670G: Auto Mode, OMG

So I started auto mode just this week (Tues., Sep 5, for the sake of the archives). I also have a thread going on TUD about this, but @sam asked if I’d start one here, and I know some people favor this space over that one, so sure. Here are a couple of things that have stuck out for me so far.

  • My first day was the best, a foretaste of how it’s supposed to work, I think. This is because I started at about 10:30 a.m., with a very good fasting BG that morning, and I think I was about 110 when we flipped the switch. System really did a great job of keeping my lunch and dinner post-prandials almost negligible, and everything was jake overnight until… the Dread DP. Since then it’s been running much higher than I’d like. 180s before lunch today, 45 minutes after pre-bolusing. To some degree I was expecting this. My initial pump experience was much the same–worse actually–so I steeled my expectations for a longish adjustment period, and so far I have not tweaked any parameters from original settings at launch, though I’m already nudging my trainer to pleeeeeeease let me off that leash a little. Pretty sure Sensitivity is cranked way higher than it should be, though there are reasons for that. More about that later.

  • Guardian 3 vs. Dexcom 5. As it happens, my first week in Manual mode, the period when the pump is “learning” you but you haven’t switched Auto on, I was still wearing my Dexcom, and I kept it running in parallel while starting with the Guardian 3 sensor, just to see the comparison. I don’t have the wherewithal to give hard data on this right now, but basically I couldn’t see any difference. Mostly they bracketed my finger stick results, but neither was consistently the higher or lower number in that bracket. If anything the Guardian 3 was closer to the mark more often than Dexcom. I also had far fewer dropouts with the Guardian (good thing, considering what was going to be riding on it). This is was good news to me because switching to Medtronic CGM was one of my biggest reservations about the whole 670G proposition to begin with.

  • This is a VERY different system to a standard pump. You have to adjust your ways of thinking about how this all works. As an experienced pump user you think you know what all these different parameters like I:C, Sensitivity Factor and the like refer to, but in this new context they don’t mean the same thing, not exactly anyway. Sensitivity Factor for instance, which is one I really want to start fiddling with because I’m clearly not getting enough correction for high BG yet. In a standard pump I think of it as the thing the Bolus Wizard is going to use to calculate said correction. But in this pump, it’s also what the pump is using in the background when it’s calculating all those individual microboluses as it’s dynamically responding to your BG. In effect, it’s “correcting” all the time. So it’s one of those Small Change Can Have Big Effect things, as it ticks along. And there are lots of other examples.

  • No such thing as a basal rate. I gather Loop does something different that does make use of traditional basal rate settings but this doesn’t. The logic is that it’s always targeting 120 BG; there isn’t any “rate” because the amount of insulin delivered isn’t the point; the BG level is. So that’s what the algorithm is always aimed at. Which raises the question of…

  • Fixed Target: 120. This is the source of a lot of criticism, but a) it corresponds to a 5.8 A1C, which ain’t nothin; I currently run 6 or 6.1 and haven’t been below that since DX; b) it doesn’t mean that you never go below 120; I had an “alert on low” of 69 last night, and I’ve had it let me stay for several hours in low 100 range–it’s not rushing to push you up to 120. Other things being equal it generally seems to pause microboluses when you hit 100 but doesn’t alert or anything unless it senses a rapid ongoing decrease.

  • Adjustable parameters. Other than that 120 target, other parameters are adjustable and you can set different values for different times of day. Main ones are I:C ratio, Sensitivity, and Active Insulin Time. The last seems generally considered to be the most important in getting this thing dialed in. Right now mine is set at 2 (!), and my trainer just said she wants to tighten it up further to help with the DP problem. That’s pretty radically different from a standard pump setting–IIRC the default on my old pump was 5 hours, which everyone shortened to 4 or sometimes 3. But again, the microbolus thing: it isn’t using that setting just to calculate the effect of a single meal bolus; it’s using it continuously to calculate how much insulin to give you for whatever the latest CGM reading is. So if it “thinks” each little blip is going to stay in effect for a longer time, overlapping with all the others, it’s going to give you less. The more you tighten it up, the more aggressive it’s going to be.

  • Temp target: 150. Knowing about that 120 thing I was surprised to learn there is another fixed target you can use as kind of the equivalent to a Temp Basal. Came up in the context of exercise lows: my trainer pointed it out and suggested switching to it for an hour or so ahead of exercise.

Like I say, I’m still very much a newbie but happy to respond to any questions. If I don’t know I can always ask–those will probably be things I’ll want to know as well.


Awesome! Keep us posted… I really like keeping up to date on latest… it’ll be interesting to see how it goes over the next few weeks or months as the dust settles…

Is this something you could do if you wanted? When you say you are nudging your trainer, does that mean you are just trying to do what the trainer says? Or are you not actually able to change it yourself?

Oh my goodness this is something that I think would be HUGE for us. We use openAPS, and generally get very good results compared to manual mode of any pump. BUT to me one of the huge weaknesses of the program is its reliance on the notion of any fixed basal rate. Because ultimately you don’t truly every *know what your basal rate should be at any given time, it’s a guess. Responding just to trends would be so much better.


Ah, good question. No, I could do it myself–it’s not locked down. Partly it’s a matter of limiting stray variables and partly it’s just that it’s different enough that I don’t really have an intuition for it yet. With a standard pump you can raise your basal rate and there’s no ambiguity about what you’ve done. With this thing all the parameters interact with each other through the algorithm, which is itself kind of a black-box, so I’m hesitant about doing anything without full participation of my team.

I actually have two trainers, one from Medtronic and the other the Endo nurse in my endo’s practice, who’s just learning this system along with me. I should probably mention I’m the first patient in the Brigham & Womens system using one, so there’s also that. They really want this to go well too, so I’m sticking to the straight and narrow so there’s a clear track record going forward.


