Reinforcement learning based automated anesthesia system for gastrointestinal endoscopy with a multicenter randomized...

Area of Science:

• Anesthesiology and Artificial Intelligence
• Medical Device Technology
• Clinical Trial Research

Background:

• Growing demand for gastrointestinal endoscopy and anesthesiologist shortages necessitate automated anesthesia solutions.
• Reinforcement learning (RL) shows promise for autonomous anesthesia control, but clinical data is limited.
• Development of an RL-based automated anesthesia system for gastrointestinal endoscopy (AAS-GE) for ciprofol delivery.

Purpose of the Study:

• To evaluate the safety and efficacy of the AAS-GE compared to clinician-managed anesthesia.
• To assess primary outcome of hypoxemia incidence and secondary outcomes including induction time, drug use, recovery, and adverse events.

Main Methods:

• Prospective, multicenter, randomized controlled trial involving 418 adult patients (ASA physical status I-II) undergoing gastrointestinal endoscopy.
• Patients were randomized 1:1 to receive anesthesia managed by AAS-GE or by clinicians.
• Primary outcome: incidence of hypoxemia (SpO2 < 92%); secondary outcomes: hypoxemia severity, induction time, drug use, recovery, adverse events.

Main Results:

• Hypoxemia incidence was similar between AAS-GE and clinician groups (14.42% vs. 14.29%, P=0.968).
• AAS-GE demonstrated significantly shorter induction times (1.55 vs. 1.90 min, P<0.001) without compromising drug dose or recovery.
• Increased intraoperative body movement in the AAS-GE group indicated lighter anesthesia depth, with no significant differences in other safety outcomes.

Conclusions:

• AAS-GE is non-inferior in safety and efficacy to clinician-managed anesthesia for gastrointestinal endoscopy.
• The system offers potential for improved efficiency and standardized sedation care.
• Further research may explore optimizing anesthesia depth control for AAS-GE.