- This morning, Novo Nordisk announced a partnership with IBM’s Watson Health to create diabetes solutions built on the Watson Health Cloud. There are no product specifics or timing, but the overarching goal is better insights for providers and patients. We speculate below on what could result from this agreement, including clinical decision support, comparative effectiveness research, predictive analytics, and better data for payers.
- The news is a major move for Novo Nordisk, who hasn’t historically been very active on the partnership front and hasn’t talked about digital health publicly. Today’s announcement follows Watson Health’s partnership with Medtronic Diabetes and a pivotal year for insulin companies moving into digital health (Sanofi and Verily [formerly Google Life Sciences], and Lilly & Companion Medical).
- We interview VP of IBM Watson Health Rob Merkel, who emphasized the rapid growth in medical data and how Watson can help all stakeholders handle it. Healthcare is clearly a major priority for IBM.
This morning, Novo Nordisk announced a partnership with IBM’s Watson Health to create diabetes solutions built on the Watson Health Cloud. The exciting announcement naturally combines Novo Nordisk’s deep diabetes expertise with IBM Watson’s computing brainpower. There are no product specifics or timing, but the overarching goal is better insights from real-time, real world evidence of Novo Nordisk diabetes treatments and devices.
We discuss below what could result from this agreement, including clinical decision support and more personalized therapy; comparative effectiveness research and more digestible real-world evidence; predictive real-time analytics for patients or providers; better payer insights (population management); stronger mining of clinical trial, medical record, or claims data; finding new drug targets and speeding drug development; and optimizing clinical trial patient recruitment. We believe there is particularly strong upside for payers, as Watson’s capabilities should allow for much more efficient analysis of “real-world” data on how certain products perform, especially in sub-groups.
The news is a major move for Novo Nordisk, for whom partnerships are rare; to date, the company hasn’t talked about digital health publicly. We see this partnership as a natural progression to enhance the efficacy of diabetes drugs – as peptide innovations become more challenging and the bar for new drugs rises, there is great potential to improve outcomes through technology and data, and to improve access by providing payers with real world data demonstrating value. In a conversation with us early today, Novo Nordisk Executive Vice President, China, Pacific, and Marketing, Mr. Jakob Riis shared that ongoing discussions started some time ago and “matured” over the past year. He emphasized this partnership’s potential at a range of levels and we found it exciting to think about the potential for patients, providers, policymakers, and payers. We see the goals and scope of the partnership as dovetailing nicely with the Cities Changing Diabetes program: using real-world evidence to improve health and diabetes management on a population level. We hope there is room for some synergies between the two going forward (see our interview with Mr. Riis here on that).
We also spoke with VP of IBM Watson Health Rob Merkel (see interview below), who shared a broad vision for Watson centered on a single concept – the rapid growth in medical data and information is far beyond the realm of the human mind to make sense of it. Indeed, this compelling Watson Health video makes the challenges clear: (i) the average person is likely to generate more than one million GB of health-related data in their lifetime (~300 million books!); (ii) medical data is expected to double every 73 days by 2020!; and (iii) less than 50% of medical decisions meet evidence-based standards.
Today’s announcement follows Watson Health partnerships this year that include Medtronic Diabetes, J&J, Apple, EPIC, CVS Health, Teva Pharmaceuticals, and many hospitals. This is clearly a high priority area for IBM – in our interview, Rob Merkel noted that the company’s CEO has referred to healthcare as IBM’s “moonshot.” He also noted how fast the Watson group has grown, from 24 researchers that built the original Jeopardy computer to 2,000 people by the end of 2014 to another 2,000 devoted specifically to Watson Health when it was launched in April.
More broadly, it’s been a pivotal year for insulin companies getting into digital health: Sanofi partnered with Google Life Sciences in August (now rebranded as “Verily”) and Lilly invested in smart pen startup Companion Medical in May. We hope to see this accelerate in 2016 and beyond as companies recognize the need to go beyond a drug in isolation.
Merck’s Global Health Innovation Fund has called data the “future currency in healthcare.” We are glad to see diabetes companies increasingly recognizing this. Who will build something truly compelling for delivering better outcomes at lower costs?
Potential Novo Nordisk/IBM Watson Health Products
- The list below reflects our speculation on what this partnership could produce; it is informed by our discussion with VP of IBM Watson Health Rob Merkel today, other IBM Watson partnerships, and gaps in diabetes care. This is not a comprehensive list of everything the companies could feasibly build. Broadly, we believe Watson can help address many seemingly insurmountable challenges in diabetes: personalizing therapy; aiding time pressed providers; getting better real world data on what works and what doesn’t; actually doing something with Big Data, particularly when it is unstructured; and preventing waste in the system.
