Can quantitative systems pharmacology guide precision dosing? An illustrative warfarin-INR example
Undine Falkenhagen, Larisa Cavallari, Julio Duarte, Stephan Schmidt, Charlotte Kloft, Wilhelm Huisinga
Universität Potsdam
Objectives
Model-informed precision dosing (MIPD) is highly relevant for drugs with large interindividual variability (IIV) and a small target window (TW), which is the case for the anticoagulant warfarin. Warfarin is widely used for the prevention and treatment of thromboembolic disorders [1]. A large proportion of the observed variability in dose requirement is not yet explained by identified covariates, thus necessitating careful dose titration [2].
Quantitative systems pharmacology (QSP) models integrate comprehensive knowledge about pharmacologic processes. Due to their complexity, however, they are typically unsuitable for parameter estimation in the context of MIPD. To leverage the physiologic knowledge, we have previously derived a small-scale warfarin-INR model by model reduction from a blood coagulation QSP model (including warfarin PK) with assumed variability on all parameters [3]. The present work aimed at externally evaluating the reduced warfarin-INR model’s predictions in a diverse patient cohort.
Methods
We performed the external validation in a diverse cohort of adult patients initiating warfarin therapy, including 51% African American patients [4]. INR data, demographic and pharmacogenetic data were available. For this analysis, only in-patient data was used (up to ten INRs per patient). Only patients with INR data at day 5 or later were considered, leaving 81 patients.
The considered warfarin-INR model inherits biological interpretability from the QSP model [3]. In the reduced model, the INR is a function of coagulation factors II, VII and X, which are indirectly inhibited by warfarin. Age, weight, CYP2C9 genotype and rs12777823 (only in African Americans) were considered as PK covariates [5-7]. INR baseline, VKORC1 and CYP4F2 genotypes were considered as PD covariates [6,8]. The reduced warfarin-INR model includes lognormal IIV with 40% CV on all parameters and lognormal residual unexplained variability (RUV) with 20% CV. We also evaluated an alternative prior with 10-40% IIV and 18% RUV.
We used a Bayesian particle filter approach [9], with individual prior parameter distribution given by the IIV, dependent on the covariates. Data until day 4 were assimilated for each patient to obtain their posterior parameter distribution. The posterior distribution allowed to predict the probability of being within the TW for a given time and dosing history. To calculate optimal daily doses, we simulated the probability of being within the TW based on the posterior distribution for a range of doses between 0.5mg and 20mg and chose the dose with highest predicted probability.
Results
We evaluated the reduced model’s performance in predicting the INR for days 5-9 based on INR data from days 1-4. The accuracy (assessed by determining the fraction of INR measurements within the 90% prediction interval) was 86% for the original and 81% for the alternative prior. The precision (average predicted probability of being within the TW at SS given optimal dose) was 52% for the original and 61% for the alternative prior. An established empirically-based PK/PD model [6] was also evaluated and found to have comparable accuracy (84%) and precision (59%).
Additionally, we compared the predicted optimal doses to the actual doses at discharge. The reduced warfarin-INR model predicted the optimal dose within ±20% of the mean actual dose for 56% of the patients. For the empirically-based PK/PD model, the percentage was 42%.
Further analysis revealed that the RUV is a key factor influencing the probability of being in the TW, independent of the structural model. For lognormal RUV with 20% CV [6], the INR can be in the TW with at most 69% probability (assuming no uncertainty for individual model parameters). Thus, it is crucial to reduce the RUV, e.g., by introducing further covariates or biomarkers. This can be done mechanistically in the warfarin-INR model thanks to its biological interpretability.
Conclusion
The reduced warfarin-INR model showed promising results in predicting clinical data from a diverse patient cohort, performing as well as an established empirically-based PK/PD model concerning INR prediction and dose optimisation. In contrast to the development of empirically-based models, deriving the model from a QSP model required no experimental patient data. This work highlights the potential of mechanistically developing PK/PD models to be employed in MIPD by reducing QSP models.
References:
[1] Takahashi H, Echizen H. Pharmacogenetics of warfarin elimination and its clinical implications. Clin Pharmacokinet. 2001;40(8):587-603. doi:10.2165/00003088-200140080-00003
[2] Hamberg AK, Dahl ML, Barban M, et al. A PK-PD model for predicting the impact of age, CYP2C9, and VKORC1 genotype on individualization of warfarin therapy. Clin Pharmacol Ther. 2007;81(4):529-538. doi:10.1038/sj.clpt.6100084
[3] Falkenhagen U, Knöchel J, Kloft C, Huisinga W. Deriving mechanism-based pharmacodynamic models by reducing quantitative systems pharmacology models: An application to warfarin. CPT Pharmacometrics Syst Pharmacol. 2023;00:1-12. doi:10.1002/psp4.12903
[4] Arwood, M., Deng, J., Drozda, K., Pugach, O., Nutescu, E., Schmidt, S., Duarte, J. and Cavallari, L. (2017), Anticoagulation Endpoints With Clinical Implementation of Warfarin Pharmacogenetic Dosing in a Real-World Setting: A Proposal for a New Pharmacogenetic Dosing Approach. Clin. Pharmacol. Ther., 101: 675-683. doi:10.1002/cpt.558
[5] Perera MA, Cavallari LH, Limdi NA, et al. Genetic variants associated with warfarin dose in African-American individuals: a genome-wide association study. Lancet. 2013;382(9894):790-796. doi:10.1016/S0140-6736(13)60681-9
[6] Hamberg AK, Wadelius M, Lindh JD, et al. A pharmacometric model describing the relationship between warfarin dose and INR response with respect to variations in CYP2C9, VKORC1, and age. Clin Pharmacol Ther. 2010;87(6):727-734. doi:10.1038/clpt.2010.37
[7] Hamberg AK, Friberg LE, Hanséus K, et al. Warfarin dose prediction in children using pharmacometric bridging--comparison with published pharmacogenetic dosing algorithms [published correction appears in Eur J Clin Pharmacol. 2013 Sep;69(9):1737]. Eur J Clin Pharmacol. 2013;69(6):1275-1283. doi:10.1007/s00228-012-1466-4
[8] McDonald MG, Rieder MJ, Nakano M, Hsia CK, Rettie AE. CYP4F2 is a vitamin K1 oxidase: An explanation for altered warfarin dose in carriers of the V433M variant. Mol Pharmacol. 2009;75(6):1337-1346. doi:10.1124/mol.109.054833
[9] Maier C, Hartung N, de Wiljes J, Kloft C, Huisinga W. Bayesian Data Assimilation to Support Informed Decision Making in Individualized Chemotherapy. CPT Pharmacometrics Syst Pharmacol. 2020;9(3):153-164. doi:10.1002/psp4.12492