The Detail
A continuous glucose monitor, or CGM, estimates glucose every few minutes and stores a sequence of values. A report can compress thousands of points into a handful of statistics. Mean glucose is usually the arithmetic average: add the included values and divide by their count. That summary is useful because it gives one center for the period. Compression also removes detail. It does not show the order of readings, the distances between them, or when missing values occurred.
Variation is a family of ideas rather than one number. A report may show standard deviation, coefficient of variation, range, interquartile range, a percentile band, or a distribution graphic. Each emphasizes a different feature of spread. NIST describes measures of scale as ways to characterize variability and notes that different measures weight the center and tails differently. Before comparing reports, identify the exact statistic instead of treating every spread measure as interchangeable.
The Average Is a Center, Not a Story
The arithmetic mean gives every included value a role, so unusually high or low values can move it. The median instead marks the middle observation after values are ordered and responds differently to extremes. A device report may use average glucose without stating mean in the headline, then define it in help text. Check the glossary or report notes. If one report uses a mean and another uses a median, identical labels such as average may hide a methodological difference.
The denominator matters too. A 14-day average is calculated from values the system retained, not from readings that were never recorded. Some products sample automatically, while others require scanning or communication to store all data. A rounded headline can also conceal small differences. Preserve the displayed precision and the period rather than copying a bare number into a long-term log.
Variation Describes Spread
Standard deviation summarizes typical distance from the mean in the same unit as the data. A larger standard deviation indicates a wider spread within that dataset, but it does not explain the sequence or cause. Range uses only the minimum and maximum, so one extreme point can dominate it. An interquartile range focuses on the middle half of ordered observations. A percentile band on a daily profile shows where a specified portion of values fell at each time of day.
Coefficient of variation expresses standard deviation relative to the mean, generally as a percentage. That can support comparisons when the centers differ, but only if reports use the same definition and adequate data. A reader does not need to calculate every statistic. The first task is to name what the report provides, find its definition, and avoid translating a technical metric into a clinical judgment that the report cannot make on its own.
Shape and Order Still Matter
A histogram can show whether observations cluster in one region or form several groups. A time-series line can show whether changes were brief, sustained, or repeated at similar hours. A daily profile can align many days by clock time. These displays may share the same mean and standard deviation while arranging observations differently. Statistics and charts are complementary views, not competitors.
Always check the window before connecting a pattern to a date. A daily profile combines days and can make separate events look simultaneous. A smooth line may be an aggregate rather than a literal trace. Legends should say whether a band represents a percentile range, a target setting, or uncertainty. When definitions are missing, the honest conclusion is that the chart cannot support a detailed comparison.
A Worked Example
Illustrative data, not patient results.Teaching Set F contains five glucose values in mg/dL: 110, 120, 130, 140, and 150. Teaching Set G contains 70, 100, 130, 160, and 190. Both arithmetic means are 130 mg/dL. Set F stays close to the mean, while Set G is spread across a much wider range. The example uses only five invented points and is not a valid clinical summary. Its purpose is to show why matching averages do not guarantee matching distributions.
Now assign the values to times. If Set G rises steadily from morning to evening, its line has one shape. If the same five values occur in the order 130, 190, 70, 160, 100, the average and unordered spread are unchanged, but the trace looks different. Even this does not explain what caused the pattern. Time labels, device context, and a complete record would be needed before a qualified professional could discuss relevance.
| Position | Teaching Set F (mg/dL) | Teaching Set G (mg/dL) |
|---|---|---|
| 1 | 110 | 70 |
| 2 | 120 | 100 |
| 3 | 130 | 130 |
| 4 | 140 | 160 |
| 5 | 150 | 190 |
| Arithmetic mean | 130 | 130 |
| Range | 40 | 120 |
Check the Denominator and Window
A summary should state its start and end dates, duration, and data sufficiency or active percentage. Two 14-day reports can contain different amounts of usable data. A daytime-heavy sample and a full-day sample might produce the same mean by coincidence, yet they describe different coverage. Look for sensor start, warm-up, replacement, connection loss, and excluded-day notes. These are properties of the dataset, not explanations of a person’s health.
Also confirm whether the period includes partial first and last days. A report labeled August 1–14 may begin at 6 p.m. on the first day and end at 9 a.m. on the last. Some summaries weight every retained reading equally; others create daily summaries first. Use the product’s definition when available. If calculation details differ, compare displays cautiously and document the mismatch.
What It Does Not Tell You
An average does not show individual highs, lows, timing, duration, or missing intervals. A variation statistic does not identify which events produced the spread. Neither number establishes whether a pattern is safe, diagnoses a condition, sets a personal range, or indicates a treatment change. A threshold line in software may reflect a setting rather than a universal rule. Its source and ownership must be checked separately.
A summary also cannot prove that two device periods are comparable. Sensor models, software versions, report settings, sampling behavior, and wear coverage may differ. A small numerical difference should not be declared meaningful from formatting alone. Preserve the reports, identify the definitions, and ask a qualified healthcare professional about personal relevance.
A Better Reading Order
Begin with the title, date window, time zone, unit, and data coverage. Next read the definition of average and identify the reported variation measures. Then inspect the distribution and time-series views for features that the statistics compress. Finally, read footnotes about exclusions, smoothing, or software changes. This order keeps a striking curve from distracting from an incomplete denominator.
When comparing two summaries, create a small header for each with device, window, number of days, active-data statement, average definition, spread definition, and display precision. Compare only fields that match. Write not stated for missing definitions. The result is a clearer question set, not a verdict about the person represented by the data.