Clinical decision support and more personalized therapy?
Could Watson’s computing power make it easier for clinicians to select the right diabetes treatment at the right time, or help titrate drugs like insulin? Personalized therapy is also a focus of the Medtronic-IBM Watson partnership, and we wonder if Novo Nordisk might consider this too. For instance, we could imagine Watson analyzing a patient’s medical history and recommending a treatment plan and dose that is most likely to work. We wonder if genomic data could eventually be incorporated into this as well.
Comparative effectiveness research and more digestible real-world evidence?
Watson has access to millions of patient medical records. Could the computer conduct large-scale cohort comparisons to better understand how well different treatments work in specific populations?
Predictive real-time analytics for patients or providers?
At DTM 2015, Medtronic showed an exploratory analysis suggesting IBM Watson can predict hypoglycemia with 80-90% accuracy three hours post bolus insulin delivery. Could Novo Nordisk and IBM Watson offer similar predictive analytics for patient and providers?
Better payer insights (population management)?
Today’s announcement mentioned “real-time, real-world evidence” – perhaps Novo Nordisk and Watson could analyze diabetes data and flag patients at risk for payers. For instance, identifying ER admits for a hypoglycemia event that are likely to return to the hospital. The key here is “real-time,” which could suggest a future connected insulin pen, connected meter or CGM, or some other source for streaming real-time information about patients (Watson does have a partnership with Apple’s HealthKit).
Better mining of clinical trial, medical record, or claims data?
These are vast sources of data about patients – could Novo Nordisk and IBM apply Watson’s computing power to gain deeper insight into patients for product development, to provide better decision support, to build better drug algorithms, etc.? Could they develop an algorithm to help translate clinical trial results to clinical practice by identifying populations most similar to those enrolled in a given trial?
Drug targets and faster drug development?
Watson has programs to identify drug targets and speed drug development – could the computer help Novo Nordisk develop new therapies?
Optimizing clinical trial patient recruitment and selection?
Could Watson more tightly define the patients most likely to benefit from an investigational intervention?
Identifying target populations for public health interventions?
Could Watson be incorporated into the Cities Changing Diabetes program and identify populations at high risk for diabetes based on social and cultural risk factors?
Interview with VP of IBM Watson Health Rob Merkel
KELLY CLOSE: Congrats on this announcement! Thanks so much for taking time to speak to our team today. Can you talk about selecting Novo Nordisk as a partner?
ROB MERKEL: We are delighted to be working with Novo Nordisk, who has shown such clear leadership in diabetes care. At Watson Health our clear leadership is in “cognitive computing” and insights. We are thinking about how to develop new insights for patients to successfully manage their diabetes.
The sobering statistic is that there are 415 million people with diabetes. That is equivalent to everyone in North America. It is a pretty big challenge. On the other hand, it is a huge opportunity. What’s very exciting in Watson Health is that it’s all about achieving new levels of insight.
With Novo Nordisk, there are a significant number of lives we can look at. We have 50 million de-identified lives we can use for diabetes. We can look across large bodies of diabetes populations and understand cohorts – patients that have similar characteristics that are meaningful in how you treat them. Then we can begin to understand what works well and what doesn’t work well. We can put in place predictive models to understand risk for co-morbidity. Those are the types of things that we hope to build out.
ADAM BROWN: It sounds like this will be mostly provider-facing work. Is that right and will there be a patient-facing element as well?
ROB: It will be both. We’re working through specific use cases with Novo Nordisk now. There’s no question from what I’ve read that there’s an opportunity to improve the way diabetes is managed, not only in developing treatments but personalizing treatments further to optimize outcomes. There are several potential use cases: research, how clinicians develop treatment plans for diabetes, or helping individuals to understand their personal circumstances. It could be a whole range of things.
ADAM: How is IBM thinking about healthcare more broadly? Where can Watson have the most impact?
ROB: What’s shaping our strategy are the foundational concepts of knowledge and data. At Watson we are providing insights into what knowledge exists, and creating new information out of large bodies of data.
In terms of knowledge, there is a proliferation of medical literature. It used to be that the amount of information that exists doubled every five years. Then it was every three years, then every two years. Now, the projection is that by 2020, every 73 days the amount of medical data will double. There are over 700,000 new research publications each year, and the average researcher reads 100-200.
Looking at clinical trials, the NIH clinical trials database has over 180,000 trials at any given point in time. In many, many lifetimes, physicians, researchers, and nurses can’t even begin to scratch the surface of how much information is being generated. It is beyond human cognition. No one could read that much, even if they were a speed reader with a photographic memory.
From the data side of the lens, it is more daunting. An article in Health Affairs in August 2014 cited study after study on what determines your health in three categories: clinical factors (those in an EMR), genomics factors, and exogenous factors (exercise, diet, etc.).
