12:00 AM
Sep 05, 2016
 |  BioCentury  |  Strategy

Back to School: Haste, not waste

24th BioCentury Back to School Issue: Revolutionize clinical development

As researchers in industry and academia rapidly unravel molecular disease mechanisms and drug developers become more adept at intervening in disease pathways with drugs targeted to specific populations, the current clinical development paradigm cannot keep pace.

The traditional staged model and frequentist approaches are too slow and uninformative to keep up with advances in biology, signals of exceptional efficacy in initial studies and changes in standard of care. To make matters worse, they do not use new and emerging knowledge to improve the odds of success or to benefit patient volunteers.

Large, cumbersome studies frequently fail to provide the kind or quality of information that drug developers need to make R&D decisions. And even "successful" trials do not provide information that physicians and patients need to make choices among treatment options, or that payers need to make reimbursement decisions.

Today's paradigm also is far too expensive to continue in the face of pricing pressures that only will get worse.

Preserving industry profitability and investor returns requires reducing clinical failures and time to market for treatments that healthcare providers will pay for and patients want.

Last year, Back to School enumerated how patients can lead the way in endpoint selection and collaborate with drug developers and regulators to improve clinical study design -- both of which can improve the speed and efficiency of getting new therapies to market. But while partnering with patients is necessary, it is not sufficient to solve industry's problem with ballooning development costs.

This year, BioCentury's 24th Back to School essay argues that drug developers must lead a profound revolution in clinical development.

Randomized clinical trials (RCTs) should remain the gold standard for validating new therapies. But drug companies must make RCTs more efficient, more predictive and more informative. This revolution will be accomplished by aggressively pursuing adaptive or seamless trial designs and embracing networked approaches that can more quickly evaluate therapies with fewer numbers of patients.

Companies also must work with health systems, patient groups and regulators to determine when and how RCTs can be complemented with other kinds of evidence better suited to answer bottom-line questions, like how a drug will perform in the real world.

To increase success rates and confidence in the results of RCTs and real-world studies alike, drug companies must enrich the information that feeds into study designs. This will require significant investments in the development of mathematical and statistical modeling, in conducting or supporting natural history studies and in accelerating biomarker research in a precompetitive setting.

The new approaches will require sponsors to relinquish control of some elements of study design and conduct, and to share more data than they have been wont to do. But ballooning expenses, late-stage failures, lengthy times to market and insufficient data to garner uptake and reimbursement are the alternatives.

Adopting the new development paradigm will force drug companies to compete on the strength of their molecules, not on the quirks of custom-crafted development programs. It will be obvious when a drug is not differentiated -- but also obvious when a drug is. And that is better for both sponsors and patients in the long run.

Skeptics will point out the executional and regulatory risks of attempting to chart a new course through the clinic. But at least in the U.S. and Europe, regulators and patients are ready for a revolution, and industry really has no choice but to make it happen.

The situation

Restructuring the "D" in R&D is essential to bending the cost curve of drug development and getting important new medicines to patients as quickly as possible.

In 2015, PhRMA estimated its member companies spent $58.8 billion on R&D. Adding just the publicly traded non-PhRMA members that develop drugs or biologics and have current market caps at or above $10 billion -- including Gilead Sciences Inc., Regeneron Pharmaceuticals Inc., Vertex Pharmaceuticals Inc., Shire plc, Incyte Corp., UCB Group, Actelion Ltd., Genmab A/S and Medivation Inc. -- brings the total to $68.4 billion.

Based on a Tufts Center for the Study of Drug Development estimate, which places non-human studies at around 30.8% of total R&D, those companies would have plowed about $47 billion into "D" last year.

Tufts' survey of development costs for 106 drugs from 10 pharma companies puts the mean out-of-pocket costs for Phase I at $29.7 million, while Phase II cost $64.7 million and Phase III cost $253.5 million (in 2013 $).

Much of that money essentially goes down the drain, because the majority of studies fail and only the trial sponsor gets to see and use the data, so the overall drug development enterprise learns almost nothing.

