A PREDICTIVE MODEL FOR THE TREATMENT APPROACH TO COMMUNITY-ACQUIRED PNEUMONIA IN PATIENTS NEEDING ICU ADMISSION

Objective: To create a predictive model for the treatment approach to community-acquired pneumonia (CAP) in patients needing Intensive Care Unit (ICU) admission. Design: Multicenter prospective study. Setting: Twenty-six Spanish ICUs. Patients: One hundred seven patients with CAP , all of them with accurate etiological diagnosis, divided in three gr oups according to their etiology in typical (bacterial pneumonia), Leg ionella and other atypical (Mycoplasma, Chlamydia spp. and virus). For the multivariate analysis we grouped Legionella and other atypical et iologies in the same category. Methods: We recorded 34 variables inclu ding clinical characteristics, risk factors and radiographic pattern. We used a multivariate logistic regression analysis to find out a pred ictive model. Results: We have the complete data in 70 patients. Four variables: APACHE II, (categorized as a dummy variable) serum sodium a nd phosphorus and ''length of symptoms'' gave an accurate predictive m odel (c = 0.856). From the model we created a score that predicts typi cal pneumonia with a sensitivity of 90.2% and specificity 72.4%. Concl usion: Our model is an attempt to help in the treatment approach to CA P in ICU patients based on a predictive model of basic clinical and la boratory information. Further studies, including larger numbers of pat ients, should validate and investigate the utility of this model in di fferent clinical settings.