- If you look at clinical factors, what’s in the EMR record determines 10% of your health. How much information is generated from that? 400 GB, which is equivalent to four top of the line smartphones.
- Genomics determines 30% of your health, which is 6 terabytes, or 6,000 GB. That’s 60 top of the line smartphones.
- Exogenous factors - exercise, what you eat, etc. – drive a large part of everyone’s health. Approximately 60%. From there, the information explodes. For one person, we estimate you generate 1,100 TB of information in a lifetime. That is 11,000 top of the line smartphones.
My smartphone holds a lot of movies, songs; I cannot imagine what 11,000 would hold. That’s where we need to get the insight on what’s driving 60% of your health.
When we think about our mission at Watson health, we want to provide insights into what knowledge and data exists so that clinicians and patients can get exactly the right insight exactly when they need it.
Within the population of IBM, we are shifting a huge amount of resources into Watson and Watson Health. To give you some background on the trajectory, the original team that built the Jeopardy program was 24 researchers over three years. In January of 2014, we announced the Watson group. At the end of that month it included 1,000 resources. It was 2,000 by the end of the year. We announced Watson Health in April with another 2,000 resources, and we plan to grow dramatically beyond that. We very rarely as a company announce an organization externally. Before the Watson group last year the last times were the PC division in the 1980s and IBM Global Services. That gives you a sense of how meaningful the announcement was for both the Watson group last year and the Watson Health organization. Our CEO described this as our “moonshot.”
ADAM: What are Watson’s data sources? EMR data is only updated quarterly at best in people with diabetes, so will there be more “real-time” data sources as well?
ROB: It will happen on a few levels. The specifics still have to be worked through. You’re right about the timeliness of EMRs, which are updated quarterly at best, and that’s even if they’re updated from the source hospital. Watson is “evergreen,” so we always have the latest patient records. We have already stated publicly several partnerships we are working on to expand upon getting that amount of information in a timely basis. We have the Apple partnership with Health Kit and Research Kit. We’re already working with them on collecting information from individuals that participate in various programs and being able to analyze that information in Watson Health Cloud.
Imagine something that captures all this information, correlates it, and analyzes it. You can do lots of things with massive data sets and analyzing real time information on behalf of physicians, and doing so in something that is HIPAA and GXP compliant. It’s a very scalable platform, it’s regulatory compliant, and we’re leveraging that with new data sources.
With Medtronic we’re integrating directly with glucose monitors and other devices. With Novo Nordisk, they are interested in capturing and analyzing information related to their products. We don’t just see this as quarterly EMR input. We see this as being a much, much tighter relationship with clinicians.
KELLY: Switching gears, can you talk about which Watson Health partnership are you most proud of?
ROB: I’m proud across the board. That’s like asking me if my house was on fire, which child would I save. [Laughter] In India last week, we announced that a large hospital system will begin using our program for oncology that we trained with Memorial Sloan Kettering. That’s a huge source of pride for me. It will be 16 hospitals, both teaching and commercial, using this to address a very tangible problem. In India, the ratio of cancer patients to oncologists is 1,600 to 1. By comparison, the ratio in the US is 100 to 1. And in the US, they say they don’t have enough time to see patients. When you’re waiting to meet with the head of oncology, you’re sitting in the oncology waiting room in a sea of people with cancer who are waiting for an extraordinary amount of time to meet with a physician to get basic information. All of us have experience with cancer personally or through others. Think about going through that experience and spending half a day waiting to get 15 minutes with a physician. Anything we can do to help an oncologist so they can understand patients faster is a meaningful contribution and something that gives me a tremendous amount of pride.
It’s the same thing with diabetes. We can analyze information sets that are beyond human cognition and are able to provide a better understanding. That’s something that also gives me tremendous pride.
ADAM: We’re interested in learning more about the relationship between Watson and clinicians. Is there any pushback from clinicians? How do you think about preserving clinical judgment and the practice of medicine while also making clinicians far smarter and more efficient?
ROB: I’ve been in this role for two years. I can tell you, I have seen a material movement in the thinking of clinicians since I’ve been in this job. The typical resistance was people saying that medicine is an art, not a science, and I know it all already. I’m saying there’s a big movement beyond it. This isn’t about you being a good or bad doctor. This is about the amount of information you would need to absorb is beyond what any human being can possibly do. And then you need to apply it in a timely basis at the point of care. I think most people get it. Then it evolves to a more constructive conversation about how can I do it in my clinical workflow so that it doesn’t create a lot more work for me? I think this will help me, how can I now integrate it into my practice?