According to a report from BIO on clinical success rates during 2006-15, only 30.7% of Phase II clinical programs advanced to Phase III, and only 58.1% of Phase III programs proceeded to a filing.

The compounded probability of approval from Phase I was a paltry 9.6%. BIO's analysis was based on a total of 9,985 clinical and regulatory phase transition events for 7,455 development programs by 1,103 companies in the BioMedTracker database.

The analysis did not attempt to elucidate reasons for programs that were dropped. Some presumably were discontinued due to lack of funding during the global financial meltdown. Others may have been nixed due to changes in pipeline or portfolio priorities or lack of commercial attractiveness. But many undoubtedly were scrapped because of disappointing data.

BIO noted that it isn't surprising that failure rates were highest in Phase II, since that is generally the first time efficacy is put to a rigorous test. But the failure rates in Phase III, which PhRMA's annual membership survey pegged at 29% of all R&D spending and 59% of pre-approval clinical trial costs in 2014, are appalling.

BIO's analysis confirms one way to do clinical development better: success rates were substantially higher for rare diseases where the genetic cause is usually known, and for programs that used a biomarker for patient selection (see "Selecting for Success," page 3).

According to the analysis, biomarker selection strategies improved Phase II success rates by 16 points to 46.7%. In Phase III, the improvement was 18.4 points to 76.5%. Despite the small number of studies in the sample that used selection biomarkers, these comparisons were statistically significant.

The analysis used data from Amplion Inc.'s BiomarkerBase.

"The large differences in Phase II and III transition success rates are quite convincing, quantitatively, of what many drug developers have long argued anecdotally -- enrichment of patient enrollment at the molecular level is a more successful strategy than heterogeneous enrollment," the report concluded.

Different data from FDA suggest poorly designed studies are frequently at fault for failures of NDAs and BLAs to get approved.

According to an agency analysis published in The Journal of the American Medical Association in 2014, FDA initially denied half of the 302 marketing applications submitted for NMEs between 2000 and 2012. In far too many cases, "preventable deficiencies" were the culprit.

"Failures late in drug development are costly, often involving the commitment of many study participants and personnel," the FDA authors wrote. The analysis was supervised by Leonard Sacks, acting deputy director of the Office of Medical Policy, and Rachel Sherman, associate deputy commissioner for medical products and tobacco.

Just under half of first-cycle failures (47%) were eventually approved, but after lengthy delays. The median time to approval was 435 days after the first action letter. The fastest of these was approved within 47 days, and the slowest within 2,374 days, or 6.5 years.

Deficiencies falling into the preventable category included failure to select suitable endpoints and/or representative patient populations. But the largest category of preventable deficiencies was a failure to select optimal doses, weighing in at a hefty 15.9% of 151 first-cycle failures (see "Reasons for Failure," page 4).

Current methods of dose-finding are little better than guessing. In cancer, companies may not even do dose-optimization studies. Instead, they test the maximum tolerated dose (MTD) from early studies in their confirmatory trials and then optimize dosing postmarket, if it all.

"I don't think we've done a good job, collectively, of getting the dose right," Merck & Co. Inc.'s Eric Rubin said at a June dose-finding workshop sponsored by FDA and the American Association for Cancer Research (AACR). "Oftentimes we take doses into later-stage studies that might not be, or then turn out perhaps not to be, the best dose -- and we end up doing postmarking the types of things to sort that out." Rubin is a medical oncologist and VP of global clinical oncology.

Even more disturbing than high rates of clinical failure are the so-called successful studies that meet their endpoints, yet fail to provide the information physicians and patients need to make treatment decisions, and that payers need to determine whether to reimburse.

This happens all the time, because study populations are narrower and more homogeneous than real-world populations, and differences in endpoints and trial designs make comparison to other therapies difficult or impossible. In addition, clinical trials may exclude prior treatment, or concomitant medications that are part of SOC.

Payers do not determine clinical value based on a drug's ability to show a statistically significant effect on a regulatory endpoint in one or two well-controlled clinical studies. Instead, they need to see improvements in health and quality of life in patients who will actually use the drugs, and clear comparative benefits over alternative treatments.

