640

Abstract

Introduction: While widely used for characterizing sepsis epidemiology, there have been only limited efforts to validate the identification of sepsis through ICD-9 hospital discharge diagnoses. Hypothesis: Hospital discharge diagnoses accurately identify sepsis hospitalizations. Methods: We used multicenter hospital data from the national REasons for Geographic and Racial Differences in Stroke (REGARDS) study, a 30,239 subject population-based cohort. Over a maximum 8-year observation period, we randomly selected hospitalizations for a serious infection. We defined “gold-standard” sepsis events as hospitalization for a serious infection with the presence of?2 SIRS criteria, determined through manual review of admission and Emergency Department records. We defined “gold-standard” severe sepsis as hospitalization for sepsis with?1 organ dysfunction. For each hospitalization, we identified discharge diagnoses for sepsis (ICD9 038, 020, 790.7) or severe sepsis (ICD9 infection + ICD9 organ dysfunction). Using the gold standard definitions, we determined the diagnostic accuracy of discharge diagnoses for sepsis and severe sepsis. Results: We selected medical records for 370 serious infection hospitalizations encompassing 156 gold-standard sepsis and 122 severe sepsis. Discharge diagnoses correctly identified 56 (35.9%) of 156 sepsis; sensitivity 27.6%, specificity 93.9%, PPV 76.8%, NPV 64.0%. Discharge diagnoses correctly identified 88 (72.1%) of 122 severe sepsis; sensitivity 41.8%, specificity 85.1%, PPV 58.0%, NPV 74.8%. When stratified by infection type (lung, kidney, skin, gastrointestinal), sensitivity for sepsis remained low (range 4.4-40.0%), while specificity remained high (90.0-98.8%). When stratified by infection type, sensitivity for severe sepsis remained low (range 36.8-43.5%), while specificity remained high (76.5-90.2%). Conclusions: Using data from the REGARDS study, discharge diagnoses underestimated the number of sepsis and severe sepsis cases presenting to the hospital. Studies of sepsis epidemiology must account for the limitations of hospital discharge diagnoses.