May 15, 2007 | Michael Krams spent 10 years at Pfizer, designing clinical trials that “learn” from experience as they go, building algorithms to enable real-time learning. His expertise has propelled him to co-chair the PhRMA working group on adaptive designs. As assistant vice president of adaptive trials at Wyeth Research, Krams heads the “Learn and Confirm” Adaptive Trials team, advising across therapeutic areas on trial simulations and flexible trial designs.
Bio•IT World’s Editor-in-Chief Kevin Davies asked Krams to provide an update on adaptive trial design at Wyeth and the industry as a whole.
Q: Tell us about your background before joining Wyeth.
KRAMS: I’m a neurologist by training. I spent formative years in research doing functional brain imaging. We looked at some very rare diseases, and patients were very, very precious. We needed to make sure that our experiments did not waste patients. At Pfizer, I approached clinical trials with this same sensibility: patients are precious and should not be wasted.
At Wyeth, the Learn & Confirm idea implemented by Dr. Charlie Gombar and Dr. Evan Loh is an extension of this basic sensibility (see “Ten Years After: Learn and Confirm”, Bio•IT World, February 2007). The adaptive trial initiative is really just one of the many enablers for the bigger Learn & Confirm effort.
How did adaptive trials get started?
KRAMS: Choosing the correct dose to take into phase III is a key decision problem in drug development. When we were developing acute stroke therapies at Pfizer, we wanted to make sure that we had the most efficient approach to establishing (a) whether the drug had legs to run, and if yes, (b) at which dose we should confirm its action in phase III. We developed the design for the ASTIN stroke study, conducted in 2000 — one of the early adaptive dose-response finding studies, learning and adapting in real time. The methodology clearly provided both scientific and business value.
In clinical drug development, the first notion of adaptive designs goes back to the 1980s, with publications on continuous reassessment methods in Phase I. Also in the 1980s, the concept of group sequential design was developed to provide an opportunity for increased efficiency of learning about efficacy, where appropriate. So although currently adaptive designs are much in the spotlight, the concept has been around for a while.
What is the relationship between your adaptive trial initiatives and Wyeth’s Learn & Confirm strategy?
KRAMS: The adaptive trial is one enabler of the Learn & Confirm paradigm. It acts as an integrator of many different approaches including modeling and simulation, translational medicine and the development of biomarkers, thinking about efficient design from a statistical perspective, and advances in capturing data in real time and making decisions based on that data.
What determines whether you will employ an adaptive trial design for a new drug candidate?
KRAMS: We have a well-structured process called opportunity review... We go through the portfolio, and wherever there appears to be an opportunity to deploy our thinking, we engage in a discussion with the Learn teams.
There are three steps. First, we put a concept sheet together, where we assemble initial information about the decision problems we have to work on. Then we develop different scenarios that might address these problems and translate them into research questions. We collect relevant information about endpoints, biomarkers, patient population, decision rules etc. We call that a scenario analysis. We compare different options and potential approaches to answer the research question.
Then we assess the value of each of these scenarios in terms of three dimensions: First, what is the information value — how much information do we get out of a given amount of resources? Second, how impactful is this design on the business case? Ideally, we want to work as efficiently as possible in terms of dollars, resources, and getting the correct decision made at the earliest time.
The third logistical aspect is, what are the implications of doing what we want to do on drug supply management? How can we ensure that from this perspective, we can optimize the approach we want to argue for?
So what are your options then?
KRAMS: So at the end of this scenario analysis, we have different options, one of which may well be the traditional approach. We can compare whether another scenario may have advantages against that traditional approach or not. If there are advantages, they will be quantified in terms of information value. For instance, what is the probability of identifying the correct dose to take into phase III in scenario A, B, and C? Do we have a more efficient way of getting the answer? Do we get the answer earlier perhaps?
The final step is to translate this scenario analysis into a simulation exercise, where we take the scenarios most acceptable to the Learn team and run large-scale simulations against it. That simulation report is the basis for the final decision made by our governance body.
Are you the chief architect of these simulation studies?
KRAMS: In any engineering environment the actual build of a system is preceded by computer simulation — think of cars, planes, etc. Simulation-guided clinical trial design was established many years ago by Lew Sheiner, Carl Peck, Don Rubin and others. My mentor has been and is Don Berry — he heads the Biostatistics unit at M.D. Anderson, and works with Peter Mueller, (his son) Scott Berry, and Peter Thall. The adaptive trials as we now build them are the product of a joint effort of many: statisticians, clinicians, programmers... The architects? Our statisticians are the architect.
