The way we run clinical trials hasn't changed significantly in decades. Yes, we're doing things online instead of on paper, and, yes, most organizations are starting to be open to new types of measurements, in the home instead of in the clinic, with wearables and other new-generation devices, and, sure, there is a growing sense that we can enhance the data set with new approaches and ideas.

But it is still early days, and trials are still on the fringes of the data revolution—for now, partly because of the huge expense involved in launching a trial—and thus the fear of taking an expensive risk on the unknown—and partly because there hasn't been much reason or industry impetus to move trial design, trial collection, or trial access forward. That has to change. It's that simple.

To reach the full potential that these new patient equations offer, clinical trials need to evolve. And they need to evolve in three ways. We need to change the way patients find and participate in trials (what people call “access”), we need to become comfortable collecting new kinds of data in new ways (going all the way in terms of scale from DNA to behavior and our environment), and we need to move to new mathematical designs. We need to innovate with new trial frameworks and techniques that will help us more effectively discover the inputs and outputs of the equations that will lead to maximum benefit for patients. In this chapter, we cover all of these issues in depth.

Expanding Access to Trials

According to the New York Times, fewer than 5% of adult cancer patients in America enroll in clinical trials, when greater rates would likely save lives—not just the lives of patients who enroll but the lives of future generations who would benefit from the increased research opportunities.¹ And while part of the reason is the necessary eligibility requirements of certain research studies, it is also the case that patients aren't always (or even often) steered toward trials that might benefit them, or made aware that even those who will not receive the study drug will still receive the same standard of care treatment that they would outside the trial—and for free.

The New York Times piece pushes for clinical trial navigators to help patients find the trials that might benefit them most—but the education needs to spread throughout the health care system, with many clinicians needing just as much information and guidance as patients. While it is true that trials won't help everyone—another New York Times article cites numbers above 90% as the failure rate of precision medicine studies—it's also the case that the patients in these trials are generally the ones whose diseases have proven most resistant to standard treatments—and the only way to improve these numbers is going to be with further trial research.²

One advocate who has been fighting for increased trial access for years is T. J. Sharpe, who was diagnosed with stage 4 melanoma in 2012. “I was originally diagnosed with stage 1b melanoma that I had removed when I was 25 years old, and then 12 years later I get diagnosed with stage 4.…[M]y son was only 4 weeks old at the time,” Sharpe told CURE magazine, a consumer publication focused on cancer.³

Faced with his diagnosis—and an oncologist telling him he'd be surprised if Sharpe survived even two years—Sharpe sought out immunotherapy trials, and after one failed trial, he tried again—and after years on the drug Keytruda, has been cancer-free since August 2017.⁴ I talked to Sharpe, and he emphasized how difficult it still is for patients to even be referred to trials.⁵ Since diagnosis, he has worked as a patient advocate and consultant to the research industry, and emphasizes the need for more information—and more tools, for doctors and patients.

Run by the National Institutes of Health (NIH), ClinicalTrials.gov is a publicly available resource that describes itself as “a database of privately and publicly funded clinical studies conducted around the world.”⁶ I receive emails and calls regularly from those who have received unfortunate diagnoses (or know people who have), and see me as their connection to the clinical trial industry. They have heard about the potential to access cutting-edge therapies through clinical trials, and they want advice about what trials they might be candidates for, and which might be of the highest potential benefit to them. ClinicalTrials.gov is inevitably the tool I use to respond. It is not, however, without its flaws.

“ClinicalTrials.gov is not a user-friendly tool, and wasn't really intended to be,” T. J. Sharpe explains. “It was designed to be a repository for results, but it's being used to find trials because right now it's the only way.…There often isn't a clear way to find the right trials for the patient, to match their health with the criteria in the database, to evaluate whether it might work well for them.”

Sharpe sees three primary problems with the current system: access to general knowledge about clinical trials, access to a useful database that can effectively match the right trial to the right person at the right time, and access to understandable results. “How do you evaluate apples to apples?” he asks. “If you have three different companies announcing trial results that all show efficacy, how can you compare them, when they might all have different endpoints, different populations, and more?”

Recalling our phase-diagram ideal of matching the right patient to the right therapy at the right time, Sharpe is highlighting the problem that there is no view into the available set of clinical trials that gets us anywhere close. There's no way to see where an individual might map across the necessary dimensions to be effectively matched to the selection of therapies that could treat them. Of course, we might not know yet if a particular experimental therapy will work for a given individual. That's why we do the research. However, even seeing where those potential matches might be—in one comprehensive, organized, accessible place—could help trial volunteers have the best possible chance of finding something that might cure or slow the progression of their disease.

It's not just patients who don't have the information they need to compare trials—it's physicians, too. There's no guide to interpretation, no way to separate out subsets of the population so someone can see how a particular trial worked for patients like them—with similar genetics, general health, and comorbidities. There are simply not enough meaningful insights being generated from the data and exposed to either practitioners or patients.

Sharpe sees the patient equation challenge as a big one right now. The data is in many different silos, and the industry isn't as motivated to evolve as he would have hoped. Being patient-centric is a buzzword, he fears, but doesn't actually translate into creating trial designs that let patients compare one drug to another, or figure out what trial might be best. His hope is that the life sciences companies will start to understand this and see that it is better business to get effective and actionable information to patients and doctors, and enable patients to become better advocates for themselves.

What might that translate into, from a practical perspective? Sharpe imagines a trusted source of trial information—far more user-friendly than the ClinicalTrials.gov site—where life sciences companies and other health care providers can contribute to creating a one-stop shop for patients, doctors, and researchers. If someone gets a new diagnosis, they or their doctor could log on and figure out what knowledge is out there, what the latest therapies are, what possible trials exist, and what the results along the range of therapies have been, not just for patients in general but for patients as similar to our new diagnosee as possible.

