Data Abstraction Assistant (DAA): a new open-access tool being developed and tested in a randomized controlled trial

Article type
Year
Authors
Saldanha I1, Jap J2, Smith B2, Dickersin K1, Schmid C2, Li T1
1Cochrane United States, and Cochrane Eyes and Vision US Satellite, USA
2Brown University School of Public Health, USA
Abstract
Background: Data abstraction, a critical systematic review step, is time-consuming and has been shown to be prone to errors. Software that can streamline and automate some aspects of the process might be useful.

Objective: To develop an open-access tool, Data Abstraction Assistant (DAA), to help data abstractors locate and mark sources of information in articles during data abstraction.

Methods and Results: Developers at Brown University (JJ, BS and CS) developed DAA using 'Ruby on Rails'. DAA takes advantage of several 'gems', reusable modular pieces of code to accomplish its task. DAA can be used with data systems such as Systematic Review Data Repository (SRDR). Abstractors can build a 'Document Store' by loading PDFs in DAA. Then, the abstractor can log into SRDR, and open any PDF from her/his Document Store. Multiple PDFs can be associated with the same form. The abstractor can view a PDF and the form simultaneously in a split screen (Figure; live demonstration possible). The abstractor also can switch between PDFs saved in the Document Store. The abstractor can flag any text/figure/table/box in the PDF to pin the location where relevant information resides, and can drag and drop text from a PDF into a text field in SRDR. Once the first abstractor completes data abstraction, a second abstractor can click on those flags, which navigates directly to the location where the pins are dropped, thus facilitating data adjudication. In March 2016, we began enrollment for a randomized cross-over trial to evaluate the comparative accuracy and efficiency of DAA-facilitated single data abstraction plus verification, traditional single data abstraction plus verification, and traditional dual independent data abstraction. The expected sample size is 24 pairs of abstractors, who will be randomized to abstract data from six articles, two under each approach. The Patient-Centered Outcomes Research Institute in the USA funds the DAA development and the trial (PI: TL).

Conclusions: We have developed DAA as a software tool to help improve the efficiency of data abstraction without comprising accuracy. We are conducting a randomized trial to empirically evaluate DAA.