Quantifying the Impact of Discrepancies Between Actual and Recorded Blood Sampling Times on Bayesian Forecasting for Area Under the Concentration–Time Curve Estimation of Vancomycin

Abstract

The effect of discrepancies in sampling time on the accuracy of area under the concentration–time curve (AUC) estimation remains unclear. We evaluated how discrepancies between actual and recorded blood sampling times impact vancomycin AUC estimation using one-point (trough) and two-point (trough and peak) strategies in critically ill patients. In total, 25 patients at five Japanese hospitals were enrolled; 21 underwent sampling on Day 1 (AUC_day1) and 14 on Days 3–5 (AUC_ss). Reference AUCs were calculated using trapezoidal methods. Bayesian forecasting estimated AUCs from one- or two-point sampling. Systematic trough (≤ 4 h earlier) and peak (−1 to +4 h) time discrepancies were introduced. Predictive performance was defined as estimated/reference AUC ratios within an acceptable range of 0.9–1.1. For AUC_day1, two-point sampling was more robust to trough discrepancies than one-point sampling (acceptable interquartile range [IQR] up to 105 vs. 30 min) but highly sensitive to peak timing (acceptable IQR −20 and +10 min). For AUC_ss, similar trends were observed. Overall, two-point sampling offers superior robustness against trough discrepancies. However, because peak sampling is highly sensitive to timing errors, if the exact peak sampling time cannot be verified, clinicians may consider remeasuring the concentration or switching to one-point trough sampling.