A patient recently came to see Brian Persing, M.D. with a list of questions about their lung cancer, every one of them written by AI. He answered them all, and he noticed that a few were tailored to that patient in ways even an attentive doctor might not think to raise. That story opens Dr. JYP's conversation with Dr. Persing, a hematologist and medical oncologist who has returned to the USA Health Mitchell Cancer Institute as its new director.

Most people hear "artificial intelligence" and picture a chatbot, but at MCI much of the work happens out of sight. Canopy Care checks in with patients on treatment by text or email and alerts the care team when a symptom can't wait. LeanTaaS studies the clinic schedule to shorten wait times, and other software follows lung nodules that turn up on scans taken for unrelated reasons, so a finding from a car accident doesn't slip through the cracks.

The two doctors are just as candid about the limits of artificial intelligence. Information generated by chatbots can be months out of date and sometimes even fully invented. Security measures dictate that patient information has to stay private, so what can be inputted has to omit any identifying information. For these reasons and more, no algorithm will ever replace the professionals in the exam room, Dr. Persing says.

Relevant Links

Brian E. Persing, M.D. | USA Health - Mitchell Cancer Institute | Medical Oncologist And Hematologist

Persing named director of USA Health Mitchell Cancer Institute

Submit your questions for the show at cancering.com

3 Key Listener Takeaways

1. Much of the AI in cancer care Is invisible

Patients on treatment at MCI receive a short weekly survey from Canopy Care by text or email. If an answer crosses a threshold, such as unexpected vomiting or new shortness of breath, the care team is alerted right away and the patient gets clear instructions on what to do next, even if they hadn't thought to call. Behind the front desk, LeanTaaS learns which appointments tend to run long and adjusts the schedule to match. Persing admits he is one of the clinicians who gives it reason to.

2. Speed is not the same as accuracy

Persing recently joined a panel that tested the NCCN and ASCO guidelines against Claude and ChatGPT. The clinicians loved getting answers in 15 to 30 seconds, but they didn't trust them, and the experts who reviewed the results found they weren't always right. Pierce has her own cautionary tale: a chatbot once sent her to a Paris café that doesn't exist. For medical questions, both doctors recommend vetted sources like UpToDate and OpenEvidence, and they encourage patients to lean on their own physician to help make sense of what they find online.

3. The human stays in the room

AI may help oncologists keep pace with the 15 to 20 new cancer drugs now approved each year, flag interactions buried in a long medication list, and turn an eighth-grade drug information sheet into one written at a first-grade level. Even so, when Canopy Care raises an alert, someone from the care team still picks up the phone to confirm it, and every new AI tool at USA Health goes before a special committee to make sure it is safe. When asked what he hopes AI will be doing in 10 years, Persing keeps it simple. He wants it to work so seamlessly that it fades into the background and gives clinicians more time face to face with their patients.

Show Overview

00:00 A patient arrives with questions written by AI

03:07 What AI really means in cancer care

05:22 When chatbots make things up

07:28 Turning medical jargon into plain language

08:43 How Canopy Care keeps tabs on patients between visits

11:01 Shorter waits through smarter scheduling

12:02 Ambient recording and protecting patient privacy

14:37 Catching cancer earlier

17:51 Moving from precision medicine to personalized medicine

21:15 The next five to 10 years, including virtual tumor boards

25:42 Why AI will never replace the human touch