Introduction: Multicenter electronic health record (EHR) databases can facilitate target trial emulation. We evaluated the accuracy of linking a single-center quality improvement (QI) cohort with similar eligibility criteria as a pediatric sepsis randomized trial to a large pediatric EHR database as proof of principle for future multicenter cohort development to emulate this randomized trial. Methods: Patients aged 2 months – 17 years in a local QI cohort who had confirmed collection of blood culture and receipt of IV fluid bolus and broad-spectrum antibiotic, hereafter called “sepsis treatment”, in the emergency department (ED) followed by inpatient admission between December 2021 and December 2023 (n=1061) were linked to records in the PEDSnet EHR database. Test characteristics were calculated to determine the accuracy of identification of these encounters within all ED encounters resulting in an inpatient admission at this center in PEDSnet in the same time frame (n=26,611). Participants in the PRagMatic Pediatric Trial of Balanced vs. nOrmaL Saline FlUid in Sepsis (PRoMPT BOLUS) interventional trial at this center (n=286) were also linked to PEDSnet records for additional evaluation of the linkage method. Results: 1061 patients in the QI cohort received sepsis treatment. After initial linkage, evaluation of discrepancies, and refinement of the code defining PEDSnet encounters with sepsis treatment, 1000 (94%) were matched in PEDSnet. Sixty-three were not matched in PEDSnet and an additional 318 patients in PEDSnet received sepsis treatment but were not in the QI cohort. Sensitivity was 94%, specificity 99%, positive predictive value 76%, and negative predictive value >99%, for an accuracy of 99%. Of 286 enrolled trial participants, 260 (91%) were matched in PEDSnet and met the sepsis treatment criteria for target trial emulation cohort inclusion. Conclusions: Linkage of PEDSnet records with a single-center QI cohort meeting sepsis treatment criteria similar to target trial eligibility criteria showed high accuracy. Evaluation of false positives and negatives improved coding for defining the cohort. Use of these cohort definitions in the multicenter PEDSnet database is anticipated to generate a large cohort emulating the trial population, facilitating accurate real-world target trial emulation analyses.
Fitzgerald et al. (Sun,) studied this question.