YouTube Facebook LinkedIn Google+ Twitter Xinginstagram rss  

Change Agents for Adaptive Designs

By Kevin Davies

May 15, 2007 | Twenty-seven years ago, Kevin Gell left IBM to found Tessella, a software consultancy based near Oxford, England. For almost a decade, the company has been working with various pharma clients on adaptive trial designs (See “Real-Time Trials,Bio•IT World, June 2006). Despite working with Pfizer on the  seminal ASTIN (Acute Stroke Therapy by Inhibition of Neutrophils) trial, Tessella’s adaptive activities drew little interest — but that is changing fast. Now, Tessella finds itself fending calls from various biopharmas eager to follow the likes of Wyeth (see “Biting the Adaptive Trials Bullet,” p. 35) and Novartis and learn more about the upside of adaptive trials.

Grant Stephen is the Scottish CEO of Tessella’s U.S. division, and also responsible for the adaptive trials business: “We feel like we’ve been in the wilderness preaching this stuff. Suddenly it’s turned around; there’s all this traction. Five years ago, we couldn’t figure out why everyone wasn’t jumping on this. [But] in the last year or two, there’s been this big uptick in interest. Before, we were saying, ‘Big pharma, talk to us’. Suddenly, this last year, they’re phoning us, they’re not just returning our calls. At last!”

“The last year’s been immense, it’s such an exciting field to be in,” agrees Tom Parke, who has managed Tessella’s adaptive trials program for the past nine years. Prior to that, he managed the development of an air traffic control system. Stephen sees a common theme there — saving lives.

Tessella’s 200 employees include physicists, chemists, mathematicians, and statisticians, most with strong software capabilities. “Tessella’s approach is to be a one-stop shop for pharma to get their head around this problem,” says Stephen. He says Tessella is working on adaptive designs with four top ten big pharmas including Wyeth.

“There are three drivers,” says Parke, explaining why pharmas hope adaptive designs will improve the failure rate of drugs in phase III. “First, fewer subjects are exposed to ineffective doses. Second, they’re better at killing ineffective compounds. They also offer hope that by characterizing the behavior of a drug in humans better in phase II, a better choice of treatment regime in phase III can be made.” Many conventional trials try too few doses in phase II, or the design has the wrong focus.

Tessella and Parke started designing software for Pfizer’s ASTIN trial in 1998, and worked on it for about five years. The trial resulted in prematurely terminating a drug trial, saving Pfizer an unspecified but evidently nontrivial sum of money.

“ASTIN was a groundbreaking design,” says Parke, “but in many ways so specific to that trial that people couldn’t see how it could be used on their [own] trial.” Each clinical trial brings a unique set of issues, as diseases and drug behavior vary profoundly. Parke also notes that electronic data capture (EDC) — a prerequisite for adaptive designs — has been a big initiative for virtually all big pharmas in recent years. The deployment of EDC across much of the industry is an important enabling step for adaptive trials.

But the slow uptake is about culture as well as technology. “Statisticians are not usually change agents,” says Stephen. “They’re not used to going out and making sweeping organizational changes. The requirements are quite different from conventional trials.”

ASTIN “really woke people up, but it was so ambitious, the rest of the world wasn’t ready for it,” Stephen continues. “Big pharma spent millions of dollars building its [trial] infrastructure, and suddenly new techniques were brought in. It requires significant change agents to change all this machinery. That delayed uptake in adaptive techniques.”

Adopting Adaptive
Pharmas pondering adaptive designs must define exactly what they want to get out of the trial, says Parke. “Because you’re going to build a model, you need to define in concrete terms what you’re after.” That typically includes the number of doses, the size of the effect, acceptable levels of toxicity, and desired confidence limits.

Parke says there are six basic types of adaptive trials in practice, with four being in common usage. In addition to dose response trials, the “seamless design” trial combines Phase II/III in one trial, allowing companies to choose which arm of the trial moves into Phase III. The design ensures that the sample size is appropriate for the remainder of the trial. “We’ve been working on a software package to help pharma companies evaluate this method, run simulations, and see how beneficial it is,” says Parke. (The toolkit just won a coveted Bio•IT WorldBest of Show” Award.)

Tessella’s role in planning an adaptive design trial often begins by working with the statistical designers of the trial, usually third party consultants like field-leader Berry Consultants, although Parke notes some big pharmas “have strong internal statisticians who do the designs themselves.” The designs are richer than conventional designs, with more parameters. Conventional trial designs are fairly straightforward, says Parke: “I want this risk of a false positive, this risk of a false negative, this is how noisy I believe this signal is — you pop it into a formula and it tells you how many subjects you need.”

But for adaptive trials, “You can’t do that,” says Parke, “because we’re doing so much more in the trial in terms of modeling the data. The only way to get a handle on it is to run simulations.” This is often Tessella’s entry point, typically taking 1-2 people 2-4 weeks. The simulations themselves are usually run on “just a few high-powered laptops.” As adaptive trials become more commonplace, companies might consider dedicated computers. 

“We got this down fairly pat,” says Parke. “We’ve done this several times with various statistical models (both Bayesian and frequentist).” Although all trials differ, Parke says, “We do have libraries of software where we use chunks of code.” The simulations estimate the “operating characteristics of the trial,” allowing pharma biostatisticians to judge the number of subjects required, the likelihood of stopping early or finding the right dose, manufacturing considerations, and more.

