Diet and AI

Hi folks,

I heard though a relative that there's a company that is trying to use AI to predict blood glucose response following food intake. The general principle, I believe, is that they gather data from CGM devices, combine it with some kind of food intake logging, have an AI analyse the data, and, in theory, once it learns enough, it could predict how a person might respond to a food based on attributes such as carb types and protein, fibre and fat content. It would then enable recommendations about specific foods via an app. Scan a barcode in the supermarket to see if it's good for you, basically.

Does this sound plausible to anyone? If there's one message that's entirely consistent on this forum it's that BG response following eating is very specific to the individual. That would suggest the AI would have to analyse data from every individual to work and only then after that individual has logged all the foods they eat. It also suggests it would only work for those with a CGM.

The company is called Welldoc and the product is called BlueStar. I am not affiliated with this company in any way.

Antje77 said:

What a complicated way of checking the package for the amount of carbs.

Click to expand...

LOL

You do have a point here, though, and I know this is a contentious point on this forum, the theory it's based on is that it's not just about how much raw carbs you eat but about how quickly those carbs hit your bloodstream. The conventional advice about wholegrains being better than refined grains for example.

On the one hand it makes intuitive sense. If you drink alcohol slow enough you don't get drunk because your body metabolises the alcohol as fast or faster than you're drinking it. Blood alcohol never reaches levels which could poison you. I believe it possible that if a healthy person were to eat one nut every 10 minutes throughout the day their blood sugar would never rise much, as compared to eating all of those nuts in the space of a few minutes.

I have no idea what to think about the slower carb release theory, about some carb containing foods being safer to eat than others.

Antje77 said:

My sensor (and sensors/meters of friends with T2 rather than T1 with insulin resistance like I have) tells me quick carbs give me a faster and higher spike for a shorter time, 'slow' carbs give me a flattened out longer high.
I have no idea which is worse, but I'm rather fond of my eyes, kidneys and toes so I prefer to avoid both the higher but shorter high and the lower but longer high to be sure.

Is this AI thing aimed at diabetics, or at healthy people of the type that use glucose sensors in hopes to optimise their food or exercise regime?
There's a very big difference between how BG reacts in diabetics and non diabetics, but there is a big market for any gadget that promises to monitor any body function among the healthy.

Click to expand...

Understood, I completely understand the desire to keep blood glucose under the tightest control possible, and I agree.

The element of the AI thing I'm referring to is aimed at diabetics. From what I can make out BlueStar in it's current form is a system that can be implemented by large health organisations. It's targeted at a range of chronic health problems. There's an app for the patient and a central system behind it which doctors can access. The system integrates with something like 400 different kinds of medical devices including the sensor/insulin pump type systems. I know next to nothing about medical devices for diabetics. The idea is that data is made available to doctors automatically so they can make decisions about treatment, no need to take notes to show the doctor or whatever, and the patient gets recommendations and motivation prompts via the app relating to things they can do to manage their health. It's like an all-in-one medical data system for chronic illnesses.

I heard about the AI diet function for diabetes through a relative who interviewed for a job with the company a couple of years ago. He didn't go to work for them. I don't know if it's still in development or if it's actually been already implemented. He told me about it and I wondered, based on my reading on this forum, whether it was even a plausible thing to be trying to do - AI-generated healthiness scores for food, targeted at diabetics, which is not just based on raw carb count.

Antje77 said:

Considering how little professionals agree on what 'healthy foods' for diabetics are (ranging from the eatwell plate to keto levels of low carb to plant based low fat, all according to not the least of professionals), I don't think there is consensus about what's healthy, so AI doesn't know either.
AI uses the available information, and in this case, this information is diametrically opposite. So the advice will likely be what has been cited most.

I'll stick to my sensor and meter, they aren't in any way influenced by popular bias.

Click to expand...

Agreed, no consensus on what is a healthy diet. If the AI has access to continuous blood glucose monitoring data though, together with a log of what was eaten, could it theoretically rate a food based on how long a person spent above a certain BG threshold for example? 'Poison minutes' or whatever?

I don't know, I'm just trying to figure out what might be possible with new technology. Instead of just measuring BG once, two hours after eating, what if an AI could analyse a BG curve based on lots of data points over those two hours and provide some kind of 'damage score' for a given food or meal. It's an intriguing idea, if there is in fact some kind of logic behind it.

*Edited to fix a sentence

KennyA said:

I am sure if someone thinks they can sell a gadget (I think "AI" is fashionable now in the way "app" used to be) and can find enough people willing to part with their money then it will be made. There are already enough dodgy "watches" on the market claiming to calculate blood glucose from sweat.

I don't honestly see how this would work or if it did, how it could help me any. It would take months to calibrate: it would need to be reset depending on any number of food and non-food things occurring each day - illness, stress, ambient temperature and exercise would be only four of the non-food ones: and it wouldn't give me any more information than I could get in real time from using a much less expensive glucometer.

In other words - I don't need to spend ages inputting data into a gadget to tell me what the weather's like when I can look out of the window.

Incidentally - I don't think using the scaremongering terms "damage score" or "poison minutes" is at all helpful in this context. Raised blood glucose for short periods after eating carbohydrate is a perfectly natural part of digestion common to diabetics and non-diabetics. High blood glucose over time - ie a failure to manage glucose in the system- is what will cause damage.

Click to expand...

Agreed on the points about AI being the latest tech fad and companies trying to sell gadgets.

I did not intend to scaremonger, just struggling for language to convey the question I'm trying to answer.

The company in question appears to believe they can eventually implement some rough equivalent of the 2 hour BG test without any actual food logging or blood testing - just scan a barcode and the app tells you how healthy or unhealthy that product is likely to be for that person. If that actually worked, even a little, it would be a boon to the masses who might otherwise not be inclined to spend much time analyzing their food choices.

Outlier said:

To me, that sounds like a sledgehammer/nut interface. But it will have a market with those of us who like gadgets. I prefer to use the Mark 1 brain and test strips, but accept that is probably a preference not shared by all.

Click to expand...

To clarify - the intended customers for BlueStar are large health organizations rather than individuals. In the UK context the intended customer would be the NHS. The product concept does exploit people's love of apps though as it wouldn't work unless people liked the ability to see all their health data in one place and liked the idea of their various doctors being able to see that data too.

KennyA said:

If it's being designed for the NHS then it will be telling me that lots of starchy carbs are the healthiest option for me, and none of that nasty fat. I can read (and ignore) that sort of information already thanks.

There's a big problem with organisations telling people that individual foods are "healthy" or "unhealthy", when most often the real answer is "it depends".

Click to expand...

It's a US company so it's being designed for hospital groups I believe, I don't know much about how the US health system works, but your point is well made.