Design: Observational, retrospective; two independent real-world CGM cohorts, linear mixed-effects models
Population: 86,779 people with diabetes; 21,011,021 measurement-days
Primary outcome: Weekly and daily patterns in CGM metrics and bolus dose
Effect: Sunday time in range -1.74 points (95% CI -1.77 to -1.72); weekend bolus +0.69 units

An observational analysis pooled two independent real-world CGM cohorts, 86,779 people with diabetes and 21 million measurement-days, and modelled how glycaemic metrics moved through the week and the day.

Time in range was highest mid-week and fell at weekends and on Mondays, with Sunday running about 1.74 percentage points below the weekly mean, while weekend bolus doses rose by around 0.7 units. Control was best on weekday mornings and worst on weekend evenings and nights.

The swings are small and will not change how a single reading is read, but they are consistent across two separate cohorts. When a download shows weekend-clustered highs, that is a behavioural pattern to explore, meals, routine and activity, rather than an automatic reason to change the regimen.

Time in range was consistently higher mid-week and lower at weekends and on Mondays.
Sunday time in range ran about 1.74 percentage points below the weekly average.
Weekend bolus insulin doses rose by roughly 0.7 units.
Weekday mornings showed the best control; weekend evenings and nights the worst.
Read weekend-clustered highs as a routine-and-meals conversation, not a reflex dose change.

Why it matters

It shows a single clinic HbA1c or a short download can miss a reproducible weekly rhythm.

The statistics, in plain English

With 21 million measurement-days, very narrow confidence intervals make even tiny differences statistically certain; the roughly 1-2 percentage-point swings are real but modest, so judge clinical size, not the p value.