Proceedings of a Workshop
Advances in cancer research have led to an improved understanding of the molecular mechanisms underpinning the development of cancer and how the immune system responds to cancer. This influx of research has led to an increasing number and variety of therapies in the drug development pipeline, including targeted therapies and associated biomarker2 tests that can select which patients are most likely to respond, and immunotherapies that harness the body’s immune system to destroy cancer cells. Compared with standard chemotherapies, these new cancer therapies may demonstrate evidence of benefit at an earlier stage of development. However, there is a concern that the traditional processes for cancer drug development, evaluation, and regulatory approval could impede or delay the use of these promising cancer treatments in clinical practice. This has led to a number
1 The planning committee’s role was limited to planning the workshop. The Proceedings of a Workshop was prepared by the rapporteurs as a factual account of what occurred at the workshop. Statements, recommendations, and opinions expressed are those of individual presenters and participants and are not necessarily endorsed or verified by the National Academies of Sciences, Engineering, and Medicine. They should not be construed as reflecting any group consensus.
2 In this proceedings, a biomarker is defined as “a characteristic that is objectively measured and evaluated as an indicator of normal biological processes, pathogenic processes, or pharmacologic responses to an intervention” (NASEM, 2016a, p. 26).
of efforts—by patient advocates, the pharmaceutical industry, and the Food and Drug Administration (FDA)—to accelerate the development and evaluation of promising new cancer therapies, especially for cancers that currently lack effective treatments. However, generating the necessary data to confirm safety and efficacy during expedited drug development programs can present a unique set of challenges and opportunities.
To explore this new landscape in cancer drug development, the National Cancer Policy Forum, in collaboration with the Forum on Drug Discovery, Development, and Translation, developed a workshop, The Drug Development Paradigm in Oncology. The workshop was held on December 12–13, 2016, at the National Academies of Sciences, Engineering, and Medicine in Washington, DC. This workshop convened cancer researchers, patient advocates, and representatives from industry, academia, and government to discuss challenges with traditional approaches to drug development, opportunities to improve the efficiency of drug development, and strategies to enhance the information available about cancer therapy throughout its life cycle in order to improve its use in clinical practice.3 Many workshop speakers discussed potential strategies for improving cancer drug development, including
- patient-centered drug development that prioritizes patient needs and employs patient-reported outcomes;
- biologically informed strategies in drug development, including the use of biomarkers to select patients most likely to respond to treatments and surrogate endpoints that are informed by a drug’s mechanism of action;
- more flexible, efficient, and continuous clinical trial designs that can evaluate multiple interventions in different patient populations, expand testing to larger numbers of patients when agents show initial promise, and conscientiously and efficiently manage limited patient availability and clinical trial resources; and
- expanding clinical trial eligibility and leveraging the use of real-world data to better understand how drugs perform in clinical practice.
3 Although some workshop speakers noted that improving the efficiency of drug development could potentially result in lower drug development costs, particularly as therapies are targeted to smaller patient populations, a detailed analysis of the cost of drug development was beyond the scope of this workshop.
This proceedings is a summary of the presentations and discussions at the workshop. A broad range of views and ideas were presented, and a summary of individual suggestions to improve cancer drug development is provided in Box 1. The workshop statement of task is in Appendix A and the workshop agenda is in Appendix B. The webcast and speakers’ presentations have been archived online.4
A number of workshop speakers discussed the goals of oncology drug development, the traditional phased approach to drug development, the regulatory processes for review and evaluation of new cancer therapies, and the ethical standards for clinical research.
Mark Ratain, director of the Center for Personalized Therapeutics and associate director for clinical sciences at The University of Chicago Comprehensive Cancer Center, summarized the goals of oncology drug development as
- demonstrating a therapy’s anticancer activity;
- identifying target populations likely to respond to the therapy;
- determining dosage and schedule that optimizes the benefit-to-risk ratio; and
- identifying individual patient factors that necessitate dose modification, such as age, body size, organ function, and molecular features affecting a drug’s disposition or interaction with a drug target.
Ratain said that oncology drug development traditionally proceeds through distinct, chronologic phases (see Table 1). G. K. Raju, executive director of Pharmaceutical Manufacturing Initiatives at the Massachusetts Institute of Technology, added that FDA weighs the risks and benefits when deciding if drugs should be approved for use in clinical practice.
Marc Theoret, lead medical officer in FDA’s Office of Hematology and Oncology Products (OHOP), said that the agency has two categories of
4 See http://www.nationalacademies.org/hmd/Activities/Disease/NCPF/2016-DEC-12.aspx (accessed February 21, 2017).
|Phase 0||Nonclinical pharmacology and toxicology studies.|
|Phase I||Small studies to assess safety and determine the appropriate dose of an investigational new drug (IND).|
|Phase II||Preliminary research into effectiveness of an IND, either in a single arm study or compared to placebo or standard of care, conducted on small numbers of patients with the targeted disease. Safety, dosing, and short-term side effects are also studied.|
|Phase III||Studies in larger and more varied patient populations, if efficacy is demonstrated in Phase II.|
|Phase IV||Postmarketing studies to amass more complete information on an approved drug’s safety, effectiveness, and optimal use.|
SOURCES: Ratain and Theoret presentations, December 12 and 13, 2016; FDA, 2014a.
drug approvals: regular and accelerated approvals. Regular and accelerated approvals both require substantial evidence of safety and efficacy based on adequate and well-controlled trials, he said. The endpoints for regular approval require direct evidence of clinical benefit—such as extending survival, improving physical functioning, or relieving tumor-related symptoms—or an effect on an established surrogate endpoint. Theoret said there is no comparative effectiveness requirement for regular approvals.
He said that accelerated approval is intended for products that treat serious or life-threatening illnesses and address an unmet medical need. There are four types of expedited drug development pathways at FDA—fast track, breakthrough therapy, priority review, and accelerated approval (see Table 2). Theoret noted that all FDA pathways for expediting clinical development and review of a new drug consider the available therapies to treat the target disease to determine whether there is an unmet medical need, i.e., if the new therapy appears to provide an improvement or advantage over available therapies (FDA, 2014c).
Theoret said that expedited pathways are widely used in oncology drug development: 42 percent of all Breakthrough Therapy Designation requests submitted to the FDA Center for Drug Evaluation and Research (CDER) between 2012 and 2014 were submitted to OHOP and one-third of the Breakthrough Therapy Designation requests to OHOP were granted. In addition, from 2014 to 2015, half of the 24 OHOP approvals of new molecular entities were accelerated approvals (FDA, 2015).
|Fast track designation||Granted when a drug demonstrates the potential to address an unmet medical need for treatments, which may be based on nonclinical information in an early stage of development. Allows rolling submission of the marketing application based on observed preclinical or clinical activity, and initiates frequent communication with FDA to rapidly resolve arising issues.|
|Breakthrough therapy designation||Granted when preliminary clinical evidence indicates a new drug may demonstrate substantial improvement over an existing treatment, and may have an effect on a clinically significant endpoint. Allows for frequent meetings with FDA based on overwhelming clinical activity and, if the observed activity holds in continued follow-up or subsequent studies, then an expedited review of the application will be conducted.|
|Priority review||Designation granted when a sponsor submits an application (new or supplemental) that, if approved, would provide a significant improvement in safety or effectiveness. Ensures an expedited time of review of the application, and requires FDA action within 6 months.|
|Accelerated approval||Approval for when a drug demonstrates meaningful therapeutic benefit over existing treatments. Endpoints based on unestablished surrogate endpoints or intermediate clinical endpoints. Confirmatory trials ongoing at the time of approval.|
SOURCES: Theoret presentation, December 13, 2016; FDA, 2014c.
In line with the expedited pathways for drug evaluation and approval, there is also interest in more flexible and efficient mechanisms for cancer drug development, Theoret said (see section titled New Strategies in Oncology Drug Development). Staff from FDA’s OHOP recently authored an article outlining an evolving drug development paradigm, indicating their willingness to consider alternatives to the traditional phased drug development process (Prowell et al., 2016; Theoret et al., 2015), said Janet Woodcock, director of CDER at FDA.
“Based on better scientific understanding of the biological underpinnings of cancer and the host response to cancer, some investigational products are demonstrating a very large magnitude of antitumor activity very early on in drug development. These unprecedented response rates have brought us into this new era of oncology drug development; the lines between distinct trial phases are becoming blurred, and a single trial may serve as the entire clinical drug development program supporting at
least an initial FDA approval,” Theoret said. He added that “discrete drug development objectives might not be evaluated in [sequential] stand-alone trials, but will be evaluated seamlessly in cohorts within an existing trial in some cases, in treatment refractory settings with clear, high unmet medical need.” For example, a single-arm trial demonstrating a treatment effect on a surrogate endpoint may provide the preliminary and primary evidence for an accelerated approval, while confirmatory trials are ongoing, he said.
Steven Joffe, associate professor of medical ethics and health policy at the University of Pennsylvania Perelman School of Medicine, outlined the ethical requirements for clinical research (Emanuel et al., 2000). For example, a clinical trial should have social or scientific value and should be conducted appropriately so it results in scientifically valid conclusions. The protocol for a clinical trial also needs to undergo an independent ethics review by an Institutional Review Board (IRB)5 prior to study launch, to ensure that adequate informed consent is obtained from participating patients, and that there is fair selection of clinical trial participants. Joffe added that a clinical trial should also have a favorable benefit-to-risk ratio for the individual patients participating in the research, as well as for future patients who stand to gain from the results of the research.
Frank Rockhold, professor of biostatistics and bioinformatics at the Duke Clinical Research Institute, noted that the ethical imperatives for clinical research are defined in the Declaration of Helsinki,6 which states that the risks to participants in clinical trials must be continuously monitored, assessed, and documented by researchers. When the risks are found to outweigh the potential benefits, or when there is conclusive proof of definitive outcomes, researchers need to assess whether to continue, modify,
5 An IRB is a panel of scientists and non-scientists in hospitals and research institutions that oversees clinical research. IRBs approve the clinical trial protocols, which describe the type of people who may participate in the clinical trial, the schedule of tests and procedures, the medications and dosages to be studied, the length of the study, the study’s objectives, and other details. IRBs make sure that the study is acceptable, that participants have given consent and are fully informed of the risks, and that researchers take appropriate steps to protect patients from harm (FDA, 2014a).
or immediately stop the study. Rockhold added that data and safety monitoring boards (DSMBs)7 serve this purpose and are integral to the ethical conduct of clinical trials. He noted that at certain intervals during the trial, the DSMB integrates interim data and determines if the benefit–risk ratio to continuing the trial is still favorable, and if there is still clinical equipoise, which he defined as a state of genuine uncertainty about the advantages or disadvantages of each therapeutic arm in a clinical trial.
A number of workshop participants discussed common challenges with the traditional approach to drug development, such as inefficiencies and long development timelines, a lack of clinically relevant information, the use of inappropriate endpoints, and inaccuracies in selecting dosages of therapies.
Ellen Sigal, chair and founder of Friends of Cancer Research, reminded workshop participants that, on average, it takes more than 10 years of testing, starting with approximately 5,000 to 10,000 compounds, to develop a single new FDA-approved drug (AACR, 2011) (see Figure 1).
Although gaps in understanding the basic science of cancer are partly responsible for the arduous process of the traditional drug development process, Sigal said that the low success rate and long time frames in clinical drug development can also be attributed to inefficiencies from lack of data sharing among companies and academic institutions.
Ratain noted that it is particularly undesirable when a drug fails late in the development timeline, after tremendous amounts of time and resources have been invested. “It is okay to fail as long as one fails pretty quickly,” he said, but frequently that does not happen. Failure can be due to a number of factors, Ratain said, such as not identifying the right dosage
7 A DSMB is a “group of individuals with pertinent expertise that reviews on a regular basis accumulating data from one or more ongoing clinical trials. The [DSMB] advises the sponsor regarding the continuing safety of trial subjects and those yet to be recruited to the trial, as well as the continuing validity and scientific merit of the trial” (FDA, 2006). They are also known as data monitoring committees or Data and Safety Monitoring Committees.
or the population likely to respond well to the therapy, as well as failing to identify individual patient factors that influence how effective or toxic the therapy is. Woodcock said that more effective ways to identify and drop ineffective drugs at an earlier stage of development would improve the success rate for drug development and reduce its cost, so more focus and resources can be devoted to interventions that will have a substantial impact on a disease, including drugs that can potentially cure disease. Often these interventions will be combinations of new agents, which are challenging to study, she noted.
Ratain added that inefficient trial designs and misunderstandings of governmental regulations or corporate bureaucracy can delay the onset or lengthen the duration of clinical trials. Clinical trial designs with more patients, sites, or data than necessary can also contribute to inefficiencies and extra expense. In addition, if historical controls are used to compare the therapy’s effectiveness, the results may not be reliable in current patient populations, Ratain said. In some cases, “we are treating patients without really learning anything,” he said.
Joffe noted that in order to make the drug development process faster, it is necessary to have a discussion about the ethical trade-offs that may result. He said that it takes time to collect data that meet traditional thresholds for safety and efficacy, and that accepting a faster pace of drug development requires changes to the amount of information gathered to inform these decisions. Two rationales have been used to support faster drug development, Joffe said: (1) some of the evidence requirements in the traditional drug development process add little or no value to information about a drug’s safety and efficacy; or (2) less evidence about safety and efficacy for drug approvals is acceptable in exchange for speed. Joffe said that it is important to be explicit about the rationale for increasing the speed of drug development, and to support the rationale with evidence-based information. Joffe added that if regulatory approvals are made more quickly based on less information on safety and efficacy, confirmatory studies will need to be completed in the postmarketing setting to gather additional information.
When a drug receives FDA approval, there is often uncertainty about which patient populations in clinical practice will most benefit from a drug, said Maria Koehler, vice president of oncology strategy, innovation,
and collaborations at Pfizer Inc. This is because clinical trials often enroll patients who are healthier than many patients in clinical practice settings, who frequently have multiple comorbidities (IOM, 2013). She noted that the problem is also exacerbated by low participation rates in trials and a lack of diversity in cancer clinical trial enrollments (Chen et al., 2014; IOM, 2010). Therefore, payers often require additional testing and confirmation of clinical trial results in more representative populations in order to make coverage decisions, she said.
As cancer is further defined by genetic mutations, several speakers noted that reliable biomarkers and surrogate endpoints will be key in regulatory and clinical decision making. Woodcock stressed that the “traditional Phase I, II, III progression cannot really provide enough information in our current era of precision medicine, as cancers are splintered into multiple subgroups and treatment categories” usually based on their underlying genetics. She added that often the scientific knowledge about biomarkers to predict patient response to therapies is evolving concurrently with the development of a therapy, and unless that information is incorporated into the drug testing protocol, development of such companion diagnostics will lag behind that of new drugs. “In general, we get agents on the market where the maturity and our understanding of the diagnostic component is still not [sufficient], and we really do not know everything we need to know,” she said.
Several participants pointed out the inadequacy of commonly used endpoints to assess patient response in clinical trials, including the Response Evaluation Criteria in Solid Tumors (RECIST). Wolfgang Weber, chief of the molecular imaging and therapy service at the Memorial Sloan Kettering Cancer Center, pointed out that the RECIST criteria were originally adopted in the 1970s based on the primary technologies to detect tumor changes at that time. These criteria are based on biomedical imaging results to assess changes in tumor size and number within a period of time following treatment. Disease progression is assumed when tumors increase in size, or if new tumors develop.
Weber noted several challenges when using RECIST criteria to determine patient response. Although RECIST criteria have been shown to correlate with patient survival and other clinical outcomes (Eisenhauer et al., 2009; Miller et al., 1981; Therasse et al., 2000; Wolchok et al., 2009),
there is a large degree of variability in patient responses. Some patients who do not show a response based on the criteria have extended periods of survival, while some responders have shortened periods of survival (Bruzzi et al., 2005; Johnson et al., 2006). Weber said that unless a therapy has an unusually strong and immediate impact on tumors, large numbers of patients in clinical trials are needed in order for RECIST criteria to reliably predict improved outcomes. Weber added that the criteria entail subjective judgment, and often cannot distinguish between scar tissue and viable tumor. The criteria also cannot measure metastases at all sites (e.g., bone) and may not be able to distinguish between slow-growing and stable tumors, Weber said (Thatcher et al., 2005; Therasse et al., 2000).
RECIST criteria can also be problematic for evaluating immunotherapies, which can initially cause an increase in tumor size before causing tumor regression (NASEM, 2016b; Wolchok et al., 2009), said Eric Rubin, vice president and therapeutic area head in oncology early development at Merck Research Laboratories. This increase in size is thought to be due to infiltrating immune cells, which is associated with therapeutic response, Rubin said. Sigal added that “immunotherapy is a new world, and we cannot rely on old standards.” Given the new developments in technologies and treatments, Gideon Blumenthal, acting deputy director of OHOP, suggested that there may be better quantitative and qualitative metrics to assess tumor burden and response to different therapies. Woodcock added, “The classic types of endpoints that have been used may not capture the effect properly, and may not help us in understanding the activity of an immunotherapy, so other endpoints are going to have to be devised.”
Woodcock also stressed using endpoints that consider the patient perspective. “Survival is really important in oncology, but so is quality of that survival,” she said, and suggested increased use of quality-of-life endpoints in clinical drug development. “Drug development has to be driven by the patients,” Sigal added, suggesting that such development take a more patient-oriented approach, rather than a product-oriented approach.
Traditionally, investigators have determined the dose of a new drug by starting with small test doses in patients and gradually increasing the amount of those doses until patients develop serious adverse reactions, said Steven Piantadosi, Phase One Foundation distinguished chair of the Samuel Oschin Comprehensive Cancer Institute at Cedars-Sinai Medical Center.
“Most of what we do in oncology is dose escalation under the ordinary belief that more drug is better,” Piantadosi said. However, that is not necessarily the case because the traditional “maximum tolerated dose”8 model is not appropriate for many of the newer classes of cancer drugs, which have stronger biological effects with relatively fewer toxicities compared with traditional chemotherapy, he said.
R. Donald Harvey, director of the Phase I Clinical Trials Section at the Winship Cancer Institute and associate professor of hematology/medical oncology and pharmacology at Emory University, added, “We look for the ceiling all the time in early drug development, but really understanding the basement is critical,” especially for therapeutics composed of monoclonal antibodies. For these drugs, modeling based on how effectively these antibodies attach to target receptors can predict the lowest dose needed to generate a response, Harvey said. Rubin called this the “minimally effective dose.” Harvey pointed out that the relationship between dose exposure and effects, both positive and negative, is rarely known before a new agent is tested in a first-in-human clinical trial. That relationship is further refined with subsequent trials, but even in Phase III trials, as many as 85 percent of patients require dose reductions (Postel-Vinay et al., 2011).
Ratain noted that historically, oncology drugs were intravenous chemotherapy drugs with a narrow range between effectiveness and toxicity. “So we would push them to the maximum tolerated dose and then we would back off if we went too far. But we are now in an era where we have drugs that do not have a narrow therapeutic index and we should have different dose testing paradigms for them,” he said.
Amy McKee, supervisory associate director at OHOP, said dose optimization is further complicated by bioavailability of oral agents. When chemotherapy drugs are given intravenously, bioavailability of the drugs given is not an issue. In contrast, bioavailability is of vital importance with oral cancer therapies, she said.
Piantadosi added that dosing designs for single agents do not extend satisfactorily to combination therapy development. Harvey agreed, and noted that the maximum tolerated dose does not often apply when com-
8 Maximum tolerated dose is “the highest dose of a drug or treatment that does not cause unacceptable side effects. The maximum tolerated dose is determined in clinical trials by testing increasing doses on different groups of people until the highest dose with acceptable side effects is found.” See https://www.cancer.gov/publications/dictionaries/cancer-terms?cdrid=546597 (accessed May 18, 2017).
bining therapies, or expanding drugs into different types of patient populations. Multiple schedules for different combination therapies are rarely explored in a comparative fashion, he added.
Harvey said a Phase I trial of vemurafenib and ipilimumab in combination was stopped early due to liver toxicity (Ribas et al., 2013). “It is the first Phase I trial I can recall reading about in the New England Journal of Medicine that was a negative trial [because] there were overlapping toxicities that were not predicted before that trial opened,” he said. Joffe and Piantadosi noted that life-threatening side effects have occurred when drugs in the same class have been combined, such as immunotherapy combinations or combinations of targeted therapies. The toxicities associated with the combinations were not evident in early-phase testing.
Many workshop speakers provided suggestions and examples of new strategies in oncology drug development, including
- patient-centered drug development;
- mechanism-informed drug development;
- new, clinically relevant endpoints;
- modeling to improve the development pathway, more accurately predict optimal dosing, and calculate benefit–risk ratios;
- innovative trial designs;
- leveraging real-world evidence from clinical practice;
- expanding trial eligibility; and
- collaboration among stakeholders.
“The patient should always come first [in drug development],” Sigal said. She suggested that patients work with FDA in meaningful ways to ensure their needs are met by clinical trials for drug development, and added that an objective of the 21st Century Cures Act (see Box 2) is to incorporate patient perspectives into the regulatory process and address patients’ unmet medical needs. Other opportunities to incorporate patients into the drug development process include FDA’s Patient-Focused Drug Development Initiative (FDA, 2017e) and the Patient-Centered Outcomes Research
Institute.9 “Patients want efficacy, but they also want quality of life,” Sigal stressed, suggesting that drug development incorporate patient-reported outcomes as a means to assess quality of life. Woodcock agreed that using quality-of-life endpoints that consider patient perspectives would improve drug development. Ratain agreed, noting that in addition to assessing tumor response in clinical trials, there should be assessment of meaningful symptom control or relief through patient-reported outcomes.
Sigal also stressed the importance of real-world clinical data to patients and clinicians when making decisions about treatments that have similar efficacy, but may have different impacts on quality of life. These impacts may be detected by analyzing real-world data from clinical practice, such as patient-reported outcomes in electronic health records (EHRs). Lynn Matrisian, chief research officer of the Pancreatic Cancer Action Network, agreed, noting that “real-world evidence is something patients really care about.” Koehler added that real-world evidence can facilitate patient-centered drug development by
- focusing on broader populations and outcomes that are important to patients;
- better informing and engaging patients through patient portals located in EHRs;
- facilitating identification of patients who are eligible for trials, better coordination of their health information, and enabling their participation in clinical trials;
- accelerating broad access to new medications; and
- promoting precision medicine in community care settings.
Several workshop participants also discussed the importance of patient-reported outcomes and related ethical considerations in patient-centered drug development. Rebecca Pentz, professor of research ethics at Emory School of Medicine, suggested developing and deploying new health applications that facilitate collection of patient-reported outcomes, such as the side effects of their treatments. “We are in a new era of social media and apps. That is what people are going to be using, so we are going to have to adapt and [collect] adverse events that way with the proper controls,” she said. Jeffrey Brown, associate professor in the Department of Population Medicine at the Harvard Pilgrim Health Care Institute at Harvard Medical
School, pointed out that many health apps do not have privacy policies and patients are not aware of this. “We . . . need to use health apps because they are the future, but we have to think about some of these privacy issues,” Brown said.
Brown said that ethicists have differing opinions on whether and how patient consent should be obtained for the use of their data for research purposes. He noted that some ethicists view it acceptable to use some patient data in research without consent when there is minimal risk to the patient or in cases of public health activities. Some ethicists do not believe that any patient data should be used for research purposes without express patient consent, Brown said. Other ethicists propose that the default option should be to allow patient data to be used in research, but to offer patients the opportunity to “opt out” of using their data in studies. “There is debate about this in the ethical field and we need to move forward, come to consensus, and come up with a solution,” he said.
