The use of physiologically-based pharmacokinetic (PBPK) modeling for drug development is well-established and is now routinely used by the pharmaceutical industry, regulators, and researchers. In this blog post, I’ll discuss a novel application that combined PBPK and Bayesian modeling to help clinicians optimize dosing at the point of patient care. This application was used to identify patients at risk of developing serious adverse effects from a standard dose of the antiviral efavirenz. Such patients usually require a decrease in dosing for effective therapy.
Why personalized dosing?
In recent years, we have realized that prescribing a standard dose for all patients is not appropriate for some drugs. The standard dose may not be effective in some patients or cause toxicity in others. This has led us to reconsider the concept of “one-size-fits-all” dosing. Drugs with a high inter-patient variability in pharmacokinetics that also have a narrow therapeutic range present special dosing challenges.
The maturation of pharmacogenetics has enabled us to identify relevant genetic differences in enzymes, transporters, and receptors. This information allows us to optimize dosing for individual patients. The best therapeutic practices now focus on giving the right drugs to the right patient in the right dose instead of using a “one-size-fits-all” approach.
The need for personalized dosing of efavirenz
Efavirenz is a potent non-nucleoside reverse transcriptase inhibitor that is highly effective against HIV-1. It is a first line treatment for HIV infection and forms a component of highly active antiretroviral therapy (HAART).
Efavirenz displays very high inter-subject pharmacokinetic variability. For example, Burger and coworkers studied a cohort of patients on the standard dose of efavirenz. They measured concentrations that ranged from as low as below 1000 nanograms per milliliter (ng/mL) to higher than 10,000 ng/mL. That’s a huge range of plasma concentrations. About 78 percent of these patients were within the therapeutic concentration range. The balance of patients had concentrations above the upper therapeutic window range, which has been associated with serious adverse reactions, such as depression, confusion and hallucinations. These side effects can lead to compliance failure or discontinuation of the drug by the clinician.
Patient differences―including genetic polymorphisms― can cause significant PK variability. CYP2B6 is the main enzyme that metabolizes efavirenz. Polymorphisms in this enzyme can lead to significant variability in the rate of drug metabolism.
Based on these studies, the extensive metabolizers (EMs) of CYP2B6 substrates can take a standard daily dose of efavirenz. However, the intermediate metabolizers (IMs) and the poor metabolizers (PMs) may require dose reduction to get a similar exposure as the extensive metabolizers taking the standard dose. Poor metabolizers frequently experience serious side effects from taking the standard dose, because of very high plasma concentrations. Their dosing must be lower to get effective and safe therapy.
Based on these observations, some clinicians recommend genotype-guided dosing of patients. First, the patients are genotyped to determine which CYP2B6 polymorphisms they have. Then, they’re classified as EMs, IMs, or PMs. Finally, they are prescribed an appropriate dose for their phenotypic status.
However, genotyping may not be available or affordable in developing countries. We were approached by a South African clinician who inquired about tools to identify patients who require dose adjustments of efavirenz. Our scientific team viewed his request as both a scientifically interesting challenge and an opportunity to help patients in nations hit hardest by the HIV epidemic.
Identifying patients requiring dose adjustment
Prior evidence suggested that plasma concentrations of efavirenz could be used as a marker for CYP2B6 activity. Thus, we asked: “Would giving patients the standard dose of efavirenz and then performing serial PK sampling enable identification of poor metabolizers at the initiation of treatment?”
Our proposed plan of action involved the following steps:
- Perform PBPK modeling to create virtual subjects that would be EMs, IMs, and PMs
- Simulate the concentration time profiles in those individuals.
- Use Bayesian modeling to determine the probability of identifying a poor metabolizer by using a drug concentration at a specific sampling time.
- Verify our models using clinically observed pairs of concentration and genotype data. The person doing the prediction would be blinded to the genotype and will be given the concentration only.
- Test the tool in the clinic.
A brief introduction to PBPK modeling
PBPK models describe the behavior of drugs in the different body tissues. Depending on the route of administration, the course of the drug can be tracked through the blood and tissues. Each tissue is considered to be a physiological compartment. 各コンパートメントにおける薬剤の濃度は、システムデータ、薬剤データ、試験デザイン情報を組み合わせて決定されます。The systems data includes demographic, physiological, and biochemical data for the individuals in the study population. The drug data consists of its physicochemical properties, its binding characteristics, and information on its metabolism and solubility. The trial design information comprises the dose, administration route, dosing schedule, and co-administered drugs.
The Simcyp Simulator uses this information as well as various algorithms to predict the PK and pharmacodynamics (PD) of the drug in the population of interest. PBPK models help determine the impact of PK variability associated with genetic polymorphisms in CYP enzymes and drug transporters as well as drug-drug interactions. Thus, this approach enables precision medicine—helping clinicians to determine which drug is right for a patient and at what dose.
