[sweston4]

I'm a type 1 diabetic and data scientist. Estimating the causal effect of a unit of insulin or food on blood sugar is an absolute crap shoot. Consider that there's a +/-20% margin of error on the reported carbohydrates on nutrition facts. We might consider this irreducible error that just cannot be modelled (Maybe you could get a calorimeter, estimate the distribution of errors, and reduce that error somewhat). Therefore, even if we created a model that explained all explainable variance, we still have a 20% margin of error. If a meal has enough carbohydrates, a 20% overestimate of insulin requirements would lead to an insulin overdose that would kill you if the resulting low blood sugar is not dealt with. In other words, the irreducible variance is so large that a "perfect" model would regularly suggest lethal insulin doses.

My "solution" is to eat low-carb/keto as a "variance reduction" strategy. Still, removing carbs also introduces gluconeogenesis (the production of glucose from protein) as a factor to consider. The synthesis of protein to glucose also occurs on a much time different time horizon than the consumption of carbs themselves which has implications for insulin dosing and insulin type.

I could go on! But long story short, modelling blood glucose is bloody hard.

[dsaavy]

This "variance reduction" strategy is also an approach by Dr. Richard Bernstein. Basically smaller doses means smaller mistakes. It's a great approach in my opinion. Obviously the key is sticking to a low-carb diet for that to work, along with tweaking your basal insulin dosage.

For me, this approach worked when I switched from Novolog to Novolin R, which has a slower effect rate. I use shots so it helps match the blood sugar profile of eating higher protein and fat meals. If you use a pump, you can adjust your bolus to give a % upfront and a % over a period of 2-3 hours so that you can match the slower glucose production from protein. In Bernstein's book, he had some typical ratios for grams of protein to insulin that turned out to be pretty close for me.

It obviously doesn't solve everything but I certainly feel better when I'm at 60 and have 1.5 units on board compared to if I had 7. The Novolin R makes me feel even better because I also know my blood sugar can't drop too fast, where Novolog I could be dropping 6-7 mg/dl per minute. So I have time to react and let my body absorb some simple sugars before it's too late.

Just a personal anecdote, maybe someone will find this useful!

[sweston4]

Thanks for this! Bernstein's book is the source of my "variance reduction" strategy too! The one thing I haven't adopted is the slower acting insulin. It's one of those things that I know I should try but just haven't gotten around to yet, so thanks for the extra incentive to try it.

[mellavora]

> that would kill you if the resulting low blood sugar is not dealt with.

My wife is type I, so I have a sense of what you live with.

She bought a book, "The Insulin Murders", which looked at a number of cases where insulin was the weapon of choice. The good news is that it is actually really hard to die from low blood sugar, assuming good medical care is available.

Coma to death is > 12 hours, more like 24 or 48. Assuming other people are around, there is plenty of time for medical response. And treatment is easy, glucogon turns it around in minutes.

And I'm not sure a 20% insulin overdose would trigger coma. Definitely hypoglycemia, but blood sugar has to be pretty low for coma.

Look, I'm not saying it is easy, and risk of harm from getting it wrong is high (as you wrote), but risk of death is much lower than you might think

[coldpie]

> The good news is that it is actually really hard to die from low blood sugar, assuming good medical care is available.

Yes, but the bad news is it only takes one mistake to do you in, and the battle never stops for your entire life. I had a fellow T1D friend die last summer from hypoglycemia. Wikipedia says (with a citation, available at link):

> In terms of mortality, hypoglycemia causes death in 6-10% of type 1 diabetics.

It's the kind of thing that hangs over you. Every time you go to sleep, you wonder if maybe you took too much at dinner and this will be your last night. (I'm sure you know this from your wife--there's a reason she was interested in that book--but the reading audience may appreciate the context.)

[1] https://en.wikipedia.org/wiki/Hypoglycemia

[Trasmatta]

> The good news is that it is actually really hard to die from low blood sugar, assuming good medical care is available.

Except it's extremely hard to get medical care when your mind and body shuts down because of a severe case of hypoglycemia. There isn't always somebody around to call an ambulance.

