And the percentage values will only be very high when the value is very low.
It's been interesting following discussion re: 505. I updated, and haven't noticed anything groundbreaking. I do see more small episodes of the numbers jumping around a bit, but as always, I find that the Dexcom provides a fairly accurate trend, and that's all. Trying to pinpoint values to say they are x% off is a game you will never win. I think the experiences of a few with accuracy concerns were most likely bad sensors, or just over-analyzing and over-calibrating. Just calibrate when it's way off, and leave it alone otherwise, as this has always been the best method.
I went back and reviewed the Dexcom G4 User's Guide and found this on page 50:
HELPFUL HINT: • Only blood glucose values between 2.2-22.2 mmol/L [40 - 400 mg/dl] can be used for calibration. If the blood glucose value(s) you entered was outside of this range, the receiver will not calibrate. You will have to wait until your blood glucose is in this range to calibrate.
I didn't read any other limits on calibration numbers.
Yes, but you better should count on users experience instead of the users manual ;) Try it out making the next calibrations when your bg is at 100-150 and you will see the improvement of accuracy.
That’s been about my observation after 5 days on the update. I’m seeing better accuracy from the first day. Maybe it was a lucky sensor, but I didn’t have the typical 12 hours of unreliable readings. I too have noticed the less smooth trend curve but I take that to mean it is “thinking” a bit more. I recommend the upgrade. Dexcom is a great product. As a LADA T1 about 1.5 years since dx, I am most grateful to have this device and my omnipod. Both great, but if I had to give one up I’d have to keep the Dex.
I agree that users can add pertinent information that's missing in any technical users' guide. I cannot, however, disregard the information given by the manufacturer.
Every word written in a medical device manufacturer's instruction manual is carefully written, vetted, and edited. In the US, it must also pass the careful review of the FDA. Unless and until my repeated experience directly contradicts what's written in a manual, I will not disregard it.
I take user experience like your report as anecdotal; I can try what you suggest and see if it works for me. I've worn Dexcom CGMs virtually non-stop since 2009. I will experiment with what you suggest but I'm skeptical that only calibrations from 100-150 work well.
This artificial pancreas 505 firmware is being used to make automated insulin dosing decisions in several ongoing clinical trials. That, to me, characterizes the blood glucose values that result from this update as more than just a good blood glucose trend indicator. I infer that this firmware is capable of producing accuracy good enough to dose insulin safely in the real world.
I agree that over-calibrating may result in degraded performance. How do you measure over-calibrating? The manual simply says that you feed the system two calibrations at the outset and then every 12 hours thereafter. It notes that other reasons to calibrate include "when your blood glucose value is very different from the sensor glucose reading." Not much detailed guidance there. Is 20 points off too much? How about 30? The manual offers no definition or warning about over-calibrating.
Over the years many here have written about their calibration preferences. I've adopted some of them but the written info from Dexcom does not explicitly support the practice. These include:
Never calibrate unless the trend is flat.
Don't calibrate if the sensor reading is within 20 mg/dl.
Don't calibrate more than twice per day.
Try to calibrate when your CGM trace is flat when it's low, medium, and high, especially in the first day.
I believe that the Dex G4 CGM with the 505 update is capable of more than being a simple trend indicator.
Does anybody have 2 receivers, one with the old firmware and one with the new firmware. It would be interesting to compare these 2 receivers being fed from the same sensor.
Yes, I do. I am waiting for the next sensor insertion to start them both at the same time. When I do that, I'll try the experiment and post some results.
I believe Lorraine is doing this experiment too... according to her blog.
Yes it is being used in clinical trials of an artificial pancreas...usually with two to avoid large errors. Also, note that the clinical trials are carefully supervised.... for example, in the Boston trials, 2 nurses followed the people in the trial around town and they were continuously monitored.
There is a difference between a medical experiment and what is safe to use for hundreds of thousands of people unsupervised.
There is no doubt CGMs are getting better and better and perhaps the G5 or G6 will be approved for dosing insulin.
I still see occasional errors large enough to cause a clinical problem if I had dosed insulin based solely on the CGM.
I still finger-stick a lot. It's an ingrained habit. If I need to add insulin or food because of my BG, I usually finger-stick. Although, I have been known to add small amounts of insulin in the middle of the night due to a mildly rising CGM high and wanting to get right back to sleep.
I agree, a medically supervised study is different than the real world. But I sense a pivot point now in CGM technology. We are evolving from never dosing based on a CGM reading to sometimes and headed towards always. Things are getting better.
The info I've been able to find on this online says exactly the opposite -- most accurate at the low end, least accurate for high numbers. I recall even reading something from dexcom -- maybe a transcript from and interview with an official or something -- stating that this was a design goal when they developed the system (actually, starting with the 7).
So, there's some vague, "that's what I heard" for you... take it for what you paid for it :-)
This artificial pancreas 505 firmware is being used to make automated insulin dosing decisions in several ongoing clinical trials. That, to me, characterizes the blood glucose values that result from this update as more than just a good blood glucose trend indicator. I infer that this firmware is capable of producing accuracy good enough to dose insulin safely in the real world.
All accurate observations, I think.
The one missing piece in judging how to use this is not knowing what the dosing algorithm/formula is. I'm pretty sure it's far more complex, factoring many variable that we don't account for when we simplistically apply IC ratios, correction factors, etc. For example, we all know that insulin sensitivity is not a static proposition but is in fact dynamic at all times. I'm certain the dosing algorithm can better predict/track this, and doses accordingly. Also, far more complex "profiles" of insulin delivery can be implemented with micro-boluses than we can program our pumps to do with the comaparatively "crude" delivery options we have (impulse, square wave, dual-wave, etc.)
Therefore, I don't think we can just make a simple connection between the "artificial pancreas" aspect of this software, and dosing based on what it says.
I don't think that matters though. If it's accurate enough, it's good enough for dosing decisions as a SMBG device. And that's what we need right now.
Mine's been telephathically accurate since I downloaded the new software a week ago. I'm on week three of a sensor, so it would be pretty accurate anyway, but it's definitely better.
I'll be starting a new sensor in the next week or so, so I'll get to see how the initial accuracy pans out then. So far, though -- better.
Actually, due to genetic differences, Canadian diabetes is an entirely different disease, caused by low phlogiston levels in the spleen, and poor phlogiston transport through the shadow limbic system.
I am using a different sensor and am on the second week with it. It has been quite accurate after the first couple of days. I’ve had some low and high numbers that have matched up exactly. I think that my problems were more sensor related than software related and I hope that continues to be the case.