Brown noted that obtaining informed consent from patients for studies that use data collected from EHRs and other sources of real-world data can be challenging. He suggested it might be feasible to have patients agree that if they enter a hospital or other medical facility that uses EHRs, they automatically agree to share their EHR data for research purposes. “That may be needed for a pragmatic, clustered randomized clinical trial,” he said. He noted that garnering such consent is currently difficult to accomplish because of current regulations that restrict access to patient information for research, such as the Privacy Rule promulgated under the Health Insurance Portability and Accountability Act (HIPAA) (IOM, 2009).
Brown suggested using broad consent forms for research on patient data in EHRs, similar to those used for patient consent for future research on patients’ biospecimens. He said that many patients do not have concerns with such broad consent forms, with the exception of a minority of patients who have expressed the concern that others will profit from the use of their biospecimens in research (Grady et al., 2015).
John Burch, a member of the Mid-America Angels Capital Investment Network, suggested using patient-centered repositories of patient data. “Not the clinician’s EHR, but the patient’s own EHR,” he stressed. A patient could put his or her data into the repository, own and control the use of the data, and could make the data available for research, he said. Pentz noted that this
could serve the patient well, and there would be an ethical advantage to having patient-centered repositories of data. “From an ethical standpoint, it is the patient’s information and they would be empowered to use and control it,” she said.
Joffe noted that there is increasing pressure to accelerate cancer drug development in order to provide patients who have life-threatening diseases with earlier access to promising new therapies. But he said an analysis of more than 100 Phase I cancer clinical trials found that faster approaches to escalating doses in studies was linked to increased toxicity rates, without improving efficacy (Koyfman et al., 2007).10 “In this case, it was hard to make the argument that these accelerated designs were better for patients than traditional designs,” Joffe said.
Joffe added that reports of life-threatening autoimmune reactions to immunotherapies have recently emerged, many of which did not become apparent until these agents entered the market after accelerated FDA approval.11 He cautioned that “as we move more quickly and have smaller numbers of patients involved in our drug development programs, we are going to be much more reliant on the postmarketing period to get a sense of what the safety data are.” He noted that the 21st Century Cures Act has provisions aimed at speeding FDA approval of new drugs, but cautioned that there are ethical concerns about whether accelerating cancer drug development will lower the bar for patient safety (Kaplan, 2016). Brown suggested that the goal should not simply be acceleration of drug development, but rather optimizing data collection and getting the right data at the right time. “We have to get the right data for the right patient without putting them at undue risk,” he said.
Joffe stressed that the push for speed and accelerated approvals “has trade-offs we can decide to make, and by ‘we’ I do not just mean the clinicians, investigators, and statisticians, but also the patients and the broader community. We can decide to accept those increased risks to get that speed.”
10 The study results may not be applicable for newer cancer therapies, such as targeted therapies and immunotherapies.
11 See https://www.nytimes.com/2016/12/03/health/immunotherapy-cancer.html (accessed May 21, 2017).
However, the ethical considerations regarding patient risk are complex because patients vary in how much risk they are willing to assume in return for a possible benefit, said Joffe, Raju, Rockhold, and Craig Tendler, vice president of Late Development and Global Medical Affairs in Oncology at the Janssen Pharmaceutical Corporation. Raju noted that regulators make decisions on whether to approve drugs based on results in populations of patients, but individual patients “should be doing their own benefit–risk analyses together with their clinicians and their circumstances.” He also said that individual patient preferences for and tolerances of risk vary over time.
Matrisian agreed, and added that how fast a disease is progressing and how serious it is can alter patients’ perspectives of their risk and whether the risk of the drug is greater than the risk of having the disease itself. She noted that patients with the most lethal cancers, which have 5-year survival rates of less than 50 percent and cause more than half of U.S. cancer-related deaths each year, could have a very different perspective on the benefit–risk ratio than patients with cancers that have a greater than 90 percent 5-year survival rate.
Mary Redman, associate member of the Clinical Research Division at the Fred Hutchinson Cancer Research Center, suggested more time be spent on the discovery phase of drug development in order to better understand the biology that underpins a therapy’s mechanism of action. In addition to early testing aimed at determining dose and the patient population most likely to respond to a treatment, Kenneth Anderson, director of the Lebow Institute for Myeloma Therapeutics and Jerome Lipper Multiple Myeloma Center at the Dana-Farber Cancer Institute, suggested that researchers conduct more preclinical modeling studies with genetic analyses to guide later clinical trials. For example, such biologic explorations have informed drug development by assessing which patients with breast cancer are most likely to respond to an epidermal growth factor receptor (EGFR) inhibitor, and by providing a better understanding of the ways in which various compounds block the growth of multiple myeloma cells in the bone marrow microenvironment of the tumor. In addition, Weber said that functional imaging can confirm mechanisms of action in both mouse models and in patients by showing, at an early stage of drug development, whether a drug is reaching its target and affecting tumor growth.
A number of cancer therapies target specific genetic mutations that drive cancer growth, such as EGFR inhibitors. Some of EGFR inhibitors initially failed in clinical trials because they were not effective in a broad population of patients, said Richard Finn, associate professor of medicine at the David Geffen School of Medicine at the University of California, Los Angeles. For example, one EGFR inhibitor, gefitinib, was initially not shown to be effective in clinical trials for patients with lung cancer. Further research demonstrated that gefitinib did improve progression-free survival in patients whose tumors had specific EGFR mutations compared to standard chemotherapy (Maemondo et al., 2010). In 2015, FDA approved gefitinib as a first-line treatment for patients with metastatic non-small cell lung cancer whose tumors harbor specific genetic mutations.12
Instead of repeating this pathway of development, Finn said that the development of palbociclib—a selective inhibitor of the cyclin-dependent kinases (CDK) 4 and 6—began by proactively determining the types of breast cancer cells that are most likely to respond to this therapy prior to initiating clinical trials. By testing a large library of cell lines, Finn and his colleagues demonstrated that estrogen receptor-positive (ER-positive) breast cancer cells responded best to palbociclib. They also demonstrated that the therapy interacted synergistically with antiestrogen treatments, suggesting breast cancer patients receiving antiestrogen treatments would receive additional benefits if they were to receive palbociclib as well (Finn et al., 2009).
These findings led to a Phase I study conducted in 2008. In 2009, the Phase II study known as PALOMA-1/TRIO 18 was initiated, which randomized women with estrogen receptor-positive, human epidermal growth factor receptor 2 (HER2)-negative advanced breast cancer to either palbociclib plus letrozole13 versus letrozole alone. Palbociclib plus letrozole, compared with letrozole alone, extended median progression-free survival by approximately 10 months (Finn et al., 2015). Shortly after the trial ended in 2015, Pfizer Inc. received accelerated approval of palbociclib as a first-in-class CDK inhibitor for treatment of ER-positive breast cancer, in
12 See https://www.fda.gov/NewsEvents/Newsroom/PressAnnouncements/ucm454678.htm (accessed May 21, 2017).
13 Letrozole is an aromatase inhibitor that inhibits the growth of estrogen-dependent breast cancer cells. See https://www.cancer.gov/publications/dictionaries/cancer-drug?CdrID=42086 (accessed May 23, 2017).
combination with letrozole. This has since been confirmed in a Phase III study (Finn et al., 2016).
During the Phase II study, Finn recalled the deliberation by the sponsoring team to determine which patients to enroll. They were concerned that they might not see a significant effect if all patients presenting with ER-positive cancer were enrolled. At the same time, they thought they might miss a treatment benefit if they selectively enrolled patients with mutations affecting proteins that interact with CDKs and might be driving cancer growth. Thus, they included two sequential cohorts in PALOMA-1/ TRIO 18: the first cohort included patients with ER-positive/HER2negative breast cancer, while the second cohort included only patients with ER-positive/HER2-negative breast cancer who also had amplification of cyclin D1, loss of p16, or both.
Finn pointed out that if they had limited patient selection in initial trials of palbociclib to only patients who had these additional biomarkers, the study would have generated positive outcomes, “but we would have missed the other 80 percent of patients who might benefit from a compound like this.” These deliberations about defining the right study population is relevant to the design of future trials, and is reflective of the choices that sponsors must make in drug development, said Finn and Mace Rothenberg, chief development officer of oncology at Pfizer Inc.
Finn noted that since palbociclib’s approval, several other CDK inhibitors have progressed from the laboratory to Phase III studies, including ribociclib (Hortobagyi et al., 2016).14 “The greatest advances in patient outcomes have come from integrating biology into clinical practice and critical use of preclinical models can guide that process. They can help us design hypothesis-driven Phase I/Phase II studies,” Finn concluded.
Anderson reported on how multiple myeloma has served as a paradigm for targeting the microenvironment of cancer cells. Because the cancerous cells in this blood cancer can be routinely accessed, researchers can more easily model interactions between the tumor and the host, he said. This has enabled discovery and validation of novel agents, alone and in combination, that can overcome conventional drug resistance. Since 2003, FDA
14 Ribociclib was approved by FDA in March 2017. See https://www.fda.gov/drugs/informationondrugs/approveddrugs/ucm546438.htm (accessed May 21, 2017).
has approved 18 new drugs for multiple myeloma, and patients with multiple myeloma are living three to four times as long as they did previously, Anderson reported. “It is fair to say that there are some patients now who end up having a normal life span,” he said.
In traditional drug development, academic laboratories complete basic cancer biology research that can suggest new pathways and mechanisms of action for how cancer grows and spreads, Anderson explained. Industry then discovers new agents that act on these pathways and develops these agents further with preclinical testing before working with academia to conduct clinical trials. However, the development of innovative drugs for multiple myeloma has instead used a more efficient, collaborative approach, with academia conducting more drug discovery and validation work along with input from patients and industry. This has resulted in quicker translation of treatments to clinical practice, Anderson said.
For example, Anderson’s lab showed that lenaliodmide, a derivative of thalidomide, affects the tumor microenvironment primarily by stimulating the host’s immune response against the cancer, but also by blocking adhesion of cells and inhibiting the generation of blood vessels. Research also demonstrated that triggering protein degradation mediated thalidomide’s effects on tumor cells. These findings fostered the development of additional drugs targeting protein degradation in different areas of the cell (Krönke et al., 2014; Lu et al., 2014; Winter et al., 2015).
Conversely, by focusing on the importance of preventing protein degradation in order to destroy cancer cells using proteasome inhibitor bortezomib, investigators were able to devise drug combinations that could inhibit that protein degradation in multiple ways and achieve higher efficacy in some patients. Anderson said that such combination approaches with different classes of drugs that have different mechanisms of action have been quite effective in multiple myeloma. “Combinations of novel agents, predicated [by] preclinical science, have achieved unprecedented results,” said Anderson. For example, when bortezomib was combined with histone deacetylase (HDAC) inhibitors to block both proteasomal and aggresomal protein degradation, dramatic suppression of multiple myeloma tumor cell growth occurred (Catley et al., 2006). This finding led to a Phase III trial testing the combination of an HDAC inhibitor with bortezomib, and the first FDA-approved HDAC inhibitor, panobinostat (Richardson et al., 2013; San-Miguel et al., 2014).
Anderson said that although immunotherapy directed at blocking the checkpoint inhibitor programmed death-1 (PD-1) has not been effective
in multiple myeloma, preclinical research suggests that combinations with lenalidomide may drive a response (Görgün et al., 2015).
Anderson noted that some mechanistic preclinical findings are relevant not only to multiple myeloma, but to other cancers as well. For example, preclinical findings suggest that there is synergy in combining a proteasome inhibitor with protein kinase B inhibitors—also known as Akt inhibitors—in multiple myeloma, prostate cancers, and other cancers. “The correlative science can inform going forward clinically, not only in one disease, but more broadly,” he said.
Weber discussed the use of functional imaging as a diagnostic tool to assess tumor response to treatments. He reported on the preclinical and clinical uses of functional imaging to assess tumor response to therapeutic agents, including assessment of tumor characteristics (e.g., blood perfusion and glucose metabolism) through magnetic resonance imaging (MRI) and positron emission tomography (PET). Weber stressed that functional imaging can overcome many of the limitations of traditional imaging approaches and RECIST criteria. Unlike traditional imaging, functional imaging can distinguish between scar tissue and viable tumor tissue; detect tumors in bone and other areas that are difficult to image with computed tomography (CT) scans or traditional MRI; quantify relevant physiological and biochemical changes; and show tumor responses more quickly, Weber said (Johnson et al., 2016). In addition, PET also can be used to assess whether an agent has hit its target, he added.
Because of the widespread use of PET-CTs for tumor staging, imaging technologies and the infrastructure to make imaging agents are also available throughout the country, Weber added (Buck et al., 2010). Weber noted that because PET infrastructure is in place at most institutions, the technology is primed for application in drug development. “We now have the opportunity to go beyond glucose imaging [with fluorodeoxyglucose] and look into molecules that are of interest in drug development,” he said. Traditional imaging is generally done infrequently (e.g., before therapy begins and then several months later), leading to a loss of information about the tumor status and changes in growth between those two widely spaced time points. By contrast, innovative PET imaging can document details during that interim time period, such as the location of the therapy and its concentration at tumor sites, and whether it interacts with the intended
target. While some of that information can be determined by examining a patient’s biopsied tumor tissue shortly after therapy, the advantage of PET imaging is that “you can do whole-body imaging studies [at different time points] and look at not only one site, one small biopsy, but all metastatic lesions in a particular patient,” Weber said, with radiologic technologists able to complete whole-body scans of patients within 15 minutes.
He said that theranostics, which he defined as use of the same or a very similar compound to both image and treat tumors, can maximally take advantage of PET technology. For example, he discussed a theranostic that combines a radioactive compound with an agent targeting prostate-specific membrane antigen. This theranostic agent led to the disappearance of bone metastases after four cycles of treatment in a patient with metastatic prostate cancer (Heck et al., 2016), shown by PET scans taken after each treatment cycle (see Figure 2). “It is a striking example of how effective this targeted radiotherapy can be if you have the right target and the right drug to treat it,” Weber said.
Weber also noted the ability to use PET for optimizing drug dosage by demonstrating in patients the concentrations of drug at the actual tumor site. This could help distinguish among several drugs in the same class, such as the PARP inhibitors olaparib and iniparib. Iniparib failed in Phase III clinical trials (Mateo et al., 2013), and PET imaging later showed that the molecule does not inhibit PARP at clinically relevant doses at the tumor site while olaparib, which has FDA approval for the treatment of ovarian cancer, is able to do so (Michel et al., 2017). Similarly, the technique has been used to determine biologically relevant doses of androgen- and estrogen-receptor inhibitors in Phase II clinical trials (Rathkopf et al., 2013; Wang et al., 2016).
Weber also stressed that such molecular imaging could serve as a companion diagnostic that aids in the selection of patients for cancer treatments. “The technical feasibility to develop a molecular diagnostic is no longer the limiting factor,” he said. Weber noted that there is a large toolbox of cancer-relevant molecular diagnostics, but use in clinical care requires validation and FDA approval for in vivo companion diagnostics, and greater participation of cancer centers with the ability to develop or use these molecular probes in clinical trials, according to Weber. Having oncologists with experience in diagnostic radiology and nuclear medicine will also be critical for translating this technology into clinical care, he added.
Anderson and Weber stressed the need for standardized criteria for use of imaging as biomarkers to enable uniform, consistent interpretation
of results that can accurately inform clinical trial designs and patient care. Weber noted that such criteria have been developed for PET imaging of tumors—PET Response Criteria in Solid Tumors (PERCIST)—and could be used in clinical trials (Wahl et al., 2009).
Several workshop participants discussed the need to develop new endpoints to assess therapeutic responses more quickly, compared with current endpoints used in cancer clinical trials, such as progression-free and overall survival. For example, Anderson said that patients with multiple myeloma may have a progression-free survival rate of 7 to 10 years when they are
treated with a three-drug therapy. If new therapies improve their survival rate by 50 percent, it will take more than a dozen years to see that improvement in a clinical trial. “Patients cannot wait for that and we cannot afford these long trials,” he said. “We need a metric that will be able to tell us at 18 months what the outcome will be at 10 years,” he added. However, Joffe cautioned that although use of surrogate markers could make drug development more efficient, there is a risk of misleading outcomes: “There is a trade-off between careful and rigorous evaluation and speed of time to approval and clinical use,” he said.
Redman said that more time should be spent identifying and validating intermediate and surrogate endpoints in order to obtain results earlier and to decrease the numbers of patients needed to complete trials. She suggested designing studies with endpoints that reflect both efficacy and futility, which is the inability of a clinical trial to show statistically significant changes in outcomes between treatment and control arms. She stressed evaluating futility separately from efficacy because there are different statistical standards and methodologies to demonstrate that a trial is unable to differentiate any effect versus a trial definitively showing no effect of the treatment.
A number of workshop participants discussed the need to have the flexibility to choose appropriate endpoints based on the type of trial, treatment, and disease. Keith Flaherty, director of the Henri & Belinda Termeer Center for Targeted Therapy at the Massachusetts General Hospital Cancer Center, said that for BRAF-targeted treatments, measuring complete responders—and not partial responders—more closely predicted patient outcomes. Flaherty therefore suggested taking into consideration the disease area, mechanism of action, and treatment type to better inform endpoint choice: “Eighteen percent of the patients had complete response to BRAF/ MEK combination therapy. So for us, I have made the case that we focus on complete response rate in early studies of BRAF/MEK treatments [when making decisions about which agents to test further], but I am not suggesting that is going to work for immunotherapy combinations,” Flaherty said. He added that “specificity by mechanism [and] by disease are variables that are going to expand and get more complicated, and I do not think we should be trying to create a single platform.”
Patricia Keegan, director of FDA’s Division of Oncology Products 2, added that response rate is more appropriate for assessing immunotherapies in melanoma than for lung cancer, the latter of which might need durability of response or progression-free survival endpoints. Anderson added that the Foundation for the National Institutes of Health (FNIH) recently initiated
a public–private collaboration among basic and clinical researchers, officials at FDA, and representatives from industry to develop a white paper15 on the use of minimal residual disease in multiple myeloma as an alternative endpoint to progression-free survival. Companies are providing de-identified data from clinical trials to create a database that should enable the group to develop such a metric, Anderson reported. Ratain noted that researchers have successfully modeled early outcomes in randomized clinical trials that suggested “time-to-tumor-growth” was a valuable outcome measure to assess (Claret et al., 2009; Wang et al., 2009).
Many workshop participants discussed the concept of modeling in a variety of contexts throughout the workshop. Several participants discussed how process modeling can help researchers predict the optimal drug development pathway for a therapy, as well as dosage modeling to determine optimal dosing for a therapy or combinations of therapies. In addition, DSMBs and FDA can use benefit–risk modeling to better ascertain the potential benefits and risks of an investigational new drug when making decisions about whether clinical trials should proceed, and whether it should receive FDA approval for use in clinical practice.
Finn cautioned that models are subject to inherent biases and need to be carefully constructed. For example, he pointed out that the development of EGFR inhibitors included preclinical models that used tumor cell lines that were highly dependent on epidermal growth factor. Finn said that these cell lines were not representative of the range of tumors seen in vivo, and that the EGFR pathway is much more complex than initial modeling suggested.
Piantadosi suggested the use of process modeling to help determine how a drug should proceed through the development pipeline, particularly for combination therapies. He noted that the overall drug development process can be characterized using Bayesian probability statistics, such as estimating the statistical power and error in each individual step in the drug
15 See http://clincancerres.aacrjournals.org/content/early/2017/04/26/1078-0432.CCR-16-2895 (accessed May 22, 2017).
development process. He pointed out that although randomized Phase II trials have become quite popular, researchers “should be careful because if they are designed with poor error properties, they can degrade the drug development pipeline.” He noted that “the optimal pipeline does not necessarily result from a sequence of seemingly optimal individual clinical trials,” because the more steps there are, the more opportunities there are for statistical error and lack of reliability due to underpowered studies. Ratain said that with good process models, “one can make drug development decisions sooner” and noted the need for more quantitative analyses to inform oncology drug development (Claret et al., 2009).
Piantadosi said that the best dosing designs are guided by mathematical dosage modeling, especially for combination therapies. “[Dosage] models are essential because they embody knowledge from outside the experimental realm, enabling incorporation of key ancillary information into dose finding,” such as pharmacokinetic data or patient characteristics, he said.
Piantadosi suggested that dosage modeling can be used to help titrate a therapy’s dose to meet a prespecified outcome. He also suggested that researchers use dosage modeling to ascertain the range of active and tolerable doses, and to define the minimum active dose rather than the maximum tolerated dose. “Our old clinical designs were not defective, but were highly optimized for the types of questions we had for agents in development then. New questions require new designs,” Piantadosi said.
Piantadosi noted that optimizing dosing for combination therapies is particularly complex because it requires determination of dose–response curves for multiple agents (Tighiouart et al., 2014, 2017). He added that this can require approximately four times more study participants compared to a standard dose escalation trials for a single agent. “Our conventional dose escalation approaches are unlikely to adequately solve the problem,” he said, and added that investigating interactions between or among drugs in combination therapies is expensive, so investigators need to know in advance whether interactions are expected and design their clinical trials accordingly. “With drug combinations, we will need larger designs and we have no alternative but to pay the price when those drug interactions are clinically important,” he said.
He noted that there is no standard dosage modeling approach for such dosing, and gave examples of several possibilities. One approach,
called envelope simulation, uses regions of possible doses on a hypothetical dose–response curve, and is continually updated as real data accumulate. He added that simulated envelope data gradually lose their influence on the predicted outcome as it is populated with actual data. With sufficient data, the distribution of peak dose estimates gradually narrows with high precision. This method can be generalized to more than one therapy, provided investigators can construct a plausible dose–response curve, Piantadosi said.
Another dosage modeling approach is the surface design method, which is widely used to optimize multiple variables in industrial or chemical processes, but is not often used in clinical trials, Piantadosi said. Surface design relies on systematically generating responses as control variables are changed, and fitting the responses on a flexible surface to determine optimum dosing. Piantadosi said this method is simple, reliable, and can be applied using a minimal sample size at each design point (Myers et al., 2016). Researchers used the surface design method to optimize the doses of doxorubicin and cyclophosphamide and a tumor vaccine to produce the highest response in patients with breast cancer (Emens et al., 2009) (see Figure 3).
A third dosage model, factorial designs, is another underused tool for studying combination therapy dosing and is well-suited to studying treatment interactions, Piantadosi said. These factorial designs are especially efficient and require small patient sample sizes when no interactions between the treatments are being tested simultaneously, Piantadosi noted.
“With [dosage] models, one can begin to make predictions about what would happen with changes in dose and schedule, which I think are absolutely critical,” Ratain said. Sigal added that as the information gathered becomes more complex with combination therapies, the ability to integrate that information into dosage models will help researchers and regulators make sense of the totality of evidence. Flaherty noted that when modeling the dosing of combinations, one cannot assume “each drug is pulling its own weight the same way. We need to be biased in our dose exploration.” For example, Flaherty pointed out that for MEK/PI3K-targeted combination therapy, giving equal weight to the contribution of each targeted treatment in the combination does not make sense because both preclinical and clinical studies have found that the PI3 kinase inhibitor is ineffective as a single agent (Garrett et al., 2011).
Keegan said, “We are not doing a good job of picking optimum doses despite about half a decade’s worth of experience with continual reassessment models for picking doses. Nobody is really taking a good look at
whether that has actually yielded better dosing information. We probably now have a critical mass of information that would tell us if we are better at picking doses with this type of model[ing].”
