In 1909 the discovery of the antibiotic arsphenamin marked the beginning of a new era in treating potentially deadly bacterial infections. In the following decades, the discovery of various new antibiotic drugs substantially contributed to a rise in life expectancy from 47.0 to 78.8 years in the United States of America. Despite this considerable progress in treating infectious diseases, bacterial infections remain a major threat to public health. Especially vulnerable patient populations, like critically ill patients, continued to suffer under mortality rates up to 60%. Worryingly, the described achievements are threatened by two alarming developments: While no truly novel antibiotic classes have been discovered and developed in the last three decades, the emergence and spread of antimicrobial resistance -accelerated by the inappropriate use of antibiotic drugs - steadily reduces the efficacy of currently available drugs. As a response to this new challenge, several national and international action plans call not only for a determined search for new antimicrobial drugs, but also for a more rational use of existing antibiotics. One vital component of rational antibiotic drug therapy is an adequate drug exposure at the site of infection, facilitated by the selection of suitable antibiotic drug(s) in combination with an appropriate dosing regimen. The antibiotic drug administered to the patient should be selected based on its efficacy against the pathogen causing the infection. Unfortunately, the pathogen causing the infection is often unknown at the start of antibiotic therapy. As a consequence, broad spectrum antibiotics – like meropenem and piperacillin/tazobactam - are frequently administered to increase the likelihood of an effective therapy. The selection of an appropriate dosing regimen can be complicated and is especially challenging in critically ill patients: The broad range of pathophysiological changes observed in this patient population leads to high pharmacokinetic (PK) variability, which results in substantial differences in drug exposures between patients receiving the same antibiotic drug and dosing regimen. Under the concept of model-informed precision dosing (MIPD), population pharmacokinetic/pharmacodynamic models and patient-specific data (e.g. patient characteristics, drug measurement(s)) can be leveraged to inform and improve dosing decisions in this vulnerable patient population. The objective of the presented thesis was the development, implementation and evaluation of MIPD tools for antibiotic drugs in critically ill patients. To enable the successful integration of MIPD into clinical practice an iterative, integrative and translational approach was followed. The initial and central question ’Is the current antibiotic dosing appropriate?’, was iteratively addressed integrating expertise from a diverse interprofessional team of healthcare professionals and can be segmented into four intermediate steps, all vital to the main objective. First, and as a prerequisite both for model development/evaluation and dosing adaptation, the establishment of a reliable and frequent antibiotic concentration measurement program was required. Second, the collected data was analysed employing pharmacometric and statistical methodology to characterise population PK/pharmacodynamics (PD) and local factors influencing antibiotic therapy (e.g. local pathogen susceptibility). Third, the gained scientific knowledge was translated into easy-to-use, model-informed dosing tools and comprehensive dosing strategies optimised for clinical practice. And fourth, the developed model-informed dosing tools were implemented into clinical routine and subsequently evaluated and optimised. This thesis focused on meropenem and the fixed drug combination piperacillin/tazobactam and addressed individual or multiple of these four steps in three different projects. In Project I, a possible adsorption of the antibiotic meropenem at the cytokine adsorber CytoSorb®, its effect on meropenem exposure and possible consequences for an adequate meropenem dosing were investigated. Despite the absence of clear evidence for a beneficial effect on patients outcomes, the CytoSorb® filter is increasingly used to reduce circulating cytokines in patients experiencing sepsis. Due to its unspecific binding and therefore elimination of molecules up to a molar mass of 55 kDa, concerns have been raised that the CytoSorb® filter unintentionally adsorbs various drugs including meropenem. To investigate if meropenem dosing needs to be increased during CytoSorb® treatment, a nonlinear mixed-effects (NLME) modelling and simulation approach was employed: A population pharmacokinetic model was developed and three distinct approaches to assess if meropenem clearance differed without or during CytoSorb® treatment were applied: (i) quantification of a possible proportional increase in clearance during CytoSorb® treatment (ii) investigation of (non)saturable adsorption at the CytoSorb® filter using different adsorption submodels and (iii) model parameter re-estimation excluding samples collected during CytoSorb® treatment and evaluating the predictive performance for meropenem concentrations during CytoSorb® treatment. In contrast to the expectation of meropenem being adsorped at the CytoSorb® filter, no significant (p<0.05) or relevant effect of CytoSorb® treatment on meropenem exposure was observed. Consequently, neither additional dosing nor a more frequent drug concentration monitoring of meropenem is necessary during the application of CytoSorb® therapy. Project II focused on improving meropenem and piperacillin/tazobactam treatment for critically ill patients at the Charité-Universitätsmedizin Berlin. For this purpose, a 3-staged clinical study was initiated as a coordinated intervention. In stage I, a frequent and reliable concentrations measurement program was implemented to evaluate the current antibiotic therapy. The assessment of the current antibiotic therapy provided insights about local pathogen susceptibility, while