Background: Approximately one million people are hospitalized for pneumonia (PNA) every year resulting in approximately 50,000 deaths. The Infectious Diseases Society of America and the American Thoracic Society recommend that in addition to clinical features, a demonstrable infiltrate by chest radiograph (CXR) or other imaging technique, with or without supporting microbiological data, is required for the diagnosis of PNA. Our study aimed to determine the accuracy of a CXR for detecting radiographic findings typical of PNA using a novel approach of language stratification.

Methods: A 2-year retrospective study was conducted of all hospitalized adults 18 years or older, who received a CXR and a subsequent non-contrast CT scan of the chest within 48 hours of arrival to the emergency department and admitted to the medicine service, in 2 large academic tertiary centers. All CXR reports were reviewed for a radiologic impression suggestive of PNA. Of the 2,420 patients that met inclusion criteria, 249 charts were excluded due to the ambiguous language used in the reference standard (CT scan). CXR interpretations were compared to the corresponding CT scan, which was used as the reference standard. CXRs and CT scans were stratified into 7 different categories based on the language used by a radiologist in the impression section. Sensitivity, specificity, positive predictive value (PPV) and negative predictive values (NPV) of a CXR in identifying radiographic images of PNA were then calculated. True positives were those with chest CT confirmed PNA using high probability language. True negatives were those where the chest CT confirmed the absence of an image suggestive of PNA. False positive CXRs were calculated against a chest CT without an image indicating pneumonia. False negative CXRs were determined with the reference standard being a CT scan with intermediate to high probability language reflective of a PNA.

Results: Of the 2171 charts analyzed, the prevalence of CT images consistent with PNA was 38.7%. As a radiographic tool for detecting an image compatible with PNA, the CXR was found to have: sensitivity 46.6%, specificity 71.4%, PPV 50.7%, NPV 67.9%. Using these numbers, for every 100 ambiguous cases that necessitated a radiographic image to aid in the diagnosis of PNA, the following can be extrapolated: approximately 39 patients would have an image suggestive of PNA on CT with 21 of those potential pneumonias being missed on CXR. In addition, up to 18 patients without an image suggestive of PNA on CT would have a false positive CXR.

Conclusions: The inclusion criteria reflect an at-risk group of patients with CXR findings that necessitated an order for a non-contrast chest CT within 48 hours. Our results suggest that in this population, out of 100 patients who receive a CXR for the evaluation of PNA, up to 38 could potentially be inappropriately managed. As clinicians, we struggle with the correct diagnosis when the clinical picture is ambiguous. Our findings demonstrate that in these situations, a CXR result may not assist to rule-out or rule-in the diagnosis of PNA.