A Bayesian Model to Describe Factors Influencing Trough Levels of Vancomycin in Hemodialysis Patients

Background/Aims: In hemodialysis patients, there is a marked inter-individual variability in the pharmacokinetics of vancomycin. This retrospective study was carried out to design a model describing the parameters that may influence the trough concentrations of vancomycin (TCV) in hemodialysis patients. Methods: A Bayesian model was constructed from data obtained during 314 hemodialysis sessions performed in 31 hemodialysis patients receiving vancomycin. The model's validity was assessed by goodness of fit. A bootstrap resampling method was used to calculate bias and accuracy for 80 predicted and observed TCV. Results: A total of 31 patients underwent dialysis 3 times a week for a mean duration of 4 h. Their mean age was 69 ± 12 years. The vancomycin infusion was started 30 min before the scheduled end of the dialysis session at a flow rate of 1,000 mg/h. The mean TCV of the study population was 16.1 ± 3.2 mg/l. The area under receiver operating characteristic curve of the constructed model was 95.2%. In the validation sample (80 randomly selected TCV), the observed mean TCV was 15.8 ± 3.6 mg/l, whereas the mean TCV predicted by the model was 15.7 ± 3.0 mg/l. If the mean bias was low between the predicted and observed TCV (-0.1 mg/l), SD was high (3.43 mg/l). The variables most closely linked to TCV were in descending order: weight after dialysis, weight before dialysis, the dose of vancomycin administered during the previous dialysis session and creatinine concentration before dialysis. Conclusion: This simple model describes patient-related and dialysis-related parameters that mainly influence TCV. Before its use in clinical practice, this model should be validated prospectively.