A decision support model to predict the presence of an acute infiltrate on chest radiograph

A chest infiltrate is needed to make a diagnosis of community-acquired pneumonia, but chest X-rays might be time consuming, entail radiation exposure, and demand resources that are not always available. We sought to derive a model to predict whether a patient will have an infiltrate on chest X-ray (CXR). This prospective observational study included patients visiting the Emergency Department of Beilinson Hospital in the years 2003–2004 (derivation cohort) and 2010–2011 (validation cohort), who had undergone a CXR, and were suspected of having a respiratory infection. We excluded all patients with possible healthcare associated infections. A logistic regression model was derived and applied to the validation cohort. A total of 1,555 patients met inclusion criteria: 993 in the derivation cohort and 562 in the validation cohort with 287 (29%) and 226 (40%) having an infiltrate, respectively. The derivation model area-under-the curve (AUC) was 0.79 (95% CI 0.76–0.82). We categorized the patients into three groups—presence or absence of infiltrate, or undetermined. In the validation cohort, 70 (12%) patients were classified as ‘no infiltrate’; 3 (4%) of them had an infiltrate, 367 (65%) were classified as ‘infiltrate’; 190 (52%) of them had an infiltrate on CXR, and 125 (46%) were classified as ‘undetermined’; 33 (26%) of them with an infiltrate on CXR. Using this prediction model for the evaluation of patients with suspected respiratory infection in an ED setting may help avoid over 10% of CXRs.