The US National Academies of Sciences, Engineering, and Medicine has published proceedings examining the policy and governance requirements for integrating artificial intelligence into cancer research and clinical care.
The publication follows a 1.5-day workshop convened by the National Academies’ National Cancer Policy Forum in collaboration with its computing research programme.
The workshop examined current and potential uses of AI in oncology, alongside the policies needed to support the responsible and effective use of the technology.
Applications considered included AI-based biomarkers, treatment planning, clinical decision-support systems, clinical workflow tools, product discovery, data curation and the generation of clinical evidence.
The proceedings identify the assessment of training datasets, model validation, clinical validation, safety and effectiveness as key areas requiring stronger and more transparent policies.
AI systems used in oncology may be trained on medical images, genomic information, electronic health records and treatment-outcome data. Participants considered how the quality and representativeness of these datasets can affect model performance across different patient populations.
The workshop also examined how AI systems should be monitored after implementation, including the need to identify performance drift as clinical evidence, treatment protocols and patient populations change.
Ethical and legal issues discussed included patient privacy, data security, algorithmic bias, liability for AI-supported decisions and transparency around the data, models and objectives used to develop AI tools.
Participants considered how accountability should be divided among technology developers, healthcare institutions, regulators and clinicians when AI-generated information contributes to a diagnosis or treatment decision.
The proceedings also address governance requirements for the responsible use of patient data, including anonymisation, consent, data-sharing practices and accountability for data misuse.
Education was identified as another priority. Healthcare professionals may require additional training to interpret AI-generated insights, while patients need clearer information about how AI systems may influence their treatment.
The National Academies said future policies will need to account for increasingly autonomous AI systems and their integration with technologies such as genomics, while remaining adaptable to changes in clinical evidence and oncology practice.
