The Impact of Scoring Method on Accuracy and Reproducibility of Hans Cell-of-Origin Prediction in Excisional Biopsies...

Aided by tissue microarray (TMA) technology, several RNA-correlated immunohistochemistry-based algorithms have been developed for cell-of-origin (COO) prediction in diffuse large B-cell lymphoma, not otherwise specified (DLBCL-NOS). However, there is currently no empirical evidence to guide the optimal application of these algorithms to whole tissue sections (WTSs).

To assess the impact of various scoring methods on the accuracy and reproducibility of the popular Hans algorithm.

We compared 3 different WTS-based scoring methods, designated as global, selective, and hotspot scoring, to a matched TMA evaluation and gold standard RNA analysis (Lymph2Cx; germinal center B cell n = 64; activated B cell/unclassified n = 68) using a representative series of 132 excisional biopsies of de novo DLBCL-NOS. Positivity scores (10% increments) were submitted by 3 expert lymphoma pathologists, with 30% or more defining positivity.

Sixty-eight of the 132 cases of DLBCL-NOS (52%) exhibited variation in Hans immunohistochemistry marker phenotype as a consequence of scoring method and/or interscorer discordance. Although this led to changes in Hans COO assignment in 27 of 132 cases (20%), none of the WTS-based scoring methods were statistically inferior to one another in terms of raw accuracy. Hotspot scoring yielded the lowest proportion of borderline scores (20%-40% range) for BCL6 transcription repressor (BCL6) and IRF4 transcription factor (MUM1) but negatively impacted the balance between sensitivity and specificity for these markers. Selective scoring was associated with significantly worse interscorer concordance compared to TMA evaluation, which it was designed to replicate.

Overall, our data favor the use of global scoring for its noninferior accuracy, solid interscorer concordance, nonnegative influence on individual Hans markers, and current widespread use.