Medtronic 670G: my endo clinic's feedback after 7 months

We had our endo visit today (or, more exactly, our Nurse Practitioner visit, since we see each one every 6 months).

The visit was actually really useful for us. We had some significant challenges, and we got some excellent feedback (I’ll report on that in another thread). In the process, we discussed the Medtronic 670G at length. The head nurse in the clinic is a good friend, and told me that they had 20 patients on the 670G.

Her feedback was quite interesting. She told us that we would not like it, because, based on what she sees with her advanced patients, it won’t allow us to control as tightly as we want to [clarifying EDIT] because it is dictating too much to the patient in terms of how to use it. She also says that it won’t work well for patients who don’t “do enough” to control their D either, because these patients need to do too much in order to stay in Auto mode: they don’t do it and the pump kicks them back into Manual mode, thereby removing the benefit of closed-loop control. So her conclusion is that the pump is failing both ends of the patient spectrum, and that it only works with a narrow band of patients: those who do take care of themselves but don’t have very good control.

I thought that was a very valuable take, in part because it is based on the observation of a lot of patients.

It appears to totally confirm @drbbennett’s personal thoughts on the 670G.


That’s strange. What do you need to do for it it stay in Auto mode?

She didn’t tell me details beyond the need to calibrate, so I am not sure.

That’s disappointing! Though not incredibly surprising. Fast-acting insulins currently approved for the pump don’t seem like they’d work fast enough for a pump to respond in time. Maybe eventually we’ll have insulin that works fast enough that makes this more practical. I’m excited for Fiasp to enter the U.S. market in 2018 and hoping it’ll be covered by my insurance.

1 Like

Well, the calibration thing isn’t a small deal for some folks—I would find that kind of annoying as someone who currently tends to calibrate about 1x/day on my Dex with good results. Would have to train myself back into the habit of promptly responding, which is ok, but not my preference.


True. A prediction algorithm is required.


Calibration is the main one, but it will also bump you out for things it finds questionable. For example, there’s a two hour limit on how long you can go without a microbolus, which can happen when you’re running persistently below 100 or so for a while. This happened to me a few times, mostly (inevitably) overnight.

@Michel–I think your endo is essentially correct, though she may be overstating how narrow the band is. My impression based on what I see in the FB 670 groups is that I see the satisfaction level rising as you get to people who have been in the ~7 A1C area pre-Auto Mode.

^^^ x 100

The trade off is that, in exchange for letting it do stuff for you, it locks up the controls you’d normally use to do it yourself. Some of this is inherent in how the thing works–some of the old parameters just aren’t meaningful in the context of a dynamically adjusting basal device. But some of it just feels like sheer nanny-ism. The need to get FDA approval means restricting its behavior to standards laid down by the medical establishment, and that establishment has always been uncomfortable with the idea of handing over control of a dangerous drug, insulin, to the patients to set dosages and administer themselves. Even in manual the number of clicks and confirmations you have to go through to do something simple like suspend verges on the absurd. If I have to have that last confirmation screen, couldn’t it at least assume that if I clicked the dang thing 9 times to get here, the default answer to the ridiculous question “Do you want to suspend?” should be “YES!!!”? Or even better, “No ■■■■ sherlock!!!”


If it kicks you out of auto, does it throw you back into manual? Would your normal basal profile pick back up? Or do you get stuck with no insulin until you reset something?

The NP told me it does.

We don’t always calibrate our son 2X a day either – and certainly not every 12 hours, as he’s not necessarily flat every 12 hours and doing so would actually be counterproductive. So if we were on this system we’d probably be kicked out at least 3 or 4 hours a day every day. Sucky.


This makes no sense.

Suppose you have a manual basal setup for x units an hour. After 2 hours of being below 100, the thing says “Oh no, this guy has been low too long. Need to switch back to manual mode…”

Now you are in manual mode getting the x units, instead of being in auto mode where it can still try to regulate you.

How is that a safety precaution? If you are low for more than 2 hours, switching over to manual would only make the low worse.



Doesn’t mean hypo, just means enough below target (120) that it hasn’t delivered anything for that long. And it yells at you with its outside voice when it bumps you out and tells you to check BG. If you’re actually approaching or over the lower end of your range, that’s a different behavior.


Are you saying that the predictive algorithm in a closed loop system is able to effectively correct for an incorrect dinner bolus?

Welcome, Katers87! Predictive algorithms work best with small differentials: in these cases, they are really great! With large differentials, they are not as powerful.

A predictive algorithm will eventually catch up with a widely inaccurate dinner bolus, but it may not be very good at it. On the other hand, it will do really well, in many circumstances, with keeping an even BG if large boluses are administered by the user with reasonable but not perfect accuracy.

Becausde it is predictive, it does not “wait” for the insulin activation time. it predicts, as well as it can (there are limitations), both the body reaction and the insulin action.


Ah, I can definitely see the benefit of that for some people. It seems like it’d be especially comforting to parents with kids that have T1 because they would know that if their kid’s blood sugar did shoot up, the pump would eventually bring them back down (regardless of how attentive the kid was).

I still think it’d be more useful if the insulin acted faster. I suppose that’s hardly a new idea though… any method of management would likely be improved with faster acting insulin.

1 Like

Faster insulin, and also the ability to interpret BG values more rapidly. I think relying on a CGM reading interstitial values is also a big drawback.

I’d love an intravenous pump reading actual BG values, and able to deliver either insulin or glucagon. Instant read and react. But that’s just fantasy land. They aren’t even working on it.


I suppose this is the closest thing, but yeah:

To me the crux is really the insulin lag, not the sensor. My frustration with Auto was that the response to a post-prandial spike was necessarily so sludgy because it takes so long for current “rapid” insulins to take effect. The pump is trying to do the right thing, but it’s like trying to steer a big ship with a tiny rudder. Gary Scheiner’s review goes to this point and is pretty near spot-on to my experience with it:


Thanks for the link to Gary’s review, it was very thorough and helpful.

1 Like

I just read an article about some non-invasive BG meter that is being approved in Europe which has similar accuracy to a CGM. IT requires calibration for two days and then none for two years. I’ll try to dig it up.
If it was a little less clunky I think that could be a better option than delayed CGM readings.