Comparing Immunohistochemical Methods for Predicting Gene Expression Profile and Survival of Diffuse Large B-Cell Lymphoma Treated with Rituximab
PN Meyer, LM Smith, K Fu, TC Greiner, P Aoun, J Delabie, RD Gascoyne, A Rosenwald, RM Braziel, E Campo, JM Vose, G Lenz, LM Staudt, WC Chan, DD Weisenburger. University of Nebraska Medical Center, Omaha, NE; Leukemia/Lymphoma Molecular Profiling Project, Omaha, NE
Background: Diffuse large B-cell lymphomas (DLBCL) have variation in survival. Gene expression profiling (GEP) shows DLBCL from germinal center B-cells (GCB) have better prognosis than activated B-cells (ABC). Immunohistochemical algorithms to predict cell of origin have been published. Our goal was to compare these algorithms and evaluate new methods to predict GEP and survival.
Design: Tissue microarrays (TMA) were created from 216 cases of DLBCL treated with rituximab and CHOP-like therapies. Clinical data were available for all cases. TMA were analyzed for GCET1, CD10, BCL6, MUM1, FOXP1, and LMO2. GEP were available on 168 of these cases. Sensitivity, specificity, positive predictive value, negative predictive value, and concordance with GEP were calculated. Overall survival for each method was determined by Kaplan-Meier curves, with differences evaluated by log-rank test.
Results: Concordance, sensitivity, and specificity were high with Hans and Choi algorithms. LMO2 alone had lower concordance, sensitivity, and specificity. The Muris or Nyman algorithms had low specificity or sensitivity, respectively. Removal of BCL6 from Hans (Hans*) and Choi (Choi*) algorithms did not lose predictive value. A tally system counting positive GCB markers (GCET1, CD10, and LMO2) versus positive ABC markers (MUM1 and FOXP1) had highest concordance and specificity while maintaining high sensitivity. All methods examined, except the Nyman algorithm, divided DLBCL patients into statistically significant prognostic groups.