PK/PD modelling - vancomycin - Mw\Pharm - therapeutic drug monitoring

Objective: To evaluate the possibility of population based dose optimisation with the aid of MW\Pharm modelling and to find the best model in the Windows version (WIN). Matherials: 25 patients repeatedly examined for vancomycin (mean age 63±14 years, body weight 88±21kg, median dose 1g/12 h). Methods: Trough concentrations predicted by WIN models “vancomycin_adult_k_C2“, ”#vancomycin_adult_C2“, ”vancomycin_adult_C2“, and ”vancomycin adult“ DOS model (DOS) were compared with the measured value. Statistics: The percentage prediction error (%PE) calculated as (predicted–measured)/measured values, Blandt-Altman plot, RMSE, Pearson’s coefficient of rank correlation (R). Data presented as mean ± SD. Student’s t-test were used for prediction precision evaluation. Results: The %PE varied from +44.4 ± 65.2% to +76.5 ± 84.3 %, P<0.001. ”#vancomycin_adult_C2“ model produced lowest %PE among WIN models as well as the lowest RMSE (79) and Blandt-Altman bias (-4.01 ± 7.59), but the Pearson’s correlation (0.6843, P=0.0002) was less tight. DOS model produced the second lowest RMSE (81), %PE (+45.9 ± 66.6%), and Blandt-Altman bias (-4.83 ± 6.97) and highest Pearson’s R (0.7847, P<0.0001). ”vancomycin_adult_C2“ produced the third best prediction: RMSE (113), %PE (62.8 ± 92.6%), Blandt-Altman bias (-6.78 ± 11.2) but Pearson’s R was the poorest (0.5773, P=0.0025). Conclusion: The lowest %PE and highest Pearson’s R were achieved by the ”#vancomycin_adult_C2“ model. Due to the poor predictive performance of all MW\Pharm versions and models we find all of them unsuitable for a-priori vancomycin dosing management. Other software should be evaluated for routine use.