Pharma's Lack of Connection to Clinical Care

Alicia Staley is Medidata's senior director of patient engagement and a three-time cancer survivor in her thirtieth year of survivorship. Staley was diagnosed with Hodgkin's lymphoma as a young adult, and then breast cancer in both 2004 and 2008. She agrees with T. J. Sharpe's take on the industry, and believes that pharma's lack of connection to clinical care is a primary source of the difficulty.⁷ “For too long,” Staley says, “pharma has operated in one lane and clinical care in another—with no crossover between the two.” This is confusing for patients, Staley argues, and holds back the ability of researchers to recruit patients into trials and then keep them in the clinical research system. It is all transactional—there is no long-term relationship building, not with patients, not with patient advocates, and not with the health care industry at large.

In fact, it's too often the case that even doctors don't really know how clinical trials work. It's simply not relevant to most of their professional lives and what they do day to day. This ought to change if we want true industry collaboration. The current lack of such collaboration plays out at every level, from education to treatment. Staley believes the life sciences industry needs to realize that patient education and the creation of long-term relationships with advocacy groups helps them—not just when it comes to selling them the next blockbuster drug, but for trial recruitment and data collection, and to truly get us on the road toward richer disease models. Just like Sharpe argues, Staley insists the data is far too siloed and that there needs to be cooperation. “The value isn't in data blocking or data hoarding,” says Staley, “because data that isn't being analyzed, shared, and used productively has no value. Data without action is valueless.”

And yet, Staley sees all kinds of things happening in the world of clinical trials that don't serve patients at all, particularly doctors refusing to share trial opportunities with patients just because they're based at another health system. Clinical trials are operated around physicians who are chosen to be the “investigators” in each individual study. If a physician thinks that a study might help a particular patient, but it's not a study that physician is an investigator in, the process will include a referral to a study investigator. That means the patient's treatment will be with another doctor, possibly at a different facility, and perhaps in a different health care system. The patient may benefit, but the doctor and hospital may lose a paying customer. This can unfortunately create disincentives for referrals out of the system. “You think it's about the loss of a potential revenue stream, but it can actually turn out to be the loss of a life,” Staley laments.

Unlike Sharpe, who sees the most promise in a technological solution to educate patients—the perfect trials database—Staley worries that an overreliance on technology can leave out lots of patients, at least right now, who aren't plugged in. “The industry needs to meet the patient where they are,” she says. At the same time, she sees a huge place for technology in collecting richer sets of patient data—as long as someone is listening to the signal. To instrument patients, “so that they can walk through life throwing off valuable data nuggets,” Staley says, “makes it easier for patients to live their lives, to avoid being constrained by constantly visiting doctors and wasting time.”

There's a mindset shift that needs to happen throughout the industry, Staley says, and it's the same kind of shift I talked about earlier, where patient equations can move us from a reactive system to a proactive one. Can we build systems that aren't just addressing things that have already happened, but can in fact educate patients even before there's a diagnosis? Can we instrument people to catch things earlier, when there are more treatment options, or to tell that a patient is becoming refractory to their treatment before it's too late?

“Pharma is still a mature industry that can make a lot of money without changing much,” Staley fears, “but they have to realize that by collaborating with patients, with advocates, with the health care industry, and with each other, they can really make a bigger difference in people's lives.”

Truly Patient-centric Trials

The frustrations expressed by both Sharpe and Staley—about the difficulties of finding trials and navigating the problem of access—stem from a particular cause: trials are designed around investigators, not around patients. This is not an accusation against those who design, sponsor, and (like me) help to run them. It is a practical limitation, imposed by important scientific and regulatory requirements. Trials are, by definition, experiments. They need to be controlled—by which I'm referring not just to the usual controls where some patients receive a new drug and others receive some standard of care or placebo treatment, but also controlled in terms of consistency.

If some aspect of care is inconsistently handled for different patients in a study, it can create a confounding effect as far as determining whether or not a particular treatment is beneficial. In any good scientific endeavor, it is always easier—and by definition more reliable—to keep as many variables as consistent as possible. The worst case is an inconsistent variable that isn't part of the ultimate analysis. It's just creating the potential for inaccurate results. Keeping a trial limited to a particular set of investigators, through whom you know that the protocol will be executed consistently (meaning that the treatments and tests that define how the therapies will be evaluated for safety, efficacy, and value to the patient will be reliably performed and recorded) is a way to meet our ethical obligation to create a valuable result.

There are also regulatory requirements, sensibly and responsibly imposed by organizations like the FDA, to ensure that the protocol is being followed consistently. Experimental medications can often lead to unknown and dangerous side effects. The data sets that are produced in studies are not only the summation of the outcomes for the volunteer subjects, but they are also the information regulators use to make assessments for approvals.

Imagine a typical phase III trial, comparing a new medicine to the current standard of care, hoping for results that lead to drug approval. Not only are lives at stake for the hundreds of patients in the study, but also for the thousands, tens of thousands, and tens of millions more future patients for whom that drug may or may not be available down the road. Care is absolutely required to ensure the most reliable result, and that care is achieved through the current investigator-centric paradigm.

But that doesn't mean it's the only possible approach. The technologies and connectivity that are enabling so many of the therapeutic breakthroughs and new measurements in today's world might also be able to change the way we structure trials. Perhaps we can flip the current approach on its head and make studies truly patient-centric. Efforts like these are happening as we speak.

