Meal Memory is now available for Android!

We’re pleased to announce the release of our first mobile app Meal Memory on Google Play.


Meal Memory takes a new approach to nutrition tracking. Our goal was to create not just a self-tracking app, but also an improved feedback loop for patients. With this approach, Meal Memory helps diabetes patients understand the effect that each meal has on their blood sugar.

Our system records a blood sugar reading and carb estimate when a meal is logged, then asks for a post-meal reading with an alert on their phone two hours after eating. Having these two data points makes it easier for patients to better manage that same meal when they eat it again. It also gives patients a simple way to scan through their eating history and see how often they’re managing meals accurately.

We started with a focus on nutrition after talking with lots of patients. Consistently people said that balancing their meals was their biggest self-management challenge. Nutrition is obviously a big and complicated topic. So we zeroed in a core habit that we saw in our own eating histories, that we are creatures of habit. We have our favorite meals, favorite restaurants and favorite recipes. But remembering the details of each meal, its effects on blood sugar and the best way to manage that food is a challenge. Meal Memory helps solve that problem by making it easy to both record a meal and related readings, but also easily retrieve that information when you’re eating that food again.

Over the last few years, I’ve tried a variety of different ways of self-tracking my own diabetes. Those experiences helped shape our design. First, I realized it’s important to create a system that is fast and easy to use. Meal Memory is effective because it uses a photo to record a meal instead of typing, letting you log a meal in just a few seconds.

Second, we wanted a simple way to give back insights to patients. Too many systems ask patients to enter lots of information without giving enough feedback. We help patients understand the relationship between their decisions about a meal and changes in their blood sugar later. A mobile phone alert two hours after eating does this well. In taking just a few seconds to enter a post-meal blood sugar, a patients can be mindful of their blood sugar and can focus on understanding this one specific meal. I’ve found that improving my own diabetes self-management has come through an accumulation of small wins like this. Over time, it also helped me become more aware of the bigger trends in my readings and diabetes control.

We hope that other patients have similar success with Meal Memory! It’s our first step in providing patients with improved self-management support tools to help us all stay healthy. Please email us with any comments or suggestions at

Posted in Meal Memory, nutrition | Leave a comment

A Year in Diabetes Data

As we close out 2013, I’ve taken one final look back at 2012. I conducted a yearlong quantified self project where I tracked every blood sugar readings, every insulin dose, every meal and all my activity data.  My diabetes control improved considerably as a result, making 2012 the healthiest year of my life.

We’ve visualized this project as a poster (pdf here). With so much data to make sense of, we printed it big. Never before in my 27 years as a patient have I been able to step back and reflect on my health through an entire year’s worth of diabetes readings.


The main image shows 91,251 blood sugar readings from my Dexcom continuous glucose monitor (CGM). Each colored line shows one day. January 1 is at the top, with days progressing clockwise around the circle. Lines grow longer with more readings for that day. Blood sugars are color-coded and grouped based on the reading. I represented the in-range readings as white to make it easier to spot the days where my blood sugar control was not as good. High blood sugars are a colder colors, low blood sugars are warmer.

Along the outside of the circle are dark gray lines showing the amount I ran each day. You can see my marathon training start over the summer and end in late October, followed by the longest line on race day November 18.

I also listed notes for significant days from throughout the year. This is a key idea behind Databetes, giving patients a way to merge medical data and information from “the rest of our life.” Food, exercise, sleep, stress, jet lag, all these things and more have great potential to influence our blood sugars. It’s important to include this information because it helps us contextualize our medical readings. Improving our health is all about moments where we connect the dots, see the cause and effect of these various factors and adjust our treatment accordingly.


Lessons Learned

Self-tracking helped me achieve the best diabetes control of my life, as evident in my A1c results. This blood test is the standard metric for measuring diabetes control. The lower the reading the better. For context, I’ve included 10 years worth of my A1c readings.

On the back of the poster, I looked more closely at my nutrition and exercise data. For me, most interesting was the change in my insulin levels relative to my exercise. Generally speaking, I need less insulin the more I run. The marathon training showed that this is true up to a certain point. My lowest medication levels were lowest during weeks with about 30 miles a week or more of running. Even if I ran more than that, my insulin rates remained about the same. It’s also interesting to watch my insulin rates spike after the marathon when I stopped running for a few weeks.


Databetes Development

This 2012 project started as a personal experiment for me. It was also the basis for my graduate school thesis at NYU’s ITP.

