Lying to the Dexcom for better calibration

Most of our sensors follow one of two models at one time or another: they are right on and stay right on, or they are 15-25 off, and see to be off again exactly that much at every calibration.

When our calibrations are very far off (that sometimes happens after a pressure low, if the sensor has been displaced enough), we sometimes simply reset the calibration by doing a quick double entry. But we rarely do that when we are somewhat off (say 25 off). For that case, we did figure out a method that gives us very good results. I have been hesitant to discuss this technique, but our calibration results are much better with it, so I finally decided to post what we do.

We only apply this method when we are 100% sure of our BG fingerprick numbers, which means, for us, that we have taken multiple fingerprick measurements over several hours. If we are reliably certain that we are, let’s say, 25 off, then we expect the next calibration to result in a starting error of about 12 (since the Dexcom typically halves an error when the real numbers are higher). So we add the expected error amount to the calibration we enter in the Dexcom, resulting in an error that is half the originally expected one.

Here is an example that happened this morning. A pressure low around 3:00am resulted in a signal that was 40 off: we fingerpricked twice to validate. We were tempted to reset calibration, but, in the end, decided to enter the real measurement. It resulted in the CGM being 20 lower than reality. A fingerprick a couple of hours later confirmed that the CGM was 20 low.

At wake-up time, we hand-tested again. The CGM gave us a read of 101, but the manual BG read was 127. We expected a post-calibration error of about 13 ([127-101]/2). So, instead of entering 127, we entered 140 (127 + 13 = 140) as a calibration. The CGM eventually calibrated to 122, which is 5 off.

Note 1: when I write that the CGM calibrated to 122, I don’t mean the number showing right after the calibration, but the “real” calibration which shows up a few minutes later.

Note 2: this technique could easily be mis-used. We ONLY use it when the error has been confirmed (a) by several manual BG measurements and (b) over several hours. It frequently happens, for us, that a CGM can be off for an hour, but back on later on.

[EDIT] Note 3: If we fully wanted to eliminate the mathematical error, we would add twice the expected error. But there is significant noise on BG measurements always, and I figure I do want to take advantage of Dexcom averaging measurement errors over a long period of time, so we only “cheat” by adding the error once.


Good tip, @Michel. I usually use the double entry. I will give your method a try next time.

1 Like

I sometimes give an Adjusted input BG, but not as methodically calculated as you describe.

1 Like

I’ve been out trying to look up “post-calibration error” to learn more about what you’ve written here when I realized I’m now a part of your site, and I can just ask. :grin: I know the Dexcom and Medtronic sensors are not the same, not by a long shot, but I was wondering if this “error” you were talking about is true of all sensors. I have lots of issues with my sensor, but the weak calibration match is at the forefront. When I attempt to calibrate at a 100, it might take me in at a 88. Other than being irritated by it, there was never too much of a consequence so I would just let it go. However, with auto mode, I would find I would receive less insulin because of this discrepancy, and that was much more than just an irritation. I began messing with 2 calibration techniques but was unsure of how “unhealthy” they might be for the sensor. I started to do something of a “step-calibration” to pull up a lagging SG when I was back to needing insulin but not getting it because of the lag (this was a back to back calibration), and I also started to do what I would think of as an “aggressive calibration” where I would bump up my number to try to take me in closer to my actual BG. I’m working completely with guesses here. Is there something I can actually calculate or even plan on as far as the “post-calibration error” you mentioned? It sounds like that explains what I see with my sensors, but I’ve never heard anything else about it.

Wow, that was long. Sorry. And thanks for the great post!

I used the old Medtronic Sof-sensors and it was routine practice back then to just give up and restart the sensor if it was way out of calibration. I can understand that you may need to tell lies to your Medtronic pump when calibrating to try to persuade it to read closer to reality. The reason they don’t directly switch to your new calibration is that they are trying to average out the errors in their sensor readings and your calibrations, on the theory that it would be dangerous to fully believe a single BG entry that might be incorrect.

I don’t think I’d try to use some formula to calculate how to lie to the pump to best overcompensate a faulty CGM reading. Your Kentucky windage approach seems likely to be as good as anything. If I remember, the old Medtronic calibration algorithm used a weighted average of the most recent 4 BG calibration values entered. They may still do that. The Dexcom is much better about being corrected, because it only uses the current calibration and the 1 previous one, so it never pays to restart the sensor rather than just calibrating again. (I think Michel explained it in another thread, something like if the BG for Dex calibration is lower, the Dex goes all the way there to protect against hypos, but if the new BG is higher than the Dex reads, they go half way there to mitigate temporary measurement errors.)

With my Dex (G5) I used to calibrate twice a day, and there was a fairly regular pattern that the Dex was reading about 10 or 15 mg/dL high in the morning and 10-15 points low in the evening. I decided that there’s probably some underlying physiologic reason such as a difference in hydration, so I switched to calibrating 1x per day. Now the calibrations are typically within 3 mg/dL which is good for looping.

