'Can AI Use Patient Messages to Identify New Cancer Care Research?'

 Hello. I'm Dr Maurie Markman, from City of Hope. I'd like to briefly discuss a very interesting topic, one that's discussed a great deal, but this is somewhat of a different theme of that topic. The paper is entitled, “Patient-Centered Research Through Artificial Intelligence to Identify Priorities in Cancer Care,” published in JAMA Oncology.

There's nothing new under the sun these days as it relates to the question of artificial intelligence (AI). Obviously, AI has been discussed extensively in the lay literature, in the technology literature, in the medical literature in general, and in the oncology literature in particular. There have been many uses proposed, including very prominently in imaging strategies and helping radiologists, as well as in pathology review — looking at a variety of gene markers, biomarkers, et cetera, et cetera. 

This was an interesting study of a novel use, I would say, of AI. The study examined de-identified patient portal messages, obviously, to the systems of the doctors, in an effort to develop patient-centered research questions. Patients were asking questions of their doctors and of their healthcare team, and AI looked at all these. Again, we're putting a human hat on a technology, but AI looked at these and said, can we come up with some questions that seem to be relevant to patients that we might pose as a research strategy? 

These investigators looked at 614,464 messages from 25,549 patients, including 10,665 with breast cancer and 14,884 with skin cancer. Three medical oncologists and three dermatologists independently evaluated what the AI came up with in two categories. One [was] meaningfulness — was it, "This is a meaningful question," but also for the novelty of the question, whether it was meaningful or not. "Was it a novel question? Was it clinically meaningful or not?"

Overall, these investigators found that one third of the AI-suggested research topics were meaningful and novel, and overall, two thirds were novel. Again, obviously, this is not surprising. Some of the questions may have been novel, but they weren't necessarily clinically meaningful. Being more conservative, one out of three of these individuals said, "These are meaningful and novel questions." Several of the examples listed in the paper, I think, support that conclusion. 

Here are just a couple of examples to emphasize the interest here: [the study of] an interdisciplinary approach to manage dental care in breast care; the development and testing of a specialized skincare regimen for patients with breast cancer; a study to examine efficacy of preventive dental care protocols for patients with breast cancer; the development and evaluation of a patient-centered digital tool for postsurgical wound care; and finally, of great interest, a longitudinal study on patient anxiety and decision-making in mole surveillance. 

Again, this was very provocative. These suggestions came from an AI analysis of patient portal messages — very interesting approach. I would personally be delighted to see other groups look at this and see what else AI can do on this front, and then consider potentially moving these interesting ideas into the actual research realm to address the questions.

https://www.medscape.com/viewarticle/can-ai-use-patient-messages-identify-new-cancer-care-2025a1000wak