Investigating the Emergence of Artificial Intelligence on Oncologic Practice

As artificial intelligence has arrived in the oncology field, experts discuss how this new technology can have implications for transforming clinical practice.

Artificial intelligence (AI), while not new to the oncology space, is emerging as a technology that has the power to transform the clinical landscape. Its applications are vast and growing, with particular focus given to its expedition of documentation, including billing and clinical trial matching; diagnostic modeling, particularly in the breast cancer sphere; and practice standardization and training to better provide uniform and optimized outcomes for patients globally.

Oncologists who are experts in AI implementation weighed in on the transformative nature of these technologies. One noteworthy proponent of AI-powered technologies, Arturo Loaiza-Bonilla, MD, MSEd, FACP, systemwide chief of Hematology and Oncology at Saint Luke’s University Health Network, expressed that he attempts to incorporate as many algorithms in daily practice as often as possible.

“I am a human but also an AI enthusiast. I believe that AI is not a replacement for us; it is a force multiplier, the same way that we have agents that [perform] certain [tasks],” Loaiza-Bonilla said in an interview with CancerNetwork®. “The true value of that is that there are going to be some tasks that we should relinquish to AI, and there are some others that are going to be part of the human experience, and some others that we need to combine AI with humans to [enhance].”

Another expert, Regina Barzilay, PhD, distinguished professor of AI and Health in the Department of Computer Science at Massachusetts Institute of Technology (MIT) and AI Faculty lead at MIT Jameel Clinic, expressed that AI tools have shown strength in use as oncologic tools, but that protocols must be developed to effectively optimize outcomes for patients.

“I would like to emphasize that the 2 main barriers to bringing AI to change outcomes for patients are designing clinical protocols that can utilize the tools effectively. Like everything else, [these are] not miracle tools. They have a lot of power but also limitations,” Barzilay explained when speaking with CancerNetwork. “What ideally would happen here is we will have clinicians who are taking these tools and [have them] develop a pipeline that shows that when we bring the tools into patient care, they improve outcomes and reduce costs. Most of the research in the next few years would be focused on this question, because, again, the tools are already strong enough.”

Furthermore, Ketan K. Badani, MD, vice chairman of Urology and Robotics and director of the Comprehensive Kidney Cancer Center and Reconstructive Urology at Mount Sinai Health System, suggested that the use of AI technologies in oncologic practice will create opportunities for more streamlined and efficient patient care.

“I believe that one of the most important things AI is going to do in the future—maybe in 5 years—is allow us to take a piece of information. For example, [during an ultrasound], a urologist took a picture and uploaded it into the AI system. The AI system will take that image, completely dissect it, put it back together, and tell you this is a grade 2 clear cell renal cell carcinoma without any significant advanced or high-risk features, the probability of this patient metastasizing in the next 5 years is 2.3%, etc, just from a single snapshot. That is what AI can do for us,” Badani expressed in an interview with CancerNetwork.

Ongoing Clinical Trials Investigating AI-Powered Technologies

Currently, numerous applications of AI implementation in oncology are being assessed in multiple clinical trials. One such application, clinical documentation, may have the ability to expedite many menial processes that take time away from clinicians that they could be spending interacting with patients.

Of note, 2 clinical trials assessed AI-powered clinical documentation, the first of which was a nonrandomized study published in JAMA Network Open.¹ Results from the study revealed that approximately half of clinicians reported time-saving benefits and improved electronic health record (EHR) experiences using AI-powered clinical documentation tools. Additionally, although a subset of clinicians reported no time-saving benefits or EHR experience improvements, the study investigators suggested potential selection and recall bias may have served as study limitations. They explained that further research was warranted to identify improvement opportunities and understand the impact of these tools on different clinician subsets.

Another clinical trial published in Cureus assessed 36 studies discussing the impact of AI technologies on accuracy and efficiency in clinical documentation.² The results from the study suggested that few accuracies and hallucinations were observed with GPT-4 generated summaries, with errors observed with large language models (LLMs) such as omissions, misinterpretations, certainty illusions, fabricated information, and attribute errors, requiring significant editing to correct.

