Sepsis incidence and mortality are underestimated in Australian intensive care unit administrative data

Objectives: To compare estimates of the incidence and mortality of sepsis and septic shock among patients in Australian intensive care units (ICUs) according to clinical diagnoses or binational intensive care database (ANZICS CORE) methodology. Design, setting, participants: Prospective inception cohort study (3-month inception period, 1 October - 31 December 2016, with 60-day follow-up); daily screening of all patients in a tertiary hospital 60-bed multidisciplinary ICU. Main outcomes: Diagnoses of sepsis and septic shock according to clinical criteria and database criteria; in-hospital mortality (censored at 60 days). Results: Of 864 patients admitted to the ICU, 146 (16.9%) were diagnosed with sepsis by clinical criteria and 98 (11%) according to the database definition (P < 0.001); the sensitivity of the database criteria for sepsis was 52%, the specificity 97%. Forty-nine patients (5.7%) were diagnosed with septic shock by clinical criteria and 83 patients (9.6%) with the database definition (P < 0.001); the sensitivity of the database criteria for septic shock was 65%, the specificity 94%. In-hospital mortality of patients diagnosed with sepsis was greater in the clinical diagnosis group (39/146, 27%) than in the database group (17/98, 17%; P = 0.12); for septic shock, mortality was significantly higher in the clinical diagnosis group (18/49; 37%) than in the database group (13/83; 16%; P = 0.006). Conclusions: When compared with the reference standard - prospective clinical diagnosis - ANZICS CORE database criteria significantly underestimate the incidence of sepsis and overestimate the incidence of septic shock, and also result in lower estimated hospital mortality rates for each condition.