High concordance of gene expression profiling-correlated immunohistochemistry algorithms in diffuse large B-cell lymphoma, not otherwise specified.

Diffuse large B-cell lymphoma (DLBCL) is classified into prognostically distinct germinal center B-cell (GCB) and activated B-cell subtypes by gene expression profiling (GEP). Recent reports suggest the role of GEP subtypes in targeted therapy. Immunohistochemistry (IHC) algorithms have been proposed as surrogates of GEP, but their utility remains controversial. Using microarray, we examined the concordance of 4 GEP-correlated and 2 non-GEP-correlated IHC algorithms in 381 DLBCLs, not otherwise specified. Subtypes and variants of DLBCL were excluded to minimize the possible confounding effect on prognosis and phenotype. Survival was analyzed in 138 cyclophosphamide, adriamycin, vincristine, and prednisone (CHOP)-treated and 147 rituximab plus CHOP (R-CHOP)-treated patients. Of the GEP-correlated algorithms, high concordance was observed among Hans, Choi, and Visco-Young algorithms (total concordance, 87.1%; κ score: 0.726 to 0.889), whereas Tally algorithm exhibited slightly lower concordance (total concordance 77.4%; κ score: 0.502 to 0.643). Two non-GEP-correlated algorithms (Muris and Nyman) exhibited poor concordance. Compared with the Western data, incidence of the non-GCB subtype was higher in all algorithms. Univariate analysis showed prognostic significance for Hans, Choi, and Visco-Young algorithms and BCL6, GCET1, LMO2, and BCL2 in CHOP-treated patients. On multivariate analysis, Hans algorithm retained its prognostic significance. By contrast, neither the algorithms nor individual antigens predicted survival in R-CHOP treatment. The high concordance among GEP-correlated algorithms suggests their usefulness as reliable discriminators of molecular subtype in DLBCL, not otherwise specified. Our study also indicates that prognostic significance of IHC algorithms may be limited in R-CHOP-treated Asian patients because of the predominance of the non-GCB type.