Validation study of a noninvasive urine test for diagnosis and prognosis assessment of bladder cancer: evidence for improved models.

PURPOSE: We validated the performance of our previously reported test for bladder cancer based on urine gene expression patterns using an independent cohort. We also ascertained whether alternative models could achieve better accuracy.
MATERIALS AND METHODS: Gene expression patterns of the previously reported 48 genes, including the 12 + 2 genes of the signature, were analyzed by TaqMan® arrays in an independent set of 207 urine samples. We pooled all samples analyzed to date to obtain a larger training set of 404 and used it to search for putative improved new models.
RESULTS: Our 12 + 2 gene expression signature had overall 80% sensitivity with 86% specificity (AUC 0.914) to discriminate between bladder cancer and control samples. It had 75% sensitivity and 75% specificity (AUC 0.83) to predict tumor aggressiveness in the validation set of urine samples. After grouping all samples 3 new signatures for diagnosis containing 2, 5 and 10 genes, respectively, and 1 containing 6 genes for prognosis were designed. Diagnostic performance of the 2, 5, 10 and 12-gene signatures was maintained or improved in the enlarged sample set (AUC 0.913, 0.941, 0.949 and 0.944, respectively). Performance to predict aggressiveness was also improved in the 14 and 6-gene signatures (AUC 0.855 and 0.906, respectively).
CONCLUSIONS: This validation study confirms the accuracy of the 12 + 2 gene signature as a noninvasive tool for assessing bladder cancer. We present improved models with fewer genes that must be validated in future studies.