Yet the spirit of this project matches our goals with Databetes as a company. There is tremendous value in aggregating all of a patient’s medical and lifestyle data. For me, it paid off with a big improvement in my health. Yet it was a very tedious effort. I was extremely motivated to do this for a year. We need to make this process much easier if we want to see wider adoption of self-tracking. But as my blood sugars show, it’s worth it.

It’s also important to remember that aggregating data is just a beginning. Improvements in diabetes treatment will also require new methods for analyzing all this data, as well as systems to help facilitate behavior change in patients. We’ll also need systems that help patients get the emotional support they need to manage their condition and make the necessary lifestyle changes.

At Databetes, we’re excited to be working on improvements in diabetes care that start with better use of data and patient-centric design.





I used 3 medical devices, 3 activity monitors and a smartphone to record data. These included:

  • Medtronic MiniMed 522 insulin pump
  • Dexcom SEVEN continuous glucose monitor
  • OneTouch UltraMini blood glucose meter
  • FitBit Ultra activity tracker
  • Nike FuelBand activity tracker
  • Garmin Forerunnner 910XT heart rate monitor with GPS
  • Apple iPhone 4s



I used 10 different software programs, apps or websites to manage my data, feeding that information into custom software we built.

  • Medtronic CareLink
  • Dexcom Studio
  • Google Docs
  • Flickr
  • RunKeeper
  • MapMyRun
  • Milermeter
  • Foursquare
  • OpenPaths
  • Garmin Connect



I tagged 2,733 photos as diabetes-related and uploaded them to Flickr. I used a Google Doc spreadsheet to record all my food and drink, topping out at 3,517 entries. I ate 47% of my meals at home (87% of the time for breakfast, 40% for lunch and 54% for dinner). I ate at 264 different restaurants.



I completed 138 runs, recording 112 in RunKeeper and 26 in MapMyRun. These apps say the mileage totaled 1,128 miles. But MilerMeter (which uses Google Maps) said the milage was actually 993, which I believe is more accurate.



I had 229 Foursquare restaurant check-ins. I also used OpenPaths to record my mobile phone’s GPS data.


The Finish

Posted in Databetes, ITP, NYU, Quantified Self | 3 Comments

Two days on a closed loop system

A few weeks ago, I completed a closed loop clinical trial at Montefiore Medical Center in the Bronx. During two visits, all my insulin dosages were calculated and delivered automatically by a Medtronic system analyzing my blood sugar data. It was the first times since being diagnosed in 1986 that I didn’t have to carb count or make any other decisions about my medication levels. As exciting and encouraging as that was, the whole experience tempered my hope that this approach will replace my current treatment any time soon.

Medtronic’s visualization of their algorithm and insulin activity.

I spent my first night in the hospital in August and returned a few weeks later in September. Each visit began in the afternoon and ran through the following evening.

The trial was run by Dr. Heptulla, the Division Chief of Pediatric Endocrinology at Montefiore. Dr. Heptulla is know for her strong interest in using technology to treat diabetes. This is one of the many clinical studies she has completed, with several more in the works.

Both times, I used Novolog insulin. On the second visit I was also given a shot of Victoza, a drug normally used only on type-2 diabetics. It slows down digestion and helps reduce post-meal blood sugar spikes. The trial’s main goal was to test whether the drug could be helpful in combination with a closed loop system to better control blood sugar levels.

The study used a Medtronic insulin pump and 3 Medtronic CGMs (continuous glucose monitors). Two of the sensors were the new Medtronic Enlite (just approved by the FDA a few weeks ago) and one older model Sof-Sensor. The data was all transmitted wirelessly to a Medtronic Comlink receiver and into a PC running Medtronic’s analysis algorithm.

The pump in the study was the same generation as my current one. The Medtronic CGM transmitters haven’t changed, even though there is a new generation of sensor.

The new Enlite CGM sensors

Medtronic Comlink wireless receiver

Mission control

The pump delivered micro-boluses of insulin (usually 0.1u) when the system sensed rising blood sugars. All other basal rates were turned off. The CGMs, which usually take readings every 5 minutes, were instead generating blood sugar readings every minute. Nurses also drew blood every half hour for 28 hours for a more accurate glucose test in plasma. The first night I learned to lay in bed with my arm extended and the IV exposed, allowing me to successfully sleep through a few overnight blood draws.

0.1u microboluses every few minutes

Medtronic software screenshot. It shows readings from the 3 CGMs, the regular blood tests taken every half hour and the microbolus dosages of insulin.

