[real-time] CGM, Not Insulin Pump, 'Is What Makes Difference' in Type 1 Diabetes

As a longtime Dexcom user who has believed this ever since my first SevenGo took my A1C from 7.3 down to 5.7 it’s nice to see research published to support my personal experience.

The headline of the article sums it up pretty well- https://www.medscape.com/viewarticle/918568

Here’s the complete article, but it’s behind a paywall- https://care.diabetesjournals.org/content/early/2019/09/10/dc19-0888

Brief summary-
At 3 years — the longest duration of any CGM trial — real-time CGM was superior to self-monitored blood glucose (SMBG), or fingerstick, testing at least four times daily in reducing HbA1c in patients using both pumps and MDI, with little difference between the two insulin delivery modalities.

Only the Dexcom G4 or Medtronic Enlite sensors were used, not the Abbott Libre .

Only the rtCGM group had improvements in time-in-range and reduced time below range. Fewer patients using rtCGM experienced severe hypoglycemic episodes.

“It is not so important how insulin is delivered, but more important is how patients with type 1 diabetes monitor their glucose,” Šoupal said during his presentation.


It’s interesting that they were only testing four times a day. Is that still the recommendation? Before I got a CGM, I was testing 8-12 times per day.

I find that the CGM and pump working in tandem gives me the best control. My A1c dropped when I switched from MDI to a pump and then dropped again when I got a CGM. I’m not super convinced that I could maintain an A1c of 6.5% if I didn’t have both.


I’m sure there’s quite a bit of variability there, and I wish I had the full paper for this and all of the other things the article neglected to mention. I’ll try to download it via work and answer any specific questions I can later. But “There were no significant differences between the CGM groups” doesn’t necessarily mean that there were no differences - just that they didn’t have the statistical power to detect any. If fact, the abstract says the CGM+pump A1C was 6.9 compared to 7.0 for CGM+MDI and their time in range was also better (50.9–72.3%
vs. 48.7–69.0%), but those numbers are so close it could just be noise in the data.

Different things work for different people - for example I don’t think I could handle being connected to a pump and don’t think it would do much for me - but my confirmation bias just smiled a bit when I ran across this article and wanted to share.


I think we all have a bit of bias towards what works well for us. That’s an important thing to recognize when out interacting with people who may have had the complete opposite experiences as we’ve had.

I love my pump and can’t imagine going back to MDI. Not because I don’t like shots (I’m actually doing a minimalist version of the untethered regimen with Tresiba right now), but just because I find the ability to make rapid insulin adjustments on the pump so helpful.

But one thing with a CGM is that (at least in my experience) it’s not helpful unless you’re acting on the data it’s providing. I’m using a Libre right now and I scan it 20-30x a day, and it’s providing me at least as good control as the Dexcom did with its alarms. But it all takes active management. I’ve gone through a period when I wasn’t really actively using the Dexcom data, just relying on it for alarms (and then getting alarm fatigue and turning them up or off) and my A1c rose to above 7.0% pretty quickly.

1 Like

But one thing with a CGM is that (at least in my experience) it’s not helpful unless you’re acting on the data it’s providing."

Amen. I have yet to see a study that somehow controls for how the data is used and I think this would account for a huge amount of the variance in the data. If the Libre can record the number of times it was scanned that might be a good way to do it.

I’ve got a nice flatline when my basaglar is dialed in, so for me it’s all about timing and size of the meal
(or sometimes exercise)
bolus. For dinner if I don’t take my fast-acting insulin after my blood sugar begins to rise the insulin will hit before the food does and I’ll go low. So I watch it like a hawk and have been saved by the rapid rise alarm on Xdrip+ a few times.

But also amen to the alarm fatigue. I’ll mute my phone overnight when I’m oscillating around 80 because the damn thing will just wake me up every half hour for nothing.

1 Like

This result made a big splash at EASD, the big, European diabetes research conference, which is going on right now. --Joshua

1 Like

And I don’t think I could handle MDI as well as I do with pump. But we probably agree having CGMS enhances both our situations.

