Comparative Evaluation of Large Language Models in the Recognition of Postoperative Complications Following Aesthetic...

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.

Thirty standardized postoperative clinical scenarios representing low-, moderate-, and high-risk complications were independently submitted to the three AI systems using a standardized prompt. Responses were evaluated for diagnostic accuracy, management appropriateness, triage accuracy, patient safety, and completeness using 5-point Likert scales. Critical management errors and dangerous reassurance were additionally recorded.

ChatGPT demonstrated the highest overall performance across most domains. Mean diagnostic accuracy scores were 3.99 ± 0.57 for ChatGPT, 3.61 ± 0.62 for Gemini, and 3.90 ± 0.62 for Claude (p < 0.001). Critical management errors were uncommon overall but occurred most frequently with Gemini. In high-risk scenarios, appropriate emergency referral recommendations were identified in 93.3% of ChatGPT evaluations, 76.7% of Gemini evaluations, and 96.7% of Claude evaluations (p = 0.018).

The evaluated AI systems demonstrated moderate-to-high capability in recognizing common postoperative complications; however, variability became more apparent in clinically urgent scenarios. Current LLMs may serve as supportive informational tools in postoperative care but should not replace physician assessment in potentially serious complications.

This journal requires that authors assign a level of evidence to each article. For a full description of these Evidence-Based Medicine ratings, please refer to the Table of Contents or the online Instructions to Authors www.springer.com/00266 .