On closed-loop insulin, needs rose and time in range fell in the luteal phase — but not for everyone

Design

Decentralised observational study with Bayesian state-space modelling

Population

77 menstruating adults with type 1 diabetes on AID, 380 cycles

Primary outcome

Glycaemic outcomes, insulin requirement and insulin sensitivity by cycle phase

Effect

Insulin sensitivity +2.6% (CrI 0.3 to 5.0) early follicular vs −2.6% (−5.2 to −0.1) midluteal

An observational study in Diabetes Care (September 2026) used donated real-world data from 77 menstruating adults with type 1 diabetes on automated insulin delivery (AID), covering 380 menstrual cycles of continuous glucose monitoring, insulin delivery and carbohydrate records.

Total daily insulin and carbohydrate intake both peaked in the luteal phase, with higher mean glucose and less time in range. A model estimating insulin sensitivity from the glucose traces put it about 2.6% higher than average in the early follicular phase and about 2.6% lower in the midluteal phase. About 85% of participants followed the population pattern, but individual variation was wide.

The design limits what can be concluded: participants self-selected, cycle phase was self-reported, and the sensitivity estimate is model-derived rather than clamp-measured. Current AID algorithms do not account for cycle phase.

The practical point is for the consultation, not the algorithm: asking about cycle timing can explain an otherwise puzzling recurring premenstrual rise.

• Ask menstruating patients with type 1 diabetes whether their glucose rises in the days before a period
• Look for a recurring luteal-phase fall in time in range when reviewing CGM downloads
• Consider a temporary higher target or more aggressive settings in the luteal phase for those with a clear pattern
• Expect roughly one in seven not to follow the typical pattern — individualise
• Record cycle dates alongside glucose data when adjusting settings

Why it matters

Closed-loop systems still leave a monthly, predictable excursion that neither guidelines nor algorithms currently address.

Don't overread it

The insulin sensitivity figures are model estimates from self-selected, self-reported data, not measured directly.

The statistics, in plain English

A credible interval is the Bayesian counterpart of a confidence interval. Both phase estimates (+2.6%, 0.3 to 5.0; −2.6%, −5.2 to −0.1) sit just clear of zero, so the direction is consistent but the size is small and uncertain at population level.

Read the rest in the app

You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.

Get Daily Doctor — free

Scan to keep reading on your phone. No account needed to start.