Diagnostic accuracy of circulating tumor DNA for early detection of colorectal cancer: A systematic review and meta-analysis.

Abstract

34 Background: Blood-based circulating tumor DNA (ctDNA) assays represent a promising non-invasive method for colorectal cancer (CRC) surveillance. To assess their diagnostic potential, we conducted a systematic review and meta-analysis to evaluate the accuracy of ctDNA assays in detecting early-stage CRC among asymptomatic adults. This meta-analysis focused on diagnostic accuracy metrics to provide robust evidence for their clinical utility. Methods: We included prospective and cross-sectional diagnostic accuracy studies of blood ctDNA tests compared against colonoscopy/histopathology. We extracted study-level 2×2 data (true positives, true negatives, false positives, false negatives) and assay type for each included study and computed sensitivity and specificity from raw counts. No continuity correction was required. Sensitivity and specificity were transformed to the logit scale; pooled point estimates were obtained from a Reitsma bivariate random-effects model, while 95% confidence intervals were derived from separate univariate random-effects meta-analyses on the logit scale (REML) and back-transformed to proportions to provide robust CIs for Se and Sp. For studies reporting AUC, per-study AUC standard errors were estimated using the Hanley & McNeil approximation and pooled on the logit(AUC) scale. Risk of bias was assessed using QUADAS-2 and certainty of evidence was evaluated with GRADE. Results: Ten studies met inclusion criteria. Reitsma pooled point estimates were Sensitivity 0.8865 (95% CI 0.8481–0.9200) and Specificity 0.9127 (95% CI 0.8974–0.9037). Pooled AUC = 0.9748 (95% CI 0.9461–0.9884; n = 5). DOR = 85.49 (95% CI ≈46.8–156.1; I² ≈ 79.8%), PLR = 9.30 (95% CI ≈7.3–11.9; I² ≈ 55.3%), NLR = 0.123 (95% CI ≈0.089–0.172; I² ≈ 64.3%). Pooled observed PPV = 0.7669 (95% CI 0.3790–0.9466; I² ≈ 99.2%) and NPV = 0.9669 (95% CI 0.8817–0.9913; I² ≈ 98.1%). Predicted PPV/NPV using pooled Se/Sp: at 0.5% prevalence PPV ≈ 4.9%, NPV ≈ 99.94%; at 1% PPV ≈ 9.3%, NPV ≈ 99.87%; at 2% PPV ≈ 17.2%, NPV ≈ 99.75%. Subgroup pooled sensitivities by assay type were 0.873 (A), 0.900 (B), and 0.926 (C); meta-regression showed no significant differences (sensitivity: B vs A p = 0.885, C vs A p = 0.179; FPR: Bvs A p = 0.484, C vs A p = 0.171). Conclusions: Blood-based ctDNA assays demonstrate high pooled sensitivity (~0.89), specificity (~0.91) and excellent discrimination (pooled AUC ≈ 0.975). PLR and NLR indicate meaningful post-test shifts, but substantial heterogeneity especially for DOR and PPV/NPV and low expected PPV at screening prevalence limit immediate population-level adoption. Given the findings, ctDNA shows strong potential as a screening option for CRC alongside FIT and Cologuard, though larger, standardized, population-based studies are required.