List of useful Apps? General thoughts on blood glucose predictions for MDI

[SimonP]

Is there a (recent) list anywhere of apps that are useful for diabetics? Ideally with a few bullet points/review of what features they each have?

I use XDrip+ (Android), I'm aware there's also an XDrip on iPhone which is different aside from the name, I've heard of mySugr, MyLife and AndroidAPS. There are doubtless others too, it would be useful if there were a list somewhere.

The reason for asking is that while XDrip+ works fine for logging data from my libre2 and Contour Next, the glucose prediction model is quite basic (useful, but perhaps it could be better) and doesn't support/detect variable insulin sensitivity, nor take exercise and different food compositions into account when predicting future blood glucose.

These aren't easy things to do, so I'm not knocking XDrip+, but it would be nice to know what else is out there.

I see that lots of apps assume a pump (e.g. AndroidAPS, which does apparently have insulin sensitivity detection), I'm on MDI so it would be useful to know what would work for me, and quite specifically it would be nice to have both MDI basal and bolus recommendations.

Similarly, are there any food logging app recommendations? I currently use a very simple App called Diet Diary, though it doesn't support recipes, but I just wanted something that I could start with and easily export the data to post-process and combine with data from XDrip+, my Garmin for exercise, etc.

What I would really like to get to is a system of apps that do their own specialised things (e.g. XDrip+ does CGM, fingerprick and insulin logging; Fitbit or Garmin, etc., does heart rate, exercise, etc.; a Diet/Food related app does food and recipe logging, etc.) but with each sharing data in real time (via Android Intents or local webservers, as XDrip+ does, rather than needing a manual CSV export) so that it could all be combined in another app and used to experiment with prediction algorithms.

I can currently export all the data and do the experimentation manually, which I plan to do, but I'm trying to think ahead - my food logging is only by description, so I need to write some code to lookup foodstuffs and pull in their nutrition/glycaemic composition (I log total mass and/or total carbs so can calculate quantities).

This level of logging is perhaps too much faff for most people, I don't know, I only planned to do it for a little while in order to see what I could do with blood glucose predictions and here I am still doing it some years data, not having had much time/motivation to work on the prediction code and intermediate data gathering steps. I always have other projects on the go and until the point that my blood sugars go a bit mad and then I wish I had the tool I've described available, it's fairly low on the priority list as it's quite an involved project. Once the blood sugars sort themselves the priority level drops once again, iteratively chipping away at the problem, such is life with hobby projects 🙂

Any thoughts and ideas appreciated 🙂

 

[littlevoice359]

I spent many years developing decision support systems for high-value, high-volume manufacturing (semiconductors). So, when I acquired T1D in late 2018, I started collecting data from the start. Over the past years I have expanded my software system and I now have a pretty sophisticated package. A component of this is a meal planner that lets me look, when planning a meal, at the last time I ate a meal with those ingredients and what the effect on my blood sugar was at the time. I find this useful, but I am also quite often surprised. The past effect is not a sure predictor of the future effect. The reason for this is that there are just too many other factors in play that you will never be able to build into the data model, simply because the data is not available (cf: the list of 42 things that affect blood sugar).

I am attaching some screenshots from recent meals, showing how I cross-reference meal ingredients with my blood glucose diary. If there is any other software system out there that does something like this, I am not aware of it.

If I were you, I would not expect to develop too accurate a predictive model, though I do wish you luck with your endeavours.

 

[everydayupsanddowns]

It’s not really the list you are looming for, but the dictionary of terms on T1 Resources will have some apps etc listed that you may not have come across?

Diabetes Tech Dictionary : Loop, OpenAPS, APS, Nightscout, New Diabetes Technologies : Terms, Definitions and links

A comprehensive list of terms and definitions for emerging diabetes technologies, Loop, openAPS, AAPS, Nightscout, CGM, and sensor-augmented insulin pumps. Concise explanations with helpful links to explain in more detail

www.t1resources.uk

Though I notice that Juggluco(?) and Tomato(?) aren‘t listed, which are two bits of software I think I’ve seen mentioned here?

 

[SimonP]

It’s not really the list you are looming for, but the dictionary of terms on T1 Resources will have some apps etc listed that you may not have come across?

