Discharge diagnoses versus medical record review in the identification of community-acquired sepsis

INTRODUCTION: We evaluated the accuracy of hospital discharge diagnoses in the identification of community-acquired sepsis and severe sepsis. METHODS: We reviewed 379 serious infection hospitalizations from 2003 to 2012 from the national population-based reasons for geographic and racial differences in stroke (REGARDS) cohort. Through manual review of medical records, we defined criterion-standard community-acquired sepsis events as the presence of a serious infection on hospital presentation with ≥2 systemic inflammatory response syndrome criteria. We also defined criterion-standard community-acquired severe sepsis events as sepsis with >1 sequential organ failure assessment organ dysfunction. For the same hospitalizations, we identified sepsis and severe sepsis events indicated by Martin et al. and Angus et al. International Classifications of Diseases 9th edition discharge diagnoses. We evaluated the diagnostic accuracy of the Martin and Angus criteria for detecting criterion-standard community-acquired sepsis and severe sepsis events. RESULTS: Among the 379 hospitalizations, there were 156 community-acquired sepsis and 122 community-acquired severe sepsis events. Discharge diagnoses identified 55 Martin-sepsis and 89 Angus-severe sepsis events. The accuracy of Martin-sepsis criteria for detecting community-acquired sepsis were: sensitivity 27.6%; specificity 94.6%; positive predictive value (PPV) 78.2%; negative predictive value (NPV) 65.1%. The accuracy of the Angus-severe sepsis criteria for detecting community-acquired severe sepsis were: sensitivity 42.6%; specificity 86.0%; PPV 58.4%; NPV 75.9%. Mortality was higher for Martin-sepsis than community-acquired sepsis (25.5% versus 10.3%, P = 0.006), as well as for Angus-severe sepsis than community-acquired severe sepsis (25.5 versus 11.5%, P = 0.002). Other baseline characteristics were similar between sepsis groups. CONCLUSIONS: Hospital discharge diagnoses show good specificity but poor sensitivity for detecting community-acquired sepsis and severe sepsis. While sharing similar baseline subject characteristics as cases identified by hospital record review, discharge diagnoses selected for higher mortality sepsis and severe sepsis cohorts. The epidemiology of a sepsis population may vary with the methods used for sepsis event identification. ELECTRONIC SUPPLEMENTARY MATERIAL: The online version of this article (doi:10.1186/s13054-015-0771-6) contains supplementary material, which is available to authorized users.