Benefit–risk modeling can help DSMBs assess whether to continue a trial after interim results have been analyzed by examining ethical considerations, such as whether the benefits of the trial outweigh potential risks to patients. Joffe suggested benefit–risk modeling should account for the properties and the performance of various study designs under multiple sets of assumptions in order to better protect trial participants and future patients. “This is work that statisticians do that I think is going to be very important,” he said.
Joffe said that assessing the risks of harm and potential benefits of a drug agent has often relied on the clinical judgment of experts on the DSMB, but benefit–risk modeling during an ongoing trial could help the DSMBs integrate the vast amount of data they need to review and lend more rigor to their discussions. He noted that researchers have made recent advances in quantifying and modeling predictable harms and benefits in clinical trials, and he suggested that leveraging these advances would help to foster more rigorous and reproducible decision making within DSMBs. He added that new methods to prevent or treat a harm, or to identify patients who are not at risk of developing severe side effects from a treatment, could be incorporated in these benefit–risk models with updated appropriate weighting.
Raju noted that FDA weighs risks and benefits when assessing whether to approve a therapy, but these assessments traditionally have been made from a qualitative review of the evidence. He suggested that FDA use a more quantitative approach by estimating the impact a therapy will have on both the length and quality of life for a patient (Raju et al., 2016a). For example, he performed a benefit–risk analysis on 22 FDA decisions for drugs used to treat non-small cell lung cancer (Raju et al., 2016b) (see Figure 4). In this analysis, he calculated the estimated benefit–hazard ratio for each therapy’s primary endpoint, the estimated benefit of each therapy, and total risk exposure compared to the control arm, as well as FDA’s decision for each therapy. He found that the analyses of estimated risks and benefits were able to distinguish among the therapies that had been FDA approved from the therapies that had been approved through the accelerated approval pathway but were later withdrawn from the market. Raju has conducted those analyses in nine other cancers, but noted that the quality of data for adverse event reporting limits the benefit–risk model. He said that the benefit of conducting these analyses is that it provides the rationale for FDA decision making, and added that this type of benefit–risk modeling could also inform earlier development decisions and trial designs, as well as at later stages when more data have been collected.
Redman stressed that clinical trials for cancer drug development need to include an explicit plan for biomarker analysis. Woodcock added that “more formal exploration of the performance of [biomarker-based] diagnostics predicting patient response [should] be incorporated into clinical
trials of therapeutics. Understanding that diagnostic and its performance is going to be critical going forward.”
Redman suggested several different approaches to incorporate biomarkers into a clinical trial design. Investigators can use a standard trial design in which all eligible patients are enrolled in the trial and the primary analysis is conducted in an unselected or minimally selected population. Secondary analyses of the trial would then include biomarker evaluation. She added that investigators use this design when a treatment is expected to have an effect in the overall patient population, but there are candidate biomarkers they wish to evaluate further.
When there is a strong rationale for using a biomarker to predict patient response in which only the subgroup of patients with that biomarker is likely to benefit from the treatment, Redman said that subgroup-focused designs may be a better option. She described three clinical trial designs that include biomarker analysis as a main objective (see Figure 5). The first approach is to include all study participants in the primary evaluation, but to include an analysis of the treatment effect in a biomarker-defined subgroup as a coprimary objective of the trial. Another option is to evaluate treatment response within a biomarker-defined subgroup of patients and compare it to the entire study population. Redman said that the final option, called the targeted or master protocol design, is where investigators specify biomarker-defined patient subgroups and treatment response is eval-
uated in each specific subgroup. If multiple biomarker-defined subgroups are part of the trial design, they are treated as independently conducted trials, Redman said.
Redman cautioned against using clinical trial designs that narrow patient participation by biomarker status if they may exclude patients who could benefit from the treatment being tested. She noted that most biomarkers are not binary categories (i.e., a patient either has a biomarker or does not have a biomarker). Instead, most biomarkers tend to be continuous variables, with varying amounts of a specific biomarker in a tumor sample. Because of this, Redman said that it can be difficult to establish appropriate thresholds for biomarker-based patient selection in clinical trials.
Another challenge with biomarker-determined subgroup analyses of clinical trial data is that the smaller size of each subgroup decreases the statistical reliability of the data, Redman noted. In a hypothetical Phase I clinical trial involving 100 patients, even if the prevalence of a biomarker is 50 percent, there is poor statistical power to detect a 20 percent difference in response rates between biomarker-negative and biomarker-positive patients (see Table 3). Even within a Phase III study with 400 patients, either large responses or prevalent biomarkers are necessary to detect differences in treatment response by biomarker status, she stressed. She said that investigators should design their biomarker studies with sufficient statistical power by considering the prevalence of a biomarker and expected treatment response, which will help determine the size of the patient population needed for the clinical trial.
|Percentage with Marker||N-Marker Positive||Power to Detect Difference Response Rate of:|
NOTES: Based on an example with response rate of 10 percent in “marker negative” group. One-sided t-test at the p = 0.05 level; N = 100.
SOURCE: Redman presentation, December 12, 2016.
Some workshop participants discussed the challenge of deciding when to use biomarkers in a study, including how to identify all patients who might benefit from a treatment, as well as consideration of resources and timing to develop a companion diagnostic. Suzanne Topalian, director of the Melanoma Program at the Sidney Kimmel Comprehensive Cancer Center at Johns Hopkins University, inquired about the appropriate timing of the development of a companion diagnostic: “Should this be done from the very beginning or do we run the risk of excluding patient populations that might benefit?” Rubin said that it takes about 1 year to validate a promising biomarker test in the laboratory, which is a risky time investment because the drug may not go to market, so “you are not really sure you need the biomarker.” He added that despite best efforts, diagnostic test development often lags behind that of the therapeutic (NASEM, 2016a), even within companies that have divisions for specialty diagnostics. Rubin’s approach has been to conduct standard trials but ensure that these trials have enough statistical power to address prespecified hypotheses regarding a potential biomarker subgroup of interest. He added that determining which biomarkers are appropriate evolves over time as more information is generated from studies, with some single biomarker tests now being replaced by genomic biomarker signatures.
Rothenberg framed the question of when to begin diagnostic test development as: “Do we know enough [about the biomarker test] to pull the trigger and modify the clinical trial design? If we do, are we missing out on a patient population who can benefit? If we do not, are we diluting the effect in the selected group of patients so it will be a false-negative study?” Rothenberg stressed that a number of factors influence this decision, including medical, scientific, ethical, regulatory, and payer issues: “All of these things come to bear in trying to [devise] the best possible decision given the information that you have, and knowing the field is moving rapidly forward. The longer you wait to make that decision, the more you are making a decision not to do it.”
Flaherty said that prospectively studying biomarkers is better than studying them retrospectively. For companies with limited resources, he suggested designing studies whose primary objective is to determine efficacy of a treatment in patients whose biomarker status suggests that they are likely to respond to the intervention “and then circle back later to explore whether a broader population might respond.” However, he suggested that companies with more resources conduct larger trials in which there is enough statistical power to detect whether certain biomarkers in subgroups
of patients predict response to treatment. “That is the ideal solution, but I do not see that being executed across much of the field,” Flaherty said.
Finn noted that the PALOMA-1/TRIO-18 clinical trial changed patient enrollment criteria midway into the study to specifically select only ER-positive and HER2-negative patients who had biomarkers for amplification of cyclin D1, loss of p16, or both (Finn et al., 2015). He said that this change slowed down the study, but investigators were concerned that if the study did not limit patients by biomarker status, there may not have been sufficient numbers of responsive patients to discern efficacy of the treatment. If he were to redo the study, he said he would have included all ER-positive and HER2-negative patients, and then stratified them by their cyclin D1 and p16 biomarker status and conducted subgroup analyses.
Several speakers highlighted the lack of standards for complex omics-based tests increasingly being used in cancer drug development (IOM, 2012). “It is a little bit of the Wild West, but validation of these tests is incredibly important because if we are going to make informed decisions on treatment for patients, it is important these tests work,” Sigal said. “We all need to operate using the same standards,” she stressed, and added that the current standards of the College of American Pathologists (CAP) and Clinical Laboratory Improvement Amendments of 1988 (CLIA)16 are not sufficient for these complex tests.
Woodcock noted that one of the challenges is that companies, academic institutions, and oversight bodies have not made it a priority to ensure diagnostic test validity: “Patients will be disadvantaged by the fact that [there is no ownership of these diagnostic tests], and there is not the kind of round-robin concordance and proficiency [testing] that needs to happen—nobody establishes reference standards and other things that are needed. It is a huge problem, but there is not one entity that owns this problem.” Sigal suggested that CAP and FDA work together to achieve concordance in standards for these complex diagnostic tests “because it is really going to hurt everyone if we do not.”
Ratain noted that the molecular pathology community has focused on technical issues in the development and analytical validation of biomarker tests, but there are other issues with clinical implementation and quality assurance, such as whether tumor samples are tested consistently at the same stage of diagnosis, because metastatic samples are likely to yield dif-
16 See https://www.cms.gov/Regulations-and-Guidance/Legislation/CLIA/index.html?redirect=/CLIA (accessed May 24, 2017).
ferent results than early-stage tumors. “We really do have to move beyond CAP and CLIA, and there are many companies selling diagnostic tests from CLIA-certified laboratories that have little or no clinical validity, and payers are paying for them,” Ratain said.
Flaherty described the efforts that were undertaken in advance of the National Cancer Institute-Molecular Analysis for Therapy Choice (NCI-MATCH) clinical trial to ensure validation and standardization of the test used to assess patients’ biospecimens for genomics biomarkers at four different institutions (see section on Innovative Clinical Trial Designs). This entailed using the same testing platform in each laboratory and having all the laboratories involved undergo proficiency testing using the same reference samples, with oversight by the FDA Center for Devices and Radiological Health, Flaherty said. Near the end of the NCI-MATCH trial, two for-profit laboratories underwent the same proficiency testing so they could also be involved with the study. The end result was a high degree of concordance across these academic and commercial laboratories using a platform that tested hundreds of genes, he said.
Many participants discussed innovations in clinical trial designs for cancer drug development. These include adaptive protocols that are amended based on new findings, seamless protocols that more easily transition between the different phases of drug testing, master protocols that enable the testing of multiple drugs and/or biomarkers in multiple diseases simultaneously, and clinical trials with common control arms. Many features of these designs can overlap and be included in a single clinical trial. In addition, some workshop participants discussed the unique challenges faced in innovative, seamless designs (see section on Issues to Consider with Innovative Clinical Trial Designs). The three categories of challenges highlighted and discussed were determining when and if a trial should progress by defining endpoints and appropriately using statistical analyses; decision making and oversight, particularly when those decisions need to be made rapidly for complex trials; and evaluating benefit and risk by balancing potential advantages with incomplete information, taking advantage of postmarket learning, and deciding when single-arm studies are warranted.
Rajeshwari Sridhara, director of the Division of Biometrics V at CDER, reported on the different types of adaptive study designs that investigators have used to assess clinical tests. Adaptive studies can be broadly grouped into two main categories. One category—sequential designs—involves analyzing and reviewing the data at periodic prespecified time points in a study and deciding whether to continue the trial based on the results. At these points, investigators can also decide to amend the study to increase the sample size or to narrow the clinical trial to a subset of patients. The second category—Bayesian design—uses probability theory to assess the likelihood that the accruing trial data suggest a trend that require adjustments to the study protocol.
A subcategory of adaptive study designs is an adaptive enrichment protocol that uses biomarkers to predict response to a therapy. With adaptive enrichment trials, all patients are tested for a potential predictive biomarker when they are treated with a therapeutic agent or standard of care. If an interim analysis finds that the biomarker predicts response to treatment, then the biomarker is used to stratify patients—those who are biomarker negative are no longer accrued into the trial, while those who are biomarker positive are randomized to the therapeutic agent or standard of care.
Sridhara stressed that an adaptive design by definition enables preplanned study design modifications based on blinded or unblinded interim results. Adaptive designs can be used for exploratory or confirmatory studies. With exploratory adaptive studies, “you can explore as much as you want without adjusting for multiple looks [or] multiple adaptations to generate hypotheses to be further tested. There is a lot of leeway here,” Sridhara said. By contrast, she said that confirmatory adaptive studies require careful planning to specify decision rules for each trial adaptation and to ensure that they are statistically valid, Sridhara reported. “We are not saying you cannot have many adaptations, but you need to prespecify your threshold when you are thinking of this adaptation,” she said, emphasizing that the requirement is for specifying the thresholds that would alter the course of the treatment, rather than specifying the expected change. Piantadosi noted that the rationale for prespecifying whenever possible is to reduce bias and statistical errors in subsequent statistical analyses of the trial data.
Richard Schilsky, senior vice president and chief medical officer of the American Society of Clinical Oncology (ASCO), noted that having prespecified adaptive designs can be challenging or counterintuitive because “part
of the issue of being able to adapt your trial design might be in response to information that you cannot foresee or anticipate and only become aware of it during the course of the trial.”
Finn added that the need to prespecify depends on the context of the study. If a randomized Phase II trial that is intended to collect data that can be used to mitigate risk in a subsequent Phase III trial has a strong hypothesis, scientific rationale, and trend, it would be acceptable to not preregister adaptations, he said. David Feltquate, head of early clinical development in oncology at Bristol-Myers Squibb, noted that analysis of results from single-arm studies often preempts design changes to ongoing randomized studies, such as a change in endpoint or study size, illustrating the practice of making an adaptation that is not prespecified.
Keegan said that prespecifying adaptations can increase efficiency and enable more appropriate decision making. She said that evaluating results at specific time points to assess whether the hypothesis needs to be altered is more appropriate than an open-ended enrollment without prespecified cohort sizes and evaluation time points. For example, an early prespecified assessment of the use of response rate as an endpoint could indicate that progression-free survival may be a more reliable endpoint, and in an adaptive trial this change can be made, she said.
Rockhold suggested using what is known about the mechanism of action of agents to model and predict potential results that would require modification of a study design. If predictive modeling is not performed, investigators respond reactively to unexpected results because no response procedure has been defined, he said.
Feltquate noted that although adaptive designs aim to make trials more efficient, the embedded feedback loops in which new data are analyzed and findings are incorporated into study amendments can be time consuming. “Our experience is that it adds an enormous amount of time onto the learning that goes on, unless you choose a surrogate endpoint, such as functional imaging, that can give you a much quicker answer. Otherwise, I do not see adaptive trials as having a lot of utility, but instead they have an opportunity to provide false negative signals, causing us to stop something too early in the testing process,” he said.
Several workshop participants described opportunities to create a more seamless drug development paradigm. Theoret illustrated a seamless drug
development paradigm that blurs the traditional phases of development, and instead includes overlapping evaluation of pharmacology, exploratory testing, and confirmatory testing, including an earlier assessment of efficacy in the process (see Figure 6). He noted that FDA is receptive to considering opportunities to use seamless drug development designs in appropriate circumstances, and noted that he and other FDA colleagues had recently authored a perspective with a conclusion that a “desire to provide earlier access to highly effective drugs should encourage further use of seamless expansion-cohort trials, particularly as drugs with unprecedented levels of efficacy advance into clinical trials” (Prowell et al., 2016).
Harvey added that the advantage of seamless clinical trial designs is that they result in “much more efficient drug development, and need a much lower patient sample size to define activity of drugs across different areas.” But Harvey stressed that seamless designs are not appropriate for all agents, and noted a number of design questions that have been proposed by Theoret and FDA colleagues (see Box 3). “There are gaps in seamless drug development that we need to mind,” he said.
Ratain suggested that cancer drug development move from the traditional phased approach to a “learn and confirm” framework (Badenas, 2010; Sheiner, 1997). With the phased approach, Ratain said that proof of concept is usually not demonstrated until the end of a Phase II trial, FDA approval is not sought until after a Phase III trial, and there are transition times between each trial. With a more continuous drug development paradigm, Ratain said that proof of concept is part of the learning stage, and is immediately followed by testing to confirm the findings, with FDA approval sought after the confirmatory stage. Ratain said that a confirming trial would include randomization, a control arm, a pharmacokinetic study, and clinical endpoints, akin to a Phase III trial, but that learning would continue by studying the agent in more heterogeneous patients, in escalating dosages, or by randomly assigning patients to different dosage regimens.
Ratain said that the learn-and-confirm drug development framework would have three overlapping phases, which he labeled alpha, beta, and gamma. The aim of the alpha phase would be to demonstrate proof of concept and the range of active and tolerable doses. This phase would rapidly escalate the dose in groups of one to two patients until there is evidence of activity and expected (mechanism-related) or unexpected (off-target) toxicity. At this point, randomized dose escalation would be used to assign patients to pharmacologically active and plausibly safe doses to
assess if greater efficacy is achieved with doses greater than the minimal effective dose (see Box 4 for more information on the challenges of dosing in seamless clinical trials).
The beta phase would include a randomized dose-ranging design that would assign patients to doses under consideration for labeling, based on the results of the alpha phase. The gamma phase, similar to the traditional Phase III trial, would confirm acceptable safety and efficacy at the selected dose or doses by using an adaptive randomized trial to confirm the results of the beta phase. Ideally, Ratain said that the gamma phase would have an 80 percent chance of leading to a drug approval.
He said that all three phases of development could theoretically fall under the umbrella of a single protocol. “It can all be done in a continuous paradigm,” Ratain said, with the next steps only implemented after the
protocol is amended based on what has been learned. This protocol would be reviewed like a traditional protocol, with the understanding that it will be amended as the data accrue and before subsequent parts of the study are activated. “It is all done as a single, efficient but phased trial,” Ratain said. “This way oncology drug development can be efficient without sacrificing scientific rigor.”
There also has to be clarity on how information and responsibilities are handed off from investigators from one phase to the other, Ratain added. “Operationally in industry, the same group may not be responsible for
preclinical, early clinical, late clinical, registrational, and life cycle development for a single medicine.” We will have to coordinate these efforts across groups, he said. Rothenberg agreed, saying, “It really is incumbent upon us to make sure we see things from end to end and coordinate.” He suggested reaching across organizational silos to ensure coordination. “We can try and have more formal and extensive discussions to make sure drugs we are getting from the early development laboratories are drugs that are going to be meaningful and successful in the clinic,” Rothenberg said.
Schilsky asked how much confirming needs to be done before a drug
enters the market with a learn-and-confirm strategy, and what aspects of confirming can be completed after market entry. Ratain said that dose optimization could be performed as a postapproval study, because often the best dose for a drug is not determined at the time of drug approval, but after the drug enters the market. However, the current business model for pharmaceutical companies does not incentivize postapproval dosing research because the results may indicate that a lower dose is needed, leading to lower drug revenues. “If we changed our pricing models to bundled payments per patient or per treatment course, companies would feel more comfortable optimizing dose after approval,” Ratain said.
Sridhara said that master protocols enable the conduct of complex clinical trials because multiple diseases, treatments, and/or biomarkers can be assessed in one overarching protocol. “With master protocols, you are trying to answer multiple questions, or you have multiple objectives,” Sridhara said. For example, investigators may use a master protocol to test one treatment on multiple types of cancer, or to test multiple treatments on a single type of cancer, or even to test multiple types of treatments on multiple types of cancer.
Many master protocols are seamless clinical trials with adaptive design features to enable investigators to set up evolving treatment arms of a single trial protocol. With a master protocol, tested agents can be more rapidly advanced for further study if they are showing good responses or can be discarded if they do not demonstrate efficacy and replaced with new agents that undergo testing, without stopping the entire protocol, Woodcock said.
Master protocols usually have a centralized governance structure, including central IRBs, DSMBs, independent review committees, and data and biospecimens repositories, said Sridhara.
Examples of trials that use a master protocol design include the Lung Cancer Master Protocol (Lung-MAP) (see Box 5), Precision Promise (see Box 6), the NCI-MATCH trial (see Box 7), as well as the BATTLE-1 (Biomarker-integrated Approaches of Targeted Therapy for Lung Cancer Elimination) trial (Kim et al., 2011) and the I-SPY TRIAL (Investigation of Serial Studies to Predict Your Therapeutic Response with Imaging and Molecular Analysis) for breast cancer (I-SPY, 2017).
Sridhara described the KEYNOTE-001 trial as a Phase I dose-
expansion cohort master protocol, in which pembrolizumab was tested in patients with different types of solid tumors (see Figure 7). With this type of trial, an agent can undergo dose escalation in patients and the protocol can be amended to expand the number of patients with the specific tumor type(s) to include those who are most likely to respond to the agent. Dose-expansion cohort trials have prespecified starting and stopping criteria, Sridhara said, as well as predetermined thresholds that indicate when the next cohort should be added. Sridhara noted that patient protection “is of utmost importance in these studies as they expose patients to unknown safety risks,” and it is important to specify how safety results will be communicated to investigators, Sridhara said.
Woodcock also suggested that adaptive master protocols be used to test single agents as an efficient means to determine dose. Woodcock noted that industry has not often used adaptive trial designs “because you lose a certain amount of control” with the way they are set up. “With an adaptive design, you agree up front that you will take certain actions based on the data as they accumulate. You lose control because you do not get to decide at the time the data accrue, but instead have to more or less decide in advance what you are going to do. But there are tremendous advantages to adaptive designs and they add to the seamlessness of drug development,” Woodcock said.
Theoret discussed several advantages of expanding cohorts under a master protocol clinical trial. The main advantages are earlier evaluation of efficacy endpoints, standardized data collection, and the potential to expedite the development of the drug because there is no need to reapply to FDA with a new investigational new drug, undergo an IRB review of an entirely new protocol, or re-engage a clinical trial network each time changes are made to the protocol (Theoret et al., 2015). Woodcock added that “a highly involved master protocol could evaluate some of the things we are not doing that well on right now, such as studying new diagnostics and how they perform, new therapies and regimens, and combination therapies.” She noted that master protocols are especially important for development and validation of diagnostics “because you probably have to do this in an adaptive fashion to be efficient and actually learn maximally about the diagnostic.”
However, Theoret also described some limitations and disadvantages of this approach. For example, the original statistical analysis plan might not be adequately detailed to justify the sample size of the subgroups or the objectives of the subgroups in the overall development plan, Theoret said. Another challenge is the potential for safety concerns related to smaller and more heterogeneous populations used in the study arms. He suggested updating protocols and informed consent documents with safety information gathered from previously treated patients, as well as informing IRBs of emerging safety information. He added that disease-specific safety monitoring may be needed for different disease-specific cohorts. He also noted that unlike for the traditional, “phased” development program pathway, when investigators deploy expansion-cohort master protocols with the intent of an accelerated approval, there are no predefined milestone meetings with FDA to discuss progress to date and how to proceed going forward. “It is incumbent upon sponsors to contact FDA to set up these meetings at
the appropriate points in their development so that various issues can be addressed without delay,” Theoret said.
Another challenge with expansion cohort master protocols is that case report forms17 may vary in the level of detail required (e.g., an early-phase trial versus a registrational trial). He suggested considering whether case report forms need to be updated based on the specific objective of the expansion cohort. Theoret also stressed the need to ensure independent oversight of master protocols. “Oftentimes, an adequate and well-controlled trial with 300 or 400 patients will have an independent panel to address not just the potential for efficacy objectives, but ongoing safety evaluations. In this type of trial design, consideration for independent oversight should be considered early on,” said Theoret.
