PII-098 - ON TRANSLATING PHYSIOLOGICALLY BASED PHARMACOKINETICS INTO CLINICAL CARE: REAL-WORLD PREDICTABILITY OF VALPROATE THERAPEUTIC DRUG MONITORING.

H. Tsai¹, Y. Chen², Y. Huang¹, Y. Huang¹, Y. Ho¹; ¹National Taiwan University College of Medicine, Taipei, Taiwan, ²Department of Pharmacy, National Taiwan University Hospital, Taipei, Taiwan.

Background: The realization of model-informed precision dosing in routine clinical care has been an ideal yet with slow progress. We have established a valproate physiologically based pharmacokinetic (PBPK) model and validated against published clinical studies (DOI: 10.1002/psp4.13045) for possible utility in patient care. This exploratory study aims to examine the predictability of the valproate PBPK model against patients’ therapeutic drug monitoring (TDM) data and to identify factors associated with noncordance between PBPK-predicted and TDM values.

Methods: With prior ethics approval, patients’ demographics, clinical status, prescription details, and relevant laboratory data were retrieved from electronic medical records of a medical center. Eligible patients were: 1) aged between 18 and 65 years, 2) received intravenous valproate therapy, and 3) had valproate TDM data between 2018 and 2020. Based on previously validated valproate PBPK model, each patient’s valproate plasma concentration was simulated according to individual characteristics and valproate regimen by Simcyp. The PBPK-predicted and observed TDM values were compared to examine valproate model predictability. Independent t-test was performed to compare predicted-to-observed ratios (Pred/Obs) between groups of various covariates.

Results: A total of 258 TDM data from 150 patients were simulated using the valproate PBPK model. The PBPK-predicted values of 202 (78.3%) TDM data from 136 patients were within two-fold of the observed concentrations, leaving 46 (17.8%; from 26 patients) overestimated and 10 (3.9%; from 8 patients) underestimated. Between group comparisons of covariates such as body mass index ( < 30 vs.  30 kg/m2), alanine transaminase ( < 41 vs. >= 41 U/L), estimated glomerular filtration rate (>= 60 vs. < 60 mL/min/1.73 m2), or on-dialysis (none vs. yes) were with no statistical differences. However, concurrent use of valproate-lowering drugs, especially carbapenems, had significantly higher mean Pred/Obs ratio than their counterparts (1.76 +/- 1.00 vs. 1.29 +/- 0.66, p < 0.001).

Conclusion: The recently developed valproate PBPK model can capture more than three quarters of real-life plasma concentrations. Refined models for resolving prediction deviations by drug interactions are highly awaited.