Anthony Costello is senior vice president, Mobile Health at Medidata and leads our involvement in the ADAPTABLE trial (Aspirin Dosing: A Patient-Centric Trial Assessing Benefits and Long-Term Effectiveness), a real-world trial that brings the study to the patient instead of the other way around, sponsored by the Patient-Centered Outcomes Research Institute (PCORI). In an interview, Costello talked about the study's transformative approach to recruitment—patients who receive their care at a PCORnet site (The National Patient-Centered Clinical Research Network, covering more than 68 million patients nationwide⁸) are identified through their electronic medical records, and sent an invitation with a code (a “golden ticket”) that allows them to log onto the study website and sign up to participate.⁹

This is the beginning of how the study turns on its head the entire idea of how we recruit for trials. Instead of finding investigators who then try to find participants, here the participants are effectively the ones who enroll themselves—and in this case, thanks to the power of a huge network of physicians, they are prescreened in advance to make sure they meet the study criteria, and provided with everything needed for them to enroll and participate. A virtual site is created around them.

ADAPTABLE, and other virtual trials like it, prove that you don't need a site to be the center of recruiting and treatment. These kinds of studies will continue to proliferate and become more of a normal practice in the future—and make everyone in the system better off, thanks to technology.

The patient's burden is lowered, certainly—they don't have to travel to clinics as often, and the study is easier to maintain as part of their daily lives. Anecdotally, ADAPTABLE has surprisingly high engagement rates, with patients compliant with the needs of the study and fewer of them dropping out, a huge problem in many studies. Costello thinks that the reason is at least partly because of how easy it is for patients to participate under this virtual model.

The companies conducting the research (or, in the case of PCORI and ADAPTABLE, the nonprofit organization conducting it) also benefit. Costs for recruiting go down, and time to recruit goes down as well. ADAPTABLE has 15,000 patients participating, a number that would be orders of magnitude harder to achieve in a traditional trial.

The outputs of the study are better as well. It's easier to get the kind of continuous monitoring previously discussed.

But this isn't an all-or-nothing option for conducting studies. Instead of patients having to travel to a physician's office for every data point to be collected in a particular study, they can be given the option to go to a pharmacy with a mini-clinic, or to a local laboratory for a blood draw. Drug supplies can be shipped to their homes. Perhaps they will go to that physician's office for an initial screening, for key check-in points, and to close out their treatment. But any chance to move even part of a study to the virtual environment—and relieve participant burden—should be taken advantage of.

Within the next five years, I expect we'll see almost every clinical trial taking advantage of virtual trial designs. Some trials will be entirely virtual—like ADAPTABLE—but more likely we'll see a bimodal distribution (two prominent peaks, if we were to plot it on a graph) where most trials are either 20% virtual or 80% virtual. The former category will largely be in cases where patients are critically ill, or the therapies are complex to administer. We'll need them to be in the clinic more, and they may even want to be there, but there will be quality-of-life adjustments by using the virtual space to make certain aspects easier.

The 80% peak will be for chronic conditions or for easily administered medications and evaluations of progression. These studies will contain elements that can be performed at home, at pharmacies, or with nurse practitioners visiting people at home, in addition to—as mentioned—perhaps a visit or two to a clinic to screen the patient, enroll and train them on whatever they need to know for the study, and then to close out their participation at the end of their course of therapy.

These kinds of virtual trials are coming at a time when the life sciences industry needs to evolve more than ever. With increasingly precise medicines, the math tells us that the number of patients who will benefit from each medicine goes down. Finding the right patient for a breakthrough precision medicine is harder than finding the right patient for a medication designed to be more broadly administered. Thus, finding the right candidates (able, willing, and appropriate) for research projects becomes an even harder problem to solve.

Accepting New Kinds of Data

The second big piece of the trial discussion is that life sciences companies need to continue moving in the direction of broader data capture, from wearables and mobile devices to genetic sequencing and the retaining of biospecimens. Richer, broader data in trials means better analysis. The more variables there are, the more likely we can find the meaningful ones.

Tarek Sherif, my co-founder and co-CEO at Medidata, talked to Pharma Times about this very issue back in 2016. “Historically in clinical trials,” he said, “we have collected more or less subjective data through diaries, paper…or by getting patients to come in to the clinic and do a test. These supposedly measure the efficacy of a treatment, but you are taking a snapshot in time.”¹⁰

Indeed, those snapshots in time don't come close to the kind of objective data we can now get with better and more advanced patient instrumentation. We can now come closer than ever before to analyzing real-world experience than merely a test result in a clinic. We can see what a patient's mobility is like, changes in step count during a trial, sleep data, and more. And while back in 2016 companies were starting to do electronic clinical trials, few were really committing to the idea of wearable trials.

Things have improved since 2016, but not by enough. And even as we see more and more wearables being incorporated—from Fitbits to Apple Watches—we still don't see sensors in trials across the board. Genetic panels are regularly being utilized in oncology and other therapeutic areas, but we don't see full gene sequencing across all studies, or the kinds of proteomics that proved so powerful for David Fajgenbaum and the Castleman Disease Collaborative Network.

Kara Dennis, Medidata's former managing director of mobile health—and one of the smartest thinkers I know about new technology in clinical trials—spoke to me about her take on these developments early on in the process of conceiving this book. “It will take some time for pharma to move away from the well-validated, well-proven measures that they've used for lots of patients over many years, but we are absolutely seeing the early steps, the process of validating the quality and usefulness of wearable data in studies,” Dennis told me.¹¹

The biggest challenges with digital data, she explains, are the quality of the sensors themselves, and whether subjects can use them properly. “Even with something as simple as a thermometer, the subjects may not be good enough at using it themselves, and there may be a difference between a clinician taking these measurements and subjects doing it on their own.” The other problem is compliance. “What kind of infrastructure do we need?” Dennis asks. “Will patients remember to use the device? Will they leave it on when they're supposed to, charge it, wear it at night if they're supposed to, or in the shower?”

