Deriving a patient de-identified clinical research database from an electronic health record system: A single center experience in determining the prognostic value of lactate, C-reactive protein, and procalcitonin in hospitalized patients

Abstract

Objective : In this study, we demonstrate the derivation of a de-identified research database from the electronic health records (EHR) and then use it in determining the prognostic value of biomarkers lactate, C-reactive protein (CRP), and procalcitonin in hospitalized patients. Methods : The database was created through a series of data export, transform, load, and visualization. A database glossary was completed, including 650 data elements per patient encounter without personal identifiers. Data visualization and statistical analysis tools were provided to those utilizing the database. Results : From July 2012 to August 2019, the database contained 240,759 distinct hospital encounters, with 2,682 patients meeting criteria for analysis, age 54.5±18.6 years, lactate 1.9±1.7 mmol/L, CRP 10.7±10.0 µg/mL, procalcitonin 4.0±17.5 ng/mL, and mortality 8.7%. ROC area under the curve for lactate, CRP, and procalcitonin was 0.670, 0.553, and 0.672, respectively. Lactate, CRP, and procalcitonin had odds ratio for mortality of 1.111 (1.037-1.190), 1.015 (0.991-1.031), and 0.999 (0.991-1.007), respectively. Conclusions : Our efforts provide a framework for creating EHR-derived de-identified patient data for clinical research. Our analysis of the prognostic value of lactate, CRP, and procalcitonin showed these biomarkers to be less accurate than expected, highlighting the challenges of using existing data.