Digitizing graphs for a meta-analysis
When a paper reports its results only as a figure, the numbers have to come out of the figure. Doing that by hand for fifty papers is days of clicking. This does a paper at a time.
What helps in a review
- Error bars are read at both ends, so a mean and its interval come out together.
- The Excel file has an audit sheet: source file and page, how each chart was read, and every value changed by hand with what it was before.
- Charts in PDFs are read exactly. When a PDF holds the chart as a drawing (most figures in papers do), the values are read from the drawing itself. Nothing is estimated from pixels, so a point is as precise as the file is.
A workable routine
- Upload the paper's PDF.
- Keep the charts you need and check the rings on each.
- Download the workbook and file it with the paper: the audit sheet is your record of how the numbers were obtained.
Be careful with
Small, crowded figures are the hard case for pictures: when a plot is only a couple of hundred pixels high, a pixel is already a visible error. Use the PDF when you have it, or a larger screenshot.
Read the audit sheet's status column. A chart read from a picture says so, and flagged values are the ones to look at first.
Questions
Does it read survival curves?
Yes: Kaplan-Meier curves as their steps, and the number-at-risk table printed under them.