Last year, I presented a webinar on Virtual Twin™ Technology that provided an overview of how this technology could be used in healthcare to predict the drug dose for an individual patient that is most likely to improve efficacy and/or lower the risk of toxicity. During the webinar, I provided a general overview of the Virtual Twin concept and demonstrated several applications of the technology for Model-informed Precision Dosing (MIPD).1 One of the applications described was its use to accurately predict olanzapine (OLZ) exposure in individual patients.2 The second application was to show how Simcyp-guided ADME biomarker discovery could be used to predict which patients are more likely to experience drug toxicity. Specifically, this was a study with the protein kinase inhibitor dabrafenib, a drug used to treat metastatic melanoma with mutated isoforms of the BRAF gene V600E and V600K.3
The webinar generated a great deal of interaction with the participants resulting in some very valuable questions. I hope this Q&A follow-up can help to further elucidate the future role that Virtual Twin could play for MIPD in a clinical setting.
Q: For building the base Virtual Twin model, (1) do you recommend including demographic data such as age, gender, glomerular filtration rate, etc. into account when you compare it to the observed data, and (2) can you talk about the model verification processes such as you would use for PBPK in drug-drug interaction studies?
A: To answer part 1 of the question, yes, you should include as much of that information as possible in the model. We are trying to incorporate all the information that we have about the individual patient. The basic demographic data such as age and weight are all important. So, the more of that information you have, the better. And, if you’ve got a sense of renal function too, that should be incorporated into the Virtual Twin model. Although you might suspect that some of this is not going to be major for describing between subject variability in pharmacokinetics, I would still include it in the model to try and mimic the actual patient as much as possible.
In regards to your second question on verification, we follow the standard FDA guidance about verification and development of PBPK models. There are lots of papers now that are beginning to do that. A really nice example is the dabrafenib work that’s been done at Flinders University in South Australia. We followed step-by-step the FDA guidance on development of PBPK models – this is the dabrafenib work that I showed during the webinar. So, it’s definitely following those standard FDA verification pathways.
Q: Can you speak to some of the challenges or difficulties, and what sort of infrastructure or personnel would be needed to help deploy model-informed precision dosing technology in healthcare settings?
A: The incredible opportunities in this area bring about many challenges. For example, the science of in-vitro, in-vivo extrapolation, a key factor in moving Virtual Twin forward will need a robust and validated way of estimating CYP abundances in the liver and in the gut. Of course, CYP3A is absolutely critical in this regard.
A non-invasive way of getting that information would be fantastic! There’s a lot of thinking about marrying top-down with bottom-up modeling approaches and having some structure around that at the moment. For example, how many of the systems parameters need to be estimated beforehand? And how much do you inform the models after clinical data become available? These are critical questions.
If we put the science aside, there are also big challenges faced with the culture of prescribers. In hospital settings and primary care, where the majority of doctors who are doing the bulk of prescribing are not clinical pharmacologists, and they are not focused on precision drug dosing, but rather selecting a dose from the approved dose range which they use all the time. The problem is that these doses don’t work for everyone!
During the webinar I mentioned that, in general, I think that between-subject variability and exposure is fairly poorly understood in healthcare. So, there are a lot of educational challenges for individuals, prescribers, and certainly at the junior level, about some basic clinical pharmacology concepts. Basically, they don’t realize there’s a problem with such variability leading to poor clinical outcomes. If the problem is not known, then there’s going to be no impetus for change. And, there will be no recognition of whether these modern, informed, dosing technologies could be utilized. So, there’s the scientific challenges and then there’s education challenges which we’re working on.
In regards to who do we see leading this type of push or who would actually be the end user in applying these technologies in healthcare, it’s probably going to be best as a collaborative effort between prescribers and clinical pharmacists. Clinical pharmacy would be essential in driving this technology forward once decisions to prescribe and drug choice has been done by the doctor. The initiation doses and then fine tuning of doses afterward could fall under the clinical pharmacies’ area.
Q: Will the clinical study planned for olanzapine use the Virtual Twin technology to decide the dose for individual patients?
A: The Virtual Twin and olanzapine study was a proof of concept study to see if we can predict steady state plasma concentrations in patients. Now that we’ve done so and we’ve developed that model, the next phase would be to apply that model and give that information to clinicians to guide dosing. The trouble with olanzapine, which we observed in the Flinders study, is that it’s used as part of the acute, agitation protocol. This is why we only had 14 individuals in that small clinical study because it’s hard to find individuals who are just starting off olanzapine without having taken it as part of an acute agitation protocol.
A better place to start a clinical impact study would be with clozapine, and that project is currently underway and advancing nicely. Developing the baseline model for clozapine is pretty well advanced, and a clinical study will investigate whether giving psychiatrists that extra information about predicted clozapine exposure could accelerate up-titration during initiations. So, we have our normal initiation protocol for clozapine – half the prescribers will follow that and then half will get the steady state plasma concentration predicted for the individual by the Virtual Twin technology, and then they will be able to personalize the up-titration based on the predictions using their clinical judgment.
So, to answer the original question, we haven’t applied the olanzapine model. But, we’re certainly looking at applying the clozapine model. And hopefully that should reveal some exciting data in the future.
