732 - Evaluation of automated data extraction from electronic medical records

Friday, April 28, 2023

5:15 PM – 7:15 PM ET

Poster Number: 732
Publication Number: 732.154

Christina Fang, University of California, Irvine, School of Medicine, Irvine, CA, United States; Mohammed Aftab, CHOC Children's Hospital of Orange County, Anaheim, CA, United States; Riya Bansal, UCI School of Medicine, San Ramon, CA, United States; Nikita R. Shirsat, University of California, Irvine, School of Medicine, Sunnyvale, CA, United States; Angela Pham, UC San Diego, San Diego, CA, United States; Aashna Patel, University of California, Irvine, School of Medicine, Anaheim, CA, United States; Caryn Bradley, University of California, Irvine, Orange, CA, United States

Presenting Author(s)

Christina Fang, BA (she/her/hers)

Medical Student
University of California, Irvine, School of Medicine
Sacramento, California, United States

Background: For infants admitted to a Neonatal Intensive Care Unit (NICU), successful oral feeding is a requirement for discharge home, but preterm infants often struggle with oral feeding skills largely due to neurodevelopmental immaturities. Multiple health factors affect a neonate’s length of stay/days to discharge from the NICU. Due to the breadth of data to be collected to identify correlated factors that impact clinical outcomes, creating an automated program for data extraction will improve the efficiency, objectivity, and accuracy of data collection that is not feasible with manual chart review. An automated data collection program will serve as a valuable resource for physicians to improve health outcomes of infants in the NICU.

Objective:

The purpose of this study was to create a data collection program that extracts NICU patient data from medical charts on Epic, and pipelines the extracted data into a clinical algorithm that will eventually be used to predict neonatal oral feeding outcomes and days to discharge from the NICU.
This retrospective study evaluated an automated process of data extraction from electronic medical records (EMR) to identify maternal factors and infant demographics potentially correlated with infant measures of feeding performance.

Design/Methods: A comprehensive literature search was utilized through multiple search engines, such as PubMed, Google Scholar to identify maternal and infant variables of interest. A clinical data collection form and a corresponding standardized protocol for manual data collection were created based on results from the literature search. Chart reviews were manually conducted on 30 mother/neonate dyads utilizing the EMR system EPIC. An automated data extraction pipeline was created through the UC Irvine Information Services to pull data based on the clinical data collection form. Data from the manual chart reviews was compared to data pulled from the automated extraction to identify the variables that were readily extractible.

Results: Table 1 represents data that was captured manually with chart review vs automatically in the extraction process. Variables that were located only by manual chart review were found in NICU admission and/or discharge notes.

Conclusion(s): Creating an automated data extraction tool that identifies and pulls variables of interest for analysis presents a powerful tool to bolster robust analyses of variables predicting neonate readiness to discharge. Identification of current capabilities of the automatic extraction process facilitates expansion of clinical projects utilizing data from chart reviewing.