Our blog

The latest thoughts and announcements from Tidepool's team.


Let’s Talk About Your Donated Data

November 2nd, 2017

Over the summer, Tidepool launched the Tidepool Big Data Donation Project, helping you donate your anonymized diabetes data to research. Since the Project’s launch, we’ve been blown away by the community’s response—more than 1,100 datasets have been donated, with more coming in every day.

We’ve been working with Ed Nykaza, a longtime contributor to Tidepool and father of Leah, who has type 1 diabetes (T1D), to help us analyze and understand this data. The results of his work are fascinating!

If you’d like to avoid having to divide by 18.01559 as you read this data, click here to read the mmol/L version.

The Basics

If you’re a Tidepool user, you’re probably familiar with the blood glucose (BG) distribution chart on the top left corner of your Basics screen. To jog your memory, it looks like this:

You may have wondered how you’re doing compared to other users. Well, first of all, you’re one of kind and there’s no one quite like you!

But based on the first 354 Tidepool users who have donated their CGM (continuous blood glucose) data, we can give you an idea of averages across each age group.

Chart of blood glucose analysis from Tidepool Big Data Donation Project

When reading the chart above, it’s important to note that the blood glucose numbers for Tidepool users in every age group tend to be lower than comparable numbers for the diabetes population overall. For more on that, see the “Bias” section at the end of this blog post.

The People Behind the Data

Before we dive deeper, let’s take a moment to better understand the demographics of our anonymized data donors.

Current Age of Donors from Tidepool Big Data Donation Project

Diagnosis Age from Tidepool Big Data Donation Project

Years Living With Diabetes from Tidepool Big Data Donation Project

Looking at donors’ age of diagnosis and number of years living with diabetes creates some really cool groupings.

For example, below you can see a cluster of users younger than 18, then a gap during the college years, then another cluster after college-age, and finally, a surprising number of datasets into the later years.

This sheds light on not just the diversity of the datasets, but also the diversity of the Tidepool community!

Age and Years with Diabetes Distribution from Tidepool Big Data Donation Project

Nerdy Notes About This Data

We’re about to dive into some trends in glucose levels. Before we do, there are a couple nerdy things to note about this data.

  1. “N” is one year of data donated by a Tidepool user. A single user may have donated more than one year of data. For example, a 35 year old who has donated three years of data from when they were 33, 34, and 35 years old. As such, they will contribute N=2 to the 30-34 bucket and N=1 to the 35-39 bucket.
  2. We are using the AACE and ACE consensus of target blood glucose being between 70 – 180 mg/dL. We know each person with diabetes sets their own personal target ranges, that’s why it’s a customizable setting within your Tidepool profile. For simplicity, we’re applying the 70 – 180 mg/dL range to all of this data.

Time in Range

Now, let’s start by looking at average glucose levels and time in range, by age group.

It’s interesting to see how often the different age groups are staying in the target range of 70-180 mg/dL.

Look at the incredible change from the 21-24 year age group to the 25-29 year age group. Those 25-29 years olds appear to make a tremendous comeback, going from 63% to 74% time in range. That’s, on average, an additional 2 hours, 38 minutes in target range per day.

There are so many incredible findings in just these two charts. What other insights do you see from this data?

And how about the datasets from folks between 65-85 years old. 82% time in range? What’s their secret?! (…a possible topic for future blog post, perhaps?)

Time Below Target

Now let’s look at the other side. How much time are Tidepool data donors spending in hypoglycemia?

Data donated by younger adults shows they experience more time in hypoglycemia than other age cohorts.

Specifically, the 30-34 year old and 35-39 year old cohorts in particular spend 5.8% and 5.5% of their time below 70 mg/dL. That’s about 80 minutes per day, on average!

These same cohorts, age groups 30-34 and 35-39, also spend about 24 minutes per day, on average, below 54 mg/dL. That’s a real cost of managing life with diabetes.

Many of us at Tidepool identify with this as our own personal experience. We get it. Trying to keep your diabetes under control is hard work.

