Machine Learning Models for Predicting Mortality in Critically Ill Patients with Sepsis-Associated Acute Kidney...

Area of Science:

• Nephrology
• Critical Care Medicine
• Medical Informatics

Background:

• Sepsis-associated acute kidney injury (AKI) is a critical condition requiring accurate prognostic tools.
• Machine learning (ML) offers potential for improving patient stratification and intervention in sepsis-AKI.
• Equitable and unbiased ML models are essential for reliable clinical decision-making.

Purpose of the Study:

• To systematically review the literature on the effectiveness of ML algorithms for predicting mortality in sepsis-associated AKI.
• To assess the performance and fairness of existing ML models in this patient population.

Main Methods:

• An exhaustive literature search was conducted across PubMed, Scopus, and Web of Science (2000-2024).
• Studies predicting mortality in sepsis-AKI using ML were included; non-English and data-deficient studies were excluded.
• Data extraction and quality assessment were performed by two independent reviewers.

Main Results:

• Five studies were included, predominantly featuring male and White patient cohorts, with limited race/ethnicity data.
• ML models showed variable performance (AUROC 0.60-0.87), with XGBoost, RF, and LR demonstrating high accuracy.
• ML models show potential for early AKI progression prediction and survival improvement.

Conclusions:

• ML models demonstrate significant potential for predicting mortality and guiding management in sepsis-associated AKI.
• Existing models exhibit performance variability and a lack of fairness, potentially exacerbating healthcare disparities.
• Development of trustworthy and equitable ML models is crucial for clinical adoption and patient benefit.