New research finds CGMs are reliable enough for dietary feedback

The study found that in a large sample of people without diabetes, CGM readings were consistent enough to be useful for personalized diet guidance


The Study:

Validity of continuous glucose monitoring for categorizing glycemic responses to diet: implications for use in personalized nutrition
Published: The American Journal of Clinical Nutrition, Feb 2022
Where: King’s College London

The Takeaway:

Continuous glucose monitor (CGM) accuracy has improved in the last decade, but there remain questions about their utility for personalized nutrition due to several sources of measurement variation. One example comes from a 2020 study published in the American Journal of Clinical Nutrition, which outfitted sixteen subjects without diabetes with both a DEXCOM (G4) and Abbot Freestyle Libre CGM simultaneously for 28 days to see how closely their readings correlated. The results were discouraging. Measurements of glycemic response to meals across CGMs varied considerably, suggesting that the poor reliability of CGMs made them unusable for personalized nutrition. Reliability is important. If one CGM tells you a meal leads to high glucose levels and another CGM tells you it doesn’t, it’s hard to make evidence-based decisions on what to eat for stable glucose.

The current study, also published in The American Journal of Clinical Nutrition and sponsored by metabolic health company Zoe, used a similar design: People without diabetes wore two CGMs simultaneously and compared the data to see how reliable the devices were. Encouragingly, wearing two of the same CGM (i.e., two Freestyle Libres) gave very similar glucose data, and wearing two different units (one Freestyle Libre and one DEXCOM G6) showed a better agreement of glucose data than in the previous study (though still showed differences between CGMs from the two manufacturers). It also found reliable measurements of related clinical measures, such as glycemic variability and time in range (TIR). The study provides evidence that CGMs are reliable enough to differentiate the glycemic response to different meals and, therefore, can be used as a tool to help people personalize their nutrition.

What It Looked At:

The findings from this study are part of a more extensive study called the ZOE PREDICT1 study, which aims to characterize the variability in post-meal glucose and triglyceride responses of 1,002 twins in the UK based on lifestyle factors, diet, genetics, microbiome, and existing biomarkers like triglycerides and cholesterol.

This study is a secondary analysis of 394 subjects to determine the reliability of CGM devices over 14 days in real-world settings. Of these 394 subjects, 360 wore two Freestyle Libre devices simultaneously, and 34 wore one Freestyle Libre and one DEXCOM G6. Subjects consumed standardized breakfast and lunch for two days and standardized breakfast for an additional eight mornings (a total of 4,457 meals). They ate their usual diet (or ad libimtum) for all remaining meals (5,738 additional meals).

The primary outcome—meaning the main thing the study was looking for—was the level of agreement of glucose for the two hours following each meal between two of the same CGMs, and agreement between the two different CGMs worn simultaneously by the same person. The main aim was not to see how a meal on a Monday compared to that same meal Tuesday—it was much more straightforward. How does device 1 compare to device 2 during that meal on Monday (commonly known as the reliability)?

To capture this, researchers calculated the coefficient of variation (CV), which measures how close the two glucose values are. The higher the CV, the further apart the two devices are, and the less reliable the measurement. Lower CV values mean more reliability and are better.

The researchers also tested as secondary outcomes two clinically relevant measures often used in people with diabetes: time in range (as defined by the American Diabetes Association) and glycemic variability, or how much a person’s glucose varies throughout the day.

In addition to what this study looked at, it is essential to note what this study did not look at. It did not look at the validity (or accuracy) of the CGMs. In other words, it did not test CGMs against a gold standard for glucose measurement like a blood draw to see if the CGM agreed—only to see if the CGMs were consistent across individual devices.

What It Found:

For standardized meals, there was strong agreement in the glucose levels over the two-hour post-meal period between the two same devices (CV 3.7%, 25th-75th percentile 1.7% – 7.1%). Across different devices, the results were slightly poorer but still better than the earlier study (CV 12.5%, 25th – 75th percentile 5.1% – 24.8%).

For meals that subjects just ate during everyday life (called ad libitum meals in the paper), the agreement was not quite as good for both same devices (CV 4.1 %, 25th-75h percentile 1.8% – 7.1%) and different devices (CV 16.6%, 25th – 75th percentile 5.5% – 30.7%).

In short, you’re more likely to get consistent readings across two of the same device (i.e., two Abbot Freestyle Libres) than between an Abbot and a Dexcom.

Interestingly, the authors also compared the same meal consumed on two days because all standardized meals were consumed twice. Here they found variation between meals using the same device was much higher than two different devices measuring the same meal. What does that tell us? That the technical variation (CV of 12.5%) between two devices is much lower than biological variation (CV 30.9%)—how your body processes a meal on a given day—and suggests CGMs are capable of differentiating different glycemic responses to meals.

For the secondary measures of TIR and glycemic variability, the authors showed similar CV differences and correlations as was seen with the two-hour glucose levels; again, two of the same device showed more minor variation than comparisons of two different devices.

Why It Matters:

An increasingly common use of CGM for people with diabetes and the general population is measuring individual glycemic responses to meals. However, if people act on inaccurate information, they may simply not see health improvements; at worst, it could be harmful. For instance, if your CGM does not accurately rank the glycemic response of pizza and cookies above a veggie-filled salad, people are more likely to make less healthy nutritional choices. Ensuring that CGMs provide accurate data that helps consumers make personalized nutrition choices is essential. The current study provides strong evidence that they do.

Remaining Questions:

While this study indicates CGMs are reliable enough for personalized nutrition, plenty of additional questions need answering in future research.

For instance, there are other CGM manufacturers besides Abbot and DEXCOM, including some using noninvasive wrist-based technology. We must see rigorous testing of all CGMs to ensure reliability and validity.

The difference in conclusions between this study and the previous one raises several vital questions. For example, how important are excess caloric intake (which occurred in the previous study), carbohydrate content in each meal, and sensor attachment sites in altering device reliability? The current study’s authors also acknowledge that more rigorous randomized matching of subjects, application of error correction, testing at a one-hour time point, and replication in an independent data set would increase the strength of the current findings.

In addition to studies that test the reliability and validity of CGMs, we need to better understand the phenomenon of significant variation in glucose values, sometimes to the same meals (see above) in the same person. The research focused on teasing apart how much of that difference is due to the device (technically variation) and how much is due to the person (biological variation) will help us better understand the limitations of CGMs in personalized nutrition.


The variability associated with CGMs is both biological and technical. While the agreement between different devices is not as robust as comparing the same device, the data is sufficient to rank most meals by the glycemic response, which is excellent for understanding the general effects of specific foods. Continued improvements in the reliability and validity of CGM will further improve the usefulness and precision of truly personalized nutrition advice.