Abstract

Background: Several severity scores have been proposed to predict patient outcome and to guide initial management of patients with community acquired pneumonia (CAP). Most have been derived as predictors of mortality. A study was undertaken to compare the predictive value of these tools using different clinically meaningful outcomes as constructs for “severe pneumonia”.

Methods: A prospective cohort study was performed of all patients presenting to the emergency department with an admission diagnosis of CAP from March 2003 to March 2004. Clinical and laboratory features at presentation were used to calculate severity scores using the pneumonia severity index (PSI), the revised American Thoracic Society score (rATS), and the British Thoracic Society (BTS) severity scores CURB, modified BTS severity score, and CURB-65. The sensitivity, specificity, positive and negative predictive values were compared for four different outcomes (death, need for ICU admission, and combined outcomes of death and/or need for ventilatory or inotropic support).

Results: 392 patients were included in the analysis; 37 (9.4%) died and 26 (6.6%) required ventilatory and/or inotropic support. The modified BTS severity score performed best for all four outcomes. The PSI (classes IV+V) and CURB had a very similar performance as predictive tools for each outcome. The rATS identified the need for ICU admission well but not mortality. The CURB-65 score predicted mortality well but performed less well when requirement for ICU was included in the outcome of interest. When the combined outcome was evaluated (excluding patients aged >90 years and those from nursing homes), the best predictors were the modified BTS severity score (sensitivity 94.3%) and the PSI and CURB score (sensitivity 83.3% for both).

Conclusions: Different severity scores have different strengths and weaknesses as prediction tools. Validation should be done in the most relevant clinical setting, using more appropriate constructs of “severe pneumonia” to ensure that these potentially useful tools truly deliver what clinicians expect of them.