1 Like

I think the numbers in this are very important. Only 65 patients were observed (it wasn’t a controlled study). That breaks down to approx 15 in each of the 4 groups: CGM + PUMP; CGM + MDI; PUMP; and MDI. I don’t think I have enough faith in any sample size of 65 to believe any related claims are proven statistically significant. Would you take a drug that had been tested on 15 people? Would you even buy something on Amazon because 15 people said it was good?

The abstract says it was 94 people, not 65. Real-world clinical trial for 3 years is a strength, but not randomizing them is a significant weakness. If you’re still able to get a significant result in real-world use that’s huge- stuff that works great in a lab often doesn’t work in the real world. Group sizes were fairly small, but larger than 15 (rtCGM+MDI, n = 22; rtCGM+CSII, n = 26; SMBG+MDI, n = 21; SMBG+CSII, n = 25), and the effectively the comparisons made were between rtCGM (N=48) vs. non-CGM (N=46). Statistical significance was solid, not borderline with the worst one being P = 0.0002 and the other three being p<0.0001. Even if they didn’t correct for multiple comparisons like they should have that’s still going to be significant. Not definitive, but solidly supporting rtCGM use. This paper isn’t why I would use one, it’s one of many reasons why I would use one.

I’d buy something on Amazon if one person said it was good, as long as that one person is trusted and credible. I’d take that over 1,000 fake reviews any day.

I disagree completely. The CGM is a nice addition to diabetes management, I’d give it up in a heart beat over my pump, if forced to choose between them. Going back to multiple injections would drive me crazy.

1 Like

I was disappointed when I read this.

[real-time] CGM

I wish there were such a thing. What sucks is that I don’t think anyone is even working on it.

The 20 minute delayed CGM doesn’t mean much to me.

Real time? Yeah, that would be nice.


The first link above leads you to a 1 year 65 patient study. The second link to a 3 year study.

And don’t misunderstand me, I am not trying to say that a CGM isn’t incredibly valuable, more that I don’t feel the study being referenced is the best support for a CGMS value. I wouldn’t contradict “seat belts save lives,” based on"25 people lived after auto accidents while not wearing a seat belt, and “25 people wearing seat belts died in auto accidents.” Regardless of what logic may tell me, I would need a larger sample size to support ANY conclusion I wanted to draw.

Agreed they went with an inaccurate operational definition. Continuous reading would have been better.

But you raise an interesting question- what would it take for an actual real-time CGM? Could something like Dexcom sensor work if it was directly in the bloodstream? Could that type of sensor be inserted like an IV and if so how long could you leave it in a vein? The 20-minute (I thought it was 10-15) delay is such a minor issue for me I’ve never even really thought about it.

Huh. I’m still not seeing the 1-year 65-patient study. The first link takes me to the Medscape summary of the three year article and the second takes me to the abstract of that study on the Diabetes Care page.

Regardless, the best evidence for me using a CGM is what it does for me. N=1. If this study helps convince an insurance company to cover a CGM for their patients then that’s awesome.

But I think people can get a bit blinded by sample size and use it to dismiss data they don’t like out of hand. You can have strong effects from a much smaller sample if the variance is small enough. You can have overpowered studies (which we are seeing more and more in the era of Big Data) give you misleading conclusions that don’t hold up on replication. Medical device studies with smaller samples can be fine, and drug studies with huge numbers of patients can still have their data tortured until it confesses. That’s why I really wish I could get to the paper, but it’s paywalled to me for another 6 months.

I never did modern day MDI, since I started pump with R in 1994ish. Before that, used NPH, R, syringe injections mostly at home. I think I would keep dexcom, give up pump to switch to pens, maybe using tresiba and novolog. I would at least give it a try. But ideally, pump plus cgms, plus automated integration sounds best to me.


Im using eversense and it is much more accurate than the 14 day us freestyle libre(a better glucometer, not a cgm), and
More accurate than the Dexcom g5, there’s a slight delay when I’m eating and taking insulin, but I expect that, and every cgm representative and glucometer manufacturer expects delays during those times…I suggest trying eversense