@Dave44, @Firenza thank you both for the detailed answers, this is just what I needed to hear.
@Firenza at this early stage the app beta only logs and shows you photo stream with glucose tags, yes. But we definitely think about auto-tracking/suggesting trends as the next step.
I am 100% Android so not an option until you have an Android compatible app. But if and when you get there may I suggest you consider an option of scanning a UPC code so that when I eat a Healthy choice steamer for example, the nutritional information can be sucked right into your app based on the code. Furthermore a factor for serving so if I am going to share a Campbell soup and eat 60% of it I could just scan UPC, put in 60% and voila, a very accurate data-set of exactly what I ate.
I like the concept of keeping a historical record of specific foods and insulin responses. Not sure I’d use it with photos though (think eating out with friends in an upscale restaurant). But I have found Tidepool to do something similar but without the photos and showing more context. On the Tidepool app (Android & Apple) I click on the hashtag “meal” then enter the food and, if desired, location. That’s it, quick and discreet, and this data can also be entered at a later time. Tidepool then inserts this event in the context of whatever insulin dose I gave via my pump along with the CGM data. You’ll see six hours of context before and six hours after each event entered. It does take a few hours for this data to populate but it’s brilliant overall. So, when I’m at a restaurant and debating ordering a dish I can do a quick search with the Tidepool app for that restaurant and see if I ordered it before and how it worked out. Here is a screenshot. Best of luck with your project! I hope my comments were useful but not discouraging.
@CJ114, thanks for the idea, do you mean that you prefer scanning rather than taking a meal pic, or that it’d be better to have both options?
@mremmers Mary, thank you a lot for detailed feedback, it really helps us decide about the next move 
We see that photo logging is not so universal and handy, as we thought it was before this discussion.
My initial thought is that food acts differently in people. Some people’s BG rises with tomato sauce, others’ doesn’t. It’s a neat idea though and probably something to consider as long as there is some kind of language indicating what I just expressed above.
Something to consider is that the Tidepool app and website is pretty much only usable as a data aggregator. It requires information from a cgm, insulin pump and/or blood glucose meter to give these events any context. There is no user supplied manual data within Tidepool. Folks who are testing with a meter and keeping a simple log might find your app fills a need, photos and all.
I think keeping a food log, whether logging a description or taking a picture, is a worthwhile task for a limited time period. I think doing this can provide you with a good education on connecting certain meals to certain blood sugar outcomes.
I’d like to describe a highly desirable fantasy app. It would be an app that used the picture you took, calculate what insulin dose you need, and if you’re using a pump, wirelessly send that dosing info to the pump and then, after user confirmation, deiiver the dose.
Any app or technology needs to produce much more value than the effort required to set up and maintain it. People who use insulin have already made dozens of compromises to their preferred lifestyle. Most of us want to enact treatments that simplify, not complicate our lives.
Yes, I agree with this. I think the idea of correlating pictures of food and BG response would have been useful in my first couple of months after diagnosis. However, at this point, I want something much more powerful. Pretty much what you describe.
I want predictive information about BG based on what choices I can make right now. Given my (current BG + IOB + x + y), choice 1 (Baby-Sized Chicken Burrito from Ketchum Burrito Company) + [no exercise] = 75 m postprandial BG of Z, while choice 2 (Spicy Chicken Salad Bowl fro KBC) + [15 minute walk] = 75 m postprandial BG of 0.75Z.
I actually think such models are more or less realistically doable at this point, given fairly large error rates, but the app doesn’t exist (so far as I know). Pull all the most recent activity and health data from Apple Health, my BG meter or CGM, my pump (if I use one), etc. and bounce the expected macros from a given meal off of that with a curve that shows me 75, 120, and 180m predicted BGs. That’s what we do in our heads already, right?
Exactly. An app that can relieve us of that cognitive burden as well as eliminate human error due to being tired or miss-applying one of the many factors, could earn its keep.
Both would be best. UPC is not always available such as in a restaurant, but where UPC is available, the exact data could be extracted. Then next time that same UPC code is scanned in advance of a meal, your software could pretty much predict and suggest pre-bolus time and units required for that same meal. Many of us are creatures of habit and repeat certain meals so over a short period of time a fairly reliable algorithm should be available to suggest a fairly tight range for both pre-bolus time and quantity.
Yes, this exactly describes our initial idea. But after talking to people it seems more of a ‘nice to have’ feature, not life-changing or significantly improving life quality. So we’re looking further.
