Ginger.io has been one my favorite subjects since joining the iMedicalApps team (see here and here), so when the opportunity arose at last week’s mHealth Summit to interview Anmol Madan, PhD, co-founder and chief architect of the company’s technology, I did not hesitate.
For those not familiar with the company, their expertise is analyzing data collected passively from mobile phones to infer behavioral clues that can help detect or treat health conditions.
Dr Madan is a recent graduate of the MIT Media Lab, easily the epicenter of digital medicine innovation in the Boston area, second only to San Diego as a regional hub of wireless medical technology.
His young company has had quite an exciting first six months. They just won $130K in the Data Design Diabetes Challenge hosted by Sanofi-Aventis, and recently closed $1.7M in Series A financing. Below Dr Madan explains his company’s vision of developing a human “check engine light,” a truly disruptive innovation if the company is successful.
BTE: When did you realize health care was the ideal application for your research at the Media Lab?
Anmol Madan: My focus at the Media Lab was not particularly in health care, I am a computer scientist by training focused on modeling of sensor data to understand human behavior while at the Media Lab, specifically using data from cell phones to understand movements, activities and interactions.
Toward the end of my thesis we conducted this big experiment where we collected over 300,000 hours of interaction data over a year with a few different questions in mind which we were trying to understand, the diffusion of political opinions, what happens to people when they are asymptomatic, and a few similar questions where you would expect interactions to play a role and we had some really compelling findings in the health context.
We were actually able to predict which days patients were asymptomatic based on their mobile data within that population for that data set. That was a very interesting moment because if you scale that up to the four billion phones on the planet and if that result does generalize across the population then you can really do interesting things with it.
BTE: What is the business model or is that still unclear?
Anmol Madan: Essentially we are building a platform that collects, analyzes and processes the data at scale and then powers different ways of intervening based on this data. Very basically, our system is a check engine light for a person. There are a few use cases for that, for example if you are running a research study and you are dependent on patient reported outcomes and surveys there are obviously going to be some biases and issues with those patient reported outcomes and there is a lot of money being spent in the industry based on these outcomes. So we see the lowest hanging fruit to be giving those researchers ways to get real world behavior data. This applies not just to clinical researchers, but also pharmaceutical and other commercial research groups as well that care about understanding some of those behavior patterns.
The second opportunity, which we have seen covered a lot here at mHealth, is how do we manage these large patient populations. especially chronic patient populations that are at risk or diagnosed with a certain condition and are potentially very expensive to the system. One of our partners, Cincinnati Children’s, has a 10,000 patient IBD population, so one of the questions for them is once these technologies are validated, how do they use them to manage this patient population and determine which patients need treatment during a given week. This is a longer term use case for the technology.
BTE: Are you guys focused on all patient verticals? Do you think your technology will work across the board with all conditions?
Anmol Madan: There is medical literature that says “here is how a patients behavior changes when they have a certain condition”, so there is a mapping between picking certain conditions when you would expect a behavior change. If you expect a behavior change, we hope to capture that using the mobile phone, our goal is to design the algorithms capable of doing that. Mental health is one big patient vertical for us, IBD and stress, congestive heart disease and diabetes are a few others.
BTE: I was following the Data Design Diabetes Challenge very closely, which you guys won, and I am curious, how did you design a solution for diabetics specifically?
Anmol Madan: That was a very interesting experience for us because we actually didn’t realize the value our system could have in the diabetes context until we actually went through the competition. One of the interesting things about patients with diabetes, especially Type II diabetes, is that there is a period of a few years where they go from being a normal healthy individual having a normal life to somebody who has to now take their health in very specific ways and in some cases you end up being insulin dependent, but all within a few years time frame, which is obviously a very difficult transition for people. The question is how to design tools to determine when a patient needs social support, as many patients are dependent on their family members, caregivers and in some cases depend on their nurses or doctors to help them make that transition seamlessly and support you through that process.
One of the things we learned through that diabetes challenge while we were doing our homework and working with some of the patients was there is actually literature that talks about how psychosocial support at the right time in that process can actually reduce the number of hospital visits, readmissions and emergence room visits, and ultimately it boils down to the same question of “how do you identify which patients need support and when?”
So there the opportunity was to use our system to understand the stress level of the patient, how much they are dealing with, get some indicator on that which could be combined with their glucometer or some other sensor data and using that to drive intelligent alerts. The key is driving those alerts with the patient’s permission using the peer groups they have selected and for each individual that group is different. For one person its their spouse, for another its their friend.
BTE: So you leave it up to the patient to determine who they want to see their “check engine light” when it starts blinking?
Anmol Madan: Exactly. One thing that is often missed is the fact that this data actually belongs to the patient. I think finding the new models for data ownership and enabling the patient to determine how it is used and shared is going to be a critical part of our system.
