Collaboration and Strategy Development -
Cephalon Uses Modeling and Simulation to Optimize its Flagship Product
by Bob Korsan1
When Cephalon, Inc. (“Cephalon”) first set out to expand the label of its flagship compound, Provigil™, it needed a factual basis to support a go/no-go decision for a previously unstudied indication—quickly, efficiently, and with as much certainty as possible. In response, Cephalon joined forces with scientists and decision analysts from Pharsight Corporation. Using the tools of modeling and simulation, the team explored the knowns and unknowns of the drug and indication, to design a trial based upon all available information.
In this initial two-month modeling and simulation effort, Pharsight and Cephalon became collaborators in clinical development. Model exploration enriched team understanding of the expected drug profile, providing a basis for quick action once recommendations were final. The proposed phase III trial design modifications were approved within days. The study succeeded in confirming efficacy with fewer treatment arms than originally planned, for a direct clinical cost savings of roughly 25 percent.
Over the next year and a half, the project’s model knowledgebase would evolve to become a central input to lifecycle management of this compound, Provigil®2, and its successor, Nuvigil™.
Growth and Wise Choices
Cephalon ranks among the world’s fastest-growing biotechnology firms, but it may be best characterized as a small-molecule pharmaceutical company. Its founders launched the organization in 1987 to perform basic research into mechanisms of cell death and survival, and to develop treatments for cancer and neurological disorders.
Cephalon’s growth streak began in 1998, with FDA approval of its new wakefulness agent Provigil (modafinil) for treatment of daytime sleepiness in narcolepsy. By 2003, Provigil would come to represent forty percent of Cephalon’s total revenue, with annual sales of $290.5M3.
Several new potential indications were commercially and therapeutically attractive. A reliable method was needed to prioritize them. Among the contenders was treatment of excessive sleepiness due to a little-studied problem: shift-work sleep disorder (SWSD).
Shift workers’ frequent changes in schedule can disrupt normal sleep/wake cycles, leading to poor quality sleep and difficulty staying alert during working hours. The situation becomes dangerous when sleepy workers operate machinery or drive a car. Taken at the start of the working “day,” Provigil promised to improve work-time wakefulness.
As in most drug development, the effort to study Provigil in SWSD was too costly to undertake without a reasonable level of confidence that the program would succeed—particularly with so many other promising indications competing for the same development resources. With these factors in mind, Cephalon elected to perform a pilot or proof-of-concept trial.
“We had the usual questions,” says Dr. Lilliam Kingsbury, Vice President of Biometrics at Cephalon4. “If we were going to do trials, what would be the best dose to use and the schedule? What would be the best clinical trial design? At what point should we continue or discontinue development? Throughout the program our thinking was guided by the central question: What are our chances for technical and market success in this indication?”
“We wanted to use modeling and simulation to try to maximize the value of our prior information,” Kingsbury says, “to then explore and evaluate our knowledge and develop some trial strategies.”
“Ultimately,” she says, “we also wanted to integrate what we knew across multiple disciplines, including even our commercial people—because we wanted to design a trial that would also allow us to interpret this resource in terms that will support the product even through its commercial development.”
Cephalon called in Senior Scientists at Pharsight Corporation (“Pharsight”), to spearhead the modeling effort. To launch the project, Cephalon Executive Vice President Dr. Paul Blake led meetings that brought together quantitative, clinical and marketing leaders for Provigil. Together the team reviewed the data they had: prior phase I-III trials for Provigil in narcolepsy, two studies in another indication (sleep apnea), one unpublished pilot study showing Provigil effects in 16 sleep-deprived volunteers, and sparse literature data on SWSD and sleep deprivation. In the next phases, Pharsight and others would combine this information into a set of models, and use those models to predict Provigil’s performance in SWSD and, finally, to explore alternative trial designs.
Knowledgebase Building and Decision Support
“One important thing that we needed to decide was how many arms we needed for this trial,” Kingsbury says, “because [SWSD] patients are scarce.” The base plan was a three-arm trial comparing two doses of Provigil against placebo. The trial would observe the clinical measure of efficacy—work-time sleepiness (alertness)—and two measures of the main side effect—insomnia— at one day toward the end of the workweek. But would this trial be sufficiently powered to capture the primary effect, or lack of one, at the study doses? Determining the best dose and observation schedule in SWSD would not be straightforward.
