Synthetic Data-Driven Early Prediction Framework for Acute Kidney Injury in Patients Receiving Vancomycin and...

Area of Science:

• Pharmacology and Toxicology
• Nephrology
• Artificial Intelligence in Medicine

Background:

• The nephrotoxic risks of combining ceftazidime/avibactam (VAN-AVI) with vancomycin (VAN) are not well understood.
• Both VAN and ceftazidime/avibactam are independently associated with acute kidney injury (AKI).

Purpose of the Study:

• To assess the risk of AKI associated with concurrent VAN-AVI therapy.
• To develop synthetic data models for early AKI prediction in VAN-AVI treated patients.

Main Methods:

• Retrospective analysis of electronic health records (2015-2022).
• Comparison of AKI incidence in VAN-AVI, VAN-piperacillin/tazobactam (VAN-TPZ), and VAN monotherapy groups.
• Application of inverse probability of treatment weighting (IPTW) and synthetic data generation (CTGAN, TVAE) to address sample imbalance.
• Augmentation of machine learning (ML) models with synthetic data for AKI prediction.

Main Results:

• VAN-AVI was associated with a higher risk of AKI (HR=3.47) compared to VAN alone, with a low incidence of de novo AKI (4.3%) but high recurrent AKI (71.7%).
• Synthetic data analyses (TVAE, CTGAN) supported the increased AKI risk associated with VAN-AVI.
• ML models augmented with synthetic data showed improved 30-day AKI prediction (F1-score=0.80).

Conclusions:

• A novel approach integrating IPTW and synthetic data generation was used for drug-associated AKI risk evaluation.
• Synthetic data frameworks show potential for scalable drug safety evaluations.
• Augmented ML models significantly improved early AKI prediction in VAN-AVI treated patients.