As these issues recede into the background—as wearables are more and more accurate, and function with less and less potential for user error (implantables, etc.)—the hope is that pharma will become more comfortable using them. An industry analyst at Gartner has said, “Seismic shifts in this market will not happen until the pharmaceutical lobby has confidence in the underlying systems supporting wearables, and that means that clinical validation expertise for wearables must improve.”¹²

But the digital clinical trial is, fortunately, becoming more of a reality as time passes and knowledge and comfort grow, making trials more accurate, more efficient, and more patient-friendly than ever before. We can use technology to remove physical barriers, geographic barriers, and temporal barriers that all make launching and completing a study more challenging and more expensive. Between video calling to connect patients, doctors, and researchers and the landscape of wearables, patients can be full trial participants without leaving their homes, and researchers can still get complete and accurate information, images, and data.

It's one thing—albeit an important thing, without a doubt—to move trials into the twenty-first century by accepting new technologies and data collection tools. It's an even bigger step to open up trial design itself, take the shackles off traditional mathematical design, and move into new statistical techniques, new ways to compare the safety, efficacy, and value of a therapeutic, and new paradigms through which we can speed up how quickly something can move from the laboratory setting into the market, helping patients far more quickly than ever before.

Unshackling the Clinical Trial

In the life sciences industry, since the days of Lind and his experiments with sailors and scurvy, we are used to having a two-to-one ratio of patients to evidence. We need one patient treated with one medication and one patient treated with another—two people—in order to make a comparison. One patient gets a traditional course of chemotherapy for their cancer, while the other gets an immunotherapy. Or, one sailor gets lime juice to drink, and the other one seawater.

This need is changing. Here in our state of data-driven disease models, as we look for the equations that define the lines between what to treat and what not to treat (or between who to treat with an existing on-market medication and who will be the best candidate for an experimental therapy), we can start to break that two-to-one paradigm and create a more steam table-like view.

With better instrumentation and richer patient data, we can begin to look at measures of safety, efficacy, and value in new ways. We will have to, in order to achieve a future state of precision medicine. If you think of the number of patients in a study—the total number—as the denominator in a fraction (where the numerator is how many of those patients benefit from the treatment), as the treatments get more and more targeted, we will have a harder and harder time finding enough of them to reach statistically-reliable conclusions. We need to get more units of evidence from each patient whose data gets incorporated into research in order to make the research possible in the precise world of tomorrow.

The phrase “digital transformation” is often thrown around by pharmaceutical executives, as they—correctly and with good intent—realize that the infrastructure and processes they use for research and development are begging for modernization. But rethinking trial design (and breaking the two-to-one patient-to-evidence ratio) goes a step further. It's a critical step when we think about the ultimate goal of building disease graphs that can truly empower better prediction and decision-making. To get our treat/don't-treat lines to be as crisp and precise as possible, we need lots more evidence than we are currently generating, lots more data from our trial patients.

It is so easy now to dive in deeper than we used to be able to—to get higher-resolution measurements from sensors, to parse through patient histories, or to use artificial intelligence to find connections that we couldn't identify on our own. We don't have to miss episodes in episodic disease, because we can now gather data in real time, 24/7. We don't just need to draw the binary conclusion of whether, say, lime juice is the right treatment for scurvy. We can go further and try to figure out how much lime juice is the right amount, and whether that changes if you're a man or a woman, a child or an adult, or if you have any number of comorbid conditions. We need this increased data to be able to say with confidence whether to treat a high PSA result or not, whether Keytruda will be better for you than conventional chemotherapy, and whether you are going to have clinical signs of Alzheimer's disease while it still matters, or not until you're projected to be 180 years old. The digital infrastructure makes this possible like never before.

Enter Thomas Bayes

Thomas Bayes was a statistician in the 1700s whose work ultimately led to a split in the world between two schools of statistical methodology: the frequentist and the Bayesian. Put simply, a frequentist approach to determining the chance that a coin toss will result in either heads or tails requires us to decide first on a number of times that we will toss the coin, measure the outcome, and then, finally, calculate our conclusions. A Bayesian approach, alternatively, allows for adjustment on the fly. Our predictions don't need to wait for the full set of data. We can modify our expectations and our hypotheses as we see more and more evidence.

With coin tosses, each toss is a trivial amount of effort—assuming we already have the coin—so deciding to toss a coin 100 times in order to figure out how many heads to expect in the future is a reasonably trivial proposition. But when it comes to patients—real people who are looking to extend their lives or increase the quality of them—it's not trivial at all. One hundred trial subjects—just to form an initial understanding of whether and for whom a treatment works—is a lot of people exposed to something that may not help them.

Using Thomas Bayes' statistical techniques, we can do better. We can expose as few patients as possible to a treatment that won't work and instead give it to the maximum number of people for whom it will. We can get our therapies through the research process more quickly, to make them more generally available. We can learn something about the nature of a coin toss every time we perform one—which means fewer coin tosses are needed to draw a conclusion. In other words, we can break the two-to-one patient-to-evidence ratio requirement.

