Continuous Glucose Monitors: Independent Evidence, Accuracy & Recalls

Table of contents

• Evidence summary
• What a CGM is, and how it is regulated
• How it works, and what "accuracy" means
• What works and what does not
• Benefits by claim
• Risks and all side effects
• All interactions and interferences
• Who should not rely on a CGM alone
• How it is used, and product-quality problems
• Follow the money: who makes it and who funded the evidence
• Related research
• Frequently asked questions
• Sources and funding notes

Evidence summary

Independent evidence and credibility scorecard

Independence tiers: 1 = regulator or government body; 2 = academic or Cochrane team with no declared commercial funding for the work; 3 = independent authors analysing mostly manufacturer-run trials, or minor declared industry ties; 4 = manufacturer-funded or manufacturer-authored. Credibility runs from A (highest) to D.

Seven of the twelve sources are wholly or partly US-based, and the US is also home to both leading makers. The UK's NICE and a Dutch meta-analysis are the main non-US checks. The best sign of reliability is agreement between sources with opposite interests. On type 2 diabetes they agree: Dexcom's own MOBILE trial, Abbott-supported IMMEDIATE and two independent meta-analyses (one with no specific funding) all point to an HbA1c benefit of a few tenths of a percentage point. On people without diabetes there is no such agreement, because no independent trial shows a health benefit.

What a CGM is, and how it is regulated

A CGM has three parts: a thin filament inserted just under the skin, a transmitter that sits on top, and a phone app or reader that shows the results. Two families dominate:

• Real-time CGM (rtCGM) sends readings automatically and can sound alarms for highs and lows. Dexcom's G6 and G7 are examples. Dexcom launched the G7 in 2023 and a 15-day version in late 2025 (Dexcom 10-K).
• Intermittently scanned CGM (isCGM, "flash") originally needed the user to scan the sensor with a phone or reader. NICE still uses this term for Abbott's FreeStyle Libre family (NICE NG17). Newer Libre models stream readings automatically, so the line between the two is blurring.
• Over-the-counter "biosensors" are sold without a prescription. The FDA cleared Dexcom's Stelo on 5 March 2024 as the first over-the-counter CGM, for adults aged 18 and over who do not use insulin. It shows readings in an app, lasts up to 15 days and has no low-glucose alarm (FDA). Abbott launched Lingo in the US in September 2024 for consumers aged 18 and over not on insulin, based on its Libre technology (Abbott).
• Subscription programmes such as Levels sell an app, coaching and lab tests, and let members "use Levels with your own CGM or get one through us" (Levels). ZOE used CGMs in its research trials (METHOD). We did not find a CGM listed among the products on its current homepage.

Regulation. In the US, CGMs are regulated medical devices. Prescription models carry insulin-dosing claims. Over-the-counter models like Stelo are cleared for people not using insulin, and the FDA says they are not designed to alert users to dangerous lows (FDA). In England, NICE sets who should be offered CGM on the NHS. The 2022 updates cover adults with type 1 diabetes (NG17) and insulin-treated type 2 diabetes (NG28). NICE refers to separate guidance for pregnancy. In the UK, buying a sensor yourself does not change the NHS eligibility rules.

How it works, and what "accuracy" means

A CGM sensor uses an enzyme to react with glucose in the interstitial fluid around cells, not in blood. Interstitial glucose lags behind blood glucose, so readings are less reliable when sugar is changing fast. Abbott's own safety information says Libre 3 and Libre 2 readings cannot be used for treatment decisions during the first 12 hours of wear when the sensor flags that it may be inaccurate, and for the older Libre 14-day system it advises a finger-prick check when glucose is changing by more than 2 mg/dL per minute (Abbott safety information). NICE asks clinicians to give CGM users enough test strips for this reason (NG28 1.7.5).