Sridhara pointed out that comparing various interventions using the same patient group as the control arm can save time and resources. In a hypothetical example, she said that the use of a common control arm when testing two interventions would reduce the number of patients needed for a trial by at least one-third.
Sridhara added that a common control arm could have been used for evaluating investigational therapies in patients with renal cell carcinoma, a relatively rare cancer. Recently, five concurrent studies from five sponsors were evaluating investigational therapies in patients with advanced renal cell carcinoma. In each of these studies, the control arm was treatment with the current standard of care, sunitinib. Many resources could have been saved if the control arm was shared among all five studies, Sridhara noted. McKee agreed and stressed, “We do not have enough patients for all of these trials.” However, many sponsors have been reluctant to pursue a master protocol that would enable them to use a common control arm. Echoing what Woodcock said earlier, Rubin reiterated that reluctance may be rooted in companies not wanting to relinquish control of the study. “Companies like to pick the sites and make sure they have the best quality data, and they might not agree with the overall approach,” he said, but
17 A case report form is “a printed, optical, or electronic document designed to record all of the protocol-required information to be reported to the sponsor on each trial subject.” See https://www.fda.gov/downloads/drugs/guidances/ucm073122.pdf (accessed May 22, 2017).
added that such cross-company collaboration “is something that should be worked on.”
Several workshop speakers discussed issues that need be considered when designing new clinical trials for oncology drug development, including what type and amount of information is needed to progress in drug development, who is involved in decision making, and evaluation of the benefits and risks of novel investigational agents.
Information needed to progress in drug development Rothenberg said seamless clinical trial designs and accelerated drug development programs raise several new questions, including what kind of information is needed for a study to progress from one stage to another; how strong and statistically reliable does that information need to be; and if a surrogate endpoint is used, how much certainty about the relationship between the surrogate endpoint and clinical outcome is needed?
He added that the determining which endpoints to use can be challenging. The relationships among outcomes such as progression-free survival, overall survival, restricted mean survival times, durability of response, and meaningful symptom control are not always well-established and may differ among cancer subtypes and treatment regimens.
Feltquate noted that “as we move toward seamless drug development, discerning true positive from true negative signals will become even more important.” He reported that when one of the first dose escalation studies for a cancer immunotherapy targeting the PD-L1 receptor in patients was completed, five types of cancer (melanoma, kidney, lung, prostate, and colorectal) were initially included. This testing demonstrated unusually large response rates in patients—between 20 and 30 percent—at a time when standard of care therapies provided response rates of approximately 15 percent. Even though the trial included relatively small groups of patients, the unusually large response rates seen in three of the five types of cancer led investigators to amend the study to expand testing to patients with more types of cancer. The results from this Phase I study led the researchers to acquire Breakthrough Therapy designation from FDA and they proceeded directly from their Phase I study to a randomized Phase III study in multiple tumor types (Robert et al., 2015).
Feltquate reported on another strong signal that occurred in one
patient—a durable complete response—in Phase I trial of a PD-L1 inhibitor (Brahmer et al., 2010). This patient with colorectal cancer was found to have a genetic mutation (microsatellite instability) that drove the response to the drug, Feltquate said. This finding led researchers to test the treatment in other patients whose cancer shared the same genetic mutation. These patients also had unusually high responses, which led to the drug receiving Breakthrough Therapy designation for this indication. “It was one patient at first so you can call it an anecdote, but there was good science tied to this, and that led to doing prospective work showing a very large signal,” Feltquate said. “If you have a few patients with outstanding responses, that can be more convincing than more patients with somewhat equivocal responses,” he said.
Feltquate also noted that researchers are sometimes confronted with false-negative signals, such as lack of activity in single-agent therapies. For example, initial tests of elotuzumab, which preclinical studies suggested would stimulate the immune response to multiple myeloma tumor cells, found little biological activity at any of the evaluated doses. However, there were biological reasons for suspecting this agent would perform better when combined with lenalidomide. This combination was tested in a Phase I study and found to have robust activity that was confirmed with a subsequent Phase III study in a larger group of patients with multiple myeloma (Richardson et al., 2014).
Feltquate referred to several other studies showing large signals when various immunotherapies are combined. These studies have shown more complete responses or near-complete responses than what was seen with either agent alone. In some cases there is a doubling of the response rate when two immunotherapies are combined compared with when they are used singly in patient populations that have response biomarkers for the therapies. “What is notable is we are seeing these big signals in small sample sizes,” Feltquate stressed, raising the question of whether the signals are real and should justify a rapid drug development and approval process for the immunotherapies, perhaps skipping Phase II trials.
Feltquate suggested that focusing on adequate sample size, clinically meaningful effects, and key clinically or biologically defined subsets may decrease chances of acting on false-positive and false-negative results, for both single therapies as well as combination therapies, which are increasingly being tested.
Joffe added that “Size matters. If there is a very large effect size, then a seamless design is unlikely to lead you to the wrong decision. But if you are
talking about something that has a small-to-moderate effect size, then I think there is a very high risk of mistaken decisions based upon seamless designs.” He noted that there will be increasing pressure to use seamless trials when there are smaller and more moderate effect sizes, but stressed that “the designs and decisions that are made along the way really have to be contingent on the effect sizes that emerge at various stages of the pathway.” He cautioned that a plan for making decisions about whether a seamless trial progresses needs to be deliberate in order to ensure that ethical requirements are met to protect patients participating in the trials and to make the safest decisions possible for future recipients of the treatments being tested.
Decision making and oversight Seamless clinical trials raise new issues related to decision making processes, including who should be making the decision to advance clinical testing and what stakeholder input should be solicited. Joffe pointed out that with rapid protocol modifications and seamless study designs, it is challenging for IRBs to quickly assess the risks and benefits to patients in the trials. “We are going to need highly specialized and nimble central IRBs to make this a practicality,” he said.
“Who should be at the table when new data—very early or intermediate data—are emerging regarding a therapy and what its real limitations may be?” Flaherty asked, adding, “Physicians and patients both are going to gravitate toward the best available option, even when it may not really be a great one,” especially when patients have advanced, life-threatening diseases and there is an unmet need for an effective therapy. “Others need to be involved in this discussion,” Flaherty said, and suggested having a small set of representative stakeholders present for such discussions. Flaherty added that it is not in FDA’s purview to provide such input, and that patients and investigators tend to be biased in favor of proceeding to the next stage.
Rockhold noted that an independent DSMB needs to assess the data and provide oversight in a continuous trial design, in order to adequately address ethical concerns and protect patients and make sound scientific decisions. He cited a paper written by FDA staff that found that independent DSMBs provide important quality control in seamless clinical trials (Prowell et al., 2016). Rockhold added that to maintain independence and minimize bias, Principal Investigators and sponsors of a study should not be on a DSMB, particularly for seamless trial designs. However, “there is a fine line between independence and ignorance. You do not want to pick a [DSMB] that is so independent, its members are never allowed to talk to the people doing the research . . . sometimes people carry the concept of
conflict of interest too far and they want people who are so removed they do not actually understand what is going on,” Rockhold said.
Joffe suggested that it would be appropriate to have Principal Investigators on DSMBs for single-center studies, but for multicenter studies, having all the investigators participate on a DSMB could pose a potential conflict of interest. He added that in hybrid Phase I/II or Phase II/III studies, there are defined points where the investigator or sponsor is part of decision making process about whether to continue the trial. He suggested that investigators and sponsors should also be involved in decision making at similar transition points in seamless trials. “These sorts of transition points cannot all be decided by the DSMB with all the sponsors and investigators blinded as to what is going on with the data,” he said.
Rockhold added that a DSMB review of a seamless trial aimed at answering multiple questions involves assessing much more information than it would for a typical trial asking a single question. “Somebody gives you 5,000 pages of tables of interim data, so one of the challenges for a DSMB is to outline upfront which is the right information to focus on,” he said. An advantage to having a DSMB established for Phase II or earlier studies is that “You now have a single group of people who are quite knowledgeable about the product, [and] once it gets into Phase III testing, [they] know what safety issues to expect,” Rockhold added.
Rockhold also suggested there be more clarity on the communication pathways in seamless trial designs. “If an independent [DSMB] wants to drop a group from testing, which requires changing the protocol, they might notify the IRB or the steering committee. Who they should notify has to be completely specified upfront to protect patients adequately,” he said.
An additional challenge with accelerated seamless trials and knowing when to transition from one phase to another is whether decision makers are “keeping up to date with changes in scientific understanding,” Rothenberg said. “I do not think clinical development is able right now to keep up with the rapidly evolving science, so how do we keep clinical evaluation from being the bottleneck?” Woodcock asked.
One participant suggested developing training programs for data safety monitors to stay current with newer technologies, study designs, and statistical methods. “I do not think we have a structure or framework now to train this new generation of monitors to do adequate monitoring jobs,” the participant said. Rockhold noted that DSMB training programs are being established based on recommendations from Duke University’s Clinical
Trials Transformation Initiative.18 However, the challenge is recruiting potential members to DSMB training programs, said Rockhold. “Being on [DSMBs] is not going to be a high-volume, lucrative career, so you have to find people who really want to do it,” he said. He noted that some have suggested an internship program for DSMB members so that they can observe what happens during DSMB meetings prior to becoming an actual member: “This is an important issue because when you set up a [DSMB], there is always the same list of seven names that come up and I think some of them desperately want to retire. So it is a challenge and as designs become more complex and involved, the greater the challenge.” Piantadosi added that there is also a paucity of trained clinical trial statisticians, due to a large migration of statisticians into the bioinformatics field.
Evaluation of benefits and risks Woodcock commented that “we all agree that the development of knowledge around an investigational drug is a continuum that might be a steep learning curve at first, then it levels off a little, but it never stops.” For example, researchers are still uncovering the best way to use the heart medicine digoxin, which has been on the market for decades, she said. However, an FDA approval decision is a binary decision. “So where do we cut that? The answer unfortunately is ‘it depends,’” Woodcock said. It depends on the clinical situation, including the number and quality of alternative therapies, as well as dosing and toxicity of the agent balanced against potential benefit and burden of disease. If the agent is showing unprecedented benefit, it may be granted Breakthrough Therapy designation; however, if it shows only incremental improvement, “usually you want to have more information because none of these drugs are a free lunch—there is always some toxicity,” she said, adding that “we try to lay out these considerations in a semi-quantitative way so hopefully there will be some stable balance of benefit and harms where the benefits come out ahead.”
Keegan added that “FDA agrees with the need to consider alternative trial designs and methods to evaluate new drugs, but all of these come with certain risks and considerations. It is important to evaluate the science and how we are extrapolating—to recognize those as risks and consider that the evaluation of the clinical development program needs to be an iterative process.” McKee commented that “no one paradigm is going to fit every-
18 See https://www.ctti-clinicaltrials.org/projects/data-monitoring-committees-dmcs (accessed May 18, 2017).
one’s needs, and we at FDA are trying to be as flexible as we can so that we get both the game changers on the market and the incremental benefit products on the market.” Piantadosi stressed that “the pipelines we use for development [need to] be designed explicitly for the clinical setting and for the properties of the drugs under study, rather than to reflexively say we need to have a seamless design. I would like to see us take control over the nature of the pipeline and make sure that it will produce the kind of product that we expect.” Redman added that “phase of development is just a signifier in terms of where you are in the development of a regimen. But it does not necessarily dictate exactly how you are going to investigate a regimen.”
“It is a balance—you get one chance in drug development. If you take the wrong dose into your pivotal trials and fail, there is no second chance. But at the same time, we work in a very competitive environment, so we have to balance these things. We need to recognize that when we bring a drug to market, we never have full certainty,” noted Rothenberg. He emphasized that “we need to know enough to be able to accurately characterize the benefit–risk relationship for the patients” and pointed out that the established process of postmarketing surveillance leads to regular label updates, usually based on experience of the drug when used in children, in individuals who have liver or kidney dysfunction, or in other rare patient populations.
Brown stressed that “much of what we know about medications we learn after their approval for market, so how can we continue to learn in the postmarket environment as well?” He suggested changing payment models to promote better flow of data so that postmarket learning can take place. “If Medicare required something, it would happen pretty quickly. Every clinician likes to get paid and they will submit the required information, and then every insurance company in the country will follow suit,” he said. He also suggested using adaptive licensing,19 which some European countries use.
Brown noted that if more effort was devoted to monitoring and
19 Adaptive licensing describes the progressive or staggered approval of a medicine based on a prospectively planned process. This iterative approach enables a medicine to be authorized for use in a restricted patient population, following which the authorization may be extended to a broader patient population (e.g., by labeling adaptation) after the collection of additional clinical evidence. The approach allows access to medicines earlier in areas of unmet medical need, while still protecting the more general public from risks associated with the product until further data can be assembled. The staggered approval also allows for a more educated value assessment by reimbursement authorities before a product is made available to a larger patient population (Taylor Wessing LLP, 2017).
documenting adverse reactions or other unexpected findings of a drug in the postmarket setting, it would “change the benefit–risk calculation, but we have to convince ourselves [that] we have a good enough postapproval system to do it.” He added that although FDA has a postmarketing surveillance system, “it does not work well for cancer as an outcome, but it can be done.”
Several workshop participants also discussed when randomized trials are necessary and when single-arm studies are sufficient for drug approval, particularly in cancers for which there are no effective treatments. Blumenthal noted that there has been a long track record of accelerated and sometimes even full approval of cancer drugs based on response rate and duration of response in single-arm trials. He noted that FDA is often blamed for not approving a drug without a randomized clinical trial, but he argued that the agency often takes an activist role in ensuring flexibility and will even, on occasion, directly and proactively ask sponsors if randomization is necessary.
Joffe suggested doing randomization unless early clinical data suggest a very large effect size, “because if we do not randomize, I think we are going to find ourselves making some very flawed decisions.” Redman suggested, “We should randomize when it is the best way to answer a question. Expediting drug development is not necessarily done by completing studies faster, but by designing trials that consider all possible outcomes, both positive and negative, of a trial, and the value of information provided in that trial.” Schilsky added, “Acceleration per se is not what we need, but rather to get the right information at the right time in the drug development process in order to optimize the process.” There are substantial risks from going too fast, not only to patients, but to the drug development process itself, Schilsky said. “If you do not get the dose and formulation right, you can ultimately derail your whole drug development process and then at the end of the day, you do not have a product that can help people,” Schilsky said. But he added that there is also a risk to going too slowly, in terms of losing a competitive position. Blumenthal noted that sometimes payers also want to see overall survival data before providing the reimbursement that enables adoption of new drugs in clinical care.
Several workshop participants noted the potential for the growing volume of clinical practice data—collected in EHRs, laboratory information, claims data, and other sources—to provide real-world evidence on the
benefits and risks to patients of new oncology therapies. Woodcock noted that the field of oncology is leading the way in using real-world evidence, with several groups integrating laboratory results, EHRs, and claims data to generate evidence on how therapies are being used, treatment outcomes, adverse events, and transitions to subsequent therapies. “To various extents [researchers] are curating these data so that they have greater reliability than the raw data from these [clinical] experiences,” Woodcock said. “We can learn rapidly many things about the outcomes in actual practice of using these drugs and how patients fare. Within a year or two we will learn things that may enable us to modify drug labels for subsequent indications in subgroups. So we are seeing progress in the use of real-world evidence in oncology,” she added. Ultimately, such use of real-world data may enable the “merging of the practice setting with the trial setting so that barriers to data collection are minimized, and for cancer particularly, so that more patients in the U.S. can participate in trials intended to optimize disease outcomes,” Woodcock said. Real-world evidence may take longer than anticipated to yield significant benefits, she added, but it represents “a real opportunity.”
Although many participants noted the potential value in using real-world data, some also highlighted potential issues to consider. For example, Rockhold pointed out the importance of using real-world data appropriately. “You run clinical trials to answer some questions and you look at real-world data to answer others. One does not necessarily substitute for the other, and people get seduced by the supposed ease and size of the data, but that is a problem. Real-world data are quite useful to answer real-world questions. They may not be useful to answer every clinical question, and studies using them are not necessarily less expensive, depending on what you want to do.” Brown agreed, noting, “It is very hard to use the data well, particularly these data that are collected for another purpose. You could prove all sorts of things that are both absolutely precise and completely wrong if you do not know what you are doing, so we have to be careful.” He added that data reliability is also a potential concern because the current practice of having nurses verify the data manually is cost intensive.
Koehler and Amy Abernethy, chief medical officer, chief scientific officer, and senior vice president of oncology at Flatiron Health, outlined some types of real-world data and the circumstances in which they are appropriate to use in drug development. Koehler defined real-world data as health care data that are not collected through randomized controlled clinical trials and used for decision-making purposes (Annemans et al., 2007; Garrison et al.,
2007). Such health care data include registries, EHRs, claims data, data gathered from health apps and social media, and large patient databases.
Registries are usually designed for research on a single disease, are expensive, lack direct access to data, and tend to have smaller, biased samples, said Koehler. In contrast, claims data are large datasets that can indicate treatment patterns and costs, and can also enable follow-up across health care settings. However, there is usually a 6-month lag between when the data are accrued and when they are reported in claims databases. In addition, claims data provide information on what interventions were provided to patients, but these data do not provide any clinical information about the patient. Because of this limitation, Koehler said that ongoing research focuses on how to use newer sources of data, such as EHRs and health apps. “These types of data could and should be incorporated in the drug label if we only knew how to do that.”
Abernethy stressed that real-world evidence may not come solely from EHRs or other databases, but rather from the production of “big data,” or the amalgamation of different datasets (including EHRs, administrative claims databases, registries, as well as patient-generated health data) that is intended to complement clinical trials by providing more generalizable knowledge. “Big data more clearly complete the picture of what is happening to patients in the real world in routine clinical practice,” she said. Big data involve rapidly accumulating, high-volume datasets with different types of data that should be verified and improved over time, according to Abernethy. She stressed that big data should not be viewed as amorphous, but rather as an organizing framework that can be applied to research questions in a competent, consistent, and reliable way. “Real-world data analyzed in a consistent manner can generate clinically meaningful and generalizable evidence,” she said.
For researchers to be able to use real-world data in a study, the data have to be accurate, of high quality, complete, longitudinal, reproducible, and traceable back to their source, Abernethy said. There also needs to be patient-level data linkages when appropriate for the research question, and endpoints and outcomes need to be embedded in the datasets. The study analysis approach needs to include objectives and strict analysis plans that are reproducible and credible, Abernethy stressed, adding, “We cannot cherry pick but need to be systematic in our cohort selection.” In addition, real-world data can be used retrospectively to identify patients with rare diseases who can be followed longitudinally over time, to observe outcomes
in patients who receive off-label medications, or to follow clinical outcomes of patients who have participated in clinical trials in the past.
Koehler said that a pragmatic trial is a study that uses real-world data and usually answers practical questions about risk, benefit, and cost of one intervention versus competing interventions. In contrast, an explanatory trial determines how well a drug works and how safe the drug is when it is evaluated under ideal conditions in a clinical trial setting (Roland and Torgerson, 1998). Randomization and placebo controls are often critical for explanatory studies to maximize the chance of revealing true biological effects of new treatments, Koehler noted. Because pragmatic trials compare multiple treatments used in clinical practice, they typically do not use control arms. Pragmatic trials are randomized, but this could include randomization in which outcomes in an entire group of patients is compared with those of another similar group of patients. Although blinding is usually required in an explanatory study, it is usually not possible in a pragmatic trial, she added.
Koehler emphasized that pragmatic trials should have strict requirements aimed at reducing many types of bias in the study. These include setting randomization requirements to address selection bias, defining similar conditions to study arms to address performance bias, and ensuring that the groups being compared have the same performance standards for collecting data on patients who drop out of the studies and conducting intention-to-treat analyses to address attrition bias.
Explanatory trials are conducted in a carefully controlled population of patients, who tend to be relatively homogeneous, lack concomitant illnesses, and often have less ethnic diversity (Roland and Torgerson, 1998). By contrast, pragmatic trials are conducted in more heterogeneous patient populations in a variety of clinical settings, Koehler said. Compared to explanatory trials, pragmatic trials tend to be simpler and can potentially be less expensive to run, she said. Koehler stressed that pragmatic trials tend to answer questions that patients and payers care about, such as how a treatment affects symptoms, patient quality of life, and costs, in addition to the mortality and morbidity usually assessed in explanatory studies. She also described nine dimensions to characterize the level of pragmatism of a trial (see Figure 8). For example, she said that a study comparing self-supervised tuberculosis treatment with directly observed treatment is considered highly
pragmatic in all factors assessed by a Pragmatic-Explanatory Continuum Indicator Summary (PRECIS) and closely resembles usual clinical care. In contrast, a study evaluating surgery for treatment of carotid artery blockage (the North American Symptomatic Carotid Endarterectomy Trial) is closer to the center of the PRECIS wheel and much more closely resembles an explanatory trial, she said.
Abernethy suggested using EHRs as the organizing framework for clinical data given that more than 90 percent of oncologists use EHRs, these
data are linkable, and they can be used to document and improve quality. Abernethy outlined several key steps undertaken at Flatiron Health to gather evidence from EHRs to derive usable real-world evidence:
- Aggregate data across different care settings or across the country.
- Standardize and harmonize the information in EHRs so data coming from multiple sources use the same terminology and units of measure.
- Abstract the information gathered from various reports and make sure key information is entered, such as biomarker data from laboratory tests, findings in radiology and pathology reports, or clinicians’ notes.
- Have quality assurance for each item of data entered. This includes having all abstracted data processed in accordance with approved procedures, quality control and auditing of abstracted information, audit trails and documentation, and traceability of abstracted information.
- Match the data in EHRs to a variety of other key datasets, such as mortality data, genomics, patient-reported outcomes, and claims data.
- Incorporate endpoints and outcomes that enable interrelating the intervention to documented differences in individual patients or patient groups.
- Organize the data for analysis and prepare an analytic plan.
Using this approach, Abernethy discussed an evaluation of the appropriateness of the black box warning on the breast cancer drug Kadcyla that she and her colleagues conducted. This warning specifies that clinicians should not use the drug to treat breast cancer patients with suboptimal heart pumping ability, specifically with heart ejection fractions of less than or equal to 50 percent. Kadcyla is composed of two anticancer drugs, emtansine and trastuzumab. Because trastuzumab had been shown to cause heart damage, patients with less than optimal ejection fraction were excluded from clinical trials. Consequently, it was not truly known how patients with the lower ejection fraction would fare with this combination drug. Despite the black box warning, many of these patients have received Kadcyla. Abernethy is conducting a retrospective study to assess the rate and severity of cardiac events in patients treated with Kadcyla using data derived from the EHRs of approximately 8,000 women with breast cancer,
1,000 of whom had been treated with Kadcyla. Sixty-six of these women had ejection fractions of 50 or less prior to being treated with Kadcyla. These patients are now being followed to assess treatment outcomes. “We now are starting to understand what happens to these women both in terms of their ejection fraction over time as well as their breast cancer outcomes,” Abernethy said.
Woodcock noted at the workshop that FDA had just published an article defining real-world evidence and how it can be used (Sherman et al., 2016). This article is both forward thinking and cautionary, she said, noting that FDA has already approved drugs for rare diseases based on such patient care data. Woodcock noted that such data will become more useful when studies can be randomized within the care setting. “That will be very helpful when it happens, but we are not there yet,” she said, adding, “The bottom line is that we are open to real-world evidence.” She pointed out studies on standard of care that “got answers rather quickly” from doing cluster or block randomization analyses of real-world data. Such studies may compare the standard of care of patients in one setting compared to another, for example.