Don Berry, a professor at the University of Texas M.D. Anderson Cancer Center and the founding chair of its department of biostatistics, is the designer of I-SPY 2, a breast cancer study that marks the largest and arguably most successful use of Bayesian statistics in clinical trials to date. Berry's work on bringing Bayesian statistics into medical science has been pioneering, and it links directly to the ideas we just talked about in the previous section. When you talk to Berry, you realize how applicable Bayesian thinking is to bringing precision medicine to research.¹³

Instead of taking the non-Bayesian frequentist approach—where we need all of a study's data in order to even make an initial estimate of therapeutic value—the Bayesian approach lets us use a probability distribution for that value, based on past knowledge, and then new data can be used to update that probability distribution as the study goes along. Simply put, the probability distribution acts as a function—an equation—that plots the expected outcome of the experiment, and whether a treatment will be effective for a patient.

Thus, rather than starting with an assumption, with no idea if that starting assumption is correct, or how to adjust it along the way if it's not, we can keep learning as a study proceeds. We can't predict perfectly, but we can create better and better estimates based on what we already know about the world, about patients, and about how they respond. We can keep updating predictions, using today's data to figure out with greater likelihood where we will be tomorrow. And, ultimately, we can move patients around during a trial in order to maximize their outcomes, and maximize what we can learn from the trial, without sacrificing the objectivity and statistical value.

Put simply, we learn as we go, explains Berry. And if data from other trials helps us make better inferences about our current one, then we can and should use it to the extent that it is statistically valuable to do so. The I-SPY 2 trial is aimed at finding the best treatments for early breast cancer in high-risk patients, where their cancer has not yet become metastatic disease. What are the best therapies for treating this disease effectively? Figure 11.1 is a graphical representation of the kind of trial design pioneered by studies like I-SPY 2.

If a therapy demonstrates poor results for a particular subtype of patients in the trial, patients with that subtype get a lower and lower probability of being assigned to that therapy, all the way down to a zero probability if the treatment proves to most likely have no value for such a patient. That is something you can't do in a standard two-arm trial: if a therapy isn't working, the trial is over, and you have failed. But in a multi-arm adaptive trial like I-SPY 2, there are multiple experimental therapy arms (as well as a standard-of-care control arm) and a set of genetic tests that are used to establish which therapies show the best outcomes for patients with particular genetic profiles.

As patients enroll in the study, and new data comes in, more knowledge about which biomarkers are associated with positive or negative outcomes is fed back into the assignment of subsequent patients to particular therapy arms. Instead of performing the statistical equivalent of tossing a coin to decide whether a patient gets the drug on study arm A or B, randomizing them to one of two arms of a study, when a new patient is enrolled in this kind of Bayesian trial, their biomarkers are used as a way to preferentially randomize them to therapies that have already been successful for patients like them.

This should sound familiar. Berry's approach is very much like building a steam table and defining the phase transitions that define what will more likely be a successful treatment for any given individual. The dimensions here are the biomarkers, not temperature and pressure. And although we can't—practically or ethically—test every therapy on every combination of biomarkers (which would be the experimental approach to generating a steam table), we can use Bayesian statistics to start with some assumptions about what those phase transitions look like (the initial probability distribution), and with each patient treated we can refine that function. In other words, we can refine the patient equation represented in the study as we go along.

The phase transition visualization of the way I-SPY 2 works is mine, not Berry's, but the thinking that led to this book truly starts with Don Berry's advocacy and mentorship for this kind of trial design. Even without an appreciation for the mathematical advantages, simply realizing that adaptive designs result in more patients being exposed to therapies that are beneficial to them should be enough to see their advantages. In Berry's words, “You learn and confirm. And you see if your predictions can be reproduced.”

Breaking the Barrier

The I-SPY 2 trial doesn't have a fixed set of therapies. There are 19 therapies that have been incorporated so far. Six have “graduated” as of this writing, and more will. Once there is enough data to confirm that a particular drug works well for patients with certain biomarker profiles, that drug is removed from the trial. The company that makes it can use the data generated for regulatory approvals, having gained further ground as far as the precision application of their therapy than would be possible in a traditional phase I–phase II–phase III clinical development program. So not only are there benefits to patients in the trial in terms of having likelier access to better therapies, but those therapies can be brought to market faster to patients waiting for them everywhere.

In the United States, the FDA has been supportive of Bayesian trial designs^14,15—but there are hurdles because of the inherently (and in many ways responsibly) conservative culture of the pharmaceutical industry, as well as industry incentive structures that are designed to support the machinery of running studies in traditional designs. Plus, there are very practical limitations to this kind of study design. It requires significant coordination and collaboration. Recalling the protocol that describes every aspect of the therapeutic administration, there is the need for a “master protocol” that governs the overall design and operation of the trial. All of the participating companies, and all the organizations with drugs being evaluated in the adaptive design, need to work within that master protocol's framework. The requirements for regulatory and scientific rigor are no different than in a traditional study design, only there are more experimental therapies, more patients, and the trial runs over an even longer time frame. This adds significant complexity. Although it is very difficult to argue against the ethical and financial benefits of adaptive, Bayesian-style trial designs like I-SPY 2, the costs of the individual studies and the complexity of coordinating them remain barriers.

These barriers, however, can, should, and will be overcome, because there is huge benefit. These types of studies break the two-patients-per-unit-of-evidence limitation we've operated with for centuries. In the simplest way, sharing a single control arm across multiple study arms with a range of new drugs means that we're able to reuse those control patients. With seven experimental arms, we effectively have 1.125 patients required per unit of evidence. And that is just the simplest view of a study like I-SPY 2. Arguably, due to the learning nature of the study and the virtuous cycle of its design, we are creating even more evidence per patient.

