How To Use AI in Your Cancer Care

Lisa Hatfield:

Artificial intelligence is becoming part of everyday life, and increasingly, patients are using a number of growing tools to ask questions about their health, understand medical terminology, prepare for appointments, and even make sense of laboratory results. But using AI effectively and safely requires more than simply typing a question into a chatbot.

Hello, I’m your host, Lisa Hatfield, and this Patient Empowerment Network DECODE podcast, we’re exploring how people living with cancer can use AI as an educational and communication tool while recognizing its limitations.

Please remember to download the resource guide for key takeaways and resources to help you better understand how to use AI as a cancer patient.

Okay, joining me today, I’m really excited, is Dr. Landgren, director of the Sylvester Myeloma Institute and a physician-scientist whose research is exploring how artificial intelligence, or AI, can be used to advance precision cancer care and support more individualized treatment decisions. Thank you so much for joining us, Dr. Landgren. I’m really excited to see you.

Dr. C. Ola Landgren:

Thank you very much for having me, Lisa.

Lisa Hatfield: 

Also joining us is Kacy Hatfield, a PhD researcher at Arizona State University’s Heart AI Lab, where she studies interpretable artificial intelligence and how people can better understand, evaluate, and trust AI systems in high-stakes environments.

Kacy also happens to be my daughter, giving her a unique perspective at the intersection of AI, caregiving, and the patient experience. She’ll help us understand the technology behind today’s AI tools, and how patients can use them more thoughtfully and effectively.

Kacy Hatfield: 

Thanks for having me!

Lisa Hatfield: 

Alright, so we’re gonna jump right in. Dr. Landgren, artificial intelligence is moving very quickly, and patients are no longer just hearing about AI, they’re using it to understand lab results, treatments, symptoms, and other health-related questions. So, from your perspective, as a physician and a researcher, how significant is this shift, and what do you think patients should understand about the role that AI can play in their cancer care?

Dr. C. Ola Landgren: 

I think patients that I see in my clinic very frequently say, I have a lot of questions, and before I ask you, I’ll tell you, I already asked the AI, so I think I know the answer. And I would say, please go ahead and ask me the questions to see if I know the answer.

And I can tell that the AI has really provided patients with new perspectives that I don’t think I heard in the past. AI sort of takes it to another level. Sometimes I notice also that the AI can not really sort of put on a scale what’s very important and maybe what is not so important, and I think that’s where the doctor comes in.

And also, there could be details that could be missed for the individual patients. So, on an overall level, the AI could do very well, but when it comes to the exact details for the patient, sometimes it’s not perfect, and I think that’s where the doctor comes in.

Lisa Hatfield: 

Ok, thank you. So how often do you have patients come in and say, hey, Dr. Landgren, here’s my list of questions? One of them is for me, and 10 of them are from AI. Do they come in very often with their AI-generated responses?

Dr. C. Ola Landgren: 

I would say patients very frequently come in, and they would tell me that they have asked the AI, but sometimes patients don’t tell me that, so they ask a lot of questions, and I would say, these are really good questions. I can tell you did a lot of homework, and as the conversation goes, sometimes patients tell me, I actually did ask the AI.

This week, when I was in clinic, I had a patient that came back for a follow-up visit, the patient came to see me just a few weeks ago and had need for new treatment, and we had talked about the clinical trial, and I said to the patient, I don’t want to pressure you, I want you to be informed, so I will tell you what standard of care would be, but I will also share with you we have a trial, and then I gave a copy of the informed consent.

So the patient went home and read the consent and came back and saw me this week and said, I actually asked the AI, and the AI said, this trial is great, so I want to go on the trial. I said, is that what you think as well? Oh, yeah, it’s the same. I just wanted to tell you that I asked the AI, the patient said.

Lisa Hatfield: Oh, that’s interesting. So, I actually heard of a patient who took the document from a clinical trial, copied it, put all of that, plugged it all into AI, and had AI summarize the important key points to it. So that’s fascinating, that’s awesome they use AI…

Dr. C. Ola Landgren: 

Yeah.

