Bayesian Secondary Reanalysis of Published Aggregate Data on Circulating Tumor DNA for Molecular Residual Disease Detection Across Cancer Types

Background Circulating tumor DNA (ctDNA)-based molecular residual disease (MRD) testing may help estimate recurrence risk after treatment. We performed a Bayesian secondary reanalysis of published aggregate data to estimate prognostic and diagnostic performance. Methods Data were extracted from Zheng et al. (80 studies, 11 cancer types). Log hazard ratios for recurrence were analysed using Bayesian normal-normal hierarchical models. Sensitivity and specificity were analysed using separate logit-scale random-effects models. Posterior probabilities were estimated from posterior samples. Results The prognostic analysis included 109 study arms (approximately 9,980 patients). The pooled hazard ratio for recurrence associated with ctDNA positivity was 7.5 (95% credible interval [CrI], 6.4-8.8). Among cancer-type subgroups reported separately, all 95% CrIs excluded 1.0. In 70 diagnostic studies, pooled sensitivity was 57.6% (95% CrI, 53.2-61.9) and specificity was 90.5% (95% CrI, 88.5-92.3). Longitudinal monitoring had higher sensitivity than landmark testing (73.6% vs. 50.0%; posterior probability >99.9%), with similar specificity (89.9% vs. 90.8%). mPCR-NGS had specificity of 93.2% (95% CrI, 90.2-95.6); in separate exploratory comparisons, P(mPCR-NGS > ddPCR) and P(mPCR-NGS > hybridization capture NGS) were both 96.0%. Conclusions This Bayesian secondary reanalysis supports the association between ctDNA-MRD positivity and recurrence risk. Longitudinal monitoring may improve sensitivity, while assay technology comparisons should be interpreted cautiously because they were based on aggregate, non-head-to-head data.