Impact of Sampling Strategy and Population Model on Bayesian Estimates of Vancomycin AUC in Patients with BMI > 40 kg/m2: A Single-Center Retrospective Study.

Abstract

Background/Objectives: Growing evidence supports the use of a single trough concentration, rather than both a peak and trough, to estimate the 24 h area under the curve (AUC₂₄) of vancomycin using Bayesian software (InsightRx^® Ver.1.71). However, patients with body mass index (BMI) ≥ 40 kg/m² are underrepresented in validation studies. Studies in patients with obesity have produced mixed results, potentially because of different population models used. Methods: This single-center, retrospective study evaluated adult inpatients with BMI ≥ 40 kg/m². Steady-state AUC₂₄ estimates generated by Bayesian software using both two-concentration and one-concentration inputs were compared. Agreement was defined as a percent difference within ±20%. Subgroup analyses were conducted for patients with defined peak and trough concentrations and for comparisons between two Bayesian population models (Carreno vs. Hughes). Linear regression assessed covariates associated with percent difference. Results: Among 82 encounters, 97.5% of one-concentration estimates based on the smaller concentration were within ±20% of the two-concentration AUC_24,SS (mean difference: 2.9%, 95% CI: 0.14 to 3.8%). Similar agreement was observed using the larger concentration (97.5%, mean difference: -3.1%, 95% CI: -4.7 to -0.1.5%). Subgroup analysis for encounters with true peak/trough levels (n = 22) also showed 100% agreement within ±20%. The percent difference did not correlate with BMI or other covariates. Comparison of Hughes vs. Carreno models showed larger variability (only 59.1% within ±20%). Conclusions: In patients with BMI ≥ 40 kg/m², Bayesian AUC_24,SS estimation using a single vancomycin concentration is feasible. Greater caution is warranted in the setting of acute kidney injury, poor model fit, or targeting AUC at the extremes of the therapeutic range. The population model used to generate the Bayesian AUC estimate has a much greater influence than the number of concentrations analyzed. Furthermore, measuring two concentrations does not ensure concordance between models.