DRIFT

Why numbers can wander

Biofouling Surface build-up slowly changes how much glucose reaches the enzyme

Enzyme loss The enzyme gradually loses activity

Local oxygen Tissue oxygen can vary from place to place

Pressure Lying on a sensor can squeeze tissue and cause a false low, called a compression low

Sensitivity is not constant, so the current-to-glucose conversion must be handled carefully.

CALIBRATION

Turning current into a number

🩸

Fingerstick calibration

The user enters a blood glucose value, and the system adjusts its conversion

🏭

Factory calibration

The maker calibrates each batch in advance, so users do not need to

Factory calibration is convenient, but it requires very uniform manufacturing.

ACCURACY SCORE

MARD

ARD = |CGM − reference| / reference × 100%

MARD is the mean absolute relative difference: the average of those percent errors across many paired readings. Lower is better. Current leading systems report roughly 8 to 10 percent.

MARD is a single, comparable number, but it is an average.

WORKED EXAMPLE

Computing a MARD

Pair 1 Reference 100, CGM 110: error 10, which is 10%

Pair 2 Reference 200, CGM 190: error 10, which is 5%

MARD The average of 10% and 5% is 7.5%

The same absolute error counts for more at low glucose, where the reference is small.

BEYOND THE AVERAGE

Averages can hide big misses

An error grid, such as the Clarke or consensus error grid, sorts each reading by how a wrong decision would affect the patient. Zone A readings are clinically accurate, while the highest zones could lead to dangerous treatment errors.

Look at the error grid as well as the MARD.

UNIT 9 STUDY COMPLETE

🎯

Ready for the Fab Challenge?

You've covered drift, calibration, and the standard way sensor accuracy is reported.