How Many Days of CGM Data Do I Need Before My Report Means Anything
A CGM report starts to mean something at about 14 days, with at least 70% of possible readings present. Three days mostly shows what you happened to eat.
A CGM report starts to mean something at about 14 days, with at least 70% of the possible readings present. That is the standard from the international CGM consensus. GMI, the A1C estimate on a CGM report, is best judged with 14 days or more. A report built on three days mostly describes what you happened to eat on those three days. The 14-day standard does not apply to a single-meal check, so if you only have a few days of data there is still a narrow use for it, covered below.
The international consensus on CGM metrics (Battelino and colleagues, 2019) recommends at least 14 days of data, with the sensor delivering at least 70% of the readings possible in that window. Seventy percent of 14 days is a little under 10 days of actual data. The aim is a window long enough that average glucose and time in range sit close to your longer-run picture.
Now the limit. That consensus was set for people with diabetes. It does not set a separate minimum for people without diabetes. Two weeks is a sensible default. Whether a flatter glucose profile settles sooner has not been established.
Why three days is mostly noise
Glucose at 1 pm depends on what you ate at noon, how recently you trained, how stressed you were and what happened the evening before. Each of those changes from day to day. Three days gives you three samples of those conditions, which is too few to show what is typical.
The AGP (ambulatory glucose profile) stacks all your days onto one 24-hour clock and draws a median line with two bands: the middle half of readings and the middle 90%. A sensor logs many readings an hour, but neighbouring readings within an hour mostly come from the same meal or event, so they are not independent evidence. What counts is how many days contribute to each time of day. With three days, each time of day has only three days behind it. The median is whichever of the three sits in the middle, and the 90% band is essentially the highest and lowest of the three. With 14 days the bands start to mean something.
As an illustration, a 3-day report might show a ragged median, a band that balloons at dinner because of one large meal, and a GMI that shifts when you add a few more days. GMI is 3.31 + 0.02392 × your average glucose in mg/dL, so every 10 mg/dL of average is about 0.24 GMI units. A single day averaging 30 mg/dL above your usual lifts a 3-day average by 10 mg/dL, or 0.24 GMI. Spread over 14 days, the same day lifts it by about 2 mg/dL, around 0.05.
Short runs also have no slack. Lose one day to a sensor falling off and you have lost a third of a 3-day run, compared with about 7% of a 14-day one.
How to count wear time and gaps
Wear is the readings you have divided by the readings possible across the whole window, not a count of calendar days. A 14-day sensor that fell off after day 9 gives you 9 of 14 days, about 64%, which is under the 70% mark.
- Work out expected readings: readings per day times days in the window.
- Divide the readings you actually have by that number.
- Then look at where the gaps fall, because that matters as much as the total.
| Reading every | Readings per day | Expected in 14 days | Needed at 70% | Most you can miss |
|---|---|---|---|---|
| 15 minutes | 96 | 1,344 | 941 | 403 |
| 5 minutes | 288 | 4,032 | 2,823 | 1,209 |
Scroll the table sideways for the rest →
Arithmetic from the consensus rule of 14 days and at least 70% of readings. Needed counts are rounded up, so they do not understate the requirement. Exports commonly use 5 or 15 minute intervals, but your sensor's interval may differ; check your export.
Gaps in the wrong place do more damage than their size suggests. Take two reports. In the first, scattered 20-minute gaps add up to 15% missing, so it is about 85% wear and every time of day is still represented. In the second, six hours go missing every night. Six hours is 25% of the day, so that report is about 75% wear, and the whole overnight profile is gone. The two look close on the wear number, and the second is far less useful.
| Pattern | Wear | Midnight to 6 am | 6 am to noon | Noon to 6 pm | 6 pm to midnight |
|---|---|---|---|---|---|
| Scattered 20-minute gaps, 15% missing | About 85% | Covered | Covered | Covered | Covered |
| Six hours missing every night, 25% missing | About 75% | No data | Covered | Covered | Covered |
Scroll the table sideways for the rest →
Illustration of the two examples in the text, not measured data. It assumes the six missing hours fall between midnight and 6 am every night.
