Interobserver variability between experienced radiologists in evaluating the number of abnormal lymph nodes seen on...

To determine the interobserver variability of experienced radiologists in evaluating the number of suspicious lymph nodes (LNs) seen on axillary ultrasound (AUS) examination, and to evaluate the effects of clinicopathological features on the discordance.

A prospective non-randomised triple-blind study was conducted from July 2016 to December 2017 at Shanghai Ruijin Hospital. Three experienced radiologists performed AUS on 462 patients (469 axillae) with invasive breast cancer independently and assessed the number of abnormal LNs during the real-time AUS examination. Interobserver agreement was determined according to the kappa statistic.

The mean numbers of metastatic LNs on final histology study were 0.35, 0.98, 3.05 and 4.70 for patients with 0, 1, 2 and >2 abnormal nodes on preoperative AUS. When the cut-off (the number of abnormal LNs seen on AUS was >1) at which the maximal sum of sensitivity and specificity for diagnosis of ≥3 tumour-involved LNs was achieved was applied, the Az values ranged from 0.776 to 0.811. When the number of abnormal LNs was grouped into two-category classification (0-1 versus ≥2 abnormal LNs detected by AUS), good interobserver agreement (kappa: 0.601-0.687) was found, and disagreement between the radiologists was more frequently found in patients with T2-stage tumour and in those with one or two tumour-involved nodes.

Abnormal LN count according to two-category classification (0-1 versus ≥2 abnormal LNs detected by AUS) was a feasible method with good accuracy and high repeatability to select patients with heavy nodal disease in the post-Z0011 trial era.