QS38).

The sequence question — retching before blood, or blood with the first vomit — is the most useful single discriminator available without a scope. A Mallory-Weiss tear follows a mechanical insult and produces streaks or small volumes; it usually settles spontaneously. Varices bleed painlessly and copiously because portal pressure drives the bleed. NSAID exposure is systematically under-reported: patients do not consider a supermarket ibuprofen a "medicine", yet NSAIDs substantially increase ulcer bleeding risk, and the combination of an NSAID with an SSRI or a steroid is multiplicative rather than additive. Swallowed epistaxis is the commonest benign mimic in older people on anticoagulants — always look in the nose.

Knowing the distribution changes the pre-test probability you carry into the referral conversation. Peptic ulcer disease still dominates despite falling H. pylori prevalence, because NSAID use has risen in an ageing population. Variceal bleeding is the minority by number but the majority of the mortality — which is why haematemesis in known cirrhosis or portal hypertension is an immediate 999 call. Aorto-enteric fistula is vanishingly rare but worth holding in mind in anyone with a previous abdominal aortic graft and a "herald bleed", because the definitive bleed is usually fatal and the window is hours.

The Glasgow-Blatchford score was designed to be calculable before endoscopy and is the tool NICE CG141 recommends at first assessment; the Rockall score requires endoscopic findings and so belongs to secondary care. A GBS of 0 identifies a group with a very low probability of needing intervention, and NICE allows early discharge for these patients — but the score requires blood results, so in a surgery without same-day phlebotomy the practical answer is referral. The disproportionately raised urea is a genuinely useful primary-care sign: it reflects absorbed blood protein plus a degree of pre-renal impairment, and a normal creatinine alongside it argues against renal disease as the explanation. Note that haemoglobin can be normal in the first hours of a brisk bleed because haemodilution has not yet occurred — a normal Hb never excludes significant bleeding.

The commonest investigative error in primary care is treating this as a diagnostic problem rather than a logistical one. Blood results do not change the disposal of a patient with genuine haematemesis — they will be repeated on arrival — so taking them must never delay transfer. The H. pylori timing rule is the detail most often missed and generates a stream of false negatives: proton pump inhibitors suppress bacterial load enough to invalidate both stool antigen and breath testing, so the test must be deferred for two weeks after the PPI stops. Since a bleeding ulcer patient will be on a PPI for 4–8 weeks, eradication testing is a planned follow-up task, not a same-visit one. A FIT test has no place in upper GI bleeding: it is calibrated and validated for colorectal cancer risk stratification, and a positive result in this context tells you nothing you did not already know.

NICE NG12 (updated April 2026) separates two thresholds for endoscopy that are easy to conflate. The urgent suspected cancer pathway referral is triggered by dysphagia at any age, or by weight loss plus an upper GI symptom at 55 and over. Haematemesis itself sits in the consider non-urgent group — which sounds contradictory until you realise NG12 addresses cancer detection, not acute bleeding: the acutely bleeding patient is captured by CG141 and admitted, scoped within 24 hours, and never waits for a suspected cancer pathway referral. The practical rule is therefore to admit on the acute pathway, and use NG12 only for the patient whose bleeding has clearly stopped, after acute GBS assessment via an agreed hospital pathway, and who is being worked up for a cause. Long-term, the hepatology referral matters more than most GPs expect: after a first variceal bleed, re-bleeding is common without specialist-led secondary prophylaxis.

NICE CG141 specifically advises against giving a PPI before endoscopy in acute upper GI bleeding — it does not reduce mortality, re-bleeding or need for surgery, and it can obscure endoscopic findings. The PPI belongs after the scope, or after discharge for a confirmed ulcer. Confirming H. pylori eradication is optional in uncomplicated dyspepsia but mandatory after a bleeding ulcer, because persistent infection is the dominant cause of re-bleeding and the consequences of failure are severe. The hardest conversation is restarting aspirin or an anticoagulant: stopping permanently after a bleed feels safe but raises thrombotic and stroke risk substantially, and the evidence favours resumption with gastroprotection once bleeding is controlled. Aspirin for secondary cardiovascular prevention in particular should usually be restarted, not abandoned.

The post-discharge medicines reconciliation is the most valuable thing primary care does in this pathway, and it is frequently skipped. Discharge summaries commonly reinstate a pre-admission NSAID by transcription error, or start a PPI with no stop date so the patient remains on it for years without review. Explicit naming of drugs matters because patients classify medicines by where they bought them rather than by pharmacology: asking "are you on any anti-inflammatories?" reliably produces a "no" from someone taking daily ibuprofen for arthritis. Written safety-netting on melaena has a genuine evidence base in reducing delayed re-presentation — the description needs to be concrete ("black, sticky, like tar, with a strong smell"), not the word "melaena".

Re-bleeding after endoscopic therapy is concentrated in the first 72 hours, so the highest-risk window is usually still inpatient — but the second peak is the patient who restarts an NSAID or whose H. pylori was never eradicated. Persistent or recurrent iron deficiency after a treated upper GI bleed is an important signal: it suggests either continued occult loss or a second lesion, and it is the classic route to a missed caecal carcinoma in a patient whose anaemia was all attributed to the known ulcer. Where iron deficiency recurs and the OGD was clean, the colon needs investigating. Finally, documenting H. pylori status and eradication confirmation in a structured, findable place in the record is a small act with a long payoff — it is precisely what the next clinician facing a recurrence will want and rarely finds.