Additional findings revealed that LLM integration into clinical workflows was associated with a lower documentation burden on clinicians, with safety risks posed through potential fabricated information. Furthermore, LLM-generated discharge summaries were found to possess greater readability and accessibility to patients while reducing clerical work for physicians. Moreover, the study outlined a need for legal and ethical considerations, noting they were crucial in safely integrating AI-powered technologies into health care systems.

Loaiza-Bonilla asserts that these technologies can help alleviate physician burnout and help them better utilize their time to treat patients.

“We can use those technologies to summarize the patient notes, [assessing biomarkers that the patient has been tested for], and then find opportunities for optimization of treatments, or at least make my life easier documenting,” he said. “There has been a dramatic shift in how we approach these tools to make our lives easier. We were burned out—and we still are—by menial tasks such as documentation, just for billing, instead of focusing on patients. Saving hours of a day by using these tools has made a [significant] difference.”

Another application, diagnostic modeling, particularly in the breast cancer sphere, may increase cancer detection rates and reduce interval cancer occurrence, potentially expediting cancer care in patients who are at risk and ultimately improving outcomes.

One trial evaluating AI-supported screening against standard double reading was the Mammography Screening with Artificial Intelligence (MASAI) trial (NCT04838756).³ Results recently published in The Lancet, revealed that the cancer detection rate was 29% (95% CI, 9%-51%) higher with AI-assisted mammography screening vs traditional double reading methodology. Additionally, AI-supported screening achieved a positive predictive value of recall of 30.5% (95% CI, 27.8%-33.3%) vs 25.5% (95% CI, 22.9%-28.3%) with standard double reading, a statistically significant increase of 19% (95% CI, 4%-37%).

An additional trial, which Loaiza-Bonilla suggests MASAI is a precursor study for, is the Early Detection using Information Technology in Health (EDITH) trial.⁴ This trial is purported to enroll approximately 700,000 women across the UK to undergo AI-assisted breast cancer screening using 5 AI platforms across 30 National Health Service sites.

Barzilay remarked that despite AI tools exhibiting improved radiologic performance and reduced false negative rates in multiple bodies of work, these tools still have yet to be integrated into clinical practice.

“With AI, you can at least increase the performance of radiologists,” she said. “You can decrease the number of patients who are called back for false negative [tests]. We already have the elite, clear body of work that supports [the adoption of] these tools, but for counseling reasons, they are still not a part of clinical practice.”

Programs and Future Considerations For AI-Assisted Clinical Practice Strategies

Numerous programs have been established with the purpose of utilizing advanced AI technologies. One such program, Advanced Analysis for Precision cancer Therapy (ADAPT), seeks to track tumor changes through treatment lines by using predictive modeling and tumor biology advancements.⁵

Barzilay suggested that in light of patients having few options after advanced cancer mutations render ongoing therapy useless, the ADAPT program will attempt to predict the development of immune escape mutations in patients utilizing measurements taken during biopsy. To that end, the program would help predict when patients would stop responding to treatment and suggest a more effective option for therapy.

Another program, one in which Barzilay has had direct involvement with, is Learning to Cure.⁶ This program seeks to utilize machine learning, natural language processing, and computer vision models to better improve disease progression models, curtail over-treatment, and ultimately cure breast cancer.

Barzilay went on to explain that the program was established to develop AI tools for screening and disease prediction and then to retroactively assess these models across global practices so that radiologists can validate whether patients abroad have certain diagnoses.

“The problem with screening tools, with these risk assessment tools, is that humans cannot validate them at a time when they are using it,” she said. “The radiologist can tell you the patient has a particular mark, and they can say, ‘It makes sense that this patient will be predicted [as] having potential cancer and [needing] a biopsy.’ When you are thinking about predicting risk from a country that is not [in the database], there is nothing that the human eye can detect. Human radiologists cannot validate the prediction that the patient is high-risk. That’s why it's important to take these tools and test them on different patients from different hospitals, countries, ethnicities, and ages.”

