Sunday, April 19, 2015

I love the Libre when it misbehaves!

UPDATE 20/04/2015: a lot of info about the methods used can be found in this Abbott patent. Very interesting and quite longish patent. One of the scenarios makes use of both the external thermistor and the internal TI thermistor. There's also a mention of a possible in-sensor thermistor. If you read the patent, you'll see why a straightforward temperature interpretation is hard to come by.

I just love when the Libre misbehaves! Really!

Let's come back to the temperature impact and the predictive nature of the Libre again.

Same very obvious scenario tonight: Libre checks at 109 then 111 mg/dL, slowly rising (expected). A few minutes later, BG meter check is at 125 mg/dL. Perfectly in line with the slowly rising trend and the RAW data interpretation I have been using since day 1 with this sensor. So far so good.

Then, Max has a very quick warm bath (Call of Duty time depends on the bath not being too long - always a guarantee for results). He dries and dresses himself and comes down.

Libre spot check at 194 mg/dL: perfectly in line with what extrapolation from past data would give. In this case, linear regression and derivative based methods give roughly the same result. Pick your points between -16 and -7, you will still end up in the same neighborhood regardless of the method.

Blood check at 118 mg/dL: BG is actually falling. We have a 50+ % difference between the spot check and BG. RAW data also seems to be falling and the trend of the fall fits with what you could expect from a standard IG-BG delay (although this could of course be due to the sensor cooling).

A couple of minutes later, as we want to track how fast the Libre spot check witll correct itself, the sensor gives us its "No result available, check in ten minutes" standard message. Yeah, we kind of expected that, didn't we?

Post mortem, here's what the Libre reports it saw. There's a possible mild overshoot at 152 mg/dL, not severe enough to be conclusive, but not in line with the later BG meter check. The 192 mg/dL and then the hypo (real) that came a bit later. I have added what I believe to be the real trend line based on BG Meter testing.

I really need to compare the response curve of that thermistor with previous Libre thermistors. If Abbott uses a look up table to compensate for the temperature, variability in thermistors could explain why this phenomenon has been as striking with this sensor.

Over enthusiastic predictive algorithm? Non-linear or bogus factory calibration curve? I have my opinion, but please make your own.

I will probably have a more in-depth look at what I have collected during this run in the next couple of weeks, but our next run starts tomorrow and it will be Dexcom + xDrip. It may be a while before I come up with new tests.

No comments:

Post a Comment

Note: Only a member of this blog may post a comment.