A survival prediction model for veterans with advanced bladder cancer (BC) treated with immune checkpoint inhibitors (ICI).

855 Background: ICIs have been associated with improved survival for patients with BC and are now an integral part of the treatment paradigm. However, predictive biomarkers for ICI in BC have not been established. Here we report outcomes for veterans with BC treated with ICI and evaluate a prediction model based on clinical and genomic characteristics. Methods: We analyzed a cohort of 456 patients with BC treated with single agent anti-PD1/PDL1 therapies from 2019 to 2024 in the VA Informatics and Computing Infrastructure (VINCI) database, and also received FoundationOne CDx testing. Genomic analysis included gene mutations, copy number variants (CNV), and key biomarkers such as PDL1 CPS expression, and tumor mutational burden (TMB). PDL1 expression was stratified into high (>10) and low (<=10); TMB was stratified high (>=10) or low (<10). Genes and CNVs mutated in at least 5% of the population were used in univariate cox proportional hazards models predicting for overall survival (OS). Significant genes (p< 0.2) were then incorporated in a multivariate model adjusting for current biomarkers and demographic variables. OS analyses were conducted from the first immunotherapy to death or last follow-up date utilizing the cox proportions hazard model. Model performance before and after removing the significant genes was compared by concordance index with ANOVA. Results: In this cohort, the median follow-up time for patients alive was 17.5 months. The 2-year OS was 41.6% and the median time on treatment was 4.1 months. Gene mutations with a statistically significant association with OS in univariate analysis included: ERBB2 amplification, CDKN1A, CDKN2A.B loss, TP53, TERT promoter, ARID1A, FGFR3, CREBBP, and TRADD. In the multivariate analysis, low PDL1 expression (hazard ratio [HR] 1.82 95% confidence interval [CI] 1.26-2.64, p<0.005), TP53 mutation (HR 1.60, 95% CI 1.18-2.60, p<0.005), CDKN1A mutation (HR 1.66, 95% CI 1.06-2.61, p=0.02), and TERT mutation (HR 1.64, 95% CI 1.15-2.34, p=0.006) were significantly associated with decreased OS. Patients with FGFR3 wild type mutation were significantly associated with increased OS (HR 0.63, 95% CI 0.41-0.95 p=0.02). High TMB did not meet statistical significance for OS (HR 0.75, 95% CI 0.56-1.01, p=0.06). When all factors are included in the multivariable model the concordance index is 0.67. Conclusions: We report outcomes for veterans with BC treated with ICI. We highlight that PDL1 expression was associated with improved OS, while TMB did not meet statistical significance. We further highlight genomic variants that were associated with OS outcomes for patients with BC on ICI and propose a multivariable model. Current efforts are focused on training and validating machine learning algorithms to predict outcomes for patients with BC treated with ICI and ICI-combinations.