Sepsis and septic shock case identification from electronic health records: an open-source workflow and comparison of...

Deploying clinical criteria to identify patients with sepsis from the electronic health record is challenging. We sought to build an open-source data workflow for sepsis and septic shock case identification using three distinct criteria and assess the identified cohorts.

We developed a workflow from raw data queries through case identification using the Centers for Disease Control and Prevention Adult Sepsis Event (CDC ASE), Sepsis-3, and International Classification of Diseases criteria. The workflow was then applied to patient encounters between 2012 and 2024 across 25 emergency departments and hospitals in an integrated health system. Separate cohorts of sepsis and septic shock patients by criteria were derived, restricted to a patient's first sepsis encounter, and compared. Pairwise agreement was evaluated with Cohen's kappa (κ) and Jaccard index.

Our programming code for the complete workflow is available on a public platform. We identified 302,112 patients with sepsis and 59,043 with septic shock who met one or more of the three criteria. Multiple criteria were satisfied for only 35% of patients meeting at least one sepsis criterion and 45% of patients meeting at least one septic shock criterion. Pairwise agreement was the highest between CDC ASE and Sepsis-3 criteria for both sepsis (κ = 0.39, Jaccard index = 0.31) and septic shock (κ = 0.53, Jaccard index = 0.37). Characteristics and outcomes of patients differed across cohorts identified with the three criteria. Patients meeting both CDC ASE and Sepsis-3 criteria for sepsis and septic shock were more severely ill and had high in-hospital mortality incidence.

Our open-source workflow is useful for individuals seeking to build and analyze sepsis and septic shock cohorts. Agreement between case criteria was fair to moderate. Clinicians, researchers, and quality improvement practitioners who identify patient cohorts with sepsis and septic shock retrospectively should consider applying multiple criteria.