LLM recognition and triage of postoperative complications after aesthetic surgery

Source article: Comparative Evaluation of Large Language Models in the Recognition of Postoperative Complications Following Aesthetic Surgery

Background: Large language models (LLMs) are increasingly being explored as supportive tools in postoperative patient communication. However, their ability to recognize and appropriately manage postoperative surgical complications remains unclear. This study comparatively evaluated ChatGPT, Gemini, and Claude in postoperative complication assessment following aesthetic surgical procedures. Methods: Thirty standardized postoperative clinical scenarios representing low-, moderate-, and high-risk complications were…

By the Truvace deskOctober 15, 20262 min read1 sourcePeer-reviewedTRV-2026-1453· cite & verify

Comparative Evaluation of Large Language Models in the Recognition of Postoperative Complications Following Aesthetic Surgery — "Chatbots linked" by Ashhstarr is licensed under CC BY-SA 4.0. To view a copy of this license, visit https://creativecommons.org/licenses/by-sa/4.0/.

Trace impact readingNegative state

P 73

Both sides are scored from claims and sources, not community votes.

G 68

In brief

Researchers gave 30 standardized postoperative scenarios, covering low-, moderate- and high-risk complications after aesthetic surgery, to ChatGPT, Gemini and Claude using one standard prompt. Responses were rated on 5-point scales for diagnosis, management, triage, patient safety and completeness, and critical errors and dangerous reassurance were logged.

The findings matter because patients and clinics are exploring chatbots for postoperative communication, where missing an urgent complication could cause harm. The tools did reasonably well on common problems, but their uneven results in urgent cases mean they remain informational aids rather than substitutes for physician judgment. The text does not say how they would perform with real patients.

Main points

ChatGPT had the highest overall performance across most domains, with mean diagnostic accuracy of 3.99 versus 3.90 for Claude and 3.61 for Gemini (p < 0.001).
In high-risk scenarios, appropriate emergency referral occurred in 93.3% of ChatGPT, 76.7% of Gemini and 96.7% of Claude evaluations (p = 0.018).
Critical management errors were uncommon overall but occurred most often with Gemini.
The authors conclude LLMs may support postoperative care but should not replace physician assessment in potentially serious complications.

The gain

In 30 standardized aesthetic-surgery scenarios, three LLMs showed moderate-to-high ability to recognize common postoperative complications, with diagnostic accuracy means near 3.6 to 4.0 on a 5-point scale.

The problem

Performance varied in clinically urgent scenarios: Gemini recommended appropriate emergency referral in only 76.7% of high-risk evaluations, and it had the most critical management errors.

The rundown

The study compared three widely used LLMs on the same set of 30 standardized clinical scenarios, so each system faced identical inputs and an identical prompt. The scenarios were deliberately spread across low-, moderate- and high-risk complications to test whether performance held as stakes rose.

Evaluation used five dimensions: diagnostic accuracy, management appropriateness, triage accuracy, patient safety and completeness. Critical management errors and dangerous reassurance were tracked separately, capturing failures that average scores can obscure.

Claude had the highest emergency referral rate in high-risk cases, while ChatGPT led on overall performance and diagnostic accuracy. Gemini trailed on both measures, showing that the tools are not interchangeable for urgent triage.

What this doesn’t fix

The study used 30 standardized scenarios rather than real patient interactions, and the authors note variability became more apparent in clinically urgent cases, so the tools cannot replace physician assessment.