Everyday Use of AI and ChatGPT in Oncology Practice

Arturo Loaiza-Bonilla, MD, MSEd

Medical Director of Oncology Research

Systemwide Chair of Research

Capital Health

Cofounder, Massive Bio

Philadelphia, PA

Targeted Oncology: What is the significance of new advances in AI for the oncology space?

The new advances in AI, particularly for oncology, are transformation and its constant evolution. But there was not a lot of knowledge, even though those developments have been happening over the past 3 decades, at least, and little by little improving optimization of workflows or using data as a way to advance AI. It can, first of all, help us to optimize what we do on a daily basis in our practices. For example, it can make our lives easier [by helping us spend] less time doing medical record input, be much more efficient in diagnosing cancer, or even identify disease beyond our current capabilities. Radiologists can find nodules and pathologists can find areas of malignancy that our eyes may not be able to detect, and are then able to make a diagnosis.

[AI] can help us to enhance early cancer detection following certain biomarkers or trends using existing data. It can personalize treatments by matching biomarkers to treatment options and clinical trials. It can help us with drug discovery—there are a number of models doing this as well— in patient care, which right now is the most important use because it’s making optimization of our time much more efficient.

As we analyze large data sets such as genomic information, medical images, and precision oncology to tailor treatments, it can improve the effectiveness of treatments, and also make them timely and available for many patients and to us, who are busy running our practices.

How does ChatGPT help oncologists and patients?

ChatGPT took everyone by storm. It started just by being used for random questions, and then it started to evolve into something much more meaningful. ChatGPT is an AI large language model that was not necessarily trained for health care, but it can be trained.

As we are part of those efforts, we can use it as a tool for our improvement. It can analyze quickly and synthetize large amounts of medical literature—so it can go into PubMed and read all the articles at once and find the most important and relevant abstract, for instance. Even for queries, it is a relevant tool. ChatGPT [will pull random sources] if it doesn’t have the data, but if we embed the data, it can be effective. For example, at the 2023 American Society of Clinical Oncology [ASCO] Annual Meeting, imagine the if tool is able to surface the most relevant ASCO abstracts that are going to be of interest to a particular person to navigate in between meetings.

If I want to go into the pathway of precision oncology or to talk about certain biomarkers only, it could give me a path in terms of the abstracts that are going to give me the information that I’m looking for.

The opportunities are endless, but they are mostly around data. ChatGPT can also empower us not only with up-to-date knowledge, but it can be a reliable source of information if we train it. It’s one of those things that can simplify information. It can simplify access to that information so we can learn it ourselves and communicate it to our patients.

Are there uses for AI and ChatGPT in all cancer types?

Yes, this is pretty much dealer’s choice, so it can work with many cases. Using the same model as a whole can personalize it for gastrointestinal malignancies and neuroendocrine cancers.

One thing that it can do is help us to surface the most relevant data in terms of the combination of biomarkers that can be utilized for better targets. Imagine that you posed the question: “What potential studies currently in development for neuroendocrine tumors have patients in second line who were exposed already to capecitabine/temozolomide and have a Ki67 of at least 20%?”

Those questions can be posed in large language models to help find the studies that are available, to find the preclinical data that came from the American Association for Cancer Research Annual Meeting or ASCO. In terms of clinical care, we can facilitate multidisciplinary care by streamlining this communication using ChatGPT and other models. It can be used for personalized resources and guidelines if I want to explain to my patient about certain therapies and be able to answer questions about it.

We can train the model to not only embed the most relevant data that we can understand, but also to explain to the patient each of the steps needed and the preparation for those procedures. It’s an interesting opportunity to leverage those models for improved patient experience.

What do you think are some of the biggest changes coming down the pipeline because of AI?

The sky’s the limit now with all these new options. One of the most significant changes is, I believe, the democratization of cancer care and information, making a specialized knowledge much more accessible and digestible. Hopefully, if we’re able to solve the digital divide in terms of access to electronics and broadband, etc, it can reduce disparities in access to care.

The internet era with the Web 2.0 was amazing because it was able to give us the universe of knowledge at our fingertips, but then it became too much; we don’t know what to look for, we don’t know where to search. Now we can use AI to guide us into the right treatment.

It can help us to improve access by meeting the patient where they are. It can translate in real time and give the patient information in their language or in their cultural context, give them what is important in their care.

Another component is precision medicine, which is now the standard of care [most of the time]. We can use these AI tools to show the best treatment options, given the circumstances of a given patient.

Or in a tumor board, we can get the information in real time so we can make decisions in a much more informed way, and not just rely on encyclopedic knowledge that we are so used to in medicine. That’s another thing that I feel is ready to [become a part of care]. We can also use AI for enhanced patient monitoring, integration of wearable technologies with AI, or even the electronic health records.

There are some efforts between OpenAI, Epic Systems, and others to do more practical disease management and give you in real time what is needed for the patient. Additionally, how we can solve and decrease adverse events, or know when the patient needs a specific biomarker test for targeted treatment, becomes a multivariable approach because now it’s all data and it’s a matter of organizing the data. We can do it for clinical care, clinical research monitoring, and clinical research by developing synthetic control arms with the same information. We can find new compounds, surface and identify exceptional responders, and look at data in a large data set, making those insights possible for us to know what’s important.

In what ways have you been using AI or ChatGPT in your practice?

I now use it as my mini assistant. I’m using it to, for example, do my presentations; I put the abstract there and I say, “Summarize it in 5 bullets,” so I can have a conversation about it. Or I’m going to talk to the patient about this specific drug, so I ask, “Can you give me a synthesized way to explain, in simple terms, what this clinical trial is about?”

I’m also using it for some form of assessment in terms of genomic alterations. There are specific mutations that I have [AI] acknowledge plus how to leverage that information when using it as an additional tool that I can explore.

I’m in this presentation mode right now, and I have my 2 hats—I’m the AI entrepreneur and clinician—so I’m trying to manage that intersection appropriately to develop those new use cases. That’s why it’s so important for me to have skin in the game and keep practicing because that’s how I’m going be able to find the best use cases [and find the right] models.

What do you think the relevancy of AI will be going forward?

I think that we all need to learn about AI, we need to be part of the conversation, and we need to embrace it as a tool to enhance our practices. I’ve heard that this is going to replace physicians—I never believed this because of my own experience living at the forefront of AI; our patients and physicians are involved in this. We still yearn for having that personal touch. We want to talk to patients; the patients want to talk to us.

Even when telemedicine is available, they want to discuss those tough questions in front of us. They want to see a human in front of them because they don’t want a machine to make the decision. Clinicians probably want a machine to solve issues such as assessing the lymph nodes so they can go to billing, document the experience that they just had and the discussion, so it can be compensated or reimbursed.

Maybe clinicians will use it to solve [issues] such as clinical trial matching or finding cohorts or biomarkers. Beyond that, I think that this is something we should try to embrace in a positive way. Use AI as a way to enhance our abilities to perform better, to be better doctors, providers, and caregivers. This is not going to replace anyone; it’s just going to make us better. It’s up to us to embrace that change and make it meaningful and use that technology for the benefit of all of us.