AI in Oncology: Opportunities and Challenges for NSCLC

Ryan Nguyen, DO, physician and researcher at the University of Illinois Chicago (UIC), emphasizes the need for personalized care in managing non–small cell lung cancer (NSCLC), considering patients' unique life circumstances, comorbidities, and treatment goals to optimize quality of life.

He also discusses the potential of artificial intelligence (AI) in cancer care, highlighting its promise in aiding complex decisions while cautioning against its limitations, such as high rates of unsupported recommendations, and stresses the importance of rigorous validation to ensure patient safety.

Nguyen participated in a panel discussion on immunotherapy advancements in NSCLC at a recent Institute for Value-Based Medicine^® event in Chicago.

This transcript has been lightly edited for clarity.

Transcript

With increasing focus on patient-centered care, how can clinicians better support patients with NSCLC in understanding and navigating complex treatment options?

We're definitely in a very complex era of advanced non-small cell and cancer management. The typical patient that I see come in is going to be an elderly patient who maybe have a few more comorbidities and is really navigating how do they manage the side effects and the symptoms that they're feeling from their cancer, with this variety of options that they have to treat the cancer, as well as the side effects that might come with that.

And so, I think it takes a very personalized approach of understanding not just what's going on with that patient's cancer, but also what's going on with that patient's life, their comorbidities, and their quality of life and what's important to them. I think it's going to require all of us as oncologists understanding the very unique side effect profile that we're seeing with all these different classes of treatment and trying to personalize that to the point where we can really try to optimize these patients' quality of life and see what's consistent with their goals.

Given the complexity and heterogeneity of NSCLC, what role do you see for artificial intelligence or machine learning in supporting diagnosis, treatment decisions, or predicting outcomes?

Well, I think we're starting to see that AI is starting to take over the world, and you're seeing it all over the news now. I certainly think it has a lot of potential in cancer care. I think the challenge is that there's so much new information coming out there that you really have to validate what's coming out in these different AI mechanisms.

One of the research projects that we've led at UIC is actually we asked ChatGPT, "If I had a patient who's got advanced non–small cell lung cancer, and if they have an EGFR, an ALK, a ROS mutation, what treatment should this patient get?" We found that ChatGPT 4.0 was definitely better than 3.5, but when you look at the actual results that come out, even though it was giving us a lot of the recommendations that were in the NCCN [National Comprehensive Cancer Network] guidelines, it was actually hallucinating at a fairly high rate as well. So about 30% to 40% of the findings that we were seeing were actually complete hallucinations—treatments that the algorithm was thinking that it was just seeing that these combinations [of treatment and disease] were in the literature, but it actually wasn't supported by the data.

So I think that's the challenge: in precision oncology and cancer care, you can't get the answer wrong. You can't give the patient the wrong treatment. I think that's the challenge with trying to integrate AI into cancer care in that there's so much data that gets there that I think we need to have very strong guardrails on making sure that there's a very low threshold for hallucinations in using AI in cancer care.