Thanks for the document! That’s awesome! One important thing one could add: Not only are the modalities for getting the sensors different in different countries, they are also geo-coded!
I tried to pair my freestyle libre sensor with the iPhone app to read out the data, and initially it claimed to not find any sensor! The problem was that my sensor was bought in France and my iPhone still ran on the US iTunes account, giving me the US version of their app!
The only way to get around this was to change my iTunes account to the German store (as that’s where I had a bank account available) and then download the German/EU version of the app. This led to the sensor being correctly identified and paired. So beware, sensors might not be compatible if you’re working around multiple countries!
I met up with @MichaelR this morning for breakfast and we had a chance to chat a bit about this self-research experiment. One simple but interesting question would be if/how the blood glucose response to the same kinds of food changes between and inside individuals.
For that we could prescribe the consumption of certain food items and monitor the blood glucose response that follows. By repeating the consumption over the monitoring period we could get replicates for the individual, allowing to see the variation for a single person. By having more than one person do the experiment, we could also see the variation between people.
@liubovv Does that sound like something you’d be interested in as well?
Next to-dos for doing this would be: Coming up with a protocol (which kinds of foods, how many repetitions, spacing of them, etc).
But I would add here another question, why I am interested in this topic:
if I understand correctly there are different factors, which may influence glucose level decay (e.g. external factors not just food right?).
So I would be interested to study how one could dissect the influence from food and other factors on this decay.
Why this is interesting? It is known that people with type 1 or 2 diabetes under mental stress generally experience an increase in their blood glucose levels. When you’re under physical stress, your blood sugar can also increase. So question is what and when one could actually see?
Or this is a bit difficult? @gedankenstuecke
I see there could be some overlap with stress-level-app.
Which are the best apps for stress control? @MichaelR@gedankenstuecke
I think that could be interesting as well, same for the influence of physical activity etc. But it certainly would be adding complexity to the experiment. I wouldn’t know a good app for stress recording, but maybe @Agaricus has an idea!
There are an interesting set of questions. For a personal research project in which we hope to gain some immediate insight, I think food+glucose is most efficient approach. While there is doubtless some influence from stress, sleep, and other factors, the question is whether a clear pattern can be seen despite these influences. I’d be more likely to participate in a simple project like this where we focused on understanding the measurement protocol, building confidence through some calibration with finger-prick measurement, and running a few A-B-A-B trials with specific foods. While the questions we can answer with this approach are very simple, the benefit is high, because we develop confidence in our method. I think adding additional data types, especially ones with their own measurement challenges, such as stress, would be for a later stage.
Specifically on stress, in my stress tracking I’ve found it useful to track only “episodes of high stress.” By focusing on incidents of high stress I avoid the hard questions about calibration of my stress measures. (Is a “2” today the same as a “2” yesterday? Who knows?) Episodes of high stress are easily detected, and for me they are the phenomenon of key concern. I don’t really care about more subtle fluctuations in activation.
While the questions we can answer with this approach are very simple, the benefit is high, because we develop confidence in our method. I think adding additional data types, especially ones with their own measurement challenges, such as stress, would be for a later stage.
Thanks, that was also my feeling. It’s probably good to start out simple for now and then complicate things at a later point once there’s some experience from the first monitoring!
If you live in Seattle, my friends and I have launched a new business to test this. We’ve partnered with a clinic (to give you the CGM prescription) and a food delivery service (to deliver precise, known quantities of fully-prepared food). “A 10 day clinical trial on yourself”. We’re not ready to publicize this widely, but I’d appreciate any feedback from OpenHumans friends. https://personalscience.com
That’s a really cool idea and I’d love to give that a try! Unfortunately all of us that are interested in giving it a try are based in Paris, so shipping ready-made food from Seattle isn’t an option.
Hi Richard, I’m with Bastian on this – the learning from blood glucose part is really interesting, both access to the device and, perhaps even more important, access to your help and knowledge. As feedback on the offering, I’d just say: that’s something I’d pay for! But the meal delivery and the associated cost puts it out of reach for me.
That’s a really interesting study. Large person-to-person variance in blood glucose response to specific foods (e.g. one person’s blood sugar might spike with white bread, but no whole wheat and vis-versa). The authors also wrote a book on their work, which I found quite interesting and a very quick read.
@gedankenstuecke, thanks for pointing me to this thread. It’s really exciting to see people interested in using CGMs to study what impacts their blood sugar. I’d love to participate as well.
I’ve got Type 2 diabetes and have recently gotten involved in more rigorous self-tracking and self-experimentation. I’m particularly interested in N>1 self-experiments to try to better understand person-to-person variations and be able to study interventions that are difficult to disentangle background in N=1 studies. Over on Reddit, I’m working with a group of diabetics studying the effect of hot showers on BG (protocol & preliminary). If any of you are interested in participating, we could really use data from some non-diabetics.
In terms of this thread, I agree that starting with food effects on glucose would be the easiest and quickest way to get useful data. I’m open to anything people want to try out, but one of the questions I’m most interested is the interaction effects between different macronutrients. There’s lots of small studies claiming that increasing fiber or fat in a meal slows glucose metabolism and reduces BG spikes, but they’re all observational, small sample sizes, and/or use fairly complex meals with lots of ingredients making it hard to be sure of what’s going on. I’d be interested in doing a study where we isolate the effects of consuming individual ingredients (e.g. pure sugar, protein isolate, oil), then mix and match to get a clean measure of the interaction effect.
Just a thought and as I said, I’m open to trying anything the group is interested in studying.
Yesterday, @MichaelR and I sat down to come up with an experimental plan for ourselves. So far we’ve stuck to an easy experimental plan that should be able to be performed within a 14 day (single FreestyleLibre sensor) period.