Woodcock reiterated that once oncology drugs enter the market, many questions about them remain unanswered, and “what we hope is that in the future, evidence from care can help inform and answer some of those additional questions we have about the use of the drug.” Koehler suggested that real-world evidence could be used to further evaluate drugs that enter the market via accelerated approval, which requires less preapproval testing, and could also support evidence generation for subsequent indications, label updates, and drug combinations within an approved indication.
Koehler suggested that pragmatic trials could play a larger role in answering regulatory questions, such as formulating drug labels. She emphasized that “the most useful source of knowledge will come from randomization in the context of clinical practice . . . therefore, the best thing to do is to potentially consider real-world outcomes for regulatory review as soon as possible.” She provided the Salford Lung Study as an example of a real-world evidence trial, which was conducted in the United Kingdom for patients with chronic obstructive pulmonary disease before the drug entered the market and used randomized and controlled real-world data to assess the drug’s effectiveness (Vestbo et al., 2016). In this study, patients were randomized as to whether they received the investigational drug from
their clinicians or the standard of care. Koehler noted that it was one of the first times that data were used from community clinician practices to support regulatory approval of an investigational drug (Vestbo et al., 2016). Koehler is currently conducting a study assessing whether patient outcomes using EHR data is similar to patient outcomes observed in a clinical trial (with a similar patient population) for women with metastatic breast cancer receiving the drug letrozole.
Koehler emphasized that although evidence on the safety and efficacy of a drug is gathered as it goes through preapproval clinical trials, the evidence generated is from a much smaller and more homogeneous patient population than in the real-world setting once the drug enters the market. Because the drug’s use in patients in the general population outpaces the collection of evidence on how it will affect them, there is an evidence gap that could be addressed by real-world data (see Figure 9).
Redman noted that real-world data can help in an iterative fashion to expand findings from clinical trials. She suggested using real-world
data to evaluate the eligibility criteria used in clinical trials and potentially modify a clinical trial design in the future that expands eligibility. Brown agreed that real-world data can be useful when determining the eligibility criteria of clinical trials. Abernethy concurred, adding that researchers are using the data when designing their clinical trials because “you have a lot of the key features that will help you pressure test, for example, eligibility criteria and how to better design your studies.” Sridhara added, “You can use all of these data to learn more about a disease, which will help us in future clinical trials to determine what we should be targeting or how we can run these studies and what should be the control.” Ratain said that such data also enable detection of adverse drug reactions.
Sridhara noted that prospective observational studies using real-world data could affect drug labeling. Studies may also randomize the clinicians rather than the patients. However, she raised the issue of determining endpoints in pragmatic oncology studies. “We cannot really use progression-free survival or tumor response rate because RECIST criteria are used by specialists and that is not going to be available in every community hospital,” she said. Although it is possible to follow patients over time, she added that in general, the formal recordkeeping and follow-up observed in traditional clinical trials are not possible with pragmatic trials because the records of patients are not often carried over when they switch from one provider to another.
Ratain noted that real-world evidence has been used to establish proof of concept at Vanderbilt University (Denny et al., 2013). This institution has a large patient database20 that includes genomic data and diagnostic coding data on tens of thousands of patients. Researchers have used the database to discover and validate genetic differences linked to various disorders. “The area where real-world data may be most useful is to discover and validate patient response biomarkers,” he said. Abernethy agreed, noting that some institutions have processed their EHR data so they can act as clinical annotation for biospecimens in their biorepositories. “This helps biologic discovery and understanding and identifying biomarkers,” she said.
Brown stressed that “there is no single data source that is going to answer all your questions, so thinking about how to match your question to the data and method you need is critically important. You cannot answer everything with the data sources we have, but you can usually get pretty far.”
20 See https://medschool.vanderbilt.edu/dbmi/synthetic-derivative (accessed May 18, 2017).
Rockhold added, “It is easy to get seduced by the size of real-world data, but an approximate answer to a good question is better than a precise answer to the wrong one. So do not just go somewhere because there is a lot of information if that information does not have the content that can answer your question.”
Several speakers described the need to ensure that real-world data are accurate and of high quality. Data need to be reproducible, said Abernethy: “One of the biggest problems with real-world evidence and the reason why it is so easily dismissed is because we are so worried about the quality of the data. So we need to be transparent about the current quality and how we are going to improve it over time.” Monica Bertagnolli, chief of the Division of Surgical Oncology at Dana-Farber/Brigham and Women’s Cancer Center, added, “We are deluged by data and sometimes the data gathering that is done is very accurate and other times it is very inaccurate. How do we understand it and make sense of it, especially since in oncology, the stakes are so incredibly high?”
Rubin noted that information technology efforts are under way to standardize and harmonize data, and make them easily transferrable from one platform to another. He described the efforts of the South Texas Accelerated Research Therapeutics (START), which uses a high-quality and innovative information technology infrastructure to ensure accurate and rapid clinical trials of novel anticancer agents (START, 2017). Bernard Munos, senior fellow at FasterCures, suggested using a model that constantly merges and extracts knowledge from raw data without the need for a common template, which is done by entities such as Google and the National Security Agency, to find answers to clinical questions. Abernethy said that she has taken an approach of starting with a parsimonious model that has flexibility to add new data points as needed over time, such as heart failure data in breast cancer patients. She pointed out that “there is great excitement about artificial intelligence and machine learning” that can curate and harmonize data, but she suggested caution in how those technologies are applied until their effectiveness for clinical applications is better understood.
Ronald Kline, medical officer of the Patient Care Models Group at the Centers for Medicare & Medicaid Services’ (CMS’s) Center for Medicare & Medicaid Innovation, noted that with the Oncology Care Model, participating physicians are required to use a data registry that collects ana-
tomical staging information, relevant molecular mutations, and histology (CMS, 2017). The cancer registry of the Centers for Disease Control and Prevention is used as the basic format; EHRs are exported to the registry in an automated fashion when possible, with the remainder of data entered manually where fields are missing or do not align. He added that EHR companies do not use a common standard for reporting oncology or other disease data, so CMS set its own data standards for the Oncology Care Model for the field to use. Rockhold added that unless an agency such as CMS requires it, standardization of real-world clinical data is unlikely to occur.
Abernethy commented on the need to clarify the role of real-world evidence in oncology drug development. She suggested that FDA guidance on real-world evidence would be helpful, including a delineation of drug development scenarios to which real-world evidence is most applicable, data quality requirements, data management and dataset production requirements, optimal analytic approaches, and how real-world evidence would be represented in a drug label. Abernethy added that a transparent multistakeholder process to develop an approach to using real-world endpoints would be beneficial.
Howard Burris, chief medical officer and president of clinical operations at the Sarah Cannon Research Institute, noted that many community oncologists are confounded by the plethora of genetic information that is often reported for patients and how to interpret it. “In practice you get your radiology and pathology report back and it is three pages long, but you read the three lines that are the assessment and conclusion, which is really what you want when you are busily seeing patients during the day,” he said. He suggested that similar synopses and clinical implications be provided for patient’s genetic profiling information, perhaps by analytic software built into EHRs.
Finally, Bertagnolli suggested that real-world evidence could be used to assess not only the impact of a specific drug or therapy, but also the impact that new treatments have on the use and impact of other treatment modalities, such as surgery and radiation therapy. She noted that her sarcoma surgery program was radically transformed by the advent of imatinib to treat gastrointestinal stromal tumors, and she wanted to study the impact of surgery on metastatic disease now that it can often be well controlled with targeted treatments.
Piantadosi asked if using real-world clinical data in research will necessitate changes to the Privacy Rule that was promulgated under HIPAA. Brown noted that the HIPAA Privacy Rule is unclear about what types of data sharing is acceptable, and therefore leads to many inefficiencies in data management and analysis. “We are so horrified of doing the wrong thing, and there is plenty of work we could do that we just do not because it involves linking information; there is no single source of information good enough. But linking it requires identifiers, and then it becomes a big mess,” Brown said.
Pentz added that “we have to be careful about how we protect patient data because there is no privacy if you are on the grid,” but “in the medical field we have these artificial [regulations] that make it almost impossible for us to do good things.” She said that we have to modify these regulations so they are more realistic and provide real abilities to prevent the wrong people from misusing patients’ medical information while facilitating access by clinicians and researchers who might actually cure diseases and help people. Pentz pointed out that some of the patient privacy concerns regarding the use of real-world data, particularly for prospective pragmatic studies, could be addressed by a broad consent process, similar to that used for performing future research on biospecimens.
Currently only about 3 to 5 percent of adults with cancer enroll in a clinical trial in the United States, said Gwynn Ison, clinical reviewer in FDA’s OHOP. Ison pointed out that one of the goals of clinical trials is to facilitate the ability to generalize study findings and apply them to all patients with the disease. “To that end, eligibility criteria are a key component of the design of a clinical trial,” she said. These criteria define patients who are appropriate to receive the investigational treatment, and exclude patients who might be at greater risk of being harmed by the agent. However, excessive or overly rigid eligibility criteria may hamper patient accruals to trials and limit the generalizability of their findings, Ison noted, adding that this can result in drug labels without adequate guidance for clinicians on how to deliver the therapy to a more heterogeneous population.
There have been recent attempts to modernize eligibility criteria, including a multiyear project led by Edward Kim (Kim et al., 2015), Ison
reported. Furthermore, one of the goals of the Cancer Moonshot is to modernize eligibility requirements. Additionally, Ison noted that at a workshop21 held in May 2016, multiple stakeholders, including ASCO, FDA, NCI, clinical investigators, industry representatives, and patient advocacy groups, addressed specifically which eligibility criteria potentially could be relaxed or eliminated within clinical trial protocols. This workshop focused on the exclusion of patients with HIV, brain metastases, or organ dysfunction, such as those with kidney, liver, or heart disease; patients with prior malignancies; and pediatric patients. Working groups presenting at the workshop recommended major revisions in eligibility requirements in all these patient groups. These working groups recommended including
- patients with treated or stable brain metastases in all phases of clinical trials unless there is a compelling safety reason for exclusion. Patients with newly diagnosed active and/or progressive brain metastases may be included in some trials based on factors related to the specific cancer and agent under investigation.
- HIV patients in all phases of trials unless there is reason to believe the drug under investigation may interfere with the control of the HIV infection. Patients with stable HIV disease—based on their blood cell counts, history of opportunistic infections, and the status of their HIV treatment—should be treated with the same standards as patients without HIV.
- pediatric patients in initial dose-finding trials, especially when there is a scientific rationale for including them. Pediatric patients ages 12 and older should also be included in late-phase trials for diseases, such as sarcomas, that span across adult and pediatric populations.
- patients with prior malignancies, depending on the type and time since cancer treatment, with more liberal time frames than are currently used to exclude patients with malignancies from trials. Patients with current non-threatening malignancies, such as non-melanoma skin cancer, should also be included in clinical trials.
In addition, the working groups recommended using liberal criteria for creatinine clearance when excluding patients because of suboptimal kidney function, especially if the agent being tested is not thought to be excreted
21 See https://am.asco.org/daily-news/eligibility-criteria-project-seeks-benefit-patients (accessed May 18, 2017).
primarily by the kidneys. Serum creatinine levels should not be used as an eligibility determinant. No changes were recommended for the current criteria for liver or cardiac functioning of eligible patients, although investigators were encouraged to include geriatric patients in clinical trials, as well as to include specific cohorts of these patients in their studies.
The working group is publishing their recommendations and plans to promote standardized and inclusive eligibility criteria that can be adopted by protocols, and to encourage sponsors to follow the recommendations. FDA also plans to develop metrics to determine whether the recommendations are being implemented, and will address practical issues and barriers to implementation when they arise. Finally, the working group plans to explore more possibilities for expanding eligibility, including revisions to criteria related to drug washout periods, concomitant medications, and other factors that may lead to exclusion of elderly patients, Ison reported.
Ison noted that the TAPUR (Targeted Agent Profiling Utilization Registry) study is unique because it has very broad patient eligibility criteria (see Box 8).
Kline encouraged the inclusion of more pediatric patients in trials. He noted that there are no ethical problems with including children in trials, as this is done all the time for studies of pediatric cancers. “I have two kids and they mean more to me than my own life, and I think most of us feel the same way. So how, as an oncology community, do we completely ignore pediatrics in developing new drugs?” he said.
Keegan added that the recommendation to loosen the eligibility criteria to include children 12 years and older is based on the understanding that children in this age group metabolize drugs very similarly to adults, so there is no reason to exclude them based on their exposure to the drug. She added that FDA also asks sponsors to consider pediatric use of a cancer drug when it is near approval,22 and consults with pediatric oncologists both to ensure that the rights of the children are protected, and to identify areas where a drug might specifically benefit children with cancer.
Rothenberg added that as more basket and other trials have been developed that can test an intervention in multiple subgroups, pediatric patients could be another subgroup added. “This would be a nice way of addressing that new drugs do not get tested in children early enough, so therefore, any benefit to them is also delayed. With the vast majority of new drugs that
22 See https://www.fda.gov/drugs/resourcesforyou/consumers/ucm143565.htm (accessed May 25, 2017).
come along, children are able to tolerate higher doses corrected for body surface area than adults, so this is something that is very much in our awareness and we are going to try and make that happen,” he said.
Sridhara agreed that subgroup analyses are a way to make trial populations more heterogeneous. “There is nothing written in the study designs that say you cannot include, for example, patients with a performance status
score23 of 2, or pediatric or geriatric patients. These are often excluded and you can always put them in the studies,” she said.
Elad Sharon, medical officer and senior investigator at the NCI, noted that the NCI has already been encouraging inclusion of HIV-positive patients in clinical trials for pembrolizumab, as it did with most of its nivolumab and atezolizumab trials. However, the NCI often finds that sponsors and investigators resist including patients with HIV, and suggested that when that happens, FDA should ask sponsors to provide justification. Ison clarified that although FDA endorses the recommendations to expand eligibility and can suggest these revised criteria to sponsors, it has no legal mandate to enforce the criteria.
Pentz commented that there also are other major barriers to clinical trial participation, including the cost or difficulty of transportation, or taking time off work, which many low-income patients cannot afford to do.
Several speakers stated that the new drug development paradigm requires greater collaboration among companies and other stakeholders, and provided several examples of how companies are sharing their data, and in some cases, participating in the same clinical trials. Sigal stressed that due to the growing complexity of cancer diagnostics and treatments, there also is a growing need for collaboration in the testing of these agents. “Everybody used to do their own thing individually, but the complexity of the science and the new models that emerged started to show us we better start to do this together because if we do not, everybody pays for it,” she said.
Blumenthal noted several collaborative data-sharing efforts to improve the development of drugs and diagnostics:
- Project Data Sphere LLC, an independent, not-for-profit initiative of the CEO Roundtable on Cancer’s Life Sciences Consortium, operates the Project Data Sphere platform. It is a free digital library-laboratory that provides one place where the research community
23 The Eastern Cooperative Oncology Group performance score helps clinicians to assess patients’ performance status and guide treatment decisions. A performance status score of 2 indicates patients are “ambulatory and capable of all self care but unable to carry out any work activities; up and about more than 50 percent of waking hours.” See https://www.ncbi.nlm.nih.gov/pubmed/7165009 (accessed May 22, 2017).
- can broadly share, integrate, and analyze historical, patient-level data from academic and industry Phase III cancer clinical trials. It is open to researchers that apply for, and are granted, status as an authorized user (Project Data Sphere LLC, 2017). “There are lots of inferences that can be made from very old legacy datasets,” Blumenthal said.
- The FNIH Volumetric CT for Precision Analysis of Clinical Trial results (VolPACT), which seeks to analyze volumetric CT imaging trial data contributed “in-kind” from completed Phase II solid tumor trials to improve quantitative prediction of Phase III results, including metrics of response to therapies (FNIH, 2017).
- Immune-related Response Evaluation Criteria In Solid Tumors (irRECIST), which was developed by the Cancer Immunotherapy Consortium, formerly the Cancer Vaccine Consortium, in collaboration with the International Society of Biological Therapy of Cancer (Wolchok et al., 2009).
- The Blood Profiling Atlas in Cancer is composed of representatives from government, academia, pharmaceutical, and diagnostic companies and aims to “accelerate the development of safe and effective blood profiling diagnostic technologies for patient benefit.” The group will launch a Blood Profiling Atlas in Cancer pilot to aggregate, harmonize, and make freely available raw datasets derived from circulating tumor cells, circulating tumor DNA, and exosome assays, as well as relevant clinical data and sample preparation and handling protocols from 13 studies (Bhan, 2017; BloodPAC, 2017).
Public–private partnerships that engage in collaboration with drug companies are valuable and constitute “a different way of doing business,” Sigal said. “It is a partnership and that means behavior has to change on all sides, and it has to be patient driven.” She gave the example of the Lung-MAP study, where there was a compelling need to test a variety of treatments in rare subsets of patients. What made this trial particularly attractive to sponsors was that FDA approved it as a registration trial. The NCI was a catalyst, but the Cooperative Groups of the NCI Clinical Trials Network, patient groups, and the companies all collaborated for this trial. “Collaboration among all sectors is very important and it is a way to get us there faster. Necessity drives us to collaborate because we need all hands on deck,” she said. Scott Fogarty, diagnostics partner at Thermo Fisher
Scientific, suggested that it would be valuable to have participation by the diagnostic industry in these collaborations as well.
Schilsky remarked that “one of the challenges we consistently face is aligning the culture of drug development with the science of drug development. The science is typically far out ahead of the cultural change that is necessary. The things that seem to drive changes to the culture ultimately come down to: what is the medical and business opportunity, what is the competition, what is the risk tolerance of the sponsor, how willing are people to fail at various points of the process, and what are the consequences of doing so?”
Rubin noted that “data sharing and cooperation for industry particularly is important because we have lots of data and in general we do not like to share.” However, he noted that there is extensive data sharing occurring among industry to develop alternatives to RECIST, including protocols for assessing a drug’s impact on a tumor. “Everyone mentions this need for an alternative to RECIST, but they all do it slightly differently so everyone shared their protocols and there was a consensus gathering around how protocols should be written to do this,” he said. The guidelines developed from this work have now been published (Seymour et al., 2017).
Another major sharing endeavor undertaken by industry is the development of a standardized PD-L1 diagnostic test that can be used to assess which patients might be eligible for certain cancer immunotherapy drugs. FDA recognized that companies developing anti-PD-L1 drugs were also developing assays for the ligand, and that each assay was different. “The concern was that there would be a lot of confusion around how pathologists would handle” these different diagnostic tests, Rubin said. Consequently, ASCO, FDA, and the American Association for Cancer Research convened a workshop to address this issue, at which an industry workgroup volunteered to develop a blueprint proposal of potential solutions using non-small cell lung cancer as the first test case (FDA, 2017a). This resulted in the Blueprint Programmed Death Ligand 1 Immunohistochemistry Assay Comparison Project, in which 39 lung cancer specimens were stained and analyzed with each assay developed by different companies (Hirsch et al., 2017). Rubin said this study “will be helpful to the community in understanding the [level of] concordance across the various assays,” adding that “it is another example of where industry is sharing. It may take some effort but it can happen.” Such sharing has to be done voluntarily, McKee noted, as FDA is prohibited by law from sharing any company’s proprietary information.
In addition to duplication of efforts in PD-L1 diagnostic tests, there is also a redundancy of 22 different PD-L1 inhibitors sponsored by different companies, Redman noted. Rothenberg said PD-L1 inhibitors were showing such extraordinary promise in early tests that every drug company wanted one in their portfolio, especially because a combination therapy with PD-L1 inhibitors was thought to be likely in the future. However, there was also recognition that the market could not support so many inhibitors: for example, Rothenberg noted that there are only four or five drugs on the market to treat high cholesterol, which is much more common than the cancers that PD-L1 inhibitors target. “You could ask me to design the most efficient trial, but it will not be efficient if it is being repeated 22 times,” he said. Driven by projections of market demand and a desire to share their expertise, Pfizer and Merck KGaA collaborated to develop the PD-L1 inhibitor avelumab, said Rothenberg. Sigal added that FDA is also concerned about this redundancy because there are not enough patients to test all of the inhibitors, and suggested a master protocol in which multiple treatments from multiple companies could be tested simultaneously. However, she said that industry was reticent to use such a model, probably because some companies were further along the development path with their inhibitors than others.
Blumenthal said companies have less incentive to share safety data than efficacy data, and suggested engaging in partnerships to identify and study biomarkers that could potentially predict rare but serious toxicities. Sigal noted that such sharing of information is especially critical with cancer immunotherapies, in order to discern important safety signals earlier. A number of workshop participants discussed barriers that discourage such sharing. For example, Rubin cited a lack of resources devoted to sharing data: “We have to convince people to carve out some internal resources to work on this data-sharing piece, because it is the right thing to do,” he said. An additional barrier, raised by Rubin and Blumenthal, is the possibility of finding a different, conflicting result when efficacy data are reanalyzed. Some participants thought the barriers could be addressed, however, and Blumenthal cited the Blood Profiling Atlas in Cancer as an encouraging initiative in support of data sharing.
Blumenthal agreed, adding that for “hot data” such as immunotherapy data, “there probably is some consternation that if they share all their data, someone will misinterpret and reanalyze it and come up with some other [incorrect] answer.” He noted that barriers to sharing may arise within the legal departments of companies and research institutions. But with the
Blood Profiling Atlas in Cancer, he said he has been encouraged by the degree of sharing, and how those roadblocks have been overcome.
Several workshop participants discussed recent policy changes and new opportunities to improve cancer drug development, including innovations at CMS, FDA’s Oncology Center of Excellence (OCE) and Expanded Access Program, and the Cancer Moonshot. A number of workshop participants also considered other potential policy strategies, such as some of the recommendations made by the Blue Ribbon Panel of the Cancer Moonshot.
Kline outlined the newly instituted parallel review process of FDA and CMS for medical devices.24 Such parallel reviews should alleviate the lag time between FDA marketing approval for a device or diagnostic and a national coverage determination by CMS, enabling reimbursement for clinical use concomitant with market approval. “It is a CMS–FDA collaboration to move things forward on a more timely basis,” Kline said. As previously mentioned, CMS has also developed an Oncology Care Model, which is an episode-based payment model25 for 6 months of care, and includes performance-based payments when clinicians achieve various quality measures (CMS, 2017). “This provides incentives to use high-value drugs and we are hoping to push the field toward value-based reimbursement,” Kline said.
Woodcock and Sigal reported on FDA’s OCE, which is expected to avert regulatory delays by fostering connections among three separate divisions in the agency (FDA, 2017c,d). OCE will support integrated evalua-
24 See https://www.federalregister.gov/documents/2016/10/24/2016-25659/program-for-parallel-review-of-medical-devices (accessed May 18, 2017).
25 Episode-based payment is when “reimbursement for medical services delivered during defined episodes of care is bundled together.” See http://www.nejm.org/doi/full/10.1056/NEJMp1105963#t=article (accessed May 18, 2017).
tion of new treatments by coordinating the agency’s expertise in regulatory science, small molecules, biologics, devices, and diagnostics. The OCE will make oncology the first disease area to have a coordinated clinical review of drugs, biologics, and devices across the agency’s three medical product centers, and it is expected to make regulatory review more seamless, particularly for combination treatment regimens, said Woodcock.
Woodcock underscored the benefits of integrated regulatory evaluation in an increasingly sophisticated and complex drug development environment. She also noted that in order to facilitate understanding of novel research and to incorporate the views and priorities of a variety of stakeholders, OCE is also expected to have intensive interaction with the scientific, medical, and patient communities.
