You Can’t Find the Needle If You Don’t Know It Exists
This blog is adapted from my book titled “Inside the Cockpit: Navigating the Complexity of Drug Development with AI and Blockchain”. Full copies are available here.
In December 2014, a twenty-nine-year-old investor named Vivek Ramaswamy bought a patent from GlaxoSmithKline (GSK) for $5 million. The patent covered an Alzheimer’s drug candidate that had failed GSK’s clinical testing. This isn’t uncommon; in fact, most candidates fail (one study found that between 2006 and 2015 only 9.6 percent of drug development programs made it to market).1 GSK, seeking to recoup some of its investment in what appeared to be a dead-end project, was more than willing to sell its intellectual property. Ramaswamy knew something GSK didn’t. He was eager to buy.
Typically, drug testing involves a “stage gate” process characterized by multiple decision points, or stage gates, along the way. If a drug fails to meet performance requirements at any decision point (DP), development is stopped. Drugs like the one Ramaswamy bought, known as SB-742457, fill decision-point graves. GSK had conducted thirteen trials involving 1,250 patients before concluding at DP-02, or decision point two, that further study was unwarranted. In 2010, it abandoned SB-742457, evidently one more in a long line of failures that characterize the hunt for blockbuster drugs.
In early 2014, Ramaswamy, a hedge fund partner, had information that helped him understand something that the GSK researchers did not: SB-742457, taken with another Alzheimer’s drug, Aricept, slowed cognitive decline for a certain subset of dementia patients. In May 2014, Ramaswamy left his hedge fund. Within eight months, he raised $360 million to launch his new company, Axovant Sciences, built around SB-742457. When Axovant went public in mid-2015, even though the firm had done no further clinical studies, the new company’s market valuation topped $2 billion—based on a single $5 million purchase of a “failed” drug.2
GSK had sold Ramaswamy a treasure for almost nothing. He had taken advantage of the information asymmetry that’s pervasive in the life sciences market. Companies like GSK often don’t even know the value of what they have because they can’t see a complete picture; data is scattered across the life sciences ecosystem in hidden pockets and walled gardens. Making connections and drawing insights from that data, consequently, is very difficult.
In order to understand the value of the data in the drug development value chain, it’s crucial to understand the context, which is dauntingly complex. To obtain a view of the landscape in, say, Alzheimer’s research, one would have to manually curate and annotate all the relevant research data, then look at the relationships within it. For most researchers today, this is essentially impossible. Companies that do this work employ armies of analysts and sell the results for a substantial premium. Big pharmaceutical firms may be able to pay for their research, which gives them a leg up in the hunt for blockbuster drugs (though no guarantee they’ll be the only ones finding them, as Vivek Ramaswamy showed). The result is information asymmetry, and the value of that asymmetry is growing exponentially, given the growth in the volume of life sciences data.
In 1950, medical knowledge was believed to double every fifty years; by 2020, it is expected to double every seventy-three days.3 Even at a huge pharma like GSK, they were—and are—sitting on an unexploited treasure trove in terms of the value of the research and development work. But they didn’t know that in the case of SB-742457. They didn’t know that the drug could be repurposed and used for a certain subset of patients. In all likelihood, they are sitting on other treasures they don’t recognize because they are not able to glean the appropriate insights from the life sciences data landscape.
DRUG DEVELOPMENT IS BROKEN
Ramaswamy and GSK’s experience is one example of what’s broken in drug discovery and development. Data is hidden away in carefully protected silos, and that secrecy means that important discoveries are not happening at the pace at which they should.
This problem is not limited to big pharmaceutical companies; consider a story I heard from a research scientist at the University of Göttingen in Germany who works on the epigenetics of pancreatic cancer—one of the deadliest cancers in the world, a cancer with one of the lowest five-year survival rates, less than 5 percent.4 Many candidate drugs for treating pancreatic cancer have failed miserably. This scientist described a drug that was in clinical trials but had been abandoned because it failed to meet the safety and efficacy criteria of the US Food and Drug Administration (FDA). The reality was that this drug actually cured the tumor in a select subset of patients. If one were to look at the epigenetics of the drug’s efficacy and stratify the patient population appropriately, it could be a wonder drug for a smaller segment of patients. But it wasn’t being pursued.
Failed, or apparently failed, drug experiments go to the valley of death. During the period from 2013 to 2015, 218 drugs failed at Stage II or Stage III trials.5 The decision has been made by their creators that they are not worth pursuing, yet valuable data is locked up in those experiments. That data still is useful. Some researcher somewhere, if he had that data, might see a connection to something that otherwise seems unrelated, and see possibility, just as Vivek Ramaswamy did. Instead, the data about those failed experiments is not published, not searchable, and not available—no one can even find out that a researcher conducted an experiment. Researchers only want to publish what seems to work, yet a broader understanding of what doesn’t work can also be useful in the search for drugs.
