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

1st
Mar 2015
Matthias Granberry
@mgranberry
Mar 01 2015 00:10
my problem with the Dexcom algorithm is that there is some systematic error in the algorithm. Mine is always around 5-10 mg/dL lower than the fingerstick. I suspect it's intended to catch hypoglycemia earlier, but it is still a little frustrating
Toby Canning
@TC2013
Mar 01 2015 00:42
I haven't researched in the last couple years what the most accurate strips and meters are on the market. Does anyone know? This might help reduce the garbage in, garbage out ratio when using the CGM.
Scott Leibrand
@scottleibrand
Mar 01 2015 00:50
I would love to see some actual quantitative real-world meter accuracy studies. The problem is not having a good source of truth...
maybe we should just do it with
control solution...
Toby Canning
@TC2013
Mar 01 2015 00:55
Doesn't control solution have a wide range? (e.ge., this solution is accurate if your meter reads between 85 - 115mg/dL)
And then we wouldn't have a accuracy across a wide range of values anyway (40-400mg/dL)
Matthias Granberry
@mgranberry
Mar 01 2015 01:00
The Freestyle Freedom Lite is one the most accurate, with a MARD around 4%
Toby Canning
@TC2013
Mar 01 2015 01:00
Awesome, do you have a link to the research?!
Toby Canning
@TC2013
Mar 01 2015 01:02
I maybe misreading this, where did you get the 4%?
Matthias Granberry
@mgranberry
Mar 01 2015 01:02
It's not given in that document, unfortunately, but it is a good comparison for meters
Toby Canning
@TC2013
Mar 01 2015 01:05
We use the Bayer Contour "Next" strips with our MM pump. I remember Bayer claiming that significantly out perform the regular contour strips, but either they are not listed on the document or my information is bogus.
Matthias Granberry
@mgranberry
Mar 01 2015 01:05
I have been extremely happy with the Freestyle meters, and I have used 4 brands in the last year looking for one I trust. The Verio just made me mad, and the Accucheck Nano wasn't bad
http://www.ncbi.nlm.nih.gov/pubmed/25260047 seems to indicate that they are relatively high accuracy
They are about the only meter I haven't used
Scott Leibrand
@scottleibrand
Mar 01 2015 01:23
Control solution ranges are a cop-out from the manufacturers. If you get (or make) a particular control solution, you can determine its actual concentration with a YSI test, or by being precise about how much glucose and how much water you make it with.
diabeticgonewild
@diabeticgonewild
Mar 01 2015 01:26
yeah, there are a bunch of really good ones on Pubmed for free via PMC
articles I mean
also, if you are anemic, that affects accuracy, and there have been studies on that with particular meters (I am anemic due to iron deficiency--that has been determined--IV iron is likely in my future)
lot-to-lot variability studies are important too
Freestyle Lite often gets top scores though.
in accuracy, I mean.
diabeticgonewild
@diabeticgonewild
Mar 01 2015 01:49
To search for "good studies" you want to find ones that are "ISO 15197" in it, but only a handful of meters meet the "ISO 151197:2013" criteria, such as the Accu-Chek Aviva and Performa ( http://dst.sagepub.com/content/8/4/906.extract# ) and Abbott Optium Neo ( sponsored by Abbott https://abbottdiabetescare.co.uk/images/uploads/documents/White_Paper__Clinical_FreeStyle_Optium_Neo.pdf --I think it is the "precision" strips, even the FS Lite doesn't meet ISO 15197:2013)
diabeticgonewild
@diabeticgonewild
Mar 01 2015 01:54
Comparison of 43 meters for ISO 15197:2003 standards http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3570840/
Factors Affecting Blood Glucose Monitoring: Sources of Errors in Measurement http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2769960/
Lot-to-Lot Variability of Test Strips and Accuracy Assessment of Systems for Self-Monitoring of Blood Glucose according to ISO 15197
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3570841/
Accuracy Evaluation of Five Blood Glucose Monitoring Systems Obtained from the Pharmacy: A European Multicenter Study with 453 Subjects
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3317395/
Evaluation of the effects of insufficient blood volume samples on the performance of blood glucose self-test meters.
http://www.ncbi.nlm.nih.gov/pubmed/24351178
Accuracy Evaluation of Five Blood Glucose Monitoring Systems: The North American Comparator Trial
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3876374/
Evaluation of Hematocrit Interference With MyStar Extra and Seven Competitive Devices. (not open)
http://www.ncbi.nlm.nih.gov/pubmed/25549636
All of these except 1 in the last post are open. Knock yourself out!
diabeticgonewild
@diabeticgonewild
Mar 01 2015 02:17
Actually, I was just searching for something unrelated @TC2013 , and I found this for the CN, and it is about closed-loop in children. I don't know what they used to determine the MARD (hopefully a YSI), but if a YSI is not available, a Hemocue meter is used as an alternative. http://www.ncbi.nlm.nih.gov/pubmed/24876427
I don't have access to it, but if you want it, I can get it though inter-library loan
Also, I think you asked about error (tolerances) awhile back, and to adjust for that, the controllers that Hovorka uses sets the "target" for the controller higher if there is a higher error (discrepancy between expected and theoretical BGs -- x hours in time that were "corrected to target"). Actually, technically, all dosages in an AP are "correction dosages".
