Aims Metastatic HER2-low breast cancer (HLBC) can be treated by trastuzumab deruxtecan. Assessment of low levels of HER2 protein expression suffers from poor interobserver reproducibility. The aim of the study was to evaluate the interobserver agreement among subspecialised breast pathologists and develop a practical algorithm for assessing HLBC.

Methods Six breast pathologists (4 juniors, 2 seniors) evaluated 106 HER2 immunostained slides with 0/1+expression. Two rounds (R1, R2) of ring study were performed before and after training with a modified Ki-67 algorithm, and concordance was assessed.

Results Agreement with 5% increments increased from substantial to almost perfect (R1: 0.796, R2: 0.804), and remained substantial for three categories (<1% vs 1%–10% vs >10%) (R1: 0.768, R2: 0.764). Seniors and juniors had almost perfect agreement with 5% increments (R1: 0.859 and 0.821, R2: 0.872 and 0.813). For the three categories, agreement remained almost perfect among seniors (R1: 0.837, R2: 0.860) and substantial among juniors (R1: 0.792, R2: 0.768). Binary analysis showed suboptimal agreement, decreasing for both juniors and seniors from substantial (R1: 0.650 and 0.620) to moderate (R2: 0.560 and 0.554) using the 1% cut-off, and increasing from moderate to substantial (R1: 0.478, R2: 0.712) among seniors but remaining moderate (R1: 0.576, R2: 0.465) among juniors using the 10% cut-off. The average scoring time per case was higher (72 vs 92 s).

Conclusions Subspecialised breast pathologists have suboptimal agreement for immunohistochemical evaluation of HLBC using the modified Ki-67 methodology. An urgent need remains for a new assay/algorithm to reliably evaluate HLBC.