Optimizing antibiotic treatment for the critically ill through pharmacometric approaches

Adequate dosing of antibiotics remains a major challenge for treating intensive care (ICU) patients, which are characterized by large pharmacokinetic (PK) variability between patients and within patients. In such cases, the conventional one-size-fits-all dosing strategy is flawed as the PK target may never be achieved in some patients. Pharmacometric modeling as an arising instrumental subject has shown its promising potentials in dose optimization. This thesis presented model development/validation, methodological optimization, and the application of such an approach in a real word setting with a focus on antibiotic treatment in ICU patients. The first section focused on the obtainment of pharmacometric models including both model validation and model development. In chapter 2, vancomycin models for ICU patients were selected from the literature and an external validation procedure was subsequently applied. The model by Roberts et al. was identified as the best model describing the data from our ICU patients with a median prediction error of -7.5%. In chapter 3, a model was developed to study the PK variability of ciprofloxacin in ICU patients. The study showed that there is still a large amount of PK variability of ciprofloxacin in the ICU population that cannot be explained by the inclusion of the most easily available covariates, such as body weight and renal function. One of the most important and common applications of pharmacometric modeling in clinical care lies within model-based therapeutic drug monitoring (TDM). In the second section, taking vancomycin as an example, we focused on maximizing the value in the dose optimization that model-based TDM can bring to clinical care, from both a practical and methodological perspective. In chapter 4, we revealed that taking more samples has no added value compared to taking only trough samples. We also found that in order to achieve this target in as many patients as possible, the dose computation should be based on a mathematical target of AUC of 500 mg·h/L. The essential technique in model-based TDM concerns maximum a posteriori estimation (MAP). In chapter 5, we explored the impact of including historical TDM data on the predictive performance of model in ICU patients. We found that the earlier the TDM data were collected, the worse the predictive performance was of the model. Additionally, we proposed two new MAP estimation methods, namely adaptive MAP and weighted MAP. The adaptive MAP method showed a better predictive performance compared to the other two and does not require to discard historical data. In the third section, we moved a step forward bringing the pharmacometric modeling technique to the bedside. In chapter 6, we systematically illustrated and discussed the design and implementation of AutoKinetics, a clinical decision support system for real-time dose recommendations. The software automates the process of collecting and organizing TDM data, performing MAP estimation, and dose computation. In order to validate the software, we conducted a randomized controlled clinical trial. In chapter 7-8, we presented the design and the results of the clinical trial with a main focus on the pharmacokinetic and clinical outcomes. Real-time bedside model informed precision antibiotic dosing was shown to be feasible and safe. No between-group differences were observed for the clinical outcome. AutoKinetics may be deployed to obtain any desired PK target provided suitable PK models are implemented. ICU patients exhibit large variability between and within individuals which make dosing in ICU patients more challenging than in normal patients. With the assistance of pharmacometric models, we are able to optimize antibiotic dosing using individual data. The trial has shown that model-based dosing strategy can outperform conventional dosing with the use of suitable models.