How to read your AGP report

What you are holding

Two pages, always: page one is the glucose summary and page two is what you logged. It stays two pages whether you asked for a fortnight or a year, because a report that spills onto a third page when the range grows is awkward to hand over and worse to fax, and a clinic still faxes things.

You choose the start and end dates before it is generated, and what comes back is a file with those dates and the time it was made printed on it. Generating the report is a Premium feature.

One thing to say before any of the parts: this is a report to hand to your clinician. It is a picture of your own data, drawn to a layout they already know. It contains no interpretation of that picture, and neither does this page.

The strip across the top

Five figures, side by side, and each is a plain piece of arithmetic over the readings in the range you asked for.

• Days of data. How many separate local days in the range had at least one reading on them. Not the length of the range: the days that actually carry data.
• Share of the window the sensor was recording. The readings you have, against the number a sensor running at its usual cadence would have produced over the same stretch. A low figure here says the sensor was off your arm, or the readings did not all arrive, and every other figure on the page is standing on that much less.
• Average glucose. The mean of every reading in the range, whether it arrived from a sensor or you typed it in yourself.
• Glucose Management Indicator. A figure worked out from your average glucose and from nothing else, by a published formula the international consensus settled on. Because it comes from the average alone, it carries no information the average did not already carry; it exists so that a clinician can read it on the scale they are used to.
• Coefficient of variation. How spread out your readings are, measured against your own average rather than against anyone else’s. It is the page’s answer to “how bumpy”, as distinct from “how high”.

Time in Ranges

Underneath the strip is a single stacked bar, with a legend beside it. Your readings are sorted into five bands, from very low at the foot of the bar up to very high at the top, with the in-range band in the middle; each band gets its share of all your readings, printed as a percentage on its legend row, and its own cut-offs in mg/dL printed beside the name. The count of readings the whole bar was built from sits underneath it, which is the thing to look at first: a bar built from a handful of readings is a bar about a handful of readings.

What the bar does is divide up your readings. What it does not do is say anything about how they ought to be divided, and you will not find that anywhere on the report either. The bands are the international consensus bands, printed with their cut-offs so that the person reading the report over your shoulder can see exactly what was counted where.

The curve

This is the part that gives the report its name, and it is the part people misread first. Every day in the range is laid over a single twenty-four-hour day in your own local time. So a bump at breakfast is not one morning: it is every morning in the range, stacked on top of each other and summarised.

Reading outward from the middle:

• The dark line is the median. At each time of day, half of your readings from that slot sat above it and half below. It is not an average day and it is not any particular day; there may well be no day in the range that looked like it.
• The inner band is the middle half of your readings at that time of day — the quarter above the median and the quarter below it. A narrow waist here says your days resembled each other at that hour; a wide one says they did not.
• The outer band is wider still, taking in almost all of the readings at that time of day and clipping only the extreme tail at each end, so that one unusual morning cannot stretch the whole picture.
• The shaded strip behind everything marks the in-range band, and the two lines drawn across the plot are its edges. It is there so the eye can see where the bands sit relative to it, not as a score.

Two things about the chart are worth knowing, because both are the chart telling you when to lean on it less. Where a time of day has readings with no neighbouring slot to join up to, the report draws a dot with a whisker there instead of a band: a lone reading gets shown as a lone reading rather than smoothed into a curve. And when there are too few populated slots for the bands to mean much, the chart says so in a line under the legend, so the thinness is stated rather than left for you to notice. If you take fingersticks a few times a day rather than wearing a sensor, that line is the one you will see, and it is worth reading before the picture above it.

Along the bottom is time of day in your local zone. Up the side is glucose in mg/dL. Readings that arrived from an import already pinned to the top of the sensor’s reporting range are drawn at the top of the plot, so a flat run along the ceiling is worth reading as “at or above this” rather than as a precise height.

Page two: what you were doing

This is the page a sensor report does not come with, and it is the reason the whole thing is worth handing over rather than just the first page.

It opens with a grid: one row for every local day in the range, including the days you logged nothing, with the hours running across. Meals, medication, exercise and hydration each get their own letter, placed at the hour you logged it, so a run of evenings or a gap of a fortnight is visible as a shape rather than as a list. The empty rows are deliberate — a day you did not log is information, and hiding it would make the record look denser than it was. Over a long range the rows would get too thin to read, so the grid shows the most recent days and prints how many of the range it had to leave out.

Below the grid is a logging adherence table: for each of the four kinds of entry, how many days in the range carried at least one of them, and what share of the days that is. It is a description of your logging, not of you.

At the foot of the page is a panel of notes, and it says four things: the date range and the timezone the times are in, when the report was generated, that everything on that page is self-reported and not independently verified and that your glucose may combine imported readings with ones you entered by hand, and that the report is a summary of consumer-logged data and is not a diagnostic device. We would rather that paragraph were printed on the report than left for someone to assume.

Handing it over

Print it or send the file; it is two pages either way, with the range and the generation time on both. A clinician who works with glucose data will recognise page one immediately, which is the entire reason we draw it to the standard layout rather than inventing a prettier one.

The useful preparation is not learning to interpret it. It is choosing the range and bringing your question. A fortnight around a change you made, or the stretch since you last spoke to them, gives the conversation an edge to work from; a year of everything gives it none. And page two is where your question usually lives, because it is the page that says what you were doing while page one was being drawn.

What this guide will not do

This guide describes the report. It does not tell you what your own bands ought to look like, and neither does the report.

Deciding whether a pattern on it matters is clinical work, and it belongs to a clinician who knows your history and your medication. A percentage that would be unremarkable for one person is worth a conversation for another, and a website cannot tell which of those you are. If something on the report worries you, take the report to them rather than to a search engine.

What we will do is draw the picture to a layout they already read, show the arithmetic behind every figure on it, and print on the page itself how much data each one is standing on.