Abstract
•Binary use of the HNCIG iENE classification showed moderate inter-rater agreement across an international cohort.•Ordinal grading of iENE showed weak inter-rater agreement.•Intra-rater agreement was moderate for binary and grade iENE classifications.•Inter-rater agreement was similar for CT and MRI.•Agreement did not differ by radiologist years of experience.
Extranodal extension detected on imaging (iENE) is an increasingly recognised risk factor for poor prognosis in patients with head and neck squamous cell carcinoma. However, its clinical usefulness depends on accurate and reliable radiological assessment. A major limitation to the use of iENE as a prognostic marker remains variability in its interpretation, attributable in part to the lack of standardised diagnostic criteria.
This international, multicentre reliability study included expert head and neck radiologists nominated to represent member groups of the Head and Neck Cancer International Group (HNCIG). Participants underwent standardised training using the HNCIG consensus classification and educational atlas, and then independently evaluated anonymised CT and MRI scans using both binary (iENE present or absent) and ordinal (grades 0–3) classifications over two rounds. Inter- and intra-rater reliability were evaluated using free-marginal multi-rater kappa statistics and intraclass correlation coefficients. Subgroup analyses were performed by imaging modality and radiologist experience.
Fourteen radiologists representing 12 research groups, spanning 25 countries, completed both assessment rounds. Overall inter-rater agreement was moderate for binary classification (κ = 0.75, 95% CI 0.66–0.84) and weak for ordinal classification (κ = 0.49, 95% CI 0.42–0.56). Intra-rater agreement was moderate for binary classification and moderate to strong for ordinal classification. Agreement levels did not differ significantly by imaging modality or years of experience.
Reliable radiological assessment is necessary for the use of iENE in clinical trials, staging, and routine decision making. Using the HNCIG consensus classification in a binary format produced consistent results across international expert radiologists and imaging modalities, supporting its wider use. However, substantial variability remains in grading the extent of iENE, indicating the need for further training support.