Physician Perspectives on AI-Enabled Voice Analysis for Disease Assessment

Michele Anzabi, MBE – Biomedical Ethics – Mayo Clinic; Hugo Botha, MB, ChB – Neurology – Mayo Clinic; Richard Sharp, PhD – Biomedical Ethics – Mayo Clinic

Abstract: Voice analysis tools powered by artificial intelligence (AI) are increasingly being studied and commercialized. These tools analyze linguistic and acoustic properties of human voice for clinical insights like classifying disease risk or severity.
To understand the anticipated impact of these tools on clinical practice, we conducted case-based in-depth interviews with 40 physicians in internal medicine and neurology. We developed case scenarios based on academic literature, commercial applications, and expert input to explore physicians' perspectives on voice analysis tools in various clinical contexts, including depression screening, Parkinson's disease classification, and Alzheimer's disease prediction. These case scenarios were selected because they demonstrate promising near-term clinical applications of voice analysis.
Our results highlighted physicians' recognition of vast potential for AI-enabled voice analysis, particularly for early disease recognition and long-term tracking. However, this optimism was tempered by physicians' sensitivities to the available evidence base and performance validation metrics before considering such tools for clinical practice. Beyond clinical utility concerns, physicians expressed significant apprehension about data stewardship practices, including storage and sharing of voice data, which some viewed as uniquely sensitive compared to other health data. Protection against data breaches was critical, especially given re-identification risks and potential misuse (e.g., deepfakes). Additionally, physicians emphasized the need for careful implementation, drawing parallels to genetic testing and patients' "right not to know" certain diagnostic information.
We contextualize physician concerns within broader ethical, legal, and social implications research and examine how this work might inform implementation and governance of these technologies given anticipated risks.

Keywords: Vocal biomarkers, Physician perspectives, Artificial intelligence