S.D.T. Sassen, A. Ozturk, B.C. M. de Winter, B.C.P. Koch
Erasmus MC, The Netherlands
Objectives: Vancomycin is a commonly used antibiotic at Intensive Care Units for gram-positive infections.1 With the current dosing regimen, the dose is adjusted based on renal function. However, in the year 2020 only 16% of the patients admitted to the ICU in the Erasmus MC reached the intended target of 20-25 mg/L at around 24 hours after start therapy, which was used in our hospital. Pharmacokinetic (PK) models can be used to predict concentrations and simulate the effect of different dosing regimen. Many population pharmacokinetic models are available in the scientific literature.2 However, implementation into the clinic is difficult, mainly due to the complexity of these models.
The aim was to develop a user-friendly tool, as a proof of concept, to facilitate the implementation of complex PK models into the clinic, which is easy to use for the prescribing physician. This tool calculates an individually optimized starting dose by extracting patient characteristics from electronic patient dossier, select proper pharmacokinetic models and perform simulations to determine the best starting dose of vancomycin in order to increase the number of patients within the target range. With this study we wanted to determine which model to select in our tool for our patient cohort and make a rough estimation on how concentrations would have changed if the tool-based dose would have been given.
Methods: A retrospective analysis was performed in 100 patients at the Intensive Care Unit (n=90) and Orthopedics department (n=10) at Erasmus Medical Center in Rotterdam the Netherlands. All patients received vancomycin as a continuous infusion preceded by a bolus infusion. Patient on continuous renal replacement therapy and extracorporeal membrane oxygenation were excluded for the analysis. A new tool was developed [in Python v3.7] to predict the concentrations of vancomycin and determine the optimal individual dosing regimen using available PK models from scientific literature.
First, PK models were evaluated for overall predictive performance and performance within patient subgroups. Patients’ characteristics and vancomycin administrations were fed into the models to predict the vancomycin concentration for each model and each patient at time of first plasma sample (around 24h after start therapy). These were compared to the observed concentrations. The models that performed poorly for certain patient characteristics (e.g. poor renal function, obesity) were excluded for those patients. These rules for model selection were built into the new tool.
Second, the tool (including model selection) was used to performed simulations to determine the best dose to achieve a concentration of 22.5 mg/L at time of sampling. This predicted dose was used to calculate the expected plasma concentrations based on the actual dose the patient received and the observed plasma concentration, assuming a linear relation between dose and concentration.
Results: A total of ten PK models were implemented into the new software. The average (range) of predicted over observed concentration in percentages for all models was 91.8% (66.8-134.0%) and the average standard error (range) was 40.9% (34.0-46.9%). Model selection for each patient was based on age, weight, BMI, renal clearance. All patients had at least four models which were combined to calculate their optimal dose. The percentage of patients within the target range of 20-25 mg/L at first plasma sample of vancomycin increased from 28% to 39% for respectively the observed concentration and tool based expected concentration. The RMSE decreased from 9.63% to 7.63% for standard versus tool-based dosing respectively.
Conclusions: The number of patients which were on target at first measured plasma concentration of vancomycin increased and had less variability. However, the standard error for the model predicted versus observed concentrations was relatively high. This might be due to the level of uncertainty regarding the exact dosing times in our patient cohort. At 24 hours steady state was not achieved in many of the patients based on the simulations. In the current study, only the continuous infusion was adjusted. Adjusting the bolus dose might be necessary to improve the treatment. In this study the target range of 20-25 mg/L was used as this is the standard care and hence model predictions were done based on target concentration range for a fair comparison as no actual AUC values of our patient cohort was available. The tool can also be used to calculate individual AUC values if a switch is to be made to AUC-based dosing. The new tool was able to retrieve patient data, select models and calculate individualized starting dosages using patient number only. The tool will be tested in a prospective study to evaluate whether we can increase the number of patients on target at the start of their treatment even further.
References:
[1] Rybak, M. J. et al. Therapeutic monitoring of vancomycin in adults summary of consensus recommendations from the American Society of Health-System Pharmacists, the Infectious Diseases Society of America, and the Society of Infectious Diseases Pharmacists. Pharmacotherapy 29, 1275–1279 (2009). [2] Cunio, C. B. et al. Towards precision dosing of vancomycin in critically ill patients: an evaluation of the predictive performance of pharmacometric models in ICU patients. Clin. Microbiol. Infect. 27, 783.e7-783.e14 (2020).
Reference: PAGE 30 (2022) Abstr 10135 [www.page-meeting.org/?abstract=10135]
Poster: Clinical Applications