Steven Lemery, lead medical officer in the Division of Oncology Products 2 at FDA’s OHOP, provided a summary of FDA’s Expanded Access Program. This program enables patients to receive investigational agents if they have a serious or immediately life-threatening disease or condition and meet all of the following conditions26:
- There is no comparable or satisfactory alternative therapy.
- The potential benefit of the new agent justifies the risks.
- Providing access will not interfere with clinical investigations that could support marketing approval of the agent.
Expanded access is intended to provide compassionate use of an investigational therapeutic agent for a patient with no other treatment options, Lemery stressed. Most requests are single-patient requests made by clinical investigators who have a letter of cross-reference from the commercial sponsor allowing access for a single patient. FDA also has provisions for an intermediate-sized population to receive expanded access, usually at the end of the drug development process. Expanded access protocols can also be submitted by commercial sponsors as part of their ongoing IND testing. In order for expanded access to be granted, patients, their clinicians, and the drug’s sponsor must all be willing to participate, and the clinician must
26 See https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfcfr/CFRSearch.cfm?fr=312.305 (accessed May 22, 2017).
submit a form to FDA (Jarow et al., 2017). Expanded access also requires IRB consent except in emergency cases.
Of 1,332 single-patient expanded access requests made to FDA for oncology drugs between 2014 and 2016, only 1 request was declined by the agency, and 4 were withdrawn prior to an FDA decision. About two-thirds were for drugs that subsequently were approved for market (Lemery et al., 2016).
Although most expanded access requests are not intended to support drug development or provide information about drug performance, in some instances data from expanded access cases could be used to support approval of a drug, according to Lemery. He noted that FDA’s OHOP approved four drugs for oncology indications based in part on data from the use of these drugs in its expanded access program. Two drugs were to provide a treatment alternative for patients who had to discontinue the standard chemotherapy because of toxicity, and two others were used to treat pediatric patients, with one drug already approved for adults. These drug approvals depended in part on safety, survival, or pharmacodynamics data collected from patients who were granted expanded access to the drugs, Lemery reported. He said the expanded access program could be useful in other cases as well, such as for patients with rare cancers or who have rare genetic mutations targeted by the drug. Lemery noted that often it is not feasible to enroll patients with rare cancer mutations into a clinical trial because they do not live in a location where the trial is offered, “so expanded access might be a good way to obtain information about the use of your drug in these rare populations. You could maybe get data on patients who do not meet eligibility criteria, but can give you some real-world experience. You may not be getting all the data you would get in a clinical trial, but you can at least get scans and demographic information,” he said. Theoret agreed, noting that expanded access could be useful once an investigational new drug has Breakthrough Therapy designation, for those patients who would like to receive the drug, but are unable to enroll in a clinical trial testing it. “As development programs receive their Breakthrough Therapy Designation, it is often very early in the discussions that we should start thinking about what their expanded access program is going to look like,” he said.
The expanded access program has not posed undue risks for drug development, according to a study of the program over a 10-year period. This study found that with more than 10,000 expanded access requests, only two commercial development programs involving more than 1,000 patients were placed on hold or partial hold due to a serious adverse event
observed in a patient who received the drug via expanded access. One hold was removed months later and the other, a partial hold limited to a specific patient population, was eventually removed as well (Jarow et al., 2017). Another study of failed oncology trials found that none of these studies failed in clinical development due to data collected from expanded access cases (Khozin et al., 2015).
Ison noted that many informed consents are based on the side effect profiles in populations being studied and are not always applicable to those subjects who are applying for expanded access. Therefore, she cautioned that patients and clinicians be accurately informed of the benefits and risks of an experimental intervention and suggested that clinicians be proactive about informed consent.
Elizabeth Jaffee, deputy director of the Sidney Kimmel Comprehensive Cancer Center at Johns Hopkins University, presented the recommendations of the Cancer Moonshot’s Blue Ribbon Panel related to improving cancer drug development. She noted that the Cancer Moonshot was initiated based on the recognition that there has been unprecedented progress in cancer research recently, but that there is insufficient coordination, data sharing, and funding to reap the benefits of this progress. “The science is ripe, but the progress made in getting it to benefit patients is too slow and lacks efficient coordination,” Jaffee stressed.
The overarching goals of the Cancer Moonshot are to accelerate progress in cancer research; encourage greater cooperation and collaboration among academia, government, and the private sector; and enhance sharing of data (The White House, 2016). To carry out its goals, a federal task force was convened to determine policy measures to pursue, and a Blue Ribbon Panel was convened to delineate research goals.
The federal task force delineated several policy goals, including the following:
- Support greater access to new research, data, and computational capabilities.
- Improve patient access to clinical trials and standard of care.
- Identify and address any unnecessary regulatory barriers to drug development and approvals and consider ways to expedite administrative reforms.
- Explore opportunities to develop public–private partnerships and increase coordination of the federal government’s interactions with the private sector with regard to drug development, treatment access, and information systems.
More specific strategic goals are to expand use of mobile devices and create tracking systems for patients to unleash the power of data, determine best practices for consent, develop a seamless data environment with open platforms, and develop the necessary workforce. The task force also recommended bringing new therapies to patients faster by modernizing eligibility criteria for clinical trials, and developing trials for patients with specific genetic defects rather than tissue-specific types of cancer, such as the NCI-MATCH trial, and using real-world evidence.
The Working Groups of the Blue Ribbon Panel made several recommendations focused on research that also have policy implications in drug development (Cancer Moonshot Blue Ribbon Panel, 2016):
- Enlist patients in a federated network that includes patient tumor profiling data, and “preregister” patients for clinical trials in which their biopsy tissues are analyzed to learn which features predict outcomes.
- Organize a network to discover and evaluate novel immune-based approaches for adult and pediatric cancers, and eventually develop vaccines for prevention.
- Create an ecosystem to collect, share, and interconnect datasets.
- Support research to accelerate development of guidelines for management of patient-reported symptoms to improve quality of life and adherence to treatment regimens.
- Create a human tumor atlas that catalogs genetic lesions and interactions among tumor cells, immune cells, and other cells in the tumor microenvironment.
- Develop new enabling technologies, such as liquid biopsies and multiplex analyses of the tumor microenvironment, to accelerate testing of therapies and tumor characterization.
Blue Ribbon Panel working groups also recommended demonstration projects, including a national pediatric immunotherapy translational science network to facilitate testing of new immunotherapies in childhood cancers, and a tumor pharmacotyping demonstration project to help
discern biomarkers that reveal the most effective therapeutic agents for individual patients.
Jaffee also described current ongoing Cancer Moonshot programs, including the following components (NCI, 2017b):
- NCI Drug Formulary that will leverage lessons learned from the NCI-MATCH trial to forge relationships with 20 to 30 public–private partners to expedite researcher access to investigational agents and approved drugs for combination trials. One goal is to reduce negotiation time for collaborations.
- Application Programming Interface, an endeavor in which several third-party innovators, including Smart Patients, Synapse, Cure Forward, and Antidote, are participating to build applications, integrations, search tools, and digital platforms to disseminate clinical information to the community.
- Strategic Computing Partnership between the Department of Energy (DOE) and the NCI to accelerate precision oncology, that is, getting the right treatments to the right patients. There are three new pilots to apply the most advanced supercomputing capabilities to analyze data from preclinical models of molecular interaction data.
- Open-access resource for sharing cancer data via the Genomic Data Commons. Foundation Medicine is participating in this endeavor and hopes to double the total number of patients who provide information to this database. This will provide a mechanism for broad sharing and partnerships among government agencies, academia, and industry that will be complemented by the NCI’s Genomics Cloud Pilots, which will provide imaging, proteomics, and immunotherapeutics datasets.
- Public–private Partnership for Accelerating Cancer Therapies that involves collaborations of 12 biopharmaceutical companies, research foundations, philanthropies and the FNIH. This partnership will fund precompetitive cancer research and share broadly all data generated for further research, with initial focus areas that include identifying and validating biomarkers for treatment response and resistance to cancer therapies, clinical trial platforms for combination therapies, and predictive modeling approaches and therapies for rare cancers.
- Applied Proteogenomics Organizational Learning Outcomes, which is a partnership between DOE, the NCI, and the Depart
- ment of Veterans Affairs to use new technologies to rapidly identify new targets and pathways for detection and intervention. This initiative plans to develop a national biospecimens bank on which researchers can conduct proteomic analyses over time as patients respond to or become resistant to treatments.
Future planned initiatives from the Cancer Moonshot include the development of predictive computer algorithms to rapidly develop, test, and validate preclinical models predicting drug response via a partnership between the NCI and DOE. Another initiative will build computational collaborative relationships, beginning with a partnership among the University of California, Intel, IBM, and GE to create a virtual data ecosystem.
The Blue Ribbon Panel also identified several current policy barriers to drug development that were not covered in its scope, but will need to be addressed in order to advance its recommendations (Cancer Moonshot Blue Ribbon Panel, 2016):
- coverage of the costs of routine care and clinical trials for patients with cancer
- uniform patient consent forms and patient access to data
- use of a central IRB, data sharing among federal agencies, and adequate clinical trial site evaluation processes
- improved delivery of cancer care in communities through reducing economic burden of clinical trial enrollment on patients and practitioners, sharing EHR data, and releasing best practices for state vaccine registries
- incentives for development of pediatric cancer drugs, especially molecularly targeted therapies
- data-sharing mechanisms, including standardization and harmonization of data, licensing, and sharing among the private sector
A number of workshop speakers discussed examples of drug development pathways and clinical trial designs reflective of the move toward a new drug development paradigm for oncology. These examples included the development of
- vemurafenib, and
- EGFR T790M inhibitors.
Rothenberg said that the development of crizotinib serves as a good example of how new scientific information led to an amendment in the clinical trial protocol. The strong signal seen in first-in-human tests, combined with the understanding of its molecular mechanism of action, led to crizotinib’s rapid approval, Rothenberg added.
Crizotinib targets c-MET, a molecular signaling pathway that is thought to be mutated and activated in a large proportion of patients who became resistant to targeted therapies. Thus, Rothenberg said that “there was a good preclinical rationale for this drug.” In a Phase I test, 3 out of 37 patients responded to crizotinib, which was not considered a strong signal. However, it was later discovered, because of preclinical research from an unrelated group, that all three responders overexpressed a fusion mutation of anaplastic lymphoma kinase (ALK).27 Therefore, it became apparent that in addition to targeting c-MET, crizotinib inhibits ALK.
This information led Pfizer to work with FDA to develop a reliable companion diagnostic for crizotinib to detect the ALK mutation. In addition, Pfizer expanded the cohort of patients expressing the ALK fusion in the Phase I study, and a Phase II trial was opened to test the drug in lung cancer patients with the ALK mutation (Malik et al., 2014). Then Pfizer opened two Phase III trials simultaneously: One to test crizotinib as a second-line treatment for lung cancer patients versus single-agent chemotherapy (Shaw et al., 2013); the other was to test it as a first-line treatment in lung cancer patients versus combination chemotherapy (Solomon et al., 2014).
While the Phase III trials were opening, it was discovered that mutated versions of the ROS1 gene, another tyrosine kinase closely related to ALK, also were prevalent in patients with lung cancer. So Pfizer amended the Phase I trial to include lung cancer patients with this mutation. Accrual
to the Phase I, II, and III trials occurred simultaneously, as can be seen in Figure 10, and was completed in 2011.
In August 2011, FDA approved crizotinib for the treatment of patients with locally advanced or metastatic non-small cell lung cancer that is ALK-positive, about 6 years after it was first tested in patients (FDA, 2011). FDA additionally approved crizotinib for ROS1-positive mutations in 2016 on the basis of the Phase I and Phase II data.28
Rothenberg said a number of lessons can be learned from the crizotinib drug development experience, including how strong signals of activity can be seen in Phase I studies and that expansion cohorts in Phase I trials can provide an opportunity to rapidly and efficiently test a clinical hypothesis when there is a good scientific understanding of the disease and the drug. Rothenberg also emphasized that Phase I, II, and III trials of the same inves-
28 See https://www.fda.gov/newsevents/newsroom/pressannouncements/ucm490329.htm (accessed May 22, 2017).
tigational agent can be run simultaneously when strong signals of efficacy are detected and confirmed. This can result in an agent with truly transformational levels of activity reaching the market years ahead of schedule, particularly if the data are sufficiently rigorous to earn regulatory approval before Phase III trials are initiated, he said.
Rothenberg also reported on the development of avelumab, which is an immunotherapy checkpoint inhibitor of PD-L1. He noted that prior to avelumab entering clinical testing in 2013, the populations and diseases most likely to respond to this type of treatment were known from clinical studies of similar checkpoint inhibitors. This led Merck-Serono to test it in a Phase I study that had multiple cohorts of likely responders, with the assumption that the cohorts showing the most response would be expanded. Currently, there is an expansive international clinical trial program exploring the use of PD-L1 inhibition with avelumab to treat multiple types of cancer, such as solid tumors, gastric cancer, Merkel cell carcinoma,29 and non-small cell lung cancer (Pfizer, 2015). Multiple actions were then taken for different indications following this Phase I study:
- The strong signal seen in Phase I studies in the absence of major safety issues led the sponsor to skip Phase II and III trials and to directly seek approval of the drug for Merkel cell carcinoma, a disease for which there is an unmet medical need.
- A Phase IB study was undertaken combining avelumab with the tyrosine kinase inhibitor axitinib for metastatic kidney cancer (NIH, 2017b). Because each of these drugs had been tested previously in a large number of patients, there was abundant evidence of efficacy and safety, and low probability of overlapping toxicities that would cause major safety issues, according to Rothenberg. In addition, there was a strong scientific rationale for the combination to have synergistic efficacy. Positive results in this Phase IB trial are expected to lead directly to a Phase III trial of the combination, he said.
29 On March 23, 2017, FDA granted accelerated approval to avelumab for the treatment of patients 12 years and older with metastatic Merkel cell carcinoma. See https://www.fda.gov/Drugs/InformationOnDrugs/ApprovedDrugs/ucm547965.htm (accessed May 18, 2017).
- A Phase IB study was also undertaken combining avelumab with another immunotherapy to test the combination therapy for overlapping side effects (NIH, 2017a). This Phase IB study could progress to Phase II and then Phase III studies or proceed directly to a Phase III study, depending on the results from the Phase IB study.
The development of avelumab demonstrates that multiple paths may be pursued following a Phase I expanded cohort study. However, significant resources must be committed for long periods of time for this type of development, Rothenberg noted, making it unsuitable for most novel agents about which little is known. In addition, FDA has taken steps to limit the size of expansion cohorts and to require specific hypotheses to be tested in them. Despite those caveats, he noted that “successful application of this strategy can shorten clinical development substantially and speed new medicines to the marketplace.”
Rubin described the development of the PD-L1 targeted immunotherapy pembrolizumab. This drug is an example of how a single-arm Phase I dose-expansion trial involving patients with different types of solid tumors led to FDA approval. The Phase I protocol prespecified expansion of the cohort of patients with melanoma if high response rates were demonstrated. Initial striking responses in melanoma patients led to the expansion of that cohort, including patients who had previously not responded to ipilimumab, another cancer immunotherapy, as well as lung cancer patients.
Additional patients were added after an analysis of the melanoma patients in the Phase I study found that those who had progressed on ipilimumab might respond to pembrolizumab. A further expansion tested the appropriate dose in both melanoma and lung cancer patients. Finally, after initial analyses demonstrated that most patients who responded to pembrolizumab had high PD-L1 expression levels, Merck KGaA began developing a companion diagnostic, which involved adding more patients within the Phase I study. While the Phase I trials were ongoing, FDA granted pembrolizumab Breakthrough status.
Rubin noted that “we were moving so fast that we did not have the results back from the randomized dosing study when we had to take cohorts into our Phase III studies so those were also tested for dose and schedule.” Ultimately, nine amendments were made to the Phase I protocol and FDA
ultimately awarded three approvals: an accelerated approval for patients with ipilimumab-refractory melanoma; an accelerated approval for patients with previously treated non-small cell lung cancer whose tumors that had high expression of PD-L1; and approval of the first companion diagnostic test for cancer immunotherapy (FDA, 2014b, 2016; Merck, 2015). Although FDA required confirmatory studies, pembrolizumab was approved just 3 years after it was initially tested in humans. While the study was considered adequate for initial approval in the United States, Canada, and Australia, it was not deemed sufficient in the European Union, where the European Medicines Agency (EMA) wanted to see data from a randomized controlled trial.
Rubin noted several challenges with adaptive trial design that were encountered in the development of pembrolizumab. It created operational burden due to rapid accrual in multiple cohorts and increased manufacturing demands. Multiple amendments also generated complexity, and required experts in multiple cancer fields to collaborate. “There is complexity in bringing together separate groups in a single site to work together on a single protocol,” Rubin said. Finally, data analyses for simultaneous hypotheses and data curation from multiple sites presented challenges.
Despite the challenges, Rubin stated that an adaptive, multiple expansion cohort design, as used to test pembrolizumab, can efficiently enable testing of multiple hypotheses about dose, biomarker, and patient population, and can be performed with sufficient rigor to support regulatory filings. Most importantly, it can accelerate development and approval for transformative drugs with strong efficacy signals by avoiding delays in initiating separate trials for each cohort and hypothesis.
He concluded that single-arm Phase I trials may be especially beneficial for quickly evaluating new cancer immunotherapies that are expected to be effective for multiple tumor types, and thus require methods “where one could have an efficient path to approval rather than requiring large, randomized studies for every single tumor type.” He added that “when you are seeing tremendous efficacy in a single cohort, it can then be very difficult to randomize patients against a standard of care, particularly when the standard of care is not very effective.”
Flaherty reported on the development of vemurafenib, which illustrates the advantages of conducting serial biopsies on patients to determine
whether the drug is acting on its intended target, and to molecularly define resistance when it develops. Vemurafenib targets the BRAF protein, which is mutated in 60 percent of melanoma patients, as well as in some colon and other cancer patients. In vitro studies suggested no plateau in the dose–response relationship, so in the first Phase II dosing studies, Flaherty examined not only efficacy, but also biomarker status to verify drug activity and determine effectiveness at lower doses. Melanoma patients had variable responses to the drug, but because it was so efficacious in a subset of patients and there was great unmet medical need for an advanced melanoma treatment, FDA agreed to a single-arm Phase II study in melanoma patients, rather than randomizing patients to receive standard of care (Sosman et al., 2012). In August 2011, after conducting a Phase III trial, FDA approved vemurafenib for the treatment of patients with unresectable or metastatic melanoma with the BRAF V600E mutation (NCI, 2013). At the time, there was some debate about whether the randomized Phase III trial was necessary or ethical, Flaherty noted (Hey and Truog, 2015).
Unfortunately, the responses melanoma patients had to vemurafenib were often transient. To assess how patients were responding to vemurafenib, the investigators conducted biopsies before and during their treatment, as well as when their tumors progressed. These biopsies showed that the drug consistently blocked its intended target, despite variability in patients’ durability of responses to it, and that only patients receiving high doses of the drug responded. “The serial tumor biopsy data were quite informative because they showed there was a fairly homogeneous molecular effect on the target pathway that was lost over time when we did biopsies at the time of relapse,” Flaherty stated.
His analysis of the biopsies suggested that tumors in the relapsed patients had activated MEK. The analyses therefore suggested a combination therapy for future trials. Accordingly, Flaherty began treating melanoma patients in 2010 with BRAF-MEK combination therapy in a Phase I study. Phase II and III trials of a MEK inhibitor in combination with vemurafenib subsequently demonstrated improved overall survival of melanoma patients (Ugurel et al., 2016).
Three randomized Phase III trials have now demonstrated that the combination of a BRAF inhibitor (vemurafenib or dabrafenib) with a MEK inhibitor is better than BRAF inhibitor therapy alone. Flaherty summed up his presentation on BRAF inhibitors by stressing, “Interrogating what happens in patients’ tumors, until the day comes that we can do it purely in blood, is important and impactful in driving forward benefit in the field.”
The BRAF-MEK combination therapy shows durable responses in patients, with 3-year overall survival rates of as high as 44 percent (Flaherty et al., 2016). This approach suggested a biologically relevant and effective combination therapy that may otherwise have not been discovered.
Most patients with non-small cell lung cancer that is resistant to treatment with EGFR inhibitors have a single mutation called EGFR T790M (Kobayashi et al., 2005; Yun et al., 2008), reported Pasi Jänne, director of the Lowe Center for Thoracic Oncology at the Dana-Farber Cancer Institute. Osimertinib targets this mutation and was developed quickly and efficiently due to solid understanding of the underlying mechanism of action, and to innovative trial design combined with the Breakthrough Designation by FDA. He discussed the development pathway for osimertinib with that of rociletinib, an EGFR T790M inhibitor whose clinical development was stopped (see Box 9).
He pointed out that in the Phase I/Phase II study of osimertinib, small groups of patients whose tumors were resistant to EGFR inhibitors were tested with escalating doses of drug as they would be in a typical Phase I clinical trial. However, the protocol stipulated that if any groups showed response to osimertinib, these cohorts would be expanded and patients would be tested for the presence or absence of T790M (Jänne et al., 2015). “This allowed in a very rapid and expeditious manner to both identify effective doses, and secondarily to evaluate the efficacy of the agent,” Jänne said.
The T790M-positive patients were shown to have high response rates, with about half of them responding to the drug. This led to an FDA Fast-Track Designation, followed by a Breakthrough Therapy Designation. Randomized Phase III studies of osimertinib showed the same high response rates as seen in previous studies (Mok et al., 2017). Furthermore, a Phase I study in patients with treatment-naïve lung cancer also showed high response rates of 77 percent, leading investigators to conduct a Phase III trial that randomized patients to receive osimertinib or two other EGFR inhibitors. Results are expected to be published in 2017. Finally, patients treated with osimertinib also showed significant neurologic improvement in the central nervous system, which is important given that lung cancer often metastasizes to the brain, Jänne noted (Yang et al., 2016).
Because of a well-defined biological underpinning; innovative, flexible trial design; and clear biomarkers to identify patients most likely to benefit
from the drug, osimertinib was very quickly brought to market and made available to patients. FDA granted accelerated approval for osimertinib in 2015,30 just 4 years after it was first synthesized, and 2 years after the
30 See https://www.fda.gov/newsevents/newsroom/pressannouncements/ucm472525.htm (accessed May 23, 2017).
31 On March 30, 2017, FDA granted regular approval to osimertinib. See https://www.fda.gov/Drugs/InformationOnDrugs/ApprovedDrugs/ucm549683.htm (accessed May 22, 2017).
Schilsky said that improving the drug development paradigm in oncology is critical, given that cancer will soon become the leading cause of death in the United States. “There are more cancer cases than ever before despite our progress in treating cancer because of the ag[ing] of the population and the increasing prevalence of the disease,” Schilsky said. He said rapid progress in cancer research has led to a better biological understanding of cancer that has fostered the identification of new drug targets, including genetic aberrations that stimulate tumor growth, and those involved in the immune system’s response to tumors. New classes of cancer drugs often have a wider range of doses that are both more effective and less toxic than previous chemotherapies. There also has been rapid development of biomarker tests to stratify and characterize patient populations with differential prognoses and sensitivities to treatment. In addition, there are new sources of data, and although they may lack the precision and accuracy of clinical trial data, Schilsky noted that this challenge could be overcome through methodological tactics. Therefore, he said, “We should not make more of the distinction than we need to between real-world data and clinical trial data.”
