Chris Wilson on Insulin Sensitivity (Correction Factor) with Loop and Learn, 2/6/2022

[@tinahammer7870]

Came across this comment by Chris on a Loop and Learn FB post: There’s some changes that have been made based on feedback from people testing in AAPS (and FreeAPS X), so the “277700” equation is largely outdated. One of the things we’ve discovered is that basal needs are at least somewhat disconnected from correction and carb dosing needs, so an “adjustment factor” related to the basal/bolus ratio needs to be applied for many individuals. There are also variances due to changes in absorption speed with different insulins, which are accounted for in the latest version which is pending inclusion in the official AAPS branch. The same adjustment also explains why the effects of BG-dependent changes in insulin sensitivity are largely masked in clinical clamp studies, and why Affrezza dosing is wonky. Most of the things referenced above have already been identified, accounted for mathematically and are moot points in the latest iterations. The shape of the curve has also changed slightly from a 1/X to a 1/ln(X), which stretches and contracts. And to Joshua Bohmer’s point, variances in resistance ARE accounted for. This has been an iterative process, identifying outliers for whom things don’t work, and adapting the underlying math to allow it to flex, change and stretch according to individual needs. I’ve even identified an additional level of natural logs inside of the current equation.

[@cudelrab]

Much thanks for this video! ^_^

[@getom9992]

I never can get my ISF exact right, I just discover trial and error and my pump has no setting on it since ISF is fluctuated, so only estimate.

[@rafaeldelrey9239]

Is the source code, or details, on the neural network you built, publicly available?

[@LoopAndLearn]

Rafael ... please contact Chris directly at the links provided.