Ultimately, that increase in evidence-generating power will result in more and more trials like this. GBM AGILE (Glioblastoma Adaptive Global Innovative Learning Environment) is an ambitious new adaptive trial design conceived in 2015 to help speed knowledge about treatments for glioblastoma, an aggressive form of brain cancer responsible for the deaths of Senators Ted Kennedy and John McCain, among many others.¹⁶

Like I-SPY 2, the GBM AGILE trial is designed to evaluate many therapies at once, with just one control group—meaning patients are more likely to get an experimental therapy, and, even more critically, more likely to get an experimental therapy that is right for them. Columbia University has been among the first institutions to enroll patients in the trial. Their news release explains, “Throughout the trial, tumor tissue from participants will undergo analyses to identify biomarkers that may be associated with a patient's response. As the trial accumulates data, its algorithm refines the randomization process, so that patients have a better chance of getting a treatment that appears to show benefit.”¹⁷

Columbia's Dr. Andrew Lassman, chief of neuro-oncology, says, “This trial design offers a way to lower the cost, time, and number of patients needed to test new therapies for newly diagnosed or recurrent glioblastoma.”¹⁸ Given the poor prognosis for patients diagnosed with glioblastoma and the lack of effective treatment options, the need for programs like GBM-AGILE is clear.

Having established the value of collaborative, adaptive designs like I-SPY 2 and GBM-AGILE, we now have to ask whether there are other barriers that can be broken down, or other ways for life science companies to collaborate that can go beyond even these studies in terms of generating evidence, getting safe and effective precision therapies to market more quickly, and creating breakthrough value for patients. I believe there are, and the discussion continues with the idea of synthetic control arms.

Synthetic Control Arms

The idea of a synthetic control was first described in a paper in 1976 by Stuart J. Pocock in the Journal of Chronic Diseases. “In many clinical trials the objective is to compare a new treatment with a standard control treatment, the design being to randomize equal numbers of patients onto the two treatments. However, there often exist acceptable historical data on the control treatment,” states the article's abstract.¹⁹

The idea is that we can synthesize historical patient data into a hypothetical control group that will function just as well as a randomized control group. As long as these two hypothetical sets of patients are equivalent based on the definition of the study—that is, if they share the same characteristics and meet the right inclusion/exclusion criteria—they should function just the same. The idea isn't dissimilar to sharing patients across the arms of a multi-arm Bayesian design like the I-SPY 2 trial. We have a protocol being rigorously followed by the investigators, and patients who all meet the criteria for being part of the study—should we not be able to reuse the data?

If we are looking for, say, a group of patients with heart disease who will be given a particular study drug, we already have data from many other trials with heart disease patients who have been given the standard-of-care treatment as part of a control group. Why not reuse their results? Or, at the very least, why not reuse the data generated in those previous studies to supplement the new data we're obtaining in the trial? The word “synthetic” can be confusing here. It's not that the control patients are somehow synthesized. These are very real participants in clinical trials, just not the clinical trial currently being performed. The synthesis is of their experiences as control subjects into a new control cohort, a new arm synthesized from data deriving from other rigorous and scientific clinical trials.

Mathematically, Pocock made the case in his paper that we can—and, from the perspectives of cost, time, and ethics, we should—approach control groups in this way. Think about those two patients, one experimental and one control—tens, hundreds, or thousands of times over—whose data is finally assembled into a data set that shows the difference in outcomes for those who are getting the experimental therapy and those who are not. If we can reuse control patients from previous studies, we should be able to save half the cost. We only need to enroll patients who are to get the experimental therapy, because the control group is already taken care of.

Time becomes an issue here as well. Assuming there are no shortcuts in how long it takes to evaluate whether or not a particular therapy works, the idea of saving time with synthetic controls may seem unintuitive to those who don't work in clinical research. If it takes 12 months to go through the course of therapy and see if, for instance, a tumor stops growing, we're not going to save any time by reusing data from other trials. However: the number of patients we have to find goes down—in the most extreme example, by half.

The time it takes to recruit patients into a study is typically one of the key rate-limiting parts of the process. Assume we need to enroll 120 patients in a theoretical study, and are able to find 10 participants a month. (Those numbers are reasonable for many studies. In some cases, it can be even harder to recruit, and the trend as we move toward more precise therapies—as we've already discussed—is going to be in that direction.)

Between the time we've enrolled the first and last patients into the study is a full year. Then add a year for the last patient recruited to get through the complete course of therapy, and then the time it takes to evaluate the effectiveness and safety. If we could reduce the number of patients we need to enroll by half, six months in this case gets shaved off the timeline. A regulatory submission could happen six months earlier. Six additional months of patients diagnosed with the condition the drug addresses could have broad access to the new treatment.

Even if the treatment doesn't work better than the standard of care, we've avoided asking 60 additional patients to take the chance to get randomized to a therapy that isn't going to help them, and give up the opportunity to be enrolled in studies that could serve them better. Particularly in therapeutic areas where there are no effective therapies currently on the market, we've created more chances for success. We've minimized the number of patients exposed to something harmful, and maximized—during and beyond the study—the number of patients who can benefit from more effective medications.

So why—with what appears to be a huge set of advantages—isn't this done more? It's certainly on the radar screen of many. Julian Jenkins—the former GlaxoSmithKline executive who worked on Flumoji—says that pharma is absolutely looking at this. “If I know the old drug worked against a particular target,” he says, “then, in looking at whether a new drug is going to work, many companies are trying to use secondary analysis, looking back to test hypotheses, to validate targets. It speeds things up, and is a huge enabler for the industry.”²⁰

And yet, the standard of care—the gold standard—for clinical evidence is still the randomized, prospectively-controlled trial. Why? The answer is partly because of the conservatism of the life sciences industry (always important to point out that this is a good thing that protects us all), and partly because creating synthetic control arms with previous clinical trial data isn't an easy thing to do. Most trials—virtually all trials, unless they are part of a master protocol like GBM-AGILE or I-SPY 2—have their own unique combination and cadence of visits to the clinic, lab tests, imaging, and so on. The way that studies are designed and run—one at a time, on one particular drug—means that every study's data set has its own unique design and particularities. It's not just a matter of taking data from studies, pooling it, and using it again. We need to ensure that the data is high-quality, standardized properly, and consistent. That consistency across trials isn't always possible to find right now.

