This observational study pooled two independent multi-year real-world continuous glucose monitoring (CGM) datasets covering 86,779 people with diabetes and just over 21 million measurement days. Mean age was about 40 in one cohort and 44 in the other. Linear mixed-effects models examined how CGM metrics and bolus insulin doses varied by day of the week and by hour of the day.
The pattern was consistent across both cohorts. Time in range was highest mid-week, with Wednesday running 0.96 percentage points above average (95% CI 0.93-0.99), and lowest at weekends and on Mondays, with Sunday 1.74 percentage points below (95% CI -1.77 to -1.72). Bolus insulin doses were higher at weekends, by 0.69 units on Sundays (95% CI 0.64-0.74). Within the day, control was best on weekday mornings and middays and worst in the evening and overnight, especially at weekends.
The authors are careful to say the effect sizes are small, and at population level they are: one to two percentage points of time in range will not change anyone's management on its own. What the study establishes is that the variation is real, systematic and in a predictable direction, which changes how a report should be read rather than what should be prescribed. When a patient's fortnightly average looks mediocre, the useful question is whether it is mediocre everywhere or whether two evenings a week are dragging it down. Those are different problems with different fixes, and only one of them is a dose problem.
- Ask which days are worst before adjusting any dose; a weekend-only pattern is behavioural, not pharmacological
- Look at the overnight and evening windows separately, since these were consistently the poorest periods
- Note that weekend bolus doses were already higher, so patients are compensating; the gap is in timing and food, not willingness
- Ensure at least 14 days of data spanning two full weekends before drawing conclusions about a weekly pattern
- Use shift patterns and actual working days, not calendar weekends, when the patient's week does not run Monday to Friday
The statistics, in plain English
The confidence intervals here are extraordinarily narrow, such as 0.93 to 0.99 for the Wednesday effect, because the dataset contains 21 million days. That is precision, not importance: with samples this large, differences far too small to matter clinically will still be highly statistically significant, and the p values below 0.001 say only that the pattern is not noise. The reported values are population averages, so an individual patient's weekend dip may be several times larger or entirely absent, which is precisely why the finding is a reason to look at each person's own weekly profile rather than to assume the average applies. Being observational, the study also cannot separate the day of the week from what people do on it.
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.
Scan to keep reading on your phone. No account needed to start.
Tomorrow morning, before your first patient
One edition a day for diabetes & endocrinology, written by the desk, every claim tied to its paper. Six minutes.
Get the app — free