MARD (mean absolute relative difference) is the headline accuracy figure. It is the average percentage gap between sensor readings and a laboratory reference, so lower is better. Two cautions matter:

• Maker-run studies tend to report lower figures. Dexcom's pivotal G7 study (funded by a Dexcom grant, several authors Dexcom employees) reported a MARD of 8.2% for arm-worn sensors and 9.1% for the abdomen in 316 adults (Garg et al. 2022).
• Independent head-to-head tests find higher errors. Germany's Institute for Diabetes Technology tested Libre 3, Dexcom G7 and Medtronic Simplera in 24 adults with type 1 diabetes. None of the three makers paid for the study. Against laboratory (YSI) values, MARD was 11.6%, 12.0% and 11.6%. Libre 3 and G7 did better in normal and high ranges, and Simplera did better in the low range (Eichenlaub et al. 2025). The authors note that all three systems were less accurate than in earlier studies, and they call for standard testing rules.

What MARD means in practice: at a true glucose of 4.0 mmol/L, a 12% error is about 0.5 mmol/L either way. That is enough to move a reading across a hypo threshold. This matters a great deal for someone dosing insulin. For someone without diabetes deciding between porridge and toast, the same error can turn a normal response into an apparent "spike".

What works and what does not

Benefits by claim

Does a CGM help in type 1 diabetes?

Yes, and this is the best-supported use. NICE's 2022 update says adults with type 1 diabetes should be offered a choice of real-time or flash CGM. The choice should rest on their needs, and the lowest-cost device should be offered when several fit (NICE NG17, 1.6.10–1.6.11). The trials behind this were largely industry-sponsored. Dexcom's DIAMOND trial (158 adults on injections) found HbA1c fell 0.6 percentage points more than with finger-pricks (95% CI 0.3 to 0.8). Time below 70 mg/dL (3.9 mmol/L) was a median 43 minutes a day versus 80, though severe hypos were equal at two people per group (Beck et al. 2017; sponsor: Dexcom). Abbott's IMPACT trial randomised 241 well-controlled adults. Time in hypoglycaemia fell from 3.38 to 2.03 hours a day with flash, versus 3.44 to 3.27 hours with finger-pricks: a 38% relative reduction (Bolinder et al. 2016; funded by Abbott Diabetes Care). Independent confirmation comes mainly from NICE's own evidence review, which reached the same conclusion with the cost-effectiveness analysis behind it.

Does a CGM help in type 2 diabetes on insulin?

Yes, for defined groups. NICE says flash CGM should be offered to adults with type 2 diabetes on multiple daily insulin injections in any of these situations (NG28, 1.7.1–1.7.3):

• They have recurrent or severe hypos.
• They have impaired hypo awareness.
• A condition or disability stops them doing finger-pricks.
• They would otherwise need to test at least 8 times a day.
• They would otherwise need a carer or healthcare worker to monitor their glucose.

Real-time CGM may be used instead if it costs the same or less. In Dexcom's MOBILE trial, 175 adults on basal insulin alone were randomised. HbA1c fell from 9.1% to 8.0% with CGM and from 9.0% to 8.4% with a meter (adjusted difference −0.4 points; 95% CI −0.8 to −0.1). Time in range was 59% versus 43% (Martens et al. 2021). Dexcom was the lead sponsor (registry); all authors' institutions received Dexcom grant funding for the study, and one author was a Dexcom employee and stockholder (disclosures). In September 2026 Abbott announced conference data from US health records. It said CGM use was linked to a 45% lower risk of death at 12 months among about 28,000 adults on basal insulin (Abbott press release). This is an observational study promoted by the seller and not yet peer reviewed, as far as we could find. People who choose CGM may differ in ways no matching can fully remove, so it does not show that CGM prevents deaths.

Does a CGM help in type 2 diabetes not treated with insulin?

A little, on average. The Dutch meta-analysis by Jancev and colleagues had no specific funding, and its authors declared no conflicts. It pooled 12 trials with 1,248 adults. CGM lowered HbA1c by 3.43 mmol/mol (0.31 percentage points; 95% CI 2.11 to 4.75 mmol/mol) versus finger-prick testing. The effect was similar in people on oral drugs only: −3.22 mmol/mol (−0.29 points; 95% CI −1.05 to −5.39). Time in range rose by 6.4 percentage points, about an hour and a half a day (Jancev et al. 2024). A US review from ECRI pooled 14 trials (1,647 people) and found about −0.32 points, with moderate certainty, from short trials that rarely measured hospital visits or adverse events (Uhl et al. 2024).