During the first visit, there were big spikes in my blood sugar after every meal, peaking at about 250 mg/dL. The system then often over-corrected and my blood sugars crashed. I took glucose tablets to treat any blood sugar below 70 mg/dL.

My blood sugars for the first visit. The vertical red lines mark my meals.

Before my second visit, I had been warned that the side effects of Victoza include nausea and headaches. But most previous participants in the study hadn’t experienced any major problems (except one patient who did get sick). I was given the shot before the 8pm dinner the first night. I still had a big spike after eating (the drug hadn’t kicked in yet). Around midnight, my blood sugars went really low. I ended up needing 48g of carbs (which meant eating 12 glucose tablets) to get back in-range (normally I need only about 15g).  But for several hours after that, my blood sugars were near perfect and I needed almost no insulin.

When I woke up however, I knew something was wrong. I was able to eat breakfast (a Boost shake) at 7am. But soon after that I became nauseous and stayed that way until about 4pm. I couldn’t eat any lunch and was only able to eat about a quarter of my dinner. So much for giving the study usable data on how well the Victoza works on meals. But I’m sure there’s value in knowing how many patients can’t tolerate the drug in the first place.


Given how I felt, I wasn’t able to keep track of my blood sugars for the second visit and can’t share them here. But I remained pretty steady most of the second day in part because I wasn’t eating.

Despite my reaction to the Victoza, I’m glad I participated in this study. It’s interesting to experience first-hand the potential of this technology. The biggest problem was the post-meal high blood sugars. It seems like the cause is the double lag in both data and insulin.  The CGM readings are 20 minutes behind current blood sugar levels because it tests outside the blood stream. When insulin is factored in, which takes 20 minutes to start working and an hour to peak, the system is always operating behind the curve.

A hybrid approach that includes pre-meal dosing seems like the most promising alternative to a pure closed loop system. This method has been integrated into some of the tests of the bionic pancreas by the Massachusetts General Hospital team. Patients use an iPhone app to enter that they are eating a small, medium or large meal. The system then delivers a percentage of a standard dosage (I think it’s around 50%) right away. The algorithm takes care of the rest of the insulin delivery and is able to reduce the severity of post-meal spikes. The bionic pancreas is also highly effective because it doses glucagon to treat low blood sugars too.

Looking back on my experience, it’s encouraging that we’re finally making real progress toward an artificial pancreas. I still have my doubts that any of these systems will hit the market within the next decade, but that’s simply my guess. Another possibility is that it will take several generations of hardware, software and medications before all the issues are resolved and patients don’t have to take any action in regulating their devices.

But until we get there, I’m still excited to be working on better self-management support systems. There’s plenty more we could be doing with existing technologies and treatments to help us stay healthy until that cure arrives.

Posted in CGM, closed loop, insulin pump, Medtronic, Victoza | Leave a comment

ADA & IN Conferences

This summer, I attended two diabetes conferences that were quite different in scope and style. I started with the American Diabetes Association’s 2013 Scientific Sessions in Chicago. This 5 day event had 18,000 participants and stretched across a massive conference center. It is centered around presentations on the newest research and studies spanning every aspect of diabetes care. One of my favorite sessions was led by Martha Funnell of the University of Michigan. As an R.N. and C.D.E., she talked about patient-centered care. She spoke about how providers “have been trained that our job is to fix people or change them, but it’s not.” Instead, she believes the right approach is to provide self-management education and support. I couldn’t agree more.

There were also more technical presentations about the accuracy of CGMs as well as latest information on artificial pancreas trials, including B.U. & MGH’s bionic pancreas. I also enjoyed a session on evolving business models and innovation within the industry. James Dudl, Diabetes Lead for Kaiser Permanente, and John Brooks III, CEO of Joslin, talked about ways that they expect the current treatment models to evolve.

As great as these sessions were, most people seem to go to the ADA conference for the networking and discussions that happen outside the events. All sides of the diabetes industry are present. Walking around the exhibition hall, you are quickly reminded that the biggest companies in the space are the pharmaceuticals catering to type 2 patients. Their booths are the largest and most centrally located. As much as I am focused on data, the device manufacturers and software companies take up significantly less space. Plus, due to the 72% drop in test strip reimbursement announced earlier this year by the Centers for Medicare & Medicaid Services, blood glucose meter companies Abbott and Bayer were both no-shows. Rumors are that several companies are looking to get out of the blood testing business as the bottom drops out of their business model.

