• Guest - w'd love to know what you think about the forum! Take the 2025 Survey »

Discussion SpotOn - an insulin dosage prediction app

This is actually really interesting stuff! Thanks for sharing! I think their approach goes one step further than what I intend to do, but it looks really promising! Do you know if they plan to build an app based on this knowledge? They should!
 
I have decided I am buying myself an android phone for this stuff. It's something that I can't do on my iPhone... It can be my "Diabetes phone". Wouldn't it be great if we could make it speak bluetooth to a pump so we could use it as a pump controller...?
 
SpotOn App said:
This is actually really interesting stuff! Thanks for sharing! I think their approach goes one step further than what I intend to do, but it looks really promising! Do you know if they plan to build an app based on this knowledge? They should!
Click to expand...
I wish! But they are wary of the FDA who wouldn't approve of them distributing this.
Some experimental branches of NightScout use bits of the DIYPS code.
 
tim2000s said:
I have decided I am buying myself an android phone for this stuff. It's something that I can't do on my iPhone... It can be my "Diabetes phone". Wouldn't it be great if we could make it speak bluetooth to a pump so we could use it as a pump controller...?
Click to expand...
Scott Leibrand & Dana Lewis of OpenAPS have already "closed the loop" and are controlling a Medtronic pump with great success.
http://diyps.org/2014/11/26/what-we...ng-the-loop-with-diyps-and-open-process-work/
 
However, if I understand it correctly they are building a purely reactive system. In control engineering you would call his a pure feedback system. What I would like to achieve with spoton is to create a model (multiple input, single output system) of the BG level in order to add a feed forward component. I think the two approaches are complementary and not competing.
DunePlodder said:
I wish! But they are wary of the FDA who wouldn't approve of them distributing this.
Some experimental branches of NightScout use bits of the DIYPS code.
Click to expand...
 
Last edited by a moderator:
SpotOn App said:
However, if I understand it correctly they are building a purely reactive system. In control engineering you would call his a pure feedback system. What I would like to achieve with spoton is to create a model (multiple input, single output system) of the BG level in order to add a feed forward component. I think the two approaches are complementary and not competing.
Click to expand...
Don't claim to understand but it sounds good! A system that "learns"?
 
@DunePlodder: Yes, that's essentially the idea. Instead of simply reacting to a currently high or low BG value, you anticipate the needed insulin/carbs to keep the BG on the desired level.
 
It's interesting, but introduces something into the mix which is totally out there. The idea of a predictive bolus. Whilst that's all very nice, it does require that you have a reasonably predictable schedule. That's not a reason to discount it, but I'm not sure how useful I personally would find that!
 
tim2000s said:
It's interesting, but introduces something into the mix which is totally out there. The idea of a predictive bolus. Whilst that's all very nice, it does require that you have a reasonably predictable schedule. That's not a reason to discount it, but I'm not sure how useful I personally would find that!
Click to expand...
I don't think you need to have a predicatble schedule in order to predict the required bolus. I do think that you need to know the effect of all the variables which determine your BG (carbs, sport, stress, base need, insulin [and many more]). Then you would just need to tell the algorithm what you are planning to do in terms of these variables and you could predict the BG value.
 
SpotOn App said:
I can't seem to find it. When was that discussion you are refering to? Or can you copy the text here?
Click to expand...
Sorry, it's just something that's cropped up there over the last month or so I've been looking at it. It wasn't a specific thread. Try searching for "bolus" in the box top right.
 
SpotOn App said:
I don't think you need to have a predicatble schedule in order to predict the required bolus. I do think that you need to know the effect of all the variables which determine your BG (carbs, sport, stress, base need, insulin [and many more]). Then you would just need to tell the algorithm what you are planning to do in terms of these variables and you could predict the BG value.
Click to expand...
But therein lies the point - you are saying predict (plan) what you are going to do in order to predict how much insulin to give. Even the body has only limited mastery of that, e.g. activation of the saliva glands leading to increase in stomach acid, potential insulin release, etc.

Given the time frame of exogenous insulin action, that may prove challenging!
 
:D I like your analogy!

But, at least for me, I usually know what I'm gonna do in the next hours. With Humalog/Lantus I need to plan around 30min to 2h in advance anyway (depending on which variable). Is this not the case for you?
 
Typically not more than thirty mins I find, especially with how I eat and train.
 
Count me in. The rate at which bolus ratios can change I would make sure that your algorithm isn't a train once solution since it will need to continually evolve.

Possibly not as interesting for end users but while developing would be good to see what confidence level your algorithm has in the dose it is suggesting.
 
Hi Andy, happy to hear you want use the app! The download link in the earlier post should still work. Let me know if you experience any issues or have questions regarding the app!

Concerning your first point: yes, the model assigns a higher weight to measurements that are recent. Thus they will have more impact on the dosage computation.

For the second point: great idea! Do you have any suggestion how this could be implemented?
 
Back
Top