http://type2diabetestreatment.net/diabetes-mellitus/softcgm-new-smartphone-app-to-continuously-monitor-and-predict-glucose/
A new venture tech startup is looking to change continuous glucose monitoring as we know it, doing away with the sensor altogether and instead focusing on smartphone algorithms to display constant blood sugar data and make glucose trend predictions.
Meet SoftCGM, a new entirely phone-based solution under development by Lancaster, Pennsylvania-based Aspire Ventures, and we’re thrilled that “one of our own” with type 1 diabetes and active in the Diabetes Online Community is on the team.
A longtime type 1, Marcus Grimm (@marcusgrimm) has been a D-blogger for years at Sweet Victory and makes some pretty awesome videos (Sh*T Diabetics Say), as well as being an avid runner and volunteer coach.
We reached out to Marcus recently to hear his personal story and learn some details about this futuristic SoftCGM tech in the works.
An Interview with Marcus Grimm on SoftCGM
DM) Marcus, can you start by introducing yourself?
MG) You bet. I am45 years old. Married with children, living in Pennsylvania. Aside from being T1 and that being my job, people sometimes recognize me from being a part of Team Type 1’s first running team a few years ago. I’ve run more than a dozen marathons and ultra marathons with T1, up to 100 miles, and I’m also the running coach for Diabetes Training Camp.
What’s your diabetes story?
I was diagnosed in 1984. I’ve been on the pump for about 16 years and CGM for several years, too. I’ve always considered myself fairly fortunate with my control, but about seven years ago, I realized that two of the three T1’s I’d grown up with had passed away. I decided then that even if diabetes was fairly easy for me, that didn’t mean it was easy for everyone, so I made it a point to become more involved.
I had one of the earliest blogs about the intersection of diabetes and exercise, but most of my diabetes outreach in recent years has occurred offline. Five years ago, I bicycled 84 miles in a single day and visited with ten legislators to gather support for the Safe at Schools Bill in PA. The same year I was named as Team Type 1’s Amateur Athlete of the Year. Two years ago, I started coaching at Diabetes Training Camp. These days, I’m a very active “lurker” in the online diabetes communities. I find there’s no shortage of great advice out there, so I try to only contribute if I feel I have a unique perspective.
Tell us about your work at Aspire Ventures, that"s creating this new tool?
I am Chief Marketing Officer, which is a fancy way of saying I’m a corporate storyteller. I spent several years managing an advertising agency before coming to Aspire. One of the Aspire managed ventures is Tempo Health, which is applying machine learning to diabetes technology. Tempo’s unique approach to creating personalized diabetes management tools with what we call Adaptive Artificial Intelligence was what 
drew me to join Aspire in the first place.
OK, so what is SoftCGM?
Technically speaking, SoftCGM is a diabetes technology tool that utilizes “sensor fusion,” which simply means it brings several pieces of related information together to make a prediction, in this case a prediction of current blood glucose values.
This video gives a pretty good intro to what SoftCGM is all about.
We call it SoftCGM because it uses software, rather than a traditional CGM sensor, to make the estimation. The first version of SoftCGM makes its estimation from fingerstick calibrations, bolus and carb information, and continuous heart rate data. The platform is flexible enough, though, to account for an ever-increasing amount of sensors that will be coming to market.
This is all presented in a mobile app?
The app serves as the user portal for SoftCGM, but when you’re talking about multiple algorithms being introduced and optimized, that level of machine learning takes place in the cloud. And with that data being stored and processed in the cloud, it opens up the possibility for all sorts of things, like decision support systems for physicians and CDEs, etc. In many ways, the app is just the beginning.
How does it actually work?
OK, this is going to get a little technical...
What’s really special about SoftCGM is that the BG estimations and predictions are based on models that use machine learning to adapt to each unique individual, instead of the typical one-size-fits-all approach that all T1’s have been used to. SoftCGM can learn how you personally respond to exercise or carbohydrates and make a prediction that’s right for you.
We’re achieving that by actually running multiple personalized models through the app at the same time. We currently have that running in the Alpha (development) version of the SoftCGM app.
Each one of these models has its own slightly unique take on diabetes -- how much impact does exercise have, for instance, or how long do carbs stay in your system?
This would be what a typical history log looks like:
On a regular basis, each model looks at all of the historical data over the past seven days and scores itself according to the MARD (Mean Absolute Relative Difference - standard measure of CGM accuracy).
And then, whichever one scores the highest is placed into action to predict current and even future blood glucose. That personalized model will continue to be in charge until the seven-day look-back declares a new winner. Along the way, the models continuously tweak themselves according to the user’s personal results. So what goes into the app is an algorithm that adapts over time to create a personalized model.