The first step was to model factors that influence sleepiness. To imitate normal wakefulness cycles, Pharsight turned to an existing model that predicts alertness and performance over a 24-hour cycle, based on interactions between an internal body clock, light/dark exposure and how long the patient has been awake5.
The next step was to simulate Provigil’s effects on that sleepiness, in order to choose the best dose. For that, Pharsight needed to relate Provigil concentrations in the blood to the drug effects—both wanted and unwanted. Cephalon’s data for narcolepsy and sleep apnea (the original FDA approved indications for Provigil) provided a clear picture of Provigil’s dose-concentration curve. To estimate efficacy in SWSD, Pharsight focused on the pilot sleep-deprivation data and relied on internal Cephalon expertise to estimate Provigil’s effect in SWSD.
The model was built from real data, wherever possible, augmented by expert opinion where data were unavailable. When tested against the pilot study data, the model reliably predicted the average responses to Provigil and placebo. The statistical spread of the study data was captured and used to calculate the spread and confidence level of the simulated trial outcomes.
“We value understanding the tradeoffs between side effects and efficacy,” says Bob Korsan, Pharsight project leader and senior scientist, “and modeling those simultaneously. We [can also] include both clinical tradeoffs and market tradeoffs, and our understanding of the prescribing physician’s perceptions of these tradeoffs, in building our model.”
“So it was actually through the simulation effort, with Pharsight, that we were able to finalize that trial design,” Kingsbusry says. “ [We] incorporated that and took the risk of going just with the two-arm study – which turned out to be just right. But the value of the simulation will speak for itself.” With two arms rather than three, and a more flexible observation period, the new design netted 25 percent direct clinical cost savings.
Meanwhile, the SWSD project had convinced Cephalon stakeholders to adopt the modeling approach in the broader program of Provigil development. The next project focused on Provigil in Attention Deficit Hyperactivity Disorder (ADHD). Children with ADHD have a tendency towards distraction and agitation that can disrupt performance at school and relationships with others. Paradoxically, these children often benefit from treatment with stimulant drugs such as Ritalin. As a wakefulness agent, Provigil was promising as an ADHD treatment without the addictive potential of stimulants. In addition, a concentrated version of Provigil was being developed to provide flexible and easy dosing for pediatric and other patients. Early trial results were promising, but inadequately powered to provide confident go/no-go decisions. The data did, however, provide the basis for modeling.
The Cephalon/Pharsight collaboration extended the Provigil models to include literature and phase IIa data on ADHD. This time the collaboration was able to craft weight and age related pediatric dosing for phase III based on the phase IIa pediatric data. Three simultaneous phase III trials were designed and executed. In August of 2004, Cephalon announced the results: three successful phase III trails in ADHD (p<0.0001)6. The trade name for Provigil (modafinil) in ADHD is Sparlon™. The FDA issued Cephalon an ‘approvable’ letter October 2005 in response to the SNDA that was filed in December 2004.
“We believe that integrating disparate data sources, literature and [quantifying certainty] through model-building can bring us a more integrated knowledge base,” Kingsbury says. “So we have a better way of understanding risks and benefits.”
“It's an approach that should enable us to take advantage of the prior information that we have to accelerate development,” she says, “not by reducing the amount of information that we have from subjects and patients, but by better targeting how we design our trials and how we present the evidence.”
Cephalon’s success with the FDA and in the market has allowed it to expand into new indications via a strong acquisition program. Scientists in Pharsight’s Strategic Consulting Services group look forward to building on its five-year collaboration with Cephalon in these exciting new areas.
1 This article would not have been possible without the invaluable assistance of Shawne Neeper of Pharsight. Many thanks.
2 PROVIGIL is a registered trademark of Genelco, SA, which is licensed to Cephalon, Inc.
3 see http://phx.corporate-ir.net/phoenix.zhtml?c=81709&p=irol-newsArticle&ID=494932&highlight
4 Quotes are from a transcript from the June 2003 Drug Information Association meeting in San Antonio, Texas. Session entitled “The Collaborative and Strategic Benefits of Exploring Drug Models in Clinical Development” by Lilliam Kingsbury, PhD, Vice President of Biometrics at Cephalon and Robert Korsan, Senior Scientist, Decision Services at Pharsight Corporation.
5 Jewett and Kronauer. Interactive mathematical models of subjective alertness and cognitive throughput in humans. J Biol Rhythms 14:588-597 (1999).
6 http://phx.corporate-ir.net/phoenix.zhtml?c=81709&p=irol-newsArticle&ID=605263&highlight