Breanne S Baker1, Jinpu Li2, Emily V Leary2. 1. Thompson Laboratory for Regenerative Orthopedics, Department of Orthopedic Surgery, Missouri Orthopedic Institute, University of Missouri, Columbia, MO, USA. Electronic address: bakerbre@health.missouri.edu. 2. Thompson Laboratory for Regenerative Orthopedics, Department of Orthopedic Surgery, Missouri Orthopedic Institute, University of Missouri, Columbia, MO, USA.
Abstract
INTRODUCTION/ BACKGROUND: Since the 1970s imaging devices have been rapidly advancing to provide researchers, clinicians, radiologists, and practitioners greater capabilities to image bone. Today, dual-energy X-ray absorptiometry (DXA) has emerged as the gold standard for the assessment of bone and body composition. Despite recent improvements in DXA image quality and reduced radiation dosages and scan times, the data extraction process has yet to be streamlined. The objective of this study was to design an application to allow DXA users to quickly and accurately transfer DXA report data directly into a user-friendly format. METHODOLOGY: Initial development of the DXA Data Xtraction Assistant (DXA2) included processing 485 DXA reports from a previous study. Using Python script, numeric data from these DXA reports were converted to .csv files and then merged into one file, effectively compiling and organizing all table data from each report, by participant and visit number. A graphical user interface was then developed and beta tested using multiple DXA manufacturers, models, and software versions. To compare the effectiveness of the DXA2 to the current standard of manual extraction, all 485 reports were manually transcribed into Microsoft Excel. The time required per report and the error rates were compared. RESULTS: Manual transcription, by an experienced researcher, took over 10 h with an error rate of 0.6% and average error magnitude of 51%. Data extraction from DXA2 took less than 25 min to retrospectively reformat all DXA reports for the DXA2 and transcribe all 485 reports with 100% accuracy. CONCLUSION: The DXA2 application automates DXA report data extraction and aggregation for clinical, research, and applied settings and drastically reduces manual data entry time and eliminates transcription errors.
INTRODUCTION/ BACKGROUND: Since the 1970s imaging devices have been rapidly advancing to provide researchers, clinicians, radiologists, and practitioners greater capabilities to image bone. Today, dual-energy X-ray absorptiometry (DXA) has emerged as the gold standard for the assessment of bone and body composition. Despite recent improvements in DXA image quality and reduced radiation dosages and scan times, the data extraction process has yet to be streamlined. The objective of this study was to design an application to allow DXA users to quickly and accurately transfer DXA report data directly into a user-friendly format. METHODOLOGY: Initial development of the DXA Data Xtraction Assistant (DXA2) included processing 485 DXA reports from a previous study. Using Python script, numeric data from these DXA reports were converted to .csv files and then merged into one file, effectively compiling and organizing all table data from each report, by participant and visit number. A graphical user interface was then developed and beta tested using multiple DXA manufacturers, models, and software versions. To compare the effectiveness of the DXA2 to the current standard of manual extraction, all 485 reports were manually transcribed into Microsoft Excel. The time required per report and the error rates were compared. RESULTS: Manual transcription, by an experienced researcher, took over 10 h with an error rate of 0.6% and average error magnitude of 51%. Data extraction from DXA2 took less than 25 min to retrospectively reformat all DXA reports for the DXA2 and transcribe all 485 reports with 100% accuracy. CONCLUSION: The DXA2 application automates DXA report data extraction and aggregation for clinical, research, and applied settings and drastically reduces manual data entry time and eliminates transcription errors.