What kind of drug candidate or therapeutic area is best suited for adaptive trial design?
KRAMS: There needs to be a research question with sufficient uncertainty around it. If we already know the answer and just have to confirm it, we can build a very efficient traditional design. But if we are not quite sure what dose to take into Phase III or which subset of the population would benefit most, then it pays off to run an experiment to hone in where the answer lies...
In many drug development programs, there is insufficient effort in exploring the dose response relationship... We’ve taken the decision to initially apply the concept of adaptive trials to the Learn paradigm, particularly around exploring the dose response. The hierarchy of questions is: first, do we have a viable product? We do want to know whether we have proof of pharmacology, proof of concept. If so, then second, what is the correct dose to take into Phase III?
At what point in development should adaptive trial designs be deployed?
KRAMS: Adaptive trial designs can be deployed from the very beginning of clinical drug development: Phase I — continuous reassessment; across phase II — response adaptive dose-ranging studies; to confirmatory phase III trials — e.g. blinded sample size re-estimation, group sequential designs, maybe even seamless phase II/III.
How do adaptive trials help stratify patients based on genotyping or bio-markers?
KRAMS: There are two approaches: one tries to enrich the population as much as possible and therefore only works with a small subset of possible patients. One can use genetic or other information to enrich the population.
An alternative approach is to take a very wide population and identify which of the many subsets of the bigger population react best. So the adaptiveness depends very much on the decision problem. If the problem is to identify an efficacy signal for a drug, then the trial needs to work with an enriched population. If the decision problem is to design a subset of patients best suited to that drug, then you’d want to start with a very broad population. Adaptation is not conditional on which of these two approaches you choose.
How are adaptive designs being applied at Wyeth?
KRAMS: Wyeth plans to launch start at least six adaptive studies this year, maybe more. There are several dozen projects in the works. One project we recently completed used adaptive designs to shave months and millions of dollars off the cost of developing a new drug.
Where is the chief ROI with adaptive trials?
KRAMS: The most important saving comes from the higher confidence in having the correct answer. When you have 50% of trials failing in Phase III, you have a problem. The effort to identify biomarkers and the effort to design an adaptive trial go hand in hand. The requirements for biomarkers allow us earlier readout, which has predictive value for the clinical endpoint we have to take to regulators.
How would you characterize the pharma industry’s overall take on adaptive trials circa 2007?
KRAMS: Ten years ago I would have said it was a fringe thing. Today, there is considerable interest across the industry, but the industry is divided into three parts: those who will never touch it; those who want to sit back and observe; and those who are ready to bite the bullet.
Wyeth is clearly at the forefront of adaptive trial development. Novartis is very engaged, so are Pfizer, GSK, and Lilly. But I think we are clearly — and have the ambition to be — at the forefront of this development. This is but one piece in the puzzle of “Learn & Confirm.” If you want to produce a good dish, it’s not enough to have only one good ingredient. Adaptive designs are but one of many enablers for the big “Learn & Confirm” effort.
Where are there opportunities for third party software developers/collaborators?
KRAMS: We invite all statistics geeks (and creative coders) to help us redesign the User-Interface of the adaptive clinical trial. It would be fantastic to open up the design process to young programmers, in much the same way that Google has given programmers “keys” they can use to design their own ‘search engines.’ It would be fantastic to have an entire toolbox of algorithms that could be used to engineer new design interfaces.
Pharmaceutical companies don’t have to be software developers, all they have to do is share their challenges with an enthusiastic community of coders, problem solvers, inventors. Not to outsource but to in-source. To recognize that the universe of design is inclusive.
Your final thoughts?
KRAMS: We are doing adaptive designs! It’s happening, big time. Adaptive trials in “Learn” clearly are in line with the philosophy of the FDA’s critical path initiative. It’s fun to plan and implement these designs and observe the impact on the efficiency of the clinical drug development process. And it’s fun to be part of a Ferrari team. We can already feel the impact. Just imagine when we will have a dozen case studies or more under our belt. I am convinced that we will be able to show how these designs added value to what we do. And it will be great to share how we managed to deal with the challenges on our way. Challenges are there to be overcome.
Subscribe to Bio-IT World magazine.