Cancer · AI doctors and chatbots
Answering patients' questions with AI before a surgery chat cut their anxiety in a trial
In a randomised trial at a Shanghai cancer centre, 268 men about to have prostate cancer surgery wrote down their questions. Half got AI-written answers before meeting the surgeon. Their anxiety score afterwards was 3.2 vs 5.7 (lower is calmer), and the meeting took 11.3 minutes instead of 19.9. Rating: In clinical trials.
How far away is this?
It's a randomised trial with real patients, so it's In clinical trials. But it was one Shanghai hospital, one cancer and a phase II trial. Patients were assigned by hospital room, not one by one, and they knew which group they were in. Anxiety was only measured straight after the talk, so a bigger trial elsewhere would need to confirm it.
Get the weekly How Far
One email every Friday morning (Sydney time) on which medical AI actually reached patients, and what it means for you. Free, in plain English.
Where is it approved?
| Region | Regulator | Status | Notes |
|---|---|---|---|
| United States | FDA | Research only | Reported as research; register not searched. |
| European Union | CE mark / EMA | Research only | Reported as research; register not searched. |
| United Kingdom | MHRA | Research only | Reported as research; register not searched. |
| Australia | TGA | Research only | Reported as research; register not searched. |
A hospital research system, not a product on sale. "Register not searched yet" means we haven't searched that regulator's register; it doesn't mean the product is or isn't available. Approval in one country isn't approval in another.
What is it?
Before an operation, patients have a lot of questions, and surgeons have limited time. Here a large language model (the kind of AI behind chatbots), run on the hospital's own computers, wrote personal answers to each patient's questions. Patients read them, then had their usual face-to-face talk with the surgeon, who didn't know which group they were in.
What's the evidence?
A prospective, randomised, single-blinded phase II trial (registered as NCT07082049) randomised 268 newly diagnosed prostate cancer patients due for radical prostatectomy: 130 to standard communication and 138 to AI-assisted communication. Patients asked 3,507 questions, 13.1 each on average. Patients were randomised by room rather than individually; only the surgeons were blinded. The trial was publicly funded and the authors declare no competing interests.[1]
Evidence stage: In clinical trials · #c1
After the consultation, anxiety on the GAD-7 scale (0 to 21, higher means more anxious) averaged 3.2 in the AI group vs 5.7 in the standard group. The AI group also reported better mood, understanding of their illness, and satisfaction.[1]
Evidence stage: In clinical trials · #c2
Doctors' workload score (NASA-TLX, higher means more strain) was 39.9 vs 56.8, and the face-to-face talk took 11.3 minutes vs 19.9 minutes, about 43% shorter.[1]
Evidence stage: In clinical trials · #c3
What does it mean for me?
If you're facing surgery, writing your questions down and getting clear answers before you see the surgeon may leave you calmer and make the meeting quicker. This was a hospital-run AI checked inside a trial, not a public chatbot. Your surgeon is still the one to confirm what applies to you.
What could I ask my doctor?
- Can I send my questions before our appointment?
- Who checks any written answers I'm given before I read them?
These are general starting points, not a recommendation to choose any particular test or treatment.
Short videos, under a minute
We make short, plain-English videos about AI in medicine, including what it can't do yet. Follow How Far for the next one.
Change history
- : First published.
General information only, not medical advice. Talk to your doctor about your situation. AI-assisted source check. Not clinically reviewed.
References
- npj Digital Medicine (2026): LLM-assisted preoperative communication in prostate cancer, randomised phase II trial. nature.com (2026). Peer-reviewed randomised trial. https://www.nature.com/articles/s41746-026-03309-7
Publisher shown as the website hosting each source. Dates are given where the source label includes them.