Machine Learning-Guided Fluid Resuscitation for Acute Pancreatitis Improves Outcomes

Area of Science:

• Artificial Intelligence in Medicine
• Critical Care Medicine
• Gastroenterology

Background:

• Ariel Dynamic Acute Pancreatitis Tracker (ADAPT) is an AI tool for predicting pancreatitis severity and guiding fluid resuscitation.
• Current fluid management strategies may not be optimal for individual patient needs.

Purpose of the Study:

• To assess if adherence to ADAPT fluid recommendations impacts clinical outcomes compared to standard management.
• To evaluate the effectiveness of AI-guided fluid resuscitation in acute pancreatitis patients.

Main Methods:

• Prospective cohort study of 1,083 patients admitted between June 2015 and November 2022.
• Compared actual fluid resuscitation within 24 hours to ADAPT recommendations.
• Primary outcome: organ failure in over-resuscitated vs. adequately resuscitated patients.

Main Results:

• Over-resuscitation was linked to increased respiratory failure (OR 2.73), ICU admission (OR 2.40), and local complications (OR 2.93).
• Adherence to ADAPT recommendations was associated with reduced adverse events.
• 1,083 patients analyzed; 700 were over-resuscitated, 196 adequately, and 187 under-resuscitated.

Conclusions:

• Adherence to ADAPT fluid recommendations improves outcomes in acute pancreatitis, reducing respiratory failure and other complications.
• Dynamic machine learning tools like ADAPT show promise for optimizing acute pancreatitis management.
• AI-guided fluid management offers a potential advancement over conventional strategies.