Personalized Fluid Management in Patients With Sepsis and Acute Kidney Injury: A Causal Machine Learning Approach.

Area of Science:

• Critical Care Medicine
• Nephrology
• Machine Learning in Healthcare

Background:

• Intravenous (IV) fluids are crucial for managing sepsis-induced acute kidney injury (AKI), but fluid overload poses a significant risk.
• Identifying patients who benefit from a restrictive fluid strategy is challenging.
• Causal machine learning (ML) offers a novel approach to estimate heterogeneous treatment effects (HTEs) of IV fluids.

Purpose of the Study:

• To develop and validate a causal-ML framework for identifying patients with AKI and sepsis who benefit from restrictive fluid administration (< 500 mL within 24 hours).

Main Methods:

• Retrospective study of patients with sepsis and AKI within 48 hours of ICU admission.
• Developed and validated a causal-ML model to estimate individualized treatment effects for fluid therapy.
• Compared model performance against a random forest model using area under the targeting operator characteristic curve (AUTOC).

Main Results:

• The causal forest model demonstrated superior performance in identifying HTE of restrictive IV fluids compared to random forest (AUTOC 0.15 vs. -0.02 in external validation).
• The model recommended restrictive fluids for 68.9% of patients in the external validation cohort.
• Patients receiving restrictive fluids showed significantly higher early AKI reversal (53.9% vs. 33.2%), sustained AKI reversal (34.2% vs. 18.0%), and lower MAKE30 rates (17.1% vs. 34.6%).

Conclusions:

• The developed causal-ML framework effectively identifies sepsis patients with AKI who benefit from restrictive fluid therapy.
• This data-driven approach enables personalized fluid management.
• Prospective clinical trials are warranted to evaluate this framework further.