Even once the data is standardized, the hard work isn't over. Although we try to eliminate biases from clinical trials, there can still be inherent issues. Take the simple example of age. A clinical trial (and, again, this is a very reasonable example) may have inclusion criteria that patients must be over 18 and under 65. With a standardized data set, it is simple enough to find patients who meet those criteria. But what is the distribution of age across the synthetic controls? Do we use a normal distribution, centered around age 42? Perhaps our synthetic control group skews younger, and perhaps the distribution of ages among the prospectively enrolled patients skews older. Should this matter? Might this create a confounding element in our synthetically controlled study that makes analysis harder, not easier?

The answer in this case is that we don't know, and therefore we must do anything we can to make the characteristics of the patients—at least all of the characteristics that we know about—as consistent as possible with the prospectively enrolled patients. We'd like the inputs of the synthetic controls to look as much like the inputs of prospectively enrolled patients in the study as possible.

The work doesn't end there. The outputs of previous studies need to emerge in a way that is matched to the new study being performed. Raw data collected in the previous trials needs to be recleaned and the results recalculated. And before you even get to all that standardization and cleaning, you need to actually find all of the raw data from those previous studies. This data may be in deeply-buried folders in servers across dozens of pharmaceutical companies. This data may not be well-cataloged or indexed in a way that makes finding all of it possible.

Our Synthetic Control Model

The effort needed to successfully reuse clinical trial data is why, over the four-plus decades since Pocock wrote his article in 1976, the idea of synthetic controls hasn't been scaled. This is where my company, Medidata, comes into the story. Barbara Elashoff, who broke her leg back in Chapter 2, along with her husband Michael Elashoff and their former colleague at the FDA, Ruthanna Davi, all reassembled at Medidata and began to work on the idea of synthetic controls with an advantage that nobody previously had: a platform where clinical trials had been run for over a decade with consistent data definitions, all the data in one place—in the cloud, where the studies themselves were run—and an organization placed centrally and able to ask more than one thousand life sciences companies if they would like to pool their control data for this greater good. Not to mention, we also had a commercially-sustainable business model, where the costs of standardizing data and managing the complexities of synthetic controls would be an incentive, not a disincentive, to make them work at scale.

Every patient who is in the volunteer pool—with consent and permission from the patients and from the companies involved—is treated as equal: as members of one giant clinical trial data set. For any given indication, relevant patients are selected and those exposed to experimental therapies are excluded. Matching algorithms—ways to look at each individual as a matrix of data—are used to ensure that the types of skews present in different variables won't impact downstream analyses, and, ultimately, from a data set synthesized across what is effectively the breadth of the life sciences industry, a synthetic control arm emerges.

It is still, to be sure, not a trivial process. This is another example where the execution of analyses is fast, but there are painstaking months—sometimes years—of necessary preparation and the testing of techniques. However, along with some of the life sciences companies that participate in the project with us, we've paved the way for these data sets to be used—both for planning purposes and as a supplemental data source to benchmark trial results. Soon—perhaps by the time you are reading this book—synthetic controls will be used as part of the statistical package submitted to regulators in the approval process for a new medication.

The inherent, incredible, game-changing value of being able to reuse patient data to create more evidence will probably offer further surprises—in a good way—for those creating therapeutic value for patients, and for the patients waiting to receive it. For a pharmaceutical or biotechnology executive, even if a synthetic control isn't part of the data they submit to a regulator, or used to prove the value of a new therapeutic to a payer or a provider, imagine the value of knowing that your control data “looks” like control data in other studies.

The need to eliminate biases in clinical trials should be clear. A drug that appears to be safe and effective in research, but has had biases introduced into that research, can result in bringing a drug to market with elevated expectations. Whether it's because of the criteria by which patients were selected, the geographies where investigators were chosen, or something else about the investigators or institutions, having elements that skew not the inputs but the outputs of the study—the endpoints related to survival or quality of life—can have tremendous consequences. Comparing a control arm to a synthetic control and seeing that the standard of care or placebo arm in a randomized, controlled trial matches your results will mean more confidence that you have guarded against that risk.

That's just the starting point. The value of generating high-quality evidence faster and more efficiently has been discussed. Once therapies have been approved using synthetic controls as a replacement (or at least as a supplement) for prospectively enrolled control subjects, I expect the life sciences industry to embrace the idea at scale. The current exception—the occasionally-presented idea at scientific meetings—will become the new standard by which therapeutics go from theoretical value in a laboratory to therapies available to the public.

There is precedent for this. There are multiple kinds of synthetic controls, and the ones discussed up to this point—the reuse of data coming from the rigorous scientific and regulatory environment of a trial, albeit a different trial than the one at hand—represent what should become the gold standard moving forward. There is also data from the world of health care outside of clinical development. If we can look at, standardize, and benchmark the data from clinical trials against these other sets of data, the same progression of value should be possible. Using this information to plan studies, estimate the value of therapeutics, mitigate biases that could be introduced in prospective controls, and ultimately supplement or replace the controls necessary for a trial are all steps along the way. Companies like Flatiron Health, for example, which is using real-world data to accelerate cancer research, are proving out this idea—scientifically and as a business model—every day.²¹

Making Every Trial an Adaptive Trial

There is an exciting progression beyond this idea of synthetic controls. If you think about the process for creating synthetic controls as described above, a key step is eliminating the subjects previously exposed to experimental therapies. This makes sense, since we want to compare a new drug to the on-market standard of care. But: imagine a trial where the new drug is better than the standard of care. It should, therefore, be on a track—or at least be a possible candidate—to become the next standard of care. So there can become a virtuous cycle of clinical trial data and synthetic controls. The old experimental cohort is the new control, if the tested therapy turns out to be the new standard. We end up with a self-perpetuating data asset that benefits both patients and the life sciences industry.

