Diagnostic agreement and limitations of the Banff automated system in kidney transplant biopsies

The Banff classification for renal allograft rejection has evolved over time, increasing in complexity. For non-pathologists conducting retrospective studies, assigning Banff diagnostic categories across different eras presents a significant challenge. The Automated Diagnosis System (ADS) is a publicly available web-based tool designed to standardize Banff category diagnoses based on Banff scoring. We retrospectively evaluated ADS using 1,071 kidney biopsy results from 544 transplant recipients, including 146 ABO-incompatible cases performed at our institution. Overall concordance between ADS and pathologists was 69.8%, with high agreement in non-rejection (97.4%) and rejection (86.3%) cases. Among rejection cases, discrepancies were noted in 27 antibody-mediated rejection (AMR) and 22 T cell-mediated rejection (TCMR) cases. Discrepancies were frequently observed in AMR following ABO-incompatible transplantation and in chronic TCMR, highlighting challenges in standardizing these categories. Despite these limitations, ADS demonstrated acceptable concordance and potential utility for promoting global standardization in rejection diagnosis.