Pages

Monday, October 29, 2018

Key Questions for Regulators Rise with the Dawn of AI-Driven Healthcare

Eric M. Fish
Much of the initial excitement and application of artificial intelligence in healthcare has been focused on population health management, data analytics, and reduction of inefficiencies within the administration of healthcare. Many of the early direct-to-consumer products using artificial intelligence are branded as devices or apps that focus on general wellness and heath, not treatment of disease. Despite growing investment in artificial intelligence for healthcare generally, only 30 percent of investment deals fund of companies developing products that support providers in direct patient care and treatment of disease. But soon, artificial intelligence will be a common component of the clinical workflow, and will offer new challenges the current regulatory framework.

To date, the regulatory discussions about artificial intelligence have focused on the activities and approvals of the U.S. Food and Drug Administration. In 2017, the FDA approved a cloud-based deep learning algorithm that serves as a decision support tool that allows for physicians to diagnose heart conditions with greater efficiency and accuracy. In 2018, the FDA permitted the marketing of a medical device that uses artificial intelligence to detect diabetic retinopathy in adults who have diabetes. These approvals, in conjunction with the ubiquitous goal of providing better, faster care at a lower cost, may drive the release of new products and fuel additional demand to create them. Although the FDA is one of many stakeholders with interest and jurisdiction to influence the development of healthcare technology, its role as the primary regulator of medical technology devices has made it the de facto agency responsible to review the introduction of new healthcare technologies that use algorithms. But other regulators need to become more engaged.

Commentators such as Pearse Keane and Eric Topol and have commented that the attraction to the transformative power of artificial intelligence creates an “AI-chasm” of misunderstanding between what is necessary for the development and approval of a scientifically sound algorithm and the proper use of such algorithms in real-world situation. And within healthcare the general maxim that the federal government regulates medical products and state government regulates medical practice frustrates the creation of a national strategy for artificial intelligence in healthcare.

So now is the time to ask important structural questions, such as, does the United States have a sufficient regulatory framework to regulate medical technologies that utilize or operationalize artificial intelligence in a clinical setting? And is the system of regulation properly designed for the day when wellness products converge with diagnosis and treatment to create a quasi-clinical product sold directly to consumers? If the full potential of artificial intelligence in the clinical setting is to be realized, it is imperative that some of the downstream effects of artificial intelligence be addressed before promised transformations become reality. All parties must re-commit to work collaboratively to develop a systemic approach restates core principles of regulation and sets standards for future integration of artificial intelligence into the medical practice.

The Federation of State Medical Boards, along with its member state boards, is taking a proactive approach to these questions and has begun to study what how artificial intelligence will change the standards for medical practice and licensure. One initial challenge that state regulators will face the delineation between a clinical decision support tool and a tool which, under current state law definitions would be engaging in the practice of medicine. If an algorithm is found to be practicing medicine, and does so to the detriment of public safety, state regulators could exert oversight and act to enforce to ensure a proper standard of care for its use in a clinical setting. As seen with previous integrations of new technologies into other regulated industries, this ex-ante scenario will have a chilling effect across the industry. Inviting state regulators to participate more fully in the development and early review of these algorithms, be it through informal review or creation of a more formal regulatory sandbox that allows regulators to see within the black-box, may serve to empower innovation and lead to a more rapid integration of artificial intelligence.

The essential characteristic of deep learning algorithms to consume data and alter the decision-making process, which leads to unpredictable outcomes over time, adds an additional layer of complexity to future regulation. The inherent mutability of deep learning systems complicates the ability to know why the algorithm arrived at a result, and frustrates the ability of a physician to explain the performance of the system to a patient. Accordingly, as more complex algorithms are introduced into the clinical setting, it may be increasingly difficult for physicians to comply with core practice requirements established by state regulations. Future regulations may also need to ensure that a physician employing algorithm-based treatment meets levels of clinical and ethical competence, including the ability to explain the rationale for diagnosis and treatment to a patient, as well as maintaining the duty to obtain any necessary consents for data collection and use

The lurking grand question, however, is the assignment ultimate responsibility and accountability if treatment using artificial intelligence results in harm to the patient. If artificial intelligence is deployed in a team-based care setting, state regulators will be pressed to determine discipline of their individual provider under their jurisdictional authority. Artificial intelligence integrated across practice areas will necessitate a greater understanding of the roles and responsibilities of all members of the health care team. Early discussions of artificial intelligence within the healthcare community have introduced, for discussion, a framework that assigns responsibility and accountability for the use of artificial intelligence to the person with the most knowledge of risk and who is in the best position to mitigate the risk. This concept may provide a guide for state regulators grappling with the same issues for purposes of licensure and discipline.

Artificial intelligence will bring about the dawn of a new day in healthcare. Rather than waiting for that day to come to develop standards to its use, it is crucial that regulators act now to understand what artificial intelligence is, what it does, and what questions must be answered to fully reap the benefits of this technology.

Eric M. Fish is Senior Vice President of Legal Services, Federation of State Medical Boards. You can reach him by e-mail at efish at fsmb.org. The views expressed in these remarks are my own and do not necessarily reflect the views of the Federation of State Medical Boards, its Board of Directors, or any member state medical board.