I assume that you guys are seeing what I call, “the flaky data pattern”? This is flaky data that will produce large errors if you verify against a BG machine.
I’m getting the sensor all confused here because I reset it’s clock. It knows its confused and throws “Error.” Do you see a “flaky data pattern” before failure? Or is it just a hard failure?
Good strategies and so true about the non reliability of equipment. I use perhaps the most expensive, yet maybe the least reliable sensors and pump… Medtronic. They are supposed to last a week. Maybe half have lasted that long, and I’ve been using them for over a year. Insurance is definitely the tyrannical overlord in the equation. I have maintained a surplus of supplies over the years by using the infusion sets days longer than “recommend”. Over the years I’m about 6 months to a year in surplus. Same goes for most equipment and insulin. Watch your carb intake and exercise decreases the volume of insulin needed. Keep that up for a while and you’ll have a surplus of insulin.
Also, I’d like to dispell a myth regarding insulin storage. I almost always store insulin in the fridge. However I also keep some out at room temperature, some vials for well over a year. It is just as effective as refrigerated. No degradation whatsoever. So if you’ve kept some out of the fridge for a while, don’t dump it. Chances are it’s just fine.
There’s a large misunderstanding about the replacement policy – including apparently among some Dexcom reps. Here’s the policy which says among other things “Dexcom will replace all sensors that we confirm as failures during the intended lifetime of the sensors through investigation. […] G-Series: Dexcom will provide a maximum of three (3) goodwill replacement sensors in a twelve (12) month period” (emphasis added) They clarify that “goodwill” means “sensor removed for a procedure, user application errors, etc”.
Don’t call for failures. Fill out the form. Often they email me asking for more answers – these partly duplicate what was on the initial form, but it’s easy enough to plug in the answers. Almost always, the next thing I hear is a notification of shipping the replacement. I think one time they called me, and even that didn’t delay the replacement. Using the form, I have never encountered a rep misunderstanding the replacement policy.
If the sensor reading and the fingerstick reading are more than 20% apart, that’s an automatic replacement. (I’ve never tried to determine whether it’s 20% of the lower number or 20% of the larger.) Quite often, my real problem is excessive variability – the jitters – what mohe0001 calls “flaky data pattern”. That’s not one of the reasons I can pick on the form – but when that happens, I can almost always watch until I think the sensor has a particularly bad reading, and do a fingerstick. Only takes one – well, they ask if you did two – and they don’t ask me for any proof of the fingerstick value and I see no evidence that they look at my sensor record. I have tried putting in “other” and have successfully argued that excessive jitters is a failure, but that’s a lot of trouble.
If a sensor fails at 239 hours, they replace it, no question. If it fails after 240 hours, IOW in the grace period, I don’t think they replace it, though I haven’t had an opportunity to test it. I have gotten replacements for a failure at 9-1/2 days.
Since going to G7, I have been getting my sensors through a local pharmacy (regional supermarket chain). I have found this much easier than dealing with a DME supplier.
I tried the back of my arm for my first three G7 sensors. One of them failed during warmup, and the other two gave radically excessive variability. I switched to abdomen. I suspect the reason is that I have very little fat on my arms. (Too much on my abdomen.) My endo is very thin and says she had similar problems using the G7 on her arm. I suspect that the short wire on the G7 means more jitters in the readings.
So I think the G7 sensor itself is slightly inferior to the G6. The improvement in the app more than balances this out for me. The 12-hour grace period means I can overlap sensors, thus giving the new one 12 hours to warm up but still getting a full ten days between sensors when there’s no failure – this is huge.
Note that you can run G6 and G7 simultaneously. I did this for a couple of months. Without that, the initial G7 problems would have been a disaster.
Terry, I think the G7 actually requires a bit more dexterity than the G6 to apply. At least, it’s different. The G6 applicator needs a fairly light index finger push. The G7 is triggered with the thumb, on a smaller button, and I find it a bit harder to locate and push.
That’s truer for newer insulins, but old insulins like R and NPH are brittle and degrade rapidly. Depends what your on.
Ok, Edward. I appreciate your real world experience and I will try out a few things you suggest to improve my CGM supply or at least not degrade it any further.
One of those things is using the online form to request a replacement (thanks for linking). I tried that a few years ago and for reasons I long ago forgot, decided that using that form was no simpler than calling in. What I now appreciate is that using the form makes it clear whether any failure is for a legitimate replacement or if it’s a goodwill replacement.
