Three years on a glucose sensor linked to lower blood sugar and less urine protein, not better kidney function

Trials of glucose sensors in type 2 diabetes have mostly lasted months. A study from a large California health system has now followed people for up to three years. Those who started a real-time sensor and kept using it had lower blood sugar than similar people who never started one, and the gap held for the whole period. By the third year they also had less protein in their urine, an early sign of kidney damage.

The study, published on October 7 in JAMA Network Open, was not a randomized trial. A blood test of how well the kidneys filter showed no difference between the groups. And nearly half of the sensor users stopped along the way, which means the results describe the people who persisted.

Why kidney outcomes were the missing piece for glucose sensors

A continuous glucose monitor is a small medical device worn on the skin, in the words of the health system’s summary of the study. Its sensor reads glucose in the fluid just beneath the skin every few minutes. The technique is called continuous glucose monitoring (CGM). A real-time device sends its readings to the wearer automatically, while other models show a reading only when scanned.

Short-term trials have established that sensors lower blood sugar in type 2 diabetes. What happens over years was less clear, and so was any effect on the complications that high blood sugar causes. The authors of the new paper put it this way: whether CGM has a secondary benefit for kidney function preservation or reducing kidney damage has not been established by existing trials.

Kidney disease is one of the most common of those complications. The health system’s summary puts the share plainly, saying that about 30% of patients with type 2 diabetes eventually develop chronic kidney disease.

Two routine tests track it. The first is the urinary albumin-creatinine ratio. Albumin is a blood protein that healthy kidneys keep out of urine, so a higher ratio means the kidneys’ filters are leaking. The second is the estimated glomerular filtration rate (eGFR), which is calculated from a blood test and reflects how much blood the kidneys clean each minute. According to the paper, chronic kidney disease in diabetes is typically defined as an albumin-creatinine ratio of 30 milligrams per gram or more, or an eGFR below 60 milliliters per minute (adjusted for body size), lasting more than three months.

How the study compared sensor users with other people with diabetes

The work was done at Kaiser Permanente Northern California, with Andrew Karter as first author and Ajit Mahapatra as senior author. The researchers used a method called target trial emulation. Target trial emulation means designing an analysis of existing health records to follow the rules a randomized trial would have followed, including who is eligible and when the clock starts.

They began with a registry of 573,663 people with type 2 diabetes who had never used a sensor. From it they kept adults whose diabetes was hard to manage. To be eligible, a person needed an HbA1c above 8% in the previous year, or a history of emergency department or inpatient treatment for hypoglycemia. HbA1c is a blood test that reflects average blood sugar over roughly the previous three months. Hypoglycemia is blood sugar that has fallen dangerously low.

That left 25,648 people. Of these, 2,771 started a real-time sensor between 2015 and mid-2023, and 22,877 served as the comparison group. For every person who started a sensor in a given month, the team randomly sampled approximately 10 eligible patients not using CGM in that same month. Nearly all, 92.9%, entered the study in 2019 or later.

The average age was 61.5 years, and 54.7% were men. Average HbA1c at the start was 9.6% among sensor users and 9.5% in the comparison group.

In other respects the groups were unalike. Among those who started a sensor, 2,468 (89.1%) were using insulin, against 8,881 (38.8%) of the comparison group. A history of hospital care for low blood sugar was more common too, at 7.5% against 2.0%. The paper adds that sensor users had a higher mean comorbidity burden, along with more kidney damage and poorer kidney function at the outset. Comorbidity means having other illnesses alongside the main one.

To make the groups comparable, the analysis weighted each person according to 74 recorded characteristics, covering demographics, other illnesses, use of health care and prescriptions. The weights were updated every three months.

One design choice shapes everything that follows. The analysis was per-protocol. Per-protocol means people are counted only for as long as they stick to the group they started in. A sensor user who went more than 180 days without refilling supplies was removed from that point on, and so was anyone in the comparison group who started a sensor. Follow-up stopped at 36 months because, as the paper explains, after that, the CGM-using sample decreased below 25% of baseline. Stopping the sensor was only one way out of the analysis. People also left it when their health plan coverage lapsed, when they died or when the study period ended, so far fewer than the 2,771 starters contributed three-year data.

What happened to blood sugar over three years

Blood sugar fell quickly in sensor users and stayed lower. The paper describes a rapid, durable and clinically substantive decline in HbA1c.

One year in, average HbA1c was 1.1 percentage points lower in people who had started and stayed on a sensor than in the comparison group, with a range from 0.90 to 1.30 points. A range here is the margin of uncertainty around an estimate. At three years the gap was about 0.8 points. Average HbA1c was 7.83% in the sensor group, with a range from 7.55% to 8.11%, and 8.58% in the comparison group, with a range from 8.49% to 8.66%.

The price of that long follow-up was a shrinking sensor group. By the end of three years, 1,228 of the 2,771 people who started a sensor (44.3%) had stopped using it and been removed from the comparison. In the other group, 4,284 people (18.7%) were removed because they started a sensor. Average time in the study was 2.2 years for sensor users and 2.1 years for the comparison group.

What the kidney tests showed in sensor users

The two kidney tests gave different answers.

For protein in urine, after weighting, nothing separated the groups at the start or over the first two years. At three years the average albumin-creatinine ratio was 153 milligrams per gram in the sensor group and 278 in the comparison group. The difference of 125 had a range from 54.0 to 196.6. Both averages are well above the threshold of 30 that marks kidney disease.

