When Algorithms Join the Consultation: A Qualitative Study of Patient Autonomy, Trust and Informed Consent in AI-Supported Oncology Decision-Making

Abstract

Background: Multimorbidity is often managed through disease-specific services, requiring patients to coordinate information, Background: Artificial intelligence (AI) is increasingly being incorporated into diagnostic, prognostic, and treatment-support processes in oncology, raising concerns regarding patient autonomy, trust, informed consent, explainability, and professional accountability. Objective: To explore how adults with current or previous oncology-care experience perceived autonomy, trust, informed consent, explainability, and clinician responsibility when considering hypothetical scenarios in which AI contributed to cancer-related diagnosis, prognosis, or treatment recommendations. Methods: A qualitative interpretive design was used. Fourteen adults with oncology-care experience were purposively recruited to provide variation in cancer experience and familiarity with digital health. Semi-structured interviews incorporated hypothetical scenarios involving AI-supported image analysis, individualized recurrence estimation, treatment recommendations, clinician–algorithm disagreement, and AI use without prior discussion. Data were analysed using reflexive thematic analysis. Results: Five interconnected themes were identified: conditional trust rather than technological trust; consent as an ongoing conversation; human accountability as the boundary of acceptable automation; explainability as practical meaning; and algorithmic pressure on autonomy. Participants generally perceived AI as acceptable when it augmented rather than replaced clinical reasoning, clinicians remained responsible for interpreting its outputs, and patients retained opportunities to question recommendations and consider alternatives. Participants also anticipated that AI-labelled recommendations could appear unusually objective and therefore become more difficult to reject, particularly when uncertainty and alternative options were not clearly communicated. Conclusion: Participants perceived AI-supported oncology decision-making as most compatible with autonomy when AI remained subordinate to accountable clinical judgment, disclosure was proportionate to its influence on the decision, explanations were clinically meaningful, and patients retained genuine opportunities to question or refuse recommendations. Because the study examined hypothetical scenarios rather than observed AI-supported consultations, the findings represent anticipated perceptions rather than demonstrated effects on clinical behaviour or autonomy