CLINICAL QUESTION: Should clinicians use artificial intelligence with natural language processing (AI/NLP) to assess patient messages and review patient concerns to further develop meaningful, novel patient-centered research topics?
BACKGROUND: Patient-centered research is crucial because it directly impacts their specific needs and health goals, which in turn can improve cancer care. Though patient portal messages are a valuable source of patients’ concerns, assessing this readily available data can be arduous. AI/NLP can be used to study this data to derive their viewpoints, but these outcomes still need to be assessed for meaningfulness and validity.
STUDY DESIGN: Retrospective case series
SETTING: Message threads from patients in Stanford Health Care and 22 affiliated centers from July 2013 to April 2024
SYNOPSIS: A total of 614,464 de-identified patient portal messages were used for the study, from a total of 25,549 patients, out of which 10,665 had breast cancer (98.6% female), and 14,884 had skin cancer (49.0% female). ChatGPT-4o [OpenAI] was used to summarize the patient concerns through these patient portal messages. Primary concerns within breast cancer patients were related to skin, urinary function, dental health, genetic testing, and liver, while skin cancer patients related to lesions on the nose, moles versus melanoma, issues with earlobes, management of surgical wounds, and side effects with 5-fluorouracil. AI also developed corresponding research ideas after searching for scientific articles related to these concerns. Oncologists and dermatologists further assessed these AI-generated research topics for meaningfulness and novelty.
Overall, mean (standard deviation) scores for meaningfulness and novelty were 3.00 (0.50) and 3.29 (0.74), respectively, for breast cancer topics and 2.67 (0.45) and 3.09 (0.68), respectively, for skin cancer topics. One-third of the AI-suggested research topics were highly meaningful and novel when both scores were lower than the average (5 of 15 for breast cancer and 6 of 15 for skin cancer). Two-thirds of the AI-suggested topics were novel (10 of 15 for breast cancer and 11 of 15 for skin cancer).
Limitations to this particular study included that only two specialties were investigated, breast and skin cancer, so generalizability is difficult to extrapolate to other types of cancer patients. In addition, certain data sets were excluded as the AI tool focused on specific concerns for this study and excluded others. Also, despite the large sample size, only experts from one single institution were involved, and this can result in bias.
BOTTOM LINE: AI can be used in the future to help guide and develop research topics that are patient-centered, given that they are priorities for patients and bring value to their care.
CITATION: Kim J, et al. Patient-centered research through artificial intelligence to identify priorities in cancer care. JAMA Oncol. 2025;11(6):630-635. doi: 10.1001/jamaoncol. 2025.0694.
Dr. Baro
Dr. Chitnis
Drs. Baro and Chitnis are clinical assistant professors in the division of hospital medicine at The Ohio State University Wexner Medical Center in Columbus, Ohio.