Learn & Confirm
The pharmaceutical industry has its share of problems —patent expirations, pressures on margins, regulatory creep (or perhaps more accurately "regulatory pounce"), battered reputation, and the like. But behind them all is One Big Problem. It takes too long and costs too much to develop new products. And so the search is on, both at drug manufacturers and FDA, for insights that will shave time and costs from drug development. Progress is already being made. Average development time, by at least one report, is down slightly in recent years. And a recent study from the Tufts Center for the Study of Drug Development noted that the tier of top performers had shortened development time by about 20 percent in the past five years, while holding regulatory time steady.
Perhaps the boldest and most interesting model to emerge in the past few years is being put in place today at Wyeth, where a year-long program of research, discussion, and planning is culminating in radical change to the way the company carries out development, including:
» A complete reorganization of the R&D function that puts control of drugs—and therapeutic-area portfolios—in the hands of project teams
» Aggressive cost-cutting initiatives that aim to save both money and time
» A shift away from traditional Phase I through III development and toward seamless use of adaptive trials
» Most important, a new approach to drug development, called Learn and Confirm, that Wyeth thinks will provide better data, improve late-phase success rates, and deepen insight into how drugs actually work in diverse patient populations.
Wyeth's new approach to clinical research originally grew out of a broader, corporation-wide initiative called Springboard, which began a year and a half ago as a search for ways to be effective within a lower-cost structure. "Initially R&D wasn't even part of the thought process," explains Bruce Schneider, Wyeth Research's executive vice president and chief of operations. "But [Wyeth Research president] Bob Ruffolo volunteered that we should be challenging ourselves to see what we could do to do things more effectively in the future.
"Initially, cost was the emphasis, but as we got into it, we also talked about ways to radically change the way that we actually developed drugs to see if we could somehow collapse phases, eliminate phases, that sort of thing. We had a strategic perspective on the one hand and an operational perspective on the other. So we tried to hit both of those at the same time."
The R&D Springboard effort started with a half dozen participants who referred to themselves as the "wacky thinker" group. One of them was Charles Gombar, vice president for product management. "We really looked for people who were not going to give us the same old, 'This is how you do pharmaceutical development, dah-dah-dah,' but who could really think of different ways of doing things," he says. "We sat around and we asked a lot of fundamental questions. We were willing to question everything.
As part of the process, Wyeth executives talked with companies in other industries that faced similar challenges. From electronics companies, they heard about rigorous focus on cost and cost-optimization tools. Product development companies explained the art of encouraging experimentation and "early kill" decision-making. Global supply and logistics companies provided a view of the culture of process engineering. But perhaps the most productive discussions, with a leading company in the aerospace industry, at first seemed the least relevant.
"I was astounded at the analogies between what they do and what we do," Gombar says. "They're highly technically oriented, highly regulated, with long development times. Their full portfolio of projects was something like 2,000 projects. But if you did a cut of their $500-million-and-up projects, it was about 30 projects—so the size of their portfolio wasn't very different than the way we do things."
What Wyeth's team carried away from the interview was the idea of project-driven teams. "That industry went through a lot of turmoil and consolidation 15 years ago," says Gombar. "At the time, they structured their development teams the way most pharmaceutical companies do—highly matrixed, everybody belongs to the aligned function. They decided to change that. They made it project-driven instead of function-driven. For major projects, the project leader is king. The project leader has the budget, the people on the project report to the project leader, and so on." That concept went on to become a centerpiece of Wyeth's new approach to R&D.
The team talked with hundreds of people inside and outside Wyeth, collected several hundred new ideas, then gradually winnowed them down to 45. Some ambitious ideas were left by the roadside, such as a plan to move to totally paperless clinical trials. ("That's going to happen someday," says Gombar. "But not today.") The final list of 14 included:
» creation of Early Clinical Development Centers (ECDC) to speed recruitment for Phase II trials
» development of a cost-optimization tool for global trials
» outsourcing of logistics for clinical trial materials
» reorganization of R&D into project teams that control not just a single candidate, but an entire therapeutic portfolio. (More about these initiatives later.)
But from the process of creating and aligning these initiatives, some big questions emerged. The team members found themselves rethinking not just their own company's structures and procedures, but the basics of how clinical research is conducted across the industry.
"We wanted to see if we could push the system and the regulators, in particular, to see if we could generate some thinking around doing things more efficiently," says Bruce Schneider. "One idea that came out of this was collapsing the phases of development—to see whether you could collapse Phase I and II somehow and just start all your trials on patients, not volunteers. Or could you collapse Phase II and III together, which would be the big win for everybody?"
"People treat phases as natural law," says Gombar. "But in fact, they're nowhere in the regs. There are definitions. But nowhere did anyone ever say, 'You have to do Phase I, then you have to do Phase II.' It was just a convenient categorization. I want to purge the world of Phase I, II, and III. It's one of my goals in life."