In terms of the future implications of AI-powered technologies on clinical practice, Badani expressed that among a ‘laundry list’ of potential opportunities, surgical standardization is a highly relevant application he highlighted.

“We now have AI systems in place that can look at video and can tell what is happening based on the video,” he said. “AI will standardize the quality of how [a procedure is] done. It will study the moves of the surgeons who have done millions of these—the best of the best—and translate it to everyone else. Now you have democratized it.”

Furthermore, Badani suggested that AI-powered applications can be used as a means of training doctors with less experience to perform optimized surgeries without having to perform them on actual patients.

“You only have so many doctors that [perform surgeries with such frequency]; they can only take care of so many patients,” Badani said. “[There is a] need to train other doctors to be up to that level. Without doing 1000 of them, how are you going to get physicians to that level? We get back to augmented reality, AI, and machine learning helping surgeons [perform] the same operation. Then, it [becomes] available to more patients across the world.”

Regarding the oncologist’s role in clinical practice in an AI-emergent landscape, Loaiza-Bonilla highlighted a phenomenon known as Moravec’s paradox, which suggests that the role of AI will help enhance concepts that are comparatively harder for humans, such as mathematical calculations, but that tasks that humans have evolved to refine over millions of years, such as motor or social skills, would still be optimally performed by humans.⁷

In that sense, humans would still have a role in delivering cancer care, performing roles that require skills that necessitate refinement through millions of years of evolutionary pressure. By contrast, computational-oriented tasks could be relinquished to AI-powered technologies to best optimize clinical practice.

Furthermore, regarding ethical considerations for the integration of AI tools into practice, Barzilay emphasized the need to apply regulatory standards to these emergent technologies like other tools being developed for patient care.

“It’s interesting to see that many of the tools that the NCCN recommends are supposed to be FDA regulated––they are not,” Barzilay expressed. “When it comes to AI, [these tools] have to be regulated, but they are not for historical reasons. We need to apply the same standards for all the tools that they use in patient care to provide patients with the best [outcomes]. [FDA] guidelines would play an important role in bringing this change.”

Loaiza-Bonilla concluded that the next steps for AI integration into clinical practice involve continuous validation of these tools, to detect cancer in earlier stages and mitigate the risk of malignancies.

“Perhaps in the future, we get [cancer diagnoses through] a multimodal scan that detects cancer before it shows up, with a blood draw and a scan. We can use the LLMs and [alternative] models, transformers, and beyond,” he explained. “You get it off the shelf, take it, and you can target multiple agents and then be done with your cancer before it even becomes a problem. That will be amazing. For that [to happen], there is still a lot to work [to do].”

References

1. Liu T-L, Hetherington TC, Stephens C, et al. AI-powered clinical documentation and clinicians' electronic health record experience: a nonrandomized clinical trial. JAMA Netw Open. 2024;7(9):e2432460. doi:10.1001/jamanetworkopen.2024.32460
2. Lee C, Britto S, Diwan K. Evaluating the impact of artificial intelligence (AI) on clinical documentation efficiency and accuracy across clinical settings: a scoping review. Cureus. 2024;16(11):e73994. doi:10.7759/cureus.73994
3. Hernström V, Josefsson V, Sartor H, et al. Screening performance and characteristics of breast cancer detected in the Mammography Screening with Artificial Intelligence trial (MASAI): a randomised, controlled, parallel-group, non-inferiority, single-blinded, screening accuracy study. The Lancet. Published online February 3, 2025. doi:10.1016/S2589-7500(24)00267-X
4. World-leading AI trial to tackle breast cancer launched. News release. Gov.UK. February 4, 2025. Accessed February 7, 2025. https://tinyurl.com/3hna99we
5. ARPA-H launches program to transform cancer care. News release. ARPA-H. Published March 7, 2024. Accessed February 24, 2025. https://tinyurl.com/4ttykr5m
6. Our mission. MIT Learning to Cure. Accessed February 24, 2025. https://tinyurl.com/bdzbje7m
7. Arora A. Moravec’s paradox and the fear of job automation in health care. Lancet. 2023;402(10397):180-181. doi:10.1016/S0140-6736(23)01129-7