321 Automated Development of a Spine Registry (Autoregistry) using an Interpretable Surgeon-Written Regular Expression Classifier. (April 2023)

• Record Type: Journal Article
• Authors: Cheung, Alexander; Laufer, Ilya; Pacione, Donato R.; Lau, Darryl; Frempong-Boadu, Anthony; Kondziolka, Douglas S.; Golfinos, John G.; Oermann, Eric
• Is Part Of: Neurosurgery. Volume 69(2023)Supplement 1
• Page Start: 46
• Page End: 47
• Publication Date: 2023-04
• Subjects: Nervous system -- Surgery -- Periodicals; 617.48005
• DOI: 10.1227/neu.0000000000002375_321
• Languages: English
• ISSNs: 0148-396X

Abstract

Abstract : INTRODUCTION: Clinical registries are critical for modern surgery and underpin outcomes research, reimbursement negotiations, device monitoring, and trial development. Existing approaches to registry construction are labor intensive, costly, and prone to user error. Natural language processing (NLP) techniques combined with broad electronic health record (EHR) datasets can theoretically automate the building and maintenance of registries. METHODS: We utilized a Hadoop datalake consisting of the EHR, billing systems, imaging systems, and other data representing all information generated by an AMC. Using this database and SQL queries, we retrieved every operative note written in the department of neurosurgery over the last decade. Operative notes were parsed using regular expressions constructed by the operative surgeons themselves. A random subset of 100 notes was selected for manual review and analysis. RESULTS: 31, 139 operative cases were identified and downloaded. An operative note processor was written by the operating surgeons and required three days of development time with approximately two weeks of validation. In 100 random cases, the regex classifier was on average 98.6% accurate at identifying whether a particular procedure was performed and correctly identified the entire list of defined surgical procedures in 88% of patients. Most errors (64.2%) occurred during the identification of the involved vertebral levels. Operative notes from ancillary specialties for assistive approaches or closures also led to false-positive results. CONCLUSIONS: This study demonstrates the feasibility of automatically generating a spine surgery registry ("autoregistry") using a Hadoop EHR datalake combined with an interpretable and customizable surgeon-written natural language processing algorithm, which may reduce labor costs and other pitfalls associated with manual registry development.