Continuous Glucose Monitoring (CGM) Algorithms

Introduction to CGM Algorithms

A CGM sensor doesn't read glucose directly. It measures a tiny electrical current from an enzyme reaction in the interstitial fluid, the fluid around cells just under the skin, and an algorithm in the transmitter converts that current into a glucose value; the Dexcom G7 15 Day reports one every 5 minutes [7]. That software has to deal with physiological lag, the time glucose takes to move from blood into interstitial fluid. Mayo Clinic researchers measured it directly with glucose tracers: 5 to 6 minutes in fasting healthy adults [2], and a median of 6.8 minutes (range 4.8 to 9.8) in fasting adults with type 1 diabetes [3].

Noise is the other problem. A Kalman filter is a method that estimates a true signal from a stream of noisy measurements, and a 2016 review of CGM algorithms describes a Kalman-filter-based smoother used to clean up sensor data in real time [1]. In that design, separate software modules then correct for blood-to-interstitial differences and drift in sensor sensitivity, and predict glucose a short time ahead [1].

Technological Advancements

Older CGMs relied on fingerstick calibration: the user entered meter readings so the software could correct each sensor's drift. Factory calibration moves that step into manufacturing. Here's what surprised us: factory calibration isn't what first let a CGM replace fingersticks for treatment decisions, known as non-adjunctive use. The FDA approved the Dexcom G5 Mobile for that use in December 2016, and the G5 still needed a fingerstick calibration at least twice a day [4].

The original FreeStyle Libre, approved in September 2017, was factory calibrated and didn't require or accept user calibration [5]. The Dexcom G6 followed in March 2018 as the first integrated CGM (iCGM); once the user entered the sensor's calibration code, no fingerstick calibration was needed for the 10-day wear [6]. Both are earlier models now. Dexcom says it will no longer manufacture the G6 after July 1, 2026, and points G6 users to the G7 or G7 15 Day [9]. Abbott sold its Libre 2 and Libre 3 sensors through September 30, 2025, and replaced them with the Libre 2 Plus and Libre 3 Plus [10].

As of October 2026, the current Dexcom and Abbott prescription sensors are all factory calibrated. The Dexcom G7 and G7 15 Day are factory calibrated, with optional fingerstick calibration [7]. The FreeStyle Libre 2 Plus and 3 Plus sensors are factory calibrated and can't be calibrated by the user [8]. Our CGM hub goes deeper into the math in Sensor Calibration Algorithms and Data Fusion.

Improved Accuracy and Integration

MARD, or mean absolute relative difference, is the average percentage gap between sensor readings and a laboratory reference, so lower is better. Software alone has moved it a long way. Using signal processing only, University of Padova researchers cut the Dexcom SEVEN Plus's MARD from 15.1% to 10.3% in data from 23 people with type 1 diabetes, and Dexcom's redesigned "software 505" took the G4 Platinum from 13% to 9% with no hardware change, which the review calls the first single-digit CGM accuracy, in 2014 [1].

Accuracy carries more weight once a pump reads the sensor. The FDA created the iCGM category with the G6 in 2018, setting special controls for sensors that send readings to connected devices, including automated insulin delivery (AID) systems [6]. The G7 15 Day is cleared under that regulation and is indicated to communicate with AID systems [7]. Whether AID suits a person is a decision to make with their care team.

The FDA's review of the G7 15 Day also shows that accuracy shifts within one sensor's life. In 130 adults, MARD was 8.6% on days 1 to 3, 7.4% and 7.3% through the middle of wear, and 8.9% on the last days, 13 through 15.5 [7].