“We incorporate nurses and clinicians and users for any tool from the very beginning. They say, ‘You know, we need help with this.’ And then we start ideation: We start understanding the problem, we meet with them, we try to see what is it that they’re trying to do, is it feasible given the data we have? We go back, we do some research, feasibility study. We say we think this is something we can predict with decent performance. Now let’s do it,” Nasim Eftekhari, MS, executive director of applied artificial intelligence (AI) and data science at the City of Hope National Medical Center in Duarte, CA, told Lenise Taylor, MN, RN, AOCNS®, BMTCN®, oncology clinical specialist at ONS, during a discussion about how the use of AI in cancer care affects an oncology nurse’s daily work.
Music Credit: “Fireflies and Stardust” by Kevin MacLeod
Licensed under Creative Commons by Attribution 3.0
The advertising messages in this episode are brought to you by LUNGevity.
Episode Notes
To discuss the information in this episode with other oncology nurses, visit the ONS Communities.
To find resources for creating an ONS Podcast Club in your chapter or nursing community, visit the ONS Podcast Library.
To provide feedback or otherwise reach ONS about the podcast, email pubONSVoice@ons.org.
Highlights From Today’s Episode
“So, there is a lot of applications of AI in cancer care, so I can't possibly give you an exhaustive list. But the ones that come to my mind, at least the ones that we are actively working on are early detection and diagnosis, treatment planning, predictive modeling for predicting unwanted outcomes, remote monitoring, radiology applications, pathology applications, improving operations and helping the resource allocation, precision medicine, and research. And we also started a year or so incorporating AI and helping with drug discovery.” TS 2:13
“We’ve been using AI for a very, very long time. Recently, we just hear more about AI, but AI is in our lives, in health care or not, all day, every day. Google Maps, Google search, all of this is enabled by AI, but we may not realize even that we’re using it.” TS 8:27
“So, for technical challenges, you have to always consider: Is this model performing in a decent manner for this application? And depending on the use case, that’s different. If you’re providing a decision support to someone that is impacting patient care, then you have to be very careful about model performance. So, model performance is one technical consideration, then how do you really technically integrate with the EMR system? It’s not easy, EMR systems are not usually very open, and that’s a whole challenge in itself to be able to read from any EMR system in real time and feed data back into it in real time.” TS 10:16
“For nurses to successfully approach and adopt this work, I think the most important thing is to keep an open mind to really realize that these technologies can, at best, take the mundane part of their work away so they can operate at the top of their license, but what AI does best is to do things that are repetitive and doesn’t require a ton of human intelligence. I think that would be very helpful. Just that mindset could make things more collaborative and cooperative, and that’s the only way that we can make these successful.” TS 12:37
“What could help is for nurses to learn the basic concepts that are involved in the development and deployment and testing of these models, so that they can really understand the limitations and capabilities and they can take an active part in the development as well. So, it’s not like we build something for you and then we’re trying to convince you this is good for you. We try to build together. As an AI and computer scientist, I’m always learning the medical language. I try to educate myself about the clinicians’ workflows and language, and I think the same needs to happen on the clinician side for us to be able to build tools that really work in their workflows for their everyday life.” TS 13:58
“We incorporate nurses and clinicians and users for any tool that will be developed from the very beginning. So, usually, the need for something, like a predictive model, comes from nurses and doctors. They say, ‘You know, we need help with this.’ And then we start ideation: We start understanding the problem, we meet with them, we try to see what is it that they’re trying to do, and is it feasible given the data we have? We go back, we do some research, feasibility study. We come back and say we think this is something we can predict, you know, with decent performance. Now let’s do it.” TS 14:30
“All of our models, even the ones that have been in production for the longest, we’re still getting feedback, we’re still improving, and we’re still retraining models, not only with new data that becomes available but also with the feedback that we get from our users.” TS 17:43
“For example, after going live, we’ve had less ICU admissions because of sepsis or septic shock, or after going live had less sepsis mortality, which is very reassuring. So that seems like we’re doing the right thing, and our model is working, but if you want to put your scientist hat on, you cannot say 100% this is the impact of the model because there is a lot of different workstreams that are trying to improve those same metrics. And unless you do a clinical trial or what we call in industry A/B testing, where you control for everything else and it’s only the model intervention that is the variable, you cannot say for 100% that this is the impact of the model. That’s why we combine our qualitative metrics that seem to be right in the right direction with the quantitative metrics.” TS 22:17
“I think for the first time, something has come up that can really make a big change in health care. I could not say this before generative AI. AI has always been helpful, but now I think it’s the time to see real change. We’re still experimenting. It’s really new technology. We are experimenting with in-house development as well as third-party tools that we are testing and evaluating. Again, there’s a huge potential in reducing manual labor and documentation, note taking, there are implications in billing and finance, data abstraction for research or whatever other purposes that we need them, tumor boards, predictive modeling, clinical trial matching is one big use case in oncology, and finding similar patients—something that we’ve been aspiring to for a really long time—seems to be very possible now with these technologies.” TS 25:30
“The users also weigh in. So, if you’re considering it to improve clinical operations, the people who will be using the tool will have a say in, ‘Yes, we think this tool will be helpful.’ So, it’s not just looking at the technical and cybersecurity and ethical and legal aspects, but also is this something that our users will use because that’s the ultimate goal. If they don’t use it, it doesn't matter how good the tool is. It won’t work.” TS 31:13
“Making it successful is not about the technology, but mostly about people and processes and operational support.” TS 33:33
“Helping people, helping clinicians, nurses to be more free of mundane tasks and be able to interact with patients, do patient care, which is what they should be doing, rather than the things that I know a lot of nurses hate. I think we have a very exciting time ahead of us.” TS 38:47
Listen on Amazon Music, Listen on Apple Podcasts, Listen on Spotify, Listen on YouTube Music
On-the-go discussions covering a wide array of clinical and leadership topics that you can earn NCPD for.
View All Podcasts