At the end of the summer, I also attended Insulindependence’s conference on diabetes and exercise. The event was refreshingly different from the ADA conference. As a patient, it was terrific to be able to connect with so many like-minded diabetics. The conference happened over three days and the vast majority of the few hundred attendees were patients (almost all T1Ds). The sessions focused on real-world issues that we face in our daily lives like balancing our nutrition, exercise and devices. For me, it was really inspiring to meet so many hardcore athletes. Everyone seemed to have just finished an IronMan, a marathon, or were training for one.

A few other sessions were also insightful. Tandem Diabetes, maker of the new t:slim insulin pump, gave us a tour of their facility in San Diego. It’s refreshing to see that their company makes design and user experience such a high priority. Another interesting session was by Gary Scheiner, a T1D, CDE and founder of Integrated Diabetes Services. He outlined the ways his company looks for insights from CGM data.

And finally, Dr. Steve Russell’s presentation on “A Bionic Pancreas in the Wild” was fascinating. It’s quite interesting to see how their work is progressing with real-world trials. Their bihormonal closed loop system uses both insulin and glucagon to control blood sugars. Check out Kelly Close’s report on her participation in a clinical trial for great perspective on the system.

A slide from Dr. Russell’s presentation


Posted in ADA, Databetes, exercise, Insulindependence, Uncategorized | Leave a comment

NYU Entrepreneurs Challenge, NYU Summer Launchpad & the Dorm Room Fund

NYU has been very supportive of Databetes these last few months. In May, we were named one of the winners of the NYU Entrepreneurs Challenge’s Technology Venture Competition. It was exciting to make our final pitch to an amazing group of judges that included Fred Wilson and David Tisch.

This summer, we also part of the inaugural class at the NYU Summer Launchpad. For 10 weeks, we received instruction on lean startup principles and guidance on business model development. The program was led by Frank Rimalovski and Lindsey Marshall of the NYU Entrepreneurial Institute.

During the 10 weeks, we were tasked with conducting 100 customer discovery interviews. We followed Steve Blank’s guidance on refining our business model canvas, prototyping and hypothesis testing. All this work was very helpful in more carefully defining who our first customer will be and the steps that will need to follow. About 2/3 of our interviews were with diabetes patients. In talking with so many diabetics, we were able to spot customer archetypes among this patient pool. The process further highlighted the challenges in designing patient-centric healthcare products, knowing that insurers and payers need to become customers later on. We’re very thankful to Frank, Lindsey and the whole SLP team for all their time and assistance this summer.

This summer, we were also excited to be named one of the first investments by the New York team at the Dorm Room Fund. First Round Capital established this “student-run venture firm that invests in student-run companies.” We appreciate both the investment and the guidance as we continue development.

Posted in Databetes, NYU | Leave a comment

A new look at blood sugar readings

My final class at ITP in the Spring was called Printing Code, a graphics design and coding class. For my final project, I took several months of CGM blood sugar data and visualized it in a new way. At that time, I was doing a fair amount of business development work on Databetes. With this project, I was interested in doing something purely for aesthetics, so I went off on this artistic tangent. As a result, I produced these two posters.

October 2012 CGM readings

July 2012 CGM readings

Each displays a month’s worth of blood sugar readings from my Dexcom 7 continuous glucose monitor, which generates a blood sugar reading every 5 minutes. The top of the circle marks midnight. Moving around the circle, time moves clockwise with the morning readings along the right. Noon is at the bottom of the circle, followed by the afternoon and evening readings.

A core element of this visualization is seeing the readings radially instead of the traditional linear approach. I also wanted to try a new approach to categorizing the readings. A normal view of blood sugars look this:

Readings are either In-Range (often categorized as 80-120 mg/dL), High or Low. This time, I had only two categories. I split the difference of In-Range readings at 100 and presented them as either above or below this threshold.

I then converted them to the radial view.

I then stacked all the day views from the month on top of each other with a light opacity. The darker the readings, the more often the readings are in that value range. Outlier days where I have an usual spike in readings can be seen faintly. The poster with the black background from October 2012 shows that I had a few more bad days than I did in July.

I coded these in Processing using the Geomerative library. Big thanks to my teacher Rude Madsen for his help getting things working.


Posted in data visualization, Databetes, ITP | Leave a comment

Catching Up: ITP Thesis & Quantified Self

The last few months have been rather hectic here in NY and I’ve fallen behind on posting. But I wanted to give a quick update on my work with Databetes.

