The Relationship Between Acuity of Organ Failure and Predictive Validity of Sepsis-3 Criteria

The Sepsis-3 taskforce defined sepsis as suspicion of infection and an acute rise in the Sequential Organ Failure Assessment score by 2 points over the preinfection baseline. Sepsis-3 studies, though, have not distinguished between acute and chronic organ failure, and may not accurately reflect the epidemiology, natural history, or impact of sepsis. Our objective was to determine the extent to which the predictive validity of Sepsis-3 is attributable to chronic rather than acute organ failure. DESIGN: Retrospective cohort study. SETTING: General medicine inpatient service at a tertiary teaching hospital. PATIENTS: A total of 3,755 adult medical acute-care encounters (1,864 confirmed acute infections) over 1 year. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: We measured the total Sequential Organ Failure Assessment score at the onset of infection and separated its components (baseline and acute rise) using case-by-case chart reviews. We compared the predictive validities of acuity-focused (acute rise in Sequential Organ Failure Assessment ≥ 2) and conventional (total Sequential Organ Failure Assessment ≥ 2) implementations of Sepsis-3 criteria. Measures of predictive validity were change in the rate of outcomes and change in the area under receiver operating characteristic curves after adding sepsis criteria to multivariate logistic regression models of baseline risk (age, sex, race, and Charlson comorbidity index). Outcomes were inhospital mortality (primary) and ICU transfer or inhospital mortality (secondary). Acuity-focused implementations of Sepsis-3 were associated with neither a change in mortality (2.2% vs 1.2%; p = 0.18) nor a rise in area under receiver operating characteristic curves compared with baseline models (0.67 vs 0.66; p = 0.75). In contrast, conventional implementations were associated with a six-fold change in mortality (2.4% vs 0.4%; p = 0.01) and a rise in area under receiver operating characteristic curves compared with baseline models (0.70 vs 0.66; p = 0.04). Results were similar for the secondary outcome. CONCLUSIONS: The evaluation of the validity of organ dysfunction-based clinical sepsis criteria is prone to bias, because acute organ dysfunction consequent to infection is difficult to separate from preexisting organ failure in large retrospective cohorts.