Dexcom G6 vs. G7: which is better for glycemic management?

Dexcom’s new G7 CGM and its claim of improved accuracy over the G6 (and all other CGMs) sounds great, right? And though it may sound counter-intuitive to ask, does greater CGM accuracy actually translate to better glycemic control? If so, how? If not, why not? Or, counter-intuitively, could it make glycemic control worse ?

It turns out, glucose levels in our bodies are far more volatile than we think, suggesting a paradigm shift in how we should think about glucose levels, from how to measure them to the way we interpret the data when making management decisions.

I just posted this comprehensive analysis on the challenges of actually measuring glucose levels, using Dexcom’s clinical trial data as a launching pad for analysis.

Bottom line: Accuracy is good, but precision is essential.


I thought your article was very interesting and the scattering of G7 points has caused me the same issues that you mentioned. For example one day after lunch the graph jumped up to about 180 which didn’t make sense based on what I’d eaten. So I took an additional bolus and 2 readings later I settled down to 140-150 which is what I expected. So not a huge overdose of insulin but still a mistake based on reacting to G7. Right now I can’t really compare my G7 time in range to G6 because I’ve had a couple of G7 sensors that started up LOW and stayed there for a couple of hours before ultimately failing. That quickly messes up statistics. I am increasingly getting to like G7 but find the insertion more difficult that G6 and also more traumatic to my tissue.

1 Like

I’m following this issue since I am in the process of accumulating G7 sensors for use with my DIY Loop automated insulin dosing system. I’ve gotten cold feet regarding the G7 and have reverted my every 90-day CGM sensor orders back to the G6.

I believe that precision is more important than accuracy for treatment performance provided that accuracy is reasonably good. Almost every sensor seems to experience some less accurate and dispersed data point performance. Loop seems to tolerate at least some degree of this. When it gets wildly off, I must open the loop and revert to manual mode.

As a 7-year user of Loop, I am confident, for me at least, that the programatic management of insulin dosing usually outperforms the human managed system. I agree with you that precision is more important than accuracy in order to make better management decisions. That is if accuracy is not too degraded!

To help the reader who may not be clear about the distinction between accuracy and precision, I cite this graphic.

When my current G6 transmitter expires late next month, I will trial a few G7 sensors and see how they go. I’ve been scared off by the online comments but I’m very aware that dissatisfied G7 users likely far outnumber the satisfied ones online. I also realize that I am not a typical (nor are you!) user of CGM data, so a trial is needed.


You say that you prefer the automatic system versus your own decision-making. But it raises the question: Do you think the loop makes better decisions because it’s smarter–that is, you’re unsure about what to do for those conditions where it’s making decisions–or is it because you simply don’t want to be bothered with having to pay the kind of close attention that would be required of you?

Many people would choose one or the other or both, but it’s an important distinction worth internal consideration. There’s no question that T1D requires very close attention to get in tight control, and that’s just not a realistic option for many. But it also requires considerable knowledge about what to do under highly diverse conditions. And that’s also not something people are experienced at.

that said, both are achievable. I think many don’t even consider it bcause they appear daunting, or literally impossible. That’s how I felt for the first 45 years of having T1D. Then when I saw someone else do it with what appeared very little effort (and didn’t seem bothered), I took the initiative to try it myself.

I think the G6+G7 experiment that I did (that prompted me to write this article) might be a similar experiment that can show some people (not all) that they may have a greater aptitude for it than they gave themselves credit for.

What wearing both sensors at the same time will do for you is motivate you to watch both more frequently, to look at how each one is handling different conditions (ie., the G7 shows big spikes, while the G6 is more gradual), and then witness the effectiveness your response to those readings in your own personally chosen actions–not that of the loop system.

I would be curious to see if you come away with this thinking that, yeah, maybe I can do this better than the loop.

1 Like

No one has the personal resource to monitor BG status every five minutes all day, every day. I don’t see the choice as doing all of the management myself or letting Loop do all the management. It’s a hybrid style that I use. I let Loop get in the trenches with all the detail and I monitor from more of a distance.

