Comparison of Two Population Pharmacokinetic Models for Predictive Performance in Vancomycin Therapeutic Drug...

Accurate model-based prediction of vancomycin concentration is essential for therapeutic drug monitoring (TDM); however, the practical criteria for selecting population pharmacokinetic (PopPK) models remain unclear. We compared the predictive performances of the Japanese Oda and Yamamoto PopPK models.

This single-center retrospective study included adults aged ≥18 years who underwent vancomycin TDM between January 2023 and December 2024. Vancomycin concentrations were predicted using both models. Predictive performance was evaluated by absolute error (AE) and ΔAE, defined as AE for the Oda model minus AE for the Yamamoto model. Decision tree analysis was used to assess patient characteristics associated with differences in prediction error between the 2 models.

Of the 396 patients, 296 were analyzed; the median age was 80 years. The mean AE values for the Oda and Yamamoto models were 3.61 ± 3.01 and 3.78 ± 2.89 µg/mL, respectively (ΔAE, -0.172 µg/mL). Creatinine clearance (CCr), body weight (BW), age, and serum albumin (Alb) were selected as splitting variables in the decision tree analysis, and most terminal nodes displayed lower AE in the Oda model. In limited subgroups, including patients with very high CCr, lower BW, and older age, the Yamamoto model showed a lower AE. Using rounded-up serum creatinine (SCr), the mean AE was 3.63 ± 2.95 and 3.96 ± 2.99 µg/mL, respectively, for the Oda and Yamamoto models (ΔAE, -0.334 µg/mL). Rounding up a low SCr level did not improve the overall predictive performance.

The Oda model showed a slightly lower prediction error than the Yamamoto model and may be a reasonable first-choice PopPK model for vancomycin TDM in elderly Japanese patients. Routine rounding at low SCr levels is not supported.