Please Critique my Experiment Design: Measuring Effect of Ingredients on Blood Sugar

Tags: #<Tag:0x00007f335494eb28>

link to full post if interested

For my next set of experiments, I want to measure the effect of different foods on blood sugar. I’m particularly interested in the effect of:

  • low-carb flour and sugar replacements (e.g. oat-fiber, lupin flour, allulose, etc.)
  • combinations of ingredients (e.g. how much does indigestible fiber, fat, or protein slow carb absorption

When I tried this before, I added ingredients to my normal meals measured the change in my normal BG trends (see Next Experiments). This proved too noisy and I couldn’t get a clean measure of the effect of even pure glucose in a reasonable number of measurements (see Next Experiments).

This time, I have a continuous glucose monitor (Freestyle Libre, post coming soon on accuracy vs. fingerstick and attempts to calibrate it) and am going to try to more carefully isolate the effects of the ingredient being tested.

This is going to be a lot of work and take many weeks, so I was hoping to get some feedback on my experimental design before I start. If you’re interested, please take a look and leave your feedback/critique in the comments.

It’d really improve the experiment to have more people participating. Let me know in the comments or by e-mail if you want to join in (see sidebar).

PROPOSED EXPERIMENT

Note: I put some specific questions at the end

  • Goals:

    • Determine effect of individual ingredients on the blood sugar of person with Type 2 diabetes
    • Determine effect of combining ingredients on same.
    • Develop model to predict the effect on blood sugar of meals that’s more accurate than standard carb+protein counting
  • Approach:

    1. Calibrate Instruments: Over several days, measure blood sugar by both CGM (Freestyle Libre) and BGM (Freestyle Lite). Develop a calibration curve to increase accuracy of CGM data
    • Note: I’m already doing this and initial indication is that ~75% of the discrepancy between the two meters can be accounted for by a simple linear gain + offset error
    1. Establish Baseline: Monitor blood sugar while skipping breakfast & lunch (both food & insulin) to identify a period of time where my blood sugar is stable for a long enough (need at least 2-4 hours).
    • Based on previous experiments, I’ll need to wait until after lunch.
    • Will collect data on at least 3 days in which I’m not exercising in the morning (M, W, F)
    • To reduce potential noise, need to be careful not to overeat or eat late the night before.
    1. Measure Food Effects: For each ingredient or combination of interest, follow the same procedure as in the baseline, but at the selected time, consume a fixed, measured quantity of the ingredient and monitor blood sugar by CGM and BGM (every 30 min.) for 2 hours or until my blood sugar is stable for at least 1 h.
    • Initial quantity will be selected based on my previous experience of what will raise my blood sugar by ~20 mg/dL.
    • Based on the initial results, I will test different quantities of the ingredients until I have a dose-response curve with BG increases from 0 to 40 mg/dL or the quantity exceeds what I would reasonably consume in a sitting, whichever is smaller.
    • Number experiments will be at least 3 per ingredient or combination.
  • Initial Ingredients to Test:

    • Glucose tablet - baseline to which everything else will be compared
    • Dissolved glucose - effect of dissolving an ingredient
    • Whey protein - effect of protein
    • Casein protein - effect of protein type
    • Allulose - my favorite “indigestible” sweetener for baking & ice-cream
    • Oat-fiber - low-calorie, low-carb flour replacement I use for muffins and cookies
    • Inulin - used in a lot of low-carb foods

QUESTIONS:

  • Current design tests one ingredient at a time. This is a lot simpler and lets me get results for the first ingredients sooner, but does introduce a systematic variation between ingredients (the week). My thought was to mitigate this by re-testing glucose at some frequency to measure week-to-week variation. Do you think this is sufficient or is there a better design?
  • I’m not planning to repeat quantities of a given ingredient multiple times, but instead vary the quantity. Since the end result of interest is change in BG as a function of quantity, I figured this would be more experimentally efficient. Are there any problems with this approach?
  • Since experiments will be done on M, W, F, there will be a 1-2 day washout period between ingredients. Is this sufficient or do I need to separate ingredients by week to ensure a two day washout?
  • Are there any other ingredients you’d like to see me test?
  • Are you interested in joining the experiment?