Now consider the Bayesian adaptive design, and the advantages already discussed of a trial where we aren't just testing the fitness of a particular drug for a broadly defined set of patients, but creating a learning environment where biomarkers are continually leveraged to pair the best therapy with each patient enrolled. Recall the complexity with administering a process like that, and with creating a master protocol that governs how the trial, across all therapeutics, is run.

How is this Bayesian adaptive environment different from the one we are creating with synthetic controls? My answer is that they aren't different at all. By combining these concepts, the life sciences industry can create a collaborative research environment that involves not just the reuse of controls, but also allows for the continuous learning necessary for the precise pairing of every available drug (on-market and experimental) to every waiting patient in a close approximation—almost a perfect implementation, save the impractical, unethical idea of testing patients and therapies like temperature and pressure in a lab—to the steam table-backed phase diagram vision of the future.

This future is one that colleagues and collaborators both inside and outside of Medidata and I hope to make a reality, ultimately creating a super-sized virtuous cycle, as represented in Figure 11.2.

Patients in previously-run clinical trials for what is the current standard of care become a continuously-refreshed control arm. Standardized data across not just a handful, but tens or hundreds of potential new drugs (or combinations of therapies) can be compared by using the same data cleaning, standardization, and benchmarking techniques already used today for synthetic controls.

Instead of a competitive environment where investigators, or the patients themselves, are recruited into disparate clinical trials, the industry can work together to create a bias—in this case a positive one—where a Bayesian adaptive approach is used to enroll every patient who could benefit from an experimental therapy to the best possible one for them, based on the industry-wide knowledge at the time.

The life sciences industry's commercial architecture is based on the value generated by drugs and devices. The cost savings generated by running more efficient clinical trials, and the revenue opportunity of getting a drug to market faster, are of course hugely valuable to the companies themselves and to patients. But a truly collaborative environment where we preferentially randomize patients to the best available known therapy of any type—for them, at that point in time—creates an unprecedented number of units of evidence per patient in research programs.

A great litmus test for conviction in medicine is whether—just like Dr. David Fajgenbaum—someone is willing to take their own medicine. A collaborative environment where volunteers' biomarkers are measured, where the past experiences of patients like them is available across therapies, and where they can be randomized to an experimental therapy regardless of what company is running a trial (effectively creating an adaptive environment for every drug being tested in it) is the world in which I want to be a patient.

A Stroke of Insight

I was asked not long ago to present ideas about the future of research at a conference on stroke, run by the American Heart Association. The extent of my academic cardiology knowledge is unfortunately limited to an afternoon crash course in interpreting ECGs, so I thought it was best to admit as quickly as possible in my talk that I was thoroughly unqualified to offer any opinions on how cardiology research specifically could or would change in the future.

However, I then explained to them what I would say to a group of oncologists: If you are looking to build a mathematical model for early diagnosis in oncology—asking what biomarkers can be measured as early indicators that a patient will be diagnosed with cancer—and you are only looking at oncology research, you've made the problem much harder than necessary, if not impossible to solve.

All of the patients in the oncology data sets have already been diagnosed with cancer. On the other hand, virtually all research projects—whether academic or funded by industry—that enroll patients prospectively include a medical history, patients' vital signs, standard blood labs, an extensive list of prescription and nonprescription medication being taken, and a list of “adverse events”—severe, life-threatening ones like cardiac arrest, as well as less severe (but not necessarily less important) ones like headaches.

If we were to look at a cardiology study instead of one in the oncology space—which might have thousands or tens of thousands of patients in it—we should be able to see which patients manifested comorbidities (like a cancer diagnosis) based on adverse events, medications taken, having to be dropped out of the study, or even death. We have a run-up to that diagnosis or death that includes a set of medical data far more extensive, and more exquisitely curated, than we would find in any integrated health system's medical record, in our personal health records, or even combinations of the data sets that governments, academic institutions, and companies around the world are trying to create. (Not that these data sets from outside of research aren't worthy endeavors. They are.)

Figuring out how to get unexpected evidence from the data collected in a research project is the ultimate manifestation of the strategies presented here. Making cardiology research data a valuable synthetic asset for oncology studies—or the converse, seeding models for heart failure or stroke with data from oncology, diabetes, or other studies—will unlock and fuel the virtuous cycle of precision medicine research even more. We can find clues that add huge insights to our patient equations from the records of patients who just so happen to be diagnosed with new conditions while they are being closely monitored.

Adaptive trial designs offer huge potential as we go forward in a world rich with data, and where we need better ways to test hypotheses quickly and accurately. But we don't just need to rethink our approach at the front end of the clinical trial pipeline. We also need to reinvent our relationship with patients at the other end in order to get the right treatments out to the public and truly make an impact.

In the next chapter, we'll turn our attention to the patient-facing piece of the data revolution: disease-management platforms and around-the-pill apps that can motivate and change behavior, measure outcomes, and match patient to treatment in ways that we haven't ever been able to do before smartphones and wearables. Not everyone will be part of a clinical trial at some point in their patient journey, but everyone has the chance to be impacted by apps and other interactive programs that bring trial results to light, and get people to care about their health and take the right actions for their future.