Multisite Analysis of Vancomycin Concentration Predictions in a Bayesian Software in Adult and Pediatric Patients With and Without Acute Kidney Injury

BACKGROUND: Bayesian software programs are recommended for area under the concentration-time curve (AUC)-guided vancomycin monitoring, but hesitancy remains for patients with acute kidney injury (AKI). Here, we assessed agreement between population pharmacokinetic model-based Bayesian predictions and observed vancomycin concentrations in patients with and without AKI. METHODS: Data were retrospectively analyzed from patients at 162 institutions in the United States. Courses with ≥ 2 vancomycin doses, ≥ 1 vancomycin level, and ≥ 2 serum creatinine values were included. Observed drug levels served as reference values and were classified as occurring in the absence of AKI, during new-onset AKI, or during persistent AKI. Adult, pediatric, and neonate pharmacokinetic models were assessed both a priori and a posteriori. Performance was measured using root mean squared error, mean percentage error, and the proportion of predictions within ±20% (P20) and ±30% (P30) of observed values. We also evaluated flattened priors and last-observation-only weighting as alternative Bayesian estimation approaches and reported agreement between model-predicted and model-estimated AUCs (classified as in or out of target). RESULTS: There were 30,561 adults, 5223 pediatrics, and 1865 neonates included. Prediction performance declined with AKI relative to no AKI, particularly in pediatrics, followed by neonates, with adults showing minimal deterioration. Flattened priors often improved performance, whereas last-observation-only typically worsened it. Overall AKI reduced P20 by 0%-33% and P30 by 0%-19% (absolute percentage points). AUC classification agreement was similar across no AKI, new-onset AKI, and persistent AKI groups (73%-83%, 72%-82%, and 72%-84%, respectively, a posteriori). CONCLUSION: Model predictiveness of vancomycin concentrations was worse in patients with AKI, especially in pediatrics and neonates, and generally less pronounced in persistent versus new-onset AKI. Further studies with denser pharmacokinetic sampling and elucidation of clinical implications of Bayesian-guided vancomycin dosing in AKI are needed.