Every trial was open label, so people knew they were being watched. No trial showed fewer severe hypos (RR 0.66; 95% CI 0.15 to 3.00) or fewer complications (Jancev). The Abbott-supported IMMEDIATE trial (116 adults on non-insulin drugs, 16 weeks) found time in range up 9.9 points and HbA1c down 0.3 points, with a confidence interval touching zero (Aronson et al. 2023; Abbott collaborator). NICE's committee found isCGM cost-effective for insulin users but "no evidence" for people not using insulin, so it limited its recommendations to insulin users. It also called for research using real-world NHS data (NG28 rationale; update history). The 2026 JAMA Internal Medicine review agreed that some individuals benefit considerably. It concluded that CGM should answer "a specific patient problem rather than" being the default (Dower et al. 2026).

Can a CGM improve health, energy or weight if you do not have diabetes?

There is no trial evidence that it does. The 2026 JAMA Internal Medicine review covered systematic reviews and meta-analyses of randomised trials, and clinical guidelines. It found only "limited and indirect evidence" for CGM in people with prediabetes or obesity, or with type 2 diabetes not on glucose-lowering treatment (Dower et al. 2026). The University of Gothenburg summarised its finding as no scientific evidence that CGM improves health or prevents disease in people without diabetes. The lead author said the technology is "tremendously valuable for certain groups of patients" (Gothenburg, August 2026).

The marketing reads differently. Abbott's Lingo launch says "compelling published research" shows that limiting glucose spikes benefits people without diabetes, citing weight management and future heart and metabolic risk (Abbott). Those references are about glucose patterns and diets, not about trials of wearing a CGM. Levels advertises that 81% "improved HbA1c". Its footnote says this is "based on retrospective analysis of Levels members" and "not a clinical trial", and no control group is described (Levels). A claim like that cannot separate the device from the motivation of people who pay for a subscription.

For calibration, normal values matter. In 153 healthy, non-obese volunteers aged 7 to 80 wearing a blinded Dexcom G6, average glucose was 5.4–5.5 mmol/L (5.8 mmol/L in those over 60). Readings were between 3.9 and 7.8 mmol/L for a median 96% of the day, and above 7.8 mmol/L for about 30 minutes a day. They were below 3.9 mmol/L for about 15 minutes a day (Shah et al. 2019; funded by the Helmsley Charitable Trust, lead author with Dexcom ties). In other words, short "spikes" and brief "lows" are normal. An app that flags them as problems can turn normal physiology into a worry.

Can a CGM tell you which foods are "good for you"?

Not reliably. US government scientists at the NIH had 16 adults without diabetes wear a Dexcom and an Abbott sensor at the same time over 28 inpatient days, covering 760 meals. The two devices ranked the same person's meals quite differently (Kendall rank correlation 0.43, where 1 would be perfect agreement). Eating the meals one device rated "best" gave about half the expected benefit when judged by the other device (Howard, Guo & Hall 2020). A larger analysis from the ZOE PREDICT programme (394 people) found better agreement: 0.9 between two sensors of the same brand and 0.7 across brands. Only 34 of those people wore two different brands. The study was supported by ZOE Ltd, with cohort funding from public bodies, and several authors were ZOE employees or consultants (Merino et al. 2022).

ZOE's METHOD trial (347 US adults, 18 weeks) compared its app-based personalised programme, which used CGM data among other inputs, with standard US dietary advice. Triglycerides fell slightly more (−0.13 mmol/L), but LDL cholesterol, the other primary outcome, did not differ (Bermingham et al. 2024). ZOE funded the trial and helped design, analyse and write it, and authors include its co-founders, staff and option-holders (same source). The trial tested a whole programme, so it cannot show what the glucose sensor added.