Diabetes Tech Dictionary : Loop, OpenAPS, APS, Nightscout, New Diabetes Technologies : Terms, Definitions and links

A comprehensive list of terms and definitions for emerging diabetes technologies, Loop, openAPS, AAPS, Nightscout, CGM, and sensor-augmented insulin pumps. Concise explanations with helpful links to explain in more detail

www.t1resources.uk

Though I notice that Juggluco(?) and Tomato(?) aren‘t listed, which are two bits of software I think I’ve seen mentioned here?

Thanks, I'll have a browse 🙂

 

[SimonP]

I spent many years developing decision support systems for high-value, high-volume manufacturing (semiconductors). So, when I acquired T1D in late 2018, I started collecting data from the start. Over the past years I have expanded my software system and I now have a pretty sophisticated package. A component of this is a meal planner that lets me look, when planning a meal, at the last time I ate a meal with those ingredients and what the effect on my blood sugar was at the time. I find this useful, but I am also quite often surprised. The past effect is not a sure predictor of the future effect. The reason for this is that there are just too many other factors in play that you will never be able to build into the data model, simply because the data is not available (cf: the list of 42 things that affect blood sugar).

I am attaching some screenshots from recent meals, showing how I cross-reference meal ingredients with my blood glucose diary. If there is any other software system out there that does something like this, I am not aware of it.

If I were you, I would not expect to develop too accurate a predictive model, though I do wish you luck with your endeavours.

The historic view of a similar meal is one of the things I would like, how useful do you find this - is this completely off? I suppose there are levels of usefulness, and at least ruling out combinations that really didn't work very well might produce a better outcome even if the exact response isn't the same.

I spend quite a lot of my day job writing simulation, data modelling and analysis code (mainly automotive), which is why I also started logging data (though I perhaps should have thought harder about the later difficulties in processing it!) when I started using a Libre, in order to have something to work with when I eventually get round to to doing it 🙂

Do you run an predictive model at all, or is it all post-hoc analysis and noting things that did or didn't work? If you do run a model, what sort of thing is it (which I realise is quite a wide question)? It would be nice to be able to simultaneously extract background insulin sensitivity (exercise, ToD, illness - all via appropriate hints), wrong quantity logged, and effects of different compositions. This will be hard, but it feels that it should be doable in an iterative fashion (in the model). Hard is fine though, as with many projects it's the journey that is interesting (though actually in this case, as with a few others, I should really try to get a barebones system up and running ASAP so I can actually make use of it 🙂).

The main reasons I've not yet got something up and running are that XDrip+ is just about sufficient and that the Android API keeps changing so much in between my being interested enough to dip into it that I feel like I need to restart. It's not helped that the software components I need to write to get a usable system (GUIs I hate writing, also things like recipe logging) are not very exciting to me, so I find I often get distracted onto something else more immediate (like bike computer software especially as the sun is shining more). Perhaps I need to set some personal milestones 🙂

 

[littlevoice359]

I don’t use a predictive model. All analysis is post-hoc. Sometimes I find my diary entries and comments after the previous meal are crucial in understanding of what the likely effect of the same meal will be this time. I know of no way to factor free-form comments like that into a predictive model.

I also deliberately chose not to use APIs to pull data from sources like xDrip, Libre, Dexcom because those are liable to change and/or not well documented. I just work with whatever data export mechanism is available for the data source - usually CSV or Excel files.

Data logging and meal planning is tedious, no question. I coded both into a web app that I access via iPhone and iPad. For meals I keep an old iPad on a book stand in the kitchen to make the process a little less tedious.

 

[Spathiphyllum]

What I would really like to get to is a system of apps that do their own specialised things (e.g. XDrip+ does CGM, fingerprick and insulin logging; Fitbit or Garmin, etc., does heart rate, exercise, etc.; a Diet/Food related app does food and recipe logging, etc.) but with each sharing data in real time (via Android Intents or local webservers, as XDrip+ does, rather than needing a manual CSV export) so that it could all be combined in another app and used to experiment with prediction algorithms. ...

