Diabetes Technology: 'Fear Not'

Today I felt like I was pulled 10 years into the future and given the opportunity to practice medicine then. This happened when I started treating my first patient with the Medtronic MiniMed® 670G hybrid closed-loop insulin delivery system.

Although I have had several patients using open artificial pancreas systems for years and have had experience treating patients using closed- loop insulin delivery, today felt different, because the 670G device is approved by the US Food and Drug Administration and soon will enter the hands of thousands of people with type 1 diabetes. Closed-loop insulin delivery will be used at a scale we have not yet seen or experienced.

A Brief Refresher on Continuous Glucose Monitoring Devices

For years, we have used continuous glucose monitoring (CGM) devices to measure blood glucose values every 5 minutes, and then used insulin pumps to continuously deliver insulin doses manually in an "open loop"—meaning that the human is telling the pump how much insulin to give. "Closed- loop" systems put a computer algorithm in between the pump and CGM, automating insulin delivery rates.

For this initial iteration, users still have to tell the 670G when they are exercising or eating (as well as estimating the carbohydrate amount), and they must calibrate the sensor with a minimum of two, but ideally four, blood glucose finger-sticks per day (hence the term "hybrid closed loop," rather than fully closed loop, because it is not yet 100% automated.)

This represents incredible progress, and I am hopeful that the 670G will perform as well in real-world use as it did in its pivotal trial.^[1] Several times over the past few weeks as I've talked to people about closed-loop insulin delivery, I've found myself choking up, thinking about the impact it will have.

For many people with type 1 diabetes, life will never be the same. I am hopeful that people with diabetes will be able to sleep at night without fear of "going low," and that they will be able to fully participate in a meeting at their job without peeking under the desk at their CGM tracing. This impact is evident in the fact that 80% of pivotal trial participants asked to stay on the 670G when the trial was over, and 96 of the original 129 participants were still using the 670G a year later.

From Physiology Expert to Machine Operator?

As a physician, a different moment also arrived for me. Many of us have imagined or joked about closed- loop systems "putting us out of the diabetes business." Today, I absorbed the reality of what it will mean to practice as a diabetologist in the future.

In an instant, I was transformed from physiology expert and disease coach to machine operator. Although Medtronic's report does offer some transparency by telling you the reasons the system goes out of "auto mode" (which it does at various times), it is, at the end of the day, still very much a "black box."

It comprises many components working together in harmony, meaning that when anything goes wrong, there are infinite possible reasons why. Is the sensor bad? Was the blood glucose reading used for calibration off? Was the algorithm wrong in its dosing choice? Is the insulin infusion site no good? Did the patient eat more than she thought she did?

We currently ask many of these same questions today in troubleshooting an insulin pump and CGM in "open loop" management, but it feels different now.

I felt a professional impotence, awed at the power of the machine, wondering how I could compete.

One of the privileges of being a diabetologist is that diabetes technology often lets us reach into the future. Although "patient-generated health data" is a recent buzzword in healthcare, we have used such data for decades in diabetes care.

Today, the curtain to the future was peeled back, as I glimpsed an era where algorithms and artificial intelligence will manage our patients, with the doctor serving as the operator—the guide—in this world. This will increasingly be the case not just in diabetes but throughout medicine, with algorithms and decision-support tools continuing to infiltrate medical care. They will start off as "suggestions" and then move to automation, with the physician as troubleshooter.

Some fear that this will reduce our role. And I understand this fear. For a brief moment, I felt a professional impotence, awed at the power of the machine, wondering how I could compete.

But ultimately, I believe our role will change rather than be diminished. Soon, in addition to making sure we teach our patients about the physiology of their illness, we will have to help them understand what the algorithm is recommending and why, and what the strengths and limitations are. Just as physicians a century ago helped explain the mysteries of the body and science to people, we will have to help explain the mysteries of why an algorithm recommended a particular diagnosis or treatment.

Importance of Informatics

This has serious implications. Although the field of informatics is growing and American Medical Informatics Association, our professional society, intends to train 10,000 informaticists in the next 10 years, this will not be enough. Medical schools must ensure that all of their graduating physicians have a basic grasp and competencies in informatics.

Professional informaticists will remain the experts and be the ones designing and training the systems, but it will no longer be acceptable for a physician not to understand the basics of an algorithm. With closed- loop insulin delivery, for example, we have to ensure that we and our patients still understand how to manage diabetes manually—the "old- fashioned" way—as well as learn how to optimize and troubleshoot the auto mode.

I find myself reaching consistently for the analogy of the airline pilot. Although much of flying today occurs on autopilot, pilots must still master and maintain the ability to fly the plane manually. They must also avoid automation complacency—the human tendency to "zone out" rather than maintain vigilance when lulled into thinking that a machine has everything under control.^[2]

So too must all physicians learn the basics of how these decision-support tools work, and learn how to optimize, interpret, and troubleshoot them when things are in "auto mode," while maintaining our grasp of the fundamentals of medicine and being able to practice in "manual mode."

Realizing this, I recovered from my brief moment of awe and fear. I quickly settled back into a comfortable and familiar conversation as a physician, discussing with my patient his expectations for this new treatment, how we would monitor progress together, adjust settings, and optimize its results for him.

The machines are here. Fear not.

Comments

Commenting is limited to medical professionals. To comment please Log-in.

Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.