Artificial intelligence-assisted multidisciplinary therapy for a complex case of cholangitis with septic shock: a case report and simulated decision-making analysis.

BACKGROUND

Acute cholangitis secondary to choledocholithiasis can rapidly progress to septic shock, with multidisciplinary team (MDT) management serving as the cornerstone of its. The potential of artificial intelligence (AI) in simulating such complex clinical decision-making remains underexplored. This study aims to examine the central role of MDT in the management of complex biliary septic shock and to evaluate the feasibility and value of AI in simulating clinical decision-making processes.

CASE DESCRIPTION: A 52-year-old female patient was admitted with septic shock (blood pressure 91/45 mmHg, heart rate 156 bpm) secondary to choledocholithiasis. Concurrently with real-world MDT management, we employed two large language models (DeepSeek and ChatGPT-5) to simulate MDT decision-making using a standardized clinical prompt. Both AI models demonstrated high concordance with the human MDT on core principles (immediate decompression, broad-spectrum antibiotics) but diverged on specific strategies, favoring percutaneous transhepatic cholangial drainage (PTCD) over the human team's choice of endoscopic intervention. Following a real-world MDT discussion, emergency endoscopic retrograde cholangiopancreatography (ERCP) with endoscopic nasobiliary drainage (ENBD) was performed. Due to rising amylase (peak 486 U/L) suggesting possible ENBD-related pancreatic duct obstruction, the ENBD was exchanged for an endoscopic retrograde biliary drainage stent on day 6. Concurrently, antibiotic therapy was escalated to imipenem-cilastatin. With this comprehensive strategy, the patient stabilized.

CONCLUSIONS

MDT is pivotal in complex biliary septic shock. AI demonstrates potential to replicate core diagnostic and therapeutic logic, but current models lack deep perception of "clinical reality feasibility". AI is a promising decision support tool, particularly in resource-limited settings, but human contextual adaptation remains irreplaceable.