highlighting the need for dose individualisation based on patient characteristics: The majority (>90%) of observed pathogens were susceptible to the two administered antibiotic drugs, but target range attainment (minimum antibiotic drug concentrations between 1 and 5 times minimum inhibitory concentration (MIC) of the pathogen) was low for the observed drug concentrations (meropenem: 35.7%, piperacillin: 50.5%) and highly variable between patients with different renal functions. To improve initial meropenem dosing (i.e. prior to the first concentration measurement) and to exploit the newly gained information about the local pathogen susceptibility, a tabular model-informed dosing tool was developed and implemented in stage II of the study. For the development of the tool, an appropriate meropenem PK model was selected from literature and successfully evaluated using the local clinical data. The PK model was then used to conduct stochastic simulations investigating clinically relevant dosing regimens, possible clinical scenarios and the probability of the dosing regimens to achieve adequate drug exposures. To inform dosing prior to pathogen identification, the local pathogen-independent mean fraction of response (LPIFR) was introduced: The LPIFR characterises the probability of a dosing regimen to reach a defined target, e.g. time above the MIC, if only the underlying MIC distribution at a hospital and not the individual MIC of the pathogen causing the infection is known. To inform dosing after MIC value determination, probability of target attainment analyses (PTA) were performed. Dosing recommendations achieving PTA>90% or LPIFR>90% for patients with different creatinine clearances (10.0-300 mL/min) were derived and summarised in one concise and clear table. To assess the potential of the newly developed model-informed dosing tool prior to implementation, the total daily dose of the dosing regimens recommended by the dosing tool for the local study population was compared to the total daily dose of the actually administered dosing regimens. For 77% of the patients with meropenem concentrations outside the target range, the dosing tool suggested a change in daily dose, highlighting the potential of the tool to optimise dosing regimens. To integrate patient individual antibiotic drug measurements and allow for more user flexibility, an interactive model-informed dosing software termed ‘DoseCalculator’ was developed for stage III of the study. In addition to the meropenem PK model already evaluated for stage II of the study, different piperacillin/tazobactam models were extracted from literature, evaluated using the local clinical data collected in stage I of the study and the best performing model implemented into the tool. Based on available knowledge about the infection, three possibilities to calculate the probability of a dosing regimen to reach adequate antibiotic exposures were integrated into the tool: (i) the LPIFR if neither the pathogen nor the MIC is available, (ii) the cumulative fraction of response (CFR) based on the MIC distribution of a specific pathogen if the pathogen is available and (iii) the PTA if the MIC is available. Furthermore, employing a maximum a-posterior (MAP) estimation approach the observed antibiotic drug measurement(s) of a patient can be used in the DoseCalculator to derive patient individual parameter estimates. If drug measurement(s) of a patient are supplied, all analyses and the resulting recommended dosing regimen are based on the individual parameter estimates of the patient. Compared to the observed dosing in stage I, the recommendations of the DoseCalculator led to a substantial relative increase in predicted target attainment (322% meropenem, 505% piperacillin) while reducing the daily dose (median reduction: 77.8% meropenem, 83.4% piperacillin). In Project III the MeroRisk Calculator, an easy-to-use Excel tool to determine the risk of meropenem target non-attainment after standard dosing previously developed at our department, was evaluated using clinical routine data. Since the direct evaluation of the MeroRisk Calculator was not feasible with the available retrospective clinical dataset, a two-step data- and model-based evaluation was conducted: In step one, a meropenem PK model was successfully evaluated using the clinical data. In step two, the evaluated PK model was used as a benchmark for the drug concentration and risk predictions of the MeroRisk Calculator. Compared to the successfully evaluated compartmental PK model, the MeroRisk Calculator provided an equally good and reliable risk assessment (Lin’s concordance correlation coefficient = 0.99) for patients with maintained renal function (creatinine clearance > 50 mL/min). However, for patients with creatinine clearances below 50 mL/min significant deviations were observed. As a consequence, the MeroRisk Calculator should not be used in patients with (severe) renal impairment. In addition to the successful evaluation, the functionality of the MeroRisk Calculator was extended. Based on CFR analysis and EUCAST reported MIC value distributions, risk of target non-attainment can now be assessed depending on the infecting pathogen informing dosing decisions prior to MIC value determination. To conclude, the presented thesis contributed to an individualised and more rational antibiotic drug therapy in critically ill patients. While PK modelling was employed in Project I to exclude a clinically relevant adsorption of meropenem at the CytoSorb® filter, Project II and Project III represent a successful example on development, implementation and evaluation of MIPD tools. As a next step, both the tabular model-informed dosing tool and the DoseCalculator should be prospectively evaluated at Charité-Universitätsmedizin Berlin. The results from this evaluation in particular and this thesis demonstrate the potential of MIPD using comprehensive examples on how to develop, implement and evaluate model-informed dosing tools and contribute to the accelerating implementation of MIPD into clinical practice.