Researchers from Geneseeq and a network of Chinese academic hospitals have validated a blood test that can detect a broad range of cancers with high accuracy using cell-free DNA. A multi-cancer early detection (MCED) test identified cancer with 87.4% sensitivity and 97.8% specificity in an independent validation cohort, and it correctly predicted the tissue of origin around 83% of the time.

Early detection remains a critical challenge in cancer care. Current screening tools contribute to late diagnoses and poor outcomes, especially in cancers lacking established screening protocols.

Cell-free DNA (cfDNA) circulating in the bloodstream, shed by tumours, has emerged as a promising target for noninvasive detection. Sensitivity for early-stage and less common cancers has remained low, yet the non-invasive nature of the tests makes them a compelling area for improvement.

In the study, "Early detection of multiple cancer types using multidimensional cell-free DNA fragmentomics," published in Nature Medicine, researchers designed a whole-genome sequencing–based blood test to detect cancer signals and predict the tissue of origin using machine learning models trained on cfDNA fragmentation patterns.

Retrospective model training used data from 3,076 cancer patients and 3,477 noