Why we are regularly wrong in the duration of insulin action (DIA) times we use, and why it matters…

Once you start using a hybrid closed loop system, whether it’s OpenAPS, Loop or AndroidAPS, or a commercial offering such as the 670G, you have an inbuilt feedback loop that rapidly identifies areas where what you thought you knew turns out to be, well, just wrong. And it appears that this is one of those areas.

When we set up our bolus advisors, whether they are on a pump or on a blood glucose meter with a bolus calculator, such as the Aviva Expert or Freestyle Libre, we are asked to enter a “Duration of Insulin Action” (DIA). So we do. What’s yours? For a long time, mine was three and a half hours. It was wrong and it has consequences.

Now, this isn’t a new topic, indeed, in 2010, John Walsh, the eminent writer of the “Pumper’s Bible”, Pumping Insulin, et al, published a paper entitled “Confusion Regarding Duration of Insulin Action” and it’s something I feel that all insulin users and their Healthcare Professionals should read.

Insulin Duration – some of the ins and outs

Firstly, when we look at the clamp studies that are published for various insulin types, we are not looking at the same thing as a DIA as entered into a pump. Let’s take a recent example:

This is a mean insulin profile graph, as provided for Fiasp (and NovoRapid) from the EMA submission. The normal mechanism for generating these graphs is to use a glucose clamp, and the insulin used in a glucose clamp is used to cover both basal and bolus insulin. As a result, the graph above is based off something like the below:

Whereas DIA is usually taken as the lifetime of the insulin over and above the basal delivery as illustrated in the below picture:

As this shows, DIA is inherently longer than IAT, so if we are using clamp study data to inform our selection of DIA length, we are likely to be kidding ourselves, and may well be out by 50%.

Adding in to this information is the data in the original picture. That is taken over five hours, and demonstrates very clearly that at 5 hours, using the mean, there is still a measureable amount of insulin absorption taking place.

If we review the EMA submission document for Fiasp, it states that on average, the half life for subcutaneous use with Fiasp is 57 minutes (which incidentally, is stated as being the same as Novorapid, and challenges some of our assumptions about the relationship between the absorption curves and IOB). Taking this, we can generate an approximation of an Insulin on Board graph based on half life, (accepting that we don’t include the period of absorption, so is in reality decaying more quickly than we would physiologically see) which, for one unit, looks like this:

I’ve run this simulation over 10 hours, but as you can see, with that length mean half-life, at 5 hours, there is still 2.6% of the insulin in the system. This may not seem like a lot, and on a single unit, it is only 0.026u, however, if you have dosed for a 120g carb meal and your IC ratio is 1u:10g, it becomes 0.31u.

This might not seem like a lot, but it’s enough to have an effect on your body, if for example, you are doing exercise that increases insulin sensitivity. To put it another way, if we are using the standard, starting values for correction factor (CF) and IC ratio (ICR) of 1u:3mmol/l and 1u:10g, 0.3u is enough to drop you 1mmol/l. And that’s five hours after eating.

If we make an assumption that someone has 20% faster insulin absorption, effectively 45 mins (and who knows how realistic this really is – my gut feel is that this is a large deviation from normal), whilst at five hours there is less available insulin (1% as opposed to 2.6% in the average), the tail is still having an effect.

Based on the half life data, we can see that short DIAs don’t make a lot of sense.

What are the consequences?

Whilst using MDI manually, this generally has less of an effect, but once you start using a bolus calculator or hybrid closed loop, the effects are compounded. The algorithm is making decisions on how much to dose to bring you back in line. If you have told it that your DIA is 3 hours, when in reality it is much longer, unless you are eating very low carb where the incremental amounts are generally tiny, you are likely to be stacking a reasonable amount of insulin, and the algorithm doesn’t know about it. That’s a recipe for an unexpected low. And that’s it. That’s really the consequence. Consequential lows.

Why is there this disparity?

I suspect there are a few reasons for this. John Walsh’s views are that a misunderstanding of the pharmacokinetics (PK) versus the metabolic effect, driven by the way information is provided in in the in-pack inserts and tied to the massive range of available options in pumps and calculators drives a lot of this. I think this is part of it.

In addition, and something that plays a key part is how we recommend to test this type of thing. The nature of the tests gives us incomplete information.

How do we test DIA?

We generally eat carbs to take us up to a stable given level (say 10 mmol/l) then apply a correction dose based on our correction factor that will bring us down safely (to say, 4 mmol/l). As previously mentioned, if your CF is 1:3, then you’d apply 2u of insulin to do that, and at that dose level, the effects once you get beyond around 4 hours (depending on CF values) are basically unobservable, so they get lost in the overall noise of a physiological system.

I mentioned in previous posts about Fiasp that under small doses, there was no noticeable tail, but at larger doses I thought it was there. More experience of it and looping clearly identified that there was more of a tail than I was accounting for, and as a result, I now have my DIA set to five hours.

Is there anything else?

With additional users also seeing similar effects to those that I saw, it has led us to do the work to update the insulin curves in OpenAPS to take this into account. This is undergoing testing that is showing a notable improvement in “late in DIA” low glucose levels as there is more insulin effect being attributed later in the IOB curve.

It has also led to us questioning the value of using DIA as a measurement for IOB at all. The equation for establishing the amount of insulin on board based on either half life, or the area under the curve of activity is not overly complex, and is a much better approximation than requesting a user input an unrealistic DIA. A better option may be “time to peak action”, as that is generally more easily observed than true DIA (although it comes with its own set of foibles).

What should I take away from this?

If you use a bolus calculator in any guise, whether that’s a pump. a loop or a bolus calculator on a meter, go away and take a look at it. It’s unlikely that any value less than four hours is realistic, and in reality, anything less than five is also likely to be unrealistic. If you’ve been having hypos that you can’t explain, and you have a short DIA, there’s a high chance it will help.

As a consequence, you may need to review your correction factor/insulin sensitivity factor and IC ratio as well.

The key thing for me is that we had a lot of data about DIA, but very few of us were really aware of it and the consequences of it. But the realisation for me of the last couple of months is that for most of us, longer is better.

As always, if you’ve been using Fiasp, please fill in this form to let others know what you’ve found, and of course, other people also use Fiasp, and they’ve shared their experiences. Those can be found here. The more people that complete the form, the better the picture we have of user experiences, so please go ahead and share!