I have a new idea for our Glucosurfer project and I would like to hear your feedback about it. In more general terms the idea can be applied to a pump too:
Imagine you have measured your blood glucose and you are entering this new value into your mobile phone or pump. At this point the device would provide an additional information: you would see the mean glucose that is to expect in 240 minutes. This mean glucose would be the mean of the last 30 days. The 240 minutes can be adjusted and are just the minutes used to calculate your insulin on board.
Now the question: do you think the knowledge of this projected value will help you to make adjustments in the present so that hopefully the coming value will be more up to target?
It would be a kind of feedback loop to show that the present is shaping the future in a way. At the current stage of our project this information would be easy to implement. I imagine that I would make more frequent adjustments to my I:C factors by having this information always present. What do you think?
I was thinking about this and I'm not sure that a "crystal ball" app would be that useful for me because, as consistent as I try to be, my schedule will change so one "Tuesday at 3:30 PM" is not likely to be the same as the next one?
Could you elaborate a bit more on how it actually would work? Saying it check my bg and it is 170. Would it then give me the mean value of all the events 240 minutes after I had a reading of around 170? Or would it just give me my mean value of all my readings over the last 30 days.
If it would be the former it would be quite helpful and it could show me if , for example, I normally overshoot treating a low or inject too much correcting a high. Maybe also how succesful I am to maintain a value that has been within my target range over the next hours. If it only shows my mean over the last 30 days off all readings, I don't knwo how much this would help me, because I normally see it at my meter anyway.
Great input. I see the benefits of evaluating in respect to the current blood glucose. At the same time this would need a round trip to the server to get the result. For WAP on the good old mobile phone we already do this round trip so it would have no penalty to do so. For the App this would mean an additional query in the background after the blood glucose has been provided by the user. In addition the query would have to pick only those future values that are of interest (within the threshold around your current blood glucose). This could be quite an expensive operation according to time and computing power. So at the moment it seems to be ideal and at the same time less likely to be implemented due to the technical constraints we have.
My intention was to use the result of the profile diagram for the projected value:
Here are some projections done with this diagram - always 4 hours into the future: 08:00 121 mg/dl
12:00 129 mg/dl
18:00 124 mg/dl
In my current situation this would motivate me to adjust something around 12:00 so hopefully the resulting mean of 129 in four hours will decrease. It is of less value than the calculation you have in mind I have to admit. To implement or not to implement that is the question ;-)
Please evaluate my idea in respect to the Verio IQ meter. At the current date and time this meter looks at the past and provides a heads up message according to your current number. In a way the informative value is not that high because you can not DO anything about it now. You just remember the outcome or reread the alerts of the device. This is why I want to use a projection into the future to at least be able to make little adjustments:
This is a neat idea that may have some applications. If you eliminate as many variables as you can for a meal or are having problems with a specific meal I could see this being beneficial.
For instance, during the work week I eat an almost identical breakfast and lunch and run very consistent BGs usually. I have figured out what to expect from my BG in 4 hours (minus DP's effect) through rotine and elimanating varables.
The big hurdle that is difficult for me and possibly glucosurfer might be the glycemic index/load of foods. If my different dinners have the same carbs and noticably different Glycemic loads, then my BG will be noticably differnt later due to when the PP peak occurs. Or, without eliminating as many variables as possible(in my routine), then I think your algorithm and/or data may have trouble with accurate predictions?
I also follow my CGM closely and would be interested in "competing" against glucosurfer and seeing who can get or stay in target more.
I like this idea as long as there is some way of marking a reading as an exception so that it doesn't influence the average and the predicted value (assuming I'm understanding this correctly). For example, if I usually eat the same thing for lunch but then one day eat out and have a higher-than-usual reading, I wouldn't want that affecting the predictions being made since it's unusual. Same with a bad set causing a high, etc.
What I really wish diabetes software had was a way to view blood sugar as a sort of "dependent variable" and things like insulin, food, activity, other "events" as independent variables, and be able to look at the effects of these various independent variables on blood sugar. I wish there was a way to track meals and mark them with certain tags ("ate out," "high-fat meal," etc.) and then view various instances of those tags on blood sugar so that you could adjust how you respond to them in future. I wish there was also a way of marking "long-term events" such as (for women) monthly hormonal cycles, or allergies, or other things that might have an impact on BG but are hard to track.
I hope all that makes sense. The reason I don't tend to stick with BG software is that it's too focused on tracking averages and blood sugar statistics. That's not a bad thing, but it doesn't really give me any tools to better manage my diabetes. Having something that let me easily track variables and see their impact on my blood sugar in detail and over multiple instances (not just a graph showing a dot for exercise, etc.) would be extremely useful, and I think the suggestion you make would also be useful in a similar way.
I would love to find better ways for data analysis and the identification of pattern. As you might know we developed a purely visual approach to identify systematic problems:
The diagram compares the quality of control between 182 days. The quality is encoded in the colors. Next you will see how one day is transformed to colors:
Rotate by 90° and project to colors:
So far it is the best approach we have found to identify systematic recurring problems visually. To automate this process with statistical methods is a big challenge.
Your input about the tags and the menstrual cycle is valuable. It is already on our wish list but maybe it is time to adjust the priorities...
I also think that the informational value will increase with a more repetitive lifestyle or with less variables. For the competition with a CGMS I would need to invest some development effords first - although I think the CGMS is far superior. These effords will first be visible in our Apps for Android and WP7. But the most important step is to decide if this path is really worth to follow. The approach is a bit like hanging a tasty steak in front of you.
The following screenshots shows an experimental prototype of the Glucosurfer App. Next to the time you will see the mean glucose of the last 30 days for this hour (⇔ 110). I entered 139 and in comparison to the typical mean of 110 I know that my current BG is higher than the typical mean. Next to my glucose you will see the mean glucose of the last 30 days that is valid in 240 minutes (⇒ 128). These 240 minutes are the length of activity of the insulin set for the insulin on board. The minutes are adjustable of course. This gives me the possibility to adjust now to have a better result in 240 minutes than the mean. Maybe I try to change the I:C. Maybe I will just put 0.5 units of insulin on top. The result is that I use my mean glucose as a source of feedback. Do you agree?
