AI's Role in Oncology: Supporting, Not Replacing, Health Care Providers

Although artificial intelligence (AI) offers great promise in oncology, some express concerns over the loss of human connection. In the second half of our interview with Travis Osterman, DO, MS, FAMIA, FASCO, associate vice president for research informatics, and director, Cancer Clinical Informatics, Vanderbilt University Medical Center (VUMC), he discusses advancing the smart use of AI in cancer care and how can health care professionals and the tech world can collaborate to build trust in AI-driven cancer care without compromising the patient-provider relationship.

You can revisit part 1 of our interview here, and to learn more about VUMC’s vision for cancer care innovation, please read the article recapping our October Institute for Value-Based Medicine^® event.

This transcript has been lightly edited for clarity.

Transcript

What are some of the most impactful applications of AI you've seen in oncology?

A big trend, not just in oncology, but nationwide is something called ambient scribe. Ambient scribe is a technology that's being implemented by a number of companies and certainly is investigated by researchers nationwide, and probably worldwide.

The concept goes like this: What if we had a microphone in the system in the room where an oncologist is talking to the patient and the oncologist could turn and actually talk to the patient and have that natural conversation? The microphone is smart enough to understand who is saying what, can capture that conversation, and then can draft what would be a note in the style that you normally write your clinical notes. That way, the oncologist isn't tied to the computer. They're able to have a more humanistic interaction with their patient while still getting their work done.

For me, I think that's probably one of the biggest opportunities. You'll hear a lot of things in the treatment and diagnosis space, and I think there are really great opportunities there. But the opportunity, ironically, for artificial intelligence to bring a more humanistic touch to how we practice oncology, I think is probably one of the biggest opportunities that we have.

How can the health care and the tech worlds best collaborate to build trust in AI-driven cancer care?

I'm a strong proponent that none of these decisions should be made in a vacuum. We need all the stakeholders at the table. Importantly, that includes patients and our communities. Sometimes researchers ask too late whether or not we should when they're trying to figure out whether we can, and they're pushing the limits of technology forward. But we need to ask those questions very early on.

What is the acceptance rate of having these things recorded in the room? Our patients, will they feel comfortable having the same conversations, for instance, in a room with the ambient scribe? How will they react to these new data points that are intended specifically for them in terms of predictive analytics and these new algorithms? I think these are things that are evolving and that we need to figure out.

On the project where we're predicting the outcomes and toxicities to immunotherapy, one of the biggest challenges that I continue to wrestle with is, what's the best way to display these data so that patients and their clinicians can have a more informed conversation? The last thing I want to do is make a cold recommendation on treatment, yes or no. What I want to do is provide the information in a way that the oncologist and patients can look at the screen together, understand that risk, understand that benefit, and then make a precise decision that takes into account that patient's value system, the kind of life they want to live, their risk tolerance, with all the information necessary. I think sometimes we jump to AI giving solutions, but I hope that the first step is AI surfacing the right information in a way that we can all understand it so we can just make better decisions.

Everyone likes to do the studies of AI vs radiologists or AI vs pathologists. I don't think we're even close to any of those things. Now, are we close to AI being a sidekick? Are we close to it being the medical assistant or the trusted nurse or the resident or fellow alongside a seasoned attending? We will get closer, and we'll get closer faster, I think, than many of us think, but I don't think it's going to replace health care delivery anytime soon.

How do care teams incorporate clinical informatics data into their treatment plans?

I really like the way Monica Bertagnolli, MD, who's our director of the NIH [National Institutes of Health], she was the ASCO [American Society of Clinical Oncology] president in 2019, and her presidential theme in 2019 for our professional society was “Learning From Every Patient,” and that's always just really stuck with me. We should learn from every patient encounter that we have as clinicians. I think that rings true. You want to take that experience and learn. It's a medical practice. I'm trying to get better every day, and I hope I can deliver better quality care every single day in my practice.

But we should be doing that not just at the individual clinician level, we need to do that at the entire medical practice level, at the system level, and in order to do that, we need to put systems in place to leverage all of these data to try and move the entire field forward. I think we owe that to our patients.

We work very, very closely with our privacy office and compliance office here [at VUMC], so when you're talking about training these models, you need to be in a HIPAA [Health Insurance Portability and Accountability Act]–compliant environment. That's a combination of both the technology and then the contracts to make sure that we have agreements in place between all the entities involved. The project that I'm a part of, with GE Health Care, that I mentioned, this is using entirely deidentified data.

We're much more interested in the outcomes, and there's not really a reason to have individually identified information in that data set, and so we've followed standard deidentification protocols and standards in order to do that. This is something that I'll say we're really lucky at Vanderbilt. We have experts like Brad Malin, PhD, here, who's one of my colleagues in the Department of Biomedical Informatics, who's an internationally known researcher in the privacy space. Our chief privacy officer, Sandra Hornsby, MSEd, CHPC, is just an incredible woman to work with and incredibly talented in the privacy space, and so we are super well supported to be able to keep these concerns really at the forethought of all the work that we do.