Chief Algorithmic Analytics Officer, Sumitovant
Bill, you are the Chief Algorithmic Analytics Officer at Sumitovant. Tell us about your role in the company.
My title arose out of the Computational Research team that I pioneered at Roivant. Our focus was using data to understand and optimize the potential value of clinical stage drugs. Within Roivant, my team developed expertise, algorithms, and code on top of some traditional and some non-traditional data sources to systematize the due diligence process. The computational ecosystem that we call DrugOme evolved out of that effort. I think the scientific approach we were taking toward the drug valuation process resonated with senior leadership at Sumitomo Dainippon Pharma, and so DrugOme became an integral part of the creation of Sumitovant.
Within our group at Roivant, the focus was on the creation of independent entities that would manage the clinical stage development of a drug and evolve independently to successful commercial organizations. Sumitovant’s focus is more on the strategic management of our portfolio of companies, and so my team has been evolving in that direction. When Sumitovant was formed, we had a roadshow to demonstrate DrugOme capabilities within Sumitomo Dainippon Pharma and to its subsidiaries. We subsequently developed pilot projects with various groups in those companies. Given that I was trained as a scientist, I have particularly enjoyed the engagements we’ve had with drug discovery teams.
The way I see it, if we can successfully value drugs, then we can successfully engage at any stage of drug development.
What’s your favorite part of the job and what are the challenges?
I work with a team of highly motivated, talented people who are passionate about their work. We work on problems that are interesting, challenging, and the solutions of which lie at the core of effective strategic decision-making in pharma. As an outsider to the pharma industry, I built that team from scratch. The value of our contribution was recognized by a storied Japanese company and became integral to the creation of Sumitovant. Through the efforts of that team we have established a center of excellence within the Sumitovant family for the use of advanced data and algorithms to support pharma strategy.
The challenges were legion. Core among them is a cultural disconnect between a computational approach and the traditional pharma approach. There are fairly deep reasons for the disconnect, but one in particular is related to the – frequently rightful – suspicion of data in both the biological and medical domains, data that only captures a shadow of complexity of human health and the patient experience.
Our organizational structure enables us to partner with eclectic groups with substantial expertise in vastly different domains. It is a constant challenge to have minimal experience in a domain, come to a partner with vastly more experience in that domain, but limited experience in ours, and create effective engagements that deliver value. Frequently, but not always, we do create value. That value becomes the core of ongoing synergistic engagements.
In your opinion, what is the biggest impact Sumitovant will have on patients and the world?
Near term, it is the impact of the drugs in our pipeline that are on the cusp of commercialization. Because I was a part of the due diligence for many of the drugs in the pipeline, I have a high degree of confidence that these drugs address the unmet need of patients and are broadly beneficial to society.
Long term, great success will only be possible if our current strengths translate into an enduring strategic advantage. If I am successful in my own role in such an evolution, then we will be world class in 1) translating the big data of electronic medical records into clearly identifying unmet need, 2) quantifying the competitive landscape of drug development based on an evolving tapestry of pharma data, and 3) utilizing natural language processing to rapidly translate large corpuses of documents into structured, actionable data. This will all support data-driven quantification of the medical landscape, which will enable Sumitovant decision-makers to make data-driven decisions against the impact of those decisions on the overall value of Sumitovant’s portfolio.
How will your role – and roles similar to yours – change along the way?
There are not a lot of roles like mine, so the first change is that a role like mine will become more common.
I have built three teams that reflect what I believe to be the three technology and data areas that will have a large impact in the intermediate to longer term on drug development. The first of these is the electronic capture of medical records and transaction data in pharma. This data is incredibly useful across almost all areas of pharma, from driving efficiency in clinical trial enrollment, to enabling “synthetic” control arms that can reduce the cost of clinical trials, to detailed stratification of indications for potential drugs, to identification of the doctors who will be most impactful at delivering those innovative drugs to patients. The maturity of the use of this kind of data in these areas is still in its infancy. The second area is a bit more complex to describe, but there is structured data dispersed through publicly available and subscription sources that enable a still incomplete, but substantially improving qualitative understanding of the landscape of unmet need in a therapeutic area. Just as an example: Regulatory changes have vastly increased the availability of structured clinical trial results in clinical trial registries. This is hugely valuable in understanding the efficacy landscape of an indication, but only if you invest the time to bring that data to clinical decision makers’ fingertips.
Finally, natural language processing has huge promise to rapidly generate structured, accessible, actionable data out of vast numbers of documents.
In your opinion, what impact will AI and wearables have on our business and industry in the next 5 to 10 years?
I’m not a huge proponent of AI as a descriptive term. But the algorithms of natural language processing and machine learning have many applications in pharma. There are obvious ones, like Google’s recent breakthrough on protein-folding. But there are some not so obvious ones. Pharmacovigilance is an example. Drug companies and doctors have a regulatory obligation to report adverse reactions to drugs. Duplicate reports are of value to no one and potentially harmful to drug companies, and they certainly creep into the data-reporting process. In addition, when adverse event reporting is of adequate quality, there are statistical techniques that can facilitate the identification of classes of adverse events that have real implications for the mechanism of action of the drug and tie them to other scientific evidence to develop a holistic picture of how drugs interact with the body. The body is an immensely complex system that no one is really capable of fully understanding, so tools that enable the identification of connections across domains can substantially improve our understanding.
Regarding wearables – when the FDA looks for efficacy of the drug, they try to focus on clinically relevant endpoints. That is, even if your drug reduces a harmful metabolic waste product that degrades the function to a patient with a rare mutation, if that patient does not see something like improving lung function then the drug may not get approved. But these endpoints are noisy: A typical clinical trial only brings in patients for a fairly small number of clinical visits and only measures the endpoints during those visits. To reach statistical significance, you need to recruit a larger number of patients to compensate for your noisy data. The promise of wearables is that you can continuously monitor every single patient instead of only monitoring them at a small number of times. Intuitively, but also mathematically, you will get a very different understanding of, say, your blood pressure if you measure it all the time instead of once a year.
For rare diseases, it is difficult to recruit patients, so statistical significance translates into fewer patients in trials and faster, more cost-effective clinical trials. I learned early on at Roivant that much of the cost of clinical trials for many diseases is related to site management, NOT clinical tests. Reduced patient count means less site management, which is critical to keeping costs low.
In your opinion, what’s the most pressing societal issue we are facing in healthcare these days?
Covid-19 has exposed some gross pitfalls of our current system. Clearly there can be progress on integration of public health data for fast regional and national visibility into disease impact. Right now there are vendors who sell this data, but it is not cheap. Broader access can enable more eyes on the problems of public health and transparent decision-making on public health questions.
Where pharma can play a role in infectious disease is developing a portfolio of de-risked technologies that enable fast development of vaccines and treatments for infectious diseases. We developed a technique to measure the actual average cost burden of a diagnosed condition on a patient using real electronic claims data on tens of millions of patients. It was very apparent in that data that the cost of an individual infection in an unvaccinated patient of a common disease can be very high for that patient. Static measures of unmet need that drive most of the drug development decisions in pharma don’t reflect the potential cost burden of situations like an outbreak.
Broader availability of the data that enable these sorts of insights will put more eyes on the problem of identifying future health crises.