Design: Decentralised observational study of donated real-world CGM and insulin data
Population: 77 menstruating adults with type 1 diabetes on AID, 380 cycles
Primary outcome: Insulin requirements, glycaemia and modelled 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

A decentralised observational study used donated real-world data from 77 menstruating adults with type 1 diabetes on automated insulin delivery (AID), covering 380 cycles. A Bayesian model estimated insulin sensitivity from CGM traces.

Total daily insulin and carbohydrate intake peaked in the luteal phase, with higher mean glucose and lower time in range. Modelled insulin sensitivity was 2.6% higher in the early follicular phase and 2.6% lower in the midluteal phase. Most participants (84.7%) followed this pattern, but individuals varied widely.

AID algorithms adapt, yet they did not fully absorb the cycle. For a patient whose CGM reports show a recurring week of poorer control, the cycle is a likely explanation worth asking about before changing settings wholesale.

Ask menstruating patients whether glucose worsens at a predictable point in the cycle
Look at CGM data by cycle phase if the patient tracks periods
Consider temporary targets or profile changes for the luteal week rather than permanent setting changes
Expect individual patterns to differ from the average

Why it matters

A monthly dip in time in range may be physiology, not poor adherence or failing settings.

The statistics, in plain English

The ±2.6% figures are model estimates of a latent insulin sensitivity parameter, not a directly measured change, and their credible intervals sit only just clear of zero. The population average is small; the clinically useful point is that most individuals shifted in the same direction, though by different amounts.