Using artificial intelligence to predict mortality in AKI patients: a systematic review/meta-analysis

Area of Science:

• Nephrology
• Medical Informatics
• Artificial Intelligence

Background:

• Acute kidney injury (AKI) significantly increases patient morbidity and mortality.
• Artificial intelligence (AI) and machine learning (ML) offer dynamic approaches for predicting mortality in AKI.
• Evaluating the performance of various ML models is crucial for improving AKI patient outcomes.

Purpose of the Study:

• To review and compare the performance of different machine learning models in predicting in-hospital mortality for patients with acute kidney injury.
• To identify the most effective AI-driven models for AKI mortality prediction.

Main Methods:

• A comprehensive literature search was conducted across PubMed, Embase, and Web of Science.
• Included studies focused on original research (cross-sectional, prospective, retrospective) evaluating AI model efficacy using metrics like AUC, accuracy, sensitivity, and specificity.
• Reviews and self-reported outcomes were excluded, with no restrictions on time or geography.

Main Results:

• Eight studies involving 37,032 AKI patients were analyzed.
• The Broad Learning System (BLS) and Elastic Net Final (ENF) models showed the highest pooled Area Under the Curve (AUC) for mortality prediction [0.852].
• The Proposed Clinical Model (PCM) had the lowest AUC [0.765] but the highest negative predictive value, suggesting its utility as a rule-out tool.

Conclusions:

• BLS and ENF models are effective for predicting in-hospital mortality in AKI patients, performing comparably to other ML models.
• Variability exists in the performance across different ML models.
• Further research is warranted to validate and refine these predictive models.