Background: Sepsis is the leading cause of death in US hospitals, and prompt recognition and treatment is critical. To improve general floor sepsis care, our hospital created a “sepsis squashing squadron” (S3) – a response team that uses scoring tools displayed in the EHR to improve efficiency, sensitivity, and specificity of sepsis identification. We previously identified workflows to improve timeliness of sepsis detection on a single hospital unit within our large academic medical center. This study is our largest to date and includes a hospital-wide effort to characterize sepsis recognition, treatment, response times, and to identify trends among multiple services and units to inform ongoing improvement efforts.
Methods: Through daily chart review, a S3 team member manually screened patients admitted to non-ICU floors. The automated scores recorded for performance characterization included modified institutional versions of the Modified Early Warning score (jMEWS) and Sequential Organ Failure Assessment scores (mSOFA) and Epic’s “sepsis score” (SS) and “deterioration index” (DI). S3 members used Epic automated score thresholds of jMEWS ≥ 2 and DI ≥ 30 and/or mSOFA ≥ 2 which we have previously identified as optimizing sepsis screening efficiency while maintaining high sensitivity. For patients with sepsis (defined as ≥ 2 SIRS criteria with an order or result suggesting likely and/or definitive infection), we assessed adherence and time to implementation of components of the Severe Sepsis/Septic Shock Early Management Bundle (SEP-1).
Results: From 4/19/23–5/31/23, 2,453 patients were screened across 13 units, diverse services, and 2 hospitals. The daily prevalence of sepsis was 274/2,453. Overall, 163/274 of those patients had a new episode of sepsis. In assessing adherence to the SEP-1 bundle, 99 patients (61%) received antibiotics with a mean order to administration time of 1.7 hours after excluding outliers of zero, negative, or ≥ 12 hours time to administration. This was followed by blood cultures (89/163; 55%; 2.1 hours), lactate draw (61/163; 37%, 0.79 hours), and IV fluid administration (49/163; 30%; 0.54 hours) (Figure 1). Only 38 patients (23%) had all bundle components completed. Of patients with new sepsis, 107/163 were adjudicated for appropriate recognition and treatment, of which 37 cases (34.5%) were deemed appropriate (e.g., not dosing fluids can be clinically appropriate). The area under the curve for jMEWS was 0.79, SS was 0.78, DI was 0.69, and mSOFA was 0.60. Using a specificity minimum of 0.2, the best sensitivity/specificity thresholds for an alert were jMEWS ≥ 6 (0.53/0.25), SS ≥ 4 (0.67/0.21), mSOFA ≥ 8 (0.04/0.4) and DI ≥ 46 (0.42/0.20).
Conclusions: Sepsis prevalence on inpatient, non-ICU floors was 11%. A minority of septic patients received all SEP-1 bundle components or were being treated appropriately, suggesting wanting recognition and treatment adherence. This and the lengthy time from order to action are potential areas for improvement. Further, automated scores to improve sepsis recognition demonstrate impractical sensitivity/specificity performance for clinical-decision support systems (CDSS). Focusing on more sensitive and specific sepsis recognition criteria such as SIRS-based screening may improve sepsis CDSS and care for a majority of patients until better tools are developed to identify edge cases.
