Klein · Annals of oncology : official journal of the European Society for Medical Oncology 2021 · Case-control diagnostic validation study · n=4077

Clinical validation of a targeted methylation-based multi-cancer early detection test using an independent validation set.

Cited 1004 times in the scientific literature.

Level 4 - case-series / case-control

Case-control diagnostic validation study using known cancer cases and non-cancer controls

PubMed 34176681 · doi:10.1016/j.annonc.2021.05.806 · record verified 2026-08-27

What was done

This pre-specified substudy evaluated a blood-based multi-cancer early detection (MCED) test combining targeted cell-free DNA (cfDNA) methylation sequencing with machine learning. The independent validation cohort included 4,077 participants (2,823 with cancer and 1,254 non-cancer controls whose status was confirmed at one-year follow-up) from the prospective Circulating Cell-free Genome Atlas study (CCGA; NCT02889978). Investigators measured test specificity, stage-specific sensitivity across more than 50 cancer types, and cancer signal origin (CSO) prediction accuracy.

What was found

Specificity for cancer signal detection was 99.5% (95% CI: 99.0% to 99.8%). Overall sensitivity was 51.5% (95% CI: 49.6% to 53.3%), scaling by stage: stage I at 16.8% (95% CI: 14.5% to 19.5%), stage II at 40.4% (95% CI: 36.8% to 44.1%), stage III at 77.0% (95% CI: 73.4% to 80.3%), and stage IV at 90.1% (95% CI: 87.5% to 92.2%). Sensitivity across stages I–III was 40.7% (95% CI: 38.7% to 42.9%) for all cancers and 67.6% (95% CI: 64.4% to 70.6%) for 12 pre-specified lethal cancers. CSO prediction accuracy in true positives was 88.7% (95% CI: 87.0% to 90.2%).

Why it matters

The study demonstrates that a cfDNA methylation test can detect multiple cancer types with high specificity and accurately identify the tumor site of origin, supporting its potential utility as an adjunct screening test.

Limits

The case-control design enrolled individuals with established cancer diagnoses rather than an unselected asymptomatic population, introducing spectrum bias. Sensitivity was low for stage I disease (16.8%). The study did not measure downstream clinical outcomes, mortality reduction, or the harms and costs of diagnostic workups.

Cited by