Derivation and Independent Validation a Gene Expression-Based Predictor for Post-Cystectomy Recurrence in Nodal Negative Muscle Invasive Urothelial Carcinoma
AS Baras, SC Smith, CA Moskaluk, HF Frierson, D Theodorescu. UVA Health System, Charlottesville, VA
Background: Despite radical surgery and lymphadenectomy, approximately half of muscle invasive urothelial carcinoma (MIUC) patients experience metastatic recurrence within two years of surgery. Within this population, patients with nodal negative disease are not commonly offered adjuvant chemotherapy, though treatment failures remain common. Predictive risk stratification in this population would be useful for decisions regarding adjuvant therapy.
Design: We used two published studies of gene expression profiling of tumor tissue to design a prediction metric for recurrence post-cystectomy among pathologically node negative MIUCs. First, we used one dataset based on the Affymetrix HG-U133A platform for training to compare gene expression between a subset of cases evincing disease free survival >2 years (N=17) to those with disease specific recurrence in < 2 years (N=14). A nearest neighbor classification system yielding a prediction score from 0 to 1 was employed to classify samples based on expression of recurrence related genes. This gene set was then applied to an independent test dataset of 47 tumors profiled on the Illumina Human WG6 V2 platform. Association between the gene expression-based prediction score and clinicopathologic parameters was tested through multivariate logistic regression and receiver operating characteristic (ROC) analysis.
Results: By a criterion of 2-fold differential expression, > 75 units average difference, and P<0.001, we found 33 candidate probes associated with disease recurrence in the training set. Prediction scores within the learning dataset were found to be independent of stage, grade, age, and sex (p=0.011). From these 33 probes, 27 Unigene IDs were able to be matched across microarray platforms. When the same prediction algorithm was applied to the independent test set of nodal negative MIUCs, we again found a significant independent association with disease recurrence (P=0.015), with a favorable area under the ROC curve (P=0.012).
Conclusions: Prediction models derived from gene expression analysis of tumor tissue samples can provide independent prognostic information about the course of patient disease in node negative stage MIUCs. This approach may show promise as an adjunct to routine histologic examination in determining patient therapy, pending validation in larger independent cohorts.
Category: Pan-genomic/Pan-proteomic Approaches to Diseases
Monday, March 22, 2010 1:00 PM
Poster Session II # 219, Monday Afternoon