Risks and all side effects

• Falsely high readings. In July 2024 Abbott corrected sensors from three Libre 3 lots distributed in the US that could read too high. That risked insulin being taken when it was not needed (Abbott).
• Accuracy drift from manufacturing changes. The FDA's March 2025 warning letter found that Dexcom had changed a component in the sensor's resistance layer for G6 and G7 without clearance. Dexcom's own clinical study of the change "did not meet" its primary endpoints, and the spread of glucose sensitivity was twice that of the original material (FDA).
• Skin problems. In Abbott's IMPACT trial, 13 sensor-related adverse events were reported by 10 people, including allergy, itching, rash, redness and insertion-site symptoms (Bolinder 2016). The FDA lists local infection, skin irritation and pain as adverse events for Stelo (FDA). NICE lists skin sensitivity and body-image concerns among factors to discuss (NG17 box 1).
• Missed lows on wellness devices. Stelo is not designed to alert users to hypoglycaemia (FDA).
• Psychological harms and over-medicalisation. Reviewers warn that healthy people may misread normal glucose swings, leading to needless food restriction, anxiety and extra use of health services (Gothenburg; Dower et al. 2026). We found no trial that measured disordered eating in wellness users.
• Treatment burden for patients. NICE notes that the choice of device can depend on alarm fatigue, smartphone access, dexterity and effects on work (NG17 box 1).

All interactions and interferences

CGMs do not interact with medicines in the body, but some substances distort the reading:

• Vitamin C (Abbott Libre): for Libre 3 Plus and Libre 2 Plus, more than 1,000 mg a day "may falsely raise" sensor readings, which could hide a low. For standard Libre 3 and Libre 2 sensors the threshold is lower, more than 500 mg a day, and Abbott says 1,000 mg cold remedies such as Airborne and Emergen-C should not be taken while wearing them (Abbott).
• Salicylic acid (aspirin, some skin products), Libre 14-day: may slightly lower readings (Abbott).
• Hydroxyurea (Dexcom G6/G7): makes readings higher than actual glucose. Dexcom says not to use G7 for treatment decisions while taking it (Dexcom).
• Paracetamol (acetaminophen), Dexcom G6/G7: standard doses up to 1 g every 6 hours are acceptable. Above the maximum dose, readings may look higher than they are (Dexcom).
• Scans and heat treatments: Abbott says Libre 3 and 3 Plus can stay on for 1.5T or 3T MRI under set conditions, though readings may be inaccurate for up to an hour. All parts must be removed before diathermy (Abbott). Check each device's own instructions, because rules differ by model.

Who should not rely on a CGM alone

• Anyone using insulin, sulfonylureas or another hypo-causing drug should keep a finger-prick meter and test strips as backup. NICE explicitly advises this (NG28 1.7.5; NG17 1.6.13).
• People with problematic hypoglycaemia should not use over-the-counter sensors without alarms, such as Stelo (FDA).
• Under-18s are outside the cleared age range for Stelo and Lingo (FDA; Abbott). Children with diabetes are covered by separate paediatric guidance that we did not review here.
• Pregnancy: NICE points to its separate diabetes-in-pregnancy guideline (NG28). Wellness sensors are not a substitute for antenatal glucose testing.
• People taking hydroxyurea, or high-dose vitamin C, need a device that is not affected, or finger-prick confirmation (see interferences above).
• People with a history of eating disorders or health anxiety may find constant glucose feedback unhelpful. This is a reasoned concern raised by reviewers, not a tested finding (Gothenburg).

How it is used, and product-quality problems

There is no "dose", but there are usage rules. Sensors are worn for a fixed period, up to 15 days for Stelo (FDA), and then replaced. NICE says CGM should be provided by a team with expertise, included in structured education, and reviewed as part of the care plan. If someone struggles with a device, clinicians should look at education and emotional support before withdrawing it (NG28 1.7.4–1.7.9). NICE also asks the NHS to monitor who is getting CGM and to close gaps in uptake between groups (NG28 1.7.10).

If you use a Libre 3 or Libre 3 Plus sensor, the FDA advises checking the serial number on Abbott's lookup site, which its notice names. Stop using any affected sensor and request a replacement (FDA). Quality problems at both leading makers within a single year show that a sensor reading is a measurement with an error margin, not a fact.

