Prediction accuracy of commonly used pneumonia severity scores in Aboriginal patients with severe community-acquired pneumonia: a retrospective study

Background: Severe community-acquired pneumonia (SCAP) is highly prevalent in the Aboriginal population. Few pneumonia severity scores are validated in this population. 

Aims: To assess the prediction accuracy of pneumonia severity scores in Aboriginal patients with SCAP and to identify risk factors for poor prognosis.

Methods: Retrospective cohort study examining Aboriginal patients admitted to the intensive care unit with confirmed SCAP between January 2011 and December 2014. Severity scores were calculated for SMARTCOP (systolic blood pressure, multi-lobar, albumin, respiratory rate, tachycardia, confusion, oxygenation and arterial pH), SMARTACOP (systolic blood pressure, multi-lobar, albumin, respiratory rate, tachycardia, Aboriginal status, confusion, oxygenation and arterial pH), CURB-65 (confusion, urea, respiratory rate, blood pressure and age ≥65 years), pneumonia severity index, Infectious Diseases Society of America and American Thoracic Society SCAP, and Acute Physiology and Chronic Health Evaluation (APACHE) II/III using medical records. Prediction accuracy of 30-day mortality and requirement for intensive respiratory and/or vasoactive support (IRVS) were assessed using logistic regression and the area under the receiver operating characteristic curve (AUROC). Multivariate analysis was used to test associations between poor prognosis and demographic/clinical variables.

Results: A total of 203 cases (49% women) was identified. Thirty-day mortality was 6.4% (n = 13), and 53% (n = 107) required IRVS. None of the tested pneumonia severity scores accurately predicted mortality. SMARTCOP and SMARTACOP predicted IRVS requirement with the highest diagnostic accuracy, but only achieved acceptable discrimination (P <0.001 and <0.001; AUROC = 0.74 and 0.75 respectively). APACHE II/III predicted both mortality (P = 0.003 and 0.001; AUROC = 0.74 and 0.73 respectively) and IRVS requirement (P <0.001 and <0.001; AUROC = 0.72 and 0.73 respectively). Multivariate analysis associated mortality with male gender, cirrhosis, immunosuppression and acidaemia, and IRVS requirement with multi-lobar pneumonia, hypotension and tachypnoea. Multivariate analysis for mortality and IRVS requirement achieved an AUROC of 0.93 and 0.87 respectively. 

Conclusion: None of the pneumonia severity scores accurately predicted mortality. We recommend SMARTACOP to predict IRVS requirement in Aboriginal patients with SCAP. Given Aboriginal patients are over-represented in Australian intensive care units, a new score is warranted for this understudied population.