Continuous glucose monitors in non-diabetic users: signal versus noise

Key takeaways

• FDA cleared the first over-the-counter CGM (Dexcom Stelo) for non-insulin-using adults in March 2024 ; Abbott Lingo (general wellness) and Libre Rio (adults with type 2 diabetes not on insulin) both received 510(k) clearance on June 10, 2024 .
• CGM accuracy is benchmarked by mean absolute relative difference (MARD) versus a reference. Modern factory-calibrated sensors have MARD around 8 to 11 percent under stable conditions but drift higher during rapid glucose changes (post-meal, post-exercise) and at glucose extremes.
• PREDICT-1 (Berry 2020) measured postprandial glucose, triglyceride, and insulin response to identical standardized meals across 1,002 healthy adults and found large inter-individual variability with low correlation to traditional risk factors .
• Published RCTs of CGM-guided dietary intervention in non-diabetic adults are few. The Personalized Nutrition Project (Zeevi 2015) showed personalized CGM-informed diets reduced postprandial peaks more than standard advice in a 26-person crossover .
• Time-in-range targets that are routine in type 1 diabetes management (70 to 180 mg per dL, with 70 percent or more time in range) do not have an established equivalent for non-diabetic adults; most non-diabetic adults already spend more than 95 percent of time in this range.
• The Klonoff 2023 international consensus on CGM in non-diabetic users recommended caution about claims of metabolic-health benefit beyond glycemic education and called for more outcome-driven trials .

How CGMs actually measure glucose

Modern CGMs measure glucose in interstitial fluid, not in blood. A small filament containing glucose oxidase (or hexokinase, depending on platform) generates an electrical current proportional to glucose concentration; the sensor electronics convert this to a glucose reading and transmit it to a phone or receiver every 1 to 5 minutes.

Interstitial glucose tracks plasma glucose with a lag of roughly 5 to 15 minutes, more during rapid changes. Factory-calibrated sensors do not require fingerstick calibration but have lower precision than blood-glucose meters during rapid post-meal rises or post-exercise drops.

Sensor accuracy is reported as mean absolute relative difference (MARD) versus a laboratory reference. Modern Dexcom G7, Abbott FreeStyle Libre 3, and the FDA-cleared OTC variants have MARD values around 8 to 11 percent. This is good enough for trend interpretation in non-diabetic users; it is not good enough to act on small absolute differences (a reading of 119 versus 124 is within the noise envelope).

The OTC clearance landscape

Dexcom Stelo received FDA 510(k) clearance for over-the-counter use in non-insulin-using adults in March 2024, the first OTC CGM in the US. It is a 15-day-wear sensor priced for monthly self-purchase without a prescription .

Abbott Lingo and Libre Rio both received FDA 510(k) clearance on June 10, 2024. Lingo is positioned as a metabolic-coaching product for non-diabetic adults with a smartphone app emphasizing post-meal glucose education. Libre Rio is positioned for non-insulin-using adults with type 2 diabetes managed with diet or oral medications, blurring the boundary between consumer and medical devices .

All three OTC products are 510(k) cleared, meaning the FDA found them substantially equivalent to predicate devices. The clearances do not validate specific metabolic-health claims (weight, HbA1c, cardiovascular outcomes); they validate measurement performance and labeling.

What CGMs can reliably show in non-diabetic users

• Postprandial glucose peaks and time to return-to-baseline. These are real signals that vary meaningfully across meals, individuals, and contexts (sleep, exercise, stress).
• Day-to-day glucose variability, expressed as standard deviation or coefficient of variation. Variability is independently associated with cardiometabolic risk in observational data.
• The directional response to specific dietary or behavioral interventions: a CGM will reliably show that a high-carb breakfast produces a larger peak than a higher-fiber, higher-protein breakfast in the same individual.
• Acute exercise effects: post-exercise glucose dips, dawn-phenomenon morning rises, the impact of sleep restriction on next-day glucose tolerance.
• Ranges that fall consistently into impaired-fasting-glucose or impaired-glucose-tolerance territory, which is a flag worth a clinician conversation regardless of HbA1c.

What CGMs cannot reliably show in non-diabetic users

• Whether a 30 mg per dL postprandial peak is a problem. Healthy young adults regularly produce post-meal peaks of 130 to 160 mg per dL, particularly with fast-absorbed carbohydrates; these are within physiologic range.
• Whether their personal post-meal pattern differs meaningfully from population norms. The PREDICT-1 study showed identical meals produce 5 to 10-fold variability across healthy adults .
• Hard outcomes (weight loss, cardiovascular events). Published RCTs of CGM-guided dietary changes in non-diabetic adults are short-term, small, and use surrogate endpoints.
• HbA1c shifts over months. A 14-day CGM run captures variability and patterns but not chronic glycemic burden.
• Hypoglycemia risk in adults not taking insulin or sulfonylureas. False-low readings during rapid drops are common and rarely indicate clinical hypoglycemia in non-diabetics.

The personalized nutrition question

Zeevi et al. 2015 in Cell showed that postprandial glucose responses to identical meals vary substantially across individuals, that machine-learned models incorporating microbiome composition could predict an individual's response, and that personalized CGM-informed diets reduced post-meal glucose peaks more than standard nutritional advice in a 26-person 2-week crossover .

PREDICT-1 (Berry et al. 2020 Nature Medicine) extended this to 1,002 adults and found that genetic, microbiome, meal-context, and circadian factors all contribute to inter-individual variability, with traditional risk factors (BMI, age) explaining less variance than expected .

These two studies are the strongest evidentiary support for the personalized-CGM use case. Both used surrogate endpoints (postprandial peaks, area-under-curve) rather than weight, HbA1c, or cardiovascular outcomes. Whether the personalization improves long-run outcomes is the next-tier question and is not yet answered.

What the evidence does not yet resolve

• Whether CGM-guided dietary changes in non-diabetic adults reduce body weight, HbA1c, or cardiovascular events over 1 to 5 years. No adequately powered RCT exists.
• Whether glycemic variability in healthy adults is causally linked to long-term cardiometabolic risk versus simply correlated.
• Whether normative time-in-range targets for non-diabetic adults can be defined or whether they would be useful given how narrow most non-diabetics already are.
• How CGM use interacts with disordered eating in vulnerable users. Reports of CGM-driven food restriction in some users have raised concerns that need follow-up.
• How CGM data should be integrated with other consumer-health metrics (HRV, sleep, activity) for actionable signal versus information overload.

Editorial summary

CGMs are powerful sensors with a strong clinical case in diabetes. Their use case in non-diabetic adults is real but narrower than the marketing implies. The personalized-response evidence is strong but uses surrogate endpoints; hard-outcome trials are pending. For non-diabetic users, treat a CGM as a behavioral feedback tool with limited diagnostic value, not as a metabolic diagnostic. The 14-day rental cost is reasonable for a learning experience; the longer-term subscription case is harder to justify on the current evidence.