The list of horribles does not stop there. Drug studies frequently fail to measure endpoints that correspond to symptoms patients care about.

For example, endpoints in cardiovascular indications are generally stroke, MI and death. These are important, to be sure, but they shed no light on whether a therapy will improve -- or reduce -- a patient's ability to function and feel well in daily life.

Finally, it's not news that clinical development today simply takes too long. But it is intolerable -- and untenable -- in therapeutic areas where science is moving quickly, or the standard of care is in flux because of new therapies coming to market.

According to the Tufts analysis, the mean length of Phase I was 33.1 months. Mean length was 37.9 months for Phase II and 45.1 months for Phase III.

It's also not news that lengthy trial timelines increase the expense of development and delay time to revenues. But the scale of improvements required to make a difference may be less obvious, and therein lies the challenge from Back to School.

In 2005, Derek Winstanly, then chairman of Quintiles Transnational Japan KK, told BioCentury that to reduce development costs by $100 million, developers needed to reduce timelines by 19%, or improve trial success by 26%, or cut out-of-pocket preclinical costs by 30%.

To reduce development costs by $200 million, he said, developers needed to reduce times by 41%, or improve trial successes by 32%, or cut out-of-pocket preclinical costs by 60%.

Quintiles declined to update these numbers, but the hurdles could not have been meaningfully improved over the ensuing decade.

The path forward

As Back to School has been documenting since 2005, when all the market and political pressures are aligning to hammer prices and slash margins, preserving industry economics depends upon upstream innovations that improve the yield of products and the time to market.

But while much industry and investor hand-wringing has focused on the cost and perceived low returns from discovery research, the real revolution has yet to take place in development. Not only is the bulk of cost embedded in "D," but the current clinical development paradigm arguably reduces industry's ability to translate "R" more effectively.

"We've achieved a plateau with randomized controlled trials. We don't want to go back below the plateau, we want to build on the plateau," said Donald Berry, a biostatistician at the University of Texas MD Anderson Cancer Center who helped design I-SPY 2, an adaptive platform trial that is testing drug candidates from multiple biopharma companies in biomarker-defined subsets of breast cancer.

The remedies have been hiding in plain sight for decades -- at least since Lewis Sheiner, then of the University of California San Francisco, published a 1997 commentary titled "Learning Versus Confirming in Clinical Drug Development." Sheiner recommended different approaches for clinical development prior to Phase III, when the objective is to learn as much as possible about a molecule, and during Phase III, when the objective is to confirm efficacy to regulatory standards.

The learning phase, which was intended to inform the design of pivotal, confirmatory studies, included techniques such as the use of Bayesian statistics, as well as modeling and simulation to predict the performance of a drug candidate in different patients and under varying dosing regimens and conditions.

A lack of necessary tools or capabilities, real or perceived regulatory risk and predictable management inertia at drug companies have prevented widespread uptake of Sheiner's insights.

Back to School argues a confluence of recent advances makes it time to unleash the revolution in drug development.

One important advance is the emerging body of evidence that non-traditional study designs can provide better information faster.

For instance, a handful of cases have demonstrated that rapid development and approval are possible when adaptive, seamless designs are used to expand studies based on early signals of efficacy.

An early example of the seamless approach came from Xalkori crizotinib from Pfizer Inc. It was approved based on data from a trial in which exceptional early efficacy led to the addition of larger cohorts in a continuous study, instead of a traditional staged approach consisting of sequential trials.

Xalkori was approved 5.3 years from the start of Phase I, and 3.8 years after a Phase II anaplastic lymphoma kinase (ALK)-positive lung cancer expansion cohort was added to the Phase I protocol (see "Seamless and Speedy," page 5).

Additionally, backed by a few forward-thinking companies, some high-profile adaptive platform trials led by academics and patient groups are now under way, led by the I-SPY studies in breast cancer (see "Platform Prototypes," page 15).

The initial results from I-SPY 2, reported in 2014, showed that the program's academic managers could quickly run a robust clinical trial across multiple tumor subtypes. In less than two years, the adaptive Phase II study in breast cancer enabled decisions to advance two compounds to Phase III in specific patient subgroups, and to stop investigation of those compounds in a combined 10 other subgroups.