Might be worth mentioning I’ve joined the two main 670G user groups on FB. One is pretty open, the other one you have to kinda qualify for and is very hardcore. But the users in the latter are very experienced, many having started in the medical trials period, and very gung-ho serious about making it work. So there’s some real depth of support there. I’m just starting to be at the point of being able to make use of it.

1 Like

Are you able to fudge this target by lying about your CGM numbers when you calibrate?

For example, if you want your target to be 100, could you just always lie every time you calibrate and report 20 points higher than what it actually is? Would this work?

1 Like

@Eric, you know, I was thinking about this as a viable option but the issue seems like it would come from the “suspend on low” functionality. For instance, let’s say you were at 140 on CGM and dropping by 10, maybe the microbolus program would not see a need to suspend your insulin delivery. But if your 140 was actually a 120 because you’d deliberately and consistently mis-calibrated, then you’re actually dropping by 10 pretty fast and maybe it would be wise to suspend insulin delivery, only the 670G wouldn’t know that.

I don’t know how the 670G works, but with openAPS, there is a micro-bolus function, and the safety built into it is in assuming that if you start dropping, whatever extra insulin you gave up front as a micro-bolus can be subtracted from the insulin you’d normally get as your basal over the next hour or two. If that calculation is wrong, then all of a sudden you’re hitting the brakes too late… Of course, the Medtronic system may work on an entirely different principle, which may mean that gaming the system isn’t as dangerous. That’s a question I’m sure someone in the trial or who has it has begun to ask – would be really curious about the answers.


Hm… I think the problem you’d run into is that the CGM is sensitive to discrepancies between calibrations and what it thinks it’s getting from the sensor and can “reject” a calibration, for want of a better term, if it doesn’t seem to make sense. I think it would eventually conclude that something was amiss. I have heard that you can approximate a lower target by lowering your pause-before-low threshold, so that it lets you run lower without pausing delivery. Effectively you’d be dropping your average BG if you stay down there enough.


On the “learning” thing you spoke of, is that based on “patterns” in your schedule, like what times you eat and go to sleep and the amount you eat and things like that?

Or is it more based on how much insulin you need, IC and ISF and so forth?

Like I don’t have a set schedule for the time I eat or amounts I eat. All of my things change daily.

1 Like

FYI, here is @drbbennett’s thread on tuD:


Very much the latter. Of course it’s a black box, but as I understand it the algorithm is learning how your particular physiology responds to what it’s doing and uses that information to keep you centered on the 120 target. There is a thing under “Options” where you can enter “Event Markers” for BG, Injection, Food, Exercise, Other. One of my questions is whether or to what extent that information is integrated into the algorithm. The markers do show up in the reports, but it may just be that that’s all they’re for–still researching that one.

Same here. To some degree I’ve been restraining things just to keep it simple, but I’ve had commitments the last couple of days that have involved non-routine exercise and eating, and I kinda decided well, it has to be able to deal with that. If it was just a way of locking you into a rigid eating/exercise time and carb schedule that would be like stepping back into the R/NPH dark ages. Been there done that, no thanks.


I’m part of the #animaswentoutofthepumpbusinesssoimnowamedtronicuser and was given a choice of whatever pump I wanted. Because of the more positive than negative things I’ve read, I chose the 670G.

I’m excited and terrified at the same time. Excited for what it could possibly do for me, and terrified of not being in total control. I know it’s a learning process…

Well, it arrived today and I opened the box and looked at everything but the only thing I will open when I get home is the manual so I can read it.

From the date of your post I see you are no longer a newbie, so I just may be asking questions. Especially because I don’t find Medtronic to be very responsive.


Hi @PerfectHorse. :slightly_smiling_face:

I’m not sure if you were looking for drbbennett in particular, but I’m also here if you have questions. I don’t even wear the pump anymore, but I still know a good deal. And I promise to offer information only and leave my opinions out. I can do that. :grin:

Let me know if you need anything.


@PerfectHorse I’m also happy to help :slight_smile: I’m currently on the 670g (an old animas user as well! RIP), using manual mode. Though I was in Automode for about 7 months. Feel free to reach out with questions!


Thank you!

Okay so I’m now at 2.5 days with my 670G. I like it. In another 1.5 weeks I will train for Automode.

But I have questions please…anybody…

I’m also using the Guardian Sensor that works with the 670G. It’s approved for use on the abdomen and BACK of the upper arm (seriously? what are single people supposed to do?). My Dexcom was only approved for use on the abdomen, but thanks to the great help I received here, I wore it on my upper arm and my thigh (and actually got more accurate readings). Does anyone else use theirs in another place?

Can the sensors be “reset” like the Dexcom?

Is there a swivel clip to be had for this thing?


/hi @PerfectHorse!

I’m happy to answer questions.

I personally ask my mom or boyfriend for help with the back of my arm normally, but I used this video to help when I’ve been in a fix occasionally. It’s not perfect or easy but helpful if you’re a fan of back of the arm for sensor placement.

I wore my guardian sensors on thighs, stomach, love handles, back of arm, inside of arm, and forearm. I found the most accurate placements for guardian sensors FOR ME to be the top of my thighs and forearm. I’m happy to explain those placements further if they’re unclear!

Yes, but I haven’t heard of as long of extensions like the dexcom extensions CAN be. I’ve heard most people who extend guardian sensors will get around ~10 days TOTAL out of the sensor. Here’s an article on extending guardian sensors.

Not that I’ve seen, but if you find one I’d love to see!

How have the guardian sensors been working out for you compared to your dexcom sensors were?


Great that you have some time to contribute here again @LarissaW, great answer.

Am I now the go-to local 670g guide when @Nickyghaleb’s off running marathons??