Thinking back, I believe some of the legitimate failure reports I made over the phone were administratively classified inappropriately as goodwill replacements. Maybe that made the transaction easier on the customer service rep or maybe the “language gap” between me, as a native English speaker, did not translate accurately to the Filipino/Filipina working the midnight shift to accurately assess my claim.
I will give the form another try. I like that your experience indicates that legitimate claims under 240 hours (10 full days) are absolutely eligible for a non-goodwill replacement.
I rarely called in without actual evidence (an actual discrepancy between the CGM sensor and a finger stick that easily exceeded 20%) of a non-goodwill failure.
Your comments about the G6 vs the G7, especially regarding the dexterity required, reinforces my decision to stay on the G6 for as long as possible.
Could you expand on this point? What exactly were the G7 problems you experienced at startup? Were they related to qualities that persist in the currently produced G7s or has the quality improved to eliminate those problems?
Thank-you for the detailed response.
This is how it should be, Edward, because that the definition of a failed sensor.
P.S. I use the following, but they have never asked for details in how I do my calculation. The fingerstick (assuming your using a reliable model of glucometer) is assumed to be the ‘actual’ blood sugar.
[ (fingerstick - sensor value)/[(fingerstick+sensor value)/2] ] x 100
(Percentage Difference Calculator)
I didn’t mention the one problem I have with the form. When I finally hit SUBMIT and it’s accepted, I don’t get any response. The browser just sits there with a spinning icon. Happens in both Firefox and Chrome, under Win10. I have learned that this situation means the form has been accepted and that I will receive a reply – sometimes in an hour, occasionally more than a day.
Yep. Before I learned about the form, I made quite a few midnight calls. I repeatedly contacted the same rep, who … spoke … very … slowly. She got the information right but it required patience.
Dexcom must have a large army of reps given the number of customers, I would suspect inadequate training over laziness.
I was just referring to the erratic readings when I placed the sensor on my arm, as mentioned elsewhere in the post.
There are multiple valid definitions of “20% difference”, for example
difference / mean > 0.2
difference / smaller > 0.2
difference / larger > 0.2
and I say this as someone who earned a BS in math a few centuries ago. The site you link isn’t controlled by Dexcom, so it doesn’t determine which definition Dexcom uses.
Some people suffer from math trauma so here’s the no math required way to figure out +/-20%: Is my Dexcom sensor accurate? | Dexcom
Pulling back the curtain on Dexcom reveals that they plot your BGM/CGM comparisons on the Parkes error grid as part of submitting every “non-courtesy” sensor replacement to the FDA MAUDE database. E.g. MAUDE Adverse Event Report: DEXCOM, INC. DEXCOM G6 CONTINUOUS GLUCOSE MONITORING SYSTEM Been a while since I looked, they definitely reported sensors in zones C, D, and E, I can’t remember, maybe B. I don’t know if the zone is part of the replacement decision.
If you want to check your own BGM/CGM comparisons I made a thing everyone can use.
Dexcom won’t inquire about this. They don’t push on this. You can use any definition you want. I typically just state, “more than 20% error.”
This is what Diabetes Technology Society uses to calculate the error in glucometers against the lab readings, I think. This is solid.
They describe their error grid here: The Diabetes Technology Society Error Grid and Trend Accuracy Matrix for Glucose Monitors - PubMed
Here’s the math I use to determine my 20% or greater inaccuracy that I base my Dexcom sensor replacement claim.
BGM = blood glucose monitor, CGM = continuous glucose monitor
(BGM-CGM)/BGM, then convert the decimal answer to percentage, doesn’t matter whether answer is negative or positive
@spdif, I played with the Parkes error grid that you linked to and found that
+/- 20% error was generally anything outside of the A zone. I think I’ll stick with my “street math” arithmetic to determine what exceeds +/- 20%!
I do it this way too sometimes. I’m typically working in excel. Sometimes I use different calculations in the same spreadsheet. It’s neat to see the different ways people do it. I’m gonna try a variety for fun.
I’m seeing such exemplary accuracy coming out of this new machine that I haven’t needed to delve into the details of the error for a while. I love this new machine!
I personally wouldn’t seek a replacement for any size error in the first 24 hours. Both the G6 and the G7 take a while to settle in for me.
I also would attempt calibration before replacement (but I don’t generally calibrate in the first 24 hours because it settles in better on its own in my experience.)
I realize that you’re saying that you CAN seek a replacement if there’s a 20% error not that you MUST but I just wanted to chime in that with some patience many sensors that read REALLY poorly at first become very accurate.
