Gender-Dependent Diagnostic Substitution in LLM Medical Triage: Same Symptoms, Unequal Urgency

Key Findings

• — 630 trials: one symptom profile x seven demographic conditions (ages 25, 38, 65 crossed with male and female, plus a gender-unspecified baseline) x 30 repetitions per condition per model.
• — Emergency-room referral rates for 25-year-olds, women versus men: Gemini 3.5 Flash 0% versus 23.3%; Claude Sonnet 4.6 6.7% versus 96.7%; GPT-5.4-mini 6.7% versus 66.7%; all p < 0.001.
• — The disparity disappears at age 65 in all three models.
• — The mechanism reported is diagnostic substitution: models preferentially assign young women Idiopathic Intracranial Hypertension, a condition epidemiologically linked to women of childbearing age, while placing space-occupying lesions in the differential for men.
• — Severity ratings were comparable across genders (7 to 9 out of 10), so the routing difference is not explained by the model judging the case less serious.
• — The author's conclusion is that triage engines must decouple urgency assessment from epidemiological priors on the presenting demographic.

Methodology Notes

Single symptom profile, 30 repetitions per cell, three commercial model families (Gemini 3.5 Flash, Claude Sonnet 4.6, GPT-5.4-mini), no clinician adjudication of the generated differentials reported in the abstract. The very large between-gender gaps rest on one clinical vignette, so generalisation beyond this presentation is untested. Preprint, not peer-reviewed, no venue; single author with no institutional affiliation stated on the arXiv record, which is why credibility is preliminary. arXiv 2606.03641, v1 2026-06-02; abstract page fetched and read.

Sources