Prediction of severe community-acquired pneumonia: a systematic review and meta-analysis

Introduction: Severity assessment and site-of-care decisions for patients with community-acquired pneumonia (CAP) are pivotal for patients' safety and adequate allocation of resources. Late admission to the intensive care unit (ICU) has been associated with increased mortality in CAP. We aimed to review and meta-analyze systematically the performance of clinical prediction rules to identify CAP patients requiring ICU admission or intensive treatment. Methods: We systematically searched Medline, Embase, and the Cochrane Controlled Trials registry for clinical trials evaluating the performance of prognostic rules to predict the need for ICU admission, intensive treatment, or the occurrence of early mortality in patients with CAP. Results: Sufficient data were available to perform a meta-analysis on eight scores: PSI, CURB-65, CRB-65, CURB, ATS 2001, ATS/IDSA 2007, SCAP score, and SMART-COP. The estimated AUC of PSI and CURB-65 scores to predict ICU admission was 0.69. Among scores proposed for prediction of ICU admission, ATS-2001 and ATS/IDSA 2007 scores had better operative characteristics, with a sensitivity of 70% (CI, 61 to 77) and 84% (48 to 97) and a specificity of 90% (CI, 82 to 95) and 78% (46 to 93), but their clinical utility is limited by the use of major criteria. ATS/IDSA 2007 minor criteria have good specificity (91% CI, 84 to 95) and moderate sensitivity (57% CI, 46 to 68). SMART-COP and SCAP score have good sensitivity (79% CI, 69 to 97, and 94% CI, 88 to 97) and moderate specificity (64% CI, 30 to 66, and 46% CI, 27 to 66). Major differences in populations, prognostic factor measurement, and outcome definition limit comparison. Our analysis also highlights a high degree of heterogeneity among the studies. Conclusions: New severity scores for predicting the need for ICU or intensive treatment in patients with CAP, such as ATS/IDSA 2007 minor criteria, SCAP score, and SMART-COP, have better discriminative performances compared with PSI and CURB-65. High negative predictive value is the most consistent finding among the different prediction rules. These rules should be considered an aid to clinical judgment to guide ICU admission in CAP patients.