Developing a tool to identify patients at risk for efavirenz toxicity
In this study, we used both PBPK and Bayesian modeling. For the efavirenz PBPK model, we used the published model by Xu and coworkers. We simulated virtual EM, IM and PM patients on the standard dose of efavirenz using the Simcyp Simulator version 13. Then, we verified the model using clinical data. Once we were satisfied with the model, we simulated concentration-time profiles in 5,000 virtual individuals of each phenotype (15,000 individuals total) to be used as a training set for our Bayesian model.
Performing rich pharmacokinetic sampling on patients is not practical in developing nations. To have a useful clinical tool, we’d need to be able to identify PMs based on one or two samples extracted at specific time periods. Therefore, we isolated virtual pharmacokinetic samples at multiple time points in the first 24 hours post dosing to see if we could use sparse sampling techniques to identify poor metabolizers.
The Bayesian model was used to calculate the probability of predicting a phenotype, given a concentration at a particular sampling time. Once we identified the model that gave the highest probability of predicting the poor metabolizers, we checked the reliability of our predictions. To do this, we used clinically observed concentration-phenotype pairs. The researcher doing this analysis was blinded to the patient phenotype. She input concentrations into the Bayesian model to predict phenotypes. Thus, we calculated the probability of correctly predicting each phenotype (the true positive rate) and the probability of correctly rejecting each phenotype (the true negative rate).
Results and performance of the tool
We verified the PBPK models using clinically measured patient time-concentration data. Our models provided good recovery of the clinical data for extensive, intermediate and poor metabolizers.
We also calculated PK parameters—the area under the curve (AUC), maximum concentration (Cmax) and clearance—using the simulated concentration-time profiles. Again, comparing the predicted PK parameters with those obtained in two other clinical studies revealed good recovery of the observed data. The predicted AUC—a measure of drug exposure—differed significantly between the EM, IM, and PM groups, which was further evidence for our model’s validity.
Then, we performed Bayesian analysis to determine the probability of correctly predicting each phenotype using a single PK sample taken 24 hours after a single dose of efavirenz. Our analysis showed our ability to identify PMs had an 82% true positive rate and an 87% true negative rate. These results suggest that our tool has a high probability of identifying poor metabolizers.
Potential application of this tool in the clinic
How do we propose using this tool in the clinic? First, a patient would visit the clinic and be prescribed the standard dose of efavirenz. The patient would then return 24 hours later for a blood draw. Next, this blood sample would be analyzed to determine the efavirenz plasma concentration. The plasma concentration would then be entered into the Bayesian model. Finally, depending on the patient’s phenotype, the clinician can adjust the dose for that patient. Of course, we’d expect further drug monitoring to ensure that that revised dose was appropriate for the patient.
This new tool could help clinicians personalize efavirenz dosing. Our next step will be to use a larger clinical data set to further confirm our findings. A similar approach using PBPK and Bayesian modeling could be used for other drugs with high intersubject pharmacokinetic variability and a narrow therapeutic index. This novel application of PBPK modeling and simulation technology has the potential to bring us one step closer to realizing the goal of personalized medicine.
 Chetty M, Cain T, Jamei M, and Rostami A. Application of PBPK and Bayesian Modeling for Prediction of the Likelihood of Individual Patients Experiencing Serious Adverse Reactions to a Standard Dose of Efavirenz. Presented at the American Society of Clinical Pharmacology and Therapeutics 2015 Annual Meeting. March 3-7, 2015, New Orleans, LA.
 Burger D, van der Heiden I, la Porte C, et al. Interpatient variability in the pharmacokinetics of the HIV non-nucleoside reverse transcriptase inhibitor efavirenz: the effect of gender, race, and CYP2B6 Polymorphism. British Journal of Clinical Pharmacology. 2006;61(2):148-154. doi:10.1111/j.1365-2125.2005.02536.x.
 Xu C, Quinney SK, Guo Y, Hall SD, Li L, Desta Z. CYP2B6 Pharmacogenetics–Based In Vitro–In Vivo Extrapolation of Efavirenz Clearance by Physiologically Based Pharmacokinetic Modeling. Drug Metabolism and Disposition. 2013;41(12):2004-2011. doi:10.1124/dmd.113.051755.
Learn more about how modeling and simulation support antiviral drug development
My colleague, Dr. Jos Lommerse, recently gave a webinar on his work to develop an adaptive trial design to support dosing of an antiviral to prevent mother to child HIV transmission in newborns. I hope that you’ll watch the webinar and let me know what you think in the comments!