MedVoiceBias: A Controlled Study of Audio LLM Behavior in Clinical Decision-Making

2025-11-10

AI Summary

This study identifies systemic bias in audio large language models (Audio-LLMs) for clinical decision-making, arising from paralinguistic cues—including age, gender, and affect—that induce healthcare inequities. We develop a rigorous evaluation framework comprising 170 clinical cases and employ text-to-speech synthesis to generate multidimensional voice variants. Integrating chain-of-thought prompting with explicit reasoning, we quantify modality-specific bias across audio and text inputs. Experiments reveal that audio input induces up to a 35% disparity in surgical recommendations compared to text; one leading model reduces surgery recommendations for elderly patients by 80%, exhibiting a 12% age-related bias; affect recognition accuracy is significantly degraded. Critically, we provide the first empirical evidence that the audio modality itself—distinct from semantic content—constitutes an independent source of bias. To address this, we propose a “bias-aware architecture” design paradigm, establishing both methodological foundations and empirical validation for trustworthy medical AI.

Technology Category

• Philosophy and Ethics of AI: Bias, Fairness & Equity
• Machine Learning: Ethics, Bias, and Fairness
• Natural Language Processing: Ethics — Bias, Fairness, Transparency & Privacy

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Social Networks and Social Media: Fairness and bias in social network and social media analysis
• Search and Retrieval-Augmented AI: Large language models for search

Abstract

As large language models transition from text-based interfaces to audio interactions in clinical settings, they might introduce new vulnerabilities through paralinguistic cues in audio. We evaluated these models on 170 clinical cases, each synthesized into speech from 36 distinct voice profiles spanning variations in age, gender, and emotion. Our findings reveal a severe modality bias: surgical recommendations for audio inputs varied by as much as 35% compared to identical text-based inputs, with one model providing 80% fewer recommendations. Further analysis uncovered age disparities of up to 12% between young and elderly voices, which persisted in most models despite chain-of-thought prompting. While explicit reasoning successfully eliminated gender bias, the impact of emotion was not detected due to poor recognition performance. These results demonstrate that audio LLMs are susceptible to making clinical decisions based on a patient's voice characteristics rather than medical evidence, a flaw that risks perpetuating healthcare disparities. We conclude that bias-aware architectures are essential and urgently needed before the clinical deployment of these models.

Problem

• Audio LLMs show significant clinical decision variations based on voice characteristics
• Models exhibit modality bias with surgical recommendations differing by 35%
• Age disparities up to 12% persist despite chain-of-thought prompting

Innovation

• Evaluated audio LLMs using synthesized voice profiles
• Uncovered modality bias in surgical recommendations
• Proposed bias-aware architectures for clinical deployment