It permits me to pay attention to any needed changes of insulin to carb ratios, basal rates, and insulin sensitivity settings. I find these changes are less frequent but regularly needed. The every five minute decision that Loop makes regarding whether to add or withhold insulin makes me tired just thinking about it!

This overall management decision is personal and I salute everyone who arrives at that juncture. Unfortunately, most people just try to “wing it” and never really balance all the required factors.

What I’m trying to say is that, for me, it’s not an either or decision to make. I use Loop and my expertise in combination. I, alone, cannot do better than Loop. I feel it is the blend of the two that performs the best. For me and possibly for others, of course.


Terry – I think you may be overestimating what it means to “watch your readings.” I glance every 1-2 hours, not every five minutes.

The only time I watch more often than that is after meals, where I check every 15-30 minutes for about an hour… maybe two, depending on whether I need to worry about an unexpected rapid change, up or down. Or, when sugars go to extremes, such as hypo and hyperglycemic ranges, where I then bolus with insulin or carbs. Here again, I only check every 10-15 minutes or so, which is only necessary for a 60-90 minute window at most.

It’s also the case that I have my dexcom readings on my watch, so the effort required is only a small rotation of my wrist.

The point is, these small observations allow me to proactively take actions before things get really out of control, where I would then have to play catch-up… the dreaded roller coaster.

It may be that your self-described “hybrid” method of using the loop and your expertise in combination is exactly what I’m describing.

But to return to the G6/G7 decision, here’s an idea: Wear both sensors at the same time as I did, but continue to use the G6 for your looping and personal management choices. As you do that, passively watch the G7 readings and see if you think the looping behavior would change from those readings, or even whether your own decisions would change if that’s all you could look at.

This may be the best way for you to truly compare these two models, given your personal management style. It might also be a good way to evaluate whether the G7 might work for you before actually committing to it. I’m sure a single 10-day session with the G7 would give you that kind of insight.

1 Like

Your simplified “bottom line” is backwards.

Accuracy is how close to something being described the description is. Precision is how detailed the description is. For making decisions - the fundamental action of management, accuracy of the information used is paramount, precision is secondary. Accuracy determines direction, precision can sometimes help to refine that - to a degree.

Two examples of how accuracy and precision relate to making decisions:

1:A digital display that says the time is 9:01:01 all the time is more precise than an analog clock with hands that point toward hour and minute marks and has no second hand.

If I have to be somewhere at noon or suffer consequences, the digital display is wrong 99.997685185…% of the time . It’s precision is useless as a time indicator.

But If I know that the analog clock indicates that it is approximately 12 when the sun appears to be directly overhead, I can use its measurement and leave early enough to assure that I make my appointment.

2.In the middle of the night, if I can smell and see smoke, I know that there is a fire. Knowing that there is a fire is more important than knowing exactly where the combustion is, more important in deciding to wake the family, to pick the evacuation route, and to call the fire department.

Accuracy, precision and effects on PWD behavior - aka “glucose management”

Comparing the two CGMS, the precision of the G6 and G7 are the same - one mg/dL in the FDA-approved versions. MARD does not measure precision; it measures statistical accuracy. A MARD of 10% means that there is a 10% chance of any one reading being wrong by more than the allowable accuracy range that is the FDA minimum requirement. That is a measure of permitted INaccuracy.

Unless you are a statistician studying group effects, the difference between a MARD of 10% and one of 8% isn’t relevant to management decisions.

Gylcemic management. A CGM is a measuring tool. Tools don’t manage, people do.

The reliability/consistency of any tool is more important than its accuracy. If a tool isn’t reliable it isn’t trusted.

No CGM made today is reliable enough or accurate enough to bet my life on its readings alone. I can’t rely on a CGM sensor -even for trending information - until after I have confirmed its behavior with a BGM. I have had enough troubie with wonky and slow starting sites to ever trust a CGM sensor during the first 6 hours after insertion.

My distrust originated with the CGM manual. It says to use a BGM when symptoms “aren’t consistent with the CGMs reading”. But I can be far above target range with no symptoms, while I have to be below 75mg/dL to have any. How trustworthy is a measuring device that is inaccurate by up to 20 mg/dL low, if it triggers a low alarm at “70 mg/dL” when I’m at 90 or 50?? That is why I switched from using Dexcom’s and Tandem’s phone apps to Xdrip+, which lets me set a series of level alert tones. They keep me trend-aware without looking. I can evaluate or independently measure BG to determine what’s happening.

