Artificial Intolerance: Stigmatizing Language in Clinical Documentation Skews Large Language Model Decision-Making

2026-05-16

AI Summary

Stigmatizing language in clinical documentation—such as expressions of suspicion, blame, or derogation—can induce implicit biases in large language models (LLMs) during high-stakes medical decision-making, thereby compromising diagnostic and therapeutic equity. This study systematically evaluates decision-making biases across nine state-of-the-art LLMs when exposed to clinical scenarios containing stigmatizing statements, revealing for the first time that even a single stigmatizing sentence can significantly alter model outputs, steering them toward more conservative clinical recommendations. Through clinical scenario simulations, targeted injection of stigmatizing language, dose–response analyses, and interventions including Chain-of-Thought reasoning and self-debiasing prompts, we demonstrate that current debiasing strategies are largely ineffective at mitigating this form of implicit bias, highlighting a critical vulnerability in the fairness of clinical LLMs.

Abstract

Large Language Models (LLMs) are increasingly deployed in high-stakes domains such as clinical decision support and medical documentation. However, the robustness of these models against subtle linguistic variations, specifically stigmatizing language (SL) commonly found in human-authored clinical notes, remains critically under-explored. In this work, we investigate whether frontier LLMs inherit and propagate this human bias when processing clinical text. We systematically evaluate nine frontier LLMs across four stigmatized medical conditions, utilizing clinical vignettes injected with varying intensities and phenotypes of SL (doubt, blame, and maligning). Our results demonstrate that all evaluated models exhibit substantial bias, with clinical decision-making significantly skewed towards less aggressive patient management. Notably, we observe a high sensitivity to linguistic framing, where a single SL sentence is sufficient to alter model outputs, revealing a clear dose-response relationship. Furthermore, we evaluate standard prompt-based mitigation strategies, including Chain-of-Thought (CoT) reasoning and model self-debiasing. These approaches show limited efficacy; models struggle to explicitly identify SL while remaining implicitly influenced by it. Our findings expose a critical vulnerability in current LLMs regarding fairness and robustness in clinical NLP, underscoring the need for rigorous algorithmic guardrails to prevent the automation of health disparities.

Problem

• stigmatizing language
• clinical decision-making
• large language models
• bias propagation
• health disparities

Innovation

• stigmatizing language
• clinical NLP
• large language models
• algorithmic bias
• fairness in AI