Acute Kidney Injury Prognosis Prediction Using Machine Learning Methods: A Systematic Review.

Area of Science:

• Medical Informatics
• Nephrology
• Machine Learning

Background:

• Accurate prediction of in-hospital outcomes for acute kidney injury (AKI) is vital for clinical decision-making.
• Machine learning (ML) models offer potential for improving AKI prognosis prediction using administrative data.

Purpose of the Study:

• To systematically review ML-based prediction models for in-hospital AKI prognosis.
• To evaluate the performance of these models compared to traditional methods.

Main Methods:

• Systematic review adhering to PRISMA guidelines.
• Searched major databases (PubMed, Embase, Web of Science, Scopus, CINAHL) for studies from 2014-2024.
• Qualitative synthesis and meta-analysis of 27 eligible studies on ML models predicting AKI outcomes.

Main Results:

• ML models demonstrated superior performance in predicting AKI mortality (AUROC 0.831) versus traditional methods (AUROC 0.772).
• Key predictors for mortality included age, serum creatinine, and white blood cell count.
• Prediction of kidney function recovery was less accurate; common predictors included AKI stage and estimated glomerular filtration rate.

Conclusions:

• ML models show significant potential for improving clinical decisions and patient outcomes in AKI management.
• Further research is needed to enhance the generalizability and validation of these models across diverse hospital settings.