Hacker News new | ask | show | jobs
by ijustlovemath 1084 days ago
Our device is for inpatient use, so it's a little more complicated than your typical DIY APS you might be used to as a T1/T2. You have to account for all kinds of different drug interactions, perfusion issues, undergoing surgical procedures, etc etc. The biggest single difference is that we use dextrose as a way to quickly recover from lows (like an automated orange juice dispenser).

Because we're in the hospital and can access IV lines, we also have rapid access to data, and the drugs we infuse get taken up much quicker (5-10 minutes for insulin, 3 minutes for dextrose).

The terminology is overlapping but the space is very different than outpatient glucose control.
1 comments
jablongo 1084 days ago
Why use dextrose instead of glucagon for lows? B/C patient liver function may be compromised more often in the in patient setting?
link
ijustlovemath 1084 days ago
In addition to liver function, you can't always rely on patients to have glycogen stores to draw from.

Additionally, dextrose is inexpensive, more available (rural county hospitals don't stock glucagon), easier to mix and store, has a longer shelf life, and most importantly, has a far quicker response time.

Glucagon has promise for outpatient work, where the volume of fluid is much more of a factor, though the stability and cost are still unsolved problems. The patients we treat are in a bed, monitored periodically by trained healthcare providers, with routine access to a pharmacy.

TL:DR Hospital control is just a different beast!

link
jablongo 1084 days ago
Cool - check out my company: https://replica.health - We automatically log a bunch of data for diabetics based on things like activity and location, and provide an LLM powered search engine for that data. You can ask our system questions w/ natural language like "show me data about whenever I get low post-exercise". If theres any crossover, I'd love to chat about it.
link
ijustlovemath 1084 days ago
I'm mostly curious about the modelling that goes on under the hood; are you just using deep learning, or also integrating something like UVA-Padova to fit your absorption curves?
link