Short, sharp overnight dips that recover within minutes may be compression artefacts from lying on the sensor. Do not assume a low is false, though. An unexplained low, or feeling unwell, should be checked with a finger-prick meter and a doctor.
What a short run can and cannot show
| Number | After 3 to 5 days | After 14 days, 70% worn |
|---|---|---|
| GMI / average glucose | Too thin | Reasonable |
| Time in range | Swings with one day | Reasonable |
| AGP shape and bands | Too thin | Usable |
| Variability (CV) | One odd day dominates | Steadier |
| Overnight lows | Only a few nights to judge | Patterns visible, confirm unexplained lows with a finger-prick meter |
Scroll the table sideways for the rest →
Built on the consensus 14-day window, which was set for people with diabetes. The ratings for short runs are a judgement, not a published threshold.
A few days can still tell you something narrow. It can show how you responded to one specific meal, or whether two breakfasts eaten on different days look very different. Treat that as a hint, since a single exposure is one data point.
It cannot tell you your typical day, your overnight pattern or your variability. It also cannot support a fair comparison between a short-run GMI and a lab A1C. GMI is an estimate from average sensor glucose, not a lab test (Bergenstal and colleagues, 2018). A measured A1C can differ from it even with complete data, and the lab value reflects a much longer stretch than your sensor did. A gap between the two does not mean either is wrong.
If you have 7 to 10 days, read the report as provisional. Check the wear percentage first, using the steps above. Then keep wearing the sensor, or start another, until you reach a 14-day window. A shorter window does not meet the consensus standard however complete the wear is.
Practical advice, not a consensus rule: if you can choose, use a typical stretch rather than a week with illness or travel. A 14-day window also covers two weekends, which helps if your weekends look different from your weekdays.
The rule in one line: 14 days in the window and at least 70% of possible readings present, before you read much into the shape of the report.
Reaching 14 days or more with two sensors
The 14 days comes from the consensus, not from how long a sensor lasts. Wear periods differ by product, so check yours, and warm-up, gaps or early failure can leave a single sensor short of a 14-day window. If you want a month, you need a second sensor and a way to join the two records.
Stitching is simple in principle. Export each sensor's file and put them in time order. Where the files overlap, you must choose one reading per timestamp.
- Overlap the two sensors by about a day. Start the second sensor roughly a day before the first ends, so it can finish warm-up while the first is still recording and the join has no gap. The overlap costs sensor days: two 14-day sensors with one day of overlap cover about 27 days, not 28.
- Where the two overlap you will have two readings for the same timestamp, and they can differ, so they are not true duplicates. If you join by hand, pick one rule for the overlap, for example using the first sensor until it ends, and apply that one rule across the whole join. That gives a single switchover point that is easy to check.
- Let a tool handle the join where you can. The free report at healthosx.com/glucose accepts several files at once, so you can add both sensors' exports together. The HealthOSX app's glucose import (Settings, Import glucose from a file) adds only what is new from an overlapping export, so a second file does not double up readings you already have.
- Stay with one brand where you can. Sensors differ by model, and this page does not quantify how much mixing brands matters.
- Check the join. A step change right where one file ends and the next begins may be the switch between sensors rather than a real change in your glucose.
If you already have an export from LibreView, Dexcom Clarity, Stelo, Lingo or SiBionics, the free CGM report at healthosx.com/glucose shows your AGP, time in range and biggest rises in the browser, with no account. If you have no sensor, HealthOSX Sensor is a month of continuous glucose with the app, two sensors of about two weeks each, at healthosx.com/glucose/sensor. That gives you enough time to reach a 14-day window, but it does not guarantee one, because warm-up, gaps or an early failure can eat into it and you still need at least 70% of readings present.
Written by
HealthOSX Research