Background: Early sepsis recognition and treatment can save lives. The Centers for Medicare & Medicaid Services defines appropriate treatment using the Severe Sepsis/Septic Shock Early Management Bundle (SEP-1) components. To help improve sepsis identification, documentation, and organizational compliance with SEP-1, our hospital implemented a “best-practice alert” (BPA) comprised of Epic’s “sepsis score,” but the rates of clinician recognition and documentation of sepsis have not been assessed robustly. Sepsis performance metrics are often based on Vizient data, which is dependent on both clinician documentation and medical coder identification of sepsis. The accuracy of these data, however, and the impact on sepsis metrics has not been explored. Our “Sepsis Squashing Squadron” (S3) sought to assess the sepsis documentation practices and documentation sensitivity of our clinicians and assess the sensitivity and specificity of our Vizient sepsis data.

Methods: Through daily chart review, an S3 team member screened patients on an assigned non-intensive care hospital unit for the presence of sepsis (defined as ≥2 SIRS criteria and an order or result consistent with suspected or confirmed bacterial infection [e.g., broad-spectrum antibiotics, blood culture]). When sepsis was identified, the chart was reviewed for if and how sepsis was documented. Vizient data was then cross referenced for those patients with sepsis to determine sensitivity of coding. A separate sample of patients found in the Vizient sepsis cohort were retrospectively reviewed for presence of sepsis (as defined above) to assess Vizient data specificity.

Results: From 4/1/23 – 5/31/23, 2,453 patients were screened across 13 units, diverse services, and two hospitals. The prevalence of sepsis was 274 (11%) which was inclusive of non-unique patients with an ongoing or recurrent episode. Of these, 118 patients were reviewed and cross-referenced with Vizient data. The word “sepsis” was documented anywhere in the chart on 79/118 (67%) patients. Of these, 42/79 (53%) were coded as sepsis (e.g., found in the Vizient sepsis data) leading to an overall accurate coding sensitivity of 42/118 (36%). Of note, of the 39 patients who met sepsis criteria and who did not have “sepsis” documented, 3 were nonetheless coded as sepsis and found in the Vizient data leading to an overall Vizient sensitivity of 45/118 (38%). Of the 65 patients found in Vizient sepsis data set with hospitalizations from 11/18/22-1/20/23, upon chart review, 64/65 (98%) had sepsis as defined above.

Conclusions: Sepsis documentation from hospital staff is poor, capturing 67% of patients with a restrictive definition of sepsis. Coding data is still worse with a capture rate of 53% for patients with sepsis documented leading to an overall coding sensitivity of 36%. Specificity of Vizient sepsis data was good at 98%. This raises significant questions about reliability and reasonability of using coded sepsis data to assess hospital performance and otherwise inform quality improvement and/or patient safety initiatives. Further interventions are needed to improve sensitivity of sepsis documentation and coding as part of comprehensive sepsis care improvement.