FDA recognizes the novel strategies being used in oncology drug development and supports improvements aimed at making the process more efficient and expedient, Schilsky noted. FDA also has broader authority than it did in the past with passage of the 21st Century Cures Act, and despite some concerns voiced that this Act might make FDA too permissive at the sacrifice of patient safety, Schilsky stressed that “I do not expect FDA would suddenly abandon all of its standards for regulatory approval.”
Given all the recent changes in cancer drug development and its regulation, much of the discussion in the workshop focused on potential ways to capitalize on all of these changes to improve the drug development process. Schilsky pointed out several recurring themes at the workshop, the first and foremost being to put patients at the center of the drug development paradigm. “At the end of the day, the reason we do what we do and try to advance new drugs into the clinical workplace is to help improve outcomes for patients. So let us continue to think about this as patient-centered drug development and not product-centered drug development,” Schilsky stressed. To that end, there needs to be a focus on clinically meaningful outcomes and endpoints in clinical trials, including patient-reported outcomes.
Another recurrent topic of discussion at the workshop was that biomarker or diagnostic test development tends to lag behind therapeutic
development. “This seems to me to be something that should be fairly easy to address,” Schilsky said.
He noted that although the workshop began by focusing on speeding up the drug development process, many of the discussions had also highlighted the value of efficiency. In characterizing possible ways to achieve efficiency, he stated that “It is important to avoid waste, and there are many sources of waste in the whole process, including the very costly ones resulting from failing late.” In addition, although companies may wish to have a unique version of a popular efficacious drug class in their portfolio, such as the PD-L1 inhibitors, “at some point there is going to be saturation, and we ought to encourage drug developers to begin to focus on new targets and continuing areas of unmet medical need,” Schilsky said. Finally, he added that in aiming for efficient development, there are risks in both moving too fast and moving too slow. “If you move too slowly, you risk competitive disadvantage, but if you move too fast, you risk not really understanding your drug well enough or the way it should be used with the population in whom it will achieve maximal effectiveness.”
He added that “in our zeal to move product development quickly, we cannot lose sight of the first principles of drug development.” Those principles are to improve understanding of the underlying biology of the disease and target, and of the drug. The basic understanding of the pharmacology of the drug should include whether it is the right chemical entity, the optimal formulation, dosing, pharmacodynamics, and pharmacokinetics.
Schilsky referred to Brown’s notion that data gathered to support drug approvals or labeling should be fit for purpose. “We need to use the right study design at the right time with the right data sources to address the question that is being posed. We also need to ensure appropriate design and timing of confirmatory studies, and to balance rigor in the study design with flexibility to adapt to new information,” he stressed. And, more emphasis needs to be put on data standardization and data quality, he said.
Another consistent theme in the workshop was the need for improved collaboration and information sharing among all parties. “When incentives are aligned and there are common needs and goals, sharing even among commercial entities that have strong proprietary interests can be accomplished and hopefully can lead to improvements in the way we collectively develop drugs,” Schilsky said. However, aligning the culture of industry with the developing science in drug development can be challenging, even within a single entity, much less across a partnership. Disconnects occur between early and late development teams, and expert academic advisors
who have been involved in drug development plans at certain points in the process can end up being excluded from key decisions later, Schilsky said.
Many workshop discussions focused on the information needed to confidently advance investigational agents in the drug development process, Schilsky noted. “We have numerous endpoints in oncology, none of which are perfect and each of which has its own challenges. We have to be able to integrate our assessment of a drug using as much information as we can glean from all the endpoints available to us at any particular point in time,” he said. A related point made repetitively throughout the workshop was the need to learn throughout the drug development process and beyond FDA approval, not just from individual trials or individual cohorts, Schilsky said.
With rapid clinical testing, many drugs are entering the market with less learning behind them, Schilsky said. Some of that learning about new drugs could come with postmarketing studies, but such research is difficult to conduct in the United States. “Once that drug is on the market, it can be prescribed and if it gets paid for, it is a huge disincentive for either the doctor or the patient to participate in a postmarketing research study, which is one of the reasons why so many of these studies take so long to complete or in some cases, are never completed,” he said.
He summarized several potential strategies for improving drug development (see Box 10). “It all comes down to clearly articulating the question we want to answer and the best approach to answering that question among an entire range of clinical trial designs and data sources available to us.” He added that the Cancer Moonshot offers many opportunities for improving drug development. “It is incumbent on us all to take advantage of them and try to use the new funding and infrastructures that will be available to continue to address some of these issues.”
“Drug development is inherently risky and the drug development paradigm is essentially about managing uncertainty in the context of unmet medical need and business opportunity,” Schilsky concluded. “The more that we, as a community, can get comfortable with how to manage that uncertainty and make key decisions about regulatory approval, clinical use, and reimbursement in the face of that uncertainty and then continue to develop the information to reduce that uncertainty, then the better off we will be and the better off our patients will be.”
AACR (American Association for Cancer Research). 2011. AACR cancer progress report 2011.https://www.roswellpark.org/sites/default/files/node-files/asset/nid91575-2011-aacr-cpr-text-web.pdf (accessed February 27, 2017).
Annemans, L., M. Aristides, and M. Kubin. 2007. Real-life data: A growing need. https://www.ispor.org/news/articles/oct07/rld.asp (accessed March 6, 2017).
ASCO (American Society of Clinical Oncology). 2017. Homepage. http://www.tapur.org (accessed February 27, 2017).
Badenas, J. M. 2010. Globalization of clinical trials. https://www.slideshare.net/josepmariabadenas/globalization-of-clinical-trials-2010-josep-m-badenas (accessed March 6, 2017).
Banerjee, A., U. B. Chitnis, S. L. Jadhav, J. S. Bhawalkar, and S. Chaudhury. 2009. Hypothesis testing, type I and type II errors. Industrial Psychiatry Journal 18(2):127-131.
Bhan, S. S. 2017. Shaping the future of liquid biopsies. https://www.foundationmedicine.com/2017/01/shaping-the-future-of-liquid-biopsies (accessed February 28, 2017).
BloodPAC. 2017. About BloodPAC. https://www.bloodpac-data.org (accessed February 28, 2017).
Brahmer, J. R., C. G. Drake, I. Wollner, J. D. Powderly, J. Picus, W. H. Sharfman, E. Stankevich, A. Pons, T. M. Salay, T. L. McMiller, M. M. Gilson, C. Wang, M. Selby, J. M. Taube, R. Anders, L. Chen, A. J. Korman, D. M. Pardoll, I. Lowy, and S. L. Topalian. 2010. Phase I study of single-agent anti–programmed death-1 (MDX-1106) in refractory solid tumors: Safety, clinical activity, pharmacodynamics, and immunologic correlates. Journal of Clinical Oncology 28(19):3167-3175.
Bruzzi, P., L. D. Mastro, M. P. Sormani, L. Bastholt, M. Danova, C. Focan, G. Fountzilas, J. Paul, R. Rosso, and M. Venturini. 2005. Objective response to chemotherapy as a potential surrogate end point of survival in metastatic breast cancer patients. Journal of Clinical Oncology 23(22):5117-5125.
Buck, A. K., K. Herrmann, T. Stargardt, T. Dechow, B. J. Krause, and J. Schreyögg. 2010. Economic evaluation of PET and PET/CT in oncology: Evidence and methodologic approaches. Journal of Nuclear Medicine 51(3):401-412.
Cancer Moonshot Blue Ribbon Panel. 2016. Cancer Moonshot Blue Ribbon Panel report 2016. https://www.cancer.gov/research/key-initiatives/moonshot-cancer-initiative/blue-ribbon-panel/blue-ribbon-panel-report-2016.pdf (accessed March 7, 2017).
Catley, L., E. Weisberg, T. Kiziltepe, Y.-T. Tai, T. Hideshima, P. Neri, P. Tassone, P. Atadja, D. Chauhan, N. C. Munshi, and K. C. Anderson. 2006. Aggresome induction by proteasome inhibitor bortezomib and α-tubulin hyperacetylation by tubulin deacetylase (TDAC) inhibitor LBH589 are synergistic in myeloma cells. Blood 108(10):3441-3449.
Chen, M. S., P. N. Lara, J. H. T. Dang, D. A. Paterniti, and K. Kelly. 2014. Twenty years post-NIH Revitalization Act: Renewing the case for enhancing minority participation in cancer clinical trials. Cancer 120(7):1091-1096.
Claret, L., P. Girard, P. M. Hoff, E. V. Cutsem, K. P. Zuideveld, K. Jorga, J. Fagerberg, and R. Bruno. 2009. Model-based prediction of phase III overall survival in colorectal cancer on the basis of phase II tumor dynamics. Journal of Clinical Oncology 27(25):4103-4108.
CMS (Centers for Medicare & Medicaid Services). 2017. Oncology care model.https://innovation.cms.gov/initiatives/oncology-care (accessed February 28, 2017).
Cziraky, M., and M. Pollock. 2015. Real-World Evidence Studies. http://www.appliedclinicaltrialsonline.com/real-world-evidence-studies?pageID=1 (accessed June 30, 2017).
Denny, J. C., L. Bastarache, M. D. Ritchie, R. J. Carroll, R. Zink, J. D. Mosley, J. R. Field, J. M. Pulley, A. H. Ramirez, E. Bowton, M. A. Basford, D. S. Carrell, P. L. Peissig, A. N. Kho, J. A. Pacheco, L. V. Rasmussen, D. R. Crosslin, P. K. Crane, J. Pathak, S. J. Bielinski, S. A. Pendergrass, H. Xu, L. A. Hindorff, R. Li, T. A. Manolio, C. G. Chute, R. L. Chisholm, E. B. Larson, G. P. Jarvik, M. H. Brilliant, C. A. McCarty, I. J. Kullo, J. L. Haines, D. C. Crawford, D. R. Masys, and D. M. Roden. 2013. Systematic comparison of phenome-wide association study of electronic medical record data and genome-wide association study data. Nature Biotechnology 31(12):1102-1110.
Eisenhauer, E. A., P. Therasse, J. Bogaerts, L. H. Schwartz, D. Sargent, R. Ford, J. Dancey, S. Arbuck, S. Gwyther, M. Mooney, L. Rubinstein, L. Shankar, L. Dodd, R. Kaplan, D. Lacombe, and J. Verweij. 2009. New response evaluation criteria in solid tumours: Revised RECIST guideline (version 1.1). European Journal of Cancer 45(2):228-247.
Emanuel, E. J., D. Wendler, and C. Grady. 2000. What makes clinical research ethical? JAMA 283(20):2701-2711.
Emens, L. A., J. M. Asquith, J. M. Leatherman, B. J. Kobrin, S. Petrik, M. Laiko, J. Levi, M. M. Daphtary, B. Biedrzycki, A. C. Wolff, V. Stearns, M. L. Disis, X. Ye, S. Piantadosi, J. H. Fetting, N. E. Davidson, and E. M. Jaffee. 2009. Timed sequential treatment with cyclophosphamide, doxorubicin, and an allogeneic granulocytemacrophage colony-stimulating factor–secreting breast tumor vaccine: A chemotherapy dose-ranging factorial study of safety and immune activation. Journal of Clinical Oncology 27(35):5911-5918.
FDA (Food and Drug Administration). 2006. Guidance for clinical trial sponsors: Establishment and operation of clinical trial data monitoring committees.https://www.fda.gov/OHRMS/DOCKETS/98fr/01d-0489-gdl0003.pdf (accessed May 4, 2017).
FDA. 2011. FDA approves Xalkori with companion diagnostic for a type of late-stage lung cancer.https://www.fda.gov/NewsEvents/Newsroom/PressAnnouncements/ucm269856.htm (accessed March 8, 2017).
FDA. 2014a. FDA’s drug review process: Ensuring drugs are safe and effective. http://www.fda.gov/Drugs/ResourcesForYou/Consumers/ucm143534.htm (accessed February 27, 2017).
FDA. 2014b. FDA approves Keytruda for advanced melanoma. https://www.fda.gov/NewsEvents/Newsroom/PressAnnouncements/ucm412802.htm (accessed March 8, 2017).
FDA. 2014c. Guidance for industry expedited programs for serious conditions—drugs and biologics. https://www.fda.gov/downloads/Drugs/Guidances/UCM358301.pdf (accessed February 27, 2017).
FDA. 2015. Breakthrough therapy designation: Exploring the qualifying criteria. https://www.fda.gov/downloads/AboutFDA/CentersOffices/OfficeofMedicalProductsandTobacco/CDER/UCM447162.pdf (accessed February 27, 2017).
FDA. 2016. Pembrolizumab (Keytruda) checkpoint inhibitor. https://www.fda.gov/Drugs/InformationOnDrugs/ApprovedDrugs/ucm526430.htm (accessed March 8, 2017).
FDA. 2017a. A blueprint proposal for companion diagnostic comparability. https://www.fda.gov/downloads/MedicalDevices/NewsEvents/WorkshopsConferences/UCM439440.pdf (accessed March 7, 2017).
FDA. 2017b. Drugs@FDA: FDA approved drug products. http://www.accessdata.fda.gov/scripts/cder/daf (accessed March 2, 2017).
FDA. 2017c. Oncology Center of Excellence. https://www.fda.gov/AboutFDA/CentersOffices/OfficeofMedicalProductsandTobacco/ucm509057.htm (accessed February 28, 2017).
FDA. 2017d. Statement from FDA Commissioner Robert Califf, M.D. Announcing FDA oncology Center of Excellence launch. https://www.fda.gov/NewsEvents/Newsroom/PressAnnouncements/ucm537564.htm (accessed February 28, 2017).
FDA. 2017e. The voice ofthe patient: A series ofreports from FDA’s patient-focuseddrug development initiative. https://www.fda.gov/ForIndustry/UserFees/PrescriptionDrugUserFee/ucm368342.htm (accessed February 27, 2017).
Finn, R. S., J. Dering, D. Conklin, O. Kalous, D. J. Cohen, A. J. Desai, C. Ginther, M. Atefi, I. Chen, C. Fowst, G. Los, and D. J. Slamon. 2009. PD 0332991, a selective cyclin D kinase 4/6 inhibitor, preferentially inhibits proliferation of luminal estrogen receptor-positive human breast cancer cell lines in vitro. Breast Cancer Research 11(5):1-13.
Finn, R. S., J. P. Crown, I. Lang, K. Boer, I. M. Bondarenko, S. O. Kulyk, J. Ettl, R. Patel, T. Pinter, M. Schmidt, Y. Shparyk, A. R. Thummala, N. L. Voytko, C. Fowst, X. Huang, S. T. Kim, S. Randolph, and D. J. Slamon. 2015. The cyclin-dependent kinase 4/6 inhibitor palbociclib in combination with letrozole versus letrozole alone as first-line treatment of oestrogen receptor-positive, HER2-negative, advanced breast cancer (PALOMA-1/TRIO-18): A randomised Phase 2 study. The Lancet Oncology 16(1):25-35.
Finn, R. S., M. Martin, H. S. Rugo, S. Jones, S.-A. Im, K. Gelmon, N. Harbeck, O. N. Lipatov, J. M. Walshe, S. Moulder, E. Gauthier, D. R. Lu, S. Randolph, V. Diéras, and D. J. Slamon. 2016. Palbociclib and letrozole in advanced breast cancer. New England Journal of Medicine 375(20):1925-1936.
Flaherty, K., M. A. Davies, J. J. Grob, G. V. Long, P. D. Nathan, A. Ribas, C. Robert, D. Schadendorf, D. T. Frederick, M. R. Hammond, J. Jane-Valbuena, X. J. Mu, M. Squires, S. A. Jaeger, S. R. Lane, B. Mookerjee, and L. A. Garraway. 2016. Genomic analysis and 3-y efficacy and safety update of COMBI-d: A Phase 3 study of dabrafenib (D) + trametinib (T) vs D monotherapy in patients (pts) with unresectable or metastatic BRAF V600E/K-mutant cutaneous melanoma. Journal of Clinical Oncology 34:Supplemental abstract 9502. http://meetinglibrary.asco.org/content/168199-176 (accessed February 27, 2017).
FNIH (Foundation for the National Institutes of Health). 2017. Biomarkers Consortium—Vol-PACT: Advanced metrics and modeling with volumetric CT for precision analysis of clinical trial results. https://fnih.org/what-we-do/biomarkers-consortium/programs/vol-pact (accessed February 28, 2017).
Garrett, J. T., A. Chakrabarty, and C. L. Arteaga. 2011. Will PI3K pathway inhibitors be effective as single agents in patients with cancer? Oncotarget 2(12):1314-1321.
Garrison, L. P., P. J. Neumann, P. Erickson, D. Marshall, and C. D. Mullins. 2007. Using real-world data for coverage and payment decisions: The ISPOR Real-World Data Task Force report. Value in Health 10(5):326-335.
Genentech USA Inc. 2017. ALK: A specific oncogenic driver. https://www.biooncology.com/pathways/alk/what-is-alk.html (accessed February 28, 2017).
Görgün, G., M. K. Samur, K. B. Cowens, S. Paula, G. Bianchi, J. E. Anderson, R. E. White, A. Singh, H. Ohguchi, R. Suzuki, S. Kikuchi, T. Harada, T. Hideshima, Y.-T. Tai, J. P. Laubach, N. Raje, F. Magrangeas, S. Minvielle, H. Avet-Loiseau, N. C. Munshi, D. M. Dorfman, P. G. Richardson, and K. C. Anderson. 2015. Lenalidomide enhances immune checkpoint blockade-induced immune response in multiple myeloma. Clinical Cancer Research 21(20):4607-4618.
Grady, C., L. Eckstein, B. Berkman, D. Brock, R. Cook-Deegan, S. M. Fullerton, H. Greely, M. G. Hansson, S. Hull, S. Kim, B. Lo, R. Pentz, L. Rodriguez, C. Weil, B. S. Wilfond, and D. Wendler. 2015. Broad consent for research with biological samples: Workshop conclusions. The American Journal of Bioethics 15(9):34-42.
Habermann, T. M., E. A. Weller, V. A. Morrison, R. D. Gascoyne, P. A. Cassileth, J. B. Cohn, S. R. Dakhil, B. Woda, R. I. Fisher, B. A. Peterson, and S. J. Horning. 2006. Rituximab-CHOP versus CHOP alone or with maintenance rituximab in older patients with diffuse large B-cell lymphoma. Journal of Clinical Oncology 24(19):3121-3127.
Heck, M. M., M. Retz, C. D’Alessandria, I. Rauscher, K. Scheidhauer, T. Maurer, E. Storz, F. Janssen, M. Schottelius, H.-J. Wester, J. E. Gschwend, M. Schwaiger, R. Tauber, and M. Eiber. 2016. Systemic radioligand therapy with 177Lu labeled prostate specific membrane antigen ligand for imaging and therapy in patients with metastatic castration resistant prostate cancer. The Journal of Urology 196(2):382-391.
Herbst, R. S., D. R. Gandara, F. R. Hirsch, M. W. Redman, M. LeBlanc, P. C. Mack, L. H. Schwartz, E. Vokes, S. S. Ramalingam, J. D. Bradley, D. Sparks, Y. Zhou, C. Miwa, V. A. Miller, R. Yelensky, Y. Li, J. D. Allen, E. V. Sigal, D. Wholley, C. C. Sigman, G. M. Blumenthal, S. Malik, G. J. Kelloff, J. S. Abrams, C. D. Blanke, and V. A. Papadimitrakopoulou. 2015. Lung master protocol (Lung-MAP)—a biomarker-driven protocol for accelerating development of therapies for squamous cell lung cancer: SWOG S1400. Clinical Cancer Research 21(7):1514-1524.
Hey, S. P., and R. D. Truog. 2015. The question of clinical equipoise and patients’ best interests. AMA Journal of Ethics 17(12):1108-1115.
HHS (Department of Health and Human Services). 2017. Homepage. https://clinicaltrials.gov/ct2/home (accessed March 2, 2017).
Hirsch, F. R., A. McElhinny, D. Stanforth, J. Ranger-Moore, M. Jansson, K. Kulangara, W. Richardson, P. Towne, D. Hanks, B. Vennapusa, A. Mistry, R. Kalamegham, S. Averbuch, J. Novotny, E. Rubin, K. Emancipator, I. McCaffery, J. A. Williams, J. Walker, J. Longshore, M. S. Tsao, and K. M. Kerr. 2017. PD-L1 immunohistochemistry assays for lung cancer: Results from Phase 1 of the blueprint PD-L1 IHC assay comparison project. Journal of Thoracic Oncology 12(2):208-222.
Hortobagyi, G. N., S. M. Stemmer, H. A. Burris, Y.-S. Yap, G. S. Sonke, S. Paluch-Shimon, M. Campone, K. L. Blackwell, F. Andre, E. P. Winer, W. Janni, S. Verma, P. Conte, C. L. Areaga, D. A. Cameron, K. Petrakova, L. L. Hart, C. Villanueva, A. Chan, E. Jakobsen, A. Nusch, O. Burdaeva, E.-M. Grischke, E. Alba, E. Wist, N. Marschner, A. M. Favret, D. Yardley, T. Bachelot, L.-M. Tseng, S. Blau, F. Xuan, F. Souami, M. Miller, C. Germa, S. Hirawat, and J. O’Shaughnessy. 2016. Ribociclib as First-Line Therapy for HR-Positive, Advanced Breast Cancer. New England Journal of Medicine 375(18):1738-1748.
IOM (Institute of Medicine). 2008. Improving the quality of cancer clinical trials: Workshop summary. Washington, DC: The National Academies Press.
IOM. 2009. Beyond the HIPAA privacy rule: Enhancing privacy, improving health through research. Washington, DC: The National Academies Press.
IOM. 2010. A national cancer clinical trials system for the 21st century: Reinvigorating the NCI cooperative group program. Washington, DC: The National Academies Press.
IOM. 2012. Evolution of translational omics: Lessons learned and the path forward. Washington, DC: The National Academies Press.
IOM. 2013. Delivering high-quality cancer care: Charting a new course for a system in crisis. Washington, DC: The National Academies Press.
I-SPY (Investigation of Serial Studies to Predict Your Therapeutic Response with Imaging and Molecular Analysis). 2017. About I-SPY TRIALs. http://www.ispytrials.org/about (accessed March 7, 2017).
Jänne, P. A., J. C.-H. Yang, D.-W. Kim, D. Planchard, Y. Ohe, S. S. Ramalingam, M.-J. Ahn, S.-W. Kim, W.-C. Su, L. Horn, D. Haggstrom, E. Felip, J.-H. Kim, P. Frewer, M. Cantarini, K. H. Brown, P. A. Dickinson, S. Ghiorghiu, and M. Ranson. 2015. AZD9291 in EGFR inhibitor–resistant non–small-cell lung cancer. New England Journal of Medicine 372(18):1689-1699.
Jarow, J. P., P. Lurie, S. C. Ikenberry, and S. Lemery. 2017. Overview of FDA’s expanded access program for investigational drugs. Therapeutic Innovation & Regulatory Science 51(2):177-179.
Johnson, K. R., C. Ringland, B. J. Stokes, D. M. Anthony, N. Freemantle, A. Irs, S. R. Hill, and R. L. Ward. 2006. Response rate or time to progression as predictors of survival in trials of metastatic colorectal cancer or non-small-cell lung cancer: A meta-analysis. The Lancet Oncology 7(9):741-746.
Johnson, P., M. Federico, A. Kirkwood, A. Fosså, L. Berkahn, A. Carella, F. d’Amore, G. Enblad, A. Franceschetto, M. Fulham, S. Luminari, M. O’Doherty, P. Patrick, T. Roberts, G. Sidra, L. Stevens, P. Smith, J. Trotman, Z. Viney, J. Radford, and S. Barrington. 2016. Adapted treatment guided by interim PET-CT scan in advanced Hodgkin’s lymphoma. New England Journal of Medicine 374(25):2419-2429.