Follow the money: who makes it and who funded the evidence

Who makes and sells it

• Abbott Laboratories (Abbott Park, Illinois, US): makes the FreeStyle Libre family and the Lingo consumer sensor. In 2025 its Diabetes Care division sold US$7,998 million, up 17.5%, and fourth-quarter CGM sales alone were US$2.0 billion (Abbott results). Lingo is based on Libre technology, which Abbott said in its 2024 Lingo launch release was used by more than 6 million people with diabetes worldwide (Abbott). In 2026 Abbott announced a consumer glucose partnership with Google (Abbott) and FDA authorisation of a combined glucose-ketone sensor (Abbott).
• DexCom, Inc. (San Diego, California, US): makes the G6, G7 and Stelo, manufactured mainly in Mesa, Arizona and Penang, Malaysia, with a new plant being built in Athenry, Ireland. Revenue was US$3.62 billion in 2023, US$4.03 billion in 2024 and US$4.66 billion in 2025 (Dexcom annual report, SEC). Dexcom describes Stelo as a biosensor for adults with prediabetes and type 2 diabetes not using insulin, launched in August 2024 (same source).
• Wellness platforms: Levels sells subscriptions from US$15 a month up to US$1,329 a year (Levels). ZOE Ltd (London) funded its METHOD trial; its co-founders include authors of that trial (METHOD disclosures). We could not verify the ownership or investors of either company from primary sources.

The commercial logic is simple. A person on insulin replaces a sensor every 10 to 15 days, for life. The non-insulin type 2 and "wellness" markets contain many more people, so both makers have a strong incentive to show benefit there. Neither has a reason to fund trials that might show no benefit for healthy people.

Who funded the trials

• Dexcom was lead sponsor of DIAMOND (type 1) and MOBILE (type 2 on basal insulin), both run with the Jaeb Center for Health Research (DIAMOND; MOBILE). It also funded the pivotal G7 accuracy study (Garg 2022).
• Abbott Diabetes Care funded IMPACT (Bolinder 2016) and was the industry collaborator on IMMEDIATE (registry). Abbott also publicised the 2026 observational mortality data (Abbott).
• ZOE funded METHOD and co-funded the PREDICT sensor-agreement analysis (METHOD; Merino 2022).
• Independent checks came from the unfunded Jancev meta-analysis, NIH intramural scientists (Howard 2020) and NICE's publicly funded evidence review. The German IfDT accuracy study took money from rival glucose-device companies but none from the three makers tested (Eichenlaub 2025).
• Regulator funding: FDA device review is partly paid for by user fees from device companies, under a system in place since 2002 (FDA). NICE is funded by the UK Department of Health and Social Care (framework agreement).

Documented integrity events

• Dexcom, 2025: FDA warning letter (4 March 2025) after inspections in San Diego and Mesa, Arizona. It cited an unapproved sensor component change, failed design validation and corrective-action shortfalls, and it called the G6 and G7 adulterated and misbranded (FDA).
• Abbott, 2025–26: Class I recall of certain Libre 3 and 3 Plus sensors for falsely low readings: 860 serious injuries and 7 deaths reported (FDA).
• Abbott, 2024: voluntary correction of three Libre 3 lots for falsely high readings (Abbott).

Related research

For wider, evidence-graded context, see these Pure City Research guides:

• Type 2 diabetes: prevention and management evidence
• Diabetes and heart disease
• Diabetes supplements: benefits, limits and safety
• Berberine: glucose evidence and safety
• Psyllium husk: cholesterol and blood sugar
• Myo-inositol: PCOS, pregnancy and glucose
• Home blood pressure monitoring
• ECGs and ambulatory rhythm monitoring
• Health anxiety: assessment and treatment

Frequently asked questions

Sources and funding notes

What we could not verify: We could not retrieve the funding statements for the Uhl et al. 2024 (ECRI) meta-analysis or the Dower et al. 2026 JAMA Internal Medicine review, because the journal pages were blocked. We could not confirm whether Abbott's 2026 observational mortality data have been peer reviewed, or the full NICE cost-effectiveness model. The US sensor count and "single production line" cause of the Libre recall come from an AARP report citing Abbott; the FDA notice itself does not give a quantity. We did not find whether the FDA has closed out Dexcom's warning letter. We could not verify ZOE's or Levels' investors, which sensors Levels supplies, or current prices for Stelo and Lingo (Lingo launched in 2024 at US$49 for two weeks to US$249 for 12 weeks, according to Abbott's launch release). We did not review paediatric or pregnancy guidance, or UCL's 2024 regulatory commentary on wellness CGMs (the page was blocked).