Lisa Hatfield: 

to try to make that decision, along with your expertise, of course.

Dr. C. Ola Landgren: 

So, I think AI can be a good help, and could be… serve as a good partner, and for me, as a doctor, I use AI when I develop new ideas, when I, for example, put together slides, or I’m outlining a new paper I’m planning on writing up. I think of the AI almost like working with a very good postdoc.

So I would ask the AI, I have all these ideas, I drafted this, what is your, what is your take on this? And the AI would give me back feedback. I would not just copy, paste, and use it, I will… I will look at it very critically. And sometimes I spot major, major issues, so…It happened more than once that the AI would say, this is what I would do, and I would say, what’s your reference for this? And the AI would say, here are 5 references.

And when I look at them, one of the references could have my name, and I say to AI, I never wrote that paper, you made that up, and the AI would say, oops, I’m sorry, I made that up. So the AI is not totally reliable.

Lisa Hatfield: 

Right, yeah. So, Kacy, building on Dr. Landgren’s comments, to get us started, when we talk about AI, what exactly are we referring to? And maybe you can briefly explain AI in some of its major subcategories, and where large language models, or LLMs, fit in.

Kacy Hatfield: 

Yeah, so AI is a very broad term. Essentially, given input goes through some calculation and returns an output. Under that umbrella, large language models is what we’re talking about here, mostly, like ChatGPT, Claude, Copilot, and Gemini, and those are a subcategory trained to understand and generate human-like text. They’re great at explaining and organizing information, but ultimately they’re predicting likely words based on patterns.

A lot of people don’t realize how common AI is in day-to-day life. Home assistants like Alexa or Google Home, as well as Siri, utilize AI. And even now, when you type in a question to Google, oftentimes it’ll show you an AI overview, and that’s Gemini coming up with those responses.

Lisa Hatfield: 

Okay, so when you talk about Gemini, that’s one that’s just automatic with Google, so there are other platforms, too, are there, like, Claude and Perplexity and all those, are those, subscription-based only, or can people just use those other AI platforms? Is there a way just to access them for free?

Kacy Hatfield: 

Yeah, so you can access them for free and create accounts to save chat history, although there are paid versions, the more memory you want stored within the chat history window.

Lisa Hatfield: 

Okay, that’s super helpful, because sometimes I’m a little bit skeptical of AI as a patient, so I like to plug in a question to one platform, and then into another platform to see if I can get them to disagree. So, okay, well, thank you for that.

Dr. Landgren, patients are increasingly using this generative AI to understand their healthcare, their treatments, their lab results. What is the most constructive way, when a patient comes in and brings in their information with something that they’ve learned from AI, how can they bring that into their appointment without creating confusion or straining the patient-doctor relationship?

Dr. C. Ola Landgren: 

Well, I think it very much depends on the individual situation. So, say, a patient is maybe newly diagnosed, has not seen a doctor before, has a lot of symptoms. I think the doctor will ask the patient, how are you feeling, what type of symptoms you have. I think it’s really about the patient side of it.

But I think when it comes to a decision-making situation, maybe you’re choosing between different therapies, or you’re choosing between different types of tests, or you’re… thinking about more subtle details, the AI can do a good job putting together and comparing a lot of different options. I do think a doctor with a lot of experience is not that different from a sophisticated AI model.

Maybe doctors that don’t see as many myeloma patients could have help from an AI to really synthesize a lot of details, but I would think that a super specialized myeloma expert probably gives similar or better results than the AI when it comes to the granularity. But having it at home and being able to ask the questions whenever you want to ask them, I think that’s an extra help to the patient.

So I think use it, if you have questions at home, you can bring the questions back to the doctor, have a conversation. But it really depends on what the exact sort of avenue for these questions are, if it’s about what to do next, or value, and compare different options. I think that could be a good situation.

Lisa Hatfield: 

Okay, thank you. And I wonder how many physicians are receptive to that question, or if when a patient walks in the door. “Hi, doctor, I have these AI… this AI information.” It sounds like you’re quite receptive to that, and I think my oncologist here is also very receptive. I just wonder how many, physicians or providers are that receptive, hopefully a lot of them.

