Using ctDNA to Predict Recurrence After Surgery in Stage II–III Colorectal Cancer: A Systematic Review and Narrative Meta-Analysis

Colorectal cancer (CRC) is a leading cause of cancer-related mortality, with 20–30% of patients experiencing recurrence after surgery. Circulating tumor DNA (ctDNA) has emerged as a minimally invasive biomarker for detecting minimal residual disease (MRD) and predicting recurrence. This systematic review and narrative meta-analysis evaluates the prognostic value of postoperative ctDNA in stage II-III CRC and its ability to guide adjuvant therapy. A systematic review identified 14 studies that reported ctDNA measurements after surgery along with recurrence outcomes, disease-free survival (DFS), and overall survival (OS). Both tumor-informed and tumor-agnostic ctDNA assays were included, with postoperative sampling ranging from 3 days to 24 months. Hazard ratios (HRs) and recurrence outcomes were synthesized at the study level using a narrative meta-analytic approach, emphasizing the direction, magnitude, and consistency of effects rather than formal statistical pooling. ctDNA performance was compared with traditional risk markers, such as tumor stage. Subgroup analyses assessed longitudinal ctDNA dynamics and adjuvant chemotherapy guidance. The results showed that postoperative ctDNA positively ranged from 8–20% and strongly predicted recurrence. ctDNA outperformed the conventional markers. ctDNA-guided therapy reduced unnecessary chemotherapy in low-risk patients without compromising recurrence-free survival. Combining ctDNA with other factors improved predictive accuracy. Overall, postoperative ctDNA is a robust predictor of recurrence in stage II-III CRC, enabling early detection of MRD and more personalized adjuvant therapy. Standardization of assays, testing intervals, and reporting methods is required before routine clinical adoption. Because of substantial clinical and methodological heterogeneity across studies, results were synthesized narratively in accordance with PRISMA 2020 guidelines, and conventional forest plots were not generated.