I’d been using an Accuchek Insight for getting on for five years, starting in 2020 with a hand me down, then obtaining one of the last ones that were officially available when my last change came up.
Given my AID preferences, the obvious replacement was the Dana-i. That precipitated a change in cannula, having used the Accuchek Flex for the past five years.
I went back to the Inset II, something I was very familiar with from many years of Medtronic use with their equivalent set.
Then I started to look at my data, and something had changed. A significant insulin dose reduction. But why?
What had I observed?
The first thing I noticed was that I seemed to not need to replace the insulin in the cartridge as frequently. This makes sense naturally, as the Insight has a 1.6ml cartridge by volume, whereas the Dana is 3ml, but rather than the replacement time doubling, it seemed to be increasing by significantly more than that.
So I started to dig, and in the digging, something unusual appeared. It looked as though my Total Daily Dose (TDD) had reduced. I should caveat here that we’re looking at a total dataset across the two periods of around 100 data points, so it’s not a large set, but it seems interesting nonetheless.
Averages
The averages in the table below show the two months pre-change (at the beginning of August) and the period so far since the change.
June and September look very similar when we view the means, however, July and August appear to be very different (with August averaging out at 70% of the previous month’s TDD).
A deeper statistical dive
Averages don’t really tell the whole story, so I’ve plotted the TDD data in a box and whisker chart. This highlights something rather different.
Pre- and post-change
What are we looking at? Here’s a quick description.
A box and whisker plot allows us to visualise the spread and shape of a dataset by highlighting its key summary statistics. Here’s what it shows:
Median (Q2): The line inside the box marks the middle value of the dataset, splitting it into two halves.
Quartiles (Q1 and Q3): The bottom and top of the box represent the 25th percentile (lower quartile) and the 75th percentile (upper quartile). The box itself shows the interquartile range (IQR), which contains the middle 50% of the data.
Whiskers: Lines extending from the box that usually represent the range of data within 1.5 × IQR of the quartiles. They indicate the spread of the bulk of the data.
Outliers: Individual points plotted beyond the whiskers, representing unusually low or high values compared to the rest of the data.
Referring back to the chart, we can see that after the pump change, all of the markers have reduced. The median and mean are both a lot lower, while the interquartile range has also tightened.
There are clearly outliers, but as long as they are explainable, then we can live with them, and in both datasets, they’re explicitly linked to excess pastry or candy consumption, or to cannula-related issues.
Let’s remove those outliers and look more specifically at the range we’re interested in.
This chart makes it easier to see the changes. The interquartile range after the pump change appears to be significantly tighter. Overall, the level off variation is much lower.
Month-by-month data
In the averages, we saw that June and September appear to have very similar means, however the larger dataset box and whisker charts show large differences pre- and post-change. Is there something going on month by month?
The following chart breaks the data out at that more granular level.
This chart shows some interesting data. It demonstrates pretty clearly that July was a bit of an outlier in terms of averages being higher than the rest of the time, and that August was lower.
What it does show is the significant reduction in interquartile range during August and September, after the pump change. This is effectively a reduction in the variation of TDD on a day by day basis.
Is there an explanation?
It would be a surprise if there wasn’t! But let’s explore what this particular pump change means and see if it’s possible to unlock one.
First, I needed to confirm there were no behavioural changes that might have driven this difference. I reviewed activity, sleep and medication data.
- Activity: pre-pump change, my average daily step count was marginally higher than post-pump change, but not enough to be statistically significant.
- Sleep: almost no difference here, with pre-change having around 4% more time asleep than post-change.
- Medication: again, no differences at all in either drugs or dosing.
- Work: no changes. Same hours, same job, nothing outstanding.
- Food: No changes to what I eat or when I eat. Consistently varied.
On the behavioural side, there was nothing that immediately indicates any change. So the next step had to be to look at the pumps.
Pump differences
The first potential hypothesis for this difference is that the four-year old Insight was faulty in some way, whether linked to the delivery mechanism, measurement system or something else.
There are ways to confirm pump functionality and I think we can discount this hypothesis on the following basis:
- There was no sign of leakage in the Insight
- The observations of AAPS regarding insulin use independently matched the rate of consumption of insulin within the reservoir
Which leaves us with functional differences. The key difference is the rate of insulin delivery. I used the Insight on the standard rate (rate data from David Burren), and have the Dana-i set to Standard as well. These rates are not the same.