Q: Are there any conflicts with using patient personal data for making these Virtual Twin simulations?
A: Most of the information is part of electronic health records, basic available information, e.g., age and weight, liver function tests and renal function is – so there’s no issue there.
There may be confidentiality issues around when individuals have had their drug metabolizing enzyme or transporter genotyped, how that information becomes available, and how it’s utilized. This information around genotyping will need to have a framework for how that information could be applied within the Virtual Twin. And, it’s going to be up to the individual to release that information – to populate their Virtual Twin with that information – if they desire.
Drug-drug interaction information will be obviously transparent because it’s going to be part of the patient’s e-medication list.
Q: If physicians start to use model-informed precision dosing tools such as Virtual Twin in their clinical practice a lot more, how could they be sure that the Virtual Twin generated by the technology actually matches their patient?
A: This is where we’re starting to build the evidence. It’s going to be a lot of effort over the next few years going into various areas of clinical medicine to try and establish good quality evidence that is translatable into clinical practice. Initially, it’s going to be from an observational point versus an interventional point. You need to build the evidence, as we see with olanzapine, where you are able to get decent predictions of final steady state concentrations focusing on pharmacokinetics initially.
Certara is working with several academic centers who are embracing and developing Virtual Twin technology for several key therapeutic areas to build evidence-based medicine information so that the technology may eventually be available to prescribers more broadly. The goal is that we can scale this to doctors outside of the specialist academic centers and make the technology intuitive and easy for all to use.
Q: What is the applicability of Virtual Twin in the clinic in the context of a new patient regarding the time that’s required for the simulations, the analysis, etc.?
A: Initially, it’s going to take some time to generate these virtual individuals. You’ll need to have a PBPK platform that you can input data prior to patients actually rolling up to clinics. In actuality, you’re going to have a Virtual Twin of that individual before they come to clinic already developed because it will take too long to generate them during a consult. But, as time goes by and people become familiar with an established tool, and the evidence builds, that part would become much faster.
But, time is not always a barrier. For example, when a patient is starting a protein kinase inhibitor for their cancer or switching to another one, the time isn’t a great factor in that case because the extra resources required for putting together the model for that patient would be absolutely critical for helping select the dose that avoids costly use if clinical resources later on, such as avoiding toxicities. The flip-side is, when you’re in an acute setting, the time spent developing a virtual twin and the potential benefits of that approach would be much more limited. So, identifying those high impact areas where the extra time that it takes to generate a Virtual Twin is worth it will be important – where the predictions is very valuable in terms of the outcome for that patient. Acute settings, of course not, but chronic, longer-term conditions such as oncology, identifying those high-impact areas of MIPD will be critical.
Subsequent to my webinar, recent publications, scientific interest in, and research into MIPD are growing and gaining momentum. It is anticipated that MIPD adoption will follow a similar path to the clinical implementation of pharmacogenomics, which has now become an important consideration in prescribing for some medical conditions and drugs. These recent studies include the application of the “Virtual Twin” approach in the cardiac drug safety arena and identifying patients who could be at higher risk, co-development of companion MIPD tools during drug development to accelerate the generation of evidence required for broader clinical implementation of MIPD, and a report of the first Asian Symposium on Precision Dosing.4,5,6
To learn more about how model-based approaches can improve precision dosing in clinical care, watch my webinar.
- Polasek TM, Shakib S, & Rostami-Hodjegan A. (2018). Precision dosing in clinical medicine: Present and future. Expert Rev. Clin. Pharmacol. 11(8), 743–746.
- Polasek TM, Tucker GT, Sorich MJ, Wiese MD, Mohan T, Rostami-Hodjegan A, Korprasertthaworn P, Perera V, & Rowland A. (2018). Prediction of olanzapine exposure in individual patients using physiologically-based pharmacokinetic modelling and simulation. Br J Clin Pharmacol84(3), 462–476
- Polasek, T.M., Ambler, K., Scott, H.S., et al. (2017). Targeted pharmacotherapy after somatic cancer and mutation screening. F1000 Research 5, 1551-1555.
- Patel N, Wiśniowska B, Jamei M, & Polak S. (2018). Real patient and its virtual twin: Application of quantitative systems toxicology modelling in the cardiac safety assessment of citalopram. AAPS J 20(6), 1-10.
- Polasek,T.M., Rayner, C.R., Peck, R.W., Rowland, A, Kimko, H., & Rostami-Hodjegan, A. (2018) Toward Dynamic Prescribing Information: Codevelopment of Companion Model-Informed Precision Dosing Tools in Drug Development. Clinical Pharmacology in Drug Development. 8(4), 418-425.
- Thomas M. Polasek, Amin Rostami-Hodjegan, Dong-Seok Yim, Masoud Jamei, Howard Lee, Holly Kimko, Jae Kyoung Kim, Phuong Thi Thu Nguyen, Adam S. Darwich, and Jae-Gook Shin (2019). What does it take to make Model-informed Precision Dosing common practice? Report from the First Asian Symposium on Precision Dosing. The AAPS Journal 21(2), 17.