Testimonial injustice in medical machine learning.

Area of Science:

• Medical Ethics
• Epistemology
• Health Informatics

Background:

• Machine learning (ML) systems are increasingly integrated into clinical practice, raising ethical considerations regarding their responsible use.
• The application of ML in healthcare necessitates an examination of its impact on patient-physician relationships and patient credibility.

Purpose of the Study:

• To analyze the ethical and epistemic dimensions of ML use in mediating patient-physician interactions.
• To investigate how ML systems can undermine patient voices and credibility, potentially causing testimonial injustice.
• To examine the role of ML-based Prediction Drug Monitoring Programmes (PDMPs) in medical encounters.

Main Methods:

• Philosophical analysis of ML systems' role in patient-physician communication.
• Case study of ML-based PDMPs used in the USA for opioid misuse risk prediction.
• Examination of how ML systems function as markers of trustworthiness and influence credibility assessments.

Main Results:

• ML systems can silence patient narratives and relativize their opinions, diminishing their credibility without valid justification.
• Withholding credibility based on ML system operation constitutes testimonial injustice, leading to adverse patient outcomes.
• ML-based PDMPs act as markers of trustworthiness, unfairly assessing patient credibility and exacerbating social inequalities.

Conclusions:

• ML-mediated medical encounters risk testimonial injustice by treating ML outputs as definitive credibility assessments.
• The use of ML systems like PDMPs can perpetuate and amplify social inequalities, disproportionately affecting vulnerable populations.
• Responsible ML implementation in healthcare requires addressing its epistemic and ethical implications on patient trust and equity.