A glucose-monitor study reached for a hard outcome — death — and found a modest signal

Continuous glucose monitoring has an enormous evidence base for what it does to glucose numbers and almost none for what it does to the outcomes those numbers stand in for. A Diabetes Care analysis is an attempt to cross that gap, and the accompanying commentary published on August 20 frames it as the field entering an era of hard outcomes [s2].

The study asks whether starting a continuous glucose monitor is associated with lower all-cause mortality in adults with type 1 diabetes [s1]. It is not a randomised trial. It is a target trial emulation — a design that specifies the trial you would have run, then reconstructs it from routine records.

The design, and why the design is the story

The data come from Veterans Health Administration health records [s1]. Eligible participants were adults with type 1 diabetes whose second endocrine visit — the study's time zero — fell between 2017 and 2020 [s1].

The mechanics matter. Each patient was cloned into both treatment strategies, "initiate CGM" and "do not initiate CGM", and then censored if their observed care deviated from the strategy they had been assigned during a six-month grace period [s1]. That artificial censoring is not random, so inverse probability weights were used to account for it, and negative outcomes were examined to probe residual confounding [s1].

Of 8,423 individuals initially assigned to both treatment groups, 1,039 were prescribed CGM devices, while 7,399 were neither censored nor assigned CGM during the grace period [s1].

What it found

Mortality was lower with CGM initiation. The adjusted risk ratios ranged from 0.90 (95% CI 0.71–0.97) to 0.84 (95% CI 0.72–0.97) over one to four years of follow-up [s1]. Similar risk ratios appeared when the grace period was set at three or nine months instead of six [s1].

People older than 65 years, or not using insulin pumps, appeared to benefit more, while effects did not vary by HbA1c, race or ethnicity, or frailty [s1].

The most informative result is a negative one. Risk ratios did not differ between the groups for incident non-diabetes outcomes, including outpatient or inpatient musculoskeletal and gastrointestinal conditions [s1]. If the mortality difference were simply the shadow of healthier, more engaged patients being the ones who get monitors, that shadow should have fallen across unrelated outcomes too. It did not.

What it does not settle

Negative-control outcomes narrow the space for confounding; they do not close it. People who are prescribed a glucose monitor differ from people who are not in ways records do not capture, and the direction of that difference is not obviously neutral.

The confidence intervals also sit close to the null. An upper bound of 0.97 at both ends of the follow-up range means the data are compatible with a benefit of a few per cent as well as with one of roughly a quarter. This is a single observational analysis in one health system, in a population — Veterans Health Administration enrollees — that is older and more male than the general type 1 diabetes population.

Nor is a mechanism established here. The analysis reports an association between initiation and death, not a chain from sensor to glucose to survival.

Where the mechanism evidence is

A separate pooled analysis, published on August 6, gets closer to that chain by asking what CGM metrics predict [s3]. It pooled individual-level data from ten type 1 diabetes studies, and among 1,550 participants recorded 65 severe hypoglycaemia and 163 diabetic ketoacidosis events over a median 16 weeks of follow-up [s3].

Higher time below range, greater glucose variability and CGM-defined hypoglycaemic episodes were associated with severe hypoglycaemia independent of HbA1c — for example, a hazard ratio of 1.52 (1.29–1.79) per one-standard-deviation higher time below 70 mg/dL [s3]. For ketoacidosis, the associations attenuated after HbA1c adjustment, but time in range, sensor mean glucose and standard deviation remained significant, with a hazard ratio of 0.76 (95% CI 0.60–0.97) per one-standard-deviation higher time in range [s3].

That is the intermediate step the mortality analysis assumes but cannot observe: CGM metrics carry risk information that HbA1c alone does not [s3].

What to watch

Whether the association survives replication outside a single integrated system, and whether any randomised trial is ever powered for mortality in this population — a question of arithmetic more than of will, given how few deaths a type 1 diabetes trial of feasible size would accrue.

This article is informational and is not medical advice. Decisions about diabetes technology belong with a clinician who knows the individual case.

Sources

• Continuous Glucose Monitoring Initiation Is Associated With Reduced Mortality in Older-Onset Type 1 Diabetes Patients: A Target Trial Emulation Study Within the Veterans Health Administration — Diabetes Care, published online 2026-06-19
• Beyond Glycemia: Entering the Era of Hard Outcomes for Continuous Glucose Monitoring — Diabetes Care, 2026-08-20
• Continuous Glucose Monitoring Metrics and Risks of Severe Hypoglycemia and Diabetic Ketoacidosis in Type 1 Diabetes: A Pooled Analysis — Diabetes Care, 2026-08-06