What is wrong with traditional phased development? For one thing, it's inefficient. That's partly because traditional blinded study design prevents researchers from ignoring signals that might emerge early in the process; partly because the phase structure works against overlapping research; and partly because the gaps between phases tend to be unstructured and unproductive.
But there are larger problems with traditional phased research. Because completion of each phase becomes a regulatory hurdle, there's an emphasis on simply getting through the phases. And that means companies don't always learn as much about their products as they should.
FDA, of course, has been working on the efficiency and effectiveness of clinical trials from its own perspective. During conversations with the likes of Janet Woodcock and Larry Lesko, from FDA's Center for Drug Evaluation and Research, the Wyeth team heard some terminology that struck a responsive chord: "At some point in that conversation, they started talking about Learn and Confirm," remembers Gombar. "We hadn't seen it before, but thought, 'Wow, that's sweet. It's instantly understandable.' It started to grow roots with us very quickly."
It turned out that the terminology wasn't unique to FDA. It actually came from the work of the late Lewis Sheiner, MD, a professor at the University of California, San Francisco School of Medicine, who introduced the concept in a 1997 article published in the journal Clinical Pharmacology & Therapeutics.
The article elegantly breaks drug development into two goals: learning things one needs to know about the drug, and confirming that knowledge with testing in a representative patient population. But the two goals often fall into conflict, Sheiner argues. "Learning and confirming are quite distinct activities, implying different goals, different study designs, and different analysis modes. The understandable focus of commercial drug development on confirmation," Sheiner wrote, "as this immediately precedes and justifies regulatory approval, has led, in my view, to a parallel intellectual focus that slights learning. The predictable result...is that clinical drug development is often inefficient and inadequate."
Missing Dose-Response Curve
Inadequate in what? Gombar is quick to point to a vital, elementary set of information that often fails to emerge from clinical research: The dose-response curve. "I'm an old pharmacologist by training," Gombar says. " In the lab, I would never ever, ever, if I'm doing a dose-response curve on a new drug, pick three doses and study it. Yet, how are 99 percent of Phase II studies done? Three doses and a placebo. When you have no idea what dosage you should even be using.
"We're not talking rocket science. I don't want you to do something that no one in the world has never ever done before. I want to know what the dose-response curve looks like. Before I go to Phase III or what we now call 'confirm,' you damned well better show me what the dose-response curve looks like."
Sheiner in his article goes beyond the dose-response curve. He talks about the "dose-response surface," a three-dimensional plot that represents not just averages, but differences in individual patients, their prognoses, and their personal responses to medications. And part of the power of the Learn and Confirm model is that it focuses the researcher on learning sources of variability early in development.
"When I went through this whole thing last year, and especially some of the discussions we had with the guys down at FDA, I finally realized that I was thinking about drug development totally backwards," says Gombar. "We'd start with a homogeneous population to control variability as much as possible so we got good data with a somewhat smaller number of patients, then we went more real world in Phase III.
"They got me to realize that's the wrong way to think about it. Actually you want to use heterogeneous populations early on, so you identify sources of variability—so that you can perhaps identify what patients are going to respond, and go to a more homogeneous population in Phase III with some confidence that these are the patients that it's actually appropriate to use the drug in."
A concept like Learn and Confirm needs to be more than just an idea—it has to be built into the structure of the R&D operation. Toward that end, Wyeth has created a new team structure for R&D. Formerly, individual drug candidates were assigned to project teams that handled both what would now be called Learn activities and Confirm activities. Now, Learn teams of a dozen people are assigned not to individual drugs, but to the portfolio for a whole therapeutic area.
"Take Alzheimer's for example. We happen to have a lot of drugs in Alzheimer's development right now—maybe seven or eight molecules," explains Schneider. "All of those are being overseen by a single team. We're trying to bring the people who have the most expertise and scientific knowledge into those teams, with the idea of learning as much as we can and being able to repeat learning from one molecule to the next."
Once a molecule gets to the Confirm phase—which for the moment will be equivalent to Phase III—it will be handled by a Confirm team focused on operations and rapid execution.
The key to making this new organization work, however, is creating a structured dialogue between Learn teams and Confirm teams to help them make practical decisions about how much learning is enough, and what, if any, additional practical steps need to be taken before a drug can move forward.
"That's a discussion that didn't happen five years ago at all," says Stiles. "It was just this finger pointing: 'The drug's not good enough.' 'Yes, it is. You're going to take it.' We haven't fixed it all, but we're getting there."
Where's the Saving?
There is little doubt that Learn and Confirm can produce better, more useful, clinical information that would serve the interests of patients, physicians, and marketers alike. But in a time of massive patent expirations and pressures on margins, the question remains: Will it save money?