Kaplan, S. 2016. House approves the 21st Century Cures Act, sending landmark bill to Senate. https://www.statnews.com/2016/11/30/21st-century-cures-act-house (accessed February 28, 2017).
Khoja, L., M. O. Butler, S. P. Kang, S. Ebbinghaus, and A. M. Joshua. 2015. Pembrolizumab. Journal for ImmunoTherapy of Cancer 3(1):36.
Khozin, S., K. Liu, J. P. Jarow, and R. Pazdur. 2015. Regulatory watch: Why do oncology drugs fail to gain US regulatory approval? Nature Reviews Drug Discovery 14(7):450-451.
Kim, E. S., R. S. Herbst, I. I. Wistuba, J. J. Lee, G. R. Blumenschein, A. Tsao, D. J. Stewart, M. E. Hicks, J. Erasmus, S. Gupta, C. M. Alden, S. Liu, X. Tang, F. R. Khuri, H. T. Tran, B. E. Johnson, J. V. Heymach, L. Mao, F. Fossella, M. Kies, V. Papadimitrakopoulou, S. E. Davis, S. M. Lippman, and W. K. Hong. 2011. The battle trial: Personalizing therapy for lung cancer. Cancer Discovery 1(1):44-53.
Kim, E. S., D. Bernstein, S. G. Hilsenbeck, C. H. Chung, A. P. Dicker, J. L. Ersek, S. Stein, F. R. Khuri, E. Burgess, K. Hunt, P. Ivy, S. S. Bruinooge, N. Meropol, and R. L. Schilsky. 2015. Modernizing eligibility criteria for molecularly driven trials. Journal of Clinical Oncology 33(25):2815-2820.
Kobayashi , S., T. J. Boggon, T. Dayaram, P. A. Jänne, O. Kocher, M. Meyerson, B. E. Johnson, M. J. Eck, D. G. Tenen, and B. Halmos. 2005. EGFR mutation and resistance of non–small-cell lung cancer to gefitinib. New England Journal of Medicine 352(8):786-792.
Koyfman, S. A., M. Agrawal, E. Garrett-Mayer, B. Krohmal, E. Wolf, E. J. Emanuel, and C. P. Gross. 2007. Risks and benefits associated with novel Phase 1 oncology trial designs. Cancer 110(5):1115-1124.
Krönke, J., N. D. Udeshi, A. Narla, P. Grauman, S. N. Hurst, M. McConkey, T. Svinkina, D. Heckl, E. Comer, X. Li, C. Ciarlo, E. Hartman, N. Munshi, M. Schenone, S. L. Schreiber, S. A. Carr, and B. L. Ebert. 2014. Lenalidomide causes selective degradation of IKZF1 and IKZF3 in multiple myeloma cells. Science 343(6168):301-305.
Lemery, S. J., S. Mailankody, D. Kazandjian, M.-A. D. Smit, G. M. Blumenthal, T. Kim, P. Keegan, and R. Pazdur. 2016. Food and Drug Administration analysis of 1332 single patient and emergency use expanded access (compassionate use) requests for patients with cancer over a duration of three years (2012-2014). Journal of Clinical Oncology.http://meetinglibrary.asco.org/content/162351-176 (accessed February 28, 2017).
Loudon, K., S. Treweek, F. Sullivan, P. Donnan, K. E. Thorpe, and M. Zwarenstein. 2015. The PRECIS-2 tool: Designing trials that are fit for purpose. British Medical Journal 350:1-11.
Lu, G., R. E. Middleton, H. Sun, M. Naniong, C. J. Ott, C. S. Mitsiades, K.-K. Wong, J. E. Bradner, and W. G. Kaelin. 2014. The myeloma drug lenalidomide promotes the cereblon-dependent destruction of Ikaros proteins. Science 343(6168):305-309.
Lung-MAP (Lung Cancer Master Protocol) Clinical Trials. 2017. Healthcare providers.http://www.lung-map.org/healthcare-providers (accessed May 24, 2017).
Maemondo, M., A. Inoue, K. Kobayashi, S. Sugawara, S. Oizumi, H. Isobe, A. Gemma, M. Harada, H. Yoshizawa, I. Kinoshita, Y. Fujita, S. Okinaga, H. Hirano, K. Yoshimori, T. Harada, T. Ogura, M. Ando, H. Miyazawa, T. Tanaka, Y. Saijo, K. Hagiwara, S. Morita, and T. Nukiwa. 2010. Gefitinib or chemotherapy for non–small-cell lung cancer with mutated EGFR. New England Journal of Medicine 362(25):2380-2388.
Malik, S. M., V. E. Maher, K. E. Bijwaard, R. L. Becker, L. Zhang, S. W. Tang, P. Song, Q. Liu, A. Marathe, B. Gehrke, W. Helms, D. Hanner, R. Justice, and R. Pazdur. 2014. U.S. Food and Drug Administration approval: Crizotinib for treatment of advanced or metastatic non–small cell lung cancer that is anaplastic lymphoma kinase positive. Clinical Cancer Research 20(8):2029-2034.
Mateo, J., M. Ong, D. S. P. Tan, M. A. Gonzalez, and J. S. de Bono. 2013. Appraising iniparib, the PARP inhibitor that never was—what must we learn? Nature Reviews: Clinical Oncology 10(12):688-696.
Merck. 2015. FDA approves Keytruda® (pembrolizumab) for the treatment of patients with metastatic non-small cell lung cancer whose tumors express PD-L1 with disease progression on or after platinum-containing chemotherapy.http://www.mercknewsroom.com/newsrelease/prescription-medicine-news/fda-approves-keytruda-pembrolizumab-treatmentpatients-metas (accessed March 8, 2017).
Michel, L. S., S. Dyroff, F. J. Brooks, K. J. Spayd, S. Lim, J. T. Engle, S. Phillips, B. Tan, A. Wang-Gillam, C. Bognar, W. Chu, D. Zhou, R. H. Mach, R. Laforest, and D. L. Chen. 2016. PET of poly (ADP-ribose) polymerase activity in cancer: Preclinical assessment and first in-human studies. Radiology 282(2):453-463.
Miller, A. B., B. Hoogstraten, M. Staquet, and A. Winkler. 1981. Reporting results of cancer treatment. Cancer 47(1):207-214.
Mok, T. S., Y.-L. Wu, M.-J. Ahn, M. C. Garassino, H. R. Kim, S. S. Ramalingam, F. A. Shepherd, Y. He, H. Akamatsu, W. S. M. E. Theelen, C. K. Lee, M. Sebastian, A. Templeton, H. Mann, M. Marotti, S. Ghiorghiu, and V. A. Papadimitrakopoulou. 2017. Osimertinib or platinum–pemetrexed in EGFR T790m–positive lung cancer. New England Journal of Medicine 376(7):629-640.
Myers, R. H., D. C. Montgomery, and C. M. Anderson-Cook. 2016. Response surface methodology: Process and product optimization using designed experiments. Hoboken, NJ: John Wiley & Sons, Inc.
NASEM (National Academies of Sciences, Engineering, and Medicine). 2016a. Biomarker tests for molecularly targeted therapies: Key to unlocking precision medicine. Washington, DC: The National Academies Press.
NASEM. 2016b. Policy issues in the clinical development and use of immunotherapy for cancer treatment: Proceedings of a workshop. Washington, DC: The National Academies Press.
NCI (National Cancer Institute). 2013. FDA approval for vemurafenib. https://www.cancer.gov/about-cancer/treatment/drugs/fda-vemurafenib (accessed March 8, 2017).
NCI. 2014. Lung-MAP launches: First precision medicine trial from National Clinical Trials Network. https://www.cancer.gov/news-events/press-releases/2014/LungMAPlaunch (accessed March 8, 2017).
NCI. 2016. Lung-MAP: Master protocol for squamous cell lung cancer. https://www.cancer.gov/types/lung/research/lung-map (accessed March 8, 2017).
NCI. 2017a. NCI-Molecular Analysis for Therapy Choice (NCI-MATCH) Trial. https://www.cancer.gov/about-cancer/treatment/clinical-trials/nci-supported/nci-match (accessed February 27, 2017).
NCI. 2017b. NCI-related Cancer Moonshot activities. https://www.cancer.gov/research/key-initiatives/moonshot-cancer-initiative/milestones/nci-activities (accessed March 7, 2017).
NIH (National Institutes of Health). 2017a. A Phase IB study to evaluate the safety, tolerability, and pharmacokinetics of avelumab in combination with M9241(NHS-IL12) (COMBO). https://clinicaltrials.gov/ct2/show/NCT02994953 (accessed March 8, 2017).
NIH. 2017b. A study of avelumab in combination with axitinib in advanced renal cell cancer (JAVELIN Renal 100). https://clinicaltrials.gov/ct2/show/NCT02493751 (accessed March 8, 2017).
Pancreatic Cancer Action Network. 2017. Precision promise. https://www.pancan.org/research/precision-promise (accessed February 28, 2017).
Pfizer. 2015. Avelumab. https://www.pfizer.com/files/news/asco/Merck-PfizerAlliance_AvelumabFactSheet_19May2015US.pdf (accessed May 18, 2017).
Postel-Vinay, S., C. Gomez-Roca, L. R. Molife, B. Anghan, A. Levy, I. Judson, J. D. Bono, J.-C. Soria, S. Kaye, and X. Paoletti. 2011. Phase I trials of molecularly targeted agents: Should we pay more attention to late toxicities? Journal of Clinical Oncology 29(13):1728-1735.
Project Data Sphere LLC. 2017. About us. https://www.projectdatasphere.org/projectdatasphere/html/about (accessed February 28, 2017).
Prowell, T. M., M. R. Theoret, and R. Pazdur. 2016. Seamless oncology-drug development. New England Journal of Medicine 374(21):2001-2003.
Raju, G. K., K. Gurumurthi, and R. Domike. 2016a. Benefit−risk analysis for decision-making: An approach. Clinical Pharmacology and Therapeutics 100(6):654-671.
Raju, G. K., K. Gurumurthi, R. Domike, D. Kazandjian, G. Blumenthal, R. Pazdur, and J. Woodcock. 2016b. A benefit–risk analysis approach to capture regulatory decision-making: Non-small cell lung cancer. Clinical Pharmacology and Therapeutics 100(6):672-684.
Rathkopf, D. E., M. J. Morris, J. J. Fox, D. C. Danila, S. F. Slovin, J. H. Hager, P. J. Rix, E. Chow Maneval, I. Chen, M. Gönen, M. Fleisher, S. M. Larson, C. L. Sawyers, and H. I. Scher. 2013. Phase I study of ARN-509, a novel antiandrogen, in the treatment of castration-resistant prostate cancer. Journal of Clinical Oncology 31(28):3525-3530.
Ribas, A., F. S. Hodi, M. Callahan, C. Konto, and J. Wolchok. 2013. Hepatotoxicity with combination of vemurafenib and ipilimumab. New England Journal of Medicine 368(14):1365-1366.
Richardson, P. G., R. L. Schlossman, M. Alsina, D. M. Weber, S. E. Coutre, C. Gasparetto, S. Mukhopadhyay, M. S. Ondovik, M. Khan, C. S. Paley, and S. Lonial. 2013. Panorama 2: Panobinostat in combination with bortezomib and dexamethasone in patients with relapsed and bortezomib-refractory myeloma. Blood 122(14):2331.
Richardson, P. G., S. Jagannath, P. Moreau, A. Jakubowiak, M. S. Raab, T. Facon, R. Vij, D. White, D. E. Reece, L. Benboubker, J. Zonder, L. C. Tsao, K. C. Anderson, E. Bleickardt, A. K. Singhal, and S. Lonial. 2014. Final results for the 1703 Phase Ib/2 study of elotuzumab in combination with lenalidomide and dexamethasone in patients with relapsed/refractory multiple myeloma. Blood 124(21):302.
Robert, C., G. V. Long, B. Brady, C. Dutriaux, M. Maio, L. Mortier, J. C. Hassel, P. Rutkowski, C. McNeil, E. Kalinka-Warzocha, K. J. Savage, M. M. Hernberg, C. Lebbé, J. Charles, C. Mihalcioiu, V. Chiarion-Sileni, C. Mauch, F. Cognetti, A. Arance, H. Schmidt, D. Schadendorf, H. Gogas, L. Lundgren-Eriksson, C. Horak, B. Sharkey, I. M. Waxman, V. Atkinson, and P. A. Ascierto. 2015. Nivolumab in previously untreated melanoma without BRAF mutation. New England Journal of Medicine 372(4):320-330.
Roland, M., and D. J. Torgerson. 1998. What are pragmatic trials? British Medical Journal 316(7127):285.
San-Miguel, J. F., V. T. M. Hungria, S.-S. Yoon, M. Beksac, M. A. Dimopoulos, A. Elghandour, W. W. Jedrzejczak, A. Günther, T. N. Nakorn, N. Siritanaratkul, P. Corradini, S. Chuncharunee, J.-J. Lee, R. L. Schlossman, T. Shelekhova, K. Yong, D. Tan, T. Numbenjapon, J. D. Cavenagh, J. Hou, R. LeBlanc, H. Nahi, L. Qiu, H. Salwender, S. Pulini, P. Moreau, K. Warzocha, D. White, J. Bladé, W. Chen, J. de la Rubia, P. Gimsing, S. Lonial, J. L. Kaufman, E. M. Ocio, L. Veskovski, S. K. Sohn, M.-C. Wang, J. H. Lee, H. Einsele, M. Sopala, C. Corrado, B.-R. Bengoudifa, F. Binlich, and P. G. Richardson. 2014. Panobinostat plus bortezomib and dexamethasone versus placebo plus bortezomib and dexamethasone in patients with relapsed or relapsed and refractory multiple myeloma: A multicentre, randomised, double-blind Phase 3 trial. The Lancet Oncology 15(11):1195-1206.
Sequist, L. V., J.-C. Soria, J. W. Goldman, H. A. Wakelee, S. M. Gadgeel, A. Varga, V. Papadimitrakopoulou, B. J. Solomon, G. R. Oxnard, R. Dziadziuszko, D. L. Aisner, R. C. Doebele, C. Galasso, E. B. Garon, R. S. Heist, J. Logan, J. W. Neal, M. A. Mendenhall, S. Nichols, Z. Piotrowska, A. J. Wozniak, M. Raponi, C. A. Karlovich, S. Jaw-Tsai, J. Isaacson, D. Despain, S. L. Matheny, L. Rolfe, A. R. Allen, and D. R. Camidge. 2015. Rociletinib in EGFR-mutated non–small-cell lung cancer. New England Journal of Medicine 372(18):1700-1709.
Sequist, L. V., J.-C. Soria, and D. R. Camidge. 2016. Update to rociletinib data with the RECIST confirmed response rate. New England Journal of Medicine 374(23):2296-2297.
Seymour, L., J. Bogaerts, A. Perrone, R. Ford, L. H. Schwartz, S. Mandrekar, N. U. Lin, S. Litière, J. Dancey, A. Chen, F. S. Hodi, P. Therasse, O. S. Hoekstra, L. K. Shankar, J. D. Wolchok, M. Ballinger, C. Caramella, and E. G. E. de Vries. 2017. IRECIST: Guidelines for response criteria for use in trials testing immunotherapeutics. The Lancet Oncology 18(3):e143-e152.
Shaw, A. T., D.-W. Kim, K. Nakagawa, T. Seto, L. Crinó, M.-J. Ahn, T. De Pas, B. Besse, B. J. Solomon, F. Blackhall, Y.-L. Wu, M. Thomas, K. J. O’Byrne, D. Moro-Sibilot, D. R. Camidge, T. Mok, V. Hirsh, G. J. Riely, S. Iyer, V. Tassell, A. Polli, K. D. Wilner, and P. A. Jänne. 2013. Crizotinib versus chemotherapy in advanced ALK-positive lung cancer. New England Journal of Medicine 368(25):2385-2394.
Sheiner, L. B. 1997. Learning versus confirming in clinical drug development. Clinical Pharmacology and Therapeutics 61(3):275-291.
Sherman, R. E., S. A. Anderson, G. J. Dal Pan, G. W. Gray, T. Gross, N. L. Hunter, L. LaVange, D. Marinac-Dabic, P. W. Marks, M. A. Robb, J. Shuren, R. Temple, J. Woodcock, L. Q. Yue, and R. M. Califf. 2016. Real-world evidence—what is it and what can it tell us? New England Journal of Medicine 375(23):2293-2297.
Solomon, B. J., T. Mok, D.-W. Kim, Y.-L. Wu, K. Nakagawa, T. Mekhail, E. Felip, F. Cappuzzo, J. Paolini, T. Usari, S. Iyer, A. Reisman, K. D. Wilner, J. Tursi, and F. Blackhall. 2014. First-line crizotinib versus chemotherapy in ALK-positive lung cancer. New England Journal of Medicine 371(23):2167-2177.
Sosman, J. A., K. B. Kim, L. Schuchter, R. Gonzalez, A. C. Pavlick, J. S. Weber, G. A. McArthur, T. E. Hutson, S. J. Moschos, K. T. Flaherty, P. Hersey, R. Kefford, D. Lawrence, I. Puzanov, K. D. Lewis, R. K. Amaravadi, B. Chmielowski, H. J. Lawrence, Y. Shyr, F. Ye, J. Li, K. B. Nolop, R. J. Lee, A. K. Joe, and A. Ribas. 2012. Survival in BRAF V600-mutant advanced melanoma treated with vemurafenib. New England Journal of Medicine 366(8):707-714.
START (South Texas Accelerated Research Therapeutics). 2017. About START: Our mission. http://startthecure.com/about_start.php (accessed February 28, 2017).
Taylor Wessing LLP. 2017. Adaptive licensing: A model approach?https://united-kingdom.taylorwessing.com/synapse/regulatory_adaptivelicense.html (accessed February 27, 2017).
Thatcher, N., A. Chang, P. Parikh, J. Rodrigues Pereira, T. Ciuleanu, J. von Pawel, S. Thongprasert, E. H. Tan, K. Pemberton, V. Archer, and K. Carroll. 2005. Gefitinib plus best supportive care in previously treated patients with refractory advanced non-small-cell lung cancer: Results from a randomised, placebo-controlled, multicentre study (Iressa Survival Evaluation in Lung Cancer). The Lancet 366(9496):1527-1537.
Theoret, M. R., L. H. Pai-Scherf, M. K. Chuk, T. M. Prowell, S. Balasubramaniam, T. Kim, G. Kim, P. G. Kluetz, P. Keegan, and R. Pazdur. 2015. Expansion cohorts in first-inhuman solid tumor oncology trials. Clinical Cancer Research 21(20):4545-4551.
Therasse, P., S. G. Arbuck, E. A. Eisenhauer, J. Wanders, R. S. Kaplan, L. Rubinstein, J. Verweij, M. Van Glabbeke, A. T. van Oosterom, M. C. Christian, and S. G. Gwyther. 2000. New guidelines to evaluate the response to treatment in solid tumors. Journal of the National Cancer Institute 92(3):205-216.
Tighiouart, M., S. Piantadosi, and A. Rogatko. 2014. Dose finding with drug combinations in cancer Phase I clinical trials using conditional escalation with overdose control. Statistics in Medicine 33(22):3815-3829.
Tighiouart, M., Q. Li, and A. Rogatko. 2017. A Bayesian adaptive design for estimating the maximum tolerated dose curve using drug combinations in cancer Phase I clinical trials. Statistics in Medicine 36(2):280-290.
Ugurel, S., J. Röhmel, P. A. Ascierto, K. T. Flaherty, J. J. Grob, A. Hauschild, J. Larkin, G. V. Long, P. Lorigan, G. A. McArthur, A. Ribas, C. Robert, D. Schadendorf, and C. Garbe. 2016. Survival of patients with advanced metastatic melanoma: The impact of novel therapies. European Journal of Cancer 53:125-134.
Vestbo, J., D. Leather, N. Diar Bakerly, J. New, J. M. Gibson, S. McCorkindale, S. Collier, J. Crawford, L. Frith, C. Harvey, H. Svedsater, and A. Woodcock. 2016. Effectiveness of fluticasone furoate–vilanterol for COPD in clinical practice. New England Journal of Medicine 375(13):1253-1260.
Wahl, R. L., H. Jacene, Y. Kasamon, and M. A. Lodge. 2009. From RECIST to PERCIST: Evolving considerations for PET response criteria in solid tumors. Journal of Nuclear Medicine 50(Suppl 1):122S-150S.
Wang, Y., C. Sung, C. Dartois, R. Ramchandani, B. P. Booth, E. Rock, and J. Gobburu. 2009. Elucidation of relationship between tumor size and survival in non-small-cell lung cancer patients can aid early decision making in clinical drug development. Clinical Pharmacology and Therapeutics 86(2):167-174.
Wang, Y., K. Ayres, D. A. Goldman, M. N. Dickler, A. Bardia, I. A. Mayer, E. P. Winer, J. Fredrickson, C. L. Arteaga, J. Baselga, H. C. Manning, U. Mahmood, and G. A. Ulaner. 2016. 18f-fluoroestradiol PET/CT measurement of estrogen receptor suppression during a phase I trial of the novel estrogen receptor-targeted therapeutic GDC-0810. Clinical Cancer Research 23(12):3053-3060.
The White House. 2016. Memorandum—White House Cancer Moonshot Task Force. https://obamawhitehouse.archives.gov/the-press-office/2016/01/28/memorandum-white-house-cancer-moonshot-task-force (accessed February 28, 2017).
Winter, G. E., D. L. Buckley, J. Paulk, J. M. Roberts, A. Souza, S. Dhe-Paganon, and J. E. Bradner. 2015. Phthalimide conjugation as a strategy for in vivo target protein degradation. Science 348(6241):1376-1381.
WMA (World Medical Association). 2017. WMA Declaration of Helsinki—ethical principles for medical research involving human subjects. https://www.wma.net/policies-post/wma-declaration-of-helsinki-ethical-principles-for-medical-research-involving-human-subjects (accessed May 12, 2017).
Wolchok, J. D., A. Hoos, S. O’Day, J. S. Weber, O. Hamid, C. Lebbé, M. Maio, M. Binder, O. Bohnsack, G. Nichol, R. Humphrey, and F. S. Hodi. 2009. Guidelines for the evaluation of immune therapy activity in solid tumors: Immune-related response criteria. Clinical Cancer Research 15(23):7412-7420.
Yang, J. C.-H., D.-W. Kim, S.-W. Kim, B. C. Cho, J.-S. Lee, X. Ye, X. Yin, Z. Yang, H. Jiang, and M.-J. Ahn. 2016. Osimertinib activity in patients (PTs) with leptomeningeal (LM) disease from non-small cell lung cancer (NSCLC): updated results from BLOOM, a phase I study. American Society of Clinical Oncology Abstract 9002.
Yun, C.-H., K. E. Mengwasser, A. V. Toms, M. S. Woo, H. Greulich, K.-K. Wong, M. Meyerson, and M. J. Eck. 2008. The T790M mutation in EGFR kinase causes drug resistance by increasing the affinity for ATP. Proceedings of the National Academy of Sciences of the United States of America 105(6):2070-2075.
Yver, A. 2016. Osimertinib (AZD9291)—a science-driven, collaborative approach to rapid drug design and development. Annals of Oncology 27(6):1165-1170.