Agreed about needed to experimentally determine whether more repeats vs. more levels is the better approach. Regarding noise in the data, in my previous attempts at this, I also saw a huge amount of variability. I’m hoping the fasting before each measurement will reduce noise by washing out any effects from the previous days food, medication, and exercise, but I’ll have to see from the initial measurements.
Regarding noise from the sensor, for the last few days, I’ve been wearing a Libre while continuing to use my normal BGM. I’ll do a full analysis once I’ve finished with this sensor, but a preliminary look indicates that the Libre has significant gain and offset errors vs. the BGM that can be calibrated out. My plan is to continue doing both measurements to enable this.
I personally didn’t compare the Libre to a standard BGM, but I did notice that the offset seems rather variable post-meals. So it’ll be interesting to see what you find!
Admin note: @skaye I’ve adjusted your Discourse default trust level settings for new users. You are now a “regular user” and should be able to link freely to whatever you want.
Definitely still OK! I’d like to convert our plan to be more supportive of ongoing joins. We have until July, after all. (And beyond that – I certainly wouldn’t want to stop something active, let’s see how this goes!)
It’s wonderful to have you join, and I’m really sorry about the forum spam issue!
Instrument calibration
It’s great to know about whether the calibration error is simply a constant error over the lifetime of a given device! That was my impression when using it. (I didn’t do much with my data, I fear – but I did manage to get the device when I had gestational diabetes last year.)
I’m only on my first sensor, but so far it looks like there’s a big change from day 1 to day 2, followed by a slow improvement in accuracy over the next several days. This is consistent with the published literature (1, 2, 3), which also indicates that there is more substantial sensor-to-sensor variation. It’ll take a while, but I should be able to confirm that as I keep using the Libre.
Definitely. I didn’t mention it in the original post, but one of the things that got me started on this experiment was the observation that adding apple cider vinegar to my recipe for oat-fiber muffins caused my blood sugar to spike (tentative conclusion, haven’t done careful enough experiments to be certain). Since apple cider vinegar has no carbs, this suggests that the vinegar is reacting with one of the other ingredients, likely the oat fiber, to make it digestible. Once I nail down the effect of the individual ingredients, I really want to test the effect of combinations, cooking, etc. Hopefully by then I can get some more people to join the experiment, otherwise this is going to take a really long time…
Ah! That’s really interesting. It sounds like a potential counterpoint to this review of vinegar indicating it might help reduce a blood glucose spike. (Presumably when taken separately, and not within a recipe. I’m guessing you’d already seen these claims and might have tried it for that reason!)
I’ve gone ahead and moved this topic to that category. We can hide the redundant post if you’d like!
I had seen those claims, but I was actually adding the vinegar to mitigate a slight bitter taste coming from the oat fiber. It did that perfectly and without any noticeable vinegar taste, but I’ve stopped using it until I sort out whether it’s actually causing the blood sugar spike.
Just finished analyzing the baseline data (details here). No window of time was perfect, but it looks like starting at 12p is my best option.
Given that, for all subsequent experiments in this study, I will fast starting 7p the night before and start the measurement at 12p.
Two other quick updates:
Did a preliminary analysis comparing fingerstick measurements to the Libre (full report next week). Using the raw, uncorrected Libre data can be way off (max error >40 mg/dL), but the majority of the error is a systematic linear gain/offset error that is consistent after the first day. The random error of the Libre is only 8.1 mg/dL vs. 3 mg/dL for my fingerstick meter. With all that, once I correct the Libre data for the gain/offset error, it’s pretty decent. Not good enough for optimizing my medication, but great for filling in gaps between fingerstick measurements and for see large effects (like the food effects we’ll be measuring in this study).
I did the first dissolved glucose experiment last Friday. Started with 4g and saw an 11 mg/dL rise, which is a good low-end, but detectable signal. I’ll be running different amounts MWF this week and will hopefully be able to get a good dose-response curve.
The analysis & calibration of the data from my CGM is more complicated than I expected, though extremely interesting. It’s going to take me another week or two to get it written up.
In the meantime, I have the results from the first ingredient, dissolved glucose (details here). Data was cleaner than I expected and showed a max BG rise of 6.7 mg/dL per gram of glucose, which is pretty close to what I’ve seen historically.
Just posted results for the second ingredient, allulose, a sugar substitute with physical properties similar to table sugar and rapidly growing in popularity for low-carb cooking (details here). The effect of allulose on my blood glucose was negligible, ~0.1 mg/dL per gram of allulose, or about 1% that of glucose. Even that could easily be experimental error (I can’t consume more than 60 g allulose in a sitting without causing gastrointestinal distress).
Next up, oat fiber, a supposedly zero calorie, zero digestible carbohydrate flour replacement.
