Preparing patients’ questions before a surgical consultation may make the conversation easier for both sides. In a randomised prostate-cancer trial, personalised answers generated by a locally deployed language model were followed by lower patient anxiety scores and shorter physician consultations than standard communication.[1][2]
In the trial, published in npj Digital Medicine, 268 newly diagnosed men scheduled for radical prostatectomy were allocated by hospital room to standard communication or to an AI-assisted preparation process. There were 130 in the control group and 138 in the intervention group. Researchers collected 3,507 patient questions, an average of about 13 per participant.[1][2]
The locally deployed large language model generated personalised responses in advance of regular physician-led, face-to-face preoperative communication. This was not a programme that left patients alone with a chatbot or substituted a bot’s decision for surgical consent. The doctor still spoke to the patient; the AI altered what the patient brought to that conversation.[1][2]
Measured benefits before surgery
After the consultation, mean GAD-7 anxiety scores were 3.2 in the AI-assisted group and 5.7 in controls. Lower scores indicate fewer self-reported anxiety symptoms. Physician workload and conversation time also fell, giving the trial a concrete result for patients and for the clinicians preparing them for surgery.[1][2]
Reported outcomes after preoperative communication
| Outcome | AI-assisted preparation | Standard communication |
|---|---|---|
| Patients randomised | 138 | 130 |
| Mean GAD-7 anxiety score | 3.2 | 5.7 |
| Mean NASA-TLX workload score | 39.9 | 56.8 |
| Mean consultation time | 11.3 minutes | 19.9 minutes |
The 8.6-minute difference in mean consultation time is calculated from the reported group averages. It describes the face-to-face conversation, not the total staff time or cost of operating the AI service. A service-level evaluation would need to include preparing, checking and delivering the answers.[1][2]
What the trial design establishes
The design was prospective, randomised and described as single-blinded. Rooms, rather than individual patients independently, determined allocation. Participants knew which preparation they received, which matters for subjective outcomes such as anxiety and satisfaction. The measured result remains a benefit within this particular clinical workflow.[1][2]
The outcome depended on the entire preparation process, including question collection and the AI-generated information patients received. Replication should assess the accuracy and readability of those answers alongside anxiety and workload.[1][2]
Preparation before the appointment
The intervention moved some explanation ahead of the appointment. Patients could absorb unfamiliar terms, revisit an answer and identify what they still wanted to ask. That offers a plausible reason for a more focused consultation, although the trial does not isolate which part of the preparation produced the improvement.[1][2]
The result points to a practical use for medical language models: helping patients arrive at an important conversation better prepared. The next test is whether other hospitals can preserve that benefit while accounting for answer review, different literacy levels and the full workload of the service.[1][2]
Sources & further reading
- Large language model–assisted preoperative communication reduces patient anxiety and physician workload in prostate cancer: a prospective randomized phase II trial npj Digital Medicine · 26 Sep 2026 · Accessed 26 Oct 2026
- AI-assisted preoperative communication in prostate cancer: A prospective, single-blinded, randomized phase II trial Journal of Clinical Oncology / ASCO · 27 May 2026 · Accessed 24 Oct 2026