Thanks for this summary, @Terry4. Completely agree about the effort-vs-value ratio. Do you think you would let the app autopilot your pump without any manual control/revision from your side? Or you’d rather get prompts, confirm its decisions?
I want predictive information about BG based on what choices I can make right now. Given my (current BG + IOB + x + y ), choice 1 (Baby-Sized Chicken Burrito from Ketchum Burrito Company) + [no exercise] = 75 m postprandial BG of Z , while choice 2 (Spicy Chicken Salad Bowl fro KBC) + [15 minute walk] = 75 m postprandial BG of 0.75Z .
Thanks for the detailed illustration, @David49. Could you explain a bit more about your decisions based on such future BG maths? Is it mostly about choosing what to eat, or how much exercise the meal would cost you (to escape insulin corrections?), or to calculate the bolus dosage?
I think a useful app would be difficult to implement. First, I should say that I don’t use a smartphone for anything! I’m not technology averse, but I just don’t see the need. But, if there were a real killer app, I would be right on it.
First, there are multiple approaches to controlling diabetes. Some, like myself, follow a low-carb approach. We choose foods at home that are low carb and when eating out request that our waiter modify what we’re served (skip the rolls, replace potatoes with low-carb veggies, etc.) I think photos in this case are not helpful. Other diabetics choose higher carb meals and take additional insulin to cover. It would seem that this latter group would be more interested.
I question whether a photo has enough information to make treatment decisions on. Sure, it might handle something like an apple. But what’s in that salad — cranberries? apple slices, croutons? And how much of each? The same issue applies to other foods as well. A sandwich at restaurant A is not the same as restaurant B.
And as Dr. Bernstein says, even food labels are only correct to +/- 20%. So even that modest spaghetti dish you are eating could have significantly different insulin requirements. Is the app/photo able to determine the size/weight of everything on your plate? As others have noted, we may not eat everything on our plate — doggie bags are still alive and well!
Many of us vary the timing of insulin relative to the meal. Some may take insulin a half hour before the meal, while others take it at the start. And that’s only when we can plan. In some circumstances we may not be able to take the insulin in advance. The outcome is likely to be significantly different.
For a new diabetic, perhaps any errors and approximations may still be useful. But I don’t see it as helpful for an experienced diabetic.
Until an app has gained many 100s of thousands of hours of experience doing this “fantasy protocol,” i would only trust using it in manual mode.
Plus, from the developers’ point of view, getting regulatory approval for hardware that would automatically dose insulin is a much harder hurdle to clear than simply making a suggestion and needing user approval before adding insulin.
I was mostly just giving a silly example (with a kernel of truth based in how predictive modeling works at a scientific level). I’m a research scientist who does a lot of statistical and mathematical modeling, and I was mostly suggesting that at this point we should be able to make some simple, stochastic models of how BG actually responds to meals and/or exercise.
Current BG + how much insulin on board + what I’m about to eat (macro content) + my bolus + what kind of exercise I have done in the last few hours + what kind of exercise I expect to do after eating == a BG curve through time (now, 75m post, 120m post, 180m post) are the
So, personally, my fantasy app would pull the following information from my connected apps & data sources:
Manual Entry:
Output:
What would I use this for? I would use it to play around with bolus doses and timing and exercises (preferably with sliders or something similar) and see how that affects a curve. If the expected BG curves turn out to be decently accurate (can use statistical techniques to compare expected to CGM/stick results), I might actually start using it as a decision making tool (timing of bolus, size of bolus, how much exercise I want to do if I choose a certain meal and bolus, etc.).
Ultimately, I want a decision support tool with a way to visualize how my daily history and current choices are going to affect my near-future BG. I have a pretty solid understanding of this for my own case (currently), but I also love tech tools and predictive models. They’re awesome.
Incidentally, in a real dream app, I’d also want to run semi-supervised learning algorithms on all auto- and manual-input and expected BG curves in order to refine the underlying models automatically. People call this “AI,” but it’s not, really… Anyhow, doing automatic improved model selection and model-fitting would lead to what Terry wants…a tool that understands the individual’s expected response and can make either automatic decisions for management (through a pump) or suggestions for management (through injection or manual control of a pump).
Yes! I’ve been wishing there was an app like this!
I love photo logging. Most of what I eat is homemade and it’s annoying to go find all the exact components of my meal in a database.
Maybe it’s a generational thing but I’m always taking pics of food anyway. Whether for Instagram or WhatsApp groups (family). My family members are foodies so we’re always sharing pics.
