Abstract

Background: Plasma circulating tumour DNA genotyping is widely adopted in advanced non-small cell lung cancer, and its performance is conventionally summarised as a concordance rate with tissue genotyping. Reported concordances of 85% to 97% are routinely cited in support of substituting plasma for tissue testing. Methods: Meta-analyses and cohorts reporting paired plasma and tissue genotyping with extractable accuracy data were eligible. Concordance, predictive values and post-test probabilities were derived analytically from pooled sensitivity and specificity across the plausible prevalence range. No new pooling was performed, since the interpretability of the pooled quantity is the object of the review. Analyses were performed in Python 3. Results: For the EGFR T790M resistance mutation, pooled analysis of 21 studies in 1,639 patients gave a sensitivity of 0.67 (95% CI 0.64-0.70) and specificity of 0.80 (0.77-0.83), with a positive predictive value of 0.85 (0.82-0.87) and a negative predictive value of 0.60 (0.56-0.63). Holding accuracy fixed, concordance varies only between 80% and 68% across prevalences from 5% to 90%, while negative predictive value falls from 98% to 21% over the same range. At a prevalence of 40%, per 100 patients tested, 27 true carriers are detected and 13 are missed, giving a concordance of 74.8%. A negative plasma result moves the probability of carrying the mutation from 40% only to 21.6%, consistent with a negative likelihood ratio of 0.46. In one real-world cohort, a raw concordance of 87.4% corresponded to a Cohen’s kappa of 0.60. Sensitivity across contributing sources ranged from 0.49 to 0.91 for nominally the same comparison. Conclusions: Concordance is a poor summary of liquid biopsy performance because it is dominated by agreement on negatives, varies little with the quantities that matter clinically, and does not answer the question for which it is cited. A negative plasma result does not exclude a targetable alteration: at realistic prevalence roughly one in five patients with a negative plasma test still carries the mutation. Reflex tissue genotyping after a negative plasma result remains necessary, and reporting should give sensitivity, predictive values and the assay detection limit rather than a concordance percentage.

Keywords: Liquid biopsy, Circulating tumour DNA, Concordance, Negative predictive value, EGFR, Non-small cell lung cancer