How accurate is circulating tumor DNA (ctDNA) for detecting cancer recurrence compared with CT imaging, and what are its false positive and false negative rates?

The short version: ctDNA can detect recurrence earlier than CT in some cancers, but its false negative rate is high enough that a negative result does not rule out recurrence, and the evidence is almost entirely from manufacturer-funded or small academic studies.

The best single-number summary comes from a 2023 meta-analysis of 39 studies covering 2,774 patients across solid tumors [5]. At a single time point (landmark testing), ctDNA detected recurrence with about 58% sensitivity and 92% specificity. That means it missed 42 out of every 100 recurrences. When testing was done serially over time, sensitivity rose to about 82% and specificity to 94% [5]. The false positive rate across both approaches was between 6% and 8% [5]. The meta-analysis found that tumor-informed panels (where the test is customized to the patient's own tumor mutations) performed far better than tumor-agnostic panels: odds ratios of 42.8 versus 7.0 [5]. That matters because most commercial tests are tumor-informed, but the cost is high and the turnaround time is days to weeks.

The numbers vary sharply by cancer type. In endometrial cancer, a 2026 study found 71% sensitivity and 96% specificity in a cohort of only 39 patients [1]. In colorectal cancer, the evidence is stronger: ctDNA often rises before CT detects anything, sometimes by months [6]. In head and neck cancer, a 2024 study found that ctDNA from lymphatic fluid (not blood) had 67-75% sensitivity, while plasma ctDNA had only 33% [2]. In liver transplant patients, a tumor-informed assay (Signatera) showed 75% sensitivity and 100% specificity, but the confidence interval on sensitivity was enormous (19-99%) because only 6 recurrences occurred [3]. In breast cancer, a prediction model hit 100% sensitivity in a validation set but only 50% positive predictive value, meaning half the positive results were wrong [4].

The evidence has serious structural problems. Most studies are small, single-center, and observational. The meta-analysis itself notes that the quality of studies was variable and that sensitivity in the better-designed studies was actually lower (48% at landmark) [5]. The meta-analysis also found that tumor-agnostic panels, which are cheaper and faster, had much worse performance [5]. Clonal hematopoiesis (benign mutations from white blood cells) can produce false positives, and many studies did not filter for it [7]. The lead time advantage (ctDNA rising before CT shows anything) is real but varies wildly: a median of about 5 months in the meta-analysis [5], but as little as a few weeks in some studies.

My call: ctDNA is a useful adjunct to CT for recurrence surveillance in cancers where the evidence is strongest (colorectal, some gynecologic), but a negative ctDNA does not rule out recurrence, and the false positive rate means a positive result requires imaging confirmation. The evidence is too thin and too conflicted by manufacturer funding to recommend ctDNA as a replacement for CT. Confidence: moderate for colorectal cancer, low for most other solid tumors.