Bladder urothelial carcinoma (BLCA) is recognized to be immunogenic and tumorigenic. This study identified a novel long noncoding RNA (lncRNA) signature for predicting survival for patients with BLCA. A univariate Cox regression model and the random survival forest-variable hunting (RSF-VH) algorithm were employed to achieve variable selection. Ten lncRNAs (LOC105375787, CYTOR, URB1-AS1, C21orf91-OT1, CASC15, LOC101928433, FLJ45139, LINC00960, HOTAIR and TTTY19) with the highest prognostic values were identified to establish the prognostic model. The nomogram integrating the signature and clinical factors showed high concordance index values of 0.94, 0.7 and 0.90 in the three datasets, and the calibration curves showed concordance between the predicted and observed 3- and 5-year survival rates. The risk score based on the 10-lncRNA signature accurately distinguished high- and low-risk BLCA patients with different disease-specific survival(DSS) or overall survival(OS) outcomes, which were stratified according to clinical factors, including T stage and tumour grade. Gene set enrichment analysis identified BLCA-specific biological pathways and enriched functional categories, such as the cell cycle, DNA repair and immune system. Furthermore, the increased infiltration of immune cells in the high-risk group indicated that lncRNA-related inflammation may reduce the survival of BLCA patients.
Keywords: bladder cancer; immune infiltration; lncRNA; machine learning; nomogram.