The Evolving AI Landscape in Cancer Care: An NCCN Summit

The National Comprehensive Cancer Network (NCCN) hosted a 1-day summit on September 9 all about artificial intelligence (AI) in cancer care: how it’s being used today, where the field is moving in the future, and challenges along the way that will require innovative and collaborative solutions. This event, hosted at the National Press Club in Washington, DC, and virtually, brought together leaders, researchers, and providers to share their insights on utilizing AI in a rapidly changing landscape.

In the morning, Clifford Goodman, PhD, a health care technology and policy consultant, led a panel discussion centered around AI’s expanding role in oncology. Experts provided perspectives spanning policy, advocacy, patient care, and drug development, with an eye toward future successes and potential challenges in the arena of AI.

Areas of Greatest Impact

Each panelist occupied a different role in the cancer care landscape, providing unique perspectives on AI’s greatest impact on the field. Danielle Bitterman, MD, an assistant professor of radiation oncology at Harvard Medical School and clinical lead for data science and AI at Mass General Brigham, cited AI’s ability to automate processes in radiotherapy. For example, adaptive radiotherapy updates radiation treatment plans based on changes in daily scans, assisting physicians in tailoring the care to the patient in real-time and with greater accuracy. “AI allows us to on-the-fly rapidly redo a plan designed to each patient’s daily anatomy with a speed and efficiency that just wouldn’t be possible without this kind of automation,” Dr. Bitterman explained.

Jorge S. Reis-Filho, MD, PhD, FRCPath, chief AI and data scientist for oncology research and development at AstraZeneca, identified precision rather than speed as one of the most valuable features in his field of drug development. AI tools are assisting in identifying novel biomarkers and matching patients with treatments that are more likely to elicit a positive response with greater accuracy than pre-AI methods. As the technology has developed, he has seen improvements in speed and efficiency as well, but the greatest gains have been in precision medicine.

In patient care settings, William Walders, MBA, MHA, vice president and chief digital and information officer for The Joint Commission, values AI’s ability to reduce administrative burden. Tools for note-taking and transcription, scheduling, and reviewing electronic health record (EHR) data take work off providers' plates. He thinks this is particularly useful for oncology nurses, who are often bogged down with administrative tasks that limit their ability to work at their maximum scope.

Challenges in Addressing Bias

One of the greatest challenges addressed by the panel was the risk of bias in AI, both in terms of data sets and how they are used. Allen Rush, PhD, MS, founder and board chairman of the Jacqueline Rush Foundation and AI technology consultant, said AI developers are well aware of the problem of bias. He feels confident that it can be removed from data sets and guardrails can be put in place as long as there is incentive for technology companies to do so.

However, some panelists were more skeptical of this possibility. Judy Gichoya, MD, MS, FSIIM, an associate professor of interventional radiology and informatics at Emory University, pointed out that there is currently no universal agreement on what a fair, unbiased AI model would look like. If the problem of bias is not clearly identified, it is difficult to develop an unbiased model.

Hopes For the Future

While there was much excitement over the benefits of AI tools, panelists also recognized challenges ahead and shared their visions for how they might be addressed.

Dr. Rush’s background in AI is broad and not specific to medicine. From his perspective, there are many AI technologies already developed that would be useful in health care, but there is a lack of communication between industries to identify these uses. “There are many models and approaches in AI that we could apply to medical  problems, and I think we’re just at the very beginning stages of that,” Dr. Rush said. He called for more collaboration between AI researchers and the medical community to identify the greatest needs and potential solutions.

Dr. Gichoya sees this same gap in her field of radiology. She explained that AI is a pattern recognizer, but so are radiologists. Both are skilled at identifying obvious patterns and diagnoses, but that is not the type of case where radiologists need help. “They need [assistance] where there’s uncertainty, where it’s difficult, and AI still very much struggles today in those types of realities,” she said. Dr. Gichoya and her colleagues often do not use available AI tools because they are not filling a gap in their practice.

Dr. Bitterman and Dr. Reis-Filho each discussed the importance of centering the end-user in the development of these tools. “We need to understand really well who the users and customers of AI solutions are,” Reis-Filho said. Including these individuals in the conversation will help keep the end result patient-focused. For example, Dr. Bitterman’s institution is developing an AI tool to improve patient engagement in clinical trials. They have a steering committee and focus groups made up of patients and patient advocates to ensure the technology developed is meeting the greatest needs of the population who will be using the tool.

Finally, the panelists all shared a desire for further research and guardrails around AI usage. While they acknowledged the great benefits of AI tools in medicine thus far, they also tempered this excitement with the reality that this technology is relatively young. Further work in development, ethics, implementation, and safety nets will all help to maximize benefit and minimize harm to patients.

Discussion and Exchange in the Room

For participants attending in-person in Washington, DC, breakout groups and report-outs explored the challenges seen by providers, researchers, and advocates who are navigating the changing AI landscape on a daily basis. One of the biggest concerns shared was transparency. Participants said there was a lack of understanding of how these tools work, which models are being used, and what data is used to train them. National standards or accreditations were suggested as solutions to encourage transparency from the developers of these technologies.

Participants also discussed the challenge of “Dr. Google,” a term for when patients look up their symptoms via internet search or AI chatbot and receive inaccurate, incomplete, or poorly communicated information. Providers cannot expect patients to abstain from searching for answers independently, but they need strategies to interpret the information provided by AI and educate patients on AI literacy regarding their health.

Some participants brought up concerns about barriers to using AI equitably. Aside from bias that may be present in AI models or data sets, there are currently significant cost barriers to adoption of many of these tools which limits access. As AI expands and improves in health care, unequal access due to cost may further drive existing care gaps in smaller cancer centers that lack the budget of a large hospital or academic center. While technology development is important, it is equally important to consider how the benefits will be delivered to patients.

Read part 2 recapping the NCCN summit on the ACCCBuzz Blog.