Validation of imaging reporting and data system of coronavirus disease 2019 lexicons CO-RADS and COVID-RADS with radiologists’ preference: a multicentric study

Abstract

Background A retrospective multicentric study gathered 1439 CT chest studies with suspected coronavirus disease 2019 (COVID-19) affection. Three radiologists, blinded to other results, interpreted all studies using both lexicons with documentation of applicability and preferred score in assessing every case. The purpose of the study is to assess COVID-19 standardized assessment schemes’ (CO-RADS and COVID-RADS lexicons) applicability and diagnostic efficacy. Results This study included 991 RT-PCR-confirmed CT studies. An almost perfect agreement was found in COVID-RADS among the three observers (Fleiss Kappa = 0.82), opposed by a substantial agreement in CO-RADS ( Κ = 0.78). The preference records favor COVID-RADS/CO-RADS in 78.5%/12.5%, 75.5%/24.5%, and 73.4%/24.5% regarding the three radiologists’ records, respectively. The distinguishability between positive and negative RT-PCR cases was 0.92 for COVID-RADS, while it was 0.85 for CO-RADS. On the other hand, both lexicons’ performance regarding clinical diagnosis and clinical suspicion index was 0.93 for COVID-RADS and 0.94 for CO-RADS. A very high to excellent agreement between the three observers for COVID-RADS/CO-RADS preference was concluded (Fleiss Kappa = 0.80 to 0.94). These results were statistically significant ( p < 0.001). Conclusion Both lexicon scores (CO-RADS and COVID-RADS) were found to be applicable in the COVID-19 structured report with the preference of COVID-RADS in more than 50% of cases. The diagnostic accuracy of COVID-RADS against RT-PCR was higher than that of CO-RADS.