Smart insulin-pen alerts matched sensor-only care in a type 1 diabetes trial

A system that turns a continuous glucose sensor into a prompt to inject insulin — rather than just a readout of numbers — worked about as well as standard sensor-guided care in people with hard-to-control type 1 diabetes, and the people who acted on its alerts did best of all [s1]. The trial met its goal of being no worse than the comparison, but it did not show the smart system was better, and the most striking finding came from an analysis planned after the fact.

Most people with type 1 diabetes who inject insulin use multiple daily injections (MDI): a long-acting dose plus mealtime shots. Many now wear a continuous glucose monitor (CGM), a sensor that tracks glucose minute to minute and alarms when it drifts too high or low. But a CGM only tells you the number; deciding what to do about it is left to the user. The "smart MDI" system tested here adds a layer on top — its alerts prompt a specific insulin-bolus action, nudging the user toward a dose rather than simply reporting a reading [s1].

What the trial did

After a three-week run-in, 143 participants with type 1 diabetes and an HbA1c of 8% or higher — a marker of poor average control — were randomly assigned to either the smart MDI system or MDI plus a standard CGM [s1]. The main question was the difference between groups in "time in range" (TIR) at six months — the share of the day spent in the target glucose band — tested for non-inferiority, meaning the smart system had to prove it was not meaningfully worse, within a margin of 7.5 percentage points [s1]. HbA1c was tested the same way, with a margin of 0.4 percentage points, and the trial also checked whether the smart system was outright superior [s1]. The trial was registered in advance [s2].

What it found

The smart system cleared the non-inferiority bar on both counts. At six months, the difference in time in range between the groups was 0.19 percentage points (95% confidence interval −4.9 to 5.3) — in other words, essentially identical [s1]. HbA1c also met non-inferiority (difference −0.13 percentage points, −0.46 to 0.19) [s1]. Neither measure showed superiority: the smart system matched standard CGM care rather than beating it [s1].

The more interesting signal came from a post-hoc look at how much people engaged with the alerts. Participants who responded to more than the median alert rate of 28.3% (31 people) gained more time in range than those who responded less (28 people) — a difference of 8.12 percentage points (p=0.04) — and lowered their HbA1c by 0.6 percentage points more (p=0.02) [s1]. That pattern was echoed in a real-world group of 5,507 users of the system [s1]. On safety, one episode of diabetic ketoacidosis occurred during the run-in and one severe low-glucose event in the control arm [s1].

How to read it

The design is sound: a randomised trial with prespecified non-inferiority margins is the right way to ask whether a new tool is an acceptable alternative to an established one, and the answer here is yes [s1]. For people who find a plain CGM gives them data without direction, a system that translates readings into an action could be genuinely useful.

But two cautions matter. First, non-inferior is not superior — the trial does not show the smart system produces better control than CGM alone, only that it holds its own [s1]. Second, the headline about engagement comes from a post-hoc subgroup analysis, which splits people by their own behaviour after the fact. People who diligently respond to alerts may differ in many ways from those who do not, so "responding to alerts drives outcomes" is a plausible hypothesis the data support, not a proven cause-and-effect [s1]. The reassuring flip side is that the benefit appeared at a fairly low bar: acting on as few as 28% of alerts was enough to see improvement [s1]. For the broader picture, see our coverage of closed-loop insulin delivery in type 1 diabetes.

What to watch

The open questions are whether a head-to-head trial powered for superiority would show the smart prompts beating standard CGM, how durable the engagement effect is once novelty fades, and whether the approach helps the people who struggle most to act on their data. For now, the result adds a credible option for injection users — and a reminder that in diabetes technology, what you do with the numbers matters as much as having them.

This article describes research and is not medical advice. Decisions about insulin and diabetes technology belong with a qualified clinician.