Predicting mortality with severity assessment tools in out-patients with community-acquired pneumonia.

INTRODUCTION In community-acquired pneumonia, severity assessment tools, such as CRB65, CURB65 and Pneumonia Severity Index (PSI), have been promoted to increase the proportion of patients treated in the community. The prognostic accuracy of these scores is established in hospitalized patients, but less is known about their use in out-patients. We aimed to study the accuracy of these severity tools to predict mortality in patients managed as out-patients. METHODS We performed a systematic review and meta-analysis according to MOOSE guidelines. From 1980 to 2010, we identified 13 studies reporting prognostic information for the CRB65, CURB65 and PSI severity scores in out-patients (either exclusively managed in the community or discharged from an emergency department <24 h after admission). Two reviewers independently collected data and assessed study quality. Performance characteristics across the studies were pooled using a random-effects model. Relationships between sensitivity and specificity were plotted using summary receiver operator characteristic curves (sROC). RESULTS Out-patient mortality ranged from 0% to 3.5%. Four studies were identified for CRB65, 2 for CURB65 and 10 for PSI. Mortality was low for out-patients in the low-risk CRB65 classes [CRB65 0 or 1: mortality occurred in 3 of 1494 patients (0.2%)] but higher in CRB65 Groups 2-4 [mortality 13 of 154 patients (8.4%)]. Similarly, mortality was low in PSI Classes I-III [mortality 8 of 3655 patients (0.2%)] managed as out-patients but higher in Classes IV and V [mortality 32 of 317 patients (10.1%)]. CRB65 showed pooled sensitivity of 81% (54-96%), pooled specificity of 91% (90-93%) and the area under the sROC was 0.91 [standard error (SE) 0.05]. For PSI, pooled sensitivity was 92% (64-100%), pooled specificity was 90% (89-91%) and area under the sROC was 0.92 (SE 0.03). There were insufficient studies to analyse CURB65. CONCLUSION The limited data available suggest that CRB65 and PSI can identify groups of patients at low risk of mortality that can be safely managed in the community.