All opinions are my own and do not necessarily reflect those of Novo Nordisk.
h/t to @Frank_S_David, @scientre, and the LinkedIn Group Big Ideas in Pharma Innovation and R&D Productivity for links and ideas
Part 1 is here.
Part 2 is here.
In the previous parts to this series I’ve covered both why the biopharma industry is ripe for disruption, and what the markets might be that could support a nascent, potentially disruptive technology until it matures enough to allow it to supplant the current dominant industry players. In this final part I’d like to ask what disruption would look like and provide some examples of directions and companies that exemplify what are, to my mind, these sorts of disruptive technologies and approaches. With, I might add, the complete and utter knowledge that I’m wrong about who and what specifically will be disruptive! But in any case, before we can identify disruption, it’s worthwhile to ask what are the key elements of biopharma drug development that serve as real bottlenecks to affecting human health, since these are the elements most likely to provide an avenue for disruption.
Bottlenecks, bottlenecks everywhere
One of the things I’ve learned in project management training is that everything comes down to the holy trinity of cost, time and quality. Every project has to find its balance within those three elements and any demand by the customer for change has to sacrifice at least one of those three qualities. This seems like a good place to start with the problem areas for drug development.
Time: drug development routinely takes ten to fifteen years. This timeframe means it’s extremely difficult to predict what the market, regulatory and payer environments, and social pressures will be when a drug eventually reaches the patient. Without accurate forecasting, drug development and drug pricing become largely a crapshoot of best guesses, sometimes with disastrous results (for the drug developer) as was seen last year with the rejection by Sloan-Kettering of the initial price structure for Zaltrap.
Cost: this goes hand in hand with time. Drug development currently costs a lot of money, and pricing models for new drugs try to amortize the past costs. However, the actual cost to develop a drug varies widely and in any case is very hard to capture accurately, given the uncertainties inherent in the process and the need to pay for failures. Also the general consensus seems to be that the cost of developing a drug is continuing to increase. So, companies need to keep charging more just to keep up, in a perverse Red Queen kind of scenario. It’s worth noting that this is one place the two customers I’ve described, the regulatory agencies and the payors, are at odds, at least some of the time. From what I can see of the FDA’s process, eventual drug price is not part of the consideration of whether to approve a new drug. But the payers are continuing to push for lower prices. It’s also worth noting that pricing is becoming more of an issue, both as mentioned above for Zaltrap, but also for the new Hepatitis C drug Solvadi from Gilead. Solvadi, with a price tag of $84,000 in the US, is creating a debate about what cost-effectiveness really means and what payers can afford to pay.
Quality: as mentioned in earlier parts of this series, the customer demand is for quality. Safer, more effective medications. Truly unmet need. And as I’ve suggested, the framework of the adaptive landscape suggests that getting that next step in efficacy is increasingly more difficult, assuming the landscape remains the same.
To that, let me add some additional bottlenecks for the industry.
Knowledge: we would like drug development to be like engineering. Identify a target, know that by modulating it we will have the desired effect, create a therapeutic (small molecule or protein) that will achieve that modulation. But unfortunately human health is not yet so straightforward. The term engineering implies a level of knowledge about the system. What are the parts? How do they work together? What are their tolerances and limits? If I make widget A in an engine of a different material, I can predict how that will change how long the engine can run before breaking. Although I’ve read some discussions of medical practice that point out the large strides we’ve made on the organ level (ie, organ transplants, knowledge of how the endocrine system works on the level of organs), we still know remarkably little about converting knowledge of cellular and molecular processes to macroscopic human health. When will we know enough? As a benchmark, let me propose we’ll know we know enough when we can take the genome and epigenome of a unicellular embryo and forecast general elements of that’s person’s future health and phenotype, given a known future environment. As the ongoing discussion of incidental findings and variants of unknown utility in exome and genome sequencing makes clear, we’re not anywhere near that.
This lack of knowledge means forecasting (to use the preferred term of Nate Silver) is extremely difficult.
Another element of lack of knowledge is that it is exceedingly difficult to forecast decades in advance what the market will want and what price it can bear. Even when a drug reaches clinical trials, it’s still five to ten years away from the market, meaning large amounts of uncertainty are unavoidable in terms of forecasting competition. There’s also the element of predicting social behavior. What leads people to choose to use specific drugs? What makes things popular?
Regulatory and reimbursement environment: this is touched on in cost and quality above, but I think it deserves its own point because, as I’ve been positing, this is the current customer for drug development. These elements have an incredible influence on the success of drug development and are in many ways volatile and unpredictable. The Affordable Care Act is an example of how things can change external to biopharma, and with dramatic potential effects. I don’t pretend to know what the ACA will ultimately do to drug pricing or uptake.
Given these bottlenecks, a disruptive technology, then, would be something that has an extreme effect on one or more of these elements, to the extent that the new technology does not simply make it easier to move up the peak of the adaptive landscape, but one that changes the landscape entirely. I’ll reiterate here that disruptive change doesn’t have to come in the form of a new method for making drugs per se. Rather, disruption may be something that fundamentally changes the way people use different products and services to create and maintain health.
Figure 1: revised version of the adaptive landscape for drug development, showing the difference between the effects of incremental versus disruptive innovation. Incremental helps get you to the top a little faster. Disruptive changes everything. Notice that the new peaks in the disrupted landscape aren’t necessarily drugs per se.
So if there were an innovation that made a dramatic difference in one of those areas relative to human health, that could be disruptive. And it would not necessarily happen through pharmacology
In part 2 I described some of the markets that might support a disruptive technology. Let me pull a few specific examples of developing technologies from those markets and how I see that technology or innovation potentially becoming disruptive.
