Advancing Sepsis ID: Open-Source EHR Workflow Insights for EDs

An open-source EHR workflow targets timely, accurate sepsis identification in emergency departments to standardize screening and cohort definition.

In a retrospective, multisite validation across multiple EDs and tens of thousands of encounters, authors compared the workflow's mappings to chart-adjudicated sepsis definitions. Discrimination fell in the 'good' range (AUROC ~0.8–0.9), with sensitivity–specificity tradeoffs aligned to surveillance versus clinical screening.

This approach departs from manual chart review and basic administrative surveillance by mapping structured EHR elements and applying deterministic rules across encounters. The workflow operationalizes multiple clinical definitions in parallel—CDC ASE, Sepsis-3, and ICD-based criteria—so implementers can compare yields and performance rather than rely on a single post hoc method. That reduces ambiguity when generating retrospective cohorts for analysis and benchmarking and is directly relevant to ED screening efforts that must balance sensitivity, specificity, and reproducibility. Inputs include vital signs, laboratory results, medication orders (antimicrobials and vasopressors), and coded fields such as ICD entries.

Algorithmic steps comprise defined trigger windows around culture and antimicrobial timing, sequential application of organ-dysfunction scoring, and logical combination of suspected infection and dysfunction signals; outputs are encounter-level sepsis flags and probable septic shock classifications. The workflow maps CDC ASE, Sepsis-3, and ICD definitions to EHR fields (for example, SOFA components from labs and physiologic measures and qSOFA from bedside vitals), and CDC ASE surveillance elements combine coded infection indicators with evidence of organ dysfunction within defined time epochs.

In the study cohort, operationalized definitions produced distinct yields and overlap: CDC ASE and ICD-based criteria captured broader cohorts, while Sepsis-3 identified a smaller, higher-severity subset. The paper provides concordance and performance metrics (including discrimination and sensitivity–specificity tradeoffs) for each mapping so teams can compare yield and predictive performance across definitions.

Noted implementation barriers include variable data completeness and timing (laboratory latency and charting delays), heterogeneity in EHR field mappings across vendors and sites, and the need for local governance to decide which criteria to operationalize. Practical mitigations are straightforward: validate local mappings against chart samples, set trigger windows to reflect typical ED timelines, and tune alert thresholds using retrospective cohorts to estimate yield and false-trigger rates.

Key Takeaways:

• An EHR-native workflow that standardizes multiple definitions enables consistent cohort generation and reduces methodologic variability, improving reproducibility for quality and research initiatives.
• Cohort composition and operational burden shift by definition: CDC ASE and ICD-based criteria increase screening and triage volume, while Sepsis-3 concentrates higher-severity cases—so definition choice must match the intended use case (surveillance, clinical screening, research).
• Local health systems should validate mappings, select definitions aligned to operational goals, and pilot alerts with workload monitoring to quantify yield, time-to-antibiotics, and unintended consequences before broad deployment.