Machine learning algorithms' accuracy in predicting kidney disease progression: a systematic review and...

Area of Science:

• Nephrology
• Medical Informatics
• Data Science

Background:

• Kidney disease progression varies significantly among patients.
• Accurate prediction of kidney disease outcomes is vital for effective management.
• Existing Machine Learning (ML) models in nephrology show inconsistent accuracy.

Purpose of the Study:

• To systematically review and meta-analyze the diagnostic accuracy of ML algorithms for predicting kidney disease progression.
• To assess the reliability of ML models in forecasting kidney disease outcomes.

Main Methods:

• Searched multiple databases (PubMed, EMBASE, etc.) for relevant diagnostic studies up to October 2020.
• Evaluated study quality using the QUADAS-2 tool and extracted data.
• Synthesized data using bivariate and hierarchical summary receiver operating characteristic (HSROC) models.

Main Results:

• Six studies involving 12,534 patients were included, focusing on chronic kidney disease (CKD) and Immunoglobulin A Nephropathy.
• The overall HSROC Area Under Curve (AUC) was 0.87, with high specificity (0.87) but lower sensitivity (0.68).
• Subgroup analysis showed AUCs of 0.82 for CKD and 0.78 for IgA nephropathy, with varying sensitivity and specificity.

Conclusions:

• ML algorithms demonstrate high accuracy, particularly in specificity, for predicting kidney disease progression.
• ML models are recommended as auxiliary tools for clinicians in treatment and management decisions.
• Further research may refine ML model sensitivity for broader clinical application.