Personally, I don’t care much what my blood sugar response to eating (let’s say) a banana after fasting all day is, because I neither eat bananas just like so, nor do I usually fast. Meanwhile, if the effect of adding or removing a banana to the smoothie I have for breakfast were so small that it’s lost in the noise, why should I be worried about it?

That said, please do share your results :slight_smile:

Fair enough :slight_smile: I’m hoping to build up to a model of my BG response that will enable me to predict the effect of a meal. Right now, I keep my meals very consistent so that I know the amount of insulin I need to keep my BG steady (simple carb/protein counting doesn’t give me the control I want). My goal is to get to a point where I can vary my meals more.

My previous issues with this experiment didn’t seem to be the effect being too small, but that there are other, equally large, systematic effects occurring that changed day-to-day. For example, I had an issue with my BG dropping in the afternoons on some days. I’m trying to figure out if this is from too much insulin or an exercise effect, but while I’m working that out, it introduces a bias into any food effect measurement. Similarly, the dawn phenomenon in the mornings varies with what I’m doing that day, making it morning measurements problematic without large numbers of repeats.

Regarding the fasting issue: I am also concerned that the effect of an ingredient while fasting is different from its effect under a normal feeding regimen. To check for this, once I’ve measured the fasting effect, I should repeat some of the measurements on normal days. This would get around the noise issue as I could pick ingredients/quantities that have a large enough effect and repeat them enough times to get a clean signal.

Thanks for the feedback! It’s really useful to bounce these ideas off people.

1 Like

Hey Steve, very cool project!

  1. You might want to control for time of day in your testing – glucose tolerance changes as a function of time of day and is a bit different for everyone.
  2. You might want to standardize what you eat in the morning before your experiment as that can acutely change your response at a later point during that day.
  3. Standardize your walking with respect to your food intake. Justin Lawler (and other folks) have shown that even a fairly small amount of walking can reduce their postprandial glucose rise. This raises in my head the idea that a small walk before meals might also do something interesting. So keeping your NEAT similar across days and in proximity to your test ingredients might be helpful.
  4. Consider an entraining period? If you’re aiming for being able to consistently isolate the effect of the food on your glucose response, getting into a rhythm of a few days or a week of eating your meals at standard times of day will help address points 1 and 2.

Hope some of this is useful, and hope to see what you learn!
-Azure

Thanks for the feedback and encouragement!

Definitely, my carb:insulin ratio is different at different times of day. Based on the baseline fasting study, I’m going to pick a single time of day for all the experiments in this study. It’s looking like 1-2p is about the right time (BG is stable from ~12-4p).

I’m going to skip breakfast and lunch on days when I’m running the experiment. I have to take insulin with meals and the effects can persist for more than 8 hours (and vary based on a whole bunch of noise factors), so even eating a consistent breakfast doesn’t get rid of enough of the noise.

Good point. I’ll be at work when I do the experiment, which is pretty consistently sitting in meetings, but I’ll make sure to note anything more active.

Very good point. My insulin sensitivity varies substantially based on time of day (e.g. for an identical meal, I need 6 units at breakfast, 3.5 at lunch, and 5 at dinner). Fortunately, as part of managing my diabetes I wake up, take insulin, and eat meals at the same time every day +/-30 min.

I’ve found that sitting in meetings can worsen my blood sugar response significantly…

Hi Steven,
Cool experiment premise. We are piloting PhysioQ platform for collecting data. Unfortunately CGM Freestyle Libre is slated for Q3, so it’s not directly compatible yet, but from your protocol, I think we might be able to work around using a survey system. If you think it might be a good fit, let me know.

@Kiipo_Chris, sounds interesting. I use a bunch of hacked together tools (pythonista app to record BGM data and some python scripts to combine the CGM and BGM data), but it would be really nice to have something less cumbersome. Sent you a PM so we can discuss offline.