Vancomycin levels for Bayesian dose-optimization in critical care: a prospective cohort study

Abstract

Background: Vancomycin dosing in critically ill patients typically requires monitoring the area under the concentration-time curve/minimum inhibitory concentration (AUC/MIC), often using at least two vancomycin levels (VLs). However, the optimal number of VLs needed for accurate AUC/MIC estimation in this population remains uncertain. This study aimed to determine the minimum number of VLs required to accurately estimate the AUC/MIC in critically ill patients treated with intermittent infusion of vancomycin. Methods: A prospective cohort study was conducted in critically ill patients, where VLs were obtained at peak, beta, and trough phases. Five AUC estimates were derived using PrecisePK™, a Bayesian software: AUC-1 [peak, beta (2 h after the end infusion), trough], AUC-2 (beta, trough), AUC-3 (peak, trough), AUC-4 (trough), and AUC-5 (only Bayesian prior, without VL). These estimates were compared for accuracy and bias (mean ± SEM) against the reference AUC calculated via the trapezoidal model (AUCRef). Results: We enrolled 36 adult patients with age of 65 (52–77) years, moderate severity [APACHE II 10 (5–14) and SOFA 5 (4–6)], 6 of them in ECMO and 4 in renal replacement therapy. A total of 108 blood samples for VL were analyzed. The AUC-3 (0.976 ± 0.012) showed greater accuracy compared to AUC-4 (1.072 ± 0.032, p = 0.042) and AUC-5 (1.150 ± 0.071, p = 0.042). AUC-3 also demonstrated lower bias (0.053 ± 0.009) than AUC-4 (0.134 ± 0.026, p = 0.036) and AUC-5 (0.270 ± 0.060, p = 0.003). Bland–Altman analysis indicated better agreement between AUC-3 and AUC-2 with AUCRef. Conclusion: Bayesian software using two vancomycin levels provides a more accurate and less biased AUC/MIC estimation in critically ill patients.