If you haven't experienced a severe episode of hypoglycemia yourself, you really don't understand fully how it can effect both the mind and body, even if you've seen it in your wife.

And like the other poster mentioned, 6-10% of T1D's die of hypoglycemia. It's a lot easier to die of than you're giving it credit for.

If COVID had a 6-10% death rate, I don't think anyone would be saying "it's actually really hard to die from COVID".

> And I'm not sure a 20% insulin overdose would trigger coma

You can experience hypoglycemia without any insulin overdose. There are many other factors that impact your BG, and sometimes a combination of those will hit a T1D with a severe hypo, even if they took what should have been the correct insulin:carb ratio.

[sgmoore]

> You can experience hypoglycemia without any insulin overdose.

One of the worst hypos that I have had (Fingerprick said 1.9mmol/l which is about 34 mg/dl) was a 'Lantus low' which is not really an overdose.

[pimeys]

It adds a bit to the risk that going down to hypoglycemia is not very good for your brain cells. And staying in hyper is not good for your cells in general, for your eyes or for your internal organs.

You might not die, but might develop some nasty problems later on in your life...

[sweston4]

I 100% agree with what you're saying.

Assume a 1:10 insulin to carb ratio (though I don't think insulin dosages have a strictly linear effect) and a meal of 100 carbs. If the meal, in reality, has 80 carbs, you've injected two extra units of insulin. We might expect a unit of insulin to reduce blood glucose by 30-50 mg/dl, so we've reduced our expected blood glucose by 60-80 mg/dl. If you target a blood glucose of 80 mg/dl, this would mean your expected future glucose is 0-20 mg/dl. In that case, you would die.

Now, the problem with this example is that I would 100% take action to avoid dying, so my actual risk of death is still minute despite having a lethal dose of insulin in my body at the time I inject myself. However, it's still a bit strange to know that, conditional on me taking no independent action, I'm hours away from death in certain situations.

[sgmoore]

> The good news is that it is actually really hard to die from low blood sugar, assuming good medical care is available.

And yet, have there not been a number of cases where medical professionals have been convicted of the murders of several patients by injecting them with insulin? I would expect that these patients should have had 'good medical care' available to them.

I have not read the book, but perhaps it is ignoring the fact that extreme hypos may trigger other problems such as heart attacks. When a condition becomes common enough to be given a non-medical name, as in the 'dead in bed' syndrome, then I you can't *assume* you will survive for 24 hours.

The risk of death might be lower than I think, but I'm not to be ignoring any alarms I get.

(Type 2 but insulin dependent)

[idealmedtech]

Insulin overdose is very hard to die from, but it can easily cause permanent brain damage. And that's just in healthy adults. In terms of immediate risk, hypoglycemia is far more dangerous than hyperglycemia (though this is not to discount the severity of DKA).

[m_kos]

> gluconeogenesis

I am very curious about gluconeogenesis. I am well aware of this pathway but have read a few times that glucose generation from protein happens very rarely. Have you ever tried eating zero carb whey protein in a fasted state? If yes, have you noticed a spike in glucose? I am genuinely very interested in your experience with gluconeogenesis as a person with diabetes.

[sweston4]

Oh, I am very happy to answer this to the extent I can. Whey protein impacts my blood glucose in confusing ways! Most days, I'll have a whey protein shake that consists of one scoop whey protein, a little bit of coffee for taste, and some almond milk. This will typically be my first meal after ~12-16 hours without eating. The correct dose of insulin for this shake can vary from 2 units of insulin to 4.

First, I believe you're correct that gluconeogenesis happens rarely. More specifically, I believe it happens in the absence of carbohydrates in the food you consume. Since I eat a low carb diet, it would make sense that I experience gluconeogenesis.

Second, have I noticed a spike in glucose? Yes! I have to take some amount of insulin if I have a protein shake. The coffee has no effect, so any effect comes from the almond milk and whey. There's maybe half a cup of almond milk in my shakes which is close to negligible. We can probably assume that whey drives most of the glucose effects from my shake.