In May, I wrapped up my graduate studies at NYU’s ITP. Databetes was my thesis. Video of presentation is available here. The foundation of my thesis was the yearlong Quantified Self project I did tracking every data point related to my diabetes throughout 2012. This included blood sugar readings (~100,00), insulin dosages, every meal eaten, location data and more. With all this self-tracking, 2012 turned out to be the healthiest of my life. My A1c blood tests improved nearly a full point to the 5.6-5.8 range. I also talked about my work designing and developing a meal tracking mobile app.

In the Spring, I presented at the New York Quantified Self Meetup. This event was held at the digital design firm R/GA, who has done extensive work with Nike and were heavily involved in the Nike FuelBand project. All the presentations focused on exercise. In 2012, I trained for and ran the Philadelphia Marathon. My presentation talked about the changes in insulin rates I saw throughout my training, race day and the month after. I also showed work I did in collaboration with Jochen Wendel, a Ph.D. student at the University of Colorado and fellow T1 diabetic. His focus is on cartography. We generated a Google Earth map that showed my marathon route that is color-coded based on my blood sugar readings. Jochen has done this type of work on his own blood sugar and mountain bike data. Geomedicine is an interesting and evolving are of focus, as noted in this article.

My marathon route, color-coded based on my blood sugar readings



The Google Earth map is interactive, allowing you to click on any point along my marathon route to see the data. I added a red arrow to this screenshot to point out my blood sugar readings from my Dexcom and OneTouch (listed as Medtronic) meters.

These two pieces of work highlight the fact that increased patient engagement with their data can lead to improved health outcomes. For me, this process was quite tedious, keeping track of all this information. With Databetes, we are developing simpler ways to archive, analyze and recall this data in a way that is easier for patients to integrate into their busy lives. I strongly believe that tools focused on improving diabetes outcomes need to begin with a focus on patient engagement and education.

Posted in A1c, CGM, Databetes, exercise, FitBit, Fuelband, insulin pump, ITP, Nike+, Quantified Self | Leave a comment

Activity Apps: Same data, different results

Recently I started using the mobile app Moves. It uses the sensors in your iPhone to measure your activity in the same way that devices like FitBit do, but without the need to buy and carry an additional advice. So far, I have been really impressed with the app. It can distinguish between walking, running and cycling. It also displays the information in a nice, simple way.

Most surprising however has been comparing its results with RunKeeper, which I activate when I run. Moves on the other hand is running in the background all the time. The two apps obviously use the same GPS data from the same phone, yet consistently reach different results about the length of my runs. My base run these days is a 5 miler near my home. Moves has been consistently pegging it at 4.9 or 5 miles, while RunKeeper usually pegs it anywhere between 5.25 and 6.5. It’s an interesting case for understanding who is to blame for these discrepancies, the software or the hardware.

5 mile run results with Moves

RunKeeper results from the same run

Posted in Uncategorized | Leave a comment

Blood sugar volatility analysis using R

I have just posted on my other blog about my data analysis of the volatility within a month’s worth of CGM readings. Using R, I wanted to explore ways of judging readings by methods other than just average daily blood sugar. This was completed for my Data Without Borders class at NYU’s ITP. The full post is here.

Posted in CGM, ITP | Leave a comment

Philadelphia Marathon

Another marathon down(my third)! And a new personal record! My time of 3:47 at the Philadelphia Marathon was a two minute improvement.

This race is an important point in my 2012 Quantified Self project. In order to best learn from the day, I carried a total of 8 devices with me: an insulin pump, 2 continuous glucose monitors (CGM) (a Dexcom 7 and my newly arrived Dexcom Gen4), a standard blood glucose monitor, a Garmin heart race monitor chest strap and GPS watch, an iPhone, Nike FuelBand and FitBit. I’ll talk more about these later.

As for my diabetes, managing my blood sugars on the day of the race turned out to be more of a challenge than I was hoping for. During my two previous marathons, I saw my blood sugars spike in the hours before the race, both from being anxious and from how i managed my meal dosages. I was determined to prevent that from happening again this year. I had a normal breakfast (an english muffin with almond butter and a banana) very early in the morning at 5am in order to get to the start line on time. I took a dose of insulin that I would normally take assuming I wasn’t about to exercise. I then had another half of an english muffin at 6am. I took a smaller dose for this, hoping that my blood sugars would start going up a bit closer to the 7am start time (they had been at about 145). I also began a temp basal rate at 6am, setting it to 20% my normal rate for 3 1/2 hours.

Waiting for the start along with Ann, my friend and running companion. We ran the Paris Marathon together back in 2010.

Continue reading

Posted in Databetes, exercise, ITP | Leave a comment