CGMs have limits, but knowing the G6’s limitations I can use them confidently. The basis of engineering is applying imperfect knowledge and known behaviors within limits to predictably perform needed functions.

A measuring tool has less to do with the results of a user’s decisions, than the user’s knowledge and ability to use that information. Confidence that the measurement is trustworthy determines how confident they are in using the measurement to do something

The most important thing to determining glycemic control is what the person does with the measurements they have. That is based on the person’s education about diabetes , their experience of how their life is affected by glucose level, diet and exercise, their understanding of the relationships between them, and their willingness to modify their behavior.

The pandemic and my retirement gave me an opportunity to study and experiment which few working adults have. Having used MDI for decades, with very limited hypo, gave me the confidence to try to do things that an endo today wouldn’t try with an old man. I decided that I should adopt using a pump before I needed one to survive. When started, I was given an A1C target of below of 7.5 - higher than I was getting with MDI. The initial pump setting were determined by the endo,. Within a day I was having problems - during a weekend. Within two days I was far above range.

I’m a slow learner, have average intelligence and a mediocre formal education. Over a long time, l became an avid reader, a competent writer and an adequate engineer. So the simple tech and programs used in CGMs and pumps today doesn’t intimidate me. I waited for 10 years for the tech to almost catch up with what I can do without it. So I understood the three settings on the pump and within a day had gotten back in range.

I spent a few months tuning my diet and starting to do measurements, before I found that the book Think like a Pancreas had a simpler way of doing it. Since then I have been dealing with a chronic issue - declining good infusion/sensor sites because of weight loss and old scarred injection bands. But the problems I’ve had have given me a limited expertise in using CGMS and the hours of reading research about them has given me a better understanding of their limitations, “care and feeding”, adn their utility.

My most recent C6 sensor took 24 hours and 2 calibrations before it stabilized and became accurate enough to use. It was immediately obvious that the new sensor was reading far low (50 vs 120mg/dL )I disabled Control IQ and the CGM low alarms. I slept confidently with my pump using my basal profile and made a small correction around 3AM. After the sensor started giving steady readings that were plausibly consistent with my BGM, I did a calibration and re-enable Control-IQ. My TIR declined for a day due to its completely erroneous low readings, but my control didn’t waver. readings, but my control didn’t waver.

This isn’t “rocket science”. Aside from packaging, less reliabilty, time delays, and indirect measurements, CGMS use the same kind of chemical reaction as BGMs. The primitive programs, “algorithms”, of CGMs and insulin pumps don’t do anything we weren’t told how to do with BGMs and syringes - and side from doing it more frequently, they don’t do it as well as we could.

We don’t need to do a calculation to see that a dropping line, if continued, will drop down to a level. It’s the same tracking ability we use watching a ball fly. If we aren’t prepared and willing to use what we know and what we can do, results we get from using tools will be limited by the tools limitations.

You see this in every clinical trial of CGMs and insulin pumps. Before and at the conclusion of clinical trials, those who have A1Cs below 7 are the minority (under 20%) those below 6 are outliers. 80% are above 8.0 and overall they drop by 0.5%. Outcomes haven’t changed in the past 20 years despite newer insulins and easier to use tech. What has changed is the cost per person for the same results has dramatically risen.

Bottom line: You can give people better tech, but you can’t make them think about what they don’t know or don’t value. Better tech won’t make them use it more effectively.


There’s a lot to your post, so I’ll try to home in on the key points.

First, the analogies you use about measuring time, temperature, and so on, all have something in common: there is no variability. The time “is what it is,” which is also the case with temperature, and so on. At least, within a small region of space. Here, your level of accuracy is determined by how closely it pairs with a standard measuring device within that same unit of space. that is, you can measure the time or temperature from a unit of space, and can rely on that measurement being standardized.

Glucose levels in the body are nothing like that. A single “sample” of biological fluid may have considerably different amounts of glucose from a different sample right next to it. What’s more, as short amounts of time pass, each sample may also be quite different.