As of July, three more compounds have graduated to Phase III, and one has been discontinued.

I-SPY is a collaboration between QuantumLeap Healthcare Collaborative, Foundation for the National Institutes of Health (FNIH), FDA, NCI, 16 academic centers, The Safeway Foundation and patient advocates. The candidate compounds have been contributed by nine pharma and biotech companies.

Unfortunately, too many of these novel approaches remain siloed within companies, limiting the efficiencies that can be gained from shared controls and networks of clinical centers and investigators that remain enrolled in standing protocols that enable platforms like I-SPY to operate continuously.

In addition, too many studies, both novel and traditional, are doomed to failure because of a poor understanding of natural disease history, dose-response relationships and other information that could be known up front if only companies would invest in discovering it.

Janet Woodcock, director of FDA's Center for Drug Evaluation and Research (CDER), told BioCentury the agency believes a lack of adequate dose-finding in Phase II is one reason for failures in the Phase II to Phase III transition. "We still get companies who do entire development programs with the wrong dose," she said.

In addition, only a minority of studies use biomarkers to enrich the patient populations and increase the chances for success. According to the BIO data set, only 512 of 9,985 phase transitions incorporated selection biomarkers.

A second key development is that regulators in the U.S. and Europe, even if not in the rest of the world, are asking for companies to use modeling and novel study designs to improve data and accelerate development timelines.

EMA is pushing for companies to adopt modeling techniques to improve the predictive value of trials throughout development, and FDA brought model-informed drug development (MIDD) to the table during PDUFA VI negotiations. The U.S. agency committed to beefing up its ability to consider both MIDD and innovative, complex trial designs.

It is now incumbent upon industry to produce the goods by investing in fit-for-purpose development programs that answer relevant questions faster using fewer resources.

There is no silver bullet. Squeezing more drugs out of development with fewer resources will require many systemwide improvements that build upon and inform each other.

Back to School sees four complementary areas where industry should lead radical change by accelerating the adoption of more efficient, more informative, and more collaborative methods of clinical study.

First, sponsors should stop automatically reaching for tried-and-true frequentist, staged study designs. Instead, whenever possible, they should evaluate and adopt novel trial designs to ensure continuous learning about the drug candidate and disease, and to eliminate delays and operational redundancies exacted by the conventional staged approach.

As long ago as 2006, BioCentury chronicled "the clear and urgent need to break free of the frequentist trial paradigm," but noted that "a combination of regulatory uncertainty and inertia has prevented the adoption of adaptive techniques."

Other barriers included lack of the required computing power, lack of software to do the calculations and lack of personnel trained in the analytical techniques.

The computational barriers have been overcome. Many pharmas and biotechs are hiring the necessary personnel. And regulators in the U.S. and EU have made it clear they are on board.

Still, adaptive studies remain the exception. As of Aug. 18, ClinicalTrials.gov included 120,155 interventional studies involving drugs. A simple search showed only 565 of these included the word "adaptive" and/or the word "Bayesian" anywhere in the study description.

"We need to try every trick we can come up with to enhance and accelerate timelines. Adaptive designs are not new approaches, but they are not utilized as much as they could be," said Doug Williams, president and CEO of cancer company Codiak Biosciences Inc. Williams was previously EVP of R&D for Biogen Inc.

Drug sponsors also should prospectively plan seamless studies. These would rapidly confirm early signals of exceptional responses and evaluate dosing regimens and candidate biomarkers without the need to set up new studies.

Industry should invest in networked, collaborative approaches to clinical testing. This investment should include helping to develop master protocols and participating in multidrug platform studies, which used together can rapidly reveal what drugs work in which populations while reducing duplication of efforts and sparing patient volunteers from unnecessary exposure to placebo or less effective controls.

In competitive areas where many companies are competing for patients, and in settings where it is hard to find subpopulations that might respond differently to different drugs, platform approaches that test several drugs in one protocol may be the only way forward.

Richard Pazdur, who was director of the Office of Hematology and Oncology Products at CDER and is now acting director of the FDA's Oncology Center of Excellence, cites PD-1/PD-L1 inhibitors as an example.

