Ethics and Regulation of AI in Precision Oncology

The ethical use of artificial intelligence in precision oncology requires a strong understanding of both the regulatory frameworks and the ethical challenges typically encountered in using artificial intelligence in clinical practice. Traditional medical ethics encompasses core concepts of beneficence, non-malfeasance, autonomy, justice, dignity, and truthfulness. Harnessing artificial intelligence tools in a clinical setting presents ethical challenges in patient privacy, informed consent, transparency, bias, and, at the core, professional competence.

State medical boards are charged with licensing medical professionals, determining standards of practice, and setting regulatory frameworks for competence and ethical practice. Disciplinary procedures can be deployed when there is professional misconduct that is contrary to the standards of care established by the medical board regulations. The duties of a medical professional are taught through formal educational pathways and then imposed on licensed professionals throughout their professional practice. State medical boards do not regulate artificial intelligence tools or technology; only the licensed professionals that use these tools. Medical boards impose a duty on the licensed professional to acquire and maintain the necessary knowledge and skills to deliver health care professionally. Central to this ongoing professional duty of knowledge and skill is understanding the benefits, risks, and application of AI within patient care.

The Federation of State Medical Boards (FSMB), a national organization that supports the state medical boards, formed a task force on artificial intelligence in 2019. Four years later, in April 2024, FSMB adopted a key policy position statement in its report, “Navigating the Responsible and Ethical Incorporation of Artificial Intelligence into Clinical Practice.” ¹ This report clarifies two important concepts in the health care professional’s duty of competence. First, overreliance on AI can have a direct impact on patient care, such as degradation of professional skills, biased dependence on the AI output, medical errors, and even misdiagnosis. Second, failure to use available AI tools could deprive patients of direct benefit and may be considered outside the standard of care. The FSMB recommends focused educational efforts around AI in health care, including how AI works, the available benefits, potential risks, and a clear understanding of the impact on patient care.

The 2024 FSMB report also addresses a related professional duty of accountability and responsible use of artificial intelligence. For the first time, the FSMB states “the physician is ultimately responsible for the use of AI and should be held accountable for any harms that occur.” Ultimately, the standards for accountability set by state medical boards will be related to the specific AI tool being used and the risk of patient harm or impact on professional obligations. The FSMB noted in the report that the regulatory accountability for the licensed professional is a function of the risk to patient safety compared to the practice of medicine (or other licensed professional category) in the specific context of use of the AI tool. For example, patient monitoring AI tools are characterized as having a higher regulatory accountability level than AI image analysis tools. Clinical decision support AI tools are designated the highest level of regulatory accountability in this risk versus function model that FSMB describes.

Interestingly, the FSMB report provides recommendations to improve responsible use and accountability that are largely aimed at the developers of the AI tools. Recommendations include: (i) developers should enable physicians to have access and agency to know when and how to use the AI tool in clinical settings, (ii) use of clinical decision support tools by organizations, such as hospital systems and insurers, should provide a robust system for deployment that includes education about the AI tool, access to performance indicators and a regular system for reviewing the efficacy of the AI tool, (iii) developers of AI tools should design the software such that the state medical boards can audit and understand the AI tool during an assessment of whether the physician reliance on the AI tool’s output deviated from accepted standards of care, and (iv) state medical boards should work with FSMB to ensure consistent and fair interpretation of responsible use and accountability expectations.

The American Medical Association (AMA) largely supported the FSMB report during review of the draft report, but in March 2024 the AMA recommended that the FSMB place a much stronger liability on the AI developers as the entities best positioned to know the AI tool risks and to mitigate harm through design, development, validation and implementation procedures. The AMA expressed a “strong belief that such developers must accept this liability with measures such as maintaining appropriate medical liability insurance and reflecting as such in agreements with users.” ² As noted above, the FSMB maintained in the final report that the licensed professional is ultimately responsible for the use of the AI tool.

Ultimately, the use of an artificial intelligence tool in clinical practice—whether a simple, low-risk administrative documentation tool or a higher-risk clinical decision support algorithm—is the responsibility of the licensed professional using that AI tool. Therefore, licensed professionals should seek to fully understand the AI tools they are utilizing, such as knowing the software design, the training data used, details on reliability, and potential areas of bias that could impact patient care. Healthcare professionals should consider what types of questions and scenarios the AI tool is uniquely suited to answer and have information available to assess the limitations of the software in the planned use context. Although it now seems that the FSMB, and presumably the individual state medical boards, are not yet ready to redefine the practice of medicine to directly include AI tools, it is now clear that expectations of professional skill, knowledge, and competency will include the selection, deployment, and reliance on AI tools in clinical practice.