Big Pharma companies might be holding hidden treasures, but they can’t get a real-time look at the entire gamut of drug candidates and the intellectual property (IP) of those candidates. They can’t combine their own research with outside research to come up with drugs that will help patients, which of course is why pharmaceutical companies should exist. The direct advantage of looking at the entire research universe, internal and external, is that researchers can see insights and correlations that they could not previously see when data was trapped in silos. The present system of siloing data and information asymmetry characterizes the life sciences ecosystem and hampers the discovery of potentially lifesaving drugs. The example of SB-742457 illustrates how broken our system of drug discovery and development actually is. Every day, people suffer and die because life sciences and pharmaceutical industries don’t have the technologies and capabilities they should and could have to bring effective drugs to market more quickly and effectively.
Imagine a physician sitting in the cockpit of a race car. What is flying against his windshield is not wind, but data—the whole life sciences data universe. If the physician wants to drive fast and safely and get where he intends to go, he drives by looking out the windshield. But in life sciences, we drive the car by looking in the rearview mirror—that is, we look at historical data. Worse, we don’t even look at recent historical data; we look at old historical data. It’s as if there’s a time delay for what we’re seeing in our car’s rear view mirror.
Because insights about that data are manually curated and delayed by the process of scientific publication, we are not able to see the data in real time. Peer review of publishable data takes up to 250 days, and the time lag from completion of research to publication can be a year.6 That’s why GSK sold Axovant a promising Alzheimer’s drug for a pittance; the firm’s researchers were not able to see what the extant, current research on Alzheimer’s looks like.
Ramaswamy and Axovant were able to get a glimpse through the windshield, and that glimpse was worth $2 billion.
FIXING WHAT IS BROKEN
To fix what’s broken about the drug discovery and development process will take a significant disruption in three areas. First, with the rapid rate that information is being added to the mountains of life sciences data, we need real-time insights. Second, the asymmetry in the amount of data that is available will need to be rectified so that data availability is democratized. And third, information about innovation needs to be widely available to avoid innovation redundancy. Scientists need to know what other scientists are working on.
The good news is that the solutions are here. They exist in the form of computers that run artificial intelligence technologies that are capable of crawling through the mountains of available data and retrieving only what is relevant. These computers can process the information in a small fraction of the time that it takes armies of analysts to do and at a much lower cost. The result is that information can be retrieved in real time. If data is reported in, say, Madrid one day, it will be retrievable by researchers in Los Angeles either later the same day or first thing the next morning, rather than a year and a half later.
And as for pharma companies holding on to their data and not sharing it, there now exists blockchain technology that makes data sharing a secure process. These companies need not worry that someone is going to steal their patents or their data, and all parties can benefit from this technology. This means that proprietary data from various companies that might have been locked away for good can instead be shared in real time.
The news that isn’t so good is that Pharma hasn’t been rushing to get on board. SOPs are difficult to change and change in the Pharma industry has never been fast. The bigger picture here, though, is that with increasing scrutiny from government and other organizations, pharmaceutical companies will have to make changes sooner rather than later. It is time for these companies to invest in artificial intelligence and blockchain, and allow these technologies to make positive, sweeping change for everyone in or affected by the pharmaceutical industry. And that’s all of us.
References: 1. David W. Thomas, Justin Burns, John Audette, Adam Carroll, Corey Dow-Hygelund, and Michael Hay, Clinical Development Success Rates 2006–2015 (Biotechnology Innovation Organization, Biomedtracker, and Amplion), 7,
2.“Did a 29-Year-Old Show GlaxoSmithKline That It Made a Billion Dollar Mistake?” PharmaCompass, June 25, 2015, https://www.pharmacompass.com/radio-compass-blog/did-a-29-year-old-show-glaxosmithkline-that-it-made-a-billion-dollar-mistake.
3. Peter Densen, “Challenges and Opportunities Facing Medical Education,” Transactions of the American Clinical and Climatological Association 122, (2011): 48–58, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3116346/
4. Boris W. Kuvshinoff and Mark P. Bryer, “Treatment of Respectable and Locally Advanced Pancreatic Cancer,” Cancer Control 7, no. 5 (2000): 428 & ff,https://journals.sagepub.com/doi/abs/10.1177/107327480000700505
5. Richard K. Harrison, “Phase II and Phase III Failures: 2013–2015,” Nature Reviews 15 (2016): 817–818, http://www.whartonwrds.com/wp-content/uploads/2017/11/Nature-Paper-Richard_ Harrison-Phase-II-Phase-III-Failure-Rates.pdf..
6.C.H.J. Hartgerink, “Publication Cycle: A Study of the Public Library of Science (PLOS),” Authorea, https://www.authorea.com/users/2013/articles/36067-publication-cycle-a-study-of-the-public-library-of-science-plos/_show_article.