AIHealthcare Analytics

Knowledge Wiki

Wiki › LLM Sensitivity to Patient Persona and Affective Framing — Unruptured Intracranial Aneurysm Benchmarking Study

LLM Sensitivity to Patient Persona and Affective Framing — Unruptured Intracranial Aneurysm Benchmarking Study

OVERVIEW
Rev 1 Aug 4, 2026 04:37 UTC 1 sources

Overview A comparative benchmarking study examined whether recasting identical unruptured intracranial aneurysm (UIA) clinical cases from third-person vignettes into first-person patient accounts—with varying affective registers (anxiety about treatment vs. anxiety about non-treatment)—altered management recommendations returned by leading large language models. The study tested Claude Opus 4.6, ChatGPT-5.4, and Gemini against 67 UIA cases previously adjudicated by a multidisciplinary team (MDT). Study Design and Context The research addresses a clinically important scenario: patients increasingly consult LLMs before specialist review, often presenting their concerns in emotionally charged language rather than neutral clinical terms. By systematically varying persona (third-person vignette vs. first-person account) and affective framing (treatment anxiety vs. non-treatment anxiety), the study isolates how narrative and emotional context—rather than clinical facts—may shift AI-generated recommendations. Significance for Clinical Safety If LLMs alter recommendations based on emotional framing rather than clinical parameters alone, this represents a meaningful safety concern. Patients presenting with high anxiety about treatment might receive recommendations biased toward conservative management, while those expressing fear of non-treatment might receive recommendations favoring intervention—neither necessarily aligned with evidence-based MDT consensus. This dynamic could…

Sources: harvested
🔒

Subscribe to read full wiki analysis

Full wiki content — including source provenance, contradiction tracking, related pages, and analyst notes — is available to Subscriber members.

View Plans & Subscribe

New users get a 7-day trial — no credit card required.

← Back to Wiki Index