Objective To evaluate external predictability of a population pharmacokinetic model of valproic acid in Thai patients with mania to ensure its appropriateness for use in other clinical settings.

Methods The published population pharmacokinetic model was evaluated for its predictive ability (at both individual and population levels) and its precision by means of mean absolute prediction error (MAPE), root mean square error (RMSE) and normalised prediction distribution error (NPDE).

Results Forty-six steady-state serum valproic acid concentration levels from 30 manic patients were retrospectively collected from routine therapeutic drug monitoring at Srithanya Hospital, Thailand. For the prediction-based diagnostics, the MAPE and RMSE were 10.44% (95% CI 8.12% to 12.76%) and 12.99% (95% CI 9.51% to 15.72%), respectively, suggesting that the proposed model was relatively predictive and had a good precision. In simulation-based diagnostics, the NPDE results also showed that the model appropriately predicted valproic acid concentration levels, as indicated by a normal distribution of NPDEs with a mean and a variance of 0 and 1, respectively.

Conclusion The predictability of the population pharmacokinetic model of valproic acid in Thai patients with mania was confirmed. This model could be applied in other clinical settings.