Just posted results for the third ingredient, oat fiber, an indigestible fiber made from grinding the hulls of oats. It’s a great partial replacement for flour in low-carb baking when you don’t want the increased calories of almond or coconut flour.
The effect of oat fiber on my blood glucose was even smaller than that of allulose, clocking in at <0.05 mg/dL/g(oat fiber), or <0.5% that of glucose.
Next up is whey protein, the most common protein supplement. I’ve already done the first trial (30 g) and preliminary results are quite interesting. While the peak is ~1/10th that of glucose per gram, it extends about twice as long, indicating much slower digestion.
As expected, olive oil had no measurable impact on my blood sugar. Whey protein isolate, on the other hand, increases my blood sugar by ~20% that of glucose (by iAUC), but with a slower rise. This result sin a lower peak, 0.68 mg/dL/g(whey) or 10% that of glucose, but a long tail of increased blood sugar, ~0.4 mg/dL/g(whey) @ 4.5 h.
This may be do to giving my body more time to produce endogenous insulin, or even directly stimulating its production, reducing the peak blood glucose. Both of these effects have been reported. Given that, it would be useful to see the same measurements in someone with Type 1 diabetes, who does not produce endogenous insulin.
Still deciding what to try next week, but am currently thinking either corn starch (to have an example from each major macronutrient), resistant starch (fiber with disputed claims to non-digestibility), or combinations of protein, fat, or fiber with sugar.
Ended up doing resistant wheat starch this week, mostly b/c I was curious if it was really as low carb as claimed. From monday’s data, it looks like it has 25-35% the effect of glucose, vs. the 12.% claimed by the nutrition label, but we’ll see once I’ve got the full data set.
I’ll do corn starch next, though. In addition to completing the macronutrient dataset, it’s the better comparator for the resistant starches.
Contrary to these claims, it increases my blood sugar by 76% that of glucose (by iAUC), though with a slower rise, lower peak (33% that of glucose), and much longer tail.
This is extremely disconcerting, as both oat fiber (iAUC 0.4% of glucose) and resistantwheat starch(iAUC 75% of glucose) are listed as insoluble fiber on nutrition labels, but have radically different impact on blood sugar. Given the lack of clarity and quantification of ingredient lists, this makes it nearly impossible to predict the blood glucose impact of a food without eating it and testing.
Next week I’ll finish out the major macronutrient groups with cornstarch. Still deciding where to go after that, but it will either be more ingredients used in low carb cooking (inulin, erythritol, soluble corn fiber, lupin flour) or mixtures of the major macronutrients (to measure combinations effects.
Overall, the experiment was successful, yielding a reliable measure of the impact of the major macronutrients and my most commonly used low-carb ingredients. I also got preliminary insight into the interaction effect between glucose and insoluble fiber.
Due to external circumstances, I didn’t get to as many ingredients as I’d have liked. I may come back and do further experiments. I’m particularly interested in testing a wider range of purportedly low-carb ingredients and diving deeper into interaction effects (maybe looking foods with a range of different carbohydrate:protein:fat:fiber).
That said, these experiments were very time consuming and the frequent extended fasts were disruptive to my normal routine. I need to either find a simpler/easier experimental protocol or get more people to join in to accelerate data collection.
My routine has changed a lot due to working from home, changing doctors, and changing medication (due to insurance requirements). Plus, I was able to get a Dexcom G6 CGM, which is showing accuracy comparable to my blood glucose meter. Blood sugars are still good, but I think I can get them better.
Re-measure blood sugar impact of glucose and insulin; compare to previous data
While working from home, I’ve gained some weight (and hopefully muscle). This has resulted in a change in my insulin sensitivity. Not huge, but I need to remeasure to have an accurate baseline for future experiments.
Test the effect of some dietary supplements that have been reported to affect blood sugar in the literature, but where data insufficient or contradictory
Vitamin C (reported to cause blood glucose meters to read higher than actual, but all measurements I can find are for injected vitamin C)
Glutamate (reported to reduce post-prandial glucose, but magnitude and timing vary widely)
Others tbd.
As always, if you have any questions, comments, or suggestions for future experiments, I’d love to hear about it.
Hey @skaye, it’s amazing to see all of those results, really well done! Will we see you by any chance on one of the upcoming self-research meetings? I know that time is short for all of us so no worries if it’s not possible, but we’re still meeting every Thursday at 10am PDT if you find some time in the upcoming weeks
One minor thing I noticed: The external circumstances link doesn’t seem to work as it goes to the blogger edit interface
@skaye this is awesome, and I hope it gets picked up and seen/shared!
In particular, the finding you’ve made for “insoluble fiber” seems like a really big deal. Do you have any sense on whether that’s likely to vary or not from individual to individual?
(I know that question applies more generally, but it seems like there’s some broad implications about diabetes management & nutritional labeling here.)
