Bacteremia Prediction Model for Community-acquired Pneumonia: External Validation in a Multicenter Retrospective Cohort

Objective: Many studies have described constructing a prediction model for bacteremia in community-acquired pneumonia (CAP), but these studies were not validated in external heterogeneous groups. The objective of this study was to test the generalizability of a previous bacteremia prediction model for CAP by external validation. Methods: This multicenter retrospective cohort analysis was performed in eight tertiary urban hospital emergency departments (EDs). We reviewed adult patients who were hospitalized after presentation to the ED with CAP. We categorized the enrolled patients into three groups according to the bacteremia prediction model score and calculated the number of patients with or without a blood culture-positive result. We performed a multivariable analysis to identify significant predictors for bacteremia. Results: Among the enrolled 2,001 patients, 1,592 (79.6%), 371 (18.5%), and 38 (1.9%) were stratified to a low-, moderate-, and high-risk group, respectively, and this proportion was similar with previous study. Each group had a bacteremia-positive rate as follows: 1.2% for the low-risk group, 7.2% for the moderate-risk group, and 31.5% for the high-risk group. The area under the receiver operating characteristic curve for the bacteremia model in the external validation cohort was 0.81, and there was no significant difference with that of the previous internal validation cohort (p=0.246). Assuming that blood cultures were not performed in the low-risk patients, the sensitivity and specificity of this model were 0.68 and 0.81, respectively. Additionally, the positive predictive value and negative predictive value were 9.54 and 98.87%, respectively. A platelet count less than 130 x 10(9) cells/L, albumin less than 3.3 mg/dL, and C-reactive protein greater than 17 mg/dL were identified as significant predictors with a sensitivity and specificity of 0.70 and 0.83, respectively. Conclusion: The bacteremia prediction model was well validated in the general population and could help physicians make the decision to reduce the number of blood cultures in patients with CAP.