Performance of Chest Computed Tomography in Differentiating Coronavirus Disease 2019 From Other Viral Infections...

An expert consensus recently proposed a standardized coronavirus disease 2019 (COVID-19) reporting language for computed tomography (CT) findings of COVID-19 pneumonia.

The purpose of the study was to evaluate the performance of CT in differentiating COVID-19 from other viral infections using a standardized reporting classification.

A total of 175 consecutive patients were retrospectively identified from a single tertiary-care medical center from March 15 to March 24, 2020, including 87 with positive reverse transcription-polymerase chain reaction (RT-PCR) test for COVID-19 and 88 with negative COVID-19 RT-PCR test, but positive respiratory pathogen panel. Two thoracic radiologists, who were blinded to RT-PCR and respiratory pathogen panel results, reviewed chest CT images independently and classified the imaging findings under 4 categories: "typical" appearance, "indeterminate," "atypical," and "negative" for pneumonia. The final classification was based on consensus between the readers.

Patients with COVID-19 were older than patients with other viral infections (P=0.038). The inter-rater agreement of CT categories between the readers ranged from good to excellent, κ=0.80 (0.73 to 0.87). Final CT categories were statistically different among COVID-19 and non-COVID-19 groups (P<0.001). CT "typical" appearance was more prevalent in the COVID-19 group (64/87, 73.6%) than in the non-COVID-19 group (2/88, 2.3%). When considering CT "typical" appearance as a positive test, a sensitivity of 73.6% (95% confidence interval [CI]: 63%-82.4%), specificity of 97.7% (95% CI: 92%-99.7%), positive predictive value of 97% (95% CI: 89.5%-99.6%), and negative predictive value of 78.9% (95% CI: 70%-86.1%) were observed.

The standardized chest CT classification demonstrated high specificity and positive predictive value in differentiating COVID-19 from other viral infections when presenting a "typical" appearance in a high pretest probability environment. Good to excellent inter-rater agreement was found regarding the CT standardized categories between the readers.