Dr. C. Ola Landgren: 

I usually say to patients when they come, I’m happy to answer any questions you may have, and my scope is to really help you the best possible way. If they want to stay with me and want me to follow and treat, I’m happy to do that.

If they’re coming just for a second opinion visit, and they want to go back to someone else, they never hurt my feelings. And after we have gone over everything, and the patients have asked me many, many questions, I usually say, is there anything we didn’t cover? Is there anything more you wanted to ask?

And sometimes they have asked questions for 30 or 40 minutes, but I still always ask that question in the end. Is there anything else you want to ask? And if they have nothing else, I would say, why don’t you write anything down? If you remember some additional things when you come back… when you come home, and when you come back next time, we can talk about it again. So I always want the patient to feel that they can ask me anything. And if I don’t know the answer, I would say, I’m afraid I don’t know the answer, or there may not be a study that has looked into this, so there is no known answer to this.

Lisa Hatfield: 

Okay, and then from your perspective, Dr. Landgren, as a physician. What concerns you the most about patients relying on AI for medical information? Have you had any concerns when they brought information in?

Dr. C. Ola Landgren: 

I do think that the AI can do a very good job, sort of, providing overviews of a lot of options, but sometimes there are a lot of sort of gradients of which is the most important, or if you put a lot of other factors into perspective, that if these and those things are also present, maybe this is a better option. The AI may not always have that granularity, and I’m not an expert on how to program all these AI models, but I would assume that the models may not be as deeply trained. I’m sure that we can have better and better models in the future, so maybe this is just a matter of time that the models get better at these things.

Lisa Hatfield: 

So, Kacy, from the technology side, what are some of the limitations that patients should understand when using generative AI for health questions?

Kacy Hatfield: 

So similar to what Dr. Landgren was saying, AI can be very incorrect, and it can be very confident in its incorrect answer. Within the field of AI, we call these hallucinations, and that just means that its output has an error, and more often than not, in terms of AI platforms, a lot of times they’ll be very confident or very sure that the answer is correct, and keep telling you that it’s correct even though it may not be. So that’s something to take any output from an LLM with a grain of salt.

Lisa Hatfield: 

Okay, good to know. And then are there… I’ve heard that sometimes there are biases in some of the data. Is that a limitation that patients need to be aware of, or can they just plug in their question and hope that what comes out knows to eliminate those biases?

Kacy Hatfield: 

Certainly be aware of the biases within AI platforms. The training data that has been collected, is just a reflection of data collection biases we’ve had over the past 100 years or so, and those will be reflected in the outputs of the models. So that’s currently a big field of AI, is trying to

make more generalizable models by making our training data more of a replication of real-world scenarios as opposed to historically, where bias has led us to collect more data from.

Lisa Hatfield: 

Okay, makes sense. All right, now we’re going to skip ahead a little bit. So this is kind of practical information for patients on how to ask AI better health questions to get the answers that we want.

So, our friend came up with this acronym called CRAFT, and Kacy’s gonna walk through the definition of CRAFT, what each letter means, and then we’re going to ask some questions and apply this CRAFT acronym, so if you want to use it in your daily life, with your health questions, you can use it. So, Kacy, spotlight’s on you to explain CRAFT.

Kacy Hatfield: 

Sure. Craft is a simple checklist for getting useful answers from AI. So it stands for

Context, Role, Ask, Format, and Tune. Context means giving the AI the background it doesn’t currently have, so that could be your diagnosis, cytogenetics, treatment history, recent labs.

And that’s to create more specificity within the questions you’re asking. Role means telling it who to be, so you would type directly into Claude or whatever AI platform you’re using, act as a patient educator, for example, which changes the vocabulary and the depth of the answer.

Ask means using a specific verb, so some of our favorites are explain, compare, and draft. And format means telling it how to package the answer, and ultimately, this is whatever is most digestible to you. So that could be a short bulleted list, or a table you can actually bring to an appointment.