I’ll give you one other example of a research project in IBM Research. An ACO in the US came to us and said, we want to better predict when a patient is going to be diagnosed with congestive heart failure, when they show absolutely no symptoms whatsoever in the context of an annual physical. The ACO is the risk-bearing entity. They gave us 5,000-10,000 patient records with tens of thousands of fields with different pieces of information per patient. Starting off with the Framingham criteria, we came up with a mediocre predictive model. What we then did was unsupervised machine learning. We got the computer to read through records based upon the question we were asking in the context we were asking it: no symptoms, an annual physical, up to six months in advance. By doing that, we came up with 500 different things. If you look at one piece in isolation, it’s just noise. In Framingham, we’re looking at 120 of those 500. The 500 in conjunction formed an incredibly powerful predictive model that could predict the risk of being diagnosed with CHF up to six months in advance and even up to two years in advance. How would a clinician ever be able to do that? Let the machine do the heavy lifting, or the heavy reading, and provide the insight to clinicians exactly when the clinician needs it.
KELLY: There are a bunch of major challenges in diabetes as you know and as Novo Nordisk is working so ambitiously on. One we really want to talk about is access: what are payers willing to reimburse. Can you talk about to what extent your work will be able to document the value of existing and new products?
ROB: There’s no question about it, and this is similar to any real-world evidence and epidemiology studies. What we can do through the Watson Health Cloud is look across large patient populations and perform comparative effectiveness research. Broadly, you are figuring out which interventions have the optimal outcome. That outcome could be an efficacy statement, a quality of life statement, or an economic statement. We can then go back to a payer and say when we follow this intervention for diabetes with this profile, here’s what works and what doesn’t work. I think all stakeholders stand to gain from analysis like this, nobody more so than the patient. We can elevate this insight into what works and doesn’t work, and the opportunity to do so today is much greater than in the past with these types of analytics.
KELLY: That’s so fascinating. Do you see the world evolving from RCTs as the gold standard to also looking in more depth at more real-world data to supplement the standardized controlled data?
ROB: From a regulatory standpoint, there are probably people better suited to comment than me. I think the entire R&D lifecycle will accelerate in the future. Even within Watson health, we can provide insights into drug discovery and accelerate clinical trials, and that will enable a major acceleration into real world evidence. That is going to compress timelines. Whether it radically shifts the regulatory process, I’m not the best person to answer that. But the level of insight that we’re going to have throughout the entire R&D lifecycle and the healthcare value chain is going to be profoundly better than in the past.
Background on Watson
- IBM’s Watson has a few features that make it exciting in diabetes:
- Watson can understand the English language (Natural Language Processing) and analyze unstructured data (80% of the world’s total data, as cited in this IBM case study);
- Watson can learn over time (e.g., incorporating the latest symptoms and test results on individual patients, the latest medical research, and the newest clinical trial outcomes);
- Watson is housed in the cloud on IBM’s servers, meaning numerous devices can access it (i.e., no need for a massive on-site server); and
- Watson is open source, meaning anyone developing products can work with IBM to use the technology. The company’s page on Watson shows applications in healthcare, finance, retail, and the public sector.
- “It would be impossible for us humans to replicate what Watson does in healthcare. Not only can it answer questions pulled from millions of individual documents, it can instantly cite the source and confidence level. Beyond empowering physicians with the most powerful Q&A tool ever created, it will fundamentally change the practice of medicine.” – Modernizing Medicine CEO Daniel Crane in Peter Diamandis’ excellent book, BOLD (email Adam if you would like his notes from this book).
- Watson has done some very interesting clinical work in oncology with partner Memorial Sloan Kettering (MSK). For instance, Watson can analyze a patient’s medical information against a vast array of data, including expert training from physicians, case histories, established treatment guidelines, and published research to provide individualized, ranked, evidence-based treatment options at the point of care. We wonder how that sort of compelling clinical decision support could be applied to diabetes.
- “IDC predicts that by 2018, half of all consumers will interact with services based on cognitive computing on a regular basis.” Whether that is true or not, it certainly suggests computers like Watson will be much more widely used in the coming years. We’re not sure what the base is now, but we assume it not very high.
Close Concerns Questions
Q: What products will ultimately emerge from this partnership?
Q: Could Novo Nordisk leverage work from the Medtronic-IBM Watson partnership?
Q: How could Watson’s analyses impact payer conversations about the value of drugs?
Q: Will Novo Nordisk seek a glucose monitoring partner?
Q: Will Novo Nordisk launch a connected insulin pen?
Q: Which insulin player will make the biggest investment in digital health? Will 2016 see more investment in digital health from traditional diabetes drug companies?
-- by Adam Brown, Emily Regier, Ava Runge, and Kelly Close