Time Above Target

What about the other side of spectrum, time above target range?

If you’re the parent of a 15-year-old with diabetes and have noticed that their BGs seem to be above target more than before, you are not alone.

According to the donated datasets, 12-14 year olds spend on average about 3 hours per day (12.7%) above 250/mg/dL. That number jumps to about 4 hours per day (16.3%) for the 15-17 year olds.

Population Bias

Let’s take a moment to acknowledge bias. Bias is an over- or under-estimation of a population sample that prevents objective consideration of an issue or situation. Bias shows up in our data in at least a couple of ways:

  1. Donors to the Tidepool Big Data Donation Project are self-selecting (they volunteered); they are not randomly selected.
  2. They are also much more likely to be on a pump and/or a CGM than the average person with diabetes, which means this data is not necessarily representative of the general public, especially those not on devices. We know from the T1D Exchange, for example, that folks on pump and CGM tend to have A1C improvements over folks on MDI without CGM.

One place we see this bias is when we compare estimated A1C of the Tidepool sample population (based on the formula: eA1C = (average glucose + 47.6) / 28.7 ) to the chart of A1C by age cohort presented by the T1D Exchange based on its Registry. We found that, while the curves are similar, estimated A1C for the Tidepool sample population is consistently 1.0% – 2.0% lower than that of the T1D Exchange Registry population.

A1c Comparison with T1DX Registry from Tidepool Big Data Donation Project

Our Next Steps

As excited as we are to help the community better understand real life with diabetes, we’ve only scratched the surface with this data. This post only examines one small, but important set of insights based on CGM data. We’ll look at more insights, including those based on insulin dosing, in upcoming posts.

But what about you? What comes to your mind when you look through these charts? Does this data resonate with you? Why or why not? What other questions would you like us to explore next? Leave a comment or send an email to bigdata@tidepool.org and let us know what you think!

Be sure to check out this guide if you’d like to donate your own diabetes data to the Tidepool Big Data Donation Project.

As for us? We’re just getting started.

Yours in big data,
Brandon & Ed

Bring this data into your next project

If you think your company would benefit from collaborating with Tidepool to better understand the needs of people with diabetes, contact us at bigdata@tidepool.org.

If you are a citizen scientist or independent researcher and need real device data to help with your project, you can also email us at bigdata@tidepool.org. We’ll have an exciting announcement for you soon. Stay tuned.

Technical notes on this data

  • Participants donated their data via the Tidepool Big Data Donation Project.
  • Tidepool users use the Tidepool Uploader and Tidepool Mobile to upload their data.
  • Data presented here is based on CGM usage. Tidepool supports both Dexcom and Medtronic CGMs. The data here includes both. Since device makers can be sensitive to comparative studies, we won’t be making a distinction between different device types.
  • Our analysis tech stack includes Project Jupyterpythonpandas and matplotlib.


  1. Wow, this is very cool, what fascinating data! I’m curious if you’ve looked at any potential effects of gender? I’m wondering if differences will be evident around puberty, or if that slight increase in average CGM level from 40-49 may be related to hormones and menopause? This is an amazing resource, and I love what you’re doing with it, keep up the good work!

    • Thanks, Angela! At the moment, we don’t ask users/donors for any information beyond birthdate and diagnosis date. But we are compiling a list of additional questions to ask like the ones you suggest that will make this data even more interesting and helpful to researchers. Thanks for the feedback!

  2. Great work. Interesting to note that the entire curve HbA1c curve on your data is lower by 1% point, wonder if this the impact of just CGM?

    • Christopher Snider

      Tidepool Team |

      November 3, 2017

      A comparison between CGM graphs associated with people wearing an insulin pump and people without an insulin pump would be an intriguing path to explore. Thanks for the great suggestion!

  3. Laurie Schatzberg

    Sensational work, team! It’s exciting to see bigger pix among Tidepoolers and while we can’t generalize and can’t be certain of causality, it sure feels good to be in this “pool” :-). Eager for your updates as your schedules allow.