This is by design @Nickyghaleb. @Michel developed a wiki which explains it. Calibration Wiki

Just read it. That’s VERY helpful. I posted the question, but do you know if this might be true for Medtronic as well? It’s almost spot on with what I see.

I do not unfortunately. We use the Dexcom G5. My guess is yes because I believe they follow much the same types of protocols and have the same types of built in algorithms, but I can’t be positive. I’m sure someone who uses the Medtonic system here will respond in short order!

I also used older MM Sof-sensors, and recall checking the ISIG factor before doing a calibration. Don’t know if that is still a factor in newer MM pumps.

With dexcom, you can delay doing entry and continue to get readings, where I think MM forces an entry to get readings.

With dexcom, I will sometimes enter the value, wait 5 minutes and enter same number. Often it jumps to very close to number entered.

as i read all of this problematic info about the dexcom, and knowing that in the past i experienced the exact same problem with erroneous BG readings and calculation issues, i am starting to wonder if i really want to bother returning to using it again.

i thought it would be a helpful tool for me, especially regarding my swimming, but i did a one week trial run from my endo’s dex sensor and i was constantly correcting.

i truly thought that the problem was me and my lack of body fat. but i figured that if eric can successfully use a dex, why not me. i read that a lot of people where the sensor on their arms and since my tummy isnt much of one, i tried it on my arm. the first 3 days it was giving crazy readings and the arrows were trending up and down like a roller coaster. whenever i was doing a finger stick, the #s were way off. but then on the end of the 3rd day, it began working well. well enough to get a sense of where i was at…but certainly NOTHING i would bolus over.

so, now after reading this thread…hummm. maybe its just not for me. i do about 20+ finger sticks a day and my last A1c was 4.9% so i must be doing something right.

if anyone has any suggestions, please chime in. i would love to hear from any and all of you.
thanks, DM

Wait and see if the G6 works any better. The sensor is changed so the potential for a significant difference in how it works with your body chemistry is certainly possible.

Otherwise even though I think the Dex is great (for us) the bottom line is if it doesn’t work for you and does not increase your overall quality of life then totally not worth it.

The main benefit I get is at night. Even if you can’t bolus from it, and even if it is off a bit, it can still wake you up if you are high or low.

At night, who cares if it says you are 200 and you are really 220. Either way, it can wake you up and you can bolus.

I know you usually set an alarm at 3am to check. Maybe the Dex would let you sleep when you are 100, and you would only need to wake up for highs or lows.

1 Like

this is my dilemma : my quality of life doing finger sticks has been wonderful. but i would love to get an Apple Waterproof Watch that i could swim with and be able to read my BGs as i swim; currently, i get out of the pool every 1/2 hour to do a finger stick and then jump back in. it would be nice not to have to do that. also, when i am prepping for the pool, after i detach from my pump, i test every 1/2 hour, and i would love to not have to do that. also, i would love to have the alarms for overnight; currently, i set my alarm for 3am and wake up to test to make certain that i am in target range. i can then correct if needed.


Totally understand. But the reality is the G5 does not seem accurate for you.
The G6 is going to be out soon. (General launch date is June 4.) I would suggest wait for the G6. Get the G6 and try it out. See how it works for you personally.

Otherwise, I think you will (again) be disappointed with the G5 enough that you may kick this issue to the curb for another couple years and potentially miss out on the new G6 which potentially might be exactly what you are looking for?

i think you said it perfectly. right on the money. i have a lot of patience for these things, so i wont bother wasting my time on the G5. i’ll wait for the G6 and see if i find that better suited for me.
(and BTW, eric thinks its a bad idea for me to bother with the G5; thinks i am doing so well on the finger sticks; so, why bother? )

thanks for your input. very appreciated.

1 Like

Unfortunately, the information will be close but not the same. Each company is going to tightly guard its algorithm, because that is really where the magic is. The sensor is just outputting a voltage that the device is logging and then the software kicks in. I would guess that they are close in operation, but only determined testing will show. The reason we have so much Dexcom info is many of us have been trying to understand the operation and have helped eachother learn. We don’t have many Medtronic adherents at the moment, but that is slowly changing.

I would argue the sensor is as much of the “magic” as the algorithm if not more so.

The sensor is part of the magic, i.e. the particular combination of coatings and so forth, but as far as it providing data, it is just a voltage. Having studied many of them, turning that voltage into actionable data is not simple. When we first studied these the algorithms were stupidly simple and the data wasn’t as useful. Especially when two different sensor will output different voltages at the same glucose level. That is why the algorithms are so interesting to me, because they still have inter-sensor variability, but they have the ability to learn and adapt. It is pretty cool

True. Voltage coming into the transmitter.
But the values of such are the most significant part of the overall system IMHO.

1 Like

And I am a self-professed geek, with all you nerd-mingos