Scott Leibrand
@scottleibrand
Mar 01 2015 02:40
It would be awesome if someone could read through all those accuracy studies and put together a summary of how precise and accurate each meter tends to be, and rank them all.
We could publish on OpenAPS.org if we want. :-)
diabeticgonewild
@diabeticgonewild
Mar 01 2015 02:48
I would agree :D
diabeticgonewild
@diabeticgonewild
Mar 01 2015 05:40
So, you can't really see it well, but I have set up the differential equations in a spatial form, so it can be solved easily through the Simulink MATLAB extension. As you can see, the spatial approach is much more simple than programming all of this. The Simulink model/diagram, with all of the different blocks can be ported into actual code. I likely made some errors in my model that I will have to trace. And no, I didn't work on this all weekend. I probably put about 15-20 hours into this diagramming, total. http://imgur.com/ugjKxnN
This is the closest thing to proof that this paper is completely solvable, besides finishing up the project. I am actually a lot closer than you would think. I pretty much have to figure out the random walks, get the differential equations working, and appropriately apply Metropolis-Hastings sampling (which is a function in MATLAB), and extract certain probability distributions through embedded MATLAB functions. Then after that I just have to trace and find errors, and check my work.
Then the hardest part of my DIY AP project is done.
Then I have:
  • Modeling Dexcom G4 error, based on a journal article
  • Create virtual patient (copy of myself)
  • Simulate virtual patient
  • Refine virtual patient and resimulate
  • Interface with MATLAB and Nightscout
  • Go live on myself
So, not bad
diabeticgonewild
@diabeticgonewild
Mar 01 2015 05:49
And for some reason I thought it was early Monday morning...crazy!
diabeticgonewild
@diabeticgonewild
Mar 01 2015 05:58
Another step with simulating the virtual patient is programming the controller, technically, but it's not hard math. It's very algorithmic and recursive.
diabeticgonewild
@diabeticgonewild
Mar 01 2015 06:26
Going to call it quits for the night. I am pretty pleased with what I have finished so far this weekend.
Scott Leibrand
@scottleibrand
Mar 01 2015 06:27
:-)
diabeticgonewild
@diabeticgonewild
Mar 01 2015 06:31
j/k blood sugar went from 70 to 44 ... have to treat that first
but life is good. I am pretty much out of the woods in this project, I think
Then bedtime...
Toby Canning
@TC2013
Mar 01 2015 18:14
I'm getting a decocare error with the carelink stick "ERROR:decocare.stick:size (0) is less than 64 and not 15, which may cause error." The message finishes, "BAD AILING" @bewest Any thoughts on what this might be?
I just loaded a ubuntu linux distro on the pi2,
I will see if raspian fixes it too
Scott Leibrand
@scottleibrand
Mar 01 2015 18:18
Did you set your SERIAL?
Ben West
@bewest
Mar 01 2015 18:19
did you use --init the first time?
depending on your version there might be an --autoinit I think
Toby Canning
@TC2013
Mar 01 2015 18:20
@scottleibrand yup
@bewest trying now
That was it. Found a missing component to installation of our version of ihawk
Thanks!
Ben West
@bewest
Mar 01 2015 18:49
the error message for missing SERIAL is much nastier
the stick gets so mad about it it goes NAK
Sulka Haro
@sulkaharo
Mar 01 2015 19:58
Medtronic 640G rollout starting in Finland, getting a lot of pings on when NS will start to work with it :)
Jason Calabrese
@jasoncalabrese
Mar 01 2015 20:13
I'm doing a some refactoring to port IOB from iob-cob to the mainline and want to add some simple IOB tests, are there some classic examples that I can use? I need a list of insulin treatments with time and the time to calc iob at
already have browserify-express working so we can share the module between the client and server
would like to add some to verify stacking, different dia's, check activity, etc
Ben West
@bewest
Mar 01 2015 21:28
nice
nice seeing that iob code re-used
nice
stub out some ttests.... very very nice
Jason Calabrese
@jasoncalabrese
Mar 01 2015 21:29
there must be some published examples that we can use for the tests
Ben West
@bewest
Mar 01 2015 21:30
man ode ?
Jason Calabrese
@jasoncalabrese
Mar 01 2015 21:30
?
Ben West
@bewest
Mar 01 2015 21:30
I agree, I'd love to figure out how to test this type of stuff
man ode was supposed to be a bit of a joke
ode is an oss tool to solve differential equations, I guess
Jason Calabrese
@jasoncalabrese
Mar 01 2015 21:32
I was thinking that we take a published IOB curve that shows stacking, etc, grab some points and make sure our calculations match
Ben West
@bewest
Mar 01 2015 21:33
did you want some edits to that roadmap thing?
