Skip to main content

Accuracy of blood pressure data abstraction from graphs

Date and Location


P4. 102


Saturday 21 September 2013 - 10:30 - 12:00


Presenting author and contact person

Presenting author

Sam Soufi

Contact person

Sam Soufi
Abstract text
Background: Assessing hourly blood pressure (BP) measurements from graphs presents a challenge in making reliable measurements. The Hypertension Review Group has undertaken a series of time-course reviews to assess the temporal BP lowering efficacy of various antihypertensive medications. The majority of included studies represented 24-hour BP data only in the form of a graph. This would involve manually printing the graph and taking measurements with a ruler, which differed between two reviewers. Objectives: To develop a new electronic method for the reliable measurement of BP data from graphs. Methods: We developed a simple method of data extraction whereby the image of the graph from the PDF file of the study is captured via the “Snapshot” tool in Adobe Acrobat (Figure 1). The image is subsequently opened in Microsoft Paint, having enabled the “Rulers” and “Gridlines” options under the “View” tab. This overlays a scaled grid on the image to take a direct measurement of the difference of BP and compare it to the scale that appears on the BP axis (Figure 2). The Line Tool is used to draw horizontal lines from the data points to the graph axis, consistently choosing either the top or the bottom of the line itself as the marker and counting all the grid squares that appear in between the lines to measure the BP difference from baseline (Figure 3). The formula for this scale can then be inputted into a Microsoft Excel spreadsheet to systematically tabulate the finalized BP difference measurements. Results: We anticipate that this new method will save time and improve the reliability of measurements of data taken from graphs. Two reviewers extracted identical data using this method (Figure 3). Conclusions: This process may be a useful tool for Cochrane reviewers when extracting graphical data from included studies.