The next problem, Parke says, “is enabling the change in the randomization regime to be admitted in the trial. The statistical model is taken out of the simulation engine and has to be embedded with the IVRS system and interface with the EDC system.” On a roughly weekly basis, the program creates a new randomization list for new patients.

In all, it might take 2-3 staff a couple of months to produce a complete clinical trial randomization system. “It allows the client to prove the benefits for an adaptive approach without spending millions messing around with a big infrastructure,” says Stephen. This proof-of-concept allows clinicians to go to management and say, ‘Look at the benefits we’ve proven here.’ Stephen says one top ten big pharma went this route to dip its foot in the adaptive waters.

 Click for larger version.
In all but one case, Parke says Tessella’s clients have been “extremely happy with the result and speed of the result. In retrospect, [even] ASTIN could have been improved.” The exception? A client “ran a conventional trial with our system running on the side... the problem was the EDC system didn’t collect the data quickly.... We realized, if all the data had been sent in when it should have been, it could have been stopped three months earlier!”

However, when asked to give specific examples of a drug brought to market using novel adaptive techniques, Stephen demures: “It’s too early. The clinical evaluation process takes years. We still have to do phase III.” After further thought, he says there are cases where “clients have saved millions of dollars.” In addition, time savings in Phase IIb have been “of the order of 3-6 months.”

Stephen says: “We’d love to do this in Phase III — that’s one of the things the industry’s waiting for. We’d anticipate much more substantial savings in time and money in Phase III.”

Tessella’s big Pharma clients have engaged the company for a variety of adaptive initiatives. One wanted early consultancy on launching its first adaptive trial, perhaps building the infrastructure or a quick bespoke system for a prototype trial. Another might be looking for detailed stats packages for a specific type of adaptive design. A third might want a simulation package for a simplified adaptive trial to get a feel for the kinds of trials that would make sense.

Despite the varying interest in adaptive designs today, Stephen is bullish: “Five years from now, the whole clinical trial environment will look very different. It’ll be the norm for big pharma to use adaptive trials.”

At this year’s Bio•IT World Conference & Expo, Tessella launched a framework software product to give people a feel for how adaptive trials work. “It’s a learning aid and a foundation for future models and things we use in real trials,” says Parke. (See “Best of Show,” p. 11)

“We’re not selling shrink-wrapped boxes of software,” says Stephen. “It’s an evangelism tool — helping biostatisticians who’ve heard about these techniques and want to understand more, get a better feel for how robust they are.”

Stephen is confident that Tessella’s relationships with pharmas in other areas of drug development will result in more collaborations in adaptive trials. “We’ve done this longer than anybody else. We don’t see an awful lot of competition in these adaptive pieces.” 

Sidebar: Trail Blazing Trials

Endocrinologist John Orloff joined Novartis’ clinical development group some four years ago. Two years later, he was entrusted with leadership of the DELPHI (Development, Excellence, Productivity, and Innovation) project, “an effort to transform drug development here at Novartis and map out where we want to be in the next 10 years.”

The goal is nothing less than “transforming the drug development process to deliver better, safer, faster medicines.” Orloff says the adaptive design piece is part of implementing that overall process, although the seeds were already laid by “a very robust statistical and modeling group that’s grown substantially in the past two years.”

“Executing an adaptive design requires careful planning and selection of appropriate trials in advance, and in agreement with the health authorities. You have rigorous methodology applied, and potential biases addressed in the planning process. The potential benefits are to reduce overall development time, increase the amount of information available from the trial, and possibly reduce the number of patients in the trials. But it does require intense planning,” he says.

“We’re in progress with approximately five compounds in trials,” says Orloff. Others are set to launch this year in therapeutic areas including transplantation, endometriosis, COPD, hypertension, and hepatitis, with even more in consideration for 2008. “We’re systematically considering the application of adaptive designs in all our programs. The strategy is to increase the use of adaptive designs for dose ranging. We’re also building up the phase II/III seamless designs methodology.” But Orloff says the best-case on record for adaptive designs is the ASTIN trials.

Despite the upsides, there are significant operational considerations for launching adaptive trials. “To ensure the integrity of the trial, you need to have procedures in place that allow individuals to be unblinded but separate from the clinical trial teams,” says Orloff. Those potential sources of bias require “limiting access to data from the clinical teams, and limiting access to results, so they have to be blinded to the results for the interim analysis,” he says. “We need the appropriate firewalls to maintain integrity.”

Novartis is synergizing its approach to adaptive designs with other model-based approaches. “We have a group here led by Don Stanski for model-based drug development that includes statistical, pharmacological, and biological modeling all integrated in one department to synergize approaches to modeling. I think that’s unique to the industry.” Model-based techniques address “dose selection, terminating compounds early in the process, shifting late stage attrition to early stages, and simulating the outcome of late phase development, so we can anticipate and predict the outcome of our late-stage trials. That is absolutely critical to our success.”

Most modeling work is done internally, “but selectively we’ve partnered with some external groups,” he says.

In addition to dose response designs, some drugs are in pivotal Phase II/III adaptive seamless trials. Novartis and the rest of the industry should know very soon whether the adaptive trial phenomenon is here to stay. Orloff has few doubts: “We’re taking experience from this first wave of trials, learning from them, and applying the principle of adaptive methodologies to the trials coming behind them. We’ll be learning from industry’s perspective as well as the FDA’s.”  -- K.D.

Subscribe to Bio-IT World  magazine.


Click here to login and leave a comment.  


Add Comment

Text Only 2000 character limit

Page 1 of 1

For reprints and/or copyright permission, please contact Angela Parsons, 781.972.5467.