Heterogeneity of Nephrotoxic Risk Across Medication Classes and Combinations in Hospitalized Adults

Drug-associated acute kidney injury (DA-AKI) is common among hospitalized patients receiving multiple medications. Nephrotoxicity is often treated as a binary categorization, but AKI risk likely varies across medication classes and combinations. We sought to characterize class-specific AKI associations and evaluate whether medication-class interactions associate with DA-AKI prediction.

This retrospective observational study included adults admitted to the University of Iowa from 2014-2023 who met high nephrotoxin exposure criteria, defined as receipt of ≥3 concurrent potentially nephrotoxic medications. The primary outcome was stage 2/3 AKI within 7 days of initial high-nephrotoxin exposure. Models adjusted for demographics, comorbidities, admission diagnosis, laboratory values, and vital signs at index exposure. Generalized estimating equation logistic models were used to estimate medication-class associations and test 66 two-way class interactions. XGBoost models were used to explore nonlinear and higher-order exposure patterns.

We evaluated 66,977 admissions with high-nephrotoxin exposure from 42,558 patients, with stage 2/3 AKI developing in 3,309 (5%). Adjusted odds ratios ranged from 0.74 (95% CI 0.64-0.87) for NSAID exposure to 1.71 (95% CI 1.55-1.88) for piperacillin-tazobactam, indicating substantial heterogeneity in class-specific associations with AKI. Among 66 medication-class pairs tested, 9 (14%) remained statistically significant after multiple-comparison adjustment. Inclusion of medication class and pairwise terms was associated with increased GEE discrimination from ROC-AUC 0.73 to 0.76 and PR-AUC 0.16 to 0.17. XGBoost achieved ROC-AUC 0.78 and PR-AUC 0.20, suggesting additional nonlinear or higher-order exposure patterns.

Among hospitalized adults with high nephrotoxin exposure, medication class composition and pairwise exposure patterns were associated with subsequent stage 2/3 AKI. These findings support evaluation of weighted nephrotoxin surveillance frameworks incorporating medication class, combinations, and clinical context.