A Retrospective Within-Subjects Analysis of Vancomycin Bayesian Modeling With Pre-steady-State vs Steady-State...

Background: Bayesian modeling of vancomycin can estimate 24-hour area under the curve (AUC₂₄) using pre-steady-state concentrations. Limited literature exists comparing Bayesian AUC₂₄ calculations derived from steady-state versus pre-steady-state concentrations. Objective: To assess the agreement between vancomycin AUC₂₄ calculations using pre-steady-state versus steady-state concentrations, employing Bayesian modeling. Methods: This retrospective within-subjects cohort study included patients with at least 1 pre-steady-state and 1 steady-state vancomycin concentration. Patients with age >100 years, weight <40 kg, height <60 inches, or renal dysfunction were excluded. The steady-state AUC₂₄ from dosing software was documented with and without hiding steady-state levels from calculations. The primary outcome was agreement between AUC₂₄ without levels hidden compared with AUC₂₄ with steady-state levels hidden from analysis. Secondary outcomes included the agreement between AUC₂₄ with pre-steady-state levels hidden and the percentage of patients with matching AUC₂₄ categories. The AUC₂₄ agreement was evaluated via Bland-Altman plot and bias via linear regression. Statistical tests were performed using SPSS statistics software (IBM Corp). Results: A total of 93 patients were included. The mean difference in AUC₂₄ compared to AUC₂₄ with steady-state levels hidden was 8.8 mg*h/L, and with pre-steady-state levels hidden, it was -3.7 mg*h/L. Linear regression analysis indicated a proportional bias when steady-state levels were hidden (β = 0.22; P = 0.038) but not when pre-steady-state levels were hidden. Category mismatch occurred more often when steady-state levels were hidden vs when pre-steady-state levels were hidden (26% vs 8%; P < 0.001). Conclusion and Relevance: The study demonstrated overall agreement between AUC₂₄ compared to AUC₂₄ with steady-state levels hidden. The mean differences in AUC₂₄ estimates were small, no matter which level was hidden, although tighter limits of agreement were observed when steady-state levels were utilized in Bayesian calculations. Further research with larger sample sizes is necessary.