Abstract

Background: Circulating tumour DNA (ctDNA) detected after curative-intent resection is proposed as a direct molecular readout of minimal residual disease, and is being used to select patients for adjuvant therapy and to design escalation and de-escalation trials. Reported hazard ratios for recurrence are strikingly large, and several tumour-specific meta-analyses now exist. Methods: An umbrella review was conducted per JBI methodology2. Reported hazard ratios were re-synthesised with a DerSimonian-Laird random-effects model on the log scale. Only estimates with a reported confidence interval were pooled; nothing was reconstructed. Prediction intervals, leave-one-out analysis and subgroup contrasts were computed. Analyses used Python 3. Results: Four systematic reviews were identified; three reported an interval and were pooled. Postoperative ctDNA positivity was associated with a pooled hazard ratio for recurrence of 6.80 (95% CI 5.41-8.55; z = 16.39, p < 0.001), with no detectable between-review heterogeneity (Q = 1.57, df = 2, p = 0.457; I² = 0.0%) and a 95% prediction interval of 4.11 to 11.24. Leave-one-out estimates ranged only from 5.94 to 7.00. Two findings qualify this. First, the hazard ratio was larger in earlier-stage disease—8.14 (5.60-11.82) in stage I-III colorectal cancer against 4.83 (3.64-6.39) in stage IV, ratio of hazard ratios 1.685 (95% CI 1.056-2.690, p = 0.029)—which reflects the lower baseline recurrence risk in early stage rather than greater biomarker information. Second, the pooled effect is roughly fourfold larger than that of typical tissue prognostic biomarkers, and the primary literature contains estimates such as a hazard ratio of 184.6 (95% CI 3.6-9577) from a 60-patient cohort, a signature of near-complete separation rather than of biomarker strength. Persistent ctDNA after adjuvant therapy carried a numerically but not significantly larger hazard than immediate postoperative ctDNA (10.59 versus 7.27; p = 0.291). Conclusions: Postoperative ctDNA is among the strongest prognostic markers reported in solid tumour oncology, with a pooled hazard ratio near 7 and remarkable consistency across tumour types. Its magnitude is plausible because ctDNA plausibly detects residual disease rather than correlating with it, but hazard ratio size depends on baseline risk and is inflated by small studies with near-complete separation. Critically, prognostic strength is not clinical utility: no estimate here shows that acting on ctDNA improves outcomes, and randomised evidence of ctDNAguided management is the necessary next step.

Keywords: Circulating tumour DNA, Minimal residual disease, Recurrence, Prognostic biomarker, Umbrella review, Liquid biopsy