All glucose measuring devices can do is measure the glucose they can detect from a given sample they’re given, but that’s it.

MARD is the difference between YOUR measuring device and another device that is deemed to be an accepted reference measuring device. But both rely on consistency between samples. If you were to use literally the same sample of fluid in each of the devices, you’d more likely get similar readings. That’s “accuracy.”

The problem with measuring glucose levels in humans is that glucose varies dramatically between individual samples, and glucose moves rapidly (in time) and dramatically (between samples). Compound that problem with the unequal distribution of glucose throughout different parts of the body, and you get the phenomenon that I described in the article.

MARD is a good way to compare measuring devices from sample fluids. Before CGMs, that’s the best you could do.

But CGMs blow up that measuring paradigm from a single measurement, such as a BGM, to a new paradigm, tracking systemic glucose movements throughout the body over time. Both of those (space and time) are like going from black and white photos to 3D images in virtual reality. The paradigm shift is enormous.

The best way to manage glucose levels is to track systemic glucose movements throughout the whole body, which is just not possible with reading individual samples. Even if each reading were 100% accurate, the volatility is so noisy, you just can’t make any sensible decisions, either in dosing or carb corrections. You have no choice but to use an algorithm that attempts to use those individual readings – which are NOT 100% accurate – along with prior trends and statistical probabilities to infer those trends.

Suffice to say, there will be a lot of errors in those estimates. And your case example is not dissimilar from my own, and everyone else’s.

But it’s the best we can get from these devices. I have found my personal management choices are more effective using the G6’s attempt at doing that than the G7’s.

I get what you’re saying. I do know that experienced sugar surfers can simply monitor the less frequent trends and be able to make the correct adjustment in a timely manner. What I was trying to say was that an automatic insulin dosing system can make rational micro-adjustments every five minutes and be correct a high percentage of the time. Do I think a sugar surfer can do this, too? Yes, of course – but I do think there’s more effort and attention required. That’s just my opinion and preference.

It’s personal style and choice, there’s no right or wrong. You live with the results you get.

We are the small minority who get decent results. The statistic that mystifies me is that 80% of T1Ds have an A1c > 7.0%. Most of our cohort does not use either of our approaches to management. I do believe people learn from a site like this – it’s certainly played a role in my diabetes education.

I find your posts interesting and our differences not as wide as it appears.


Indeed – we’re more on the same page than not. let’s not lose sight of the main thesis, though: The G7’s high volatility in glucose values is going to make it very hard for looping algorithms (or sugar surfers) to truly know what to do. If several adjacent readings show many divergent ups and downs, the surfers and the looping algorithm may inadvertently dose a larger amount that would be necessary (just as I did in the example I gave in my article). To avoid this problem, one has to wait longer for more G7 data to arrive before one can truly discern the more truthful systemic glucose movements. This wasn’t a problem with the G6, since each reading was more “reliable” insofar as an indicator for speed and direction.

Whether surfing or looping, the G7’s data presents too many problems and no useful information in small time windows. Macro-level movements are all fine, of course. They match the G6, so my bottom line assessment is that you will never do “better” than the G6 in your management, but you could well do worse.

1 Like

Thank you for doing this experiment and raise the question.

I do not have an answer for you, but it made me think of how the G6 used to smooth all but your current G6 reading on your trend graph to make it easier to track trends on your graph. They stopped data smoothing with v.1.9 of the app. Here:
Dexcom data smoothing

Maybe an idea to bring back if enough users of the G7 experiences the same as you did?

This is a bit on the side, but my G6-readings ofte look like your G7-readings do when I´m dehydrated or under a lot of stress. My diabetic nurse says she sees curvs like that from people going through crisis in life or in other ways experience massive bodily stress.

1 Like

My G7 is sometimes correct and sometimes off as much as 50 mg/dl. I wouldn’t dare use it for insulin dosing. I use a finger stick for insulin dosing. I the dexcom because I have hypoglycemic unawareness.


@argv I have absolutely noticed that with the G7 I need to consider more data before deciding on an insulin dose. I’m MDI after giving pumping a good long run. I like the control I get with MDI (same as with pumping, actually) but with less work as I no longer have to consider if the pump has it right. The G6 was less volatile over, say, a 15 minute chunk of time, but I’ve not experienced wild jumps that would have significantly changed a dose.

