Extracting data from figures with software was faster, with higher interrater reliability than manual extraction.

Area of Science:

• Biomedical Informatics
• Data Science
• Scientific Publishing

Background:

• Extracting data from graphical representations in scientific literature is crucial for meta-analyses and systematic reviews.
• Manual estimation of data points from figures can be time-consuming and prone to inaccuracies.
• The advent of open-source software offers potential for more efficient and accurate data extraction.

Purpose of the Study:

• To compare the speed and accuracy of graphical data extraction using manual estimation versus open-source software.
• To evaluate the impact of software-assisted extraction on interrater reliability.

Main Methods:

• Data points were independently extracted from graphs in randomized controlled trials (RCTs) published between 2009 and 2014 using both manual estimation and the Plot Digitizer software.
• Corresponding authors were contacted for exact data values to serve as the ground truth.
• Accuracy was assessed by comparing extracted data against the source data.

Main Results:

• Software-assisted data extraction was significantly faster, reducing extraction time by 47%.
• Percent agreement between raters was slightly higher with software (53.5%) compared to manual methods (51%).
• Agreement between extracted data and original source data was higher using software (75% and 73%) than manual estimation (66% and 69%) for both raters.

Conclusions:

• Open-source software is recommended for graphical data extraction due to its speed and improved accuracy.
• Manual estimation from figures should be minimized to enhance efficiency and reliability in data extraction.
• Software-based data extraction enhances interrater reliability and is a more efficient approach for data presented in figures.