Back
# Let’s Talk About Your Insulin Pump Data

##### February 15th, 2018

## TL;DR

## Reviewing Donated Data

## Total Daily Dose of Insulin

## Insulin Pump Settings

## Insulin Sensitivity

## Boluses per Day

## Carbs

## Basal Rates

## Basal : Bolus Ratio

## Bias

## Bring this data into your next project

## Technical notes on this data

##
One Comment

### Leave a Reply

With this post, we’ll explore bolus and continuous infusion basal information from people who have donated their diabetes data through the Tidepool Big Data Donation Project. Here’s a quick summary of our findings:

- The analyzed dataset contains data from 803 donors, totaling more than 479 years of insulin pump use.
- Children are using less insulin per day than our older cohorts, as expected, with a general trend of insulin needs increasing until age 18, when the amounts start to level off.
- Median basal rates increase until around age 18, and then stay fairly consistent throughout the older ages between the range of 0.75 to 0.9 U/hr.
- Insulin-to-carb ratios and insulin sensitivity (a.k.a. correction factor) are highly variable at the younger ages, leveling off around age 15.
- Despite high variability around total number of carbs per day across all age groups, median total daily carbs falls between 125 and 160 in older age groups, based on carbs input into the insulin pump bolus calculator.
- Median basal:bolus ratios also increase until age 20 (from around 35:65 to 50:50), and level off close to 50:50 thereafter.

In our last analysis, we looked at average CGM levels across different age groups. This time, we look at *insulin pump use*.

The dataset analyzed contains over 479 years (175,000+ days!) of insulin pump data from 803 donors. All of the results come from data that has both bolus and continuous infusion basal information. The representation of insulin pumps is spread across Tandem (282), Omnipod (217), Medtronic (210), and Animas (94). We will not be comparing across brands.

For this analysis, we’ve excluded data from donors who are using a closed-loop system. For more details on how we conducted this analysis, see the technical notes section at the end of this post. Here is an overview of the data:

As is always the case with these conversations, please keep in mind that your diabetes will vary from what you see here. You may require more or less insulin than the medians we present, and that’s cool. The point of this analysis is to see what we can learn from each other.

Total Daily Dose of Insulin includes all the boluses (that is, insulin for food and blood glucose corrections), as well as all basal insulin delivered. Said simply, it’s how much insulin each person got from their pump each day. Here we’re showing the median (or 50th percentile) daily dose for each age group to get a sense of the general trends.

For the data nerds reading this blog post, below is the distribution of the total amount of insulin per day. Just like the Trends view in the Tidepool web app, we are showing the median (the colored square in the middle of each bar), the distribution of 50% of the data (25th to 75th percentile) in dark gray, and the distribution of 80% of the data (10th to 90th percentile) in light gray. One thing to note: we’ve removed the outliers (the top and bottom 10%) to make the graphs a little easier to read.

We see that the total amount of insulin given each day is highly variable. As expected, children are using less insulin than our older groups, with a general trend of insulin needs increasing until age 18, where the amounts start to level off. We’ll revisit this trend shortly.

As you may know, there are a few key settings on an insulin pump:

- insulin-to-carb ratio
- insulin sensitivity, also known as correction factor
- basal rates

These three settings work together along with the blood glucose readings and any carb counts you enter to determine how much insulin you need at a given point in time.

So how do the insulin-to-carb ratios vary by age?

The first thing to note about insulin-to-carb ratios is that they are highly variable at the younger ages, and decrease until age 12 or so, where they begin to level off. It may be important to note that very young children, in addition to having low insulin-to-carb ratios, may also be using diluted insulin, which this data does not capture. (Side note: We think it’s silly to call this “insulin to carb” ratio, when it’s really “carbs to insulin ratio”, but that’s what everyone does, so we will too…)

We see a similar trend for insulin sensitivity (a.k.a. correction factor), which represents the expected blood glucose drop for 1 unit of insulin:

You can view this chart in mmol/L here.

You can view this distribution chart in mmol/L here.

Let’s walk through what we’ve learned from your food and correction boluses. Remarkably, it’s rather surprising — or boring depending on how you look at it — to see how uniform much of the bolus analysis appears across all age groups. The median number of daily boluses, regardless of type (normal, dual, extended), is consistently 6-7 per day. Remember, the dataset we analyzed is limited to people on insulin pumps. We might assume people using insulin pens bolus less often, but we don’t have the data to prove it. Yet.

Here’s a look at variability of total number boluses per day, across all age groups.

Based on the carb values that you all put into your insulin pumps, we were able to calculate the total number of carbs you eat per day.

There’s a significant increase in carbs around the ages of puberty, and a general trend of the median number of daily carbs decreasing with age. However, the number of carbs per day really varies, especially when you look at the distribution of the data.

We also analyzed the number of carbs given per bolus to see how much food you’re covering with each bolus when carbs are entered into the bolus calculator. Here we see a trend of the number of carbs per meal bolus increasing until age 12-14, remaining around 40g per meal until age 21-24, and then dropping to around 30-32g thereafter.

Here’s the distribution of carbs per meal bolus.

Here’s a look at median basal rate of insulin delivered per hour for each age group. For this analysis we included both scheduled and temporary basals to get a sense of the actual amount of basal insulin given per hour. In addition, we weighted the basal rate by basal rate duration, so the median here is actual a weighted median. See the technical notes for details on that one.