- Insight: 12u/min (5 secs/u)
- Dana-i: 5u/min (12 secs/u)
This presents one hypothesis in relation to why TDD varies so significantly. The insulin delivery rate when bolusing (given the heavy reliance on Super-Micro Boluses in my set-up) may be a player.
Cannula differences
As I mentioned at the start, the pump change necessitated a change of cannula type, and I switched from the Roche Flex to the “Sooil” Inset II. Whilst the set designs are different, I can’t speak for the materials involved. Perhaps more importantly, both are made by Unomedical for the respective manufacturers.
Are there material differences between these sets? Not really. Both are “soft” and both are 6mm long.
User experiences
I asked ChatGPT and Gemini to try and consolidate studies and user feedback from across the internet to see if there were any trends. Both returned slightly different responses, with one highlighting more favourable reports on the Inset, and the other taking the opposite view. Neither could find any real studies, suggesting that both sets have similar feedback from users.
What ChatGPT did highlight (and provide sources for from manufacturer specs and third party measures) was the guage (diameter) of the cannulae. Roche’s Flex was stated as 25G (0.5mm) and the Inset II (Mio) was stated as 27G (0.41mm).
Design and materials
With regard to materials, I asked ChatGPT to deep dive into regulatory documents associated with both sets. While the data on the Inset II explicitly called out PTFE, the Roche filings explicitly mentioned using Unomedical to manufacture the cannulas that ” use components and manufacturing processes … used for other commercially available infusion sets”.
It seems reasonable to infer that the sets themselves use the same materials.
Whilst this highlights limited difference between the two cannulae, it doesn’t completely remove the possibility that they are the cause.
Where do we go from here?
We have two hypotheses about the driver of the change in TDD (whether that’s overall volume or reduced variation):
- Changing the cannula has resulted in the observed changes;
- Changing the pump delivery speed has resulted in the observed changes.
Given that these could both be the cause, I think we need a further experiment. And I have a proposal.
At pump change time, I retained a couple of boxes of the Roche Flex cannulas. It would be very straightforward to swap back to the Insight and instead of using it on the “Standard” bolus delivery setting at 5s/iu (12iu per minute) I will change the setting to “Slow”, which is 10s/iu, or 6iu per minute, close to the Standard offering from the Dana-i.
Come back and see what happened when that test is complete, sometime in early November!
Very interesting insight into the pump and cannula change. Been using the ‘Inset II’ from Medtronic and now the same thing with the Dana-i for years (and years).
I had resistance with a 6mm cannula and switched to 9mm which made a big difference on absorption, but was using 6mm for 12+ years before that (maybe ageing has meant thicker skin….or a bit more fat).
I ‘think’ I noticed a reduction in units-per-day on the Dana-i as went from Medtronic delivery to Dana slowest possible to reduce cannula/skin damage and seemed to drop about 10-20%. That might suggest that slower delivery causes less pooling of the insulin which is somehow linked to a higher ISF.
Sadly didn’t capture the data at the time as was going from non-looping to setting up APS (which was a long-haul)
What about carbohydrates ?
I perhaps forgot to mention it in my “What might have changed” section, but the answer there is “No changes to food”.
That has now been updated.
Same change here, and same effect. Switching from insight to dana i (with orbit) My child need less insulin. Pump vendor told me we would probably need 10 to 20% less insulin and he was right (around 11% less if I remember well). He said that’s because of more accurate delivery and should also impove TIR. But with a growing stubborn child, I was unable to confirm the later.
Thanks for the feedback. Very interesting.
The question is “What do they mean by more accurate?”. If the dosing sizes were that far out, the stated delivered insulin would be different to the life of the reservoir (essentially, you measure delivery based on how long the reservoir lasts for compared to volumes supposedly delivered). When the two values correspond, which in my assessment here, they seem to, I’m not clear how one pump’s delivery can be “less accurate” than the others.
Maybe the insight deliver a little less than asked. And in the ends it count 40ui a day instead of 36ui realy delivered. 3 days later, that’s 12ui, and it’s difficult to see in the pump cart.
Or it may be more about the pump being able to deliver the same amount when asked to. Ex: I want 1ui, if I sometimes got 1.05 sometimes 0.95, it makes things more complicated to adjust dosing. So your TIR is higher and your overall insulin needs increase.