  4. Some interesting analysis. Pity that Freestyle Libre data is not collected particularly as the Libre software already has a .csv datadump built in that they use to send it to Abbott.

    • Hey Joe, we actually just released Beta support for the mmol/L version of the Libre, and will release support for the mg/dL version as soon as we can get our hands on some for testing.

  5. Lori Schlosser

    Bravo! Stellar work! Easy to follow and packed with insights! On the journey with DS, dx age 8, now a college sophomore. Affirming to see these numbers! Sharing with our support group parents. Thank you!!!👍🏻

  6. Perhaps you could expand your findings and show graphs for average hourly BG per age group in a future post?

  7. This is super cool. Thanks for putting it together and sharing.

  8. Michael Lipschultz

    Some very interesting analysis so far! I’m sorry this comment got so long, I guess you could say it resonated with me because it combines two of my favorite things: my T1 wife and data analysis! I should also thank my wife for letting me talk about some of her data here :).

    Regarding the Time In Range analysis, there appears to be a general correlation between age and better numbers. If there actually is a correlation, I wonder if it’s caused by more experience controlling their BG. Of course, not everyone’s diagnosed at the same age, so I’d probably do the correlation between years living with diabetes and average CGM level. Or maybe age is actually an indicator for something else (stress? free time?) that better-correlates with BG control. I see a lot of interesting directions for investigation here!

    For your insulin analysis, I’d be interested to see whether there’s a difference between insulin brands/types (or are insulin manufacturers also sensitive to comparative studies?). My wife got worried when her insurance required her to switch from Novalog to Humalog, that her BG would behave differently with the new insulin. Luckily, other than a small change in when she needed to take a bolus before a meal, there weren’t any changes.

    Some other analysis that’d be interesting is how much insulin is taken (in relation to carbs consumed) and when boluses are taken, broken down by age. I wonder if it’d be possible with the data you have to estimate the insulin activity curve — I’ve been trying that with the data I have.

    What analysis do you have planned after the insulin work? My wife and I have been examining her data to see whether there are differences across days of the week and between semesters (she’s a college professor) — there are! Maybe for Tidepool, looking at weekday sugar control vs. weekend sugar control could be interesting (or looking at each day individually). Or, looking at how BG responds at different meals. I’ve started using clustering and some assumptions to analyze meal-time data. I wonder if something similar is possible with your data. Although, I guess Thomas’s suggestion above would be nice to do first.

    A small critique/question: What’s up with the y-axis labels on your “Current Age of Donors” and “Diagnosis Age” graphs? The tick marks seem to go out of sync with the labels (e.g. the last bar in the Current Age of Donors graph actually corresponds with 76-80, but it looks like it belongs to 81-85). I don’t think it’s a major problem (the meaning’s still clear), just something curious I noticed (all the more curious because the later graphs don’t have this problem).

    Thanks for sharing this analysis and for developing Tidepool! I’m looking forward to further posts on the topic!

  9. Brandon Arbiter

    Tidepool Team |

    December 12, 2017

    Michael, thank you for your feedback! We have collected a bunch of ideas for what to do next and we’ll be discussing them after the holidays. We’re so excited to dive back in for more!

    I, too, am curious to see glycemic control as a function of years with diabetes. One thing I often think about is that the human brain is fully developed around 25 years old, and this seems to correlate with the steepest drop in A1C.

    With respect to tick marks, if I told you they may have been the most complicated part of this project, would you believe me? 🙂

    • Michael Lipschultz

      Yep, I can believe using matplotlib was the hardest part of any analysis :).

      Good point about the brain development. I wasn’t sure what to make of that noticeable drop at 25-29, but that makes sense.

  10. Larry Martin

    I find it interesting that my lows and highs are opposite for my age group. I am 61 and it appears that most in my age group are out of range on the high side. I am out of range on the low side and while I know that can be dangerous, I think overall it is more healthy. Working on fixing that with a lower carb and higher protein diet. Would also love for Medtronic to get an all in one pump/cgm. Having 2 insertion points is very annoying to me.

Leave a Reply

Your email address will not be published. Required fields are marked *