Metastatic pattern accounts for the apparent predictive value of immune gene signatures in urothelial carcinoma treated with checkpoint inhibitors

Abstract Purpose: Transcriptomic immune signatures are widely proposed as predictors of response to immune checkpoint inhibitors in metastatic urothelial carcinoma, but are almost always evaluated against chance rather than against the markers a clinician already has. We asked what such a signature adds to information that is already measured. Methods: We analyzed IMvigor210 ( n = 348 with pre-treatment tumor RNA sequencing; 298 evaluable for response). Six published immune signatures and two pre-specified control signatures were compared with two baselines: routine clinical variables (metastatic pattern, ECOG performance status, prior platinum), and those plus PD-L1 immunohistochemistry and tumor mutational burden. Discrimination was estimated as out-of-fold area under the curve over 50 repeats of stratified five-fold cross-validation; incremental value was tested by likelihood-ratio test. Overall survival was analyzed by Cox regression with Harrell’s C-index. Results: The clinical baseline reached an AUC of 0.697 (95% CI 0.619–0.767) and the extended baseline 0.778 (0.708–0.844). Tumor mutational burden alone reached 0.720 (0.639–0.796) and exceeded every immune signature in a paired comparison (ΔAUC −0.122 to −0.189). No immune signature added measurable information to either baseline (ΔAUC ≤ +0.003; likelihood-ratio p = 0.06–0.99), whereas a proliferation control did. Metastatic pattern explained 6.8–11.0% of signature variance and accounted for nearly all of the attenuation seen on adjustment. Conclusion: Established immune signatures carried no detectable incremental value over markers already measured in this cohort.