Analysis of high-level electronic operative report documentation structure

Abstract

Operative report documentation (ORD) is a fundamental part of surgical practice that has a direct and significant role in quality, medico-legal and billing, and other secondary uses of operative reports. Traditionally, operative reports are created by the primary surgeon through dictation and subsequent transcription of details of the operative procedure after its completion. With the advent of widespread electronic health record (EHR) adoption, there is a potential opportunity to improve aspects of ORD through synoptic reporting and templating of operative reports, as well as to leverage operative report sections and high-level phases described with surgical procedures in operative report. This research seeks to illuminate knowledge about high-level ORD structure including an assessment of attitudes and opinions towards ORD in surgical education, evaluation of structured operative report section names, and appraisal of high-level phases of surgical descriptions in operative reports with the application of automated methods to classify these phases. An electronic survey about ORD teaching and practices was sent to surgical program directors. While most program directors responding to the survey consider ORD teaching a priority, few provide ORD instruction, and significant barriers were perceived by program directors for ORD instruction and ORD using synoptic reporting. To evaluate operative report sections, the HL7 Implementation Guide for Clinical Document Architecture Release 2.0 Operative Note Draft Standard for Trial Use (HL7-ON DSTU) Release 1 and Logical Observation Identifiers Names and Codes (LOINC®) structured sections were evaluated on 384 unique section headers from 362,311 operative reports. HL7-ON DSTU alone and HL7-ON DSTU with LOINC® section headers covered 66% and 79% of sections headers (93% and 98% of header instances), respectively. Section headers contained large numbers of synonyms, formatting and word form variation, as well as coverage gaps in the current terminology sources. In a third study, high-level phases of the "Surgery Description" from operative reports were identified by surgeons. Automatic classification with support vector machines using topic analysis and information gain for feature selection was then used to identify these high-level phases, followed by assessment of subtopics for phases. Five high-level "Surgery Description" section phases were identified: Preliminaries, Getting Started, Main Part, Closure, and Epilogue, as well as need for Observations to describe findings, events, and other important clinical information. Automated classification of high-level sections performed well on a set of laparoscopic cholecystectomy and random surgery operative reports and was associated with the identification of meaningful subtopics. Overall, this research demonstrates the variability in ORD practices for surgeons nationally and individually with respect to section structure, as well as the value of high-level phases to group content in these important clinical documents. Future work will seek to leverage our understanding of ORD structure to improve information extraction and natural language processing (NLP) techniques for secondary use of operative reports.