Diagnostic performance and inter-observer reproducibility of glomerular PLA2R immunohistochemistry in primary membranous nephropathy: a multi-laboratory blinded observer study

Abstract Background PLA2R immunohistochemistry (IHC) and serum anti-PLA2R antibodies are key diagnostic tools in PLA2R-associated membranous nephropathy (MN). However, diagnostic performance and inter-observer reproducibility of PLA2R IHC across different laboratory protocols and cut-offs remain insufficiently validated. We performed local validation of two routine PLA2R IHC protocols and report the methodological considerations that emerged, including the interaction between cut-off selection, diagnostic accuracy, and observer reproducibility. Methods In a multi-laboratory study, kidney biopsies from patients with primary MN (pMN), secondary MN, and non-MN controls were stained using two routine PLA2R IHC protocols (Aarhus University Hospital [AUH] and Herlev Hospital [HEH]). Seven blinded observers scored granular capillary wall staining intensity on a 0–3 scale. Diagnostic performance was assessed at a prespecified cut-off (≥ 2) and at protocol-specific optimal cut-offs determined by Youden’s J index. Inter-observer agreement for the ordinal scale was quantified using intraclass correlation coefficients (ICC). Reproducibility of dichotomized classifications was evaluated using the ONEST framework, examining overall percent agreement (OPA) across increasing numbers of observers. Results At the prespecified cut-off (≥ 2), sensitivity/specificity was 70%/70% for AUH and 55%/100% for HEH. Applying Youden’s J calculated optimal cut-offs (AUH: ≥3; HEH: ≥1) yielded sensitivity/specificity of 65%/100% for AUH and 65%/93% for HEH. Single-rating reliability was good-to-excellent overall [ICC (2,1) = 0.85], with protocol-specific differences (AUH = 0.78; HEH = 0.93), and the seven-observer panel demonstrated excellent reliability [ICC (2,7) = 0.96–0.99]. ONEST analyses demonstrated higher agreement for HEH at the prespecified cut-off (OPA 93% vs. 60%) and earlier stabilization of agreement curves, whereas AUH remained sensitive to panel size throughout the seven-observer range. Diagnostic accuracy and reproducibility were not fully aligned for the HEH protocol: Youden’s J favored cut-off ≥ 1, whereas reproducibility was highest at ≥ 2. Conclusion PLA2R IHC provides clinically useful diagnostic information for PLA2R-associated MN, but performance and reproducibility are protocol-dependent and were optimized at different cut-offs. Laboratories should therefore validate both dimensions locally rather than adopting a single fixed cut-off. We propose a stepwise framework combining Youden’s J and ONEST as a practical guide for laboratories implementing PLA2R IHC.