Snap! I'm not 'an IT guy' nor certified as a quantoid of any kind-- but have always had as it were quantoid proclivities; I love data and thinking about things one could theoretically or in principle do with data. It annoys me when I see data out there that could be better used and isn't. And-- before anyone else says it-- yes, this does mean I am frequently annoyed. ; )

So it annoys me enormously that I have all this data on my Libre 2-- not only re glucose levels but also notes in Logbook re carbs and insulin and what sort of food I've eaten-- but the Libre 2 app does not provide any proper way to analyse this data, let alone make use of it. And in addition I have a Fitbit, and it would obviously be sensible to integrate Fitbit exercise data.

I am hoping to get a pump, and indeed HCL, so I'll wait to see about that. Depending on how that goes-- in my thinking/annoyance about my existing tech, the thought occurred that I know a couple of fully fledged geeks whom I could ask, who might find it fun to have a go.

In summary: All power to your elbows, and please do let us know how you get on!

 

[SimonP]

I don’t use a predictive model. All analysis is post-hoc. Sometimes I find my diary entries and comments after the previous meal are crucial in understanding of what the likely effect of the same meal will be this time. I know of no way to factor free-form comments like that into a predictive model.

I also deliberately chose not to use APIs to pull data from sources like xDrip, Libre, Dexcom because those are liable to change and/or not well documented. I just work with whatever data export mechanism is available for the data source - usually CSV or Excel files.

Data logging and meal planning is tedious, no question. I coded both into a web app that I access via iPhone and iPad. For meals I keep an old iPad on a book stand in the kitchen to make the process a little less tedious.

I must admit I don't bother with notes and instead trust that I'll remember, which may not be very effective in reality, though I do use my Diet Diary to add entries about things like how I feel (if I'm poorly that is rather anything psychology-based).

I can understand why you've taken the approach you have. I don't want a complete hand-holding predictive model but rather an historical view so I can make informed future decisions about what/how much to eat and what to dose, with a relatively short term model to deal with things that I've already eaten and insulin I've already taken. I'm not sure how much sense that makes, I'll try to give it a bit more consideration and explain myself/my usecases better (which is what I need to do for a GUI workflow anyway).

Re datasources I would prefer automation if possible, but equally writing the automation code for each data source is a pain/not something I want to have to do when I don't have e.g. a Fitbit (any more, I now use a Garmin). My preference here is a plugin based system allowing people to write and share their own, or to fallback to something like CSV import if purely historical review is sufficient.

 

[SimonP]

Snap! I'm not 'an IT guy' nor certified as a quantoid of any kind-- but have always had as it were quantoid proclivities; I love data and thinking about things one could theoretically or in principle do with data. It annoys me when I see data out there that could be better used and isn't. And-- before anyone else says it-- yes, this does mean I am frequently annoyed. ; )

So it annoys me enormously that I have all this data on my Libre 2-- not only re glucose levels but also notes in Logbook re carbs and insulin and what sort of food I've eaten-- but the Libre 2 app does not provide any proper way to analyse this data, let alone make use of it. And in addition I have a Fitbit, and it would obviously be sensible to integrate Fitbit exercise data.

I am hoping to get a pump, and indeed HCL, so I'll wait to see about that. Depending on how that goes-- in my thinking/annoyance about my existing tech, the thought occurred that I know a couple of fully fledged geeks whom I could ask, who might find it fun to have a go.

In summary: All power to your elbows, and please do let us know how you get on!

XDrip+ does provide a way to share data and indeed even to upload it via the local webserver, but as you say e.g. the Libre App doesn't do this, which is a pain, as I would have probably kept using the app had there been a more convenient and timely way of exporting data from it.

Garmin isn't all that much better, things can be exported but in bulk form or from multiple places. There is Python code that will call their API to grab data from online, but even this is not really timely enough, it would be much better to have local APIs available.

The one that annoys me most is the food tracking aspect, there appear to be some really nice apps for those looking to track their calorific intake, taking photos of food and guestimating how much is there (no idea how well that works, but cool if it does), but again you're locked into their app with often no way of exporting data and certainly not getting it as soon as it's been entered.

For the Libre app I think the argument will be something along the lines of data privacy, which I think is probably a moot point - my device, I ensure safety. For Garmin and the Food tracking apps I don't see what harm there is in providing the data (I've no idea who to suggest this to mind you), it's not like having access to the data is going to prevent you using the app - in the case of Garmin you have the hardware so need the app, for the Food tracker apps it's the app itself that is providing value in terms of the database, entry method, etc.

Interoperability can only be a good thing for the "customer".

</rant>