And tune means following up in the same conversation instead of starting over. A lot of times, these AI platforms, especially if you’re logged into an account, will keep a chat history, and some of the context will be saved in memory. So one clarifying question almost always turns a mediocre answer into a genuinely useful one.

Lisa Hatfield: 

Okay, thank you for defining that. So, we are going to put this into practice. We have two real-world type of questions that a patient facing a cancer diagnosis might ask, and we’re going to look at how changing the prompts can change the usefulness of the response. So, we chose a blood cancer and a solid tumor cancer, because this is a broad audience we have.

So, the questions are similar, and once we’re done, Kacy, if you can apply this CRAFT acronym to both questions at the same time, that would be great. So, first question.

Lisa Hatfield: 

I was diagnosed with multiple myeloma, a blood cancer. What does this mean? What is my prognosis and treatment? And will I be on chemo forever? Same question for lung cancer. I have, I was diagnosed with stage 3B lung cancer. What does this mean? What is my prognosis and treatment? Will I be on chemo forever? Same type of question. So, using this CRAFT acronym, how could I, as a patient, write that a little bit better to get a response that’s more useful to me?

Kacy Hatfield: 

So starting with context, instead of, I was recently diagnosed with stage 3B lung cancer, perhaps something like, I was recently diagnosed with stage 3B lung cancer, with an EGFR mutation, or similarly for multiple myeloma, translocation t(11;14), any of the cytogenetic information. And then in terms of role, this can be right at the beginning. You can use act as a Patient Educator, telling the AI that depth and type of information you want to receive.

In terms of ask, instead of asking what is my prognosis and treatment, you could frame it as, explain to me the diagnosis, and please list standard of care chemo regimens. That goes along with format, which is how it packages the answer, so you could ask it to explain in a bulleted list, or perhaps define any jargon used.

And then T for tune, that wouldn’t be within this question, but after. You could save this question in a chat, and then come back to it later for clarifying questions, either after a visit, or if you have questions that come up, and iterate within the context you’ve already given.

Lisa Hatfield: 

Okay, so kind of just to summarize for patients listening to this, so the more information I give it, it’s not… the more detailed information that I give it, the better, without giving any identifying information. So, in context, I will say the type of cancer I have. I have kappa light-chain–only myeloma or I have non-small cell lung cancer (NSCLC), plus the stage, plus any cytogenetics, plus can I enter lab work in there, too? Like, these are the labs that I have, could I put that in there? Is that… that’s probably going to be a little more useful.

Kacy Hatfield: 

It’s up to user discretion with what you’re willing to share privacy-wise. However, it could be uploaded into an AI platform.

Lisa Hatfield: 

Dr. Landgren, what are your thoughts on that, on how AI responds to these questions?

Dr. C. Ola Landgren: 

I work every day thinking about what would be the next question, and trying to push the boundary, and provide patients with access to the newest therapies, the newest concepts, and so forth.

But I think inherently, if we look online or wherever we look for information, if you go to a traditional old library and pull all the books, everything that is published, presented, will always have a bias of being old information.

So, all the old treatments have been documented over and over and over again, but the newer therapies have maybe not even been published yet. They may only be presented at the conference, that could be one talk, or so forth. Could be maybe one talk in Europe, one in the United States, while the older therapies, there could be thousands of references pointing at these old studies. So when the AI takes all this into account, it will have a bias towards the history, because that’s where the information is. And I do think that’s where the expert can help the individual patient.

So, striking examples I could see in the myeloma field are, for example. Should you use immunotherapy for newly diagnosed patients? What type of immunotherapy should you use? Is the role of minimal residual disease detection? Is that important? Is transplantation necessary for every patient going forward with better and better drugs, if you reach MRD negativity, is it really necessary to do a transplant, or is that maybe not necessary? Is that going to prolong your life? Is it going to cause other toxicities?

