New psychosis, chloride 126 and an anion gap of −21: which result was lying?

The presentation pointed in two directions

A 60-year-old man with no recorded psychiatric or major medical history came to an emergency department convinced that his neighbour was poisoning him. He was alert, his observations were stable and his initial neurological examination was normal. The story sounded psychiatric. The chemistry panel did not.

His reported chloride was 126 mmol/L, above the laboratory reference range, while sodium was 141 mmol/L and bicarbonate 36 mmol/L. The calculated anion gap was −21 mEq/L. Albumin-related shifts and analytical error can lower an anion gap, but a value that negative should make the result itself part of the diagnostic problem. It was less a description of the patient’s acid–base state than a signal that one measured ion might not be what the analyser thought it was.

The mental state deteriorated

Within 24 hours, the patient became more paranoid and developed auditory and visual hallucinations. He attempted to leave and required involuntary psychiatric care. Once he was calmer, the history widened: excessive thirst, insomnia, fatigue, poor coordination and characteristic skin changes had accompanied the change in behaviour.

A first episode of psychosis at 60 already warrants a deliberate search for medical, neurological, toxic and medication-related causes. Here, the negative anion gap offered a narrower question. Could another halide—a chemical relative of chloride—be interfering with the chloride electrode and creating pseudohyperchloraemia, a falsely high result?

The three-month experiment

The patient eventually described a restrictive self-directed diet. After reading about sodium chloride, ordinary table salt, he had wanted to eliminate chloride rather than simply reduce sodium. He reported consulting ChatGPT and then purchasing sodium bromide online. For about three months, he used it in place of sodium chloride.

That history identifies the exposure, but it does not recover the missing conversation. The authors could not inspect his original chat log, so they could not verify the exact question, response or warnings he saw. They later tested a chatbot with a similar query and obtained a response mentioning bromide, but that reconstruction is not the patient’s transcript. The defensible conclusion is that internet and AI-derived information influenced the case, not that a particular reproducible sentence directly caused the poisoning.

Why bromide can make chloride look impossible

Bromide and chloride are both halide ions. Many routine analysers estimate chloride using an ion-selective electrode. At sufficiently high concentrations, bromide can cross-react with that system and be counted as chloride. Sodium remains genuinely measured, while the chloride value is pushed upwards. When the anion gap is calculated from sodium minus chloride and bicarbonate, the inflated chloride can drive the answer far below zero.

The patient’s serum bromide was reported at 1,700 mg/L, compared with a reference interval of 0.9 to 7.3 mg/L in the case report. A negative gap is not diagnostic of bromism on its own: hypoalbuminaemia, laboratory problems, paraproteins, lithium, iodide and salicylate-related assay interference belong in the differential. But a markedly negative value, apparent hyperchloraemia and compatible neurological or psychiatric features should prompt a targeted exposure history and discussion with the laboratory and toxicology service.

The reveal was a toxidrome, not a psychiatric label

Bromism is the syndrome produced by excessive bromide accumulation. Modern cases are uncommon because bromide salts largely disappeared from routine human medicines, although they remain obtainable online and appear in some products internationally. The presentation can be diffuse: confusion, delusions, hallucinations, ataxia, tremor, fatigue, skin eruptions and gastrointestinal symptoms. That spread across psychiatry, neurology, dermatology and chemistry makes the diagnosis easy to fragment.

The patient’s exposure was stopped. He received hospital treatment including fluid and electrolyte management, nutritional replacement and temporary antipsychotic medication. Over a three-week admission, the psychosis improved and the chloride and anion gap normalised; risperidone was stopped before discharge, and he remained stable without it at follow-up. Severe bromide toxicity can require extracorporeal removal, but management depends on the exposure, renal function and clinical severity and belongs with specialist toxicology advice.

What changes for a UK clinician?

The case does not create a new UK pathway. It reinforces existing practice: suspected poisoning needs urgent assessment, and clinicians should use local toxicology support and TOXBASE rather than extrapolating a case report into a protocol. The NHS advises immediate medical help when a harmful substance may have been swallowed and advises against inducing vomiting or giving food or drink unless instructed.

The transferable lesson is to ask where a patient obtained health information without turning the consultation into a judgement about technology. Ask what they bought, how they used it, for how long, and whether packaging or screenshots are available. Then look again at results that appear physiologically impossible. In this case, the strange number was not noise. It was the shortest route from a new psychiatric presentation to a forgotten poison.

Source note

The primary case was published in Annals of Internal Medicine: Clinical Cases on 5 August 2025 and resurfaced independently during August 2026. The original chatbot log was unavailable. No correction, expression of concern or retraction was linked from the journal, DOI or indexing records when checked on 22 August 2026. No patient image, quotation, location or unnecessary identifying detail is reproduced.