These are chat archives for nightscout/intend-to-bolus

29th
Jan 2015
diabeticgonewild
@diabeticgonewild
Jan 29 2015 08:43
Guess what?! I am going to finish this project sooner than expected! It turns out that programming the differential equations and the statistical stuff (Bayesian modeling using Markov Chain Monte Carlo) is a lot easier than I thought using WinBUGs in the WBDiff patch/extension! WinBUGS solves the DiffEQ numerically using Runge-Kutta methods. Anyways, I still have to install BlackBox, to "hardwire" the codes for efficient computing. Unfortunately that software is only 32-bit compatible, so I have to use an old netbook. But the heavy computations needed for this project can be done in WinBUGS, MATLAB, and R in 64 bit Windows, yay! I am so relieved!
Gustavo
@bustavo
Jan 29 2015 15:29
:D
@diabeticgonewild is the "A robust sliding mode controller with internal model for closed-loop artificial pancreas" paper available to download/read somewhere?
Scott Leibrand
@scottleibrand
Jan 29 2015 16:28
Nice! :-)
diabeticgonewild
@diabeticgonewild
Jan 29 2015 18:41

Thank you! I can send it to you @bustavo. I am about to private message you.

I also happened to find MATLAB code of the controller I wanted to use, and it's very close to "all of the code". It is designed to minimize risk of hypoglycemia and hyperglycemia and is the only controller that can adjust for "missed meals" (I forget to bolus like once a month). The thesis with the MATLAB code can be found here, and the target is set to 112.5 (it's a nice number geometrically), https://stacks.stanford.edu/file/druid:jf647rq8653/Thesis-augmented.pdf . However, the risk of hypoglycemia is minimized at 140, and has been used as a target of 140, as seen here, with unannounced meals http://online.liebertpub.com/doi/abs/10.1089/dia.2014.0093 . I am going for 112.5 though.