When Dan Heller received his first batch of Dexcom’s latest continuous glucose monitors in early 2023, he decided to run a small experiment: He wore the new biosensor and the previous generation at the same time to see how they compared in measuring his glucose levels.
The new, seventh-generation model (aptly called the G7) made by San Diego-based healthcare company Dexcom had just begun shipping in the United States. Dexcom claimed the G7 to be the “most accurate sensor” available to the thousands of people with Type 1 diabetes who use continuous glucose monitors to help manage their blood sugars. But Heller found that its real-world performance wasn’t up to par. In a September 2023 post on his Substack, which is dedicated to covering Type 1 diabetes research and management, he wrote about the experience and predicted an increase in adverse events with the G7, drawing on his past experience leading tech and biotech companies.
In the two years since Heller’s experiment, many other users have reported issues with the device. Some complaints regard failed connection and deployment issues, which Dexcom claims to have now addressed. More concerning are reports of erratic, inaccurate readings. A public Facebook group dedicated to sharing negative experiences with the G7 has grown to thousands of users, and several class action lawsuits have been filed against the company, alleging false advertising and misleading claims about device accuracy.
Yet, based on a standard metric in the industry, the G7 is one of the most accurate glucose sensors available. “Accuracy in the performance of our device is our number one priority. We understand this is a lifesaving device for people with Type 1 diabetes,” Peter Simpson, Dexcom’s senior vice president of innovation and sensor technology, told IEEE Spectrum. Simpson acknowledged some variability in individual sensors, but stood by the accuracy of the devices.
So why have users faced issues? In part, metrics used in marketing can be misleading compared to real world performance. Differences in study design, combined with complex biological realities, mean that the accuracy of these biosensors can’t be boiled down to one number—and users are learning this the hard way.
Dexcom’s Glucose Monitors
Continuous glucose monitors (CGMs) typically consist of a small filament inserted under the skin, a transmitter, and a receiver. The filament is coated with an enzyme that generates an electrical signal when it reacts with glucose in the fluid surrounding the body’s cells. That signal is then converted to a digital signal and processed to generate glucose readings every few minutes. Each sensor lasts a week or two before needing to be replaced.
The technology has come a long way in recent years. In the 2010s, these devices required blood glucose calibrations twice a day and still weren’t reliable enough to dose insulin based on the readings. Now, some insulin pumps use the near-real-time data to automatically make adjustments. With those improvements has come greater trust in the data users receive—and higher standards. A faulty reading could result in a dangerous dose of insulin.
The G7 introduced several changes to Dexcom’s earlier designs, including a much smaller footprint, and updated the algorithm used to translate sensor signals into glucose readings for better accuracy, Simpson says. “From a performance perspective, we did demonstrate in a clinical trial that the G7 is significantly more accurate than the G6,” he says.
So Heller and others were surprised when the new Dexcom sensor seemed to be performing worse. For some batches of sensors, it’s possible that the issue was in part due to an unvalidated change in a component used in a resistive layer of the sensors. The new component showed worse performance, according to a warning letter issued by the U.S. Food and Drug Administration in March 2025, following an audit of two U.S. manufacturing sites. The material has since been removed from all G7 sensors, Simpson says, and the company is continuing to work with the FDA to address concerns. (“The warning letter does not restrict Dexcom’s ability to produce, market, manufacture or distribute products, require recall of any products, nor restrict our ability to seek clearance of new products,” Dexcom added in a statement.)
“There is a distribution of accuracies that have to do with people’s physiology and also the devices themselves. Even in our clinical studies, we saw some that were really precise and some that had a little bit of inaccuracy to them,” says Simpson. “But in general, our sensor is very accurate.”
In late November Abbott—one of Dexcom’s main competitors—recalled some of its CGMs due to inaccurate low glucose readings. The recall affects approximately 3 million sensors and was caused by an issue with one of Abbott’s production lines.
The discrepancy between reported accuracy and user experience, however, goes beyond any one company’s manufacturing missteps.
Does MARD Matter?
The accuracy of CGM systems is frequently measured via “mean absolute relative difference,” or MARD, a percentage that compares the sensor readings to laboratory blood glucose measurements. The lower the MARD, the more accurate the sensor.
This number is often used in advertising and marketing, and it has a historical relevance, says Manuel Eichenlaub, a biomedical engineer at the Institute for Diabetes Technology Ulm in Germany, where he and his colleagues conduct independent CGM performance studies. For years, there was a general belief that a MARD under 10 percent meant a system would be accurate enough to be used for insulin dosing. In 2018, the FDA established a specific set of accuracy requirements beyond MARD for insulin-guiding glucose monitors, including Dexcom’s. But manufacturers design the clinical trials that determine accuracy metrics, and the way studies are designed can make a big difference.
For instance, blood glucose levels serve as the “ground truth to compare the CGM values against,” says Eichenlaub. But glucose levels vary across blood compartments in the body; blood collected from capillaries with a finger prick fluctuates more and can have glucose levels around 5 to 10 percent higher than venous blood. (Dexcom tests against a gold-standard venous blood analyzer. When users see inaccuracies against home meters that use capillary blood, it could in part be a reflection of the meter’s own inaccuracy, Simpson says, though he acknowledges real inaccuracies in CGMs as well.)
Additionally, the distribution of sampling isn’t standardized. CGMs are known to be less accurate at the beginning and end of use, or when glucose levels are out of range or changing quickly. That means measured accuracy could be skewed by taking fewer samples right after a meal or late in the CGM’s lifetime.
According to Simpson, Dexcom’s trial protocol meets the FDA’s expectation and tests the devices in different blood sugar ranges across the life of the sensor. “Within these clinical trials, we do stress the sensors to try and simulate those real world conditions,” he says.
Dexcom and other companies advertise a MARD around 8 percent. But some independent studies are more demanding and find higher numbers; a head-to-head study of three popular CGMs that Eichenlaub led found MARD values closer to 10 percent or higher.
Eichenlaub and other CGM experts believe that more standardization of testing and an extension of the FDA requirements are necessary, so they recently proposed comprehensive guidelines on CGM performance testing. In the United States and Europe, a few manufacturers currently dominate the market. But newer players are entering the growing market and, especially in Europe, may not meet the same standards as legacy manufacturers, he says. “Having a standardized way of evaluating the performance of those systems is very important.”
For users like Heller though, better accuracy only matters if it yields better diabetes management. “I don’t care about MARD. I want data that is reliably actionable,” Heller says. He encourages engineers working on these devices to think like the patient. “At some point, there’s quantitative data, but you need qualitative data.”
Reference: https://ift.tt/dAXkyVZ
No comments:
Post a Comment