(896) Development of a Multicenter Cloud Database and Linked Biorepository for Pediatric Sepsis Research

Introduction: A major goal in pediatric sepsis research is to identify and characterize sepsis phenotypes and their underlying biology. Collection and analysis of biospecimens, particularly early in presentation, is a cornerstone of this research agenda. Many biorepositories have associated clinical case report forms; however, these often lack granularity, particularly in dynamic, complex syndromes like sepsis. We aimed to test the feasibility of developing a multicenter, scalable platform to collect granular electronic health record (EHR) data linked to biospecimens.

Methods: We performed a prospective two-center pilot study. All patients who triggered sepsis pathway care in the emergency departments between 7/1/22-12/31/22 were included. Files with EHR-derived data were extracted using standardized queries and transferred to a cloud database. A single-site subset of patients were enrolled in the biorepository aim of the study. Biospecimen metadata were linked to subject identifiers locally prior to data transfer. Data were harmonized and loaded into a common data model (CDM) based on the Observational Medical Outcomes Partnership (OMOP) standards, and underwent a structured data quality assessment (QA) based on measures of completeness, conformance, and plausibility.

Results: There were 665 patient encounters included (294 and 371) in both sites, of which 13 had blood collected as part of the biorepository aim. During the data QA and harmonization, 503,371 unique values corresponding to 186 clinical measurements (observations, interventions, tests, etc.) were mapped to 76 LOINC codes and 47 unique clinical concepts used to determine organ dysfunction and clinical outcomes. Metadata corresponding to the collected biospecimens were harmonized into a Specimen table and linked to the clinical data in the OMOP CDM. The structured data QA report identified 1 missing visit identifier, 4 visits with missing blood pressures in one site, and 2 variables with 1 or more non-plausible values in each site.

Conclusions: We demonstrated the feasibility of developing a scalable and generalizable cloud-based platform to collect, harmonize, and perform QA on granular EHR data linked to biorepository metadata across multiple sites. In future work we will expand the number of sites and scope of biospecimen collection efforts.