There are reasons to be optimistic. Today, far too many drugs fail in Phase III because of inadequate data on dosing. If Learn and Confirm is only moderately successful in improving the success rate (or, better yet, brings it back to the historical rate of 80 percent or so, which has seemed like a pipe dream in recent years), it would be worth the effort.
The new system also promises the potential to save development time, which pays off both in lower costs and a longer period of time selling on patent. There could well be additional savings just from collapsing Phase I and II together. But the real savings, says Schneider, will come when regulatory agencies are willing to let the company go all the way to a seamless Phase II/Phase III approach. "Right now, it's sequential," he explains. "You do Phase II. Say it takes a couple of years. Even if you do interim analyses along the way, you have a transition zone, at least two and a half years, and then you have Phase III.
"If you combine those, that would be a huge saving," Schneider goes on. "If you collapse it all into a single sort of protocol concept—maybe dueling adaptive trials—you greatly collapse the time. That to me is the big win, if we can ever get ourselves to that point."
Getting to that point depends to a great extent on the comfort level of regulators, and they haven't shown themselves ready yet. But Schneider wonders whether there aren't other possibilities that would ease regulators into a new world. "Is there a halfway point that would allow us to do one adaptive, combined, seamless Phase I/II/III protocol, then somewhere along the way, you kick off a single confirmatory Phase III trial without waiting until the end? Maybe it's a little bit staggered, but you're still saving some time there and you're not creating two confirmatory studies. You're focusing on one."
Still, there is reason to fear that Learn and Confirm will lengthen the old Phases I and II rather than shorten them. As Sheiner pointed out in his original article, learning is sequential, and we don't always control what we learn and when.
"Given the pressures for greater short-term rather than long-term thinking in our society in general and in the pharmaceutical industry in particular," he wrote, "the specter, even if unjustified, of longer total development time associated with sequential science based development may prove a formidable barrier to its acceptance. If so, then it will fall to a few brave (or foolhardy) research and development managers to judiciously choose from promising examples to demonstrate the value of fully science-oriented development."
Gombar agrees, but he takes the argument a step further: Managers need to seek efficiencies anywhere they can to make it possible to pursue a Learn and Confirm strategy. The long-term benefits are just too great to pass up.
More Springboard Initiatives
The search for new efficiencies makes Wyeth's other Springboard initiatives especially important. If they succeed, they will allow the company to afford the added work of acquiring more information about its drugs—without increasing overall costs.
Early Clinical Development Centers The Springboard team took an especially close look at recruitment for Phase II trials. There's a double-whammy to slow enrollment at this phase: On the one hand, it slows down the trial. On the other, it encourages companies to cope with slow enrollment by adding multiple additional research sites, many with very few patients, and accrue additional logistical costs for identifying sites, making qualifying visits, filing IRB paperwork, and monitoring performance.
What's more, explains Robert Maguire, vice president and chief of operations for clinical R&D, is the ample amount of waste built into the system. During a typical year, the company uses 400 to 500 sites for Phase II research—and 30 to 40 percent of them accrue either no patients or only one.
The team asked a question that hadn't really been asked before: Could they dramatically reduce the number of sites involved in a study while maintaining quality—and perhaps reducing cost? The company already did something similar with Phase I trials, which were mostly run out of large centers with cohorts of patients already lined up as volunteers. Could they build an equivalent mechanism for Phase II?
"One thing we saw over and over again was that 80 percent of our patients came from 20 percent of our sites," says Sheila Ronkin, assistant vice president for clinical development, who heads the ECDC program. "And we found that there were steps to take that could anticipate how they were going to enroll and conduct the study."
Was it possible, the team asked, to conduct Phase II trials in which 80 percent of the patients would come from a very limited number of sites—say, 10 or 20? They tried the process in Latin America, and received some encouraging evidence. In one trial, reports Ronkin, two Latin American hospitals provided 1,000 patients, about one-third for the entire trial.
As a result, the team came up with the idea of Early Clinical Development Centers—major hospitals, internationally sited, with which the company would create formal relationships for performing early-phase trials. The target was to find about 15 high-volume hospitals in the United States, Western and Eastern Europe, India, China, Hong Kong, and Latin America. Each site is required to have experience in electronic data capture (Wyeth uses EDC in 95 percent of new studies), and sites are audited in advance to identify weaknesses and training needs.
The expectation is that in a typical Phase II trial of 200 to 300 patients, five or six sites will provide 40 to 60 patients per study. Given the size of the hospitals Wyeth is currently negotiating with (which had not been made public at press time), that seems reasonable: One, for example, sees 9,000 outpatients a day. Another has 30,000 new cancer patients a year.
Outsourcing clinical logistics Stiles estimates that Wyeth is involved in clinical trials with 45,000 patients at 5,000 sites around the world. And that represents a logistical nightmare. "We have to get the drugs out to the sites. We have to get out all of the paperwork. We have to have blood drawing equipment and EKG machines if that's necessary. And then we have to get them back. After a study closes, you have to account for every one of your test article drugs. Yet, that is not a core competency, certainly not a physician's and not our team's."