Mostly an aside here with my global perspective on my self care - it’s not about G6 or G7. It’s all about what I’ve eaten lately. If I’m having a bad day it’s tempting to lay it on the technology, or insulin action, lack of exercise, too much exercise, how long my insulin has been in use, etc. - but that’s mostly obscuring the real issue that when I make risky food choices I’m going to struggle more whether on a pump or MDI. All my bad days are food related. The best way I’ve found to get good glucose control is to take full control of all of it. It’s also the easiest way. Go figure.


Found your post and this thread very interesting. I wonder if a similar comparison was done along with meter BG testing. (cost prohibitive to do and that would be an awful lot of finger sticks) In all of the thread comments are we assuming the accuracy of both G6 and G7 devices or just in regards to comparison of the two devices?

Thanks for this #argv!

The smoothing the G6 is doing is essentially mini-calibrations, every 5 minutes. When I calibrate Dexcom to more closely match a meter reading it’s pretty simple math - Dexcom adds the latest reading, my meter reading, then divides by 2. With each successive calibration it’s the same. I found even when Dex was wildly off that after three successive calibrations it would be within just a few points from the meter reading.

So on G7 if my readings are varying over the last 15 minutes, with say 100, 92 and 125 I could add those three numbers, divide by three, and that’s what I’d dose for.

My article was comparing the G6 and G7, but the principles discussed in my article apply to the larger concept of glucose in the body: It’s highly volatile, and not evenly distributed. Therefore, it’s hard to track systemic glucose movements from individual readings. What you really want to see are glucose trends–these are what allow you to track systemic movements of glucose changes in your body. The G6 is very good at showing trends, whereas the G7 requires far too many consecutive readings before useful trends can be identified.

For the same reason, blood glucose testing (finger prick tests) are also “single value” measuring instruments–they don’t tell you trends unless you, well, act like a G6 and take a ton of measurements and do math to figure out direction and rate of change.

@argv This login is Dave Heller, right? I somehow had my email address lifted from this conference and am now getting articles, written by you, directly in my inbox. Is that appropriate? I’d be happy to read your content in this forum but I don’t want to be added to your email list or subscription service. Please remove me. Thanks.

“Dan”, not “Dave.”

I think you added yourself to my substack mailing list, as your name looks familiar. As I am just starting my substack, I am currently cross-posting to a few T1D platforms (such as this) to gain an audience and traction. Hence, you’re getting two copies of my articles if you also sign up for tudiabetes lists as well.

You’re welcome to unsubscribe from my substack list, of course. I don’t manage the list, so you’ll have to do that using a link in the email you receive from when I publish on substack.

I appreciate that you don’t want to receive two copies, and you probably won’t be alone in this feeling. Therefore, I am not likely to cross-post like this in perpetuity, at which point, you’ll only get the newsletter through substack.

Not sure what to recommend in the interim, but am open to suggestions if you have any.

Dan, (sorry about the Dave), I didn’t sign up for a substack mailing list. In fact, when I click on the Subscribe Here link it only presents a place to enter an email address to subscribe. How can I unsubscribe if I never subscribed? Is my only option to block you? That seems hostile but maybe the best way to go? Sincerely looking for advice on how to get off this list.

I’m not sure, but I think the “subscribe here” link you’re referring to is part of the Substack newsletter you got, not on TuDiabetes. So, if that’s what you did, that’s how you got on my list. (If not, it’s hard for me to say how you got on the list)

In any event, I looked up this problem on substack and they said that it’s a common mistake when a newsletter is posted to other sites, usually because someone else cross-posts. (In this case, I did.) Anyway, when that happens, people think they’re signing up at the other site, but are actually subscribing for the substack newsletter.

If you have the email that came to your inbox with the article you received, there should be a link to unsubscribe in that email. If you don’t see it, click on the link at the top “open in browser.” That will give you the same article on the substack page, where you will then have a link to unsubscribe.

I tried it, and see that the button/image says “Subscribed” – click on that to unsubscribe.

Let me know if you still have questions/problems.