Evaluation of two model-based Bayesian approaches for vancomycin dosing in general hospitalized patients.

Abstract

Recent guidelines recommend Bayesian forecasting for vancomycin dosing. However, the agreement and predictive performance of different pharmacokinetic models remain unclear in hospitalized patients in general. This study evaluated two model-based Bayesian approaches implemented in platforms: the Goti et al. model in DoseMeRx and the Rodvold et al. model in PrecisePK. This single-centre prospective cohort study was conducted at Sultan Qaboos University Hospital, Oman, between April and October 2025. Adult inpatients aged ≥ 18 years with normal renal function were included; patients who were critically ill, pregnant, transplant recipients or receiving renal replacement therapy, or had malignancy were excluded. Agreement between the Goti et al. and Rodvold et al. models was assessed using the concordance correlation coefficient (ρc) and weighted Cohen's kappa (κw). Predictive performance was evaluated using relative bias (rBias) and relative root mean squared error (rRMSE). Models were considered clinically acceptable if rBias was within ± 20%. 90 patients were included. The models showed poor concordance (ρc = 0.59; 95% CI 0.44-0.71) and fair agreement (κw = 0.36; 95% CI 0.23-0.50). The Goti et al. model displayed poor predictive performance in prior predictions (rBias 33.92%; rRMSE 72.86%) and posterior predictions (rBias 20.17%; rRMSE 50.49%). The Rodvold et al. prior model also exhibited bias (rBias - 19.17%; rRMSE 51.52%). In contrast, the Rodvold et al. posterior model achieved clinically acceptable performance (rBias 0.47%; rRMSE 13.18%). The disagreement between the models suggests they may lead to different dosing decisions. The Rodvold et al. posterior model demonstrated acceptable accuracy and precision, making it more suitable for model-informed precision dosing. Further studies in larger and more diverse populations are needed.