Time-varying prognostic value of circulating tumor DNA in urothelial carcinoma: A systematic review and reconstructed patient-data meta-analysis.
Shugo Yajima, Gaku Okumura, Soichiro Yoshida et al.
✦ AI-curated · Sources linked
In 30 seconds
This systematic review and reconstructed patient-data meta-analysis assessed the prognostic association of circulating tumor DNA (ctDNA) across non-muscle-invasive, muscle-invasive, upper-tract and metastatic urothelial carcinoma. Twenty-five studies were reviewed qualitatively and 17 contributed to quantitative analyses. Unfavorable versus favorable ctDNA was associated with worse survival, with a pooled hazard ratio of 3.85 (95% credible interval 2.92-5.26). Proportional hazards were not supported in 5 of 11 cohorts under the prespecified rule, suggesting that a single pooled hazard ratio may not fully describe the association over time.
Key findings
- The pooled hazard ratio for unfavorable versus favorable ctDNA was 3.85 (95% credible interval 2.92-5.26; I² = 50.7%; prediction interval 1.67-9.38).
- The hazard ratio was higher in MIBC than metastatic UC (4.89 vs 2.52; p = 0.034), but this difference did not survive multiplicity adjustment.
- Tumor-informed assays had a higher hazard ratio than tumor-agnostic assays (5.46 vs 2.68; p = 0.009); this difference did not survive multiplicity adjustment and was confounded by setting and sponsorship.
- Pooled RMST loss in the unfavorable ctDNA stratum was 2.2, 6.6 and 12.2 months at 12, 24 and 36 months, respectively, with certainty declining at longer horizons.
Why it matters
The findings support ctDNA as a prognostic marker across urothelial carcinoma settings, while showing that its association with survival may change over time. When proportional hazards do not hold, a single hazard ratio can obscure the timing and magnitude of prognostic differences. Reporting time-specific measures such as restricted mean survival time alongside hazard ratios may therefore give clinicians a more complete account of the evidence.
What to keep in mind
The analysis reconstructed individual patient data where possible, and proportional-hazards findings varied across cohorts and decision rules. RMST estimates drew on fewer cohorts at longer horizons, with certainty declining over time.