With the past 18 months or working on Boost and using it, various patterns have shown their heads. In the past I’ve written about the “hangover” effects of weight training and insulin sensitivity, and this past summer of participating in the 1mn steps challenge has provided an alternative view of how insulin sensitivity works in a relatively sedentary working environment.
I passed the 1 million steps in roughly 75 days, which means an average of 13,333 steps per day. To achieve this, during the week, that meant walking to and from stations in the morning and evening, and generally fitting it in around being at work. At the weekends, it was more a case of continuous low level movement around the locale, house and garden. Again, in general, my weekly steps were split up with weekdays being 11,000 to 12,000 steps and my weekends 15,000 to 16,000 steps.
What I wasn’t expecting was the difference in insulin sensitivity pattern that this generated, which in hindsight was perhaps a bit naive of me.
What follows is an analysis of the data I’ve been observing, with some thoughts on what that might do for AID systems. But if that’s TLDR, then here’s the key point. Maybe the way we tend to look at step data is incorrect. It shouldn’t be about identifying exercise. It should be about identifying inactivity and adjusting systems for that.
If you want to see where this comes from, carry on reading through.
Over the summer, my weekly working period typically resulted in a TIR that looked similar to the two examples below:
Where typically, using the Boost post prandial dosing model, more than 15% of my time was spent higher than 10 mmol/l (180 mg/dl), with the average at a level of around 12-12.5.
On top of this, my average TDD during these periods is around 50iu of u200 Lyumjev insulin, or 100iu of u100.
For many people, the TIR around 80% is still pretty damn impressive, especially where the mealtime dosing is post prandial, and not calculated by carb counting, but the key thing here is the difference it shows to weekends.
The weekends show a noticeably different set of data, with a greater time in range, much lower insulin usage and much lower time above 10mmol/l (180mg/dl):
Insulin usage in these two examples is between 25iu and 30iu per day, again of u200 insulin, post prandially applied, ie 50-60iu of u100.
What we can also see from the data is that the weekend readings have around a 35% lower standard deviation than during the week.
But what do we attribute this to?
Factors affecting glucose levels
We all know of the list of at least 42 things that affect glucose levels from Adam at Diatribe:
In the context of my weeks, Food, Medication, Environment, Behaviour and Biological are mostly the same. The one factor that varies in the Biological list is that at the weekends, the stress levels are definitely lower. On the Activity front, the number of daily steps isn’t vastly different, however, the style of them is.
As the two images below from Oura, showing a typical weekend and weekday, show, that the dispersion (and amount ) of activity is quite different. During the daytime on weekdays, my activity levels are essentially background, with the peaks around start and end of day.
During weekends, the activity tends to be consistent throughout the day with a reasonable level of movement, before settling down for the evening
Given the differences, the more constant movement over the course of the weekend, associated with the lower stress seems to make a significant difference to sensitivity and ability to keep glucose levels in range.
Perhaps more importantly, doing a high level of steps in bunches at the start and end of day doesn’t have the same effects on sensitivity as something like weight training, which depletes glycogen storage in both the muscles and liver, and requires the body to replenish it, leading to greater sensitivity, even with an ongoing increase in stress during work periods.
So what’s your point?
My point is fairly simple. For long term increased insulin sensitivity for most of us, exercise definitely helps, but we need to consider what that exercise looks like. Just walking to and from work is unlikely to be enough to maintain that level throughout the day. Being active for longer periods, or deliberately doing targeted exercise that has a longer term impact is likely to help.
It also comes back to the design of automated systems. One of the discussions that has been an ongoing one in relation to open source systems is whether it’s worth integrating step data into the algorithms, and if so, what you do with it?
The view above suggests that there is a clear reason to do so, but it’s not the received wisdom of identifying exercise and presenting higher targets.
We already know that by the time that happens, with insulin on board it’s probably already too late.
If, on the other hand, we use step data to identify lack of activity, and then allow that to adjust some sort of modification factor that allows sensitivity to be decreased as a result, it might provide better outcomes.
Essentially, something along the lines of:
If Steps per 15 mins is less than x;
Increase sensitivity factor by 1/(x/xmean)
It’s certainly something to think about in algorithm design for both commercial and open source AID systems, and maybe it’s time for a further collection of crowdsourced data to create some models that might address this.