Gene Expression Profile Signatures to Predict Survival in Diffuse Large B-Cell Lymphoma: A Meta-Analysis of Early Results.
Lyman, GH; Kuderer, NM
Published in: Blood
Background: Diffuse large B-cell lymphoma (DLBCL) represents the most common lymphoma in adults. Although responsive to chemotherapy, less than half of patients are currently cured. In an effort to better assess prognosis, several investigators have attempted to define gene expression signatures based on their prediction of survival. The preliminary analysis reported here is part of an ongoing effort to systematically review and catalogue classes and test performance characteristics of reported gene expression signatures in patients with DLBCL.Methods: All reports of gene expression profiling in patients with DLBCL were sought through an extensive search of the published literature including MEDLINE, EMBASE, the Cochrane Library and hand searching of references. Eleven reported studies in patients with DLBCL were identified which reported the relationship between a gene expression signature and overall survival. Weighted summary measures of test performance were estimated including sensitivity, specificity, likelihood ratio, posttest probability (PP) and the diagnostic odds ratio (DOR) as an overall measure of test discrimination. An inconsistency index (I2) was used to reflect the proportion of variation in estimates due to heterogeneity. Seven studies also provided signature results stratified by International Prognostic Index (IPI).Results: Reported series included 1027 patients ranging from 22 to 240 per study. Five studies utilized cross validation techniques while six were based on independent cohorts. Five-year survivals were 28% and 69% among the 501 high-risk profile and 525 low-risk patients, respectively. Summary gene expression signature performance characteristics for survival are shown:Measure Overall [95%CL] Range Median Mean [95%CL] Sensitivity 75% [66, 85] 52% – 96% 74% 75% [65, 85] Specificity 75 % [69, 82] 31% – 90% 75% 71% [60, 83] PP + 76% [69, 83] 44% – 92% 76% 75% [65, 84] PP − 30% [22, 42] 8% – 61% 31% 29% [18,3 9] DOR 7.34 [4.70, 11.46] 3.6 – 170.0 8.04 23.5 [0, 56.3]The estimated I2 was 43%. The false negative rate and false positive rate were over 20% in 7 (64%) and 9 (82%) of studies, respectively. Although not reaching statistical significance, test performance measures were generally poorer in studies with independent validation. The number of genes in the assay correlated inversely with the DOR [rsp =0.59, p=.05], the likelihood ratio negative [rsp=0.70, p=.02] and the posttest probability negative [rsp=.78, p<.01]. Among reporting studies, 263 patients were IPI high and 374 IPI low-risk. The 5-year survival rate among IPI high risk patients were 8% and 52% in high and low risk profile subjects, respectively. Likewise, among IPI low risk patients, 5-year survival of 46% and 81% were observed among high and low risk profile patients. The DOR of the gene expression signatures among IPI low risk and high-risk patients were 6.66 [3.15,14.07] and 10.98 [5.51,21.89], respectively.Discussion: Gene expression profiling based on microarray analysis shows early promise for improving clinical estimation of survival in patients with DLBCL. However, the use of these assays in therapeutic decision-making must consider both the limitations of assay test performance and the specific patient population being evaluated.
