SIR 2024
General IR
Ankit S. Baghel (he/him/his)
Medical Student
Stanford University
Financial relationships: Full list of relationships is listed on the CME information page.
Lawrence Hofmann, MD, FSIR
Division Chief Interventional Radiology
Stanford University Medical Center
Financial relationships: Full list of relationships is listed on the CME information page.
Alexander Vezeridis, MD, PhD
Assistant Professor, Interventional Radiology
Stanford University Medical Center
Disclosure information not submitted.
Since 2015, our IR division has phased in procedure report templates containing predefined fields filled in by procedure personnel. We define three types of fields for our framework: headers that indicate sections of the report (e.g. PROCEDURE COMMENTS AND FINDINGS), module names that indicate parts of the procedure performed (e.g. CATHETERIZED ARTERY), and procedure details (e.g. Antibiotics, New tube size). A retrospective single-center database was constructed consisting of 42,705 procedure reports from 2015 through mid-2022 with increasing usage of templates over time. Manual inspection of a subset of reports generated a list of over 2400 field names across all three types. Using this list, the parser automatically generates regular expressions for rule-based information extraction. Furthermore, by leveraging hierarchical information regarding the three types of fields, the parser represents procedures as a series of consecutive modules, each with their own corresponding details describing approach, equipment, and findings.
Results:
With adoption of semi-structured templates over time, our parser could extract more pertinent information regarding medications, radiation doses, anesthesia, procedure personnel, and procedure/module fields. In 2015, only medication information could be automatically extracted from more than 30% of reports. However, by 2017, 70% of reports had automatically extractable information for across all categories of interest, and from 2018 onward, over 90% of reports had easily parsable information.
Conclusion:
The integration of semi-structured reporting in our practice has proven invaluable for enhancing data extraction and analytics, streamlining retrospective analysis. As we transition toward a learning healthcare system, automated data extraction tools like this parser will be crucial in harnessing procedural data for continuous quality improvement.