pyAKI-An open source solution to automated acute kidney injury classification

Area of Science:

• Critical Care Medicine
• Biomedical Informatics
• Nephrology

Background:

• Acute kidney injury (AKI) affects up to 50% of critically ill patients.
• Current methods for applying KDIGO criteria to time series data lack standardization, leading to resource-intensive, variable implementations.
• This variability can impact the quality and reproducibility of AKI research.

Purpose of the Study:

• Introduce pyAKI, an open-source pipeline for consistent KDIGO criteria implementation.
• Address the need for standardized, accessible tools in AKI research.
• Facilitate accurate AKI classification in time series data.

Main Methods:

• Developed and validated the pyAKI pipeline using the MIMIC-IV database.
• Constructed a standardized data model for reproducibility.
• Implemented KDIGO guidelines for AKI diagnosis using serum creatinine and urinary output data.
• Compared pyAKI's diagnostic accuracy against physician annotations.

Main Results:

• pyAKI demonstrated robust performance in implementing KDIGO criteria.
• Validation against expert annotations confirmed high accuracy.
• Comparative analysis showed pyAKI surpassed human label quality with 1.0 accuracy across all categories.

Conclusions:

• pyAKI is the first open-source solution for KDIGO criteria implementation in time series data.
• Provides a standardized model for consistent AKI classification in research.
• Offers a valuable, accurate tool for clinical research and decision support systems.