Prognostic Value of Computed Tomography in Empyema: A Multicenter Retrospective Cohort Study.

Chest computed tomography is performed in patients with empyema for various reasons. However, its predictive ability for patient outcomes in empyema has not been evaluated. To evaluate the predictive ability of computed tomography findings (pleural thickness, loculation, interlobar pleural effusion, lung abscess, and bronchopleural fistula) for 90-day mortality in empyema. This multicenter retrospective cohort study was conducted across six acute care hospitals in Japan. We included patients with confirmed empyema diagnoses who underwent chest computed tomography within 7 days of diagnosis. Imaging findings were defined as pleural thickness, loculation, interlobar pleural effusion, lung abscess, or bronchopleural fistula. One radiologist interpreted the computed tomography scans without patient information. The primary outcome was 90-day mortality. We calculated the differences in 90-day mortality between the presence and absence of each computed tomography finding using logistic regression with or without adjustment for early thoracic surgery. A total of 711 patients were included in our study. Thoracic surgery was performed in 27% of patients, and the 90-day mortality rate was 10%. The differences (95% confidence intervals) in 90-day mortality without and with adjustment for early thoracic surgery were as follows: pleural thickness, 3.09% (-1.35% to 7.54%) and 2.70% (-1.80% to 7.20%); loculation, -4.01% (-8.61% to 0.60%) and -3.80% (-8.41% to 0.81%); interlobar pleural effusion, -9.15% (-14.58% to -3.72%) and -8.96% (-14.39% to -3.53%); lung abscess, 7.04% (-1.16% to 15.2%) and 6.86% (-1.34% to 15.05%); and bronchopleural fistula, 13.80% (7.66% to 19.94%) and 13.63% (7.50% to 19.77%), respectively. Although interlobar pleural effusion predicted lower 90-day mortality regardless of early thoracic surgery, the presence of bronchopleural fistula predicted higher 90-day mortality with empyema. Our results warrant further validation.