EVENTS | VIEW CALENDAR
Optimized dosing: We can do better, but when?
It is widely agreed that there are limitations in using frequentist statistics to conclude what should be the right drug at the right dose at the right time for any individual patient. Since the 1990s, there has been optimism for pharmacogenomics aiding in these choices.
At the same time, practical guidance based on genome information has been very disappointing and thus far very limited. A patient’s individual phenotype as revealed by their lifestyle choices, gender, age, comorbidities, individual organ performance, microbiome and co-administered drugs can be even more impactful. Diurnal variations in metabolism are typically ignored.
To make matters worse, many drugs induce the enzymes which metabolize them and the transporters which move them about, changing the optimum dose over time (e.g., months). Ideally, a clinical decision support (CDS) system would integrate all of these, but even a few can help at the subpopulation level.
The only way to include all of them together is with data. A recent FDA workshop on Aug. 12 had the title “Precision Dosing: Defining the Need and Approaches to Deliver Individualized Drug Dosing in the Real-World Setting.” I had to be there.
Today, the notion of companion diagnostics to improve drug efficacy and safety is well advanced. Preclinical modeling from research models to first human dose is also getting much attention, along with adaptive clinical trials. Nevertheless, while choices for subpopulations are at hand, optimizing dosing for an individual is an aspirational goal and not one close at hand.
An elephant in the room is the many drugs approved at an earlier time when science was less advanced.
Dosing of many of these drugs can no doubt be optimized with modern tools. The incentives for funding the real-world clinical pharmacology research and practice are not evident. Many of these drugs have a narrow therapeutic index and are subject to drug-drug interactions that can alter blood concentrations manyfold. Today, in-vitro absorption, distribution, metabolism and excretion work with known metabolizing enzymes and drug-transporting enzymes enables forecasting some interactions, but not quantitatively in an individual patient. FDA, NIH, pharma and in-vitro diagnostics firms are neither motivated nor adequately resourced.
When a chemist or cook prepares a solution of known concentration, we first want to know the volume. In animals or humans, there is a tradition of using the mass of the subject as a surrogate for blood volume. Traditional doses are referred to as mg/kg, or in the jargon “migs per kig.” For decades, there has been no other choice, given that blood volume was not precisely known for an individual. This remains the case today. To further this point, when we prepare a dosing solution for a rat, we confirm the concentration with analytical chemistry. When we prepare a drug in an IV bag to be infused in a patient, we will not confirm the concentration very often. Only rarely is the circulating drug concentration determined in a timely manner even one time, if at all.
Given the recognition of the problem, it is helpful to contrast the notion of patient exposure to a drug vs. the dose of the drug. Exposure is one step closer to the site of drug action. Given that time matters and concentration is dynamic, exposure may be expressed as “area under the curve” (AUC), meaning the concentration vs. time curve. This is available in animals and only in the early phases of drug trials. Given the lack of data for individual patients, mathematical models are beginning to be developed based on population data. Several young companies are offering this CDS software as an augmentation to electronic health record (EHR) systems. They are based on rather sparse data, but offer a real advance that benefits care.
To improve CDS with respect to dosing, more population and then individual data vs. time is needed to allow the model(s) to learn collectively and better apply individually. Artificial intelligence purports to be able to do this, but not without data. Searching EHRs is unhelpful; the data are not there. AI is cool, but is hungry for valid data.
Exposure data requires:
While it appears very difficult to revisit older drugs from the regulatory and financial perspectives, the opportunity for better care is compelling. Serious consideration is needed to develop a financial model to enhance optimized use for all drugs in critical cases. Some of this has been supported by NIH for pediatrics, but there is not a good home for advancing clinical pharmacology at the pace needed and now feasible. I’ve asked what pharmacology model was used to arrive at an 81 mg aspirin tablet for all? That model was dividing 325 mg by four and rounding to the nearest mg. Most readers of this column will be gambling on one of these tablets per day with little justification. We can do much better than that!
Peter T. Kissinger (who can be reached at firstname.lastname@example.org) is a professor emeritus at Purdue University, founder of BASi, chairman of Phlebotics and director of both Prosolia and Tymora.
The FDA workshop contents, including slide decks and compelling talks are available here.
Kudos to FDA for their attention to this very translational topic. The patients are waiting and most don’t know why. More attention is paid to repurposing older drugs than to judiciously using them as originally intended.