Inter-reader reliability of Ovarian-Adnexal Reporting and Data System US: a systematic review and meta-analysis

Ovarian-Adnexal Reporting and Data System (O-RADS) US provides a standardized lexicon for ovarian and adnexal lesions, facilitating risk stratification based on morphological features for malignancy assessment, which is essential for proper management. However, systematic determination of inter-reader reliability in O-RADS US categorization remains unexplored. This study aimed to systematically determine the inter-reader reliability of O-RADS US categorization and identify the factors that affect it.

Original articles reporting the inter-reader reliability of O-RADS US in lesion categorization were identified in the MEDLINE, EMBASE, and Web of Science databases from January 2018 to December 2023. DerSimonian-Laird random-effects models were used to determine the meta-analytic pooled inter-reader reliability of the O-RADS US categorization. Subgroup meta-regression analysis was performed to identify the factors causing study heterogeneity.

Fourteen original articles with 5139 ovarian and adnexal lesions were included. The inter-reader reliability of O-RADS US in lesion categorization ranged from 0.71 to 0.99, with a meta-analytic pooled estimate of 0.83 (95% CI, 0.78-0.88), indicating almost perfect reliability. Substantial study heterogeneity was observed in the inter-reader reliability of the O-RADS US categorization (I² = 96.9). In subgroup meta-regression analysis, reader experience was the only factor associated with study heterogeneity. Pooled inter-reader reliability of the O-RADS US categorization was higher in studies with all experienced readers (0.86; 95% CI, 0.81-0.91) compared to those with multiple readers including trainees (0.74; 95% CI, 0.70-0.78; P = 0.009). The inter-reader reliability of US descriptors ranged from 0.39 to 0.97, with ascites and peritoneal nodules showing almost perfect reliability (0.79- 0.97).

The O-RADS US risk stratification system demonstrated almost perfect inter-reader reliability in lesion categorization. Our results highlight the importance of targeted training and descriptor simplification to improve inter-reader reliability and clinical adoption.