Abstract:Renal function is an important factor affecting the pharmacokinetics of vancomycin. The renal function in elderly patients gradually decreases with age. An accurate estimated glomerular filtration rate (GFR) is essential in drug dosing. The study aimed to determine the most appropriate renal function estimation equations to describe vancomycin pharmacokinetics in elderly patients using population pharmacokinetic analysis. Data were obtained retrospectively from elderly patients aged ≥65 years who received vancomycin for infection from September 2016 to January 2022. Renal function was estimated using the Cockcroft‐Gault equation (CG), Modification of Diet in Renal Disease equation (MDRD), three Chronic Kidney Disease Epidemiology Collaboration equations (CKD‐EPIcys‐scr, CKD‐EPIscr, and CKD‐EPIcys) and two Berlin Initiative Study equations (BIS‐1 and BIS‐2). The CKD‐EPIcys‐scr and BIS‐2 equations were based on cystatin C (Cys C) and serum creatinine (Scr). The others were based on Cys C or Scr. A nonlinear mixed effects model (NONMEM) was used to develop the population pharmacokinetic model. A total of 471 serum concentrations from 313 elderly patients were used to develop the population pharmacokinetic model. Weight and GFR were identified as significant covariates affecting the pharmacokinetics of vancomycin. Cys C and Scr‐based GFR (CKD‐EPIcys‐scr and BIS‐2) yielded significant improvement performance compared with the other equations in model building. The interindividual variability of CL was reduced from 49.4% to 23.6% and 49.4% to 23.7% in CKD‐EPIcys‐scr and BIS‐2 based models, respectively. However, greater interindividual variabilities of CL (from 26.6% to 29.0%) were represented in the other five models which were based on either Cys C or Scr. The GFR estimated by EPIcys‐scr and BIS‐2 equations and vancomycin CL exhibited a good correlation (r = 0.834 and 0.833). In the external validation with 124 serum concentrations, the predictive performances of the CKD‐EPIcys‐scr and BIS‐2 based models (the mean relative prediction errors were less than 1%, the mean relative absolute prediction errors were about 23%) were also superior to the other five models (the mean relative prediction errors were about 2%, the mean relative absolute prediction errors were greater than 25%) which are based on either Cys C or Scr. In this study, we determined that the equation used to estimate GFR can affect the population pharmacokinetic model fitting result. Population pharmacokinetics model with CKD‐EPIcys‐scr or BIS‐2 can be used to optimize vancomycin dosage in elderly Chinese patients.