I'm hoping it reflects people's thoughts, would like to use it as basis for co-ordinating everything
Jason Calabrese
@jasoncalabrese
Mar 01 2015 21:36
I guess the big thing is if we want Nightscout to be the umbrella for everything
Ben West
@bewest
Mar 01 2015 21:37
doesn't matter to me what it's called, but think community as a whole needs to be able to point to unified roadmap
different groups doing different things on different timelines is fine
dexcom, helmsley, jdrf, medtronic... no one knows what to expect
those organizations work on committment based agreements
open source community has a tough time committing to deliverables
Jason Calabrese
@jasoncalabrese
Mar 01 2015 21:38
the other issue is that I think there is still a lot of personal itch scratching going on, as a side project we can't really commit to much
Ben West
@bewest
Mar 01 2015 21:38
that's not the way we roll
exactly
so hopefully what this does is help identify across all these projects? what is the point?
why are we doing it? there needs to be more on this I guess
that will guide the terms of "committment" means
we're not going to commit to time tables and specific deliverables
Jason Calabrese
@jasoncalabrese
Mar 01 2015 21:39
yeah, I liked in the other roadmap wiki page where we listed our current personal interests
Ben West
@bewest
Mar 01 2015 21:40
but we can say the kinds of things we're interested in
yeah, that's a low level thing
per project technical roadmap
this is high level: what is Nightscout going to be in 2 years?
so in terms of regulators, if they want to work with open source community at large
if they are really serious about "we want more experience with open source"
they can expect the diverse projects to kind of arch in this direction
Jason Calabrese
@jasoncalabrese
Mar 01 2015 21:42
we'll see how many people stay with NS when that share receiver is out, guessing that the pebble, raw data, care portal, and some other features will make it a better tool for some people.
if the less technical people leave I could see the roadmap changing and us focusing on building more of a platform that can be used for other projects
Ben West
@bewest
Mar 01 2015 21:43
sure, there's no mention of tooling in this anywhere
whole process since april has been introducing some features, but mostly tooling... unit tests, etc
you just got first bit of UI tested, btw
nice
trivially, but still
Jason Calabrese
@jasoncalabrese
Mar 01 2015 21:44
will be easy to move more into this
Ben West
@bewest
Mar 01 2015 21:45
that's huge landmark
Jason Calabrese
@jasoncalabrese
Mar 01 2015 21:45
all the utill like code can be factored out now
Ben West
@bewest
Mar 01 2015 21:46
should I stub out calendar for next 16 weeks with suggested calls/hangout discussions?
2 - 3?
Jason Calabrese
@jasoncalabrese
Mar 01 2015 21:46
sure
Ben West
@bewest
Mar 01 2015 21:46
cgm-remote-monitor is primed for some interesting things now
Jason Calabrese
@jasoncalabrese
Mar 01 2015 21:46
I'd like to have a E release planning meeting
Ben West
@bewest
Mar 01 2015 21:46
that's a huge turning point
Jason Calabrese
@jasoncalabrese
Mar 01 2015 21:47
think we're close
Ben West
@bewest
Mar 01 2015 21:47
combined with the refactoring experiments...
Jason Calabrese
@jasoncalabrese
Mar 01 2015 21:47
IOB + MQTT will be a nice release
Ben West
@bewest
Mar 01 2015 21:47
eg, we need to refactor soem stuff to avoid needing to change signatures everywhere
that's crazy
dependency injection is good, but needs to be attached to like app.db or something
Jason Calabrese
@jasoncalabrese
Mar 01 2015 21:47
we can have a shared model now
Ben West
@bewest
Mar 01 2015 21:48
in express, in your handler you can get at it through req.app.db.Entries or whatever
Jason Calabrese
@jasoncalabrese
Mar 01 2015 21:48
or maybe IOB should be 0.6.5, lots of people are going to setup iob-cob just for IOB
Matthias Granberry
@mgranberry
Mar 01 2015 21:55
I only barely use NS now, mainly for the iob-cob part or if I want to see the CGM data on my pebble. I'm a lot more likely to use it if I don't have to tether a device to my phone. You might find that the share doesn't reduce the audience at all.
Jason Calabrese
@jasoncalabrese
Mar 01 2015 21:58
thats true, if/when we can use BLE to connect to the receiver the system will be much easier to use. Just being able to charge and upload at the same time will be huge.
@bewest sent you a pm to reduce the clutter here
Ben West
@bewest
Mar 01 2015 22:05
yeah, 'share has weird limit of 5 followers, that seems to be a pretty heavy restriction
they also don't have clock mode... they don't quite seem to understand glanceable display
iob + mqtt would be good, yeah
Matthias Granberry
@mgranberry
Mar 01 2015 23:32
Too bad chumby is dead. They would be great for nightscout.