Background and objectives Critically ill patients display pathophysiological features that cause high inter- and intra-individual pharmacokinetic (PK) variability, which can result in supra or sub-therapeutic drug concentrations. Among the factors involved in PK variability, renal clearance appears as an important contributor, whether as renal failure1 or augmented renal clearance2. Beta-lactams, and more specifically cefotaxime, are a good example of a drug commonly used in intensive care unit (ICU) patients, for which a high risk of not reaching the PK/PD target3 has been observed due to renal clearance changes. However, factors leading to intra-individual variability during the treatment course remain poorly described. A better integration of the evolution of renal function in ICU patients in population pharmacokinetic (PopPK) models is therefore suitable to guide a priori dose adjustment of cefotaxime. Materials and methods In a prospective, observational study, adult ICU patients (October 2015 - May 2017) were treated with a loading dose (2-4g/0.5h) followed by a continuous infusion (1-24g/24h) of cefotaxime. Cefotaxime concentrations 30 minutes after the loading dose, during the continuous infusion at day 1, 4 and 7 after drug initiation, and 4 hours after the end of the infusion, were measured by a validated HPLC-UV method. Demographic, clinical and biological data were collected at inclusion and throughout the study. Data were analysed by nonlinear mixed-effects modeling using MONOLIX version 2023R1. Data below the limit of quantification were treated as left-censored. The impact of covariates values at baseline was evaluated using the stepwise covariate model procedure. Time-varying covariates reflecting the evolution of renal function during the course of treatment were first included in the model as a regressor, then as a biomarker model. Results The study included 76 ICU hospitalized patients (31 females, 45 males, age 57.5 ± 17.6 years). 251 cefotaxime plasma concentrations were available for PK analysis. 597 serum creatinine levels were collected to monitor the renal function. A one-compartment model with linear elimination and proportional residual error best described the PK data. Inclusion of baseline serum creatinine concentration as a covariate on cefotaxime clearance significantly improved the model and reduced the unexplained inter-individual variability of clearance from 54 to 31%. Inclusion of time-varying serum creatinine as a regressor further improved the model. Backward interpolation of the regressor performed better than forward interpolation. In order to anticipate changes in kidney function, particularly from normal kidney function to kidney injury, a joint model with a biomarker model of serum creatinine impacting the cefotaxime PK model was developed. A mono-exponential increase followed by a mono-exponential decrease best described kidney injury and recovery. Conclusions A PopPK model of cefotaxime was developed, encompassing the variation of renal function over the treatment course. Serum creatinine concentration explained a significant part of the high inter- and intra-variability of cefotaxime PK in ICU patients. Since the impact of renal function might be similar for other β-lactams4, our approach might be applied to a common PopPK model for various β-lactams in order to optimize a priori dosing adjustment in the context of drug switches.