Oh, and if you are curious to know how the median values change when you only look at the scheduled basal rates… well, they don’t! The only change in in the 18-20 year old group. The median value jumps from 0.75 to 0.80 U/hr.

We can also look at the distribution of basal rates. Again, we see a noticeable leap once puberty gets added to the mix, but by and large, the basal rate levels out by the time you reach your 20s.

Finally, here’s a look at the weighted median basal rate duration. For this analysis, we only consider scheduled basal rates and durations.

Most people run their basals for 4-6 hours, which means your active basal profiles have about 4 to 6 basal rates. Though, we do see a lot of variance across all age groups.

After all this discussion about and around your basal and bolus data, where does that leave us? Does it all balance out in the end? Well, that depends on your age!

Here’s a look at the typical amount of basal (top) and bolus (bottom) insulin delivered per day. As we noted in the beginning, insulin amounts increase with age and then plateau around age 18-20.

And, keeping with the other theme in this blog post…there is much variation! In this case, the amount of variation increases with age and levels off around 18 or so.

Many conversations in clinic around insulin usage are informed by the basal:bolus ratio. Some endocrinologists and diabetes educators like you to aim for a 50/50 split, though others prefer less basal and more aggressive meal bolusing. We don’t know of a consensus statement on this, and remember, your diabetes may vary.

This chart shows the median percent of insulin delivery that came from your basals by age group. Like the trend we saw with basal rates, the median basal:bolus ratios increase until age 20 (from around 35:65 to 50:50), and then level off close to 50:50 thereafter.

As we discussed in our previous analysis, there’s always bias to consider. For the datasets analyzed and presented in this post:

- Donors to the Tidepool Big Data Donation Project are self-selecting (they volunteered); they are not randomly selected.
- Because we’re specifically analyzing insulin pump data, the basal and bolus data presented do not include data from people using multiple daily injections (MDI). This, we hypothesize, will skew the number of boluses per day higher than the insulin-requiring population overall.
- A disproportionately high number of Tidepool users are within five years of their diagnosis with diabetes. (See previous analysis). This may lead to a disproportionate representation of people in their honeymoon period (the period soon after diagnosis when the pancreas is still producing some insulin).

What do you think about the results of our analysis? What other questions would you like us to explore? 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.

Yours in big data,

Ed, Brandon, and Christopher

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 academic researcher and need real device data to help with your project, you can also email us at bigdata@tidepool.org. We’re working a program to make datasets available for free or at greatly reduced cost.

- Participants donated their data via the Tidepool Big Data Donation Project.
- Tidepool users use the Tidepool Uploader and Tidepool Mobile to upload their data, and may also connect their Tidepool account to their Dexcom account to automatically sync their CGM data.
- Data presented here is based on insulin pump usage. At the time of this analysis Tidepool supports Tandem, Animas, Insulet/OmniPod, and Medtronic insulin pumps. Since device makers can be sensitive to comparative studies, we won’t be making a distinction between different device brands.
- “Unique Donors in Age Group” is the number of Tidepool Donors that had data in a given age group. Please note, a single donor may have donated more than one year of data, and may have data that spans multiple age groups. For example, a 35 year old who has donated three years of data from when they were 33, 34, and 35 years old, will contribute to the 30-34 age data and the 35-39 age data.
- We removed the top and bottom 10% from all of the distributions to make them easier to read. Most of the distributions discussed are not normal (in the statistical sense of normal distributions); they are skewed towards the right of the median. This is also why we present median values throughout the post rather than mean values.
- For the basal rate analysis we weighted each basal rate by the basal duration (rounded to the nearest hour), so that we were properly accounting for the fact that some basal rates run longer than others. For example, if a basal rate of 1.5 U/hr was used for 23 hours and 0.5 U/hr for 1 hour, the median basal rate is not 1.0 U/hr, but rather would be 1.5 U/hr and is a more accurate representation of this person’s typical basal rate.
- Days where donors were using a closed-loop system (identified as having more than 30 temp basals used in a day) were excluded from the dataset we analyzed.
- This dataset does not consider CGM use. Some donors in this analysis use CGM, some don’t.
- Our analysis tech stack includes Project Jupyter, spyder, python, pandas and matplotlib.

- October 2018
- September 2018
- August 2018
- July 2018
- June 2018
- March 2018
- February 2018
- January 2018
- December 2017
- November 2017
- October 2017
- September 2017
- July 2017
- May 2017
- September 2016
- February 2016
- October 2015
- August 2015
- July 2015
- April 2015
- November 2014
- October 2014
- September 2014
- August 2014
- July 2014
- June 2014
- February 2014
- January 2014
- December 2013
- November 2013
- September 2013
- August 2013
- July 2013
- June 2013

Very interesting. I wonder, whether in a followup article, it would be possible to publish:

– basals by age range by time of day

– carb ratios by age range by time of day

– insulin sensitivity by age range by time of day

On a different note, would you be able to publish information re. time in range/BG average/variances for users of closed-loop systems vs ‘regular’ users ?

Thanks for all that you guys do–I relied on Tidepool until switching to an older (uncompatbile) pump for use with Loop/Nightscout.