Interventions we plan:
After waking up: Eating 3, 4 or 5 cubes of sugar with at least 1h time after taking in the sugar cube before having actual breakfast. For each of the 3 quantities we will get at least 4 glucose response curves over the 14 day period.
For dinner: Having a meal that is either: carbs, protein or carb+proteins. Again, 3 different conditions would mean having at least 4 data points for each category. We’d also shuffle these against the breakfast conditions, in order to not have a fixed correlation with those.
It’s pretty simplistic, but maybe a good starting point? @skaye Given that you’re clearly more expert on this we wondered though whether it might be worthwhile for us all to join forces and follow one shared protocol in order to maximize our learning? Would you be interested in that?
@gedankenstuecke, I’d love to join in! A few thoughts on the protocol:
Sugar cubes weigh ~2.3g, so this would be 7, 9, and 12 g. For a non-diabetic adult, a 75 g glucose tolerance test results in an increase of <50 mg/dL (2.7 mmol/L) after 2 h. If dose-response was linear, that would mean the 12 g dose would produce a 2 h response of <8 mg/dL (0.44 mmol/L), and that’s if you were borderline pre-diabetic. Actual response should worse as sugar cubes are solid sucrose (lower glycemic index than dissolved glucose), dose-response is less than linear for non-diabetics, and you’re unlikely to be so close to the pre-diabetes threshold for glucose tolerance. You might have a shot of seeing something in the 30-60 minute range, but I wouldn’t be confident in that.
If you want to see any effect, I’d suggest: 1) A wider range of carbs, maybe 10, 20, 40g (40 is the amount in one can of soda), and increase from there if you don’t see a measurable signal? 2) Use glucose dissolved in water (available from any pharmacy or online, faster impact on blood sugar and removes rate of eating as variable). 3) Do the experiment at some time other than the morning (your insulin sensitivity is lower in the morning and your liver dumps glucose into your bloodstream. This effect is variable depending on time you woke up, level of activity, and a bunch of other factors which will introduce noise into the experiment)
For this part of the experiment, I’d be able to do the same protocol, but would need to use a lower amount of glucose. Even 10g would raise my blood sugar by 50-60 mg/dL, which would not feel good. I would use 2, 4, and 6g to try to limit myself to a 35 mg/dL rise.
To get an interpretable result from this, you’re going to need to hold something constant (e.g. calories, mass, etc.). What are you trying learn and/or what hypothesis are you trying to test? For me, I’d be curious to test if eating carbs + protein together reduces the glucose response from the carbs (review for dairy proteins) To do this, we’d want to fix the total amount of carbs and protein and pick an amount of carbs that would produce a measurable BG spike.
Again, I’d need to pick a different amount of carbs & protein to stay in the same effect size, but totally doable.
Actually, based on your experience we thought that maybe it should be us joining your efforts! And following this feedback that seems even more relevant!
The range was set rather low, to allow non-diabetic and (pre-)diabetic people to participate in the protocol without putting anyone at risk. But maybe the approach we took is not useful as the response for non-diabetics would be too small? As a non-diabetic I tried some sugar cube eating (or rather dissolving varying amounts of them in my coffee) and could see varying spikes in my data. But maybe it’s more useful to escalate more quickly and ask people to stop increases once they hit a threshold?
We had assumed that in the morning might be best to have a stable BG baseline, not being aware of the other factors that make this a bad setting!
That works too . I’ve started the protocol I posted. I’m currently on the second of my fasting trials to establish the right time of day for me to run the experiments, but it’s looking like early afternoon is optimal for me (probably ~1p, will post data and make a decision by Sunday). You could do the same time, but it would require skipping lunch. Since your blood glucose is a lot less sensitive than mine, I think any time that’s at least 4 hours after a meal and is consistently stable based on your CGM data would be fine.
I think this is the right way to go (escalating dose until you hit a threshold). Responses between people vary wildly (even within diabetic/non-diabetic) and the we have a more narrow window in-between large enough to see above sensor noise, but small enough not to cause symptoms. I would suggest the following protocol:
Start with a dose of glucose that you suspect will cause a measurable increase in BG
If you observe an increase in BG, test linearly increasing amounts until you see a BG response of >36 mg/dL (2 mmol/L) and have at least 3 data sets with measurable BG rise.
If you don’t see a BG response, increase the amount of glucose geometrically (2x each time) until you do.
From this protocol, we’ll get a dose-response function for each person (i.e. BG = f(glucose consumed, t)), which will allow us to look at person-to-person variation, serve as a baseline for comparison with anything else we eat, and a baseline for comparison for any future interventions we want to study.
Yeah, I don’t know how large an effect it is for non-diabetics, but for diabetics, it’s a real pain. Means I need an extra dose of insulin and that I have to calibrate my insulin doses separately based on time of day…
Do you want to just join in for measuring the glucose effect or other foods/ingredients as well?
Are you ok if I try to recruit some others to join in? I think there are some people I’ve been working with on the shower effect study who might be interested and I’m curious to see what person-to-person variation looks like on the more complex ingredients. Segal and Elinav have published several papers showing that BG response to foods varies relative, not just absolute magnitudes. I’m curious to see if it replicates under more controlled conditions and to get a better measure of the effect size.
Great! I’ll send out some recruiting posts over the weekend and reach out specific individuals I think might be interested.
BTW, I finished collecting my fasting baseline (will add report over the weekend). Looks like I’m stable around 12p, so I’ll start the glucose measurements tomorrow with a 4g dose (should raise me ~22 mg/dL). I’ll let you know how it goes.