If you use the online form and pick “inaccurate readings” (actually phrased in wimpy words), then the form requests the sensor and fingerstick readings. (Forcing you to pick from a menu of numbers that’s hundreds of choices long, ugh.) I once submitted one that I thought was wide enough and got rejected. Unfortunately I don’t have a record of what the specific numbers were. Perhaps using the “other” choice or calling would enable one to just say “more than 20%”.
No basic disagreement here, although sometimes the error clearly indicates a failure. For example, I recently had one start out reading about 50% of true values. There had been some bleeding on insertion, so that was probably the reason. The bleeding on its own was not enough to reject the sensor, so it was easier just to report it as inaccuracy.
The first-day issues are why I use the “grace period” to warm up the next sensor. That gives it half a day to settle before activating.
I agree with you here. I also generally have good experience with reining-in freshly inserted yet inaccurate sensors (G6) with timely calibrations. Early (1st hour especially) calibrations work well. I limit these early cals to <= 30 mg/dL and, if needed, wait for at least three CGM relatively flat data points before doing a follow-up calibration.
Where I usually run into problems is along days 6, 7, or 8. If the CGM is running > 30 mg/dL + or -, then I enter a 30 mg/dL calibration. Unlike its first day agreeable calibration attitude, it will often only move the CGM reading a few mg/dL and then quickly, within 5 or 10 mins, move back to the pre-cal CGM reading.
This stubborn behavior baffles me. Not every sensor works this way, so I know that some sensors, at least, will respond more appropriately to my modest calibration cues.
This is exactly the kind of behavior, on Dexcom’s part, that has frustrated me. I was hoping that adopting the online replacement request would dependably get me the sensor that I need. Doing this without interacting with the more subjective judgement of the phone customer service staff is a plus.
I still intend to give the online form a sustained effort. Overall, Dexcom CGMs, have been a net positive to me. They are worth all this effort manage them.
Unfortunately, this behavior is still present in the G7.
To clarify, when I said I got rejected when I “thought” it was a wide enough gap, I hadn’t analyzed carefully. The memory is old, but it was probably more than 20% of the smaller reading but less than 20% of the larger reading. Now, I make sure the difference is more than 20% of the larger reading, and when applying that criterion, I have not been rejected.
I’ve never used the form before! Thanks for the warning. That would catch me completely by surprise.
This is unacceptable. How do we know that Terry hasn’t discovered something interesting that is worthy of having the company explore in order to determine the source of failure? What if other people have failures like this and don’t notice? What if people are running an Omnipod 5 off that bad data?
Sensor error is not OK. It’s not safe.
I think Terry has found an interesting technical project and a new responsibility to investigate this and report it EVERY time it happens. I think there should be a detailed spreadsheet evaluating sensor behavior that accompanies each report of failure.
Are you using tidepool? You have a digital way to record fingerstick values into your software, right? Where it shows the finger sticks clearly against the sensor data w/o requiring you to calibrate? If you calibrate, then you will loose access to the error. So, you might need paper records.
This is how the finger sticks look in glooko when I enter them into my Dash PDM.
It’s pretty easy to go back and retrieve the error from the graph, as long as I don’t calibrate.
How could anybody argue with you if you have the data to support what you are saying?
For the record, I would be enraged if I were Terry.
I appreciate your outrage on my behalf but I want you to know that I am not personally interested in spending any more of my attention beyond a reasonable pragmatic effort. Sorry that I’m not more engaged! I’ve noticed that as I’ve aged I’ve limited where I’ll expend my personal resources.
I do think our community has a legitimate issue with the inconsistent behavior of Dexcom’s algorithm. I don’t use Tidepool anymore. My digital “attention” is limited to DIY Loop, the Dexcom app, and the related Apple Health app.
Apple Health (AH), by the way, contains an amazing amount of diabetes data. Here’s a depiction of my insulin weekly usage showing total daily dose as well as the basal and bolus amounts.
I like your energy but I believe Dexcom is not really interested and I’ve decided that their system works well-enough for me. I just need to finesse through a few of their deficiencies.
We don’t have absolute knowledge, but as a retired software engineer, I’m quite certain that the software engineers at Dexcom are fully aware of the behavior. I disagree with the choice, but my attempts to comment on the software have gone into black holes. There’s also issues of FDA interaction.
I mostly agree, and this along with the many inaccuracies of the sensor is why I’m not on a closed loop pump. For some people, the sensor works well enough for this purpose, even if very imperfectly. Given how the sensor reacts to my body, I find it inadequate.
The worst case was a couple of years ago, out on a bike ride, feeling hypo symptoms even though the sensor was reading 120. Did a fingerstick, 53. OK, looping in exercise mode, presumably insulin would be cut off at 120. But what if the numbers had been 200 and 100? That would be dangerous.