What are we seeing on that last screen with "Adaptive Algorithms"?
That fourth screen is the most boring, but it"s really the most important thing that makes this approach different. What you"re seeing is that the app is pulling from four different adaptive algorithms. Each algorithm is "scored" against its ability to predict MARD over the past 7 days of data. The one that scores the highest is the one that the app uses to predict current and future BG. In this scenario, GeneralT2D is the best performing with the data set, scoring 85.6. Right now, the models optimize themselves nightly and the highest scoring one is "put in the game." As we add more nuances to the app, it will be easy to do things like pull up the model that scores best for exercise when an increase in heart rate is detected or pull up the one that scores best when large amounts of carbs come from the pump or pen. That"s called scenario training and it doesn"t exist for us yet, but in this Alpha version you can see how the concept works -- with personalized models competing to be used. It really is the heart of the story.
Wow, this sounds pretty unique and different from current CGMs, no?
The personalized model approach is definitely the most unique piece; we haven’t seen this approach attempted before. The other comparisons to traditional CGM are more obvious -- no invasive sensor being the primary one.
There are really two key aspects that make SoftCGM unique in the diabetes space. The first is obvious, and that’s that we’re bringing in heart rate data to help determine what blood glucose is likely to do in the future. As diabetics, we know that exercise has a potent impact on BG, but other than educated guesses, there are no reliable formulas -- and worse, what worked yesterday might not work tomorrow. Because we’re using machine learning algorithms that can adapt to each user, the personalized models are able to measure the impact of exercise on BG.
Have you used SoftCGM yourself in Alpha testing?
Yes! We had three Alpha users of the app: myself, one other T1D and another T2D. Just last week, we went into Beta, currently set up with 12 participants. The Alpha results were encouraging -- roughly the same accuracy as Medtronic"s EnLite CGM sensor. To be clear, it"s not an apples-to-apples comparison. Our version requires a LOT more data input at this time, but in terms of a first-pass at accuracy, like I said, it’s encouraging.
It sounds a bit like InSpark"s new Vigilant app… any big similarities or differences that come to mind?
I think Vigilant is super interesting and I will be testing it myself. What we share with them is the idea that different users are looking for different ways to manage their diabetes. And by focusing on doing one piece of the puzzle extremely well, I think they"re looking at the problem appropriately.
Without digging into their product, the key difference I think between their approach and ours is that it appears they have one very good algorithm for predicting lows, and I would suspect it will work very well for some people and less well for other people.
Not to mention that if the algorithm works well for me today, what happens when something major changes with my metabolism -- like if I start exercising or get the flu, etc. Those types of algorithms often break in given scenarios.
Our underlying technology is based on multiple algorithms, so we could actually (if they let us) take their algorithm and tweak it for the individual person and their individual scenarios. As we all know, there are times when the math that all diabetics use doesn"t work for us in a given situation. We"re trying to fix that.
Vigilant apparently didn"t require FDA approval. Will you need that for SoftCGM"s unique use of algorithms?
Absolutely, but what that approval might look like is very much up in the air this early on. For instance, the current Alpha version in my hands predicts blood glucose into the future. How the FDA feels about that -- and how we present that data -- will certainly have impact on the process and the product.
Does this have closed loop/Artificial Pancreas potential?
There’s a potential for adaptive artificial intelligence to be used wherever truly personalized medicine is the goal, and a closed loop system could likely benefit from such an approach. But there are just as many potential applications outside of the high-tech AP population, because it’s a personalized approach.
What’s the timeline on this?
We’re looking at having two small Beta tests this summer. The results from that should be sufficient to have discussions with the FDA.
How can our D-Community get more information or get involved if they"re interested?
People can sign up to be part of the feedback process directly online. Like every product of this nature, sometimes we’re looking for Beta users and sometimes we’re looking for feedback from specific subsets of users. But the Alpha version of SoftCGM was built with phenomenal insight from a group of T1s who attended a webinar we hosted, so user feedback is absolutely critical to this process.
Very exciting stuff, Marcus! Thanks for everything you do in helping develop these innovations, and we look forward to seeing SoftCGM materialize.
Disclaimer
This content is created for Diabetes Mine, a consumer health blog focused on the diabetes community. The content is not medically reviewed and doesn"t adhere to Healthline"s editorial guidelines. For more information about Healthline"s partnership with Diabetes Mine, please click here.
Type 2 Diabetes TreatmentType 2 Diabetes Diet
Diabetes Destroyer Reviews
Original Article
#DiabetesMellitus
No comments:
Post a Comment