Original Title: A randomized controlled trial of a WeChat-based artificial intelligence agent for postoperative care in orthopedic patients
Journal: NPJ digital medicine
DOI: 10.1038/s41746-025-02269-8
Overview
This randomized controlled trial evaluated a GPT-4-powered artificial intelligence agent delivered via the WeChat platform to support postoperative recovery in orthopedic patients. The study included 261 participants, with 140 assigned to the AI-driven intervention and 121 to traditional physician-led communication. Effective postoperative management is often hindered by limited access to timely support and poor adherence to rehabilitation protocols. The AI system demonstrated a significantly faster response time of 0.5 ± 0.6 minutes compared to 358 ± 47.5 minutes in the doctor-led group (p < 0.05). At the 1-month and 3-month follow-up assessments, the AI group showed significantly better outcomes in knee function, physical health scores, and overall patient satisfaction (p < 0.05). For instance, at 1 month, the AI group achieved a mean function score of 57.69 ± 9.64 versus 54.72 ± 10.3 in the control group. While the AI achieved a slightly lower accuracy rate of 93.9% compared to 98.1% for human physicians, the perceived quality of its responses was rated higher at 8.4 ± 0.9 versus 7.2 ± 0.9. By the 6-month mark, differences between the groups were no longer statistically significant, indicating that the AI agent provides specific benefits for accelerated early recovery.
Novelty
The study introduces a specialized application of large language models by integrating GPT-4 with a localized, expert-validated medical knowledge base using Retrieval-Augmented Generation. Unlike previous studies that developed standalone mobile applications, this intervention utilized WeChat, a ubiquitous social media platform, to maximize accessibility and user engagement. The research provides evidence through a rigorous prospective randomized controlled trial, comparing an autonomous AI agent directly against standard human-led postoperative care. Furthermore, the implementation included a two-tier auditing protocol where orthopedic surgeons performed daily and monthly reviews of interactions to identify hallucinations, which occurred at a rate of 6.3%. This structured approach to safety and reliability in a real-world clinical setting represents a progression in digital health research.
Potential Clinical / Research Applications
This technology offers a scalable framework for postoperative monitoring in various surgical specialties beyond orthopedics, such as general surgery or oncology, where timely patient education is essential. It can be utilized to standardize discharge instructions and rehabilitation guidance across large hospital systems, ensuring all patients receive consistent, evidence-based information. In research settings, the data generated from these interactions provide a rich source for analyzing patient concerns and identifying common recovery hurdles in real-time. Future iterations could incorporate multimodal capabilities, allowing the agent to analyze patient-submitted photographs of surgical wounds to detect early signs of infection. This model also serves as a supplementary tool in resource-limited environments where the patient-to-doctor ratio is high, providing a first line of support while reserving human expertise for complex clinical decision-making.
Leave a Reply