Machine learning-based prediction of a high-risk kidney function trajectory class after acute kidney injury

Area of Science:

• Nephrology
• Data Science
• Public Health

Background:

• Monitoring kidney function post-acute kidney injury (AKI) is crucial for predicting long-term outcomes.
• Identifying patients at high risk for progressive kidney disease is essential for timely intervention.

Purpose of the Study:

• To develop and validate a machine learning (ML) model to predict high-risk estimated glomerular filtration rate (eGFR) trajectories after AKI hospitalization.
• To identify patients likely to progress towards kidney replacement therapy (KRT).

Main Methods:

• A cohort study included 88,632 adults with at least two post-discharge eGFR measurements.
• Joint latent class mixed models identified distinct eGFR trajectories.
• Extreme gradient boosting (XGBoost) and random forest models were trained using clinical data within 3 months post-discharge.

Main Results:

• Three eGFR trajectories were identified: Steady Low (high-risk), Slow Decline, and Recovery.
• The 'Steady Low' trajectory was strongly associated with subsequent KRT initiation.
• A 10-variable XGBoost model achieved high discrimination (AUROC 0.974), with the eGFR slope from discharge to 3 months being the most influential predictor.

Conclusions:

• The ML framework effectively stratifies patients at risk for rapid kidney function decline post-AKI using routine clinical data.
• This model offers a practical approach for personalized post-AKI care and timely kidney-protective interventions.
• Further validation and benchmarking are recommended.