Abstract
Vancomycin is an antibiotic administered intravenously for severe invasive infections and due to its low therapeutic index, therapeutic drug monitoring is recommended. Clinical staff at the Canterbury District Health Board (CDHB) have been monitoring serum concentrations and targeting a Cmin of 15-20 mg/L. The 2020 guidelines from the Infectious Diseases Society of America (IDSA) recommend directly monitoring the AUC0-24h/MICBMD for Methicillin Resistant Staphylococcus Aureus due to: increased rates of target attainment, reduced overexposure and that this metric accounts for bacterial susceptibility compared to Cmin values. The value of AUC24 can be difficult to estimate, requiring several concentrations across a dose profile, using empirical estimates or specialised software. Clinical staff at the CDHB have access to a Bayesian forecasting software, TCIWorks with the Thomson et al. model implemented; to date there has been no evaluation of the accuracy of a limited sampling strategy combined with this solution.
This research had three aims: 1) To determine whether a limited vancomycin sampling strategy (1-2 samples per dosing interval, using Bayesian methods) can accurately predict the AUC0-24h at steady state (AUCss0-24h) of vancomycin in adults, 2) to determine whether accuracy in the AUCss0-24h predictions differs between obese and non-obese patients, and, 3) to investigate factors contributing to bias in the AUCss0-24h estimate. A simulation study was undertaken using demographic dosing data from patients administered vancomycin at the CDHB between 2016 and 2019 and nine sampling strategies where 1-2 simulated samples were included in the Bayesian forecasting per course, varying the day of the sample(s). Values of mean prediction error (mg.h/L) and AUC of the sample strategy versus a simulated reference AUC (AUCssTest: 0-24h/AUCssRef: 0-24h) were calculated for each sample strategy and the latter was compared to a bioequivalence range recommended for low therapeutic index medications (0.900-1.111). To estimate the accuracy of each sampling strategy, the proportion of values of AUCssTest: 0-24h within 20% of the AUCssRef: -24h was calculated (P20).
It was found that the sampling strategies were unable to estimate an unbiased result across the dataset; all strategies significantly underestimated the value of AUCssTest: 0-24h (mean prediction error: -70.52 (95% CI -97.37, -43.68) to -59.08 (95% CI -77.13, -41.02)) and only one strategy met the bioequivalence criteria; using the Cmin for doses 2 and 3. Stratification by obesity status found bias in obese individuals but not non-obese individuals. This was corroborated by the multilinear regression model which found that BMI was a significant predictor of the prediction error (coefficient -4.50 (95% CI -7.06, -1.94); p < 0.001).
This research provided useful information that any of the strategies implemented could estimate AUCss0-24h in non-obese individuals with sufficient accuracy. Further research into strategies for obese individuals is required.