Mark Mai, MD, MHS
Fellow in Pediatric Critical Care
Childrens Hospital of Philadelphia
Philadelphia, Pennsylvania
Disclosure information not submitted.
Sharon Irving, CRNP, MSN, PhD, PNP
Associate Professor
University of Pennsylvania
Philadelphia, Pennsylvania
Disclosure information not submitted.
Nadir Yehya, MD, MSCE
Children's Hospital of Philadelphia
Cherry Hill, NJ
Disclosure information not submitted.
Vijay Srinivasan, MD, MBBS,FCCM
Asst Professor, Attending Physician
The Children's Hospital of Philadelphia
Philadelphia, Pennsylvania
Disclosure information not submitted.
Title: Feasibility of Using EHR-based Extraction of Nutrition Data in Critically Ill Children
INTRODUCTION/HYPOTHESIS: Current approaches to clinical decision support and research studies of nutrition therapy rely on prospective nutrition surveys or retrospective manual chart review with numerous limitations. We hypothesized that an automated approach to extract nutrition-specific data from the electronic health record (EHR) is feasible and sought to validate an EHR-based algorithm in comparison to traditional methods.
Methods: We compared the EHR-based algorithm against two different retrospective cohorts of patients in a 65-bed PICU. The first consisted of 91 patients (6/2014-7/2016) prescribed a clinical pathway for enteral nutrition, in whom demographic and nutrition information had been previously recorded by clinicians for quality improvement. The second consisted of a random selection of 60 patients (11/2011-11/2019) from a registry of patients with pediatric acute respiratory distress syndrome (PARDS), whose nutrition information had been collected via chart review by trained research assistants. We identified select enteral and parenteral nutrition variables from a query of EHR elements (e.g. delivered TPN volume, delivered feed volume, route of delivery) and summarized these per 24-hour period. We report correlation using the Pearson correlation coefficient.
Results: For the clinical pathway cohort, 273 patient days of enteral nutrition were studied. Comparing the EHR-based approach against manual entry, there was modest overall agreement, with improved correlation each subsequent day (Total R = 0.61; Day 1 R = 0.48, Day 2 R = 0.91, Day 3 R = 0.94). The 10 most discordant subjects by absolute difference demonstrated strong agreement with the EHR-based approach, upon repeated manual review (R = 0.99). For the PARDS cohort, 1440 patient days of enteral and parenteral nutrition were studied. In comparison to chart review by a trained research assistant, the EHR-based approach demonstrated extremely strong agreement across all days (R = 0.94-0.99).
Conclusions: An automated approach to nutrition specific data extraction from the electronic health record (EHR) is feasible and may limit errors related to manual chart review. This pipeline can be used to inform clinical decision support and improve research investigations to optimize the provision of nutrition therapy in critically ill children.