490 Background: Integrating molecular subtypes, gene transcripts associated with disease recurrence (DR), and clinicopathologic features may help risk stratify muscle-invasive bladder cancer (MIBC) patients & guide therapy selection. We hypothesized that combined transcriptomic & clinical data would improve risk stratification for DR (local or distant) after cystectomy +/- adjuvant chemotherapy. Methods: We identified 401 MIBC patients (pT2-4 N0-N3 M0) in The Cancer Genome Atlas with detailed demographic, clinical, pathologic, and treatment-related data. We split the data into training (60%) & testing (40%) sets. We produced RNA gene expression scores for molecular subtype using 48 established, relevant genes (PMID 28988769). In the training set, we performed feature selection by conducting random forest modeling of an additional 108 genes associated with DR. We kept genes of highest importance based on the evaluation of increasing mean-squared error & node purity. We excluded highly correlated genes & used the false discovery rate method for multiple hypotheses testing. We performed univariable analyses on genes of highest importance, molecular subtype, & clinicopathologic variables. Using adjusted multivariable analyses (MVA), we built two models: with & without transcriptomic data. Using the testing set, we compared the final models' performance to predict DR, using receiver operating characteristics & area under the curve (AUC). Results: Median follow-up was 18 months (range 1-168). 104 patients recurred with a 5-yr cumulative incidence of 34.6%[28.6-40.5%]. Using the training set, we identified 6 genes significantly associated with DR (VEGFA, TRMT1, FGFR2B, ERBB2, MMP14, PDGFC). The final MVA showed that the new 6-gene signature (HR 1.61, 95% CI 1.27-2.05, p < 0.001); immune molecular subtype [increased expression of PD-L1, PD-1, IDO1, CXCL11, L1CAM, SAA1] (HR 0.52, 95% CI 0.29-0.94, p = 0.03); smoking status (HR 1.17 per 10 pack-years, 95% CI 1.05-1.29, p = 0.005); and local failure risk factors [≥pT3 with negative margins & ≥10 nodes removed (HR 1.63, 95% CI 1.15-2.32, p = 0.006); ≥pT3 and positive margins OR < 10 nodes removed (HR 3.26, 95%CI 2.43 to 4.09, p = 0.007)], were all significantly associated with DR. This combined model outperformed a stand-alone clinicopathologic model (AUC 0.75 vs. 0.66) in the testing set. The combined model stratified patients based on DR risk into 3 groups with 5-yr cumulative incidences of 19.8%[7.7-31.9%] (low-risk); 34.5%[26.1-42.8%] (intermediate); and 49.8%[37.7-61.9%] (high), Gray’s Test p < 0.0001. Conclusions: To our knowledge, this study is the first to integrate clinicopathologic & transcriptomic information (including molecular subtype) to better stratify MIBC patients by risk of recurrence. This stratification may help guide decision-making for adjuvant treatment. Further validation is warranted.
Baseline characteristics of participants in pl-208: A multi-center trial of the prolung test™ (Transthoracic Bioconductance Measurement) as an adjunct to CT chest scans for the risk stratification of patients with pulmonary lesions suspicious for lung cancer
