Continuous glucose monitoring is having its standard-of-care moment. The 2026 ADA Standards of Care broadened CGM from a tool for insulin users to something recommended at the onset of type 2 diabetes and “anytime thereafter” including for patients on non-insulin therapies. At the same time, over-the-counter sensors are being sold to people with no diabetes at all as wellness devices. That combination a genuine evidence-based expansion happening simultaneously with a marketing wave makes CGM the perfect Tier Check. The technology is the same; the evidence behind each use is wildly different.
Here’s the discipline this whole professional series is built on: the strength of evidence for “CGM” depends entirely on who is wearing it and why. Sort it, and the picture clarifies fast.
Demonstrated CGM in insulin-treated diabetes
This is settled. In type 1 diabetes and in insulin-treated type 2 diabetes, randomized trials show CGM improves glycemic control and reduces hypoglycemia. DIAMOND and GOLD established the benefit in type 1 adults on multiple daily injections; MOBILE extended it to type 2 patients on basal insulin in primary care, showing meaningful HbA1c reduction versus fingerstick monitoring. The mechanism is intuitive real-time data and trend arrows let a patient (and clinician) titrate insulin and catch lows that a few daily fingersticks miss. When the therapy carries hypoglycemia risk and requires active dose adjustment, CGM earns its place on hard, repeatedly-replicated grounds. Tier: demonstrated.
Demonstrated on the surrogate CGM in non-insulin type 2 diabetes
This is the newer and more interesting story, and it’s where the 2026 ADA expansion lives. A growing body of RCTs and a 2025 systematic review and meta-analysis show that CGM in non-insulin-treated type 2 diabetes produces a modest but real HbA1c reduction, on the order of 0.5%, compared with usual care. The CONNECT trial, presented at the 2026 ADA Scientific Sessions, randomized 283 non-insulin type 2 adults in primary care to a CGM or standard care and reported that most CGM users improved their glycemia over six months. The presumed mechanism is behavioral, not pharmacologic: seeing the glucose consequence of a specific meal or walk in near-real-time drives dietary and activity change in a way a quarterly A1c never could.
But note the ceiling precisely. What’s demonstrated is a surrogate benefit HbA1c and time-in-range improve. What is not yet demonstrated in this population is a hard-endpoint benefit: fewer complications, events, or deaths. The effect is real, the magnitude is modest, and it lives on the surrogate. That’s still enough to justify the ADA’s recommendation a 0.5% A1c improvement from a safe, behavior-driven tool is worthwhile but the honest label is “demonstrated on the surrogate, inferred on outcomes.” Tier: demonstrated (surrogate); outcomes inferred.
Inferred time-in-range as a stand-in for complications
CGM’s signature metric, time-in-range (TIR, typically 70–180 mg/dL), is now endorsed by the 2026 ADA Standards as a valid glycemic assessment, and it’s genuinely useful it captures variability and hypoglycemia that A1c hides. The International Consensus on Time in Range gave it standardized targets. And observational analyses (largely re-analyses of DCCT data) show that lower TIR correlates with more retinopathy, albuminuria, and neuropathy.
But apply the surrogate discipline from the food series here. TIR’s link to hard complications is correlational derived from observational reanalysis, not from a trial that raised TIR and measured whether complications fell independent of A1c. No RCT has used TIR as a primary outcome to prove it’s a causal, treat-to-target surrogate. So TIR is an excellent descriptive metric and a reasonable management target, but the claim “raise TIR and you’ll prevent complications” is inferred from correlation, not demonstrated. Use it; don’t overstate what moving it proves. Tier: inferred.
Speculative CGM for people without diabetes
Now the marketing frontier. Since 2024, over-the-counter sensors Dexcom’s Stelo (FDA-cleared March 2024) and Abbott’s Lingo (June 2024) are sold without a prescription to adults not on insulin, explicitly as wellness and “metabolic optimization” tools. The pitch: see how your body responds to food, sleep, and exercise, and optimize.
Here the tier is unambiguous, and the ADA says so plainly: there is insufficient evidence to support CGM for screening or diagnosis of prediabetes or diabetes, and there is no strong evidence that CGM improves health outcomes in people without diabetes. A 2024 review across diabetic and non-diabetic users found CGM feedback moved A1c by only ~0.3% and TIR by ~7% overall small, and unproven to matter in healthy people. Meanwhile there are plausible harms specific to this population: glucose data are easy to over-interpret, normal postprandial excursions can be misread as pathology, and clinicians have flagged the risk of anxiety and disordered eating in people who fixate on a number that was never abnormal. Tier: speculative heavily marketed, outcome-unproven, and not without downside.
The honest version
CGM is a genuine standard-of-care advance for the right population, and the 2026 ADA expansion into non-insulin type 2 diabetes is evidence-based, not hype a modest, real, surrogate-level benefit from a safe behavioral tool. That legitimacy is exactly what the wellness market is borrowing to sell sensors to people with no diabetes, where the evidence collapses. The skill is to not let the strong end of the spectrum launder the weak end. Same device, four very different claims.
Clinical takeaway
When “CGM” comes up, ask who and why before you assign a tier. Insulin-treated diabetes: demonstrated, recommend it. Non-insulin type 2 diabetes: demonstrated on the surrogate, now guideline-endorsed, worth offering while being honest it’s an A1c/behavior benefit, not a proven outcome benefit. Time-in-range: a useful metric whose complication link is inferred from correlation, not proven causally. And a healthy person asking whether to buy a wellness sensor: no outcome evidence, real potential for harm, and an honest “probably not, and here’s why.” One technology, four tiers which is the whole point of a Tier Check.