Background: Early aggressive therapy can reduce the mortality associated with severe sepsis but this relies on prompt recognition, which is hindered by variation among published severity criteria. Our aim was to test the performance of different severity scores in predicting mortality among a cohort of hospital inpatients with sepsis.
Methods: We anonymously linked routine outcome data to a cohort of prospectively identified adult hospital inpatients with sepsis, and used logistic regression to identify associations between mortality and demographic variables, clinical factors including blood culture results, and six sets of severity criteria. We calculated performance characteristics, including area under receiver operating characteristic curves (AUROC), of each set of severity criteria in predicting mortality.
Results: Overall mortality was 19.4% (124/640) at 30 days after sepsis onset. In adjusted analysis, older age (odds ratio 5.79 (95% CI 2.87-11.70) for ≥80y versus <60y), having been admitted as an emergency (OR 3.91 (1.31-11.70) versus electively), and longer inpatient stay prior to sepsis onset (OR 2.90 (1.41-5.94) for >21d versus <4d), were associated with increased 30 day mortality. Being in a surgical or orthopaedic, versus medical, ward was associated with lower mortality (OR 0.47 (0.27-0.81) and 0.26 (0.11-0.63), respectively). Blood culture results (positive vs. negative) were not significantly association with mortality. All severity scores predicted mortality but performance varied. The CURB65 community-acquired pneumonia severity score had the best performance characteristics (sensitivity 81%, specificity 52%, positive predictive value 29%, negative predictive value 92%, for 30 day mortality), including having the largest AUROC curve (0.72, 95% CI 0.67-0.77).
Conclusions: The CURB65 pneumonia severity score outperformed five other severity scores in predicting risk of death among a cohort of hospital inpatients with sepsis. The utility of the CURB65 score for risk-stratifying patients with sepsis in clinical practice will depend on replicating these findings in a validation cohort including patients with sepsis on admission to hospital.