Et cetera, et cetera. They’re all historical therapies were based on 3 drugs, they had always transplantation, they didn’t have immunotherapy, and there were no MRD testing tools at the time. But in the future, I think the field is going in that direction, so…if you go to the AI, and you… you don’t really know all the details of the field, as Kacy said, the AI is going to sound extremely confident, and they’re going to say there are these many studies that have looked at this, and this is what the literature shows. And that is, of course, factually correct.

But that may not be really what you as an individual, are looking for. You may look at what’s more relevant to me, and where’s the field going in the future. And I think an expert in the field will be the best person to help and answer, but the AI can provide a lot of perspectives and raise a lot of questions that can be used to the discussion.

Lisa Hatfield: 

Yeah, and that’s a really great thing that you point out, because I also attend, just as a patient advocate, the ASCO, American Society of Clinical Oncology, and then also the ASH meetings.

And when I tried to ask… AI gave 4 pages of information. When I tried to ask to summarize what was presented June 2026 ASCO meeting, or ASH December 2025 meeting, it cannot do that. I don’t think it can access that data, so that’s a really great point you bring up. It’s going to be based on historical data. That’s what’s in this database, in this in this data set. So, that’s really helpful. And another reason why using AI as a tool versus a decision maker is really important for patients. We need to get the expertise of our doctors and not rely on AI for decision making. We can use it as a tool, but not for decision making, I think. So, yeah, I appreciate that perspective, Dr. Landgren.

Then, just backing up just a little bit, so do you have a preference? Do you use different AI platforms for health-related questions? You probably use very sophisticated ones, but do you have a preference for one or two of them, and what are those?

Dr. C. Ola Landgren: 

Yeah, for my own work, I use Claude, I also use ChatGPT, and I also use Gemini, and I use them for different reasons. I think ChatGPT I use for certain types of searches I do. Claude, I like if I… like to use if we are trying to put together information to write up for papers and come up with ideas.

I think ChatGPT can be pretty good if you do coding or other things. And I think Gemini, I think it’s built into the Google platform, so if you do Google searches, you get that whether you like it or not. So I use all of them, and sometimes I ask all the three of them the same question, and you get a little bit different answers, which I think it’s not different from life in general. You can ask three different people, they will tell you three different answers, right?

Lisa Hatfield: 

Yeah. All right, thank you. And Kacy, do you have a preference for health-related questions? Which platform?

Kacy Hatfield: 

While I am not as well-versed with health-related questions, because I am in the explainability, programming, and ethics space of AI, I similarly utilize Claude most commonly, and then secondarily, I use Notebook LM, which is from Google, and it allows me to upload all the papers I’m trying to read and get audio overviews in the form of a podcast, which is one of my favorite modalities to get information.

Lisa Hatfield: 

Oh, it’s great, yeah. So, final question, if you could give a patient just one rule to remember before using AI for health-related questions, what would that rule be? Kacy?

Kacy Hatfield: 

I would say use AI to help you develop better questions, and maybe promote new perspectives, not getting final answers. If you leave feeling like you know your prognosis or treatment plan, that maybe is an indication to take it to your care team.

Lisa Hatfield: 

Okay, and Dr. Landgren, how about you? One rule for a patient to remember.

Dr. C. Ola Landgren: 

I think, never upload information about yourself, like, our HIPAA data, because everything that you upload in the cloud, maybe someone else can find it, so you don’t want to take those risks, so be careful when you upload information, so you don’t put your own identifiable information out there. And I also think, as we spoke, and I agree with Kacy, that don’t take all the answers as necessarily the full truth and the law, so it could be viewed as one perspective, and I think it’s important to have a strong provider and a providing team that can provide long-term care and also guidance on the optimal treatment route and monitoring for clinical care.

Lisa Hatfield: 

Well, Kacy and Dr. Landgren, thank you so much for helping us better understand what these tools are, what they can do, and how patients can use them more thoughtfully and responsibly. And to everyone joining us, remember, AI can help you learn, prepare, and ask better questions, but it does not replace the expertise of your healthcare team or the context they bring to your individual care.

Thank you for joining Patient Empowerment Network’s DECODE podcast, I’m Lisa Hatfield. Thank you for being with us.