Community-acquired pneumonia: development of a bedside predictive model and scoring system to identify the aetiology.

Although initial presentation has been commonly used to select empirical therapy in patients with community-acquired pneumonia (CAP), few studies have provided a quantitative estimation of its value. The objective of this study was to analyse whether a combination of basic clinical and laboratory information performed at bedside can accurately predict the aetiology of pneumonia. A prospective study was developed among patients admitted to the Emergency Department University Hospital Arnau de Vilanova, Lleida, Spain, with CAP. Informed consent was obtained from patients in the study. At entry, basic clinical (age, comorbidity, symptoms and physical findings) and laboratory (white blood cell count) information commonly used by clinicians in the management of respiratory infections, was recorded. According to microbiological results, patients were assigned to the following categories: bacterial (Streptococcus pneumoniae and other pyogenic bacteria), virus-like (Mycoplasma pneumoniae, Chlamydia spp and virus) and unknown pneumonia. A scoring system to identify the aetiology was derived from the odds ratio (OR) assigned to independent variables, adjusted by a logistic regression model. The accuracy of the prediction rule was tested by using receiver operating characteristic curves. One hundred and three consecutive patients were classified as having virus-like (48), bacterial (37) and unknown (18) pneumonia, respectively. Independent predictors related to bacterial pneumonia were an acute onset of symptoms (OR 31; 95% CI, 6-150), age greater than 65 or comorbidity (OR 6.9; 95% CI, 2-23), and leukocytosis or leukopenia (OR 2; 95% CI, 0.6-7). The sensitivity and specificity of the scoring system to identify patients with bacterial pneumonia were 89% and 94%, respectively. The prediction rule developed from these three variables classified the aetiology of pneumonia with a ROC curve area of 0.84. Proper use of basic clinical and laboratory information is useful to identify the aetiology of CAP. The prediction rule may help clinicians to choose initial antibiotic therapy.