Angela Elma Edwina1, Nada Dia1, Matthias Gijsen1,2, Lorenz Van der Linden1,2, Isabel Spriet1,2, Jos Tournoy1,2, Erwin Dreesen1
1KU Leuven, 2UZ Leuven
Introduction: Intravenous vancomycin is commonly used to treat severe gram-positive infections in older adults (=75 years). Yet, its narrow therapeutic window and high pharmacokinetic variability complicate its clinical use (1-4). Model-informed precision dosing (MIPD) may help guide vancomycin dosing. Conversely, the suitability of population pharmacokinetics (popPK) models for older adults has not been evaluated. Objectives: We aimed to evaluate the predictive performance of both single- and multi-model approaches (i.e., a model selection algorithm [MSA] and a model averaging algorithm [MAA]) using published popPK models to guide intravenous vancomycin dosing in older inpatients. Methods: We encoded nine published popPK models of vancomycin – developed to describe the pharmacokinetics in distinct populations – from literature in NONMEM 7.5 (5-13). We used prior data from 255 older inpatients with vancomycin dosing information for the 48 hours preceding therapeutic drug monitoring (TDM), with each patient contributing three consecutive TDM concentrations. We evaluated the a priori (solely based on covariates) and a posteriori (based on covariates and vancomycin concentrations) predictive performance (accuracy and precision) of single-model and multi-model approaches (MSA/MAA) visually and statistically. Three a posteriori prediction settings were compared using 1) the most recent TDM concentration (CT0), corresponding to the second measured concentration, 2) an earlier TDM concentration (CT-1), corresponding to the first measured concentration, and 3) both CT0 and CT–1, to predict the third measured vancomycin concentration (CT1). Accuracy was evaluated using the relative Bias (rBias), considered clinically acceptable if within ±20% with a 95% confidence interval including zero. Precision was assessed using the relative root mean squared error (rRMSE), where lower values indicate more precise predictions (14). The number of popPK models in the multi-model algorithms was reduced in a robustness analysis without compromising predictive performance. Agreement between predicted and measured CT1 values over time intervals from CT–1 or CT0 to CT1 was visually assessed in Bland-Altman analysis. The classification accuracy was also evaluated; it was defined as no change in exposure category, i.e., subtherapeutic (<12.5) mg/L, supratherapeutic (>17.5 mg/L), or therapeutic (12.5–17.5 mg/L) between the predicted and the measured vancomycin concentration. Results: The Goti et al., Ji et al., and Thomson et al. models were identified as the best-fitting models in terms of the normalized prediction distribution errors. A priori predictions varied largely (inaccuracy range -37%–44%, imprecision range 47%–77%). A posteriori prediction using CT0 (time between TDM sample and predicted occasion: median 36.0 hours, interquartile range 24.0–48.6 hours) provided the best predictive performance (inaccuracy range -18%–24%, imprecision range 25%–51%). Five candidate models were selected in the robustness analysis based on its overall predictive performance: the Thomson et al., Ji et al., Zhou et al., Adane et al., and Roberts et al. models. The MSA using these five models had the best predictive performance, outperforming the MAA and single-model algorithms (inaccuracy range -5%–6%, imprecision range 26%–42%). Its Bland-Altman plots showed an agreement between the predicted and the measured vancomycin concentrations in the a posteriori predictions over time intervals. It had also the highest classification accuracy (72%). Conclusions: We evaluated the predictive performance of both single- and multi-model approaches for MIPD of vancomycin in older adults using nine popPK models. The MSA using five models and the most recent TDM concentration had the best performance. This approach enhances precision dosing of vancomycin, potentially improving safety and efficacy in clinical practice.
[1] Buurman et al. PLoS ONE (2011) 6 (11), e26951. [2] Tacconelli et al. J Hosp Infect (2006) 64 (3), 251–6. [3] Liu et al. Clin Infect Dis (2011) 52 (3), 285–92. [4] Rybak et al. Am J Health Syst Pharm (2020) 77 (11), 835–64. [5] Adane et al. Pharmacotherapy (2015) 35 (2), 127–39. [6] Revilla et al. Br J Clin Pharmacol (2010) 70 (2), 201–12. [7] Thomson et al. J Antimicrob Chemother (2009) 63 (5), 1050–7. [8] Goti et al. Ther Drug Monit (2018) 40 (2), 212–21. [9] Medellin-Garibay et al. J Antimicrob Chemother (2016) 71 (2), 471–9. [10] Mangin et al. Clin Pharmacokinet (2014) 53 (9), 849–61. [11] Roberts et al. Antimicrob Agents Chemother (2011) 55 (6), 2704–9. [12] Ji et al. Acta Pharmacol Sin (2018) 39 (2), 286–93. [13] Zhou et al. Eur J Drug Metab Pharmacokinet (2019) 44 (3), 361–70. [14] Uster et al. Clin Pharmacol Ther (2021) 109 (1), 175–83
Reference: PAGE 33 (2025) Abstr 11407 [www.page-meeting.org/?abstract=11407]
Poster: Clinical Applications