Time-varying prognostic value of circulating tumor DNA in urothelial carcinoma: a systematic review and reconstructed patient-data meta-analysis.

Circulating tumor DNA (ctDNA) increasingly guides adjuvant therapy and surveillance in urothelial carcinoma (UC), yet whether its prognostic effect is adequately summarized by a single pooled hazard ratio (HR), which assumes proportional hazards (PH), has not been formally evaluated. We searched five sources for studies reporting ctDNA prognostic associations across non-muscle-invasive, muscle-invasive (MIBC), upper-tract and metastatic UC. HRs for unfavorable versus favorable ctDNA were pooled by Bayesian random-effects meta-analysis, with subgroups by stage and assay strategy. Where individual patient data could be reconstructed, PH was tested under four decision rules and restricted mean survival time (RMST), a PH-independent estimand, estimated at 12, 24 and 36 months. Twenty-five studies were synthesized qualitatively and 17 quantitatively. The pooled HR was 3.85 (95% credible interval 2.92 to 5.26; I² = 50.7%; prediction interval 1.67 to 9.38), and was higher in MIBC (4.89) than metastatic UC (2.52; p = 0.034) and for tumor-informed (5.46) than tumor-agnostic (2.68) assays (p = 0.009); neither survived multiplicity adjustment, and the latter is confounded with setting and sponsorship. PH was not supported in 5 of 11 cohorts under the pre-specified rule (2 to 5 across rules); pooled RMST loss in the unfavorable stratum was 2.2, 6.6 and 12.2 months at 12, 24 and 36 months (11, 8 and 5 cohorts), with certainty falling as the horizon lengthened. Unfavorable ctDNA was consistently associated with worse survival, particularly in perioperative MIBC, but non-proportional hazards in many cohorts indicate that its prognostic value should be reported with time-axis estimands alongside HRs.