BTE: So is it a machine-to-machine solution that uses apps on both the patient’s phone as well as the phones of the patient’s identified peer group?
Anmol Madan: You actually don’t need to have an app on the peer’s phone because you can use SMS or email to send notifications. You really only need an app on the patient’s phone.
BTE: So is the system totally passive with truly zero patient input?
Anmol Madan: The system was built to be totally passive with no patient input, but what it does support is asking the patient questions and sending them notifications to collect data that otherwise would be difficult to collect passively or as part of the validation process. We have spent a lot of time working with providers who have thought of a lot of questions that ought to be asked when they think a patient is having a flare up for any of these conditions, whether its mental health or diabetes, and they have realized one of the challenges they have is knowing when to ask those questions because they have thought of behavior data as uninteresting but if you get a question about how your mood is every six hours it gets annoying after awhile.
However, if the system was saying “here is how you are behaving”, noticed a difference from your normal behavior patterns and then asked you the question, be it once a week or once a month, then you are more likely to get useful responses. So that is the point behind some of the notification and the ability to as questions, and it gives us better training data, which means overtime we should get better at mapping the questions to the behavior data.
BTE: That connection between passive data collection and patient notification and questioning is powerful. By simply collecting data passively you could easily get a false positive because their change in phone usage could be totally unrelated to their condition, right?
Anmol Madan: Correct, it could be totally unrelated. I think that is the primary difference between our system and some of the precise physiological sensors. I think there is going to be an ecosystem where these all co-exist, but I see our system as the least common denominator because it uses the cell phone, which everyone has, but you don’t necessarily have a blood sample for every person, so there can be some confounding issues. Our system will trigger the actions and determine whether a family member reaches out to you and asks you to take a blood sample.
BTE: So what is the next step for you guys? You just raised some venture funding, $1.7M if I am not mistaken.
Anmol Madan: Yes, we received investment from both angels and venture capitalists. The round was led by True Ventures, which is an early stage firm that has invested in companies like fitbit. One of the other investors was Mitch Kapor from Kapor Capital, and some of the other investors were strategic individuals within the health care space.
BTE: How do you plan to spend the money?
Anmol Madan: Well, first we bought a machine that cooks Ramen faster. We are engaging new partners which we will probably announce first quarter of 2012. We probably won’t expand to too many new partners, keeping it to just four or five, and really focus on expanding our research with those groups. We are growing the team, specifically on the engineering side to be able to support some of the bigger scale deployments and focus on validating our existing partnerships and then sometime in the middle of next year we will probably make a broader announcement and seek wider distribution.
BTE: What is an ideal partner for you guys? What types of organizations are you looking to engage?
Anmol Madan: We are looking for partners who believe in the value of the passive sensing approach or even remote sensing and are looking to work with companies leading that space. I think across the industry there are individuals and groups that have that mindset. So the partners could be on the provider side, specifically providers looking to manage or better understand a particular patient population. We are looking at partners using hybrid model such as bundled payments or ACOs. We are also looking for partnerships on the pharma side, specifically pharma companies that want to think outside the box and think around patient services and identify how patients are adjusting to the drugs they are selling.
BTE: How do you anticipate distributing the app in the future? Will you use app stores or do you foresee it as a prescribed solution?
Anmol Madan: We don’t really see it as a prescribed solution. We may use app stores but really our current participation is based on affiliation with one of our partners. We have thought carefully about launching direct to consumer and we revisit it every other month. We think there are a lot of interesting consumer models and consumer apps, but our model has been a little bit different. Our model has been to work with the people who are really feeling the pain point, be it financially or clinically responsible for managing those patients and give them the tools they need to do their job better. So the model is as long as people are affiliated with those groups they will be able to sign up and use the app.
BTE: So I as an individual can’t go to the app store today or in the near future and download the DailyData app?
Anmol Madan: No, you as an individual cannot today but this is something we revisit quite often so it could change in the next 3-6 months, but as of today we are focused on making sure we are focused on our four or five partner groups and their patients. This really allows us to iterate very quickly because you are working in a semi-controlled hospital network which means we can get things out and updated very quickly.
BTE: So I assume you guys have collected a whole lot more than 300,000 hours of data now?
Anmol Madan: Yes, and the great thing about this data is that it scales like crazy and the challenge becomes not simply collecting the data but determining what data is relevant and important. This is a difficult task. Fortunately we have been working with a lot of individuals out of the Media Lab, such as Sandy Pentland, Media Lab professor and one of our co-founders, and Frank Moss, Media Lab founder and a member of our advisory board, who have been thinking about this problem for a long time and that was my background before. But what’s interesting about going beyond the 80-100 people who would be able to use our system in an academic setting, we can now see how this technology works when scaled to tens of thousands or hundreds of thousands of users and anyone with a phone can install it.