1. NICE NG17: Type 1 diabetes in adults, recommendations 1.6.10–1.6.16 (2022) — UK guideline; funded by DHSC.
2. NICE NG28: Type 2 diabetes in adults, blood glucose management, section 1.7 (2022) — UK guideline; DHSC.
3. NICE NG28 update information — March 2022 CGM review date.
4. DHSC–NICE framework agreement — NICE funding.
5. FDA warning letter to Dexcom, 4 March 2025 — US regulator.
6. FDA: Abbott removes certain FreeStyle Libre 3 and 3 Plus sensors (Class I) — US regulator.
7. FDA clears first over-the-counter CGM (5 March 2024) — US regulator.
8. FDA: Medical Device User Fee Amendments — FDA funding model.
9. AARP: FDA warning on faulty glucose monitor sensors — consumer media (member-funded non-profit); secondary report.
10. Jancev M et al. CGM in adults with type 2 diabetes: systematic review and meta-analysis. Diabetologia 2024 (PMID 38363342) — no specific funding; no conflicts declared.
11. Uhl S et al. Effectiveness of CGM on glycemic control in T2DM: meta-analysis. J Clin Endocrinol Metab 2024 — ECRI/University of Arizona/Phoenix VA; funding not retrieved.
12. Dower JA, Johansson M et al. CGM in type 2 diabetes and beyond: a review. JAMA Intern Med 2026 — academic; funding not retrieved.
13. Johansson M et al. CGM for persons with and without diabetes. JAMA Intern Med patient page 2026 — academic.
14. University of Gothenburg news release, August 2026 — university press office.
15. Beck RW et al. DIAMOND trial. JAMA 2017 — sponsored by Dexcom (NCT02282397).
16. Bolinder J et al. IMPACT trial. Lancet 2016 — funded by Abbott Diabetes Care.
17. Martens T et al. MOBILE trial. JAMA 2021 — sponsored by Dexcom (NCT03566693).
18. Aronson R et al. IMMEDIATE trial. Diabetes Obes Metab 2023 — LMC Diabetes & Endocrinology, Abbott collaborator (NCT04562714).
19. Garg SK et al. Accuracy and safety of Dexcom G7. Diabetes Technol Ther 2022 — Dexcom grant; Dexcom-employed authors.
20. Eichenlaub M et al. Performance of three CGM systems in adults with type 1 diabetes. J Diabetes Sci Technol 2025 — IfDT; part-funded by BIONIME, i-SENS, Roche, Diabetes Center Berne; not by tested makers.
21. Shah VN et al. CGM profiles in healthy nondiabetic participants. J Clin Endocrinol Metab 2019 — Helmsley Charitable Trust; lead author Dexcom ties (journal page).
22. Howard R, Guo J, Hall KD. Imprecision nutrition? Am J Clin Nutr 2020 — NIH (NIDDK) scientists.
23. Merino J et al. Validity of CGM for categorising glycemic responses to diet. Am J Clin Nutr 2022 — supported by Zoe Ltd; TwinsUK cohort funding from Wellcome, MRC, EU, NIHR, Zoe Global and others; ZOE-employed and consultant authors.
24. Bermingham KM et al. ZOE METHOD trial. Nat Med 2024 — funded by ZOE Ltd.
25. Abbott: Lingo now available in the US (September 2024) — seller.
26. Abbott: voluntary correction of Libre 3 sensors (July 2024) — seller.
27. Abbott fourth-quarter and full-year 2025 results — seller financials.
28. Abbott: CGM and mortality in basal-insulin type 2 diabetes (September 2026) — seller press release.
29. Abbott and Google partnership (August 2026) — seller.
30. Abbott glucose-ketone sensor authorisation (August 2026) — seller.
31. DexCom, Inc. Form 10-K for 2025 (filed February 2026) — company filing to US regulator.
32. Abbott FreeStyle Libre safety information — seller labelling.
33. Dexcom interfering substances — seller labelling.
34. Levels homepage — seller marketing.

Educational research reviewed 9 October 2026. This article is not medical advice; talk to a qualified clinician or pharmacist before starting, stopping or combining treatments.