"One of our concerns is that these drugs are expeditiously developed in an appropriate manner," he told BioCentury in June. "Some of the discussions that we are having internally and that we will have with other stakeholders is, for example, doing studies with multiple PD-L1 drugs in one trial comparing them to a common control."

As companies seek to expand their drugs' labels, Pazdur noted, there may not be enough patients to go around for each sponsor to run its own trials.

The platform approach does require relinquishing control of a study's conduct to a third party, which companies have been loath to do. But I-SPY has demonstrated that an academic organizer can set up and train a standing network of sites in complex data collection, all while using a master protocol that ensures the data are collected in a consistent manner.

These studies also provide a forum for simultaneously testing multiple hypotheses across a range of disease subtypes -- something that smaller biotechs may not have the resources to do.

Second, to maximize the value of novel trial designs, drug developers should invest in advancing methods to develop real-world evidence, including in the premarket setting.

Many novel approaches to randomized controlled trials will allow regulators to approve drugs based on smaller data sets. But this accomplishment will make it even more difficult for patients, physicians and payers to evaluate new drugs. Real-world evidence can be a cost-effective way of developing the data they need.

Real-world evidence can include data from pragmatic trials, observational studies, registries, retrospective database studies, case reports, administrative and healthcare claims or electronic health records.

Under its adaptive licensing pathway, EMA is studying how to use such evidence to expand the labels of drugs that were approved for restricted populations based on small data sets from RCTs.

Draft guidance issued by FDA's Center for Devices and Radiological Health suggests that national data systems conceivably could be mined to support prospective, randomized pragmatic studies in the premarket setting.

Pragmatic trials are conducted with broad enrollment criteria in the setting of routine clinical practice. One example is the Salford Lung Study, which began testing GlaxoSmithKline plc's Breo Ellipta fluticasone furoate/vilanterol before the drug was approved for chronic obstructive pulmonary disease (COPD). The study was done to support uptake and reimbursement.

To expand global adoption of this approach, drug developers need to support the development of standardized analytics and data systems to ensure that real-world data are reliable.

Drug companies also should train community practitioners to participate in pragmatic trials. Initially, this represents an investment of money and manpower. But these efforts can result in standing networks that can be used for additional studies.

Third, companies need to improve the inputs into clinical trials by investing in the development of mathematical and statistical models and by designing and funding natural history studies that lead to more informative endpoints.

Natural history studies can lead to the selection of more relevant endpoints that take less time to measure, and provide additional confidence in comparisons to controls.

They also can inform mathematical and statistical modeling to both improve the quality of information generated in drug development, and increase efficiency in terms of time, cost and patient volunteers required.

Modeling can be used to assess PK/PD and safety properties to select clinical candidates, predict safety and efficacy outcomes and -- importantly -- select the correct doses.

EMA has suggested it could be possible to approve a drug based on a single Phase III study if accompanied by a well-run Phase II dose-finding trial.

Guidances and other regulatory assurances would serve to remove industry excuses for inaction on the new paradigm. But, frankly, drug developers should need no more incentive than the prospect of reducing the proportion of Phase III studies and marketing applications that fail because the dose is wrong.

To exercise the industry's leadership, drug developers should be collaborating to hone these techniques and bring them to regulators, rather than waiting for guidances that may be years in the making.

Fourth, Back to School calls on industry to invest more heavily in collaborative efforts to identify and validate biomarkers for broad use rather than waiting for regulators to figure it out.

Hunting for biomarkers has been a routine part of drug development for well over a decade, yet the number of markers that can be used to match patients to therapies remains minuscule. Despite common wisdom that a lack of regulatory endorsement is the bottleneck, it's more likely that a lack of scale, data sharing and consistent assays and analytics are what stand in the way.

A recent example comes from the checkpoint inhibitor field, where a lack of consistency in assays is making it impossible to tell how much PD-L1 expression makes a difference in responses -- or, for some of the drugs, whether any amount of PD-L1 expression does. Differences among the assays used in clinical trials also preclude comparisons of the treatments to each other, complicating choices for physicians and patients.

