MiniMed 670G: Performance, Auto Mode, and User Challenges

The Medtronic MiniMed 670G was the first commercially available insulin pump to automatically adjust basal insulin delivery based on continuous glucose readings, making it a landmark in type 1 diabetes management when it launched in 2017.¹ Its “Auto Mode” brought real, measurable improvements in blood sugar control for adults, adolescents, and children as young as two. But the system also earned a reputation for demanding a lot from its users, and those demands pushed a significant number of people to stop using its automated features altogether.

How Auto Mode Works

The 670G is called a “hybrid” closed-loop system because it only automates part of insulin delivery. The pump reads glucose levels from a sensor worn under the skin (the Guardian Sensor 3) and adjusts background, or basal, insulin every five minutes to keep readings near a target of 120 mg/dL. What it does not do is handle meals. You still need to count your carbohydrates and tell the pump to deliver a bolus before eating, just as you would with a traditional insulin pump.² The system consists of the pump itself, the Guardian sensor and transmitter, and a paired blood glucose meter used for calibrating the sensor.³

The “closed-loop” part refers to the feedback cycle between sensor and pump. When glucose is rising, the algorithm increases basal delivery. When glucose is trending down, it reduces or suspends it. In a truly closed loop, the system would handle everything, including meals. The 670G sits at a halfway point, hence “hybrid.” That distinction matters because it sets expectations: this is not a device you strap on and forget about.

Glycemic Improvements Across Studies

The pivotal trial that led to FDA clearance enrolled adolescents and adults with type 1 diabetes for three months of in-home use. Compared with a two-week run-in period, participants saw more time with glucose readings in the target range of 71 to 180 mg/dL, lower HbA1c values, and no episodes of severe hypoglycemia or diabetic ketoacidosis.⁴ Among adolescents specifically, time in range jumped by about 14 percentage points over three months, and HbA1c dropped by 0.75%.⁵

Real-world data told a consistent story. An analysis of over 3,000 U.S. users who completed three months on the system found that average time in range rose from about 66% in Manual Mode to roughly 73% in Auto Mode, while time spent with glucose below 70 mg/dL fell from 2.7% to 2.1%.⁶ A larger European dataset of nearly 15,000 users showed similar results: after starting Auto Mode, mean time in range was about 72%, with a glucose management indicator (a sensor-derived estimate of HbA1c) averaging 7.0%.⁷ These users spent roughly 81% of their time in Auto Mode, suggesting most people found a way to stay engaged with the system’s demands, at least in this dataset.

A twelve-month observational study from Denmark tracked 62 adults who went through a structured training program before starting the 670G. After a year, HbA1c fell from 7.4% to 7.1%, time in range climbed from about 59% to 72%, and time spent in the mild low range (54 to 70 mg/dL) was nearly halved.⁸ The improvements were tied to spending more time in Auto Mode and to total daily carbohydrate intake, which makes intuitive sense: the algorithm works better when users consistently interact with it and when meals are adequately bolused.

How the 670G Performs in Children

The 670G was studied separately in children ages 7 to 13, and results paralleled the adult and adolescent findings. Three months of in-home Auto Mode use was safe, with reduced HbA1c and more time in the target glucose range.⁹

More striking was the trial in very young children, ages 2 to 6. These kids started with an average HbA1c of 8.0%, which dropped to 7.5% over the study period. Time in range rose from about 56% to 64%, and time above range fell considerably, from 41% down to 33%. Overnight glucose control improved further, particularly between midnight and 6 a.m. Throughout the study, there were no episodes of severe hypoglycemia, no diabetic ketoacidosis, and no serious device-related adverse events.¹⁰ For parents of toddlers with type 1, overnight glucose management is among the most anxiety-producing aspects of the disease, so the safety profile in this age group was meaningful.

A systematic review that pooled data from children, adolescents, and young adults concluded that the 670G was safe and effective at improving glucose control for up to twelve months, with HbA1c reaching as low as 7.1% and time in range up to 73% in some groups. Time spent in hypoglycemia was described as nearly negligible.¹¹ The review also noted that follow-up beyond a year was lacking, leaving open questions about long-term outcomes.

Why So Many People Stop Using Auto Mode

Despite the glycemic benefits, the 670G developed a reputation for being burdensome. The numbers on discontinuation are sobering. In a cohort of 92 young people, 30% stopped using the hybrid closed-loop features within six months. Most of those who quit did so between months three and six, suggesting that users gave the system a real chance before deciding it was not worth the hassle.¹² Youth who stopped rated calibration difficulty, the number of alarms, and the sheer time the system demanded as the biggest problems. Qualitative interviews turned up recurring themes: error alerts, the system “not working correctly,” too many fingersticks, and, for some families, the ongoing expense of sensor supplies.

Among those who continued using the system, Auto Mode usage itself eroded over time. At one month, youth averaged about 66% of their time in Auto Mode. By six months, that had dropped to roughly 51%.¹³ That means even the people who kept the system were spending nearly half their time in Manual Mode, which largely defeats the purpose of the technology.

Adults showed a similar pattern. In one real-world cohort of 79 adults, at least 28% had abandoned Auto Mode by three months, and by twelve months, a third of users were no longer using it. The single most common reason was sensor-related problems, cited by 62% of those who quit. Difficulty obtaining supplies, fear of hypoglycemia, and a preference for returning to injections accounted for smaller fractions.¹⁴ The primary motivation for those who stuck with Auto Mode was straightforward: they wanted more time in range, and the system delivered on that metric.