Third, why is there so much variance in my blood glucose response and insulin requirements? Here, I do not have a defined answer. One aspect may be that whey is quickly digested. The quick digestion may accelerate the effect of gluconeogenesis. Another factor may be the state of the glucose reserves in my liver before I consume the whey. If I'm in a state of ketosis, it may be that my body accelerates gluconeogenesis because it believes it's in a carbohydrate shortage. In this situation, the glucose spike may exceed what would otherwise be expected. There's a few other things such as exercise I'd include here, but I don't have a single definitive answer to this.

A final note: In some sense, I can feel/anticipate my insulin sensitivity during the day. I cannot explain this in writing in any coherent way, but I have decent intuition on what insulin dose between 2 and 4 units I should pick each day. Or, before bed, even if my blood sugar appears constant and in-range, I may anticipate that it'll go up and down as I sleep and eat/inject insulin accordingly. One part of this intuition is "knowing" rather a protein heavy meal will kick into gluconeogenesis while I sleep. Anyways, I can answer more, but as you can perhaps tell, most of my explanations are of the waves hands variety.

[m_kos]

Thank you for such a thorough and informative response!

[consp]

> have to take some amount of insulin

There are quite a few pathways for your muscles and liver to release "stored glocose". If you want to artificially do it, get a glucagon siringe and get the worst hyperglycaemia you ever had since your liver starts converting glycogen into glucose en-masse. Natural release of glucagon is also regulated by having a low carb diet.

[LtdJorge]

I just want to point out that gluconeogenesis can also use fatty acids to obtain glucose.

[ineedasername]

In addition to the margin of error on food labels, there's also the margin of error on blood tests, which IIRC from the spec's on the one my wife uses it too is about +/- 20%.

If you have the misfortune of having those errors match up (and they will at times) they you eat something that overestimates carbs by 20% and a blood test that overestimates your current glucose level by 20%. Give yourself a corresponding dose of insulin and drive your levels through the floor.

[whatshisface]

>Consider that there's a +/-20% margin of error on the reported carbohydrates on nutrition facts.

Is this accounted for by product-to-product variation or package-to-package variation?

[stonemetal12]

The 20% seems to be how much you are allowed to lie by. You get another bit for variability of the test, and a third error term for variability of "good manufacturing practice".

Here is the actual rule from https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfcfr/cfr...

A food with a label declaration of calories, total sugars, added sugars (when the only source of sugars in the food is added sugars), total fat, saturated fat, trans fat, cholesterol, or sodium shall be deemed to be misbranded under section 403(a) of the act if the nutrient content of the composite is greater than 20 percent in excess of the value for that nutrient declared on the label. Provided, That no regulatory action will be based on a determination of a nutrient value that falls above this level by a factor less than the variability generally recognized for the analytical method used in that food at the level involved.

Reasonable excesses of vitamins, minerals, protein, total carbohydrate, dietary fiber, soluble fiber, insoluble fiber, sugar alcohols, polyunsaturated or monounsaturated fat over labeled amounts are acceptable within current good manufacturing practice.

[bosie]

I am not a lower but the 20% do not seem to be about carbs. Only sugar. Carbs are only covered by the second paragraph and must be 'within current good manufacturing practice', whatever that means. Am I reading this wrong?

[kahrl]

The toothless FDA allows for a 20% margin of error on nutrition facts labeling, so it could possibly be one or the other or both.

Some products may just have variation. Some foods will be maliciously mislabeled with 19% less calories/sugar/fat but may have little to no variation within the same product.

[MattGaiser]

I am beginning to believe those who count calories and lose less weight than they anticipate.

[jamiek88]

Yeah 20% is massive if most of your calories come from carbs.

Sensible dieting talks about 10% reductions in intake along with light exercise.

Trying to manage that by those labels would be impossible.

However eating packaged processed foods isn’t a very good way to lose weight anyway.

I lost 180lbs a few years ago by cutting processed food, soda and alcohol out of my diet. Didn’t change anything else.

I presume the carbs in veggies are pretty much accurate by weight a carrot is a carrot (except for water content).

[btowngar]

And there can be large errors in the CGM data as well!