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

Summary

pyAKI standardizes KDIGO-based AKI classification from ICU time-series using a reproducible data model and dual-input algorithm (serum creatinine and urine output). Validated against expert annotations on MIMIC-IV, it achieved perfect accuracy across categories, offering the first open-source, end-to-end solution for consistent AKI labeling.

Key Findings

• First open-source, standardized pipeline implementing KDIGO AKI criteria on ICU time-series.
• Validated on MIMIC-IV subset with accuracy of 1.0 versus expert annotations across all categories.
• Implements a reproducible data model using serum creatinine and urine output.

Clinical Implications

While a research tool, pyAKI can underpin robust decision-support, audit, and quality-improvement pipelines in perioperative and critical care settings where AKI is common.

Limitations

• Validation on a subset of a single database (MIMIC-IV) may limit generalizability without multicenter testing
• Clinical impact on outcomes not assessed; tool performance in noisy real-time streams unproven

Future Directions

Benchmark across diverse EHRs and institutions, integrate into real-time clinical decision-support, and extend to AKI staging/trajectory prediction using probabilistic or ML approaches.