For filtering, the result was null. The summary of results states that there were no differences in eGFR during follow-up.

The authors offer a reason the two might diverge. Microalbuminuria typically precedes decreases in eGFR, although the timing is poorly understood, they write. Microalbuminuria is a small but abnormal amount of albumin in urine. On that reading, three years may be too short to see filtering change.

Mahapatra, a kidney specialist, made the same argument in the health system’s summary. “Kidney function is the ultimate marker of kidney health, but reduction in urine protein is an accepted early marker for preserved kidney function and reduced kidney decline,” Mahapatra said. Mahapatra went on: “With a longer study period, I think we may have also seen preservation of kidney function.” That is a hope about a study not yet done. The paper’s own wording is that longer follow-up may be required to investigate whether sensors go with preserved kidney function.

The paper is also careful about why the urine marker moved. The authors suspect that better blood sugar control is responsible, and state that an observational study cannot verify it. Their conclusion is hedged: sensor use may extend the time before pharmacologic escalation is required and potentially delay diabetic kidney disease progression. Pharmacologic escalation means adding or increasing medicines.

How this fits with trials of glucose sensors in type 2 diabetes

Randomized trials have found smaller effects on blood sugar, over much shorter periods.

A meta-analysis published in the journal Diabetologia in 2024 pooled 12 randomized trials with 1,248 participants. Sensor use lowered HbA1c by 0.31 percentage points on average compared with fingerstick testing, and by 0.36 points in the trials of real-time devices. The reviewers flagged two gaps. One was that all studies were open label. Open label means participants and their clinicians knew who had a sensor, which is hard to avoid with a device. The other gap is the one the California study set out to address. Small-vessel complications of diabetes, which include kidney, eye and nerve damage, are called microvascular complications, and the review found that no trials reported data on microvascular complications.

Two of the larger US trials show the pattern. One enrolled 158 adults who had had type 2 diabetes for a median of 17 years and were injecting insulin several times a day. After 24 weeks, HbA1c was 7.7% with a sensor and 8.0% with usual care. The other was conducted at 15 centers in the US among 175 adults on once- or twice-daily long-acting insulin. At eight months HbA1c had fallen from 9.1% to 8.0% with a sensor and from 9.0% to 8.4% with a standard meter, an adjusted difference of 0.4 points.

Why would the California figure be two to three times as large? The authors give two possible reasons. Their participants started with higher blood sugar, and they note that larger treatment effects are expected among higher-risk patients. The second reason is a possible bias. The gap could partly reflect the health system prescribing sensors selectively to more engaged or better supported patients.

The same group has met this problem before. In 2021 it reported a one-year study of insulin-treated patients with diabetes selected by physicians for real-time continuous glucose monitoring. HbA1c fell by 0.40 points more in those who started a sensor than in those who did not, and the authors cautioned then that selection could have shaped the result.

On the kidney question there is indirect support for the idea that steadier glucose matters. A 2018 reanalysis of a landmark type 1 diabetes trial looked at time in range, the share of readings between 70 and 180 milligrams per deciliter. For each 10 percentage points less time in range, the rate of developing microalbuminuria was 40% higher. The authors concluded that time in range is strongly associated with the risk of microvascular complications. That analysis concerned type 1 diabetes and fingerstick readings, and it measured association only.

What the glucose sensor study cannot show

Not a randomized trial. Clinicians chose who got a sensor, and patients chose whether to keep using it. Weighting can only balance what was recorded. The authors name what was not: residual, unmeasured confounding by baseline or time-varying factors, such as clinician support or baseline motivation. Confounding means that some other difference between the groups explains part of the result.

People who stopped. With 44.3% of sensor users removed for stopping, the three-year figures rest on those who persisted. The paper says so directly, stating that substantial attrition is a limitation and should be clearly acknowledged when interpreting these results. Its conclusion limits the urine result to people who were not lost to follow-up.

A marker, with a late and single signal. The urine protein difference appeared only at the final time point, after two years of no difference. The study did not count kidney failure, dialysis or deaths.

No change in filtering. The measure that Mahapatra called the ultimate marker of kidney health did not move.

One health system, one kind of device. Everyone was a member of an integrated system in Northern California. Only real-time sensors were studied, and the authors say it is unclear whether the results apply to devices that must be scanned. Some people in the comparison group may have bought a sensor elsewhere, and the paper allows that controls could have purchased CGM outside Kaiser.

Glucose patterns were not captured. The study had HbA1c, an average. It did not have day-to-day glucose patterns, which, the authors write, likely have a greater influence on kidney outcomes than measures of mean glucose or HbA1c.

Who paid. The health system’s summary states that the study was funded by Dexcom Inc., a sensor maker, together with The Permanente Medical Group, the Kaiser Foundation Research Institute and two US National Institutes of Health institutes.

Who it covers. Participants had poorly controlled diabetes or a history of dangerous lows, and most sensor users were on insulin. The Kaiser Permanente summary notes that whether the monitors should be offered to all individuals with type 2 diabetes is currently an open area of research.

Do not stop or change a prescribed medicine without talking to your doctor. Whether a glucose sensor makes sense for a particular person with type 2 diabetes is assessed by their doctor.

Three years of sensor use went with lower blood sugar and, late on, less protein in urine, and whether that translates into kidneys that last longer is a question the study leaves for longer follow-up.