Nov 09, 2023
By Geraldine Carroll, ASCO Publishing
Artificial intelligence (AI) is poised to transform cancer care delivery from diagnosis to treatment selection, and even the communication between physicians and patients. AI applications also have the potential to improve clinician resilience by reducing administrative tasks, but as tools like ChatGPT have become available and many health care applications have emerged, the oncology community will need to address how to ensure the quality, safety, and efficacy of emerging AI tools.
“I think generative artificial intelligence will have a profound and transformative impact across all of society in every aspect of human experience and of course inclusive of computational medicine and health care,” said Peter P. Yu, MD, FACP, FASCO, physician-in-chief at the Hartford HealthCare Cancer Institute, associate editor of JCO Clinical Cancer Informatics (JCO CCI), and a member of a new AI task force at ASCO.
Outside oncology, one of the biggest concerns about the increasing reach of AI is whether its users control its development or whether it creates its own unstoppable momentum and evolution. Experts in the use of AI in cancer research say that it is important to ensure that the sources AI uses to generate its models rely on peer review data.
With any new frontier, the fear of the unknown can cloud accurate analysis of the potential of a new tool. With AI, more than any other technology since the dawn of the information age, the question is whether artificial intelligence could replace humans.1,2
Dr. Yu observed that, in the past, advances in technology has primarily replaced jobs utilizing manual labor that can be automated by machine. “What’s different with generative AI is that it impacts knowledge workers,” he said. “Among knowledge workers, there are those who create knowledge and will not be replaced by generative AI, which draws from pre-existing information, and those who receive and implement knowledge whose jobs might be either replaced or alternatively enhanced by AI.” He noted that “another safe harbor, if you will, for physicians is in supporting patients’ and their families’ emotional needs as distinct from their knowledge needs.”
How Generative AI Will Be Used
The impact of AI on oncologists and the practice of cancer medicine will be multipronged and is emerging from a variety of directions, Dr. Yu added. The three imminently emerging uses of AI include (1) predictive algorithms that are clinical decision support tools; (2) vision recognition applied to pathology and radiologic images; and (3) large language models (LLMs) in clinical documentation to relieve the burden on physicians and allied health professionals as well as helping patients understand their electronic medical record.
“AI has the potential to restore the satisfaction we derive from patient care by removing the burden of repetitive administrative tasks and thereby giving us back time with our patients,” Dr. Yu said.
Using generative AI to reduce the time cancer care professionals spend on tedious documentation tasks will be a key benefit of LLMs, according to Douglas Flora, MD, LSSBB, executive medical director of oncology services at St. Elizabeth Healthcare in Northern Kentucky. Motivated by the promise of AI innovation to transform cancer care, reduce clinician burnout, and improve the doctor-patient relationship, Dr. Flora launched AI in Precision Oncology, the first peer-reviewed academic medical journal dedicated specifically to advancing the applications of AI in oncology. The journal launched in a preview issue in October 2023 and its premier issue will publish in January 2024.
“I think as these large language models start to make their way into clinic, we are going to give doctors back several hours a day that they currently spend documenting their care,” Dr. Flora said. “Ultimately, as we make our care more human, these tools might actually give us time back in the room with our patients to repair the doctor-patient relationship that’s been so transactional for the last 4 or 5 or 10 years. And my hope is that we’re going to go back to doing what we went into oncology to do, to care deeply about the patients in our care and let the computers handle the rote, mechanical stuff.”
Using AI to for medical documentation will be an early gain for hospital systems, according to Dr. Flora. “I think this will be one of the early salvos that hospital systems recognize because there’s a higher ROI if you can give 400 doctors back 2 hours a day,” he said. “The notes will be carefully curated, and they may bring in order sets for the MRI with gadolinium that you forgot you wanted to order; the digital assistant will get that,” Dr. Flora said, adding that reminders will be automatically programmed on the physician’s calendar to call the patient back with their test results and the next set of labs ordered.
AI as the Clinician Co-Pilot
In radiology and pathology, AI is enhancing the resolution of images and increasing the speed of diagnostic processes. In a study published in Lancet Oncology, Kristina Lång, MD, PhD, associate professor of diagnostic radiology at Skåne University Hospital, in Malmö, Sweden, and colleagues reported that AI-supported mammography resulted in a cancer detection rate comparable to the work of two breast radiologists and reduced the screening workload.3 The study found there were 36,886 fewer screen readings by radiologists in the AI-supported group than in the control group, showing a 44% reduction in the screen-reading workload of radiologists. The study also showed that AI in mammography screening is safe.
“The clinical work for a breast radiologist has become more complex, with more examinations and interventions per patient, such as targeted axillary dissection (TAD) and vacuum assisted excision (VAE),” said Dr. Lång, who believes that AI will enhance the work of radiologists rather than replace them. “The use of AI can let us shift our focus from the relatively simpler task of screen reading to the more complex patient-centered tasks. More efficient screen reading is needed in order to sustain the screening program due to the lack of breast radiologists.”
