Preparing KM coordinates

What you need from the publication

Before you start, find these in the paper or its supplement:

• The Kaplan–Meier figure for the arm you want to model, as a clear image (vector PDF or a high-resolution screenshot).
• The numbers-at-risk table printed beneath the curve, with the time points it reports.
• The total sample size (N) for the arm.
• The total number of events, if reported. This is optional but improves the reconstruction by constraining inferred censoring.
• The time unit (months or years), so your coordinates and at-risk interval are consistent.

Digitizing the curve

Digitizing turns a picture of the curve into numeric coordinates. Any plot digitizer works; a common free choice is WebPlotDigitizer. Whichever tool you use, the steps are the same:

1. Calibrate the axes. Mark two known points on each axis (for example survival 0 and 1 on the y-axis, and two time labels on the x-axis) so the tool can map pixels to values.
2. Capture the origin. Make sure you have a point at time = 0, survival = 1; survival curves start at 1.
3. Sample enough points, especially around steps. Place a point just before and just after each visible drop so the step pattern is preserved. Smooth stretches need only a few points.
4. Export to CSV with time and survival columns.

CSV format

The upload needs a plain CSV with just two numeric columns, time and survival (survival as a probability between 0 and 1). Headers are auto-detected, so column order and extra columns don’t matter.

• Save as a plain .csv (comma, semicolon, tab, or pipe separated), not an .xlsx spreadsheet.
• Survival must be a probability between 0 and 1. If your tool exported percentages (for example 95 instead of 0.95), divide the column by 100.
• Numbers only: no patient identifiers or text labels.
• The at-risk table does not belong in the CSV; it is entered separately in the app (see below).
• Keep the file under 10 MB and 10,000 rows.

The at-risk table

The numbers at risk are entered in the app during reconstruction, not in the CSV. They tell the Guyot algorithm how many patients remained in follow-up at each reported time, which is what makes patient-level reconstruction possible.

• Read the counts at the interval the publication reports (the time step, often 3, 6, or 12 months).
• Enter one value per grid point: time 0, one step, two steps, and so on up to the max time. The form shows the exact times it expects.
• The first at-risk value should equal (or be below) the total sample size N.
• A single trailing “0 at risk” is tolerated.

Pre-upload checklist

• Survival values are probabilities between 0 and 1, not percentages.
• The origin (time 0, survival 1) is captured.
• There are points just before and after every visible step.
• The at-risk interval matches the time step you will enter.
• The first at-risk value is at or below total N.
• Time units are consistent between the curve and the at-risk table.