A good place to start is in the agricultural market, and the company that seems headed towards developing a disruptive, landscape-altering approach is Monsanto. When one talks about rivals to Biopharma in the race to the bottom of the reputation bobsled run (where the US Congress is already waiting, having secured the top of the podium), Big Ag is one of the leading competitors with Biopharma for the Silver. However, Monsanto has been taking a kind of an inside-out approach to developing new products which might have implications beyond just a better tomato. As described in this profile from Wired, Monsanto is employing their research teams to figure out ways to naturally introduce improvements into commercial crops, with an emphasis on figuring out how to forecast phenotype from genetic sequence during cross-breeding experiments.
Particularly intriguing is this description of Monsanto’s ramped up plant breeding methodology:
“Monsanto computer models can actually predict inheritance patterns, meaning they can tell which desired traits will successfully be passed on. It’s breeding without breeding, plant sex in silico. In the real world, the odds of stacking 20 different characteristics into a single plant are one in 2 trillion. In nature, it can take a millennium. Monsanto can do it in just a few years.”
What seems disruptive here is the element of forecasting. Monsanto is developing tools to make the process of breeding better traits into plants less uncertain, and one element of that is having a reasonable idea of what will happen when different genetic combinations are put together. When phrased that way, it doesn’t seem that different from what a host of startups are trying to do now. Companies like 23andMe or Personalis have also made forays into this space, trying to forecast human phenotype from genetic sequence information.
Why should I think Monsanto might be able to get there first, and develop algorithms and methods that those other companies won’t? The main reason is that Monsanto isn’t working on human health, and because of that they have a freedom to operate and experiment. As was seen recently, the FDA recently ordered 23andMe to cease their activities in health-related genome interpretation precisely because 23andMe was trying to act directly within the existing human health framework. Another reason to be intrigued by Monsanto is that they have been making several strategic purchases of companies that deal in Big Data, such as the Climate Corporation, and recently the soil analysis part of Granular (formerly Solum), which can be thought of as being in the data acquisition part of a Big Data pipeline. While the proximal reason is to help farmers by gaining expertise in areas such as more accurate weather forecasting and optimal planting strategies, the generic expertise Monsanto is gathering is in the area of Big Data analytics and forecasting.
If Monsanto should succeed in figuring out how to forecast phenotype more accurately and reliably from genetic information, they’ll be sitting on a powerful tool that could greatly affect how we maintain human health. Possibly this could be used to derisk conventional drug development by increasing certainty about forecasts of eventual drug success. Or possibly Monsanto comes up with a more holistic model of how to affect health through a combination of fertilizers–uh, I mean existing drugs–and managing environment.
Another market for disruption is found with patients’ groups. As briefly described in part 2, the power of peer-to-peer networks may itself become a disruptive technology for human health. There is a fascinating struggle going on now between established businesses and various peer-to-peer services such as AirBnB, Uber, Lyft, and other companies which allow users and suppliers to connect directly with each other. In Seattle there has been a heated debate about whether and how to regulate Lyft and Uber versus conventional taxis. Certainly no one wants to compromise rider safety, but at the same time, it’s unclear how much regulation and licensing are truly needed versus allowing different business models to thrive.
How would peer-to-peer networks operate to disrupt the current status quo of drug development by Biopharma? There are a couple of possibilities. One is the previously described example of Kalydeco being developed due to the organized efforts of a specific patients’ group. It’s possible that other drug development programs will benefit from this kind of patient organization. First, programs can benefit from multiple layers of incremental innovation enabled by an activist patients’ group. Two possible ways could be the patients’ group providing samples to help target discovery scientists perform disease specific analyses and also the group supplying an eager, highly available and motivated population for clinical trial enrollment.
Second, patients groups could simply do an end-around the current regulatory framework. Just as Uber allows car owners and people who want a ride to connect directly, and AirBnB allows visitors and property owners to rent out rooms one-on-one, patients’ groups could connect with scientists, manufacturers and healthcare providers directly to create a parallel process for finding and testing drugs. This would be riskier in many ways, but that’s the shift in the Value Network. Patients groups could decide that increased speed and lower costs could be worth the risk. They could invent their own processes to ensure reasonable safety while accelerating drug development in different ways. And with the increasing sophistication of crowdsourcing and crowdfunding mechanisms, funding these activities seems quite doable.
Patients’ groups could develop their own standard of care, their own support networks and recommendations for medicines, regimens, tests, and practices. Use of self-monitoring and digital health tools would allow continual accumulation of data across the patient population, and the ever dropping cost of storage and data analytics would eventually allow Big Data mining of this dataset to allow a process of continual, on-the-fly testing and refining of care. Mobile apps would help people keep to their regimens and report back interesting observations and correlations.
Just as an example, and to suggest this isn’t completely far-fetched, I’d again point to the evolving debate about fecal transplants. After publication of initial studies showing the Clostridium difficil infections could be more effectively treated by fecal transplants than by antibiotics, the FDA moved to regulate fecal transplants as a kind of a drug. However, outcry from the medical and patient communities led the FDA to pull back and not enforce this interpretation. The recent opinion in Nature (linked above) proposes a different approach, classifying fecal transplant material as an organ or a custom tissue, similar to blood. So organized effort can influence policy. It’s an open question whether such efforts can take several steps further and move health out of the current regulatory framework entirely.
Getting to the future, but more quickly
When taking a probabilistic worldview, it’s really not so much predicting exact things but rather forecasting trends. What I’ve tried to do is look at some current trends and ask where that might take the process of optimizing human health, especially given that the current system in drug development appears to be in a difficult position with respect to making substantial progress.
I don’t really know what it will look like, but it’ll be fun and interesting and hopefully better for human health when disruption finally does appear, and probably from where we’d least expect it.