She noted that the study shows that AI can be a valuable tool to increase efficiency and potentially accuracy as well, which is particularly important to reduce false positive results.
Melvin L.K. Chua, MBBS, PhD, FRCR, leads head, neck, and thoracic cancers in the Division of Radiation Oncology at the National Cancer Center Singapore. He said AI applications will enable him to do his job more efficiently.
“Planning treatment in radiation oncology is a complex, multistep process and there is a quality assurance process as well,” Dr. Chua said. “All of this could be automated in some way, assisted by AI, where the human brain is still required to think through individual cases, but AI can start the process rather than requiring the oncologist to start everything from scratch. It will be more efficient for the oncologist to edit the work done by AI, while ensuring quality care for the patient.”
Challenges With Large Language Models
The generation of false information, often referred to as hallucinations, in LLMs is a major problem,4 as are significant concerns about patient privacy.
“We will have to be concerned about privacy and be aware about where these large language models are going to take us,” Dr. Flora said. “I think we should be leaning into this challenge and figuring out how to best use the technology to improve the efficiency of health care delivery and to break down some of the barriers to access that exist in cancer care today.”
Dr. Yu proposed mitigating the problem in the setting of the electronic health record (EHR) where the veracity of data can be trusted, as opposed to tapping into the entirety of online content, noting that accuracy could be further improved if the HER’s data is internally organized into structured data models so the data is not misinterpreted by the LLMs.
“Embedding the LLMs into the EHRs will reduce the risk of data being sent outside the data system firewall which could compromise data privacy,” Dr. Yu added. Exercising caution in allowing external apps to tap into the EHRs will be crucial. He warned that the greatest risk may be the human tendency to overly trust automated systems, such as when hurried clinicians fail to check the content carefully.
Additionally, being aware of some of the limitations of AI products will be important to determine whether the product’s utility is reproducible across different populations or flawed in some way. Dr. Flora said oncologists should have a say in policing bias in AI innovation. “A lot of language models were programmed by people who look like me, white Caucasians with Eurocentric features, and did not include things that were culturally competent… I’m worried about that for clinical trial selection and screening,” Dr. Flora said. “Building databases that don’t represent the patients in our charge is a real concern and so bias is a big deal and that has to be transparent.”
ASCO has launched a task force to look how generative AI and LLMs could impact the lives of oncologists and their patients, as well as ways in which ASCO can support the oncology community in this transformation. Meanwhile, ASCO’s Center for Research and Analytics (CENTRA) is collaborating with the University of Chicago on a project to assess the prevalence of abstracts potentially generated by AI, among the abstracts submitted to ASCO Annual Meetings since 2021.
ASCO Journals will continue to help oncologists navigate the world of AI with authorship guidance on the use of generative word content, and the journals will be a critical medium for the publication of research into the safe and effective implementation of AI in the clinic and in clinical research.5,6
“As the official journals of the largest oncology professional society, ASCO Journals will play a critical role in establishing the role and scope of emerging technologies,” said Jeremy L. Warner, MD, MS, FAMIA, FASCO, a professor at Brown University, deputy editor of HemOnc.org, and editor-in-chief of JCO CCI. “While the landscape is changing rapidly, rigorous approaches such as proving a new technology through well-conducted prospective clinical trials will remain highly relevant.”
AI is increasing in capacity every day, every week, and every month, and it will require vigilance and open-mindedness to best maximize its effectiveness for cancer care professionals and patients.
- De Cremer D, Kasparov G. AI Should Augment Human Intelligence, Not Replace It. Harvard Business Review. 2021 Mar 18.
- Anderson J, Rainie L. Artificial Intelligence and the Future of Humans. Pew Research Center. 2018 Dec 10.
- Lång K, Josefsson V, Larsson AM, et al. Artificial intelligence-supported screen reading versus standard double reading in the Mammography Screening with Artificial Intelligence trial (MASAI): a clinical safety analysis of a randomized, controlled, non-inferiority, single-blinded, screening accuracy study. Lancet Oncol. 2023;24:936-44.
- Meskó B, Topol EJ. The imperative for regulatory oversight of large language models (or generative AI) in healthcare. NPJ Digit Med. 2023;6:120.
- Miller, K, Gunn E, Cochran A, et al. Use of Large Language Models and Artificial Intelligence Tools in Works Submitted to Journal of Clinical Oncology. J Clin Oncol. 2023;41:3480-1.
- Warner, JL. Looking Back and Looking Forward: My Themes for the Continued Success of JCO Clinical Cancer Informatics. JCO Clin Cancer Informatics. 2023;7:e2300107.