Large-scale collaborative studies are one way to identify and validate selection markers, and possibly even prognostic markers that could form the basis of surrogate endpoints.

One example is the 1,150-patient CoMMpass study by the Multiple Myeloma Research Foundation (MMRF), which has identified previously unknown molecular drivers of MM. Companies that are collaborating on the study are using the data to investigate new targets and selection biomarkers in their development programs.

The CoMMpass data also will inform the treatment arms in a collaborative MM master protocol trial that is expected to begin next year.

Platform trials such as I-SPY also can validate combinations of markers and therapeutics. And natural history studies and shared registries can provide large, high-quality data sets that can be mined for candidate markers.

None of Back to School's prescriptions will solve the "D" dilemma on its own. They are a package deal in which each element can further the gains from the others.

The following sections describe each of these transformations in more detail and, importantly, show that forward-looking companies, regulators and patient groups already are breaking ground.

Taken together, Back to School's four prescriptions will create a virtuous cycle of learning that will result in faster and cheaper development of a greater number of medical products that improve quality and quantity of life for patients.

Innovating for efficiency

Sponsors should stop automatically reaching for tried-and-true frequentist, staged study designs. Instead, whenever possible, they should adopt novel trial designs to ensure continuous learning about the drug candidate and disease, and to eliminate delays and operational redundancies exacted by the conventional staged approach.

Companies are beginning to use novel trial designs, but in fairly limited fashion, and almost exclusively in early development. Expanding use to late-stage development can unlock even greater efficiencies because of the impact on the cost and duration of these large studies.

Generally, the new strategies fall into two categories. One uses Bayesian statistics to enable adaptive trials; the other applies more efficient protocols, including seamless designs, platform approaches and basket studies, to reduce setup time, infrastructure and redundant controls. The most powerful approaches combine both categories.

In comparison with frequentist designs, Bayesian trials can reduce time and expense by making better use of emerging data about a compound or population to expand, reduce or stop trials or study arms based on probabilities of success or failure.

A conventional trial using frequentist statistics enrolls a fixed number of patients who receive a predetermined dose or placebo, administered in a preset regimen. No modifications are made as the trial progresses. Conclusions are drawn at the end of the study, using only the data from the trial itself. Previous knowledge, such as data from previous studies, is not included in the final analysis.

These studies are vulnerable to failure if the characteristics of the patient population, the effect of the medicine, or variation in the data are different than expected.

Bayesian statistics provide a formal methodology for changing or modifying study parameters as new information comes in. Interim analyses are used to make adjustments in randomization, doses, study duration and/or total enrollment based on prespecified criteria.

Results are reported as probabilities, and the statistical analysis can include not only data from the study, but also previous data. In a nutshell, the Bayesian idea is to consider all of the data as part of a stream of information in which inferences are continually updated.

"Just like it always helps to look and see where you're going, it helps to monitor the trial and have a prospective design that will let you make changes based on the data you're accumulating," said MD Anderson's Berry. "Under the current paradigm, we run trials that are 5, 10 years and never look at the data. It's anti science."

Designing a Bayesian trial requires upfront work to establish prospective criteria for making modifications. These criteria are established using simulations based on data from the literature and earlier studies of a compound. The simulations allow sponsors to model how a range of variables -- including accrual or dropout rates, missing data, characteristics of the study population and performance of the medicine -- can affect study size, cost, duration and outcomes. That information is used to optimize the study design.

The decision rules are applied automatically by an algorithm as data come in from the ongoing study.

The risk of type I error (false positive), which is a concern any time a trial is altered midstream, is managed by conducting thousands of simulations of the study in which it is assumed that the treatment has no effect. The simulations vary the characteristics of the trial such as accrual rate and the expected event rate in the population. To be confident in the study, no more than 2.5% of the simulations can show a false-positive result.

For drug developers married to frequentist designs, it may be unsettling to prospectively determine decision rules that will be implemented...

Read the full 24927 word article

User Sign in

Trial Subscription

Get a 4-week free trial subscription to BioCentury

Article Purchase

$150 USD
More Info >