Résumé
Objectives: Develop a coding system to extract EHR data and establish research validity to lessen need for manual data extraction Methods: As part of a data collection project for COVID + patients requiring ICU care, we established data elements able to be extracted from the Epic electronic health record (EHR). Collaboration between Information Technology (IT), research and clinical personnel established where data elements were located within the EHR and what data could be extracted with minimal manual assistance and uploaded to a research database. Coding was developed using Structured Query Language (SQL) with best practices (includes indexes, execution plans, optimized range keys, avoiding large reads inside read-write transactions as instructed by the Epic consultant). Accuracy of extracted data was evaluated by manual validation of data against Epic records via random selection of patient data within the cohort. Result(s): From July-December 2022, coding was developed which extracted over 130 fields of data from 3093 COVID patients across 5 INOVA ICU sites (demographic, physiologic, lab, interventions, outcome). Prior efforts at data extraction of these elements from research personnel (ZS) who previously performed this task noted an average of 4 hours/patient to complete coded fields. Coded data was also noted to be more accurate when accessed by the same personnel to manually extracted fields. Assuming 4 hrs/pt, manual extraction would require 12,372 hours, which equates to over 6 full time human research personnel. Data coding required 446 hours. Coded data extraction can be almost immediate once fields requested are established, decreasing personnel costs and effort significantly. Conclusion(s): Reduction in need for manual data collection using automated coding extraction can reduce costs, personnel time and enhance research efforts. Sharing coding mapping to other EPIC sites or use of similar methods may improve timeliness of ongoing data extraction and will be useful to develop earlywarning and patient-centered care algorithms to improve care.