Ethical Integration of Patient-Reported Outcomes and Digital Biomarkers in AI Healthcare Models: An Expert Consensus Framework

Abstract

Background: Alongside expected benefits, several ethical concerns arise from Artificial Intelligence (AI) based models. From the design to the implementation and subsequent evaluation, it is crucial to map potential ethical concerns regarding the use of AI models in healthcare. Patient-Reported Outcomes (PROs) and Digital Biomarkers (DBs) are being increasingly collected to improve patient-centered healthcare systems. However, due to the sensitive nature of this data, its processing into AI models may raise ethical concerns that should be considered. While general AI ethics frameworks exist, no expert consensus has specifically addressed the unique ethical challenges of integrating PROs and DBs in AI healthcare models. Objective: This study aims to address this gap by establishing expert consensus on ethical, legal, and social considerations for integrating PROs and DBs into AI-driven healthcare models. Methods: A mixed-method study was performed. Phase 1 consisted of a narrative review to map the ethical landscape and generate an initial pool of recommendations. Phase 2 involved a two-round modified e-Delphi survey to validate and refine these recommendations among a multidisciplinary panel of experts (n=27). The panel included experts in AI, bioethics, clinical research, and data protection, primarily from Southern Europe. Results: The findings of the two complementary components of this study (narrative review and modified e-Delphi study) were organized around five core ethical principles: autonomy, beneficence, non-maleficence, justice, and transparency and accountability. The modified e-Delphi study achieved high consensus (≥80%) on 55 specific recommendations across these principles. Key recommendations included implementing dynamic consent models, establishing continuous model validation protocols, conducting regular impact assessments, ensuring diverse stakeholder engagement to mitigate biases, and maintaining human oversight within AI systems. Conclusion: This study provides the first comprehensive expert-validated ethical framework specifically designed for PROs and DBs integration in AI healthcare models, filling a gap in the literature that has primarily focused on general AI ethics rather than the unique challenges posed by patient-generated health data.