Sensor Accuracy and the Calibration Burden

Much of the user frustration traced back to the Guardian Sensor 3, the continuous glucose monitor that feeds data to the pump’s algorithm. The 670G required fingerstick blood glucose calibrations at least every twelve hours, and the sensor performed measurably better with three to four calibrations per day. When calibrated twice daily, the sensor’s overall mean absolute relative difference (a measure of how far its readings stray from a reference blood glucose) was about 10.6% for abdominal placement. With more frequent calibrations, that improved to roughly 9.6%.¹⁵

Those numbers are reasonable for glucose sensor technology, but they created a practical catch-22. The system needed accurate sensor data to stay in Auto Mode. If a calibration was late, or if the sensor and fingerstick readings diverged too much, the pump would kick the user out of Auto Mode with an alarm. Getting back into Auto Mode required another calibration and a waiting period. For people with busy lives, and especially for parents of young children, this cycle of alarms, calibrations, and Auto Mode exits became exhausting. Several studies cited alarm fatigue as a distinct contributor to discontinuation, separate from any specific technical failure.

Psychological and Quality-of-Life Effects

You might expect that a system delivering better glucose numbers would automatically improve quality of life, but the evidence is more nuanced. A study of youth and their parents found that most survey measures of sleep and quality of life did not change significantly after starting the 670G. The exceptions were modest: parents reported moderately improved sleep quality, and adolescents reported less fear of hypoglycemia.¹⁶

A year-long observational study in adults painted a more detailed picture. Physical functioning scores improved by about 23%, worry about hypoglycemia dropped by roughly 25%, and diabetes-related emotional burden decreased by about 45%. Those are substantial shifts.¹⁷ But the same study found no improvement in general wellbeing scores (using a standard WHO measure) and no difference in sleep quality or sleep duration from baseline. The system appeared to relieve specific diabetes-related anxieties while leaving broader well-being essentially unchanged, and potentially adding new stressors in the form of alarms and device management.

That split outcome helps explain why satisfaction with the 670G was so polarized. People who valued tight glucose control and could tolerate the device’s demands often became strong advocates. Those who found the alarms and calibrations more stressful than the glucose improvements were worth ended up either dropping Auto Mode or switching to a different device entirely.

Using the 670G During Pregnancy

Managing blood sugar during pregnancy is critical, and the targets are tighter than for non-pregnant people with type 1 diabetes. Published experience with the 670G in pregnancy is limited. A case report documented one woman who used the system throughout her pregnancy with glucose targets adjusted to the tighter pregnancy range of 63 to 140 mg/dL. Her time in that range increased from about 47% to 51% after starting Auto Mode, her HbA1c held near 6.5% until delivery, and her time spent low stayed below 4%, which is considered acceptable. She delivered a healthy baby at 37 weeks with no safety events.¹⁸

The 670G’s fixed glucose target of 120 mg/dL in Auto Mode is higher than what most endocrinologists aim for in pregnancy, which complicated its use. The pregnancy case relied on adjusted settings and close clinical oversight. This was a single case, not a recommendation, but it showed the system could be used safely with appropriate medical supervision.

Cost-Effectiveness

A cost-effectiveness analysis modeled the 670G against a standard insulin pump (without automated basal adjustment) over a patient’s lifetime. The hybrid closed-loop system was associated with a gain of about 1.9 quality-adjusted life years but came with higher upfront and ongoing costs. The analysis estimated an incremental cost-effectiveness ratio of roughly 164,000 Swedish kronor per quality-adjusted life year, which falls within what many health systems consider acceptable.¹⁹

The system looked far more cost-effective for people with poor baseline control. In that subgroup, the model projected a gain of 2.25 quality-adjusted life years at an incremental cost of just 15,830 kronor per quality-adjusted life year, primarily because more complications were averted. The takeaway is intuitive: automated insulin delivery offers the most economic value to those who stand to gain the most from better glucose management, namely people whose control on their current therapy is suboptimal.

Cybersecurity Concerns

The 670G drew attention for a less expected reason: cybersecurity. The pump communicates wirelessly with its sensor and meter, and vulnerabilities in that wireless communication raised concerns about unauthorized access. An FDA safety notice led to discussions about how connected medical devices should be secured, and the episode was discussed in the literature as a first-of-its-kind event for a diabetes device.²⁰ No reports of actual patient harm from cybersecurity exploits surfaced, but the incident underscored that insulin pumps increasingly share the security challenges of any wireless networked device.

From the 670G to the 780G

Medtronic addressed many of the 670G’s frustrations with the MiniMed 780G, a second-generation system described as an “advanced” hybrid closed-loop.²¹ The 780G can deliver automatic correction boluses for high glucose, something the 670G could not do. It offers an adjustable glucose target as low as 100 mg/dL rather than the 670G’s fixed 120 mg/dL target. When paired with the Guardian 4 sensor, it eliminates routine fingerstick calibrations entirely, directly tackling the single biggest source of user complaints with the prior generation. The 780G also added smartphone connectivity via Bluetooth, allowing users to view glucose data and system status on their phones instead of relying solely on the pump screen.

These changes matter because the 670G’s usability problems were not incidental. They shaped who could benefit from the technology. A system that works well but demands too much of its users will, in practice, work well for fewer people than its trial data would suggest. The 780G’s design reflects Medtronic’s response to real-world feedback, and early studies of the newer system have reported higher Auto Mode usage rates and improved glucose outcomes compared with the 670G.²² The competitive landscape also shifted. Other commercial systems launched around the same time, offering alternatives that further pushed the field toward less burdensome automation.²³ Do-it-yourself open-source loop systems also gained traction during this period, built by parents and patients who felt commercial development was moving too slowly.

The 670G will likely be remembered as a proof of concept as much as a product. It demonstrated that automated basal insulin delivery could work safely across age groups, from toddlers to older adults, and that even imperfect automation improved glucose control. It also demonstrated, unintentionally, exactly how much user burden a device can impose before the technology’s clinical benefits start to be undermined by people simply choosing not to use it.