Moderated Poster

Poster, Podium & Video Sessions

MP96-10: Initial validation of automated data extraction methods in urologic oncology practice

Tuesday, May 16
9:30 AM - 11:30 AM
Location: BCEC: Room 153

Presentation Authors: Renu Eapen*, Samuel Washington III, Annika Herlemann, David Tat, Mark Bridge, Niloufar Ameli, Janet Cowan, Frank Stauf, Peter Carroll, Matthew Cooperberg, San Francisco, CA

Introduction: The Urological Outcomes Data Base (UODB) has existed for 15 years and contains data on over 6,000 patients treated for prostate cancer at University of California, San Francisco (UCSF). Until recently, clinical data in UODB have been manually abstracted from patient records. We are now implementing automated data extraction from the EPIC electronic health record system. EPIC is supported by a research database that automatically extracts patient data. We aim to study a set of chosen variables and compare the types and degrees of miscorrelation between automated and manual data extraction, to see if manual data extraction can be minimized or eliminated.

Methods: In early 2016 we developed a set of Smart Data Elements (SDEs) for urologic oncology, including SDEs for men with prostate cancer. These SDEs are populated automatically from the EPIC clinician interface during routine clinical documentation, using either SmartForms or SmartLists embedded within a dozen new standardized templates. SDEs are available immediately in EPIC's Clarity database, and can be populated in future documentation notes. We selected 15 core sample SDEs for validation against manually abstracted data in UODB for patients seen in 2016. Manually abstracted values were compared directly to SDEs values to assess match frequency.

Results: The 15 SDEs encompassed a wide range of variables from diagnosis to pathologic staging, including clinical risk characteristics at diagnosis, biopsy Gleason score and surgical pathology findings. The median number of patients per variable was 37 (IQR 17-39). Median number of matches per variable was was 14% (IQR 5-37) with median match rate of 70.6% (IQR 35.7-97.4%). Detailed match rates are shown in the table.

Conclusions: Next steps include expanding validation rules across a larger set of variables and exploring the limitations of the match strategy. In some cases, data sources such as a computerized system may prove more accurate than manual entry. Working with the AUA Quality (AQUA) registry, we plan to transfer subsets of SDEs to the EPIC Foundation repository, allowing access to any EPIC center. Automated data extraction can improve clinical workflow and streamline data collection within urologic oncology.

Source Of Funding: None.

Send Email for Renu Eapen


Assets

MP96-10: Initial validation of automated data extraction methods in urologic oncology practice



Attendees who have favorited this

Send Email for Initial validation of automated data extraction methods in urologic oncology practice