The expertise is readily available, however, at large global logistics companies like Federal Express. And Wyeth is currently negotiating with an unnamed company to take over its entire clinical supply logistics function. "We're almost finished," says Stiles. "Their computer systems will tie into ours because we have to make the drug and keep it in our warehouse, but then they'll be responsible around the world. It'll be just like when you order a book on Amazon.com You enter it in and it says, 'OK, it has left Seattle. It will be in Philadelphia tomorrow morning.' We can do that."
24x7 operations "We don't believe in working 12-hour days," says Stiles. "Yet, we're not interested in outsourcing everything that we do." And so Wyeth's R&D operation is dispersing its centers worldwide to allow for an around-the-clock pass-off of work.
"We will have a site in Mumbai with 400 Wyeth people working on data management, statistics, writing," Stiles explains. We're not sending them work that we don't want or cleanup work. They are part of our teams, and likely we'll have another one of these sites, perhaps in Eastern Europe."
Cost-optimization tool A major cost of clinical trials comes in the form of lab tests, x-rays, and purchases of various sorts—all of which vary considerably by geographical location. One key Springboard initiative is to develop a tool that tracks and compares these costs. "Let's say we need a MRI of the brain for Alzheimer's disease every three months for two years, and we want to do this study in Buenos Aires and Paris," Stiles says. "We'll have a system that tells us what's the cost range at different sites for different vendors so the people writing the protocol can say, 'OK, I can get this for this price. The tradeoff of doing it over here for that price is the following and build it in.'
"We're beginning to use that right now as we speak. We think we can drive an awful lot of cost out. It also helps us to be better negotiators when we're going after pricing. We can look at it and say, 'Sorry, but we can do better than that.'"
Data standards For Stiles, perhaps the most exciting initiative has to do with standardizing safety data. "All drug companies tend to be compartmentalized by therapeutic areas, and how they do their studies is very team-dependent," he says. "Each team devises how they want to gather the data, create their statistical data packages, analyze the data, and do their own tables. Each one of those requires the statisticians and programmers to create their own statistical programming model."
But many data can be standardized: demographic data on patients, their conditions, what drugs they are taking, and so forth. On the safety side, in particular, it is possible to standardize data collection and analysis. "It doesn't matter whether you're collecting it on somebody with Alzheimer's disease or breast cancer," says Stiles. "The safety data are one and the same."
The new project aims at a deep level of standardization. "It starts with the simplest things and then goes down the line: tagging the data, how it will be analyzed, how it'll be pulled out, how it will be stored, and get to a package," Stiles says. "The agencies we've talked to about this are absolutely thrilled because when they get a data package from Wyeth for safety, they'll know what it will look like and where to look for what they want."
So Far, So Good
The transformation of Wyeth's R&D operation is a work in progress, and will remain one for a good time to come. Implementation began in January, under the guidance of Matt Bell, senior director of the Learn and Confirm implementation office. In general, he says, the projects are on track and on schedule. "We're exactly where we expected to be right now," he says. "So far, so good."
» The Learn teams were put through a pilot program involving oncology, respiratory diseases, and a high-impact alliance team working on methylnaltrexone, which Wyeth is developing with Progenics for the treatment of opioid-induced bowel dysfunctions. Based on that experience, Learn teams were rolled out across clinical areas in June
» The ECDC team is on track to get 10 Phase II centers up and running by the end of the year
» A contract for logistics management is in negotiation, but Bell expects that full implementation of the complex program will take several years.
For the next six months, Bell says, the goal will be to see how the program works when applied to the entire pipeline. "We're at the start of the journey. And one of the things that is going to help us manage is Wyeth's focus on metrics—understanding the outcome at the team level in terms of time saved or quality improved or value added and then holding the teams to that. Not just saying you're doing something, but actually measuring it.
"Through the rest of this year and next year the organization should, in theory, start to see improvements," Bell says. "And if you don't see things changing, then you go back and ask the question, 'Why are we not seeing any changes?'"
In truth, Wyeth's R&D leaders seem to be expecting more than measurable change—they seem to be expecting transformative change. A year or so of asking unaskable questions and poking at the very foundations of R&D seems to have created a sense that R&D could be a very different thing than it is today.
And there is an impatience to get there. "Why can't we cut drug development time in half?" asks Schneider. "As you go from Phase I to II, II to III, you actually have some blank space—some transition zones you have to go through. If you can eliminate those, you save time right there. If you can collapse them in such a way that you don't have to repeat the whole phase or do it all separately, you can save even more time. So why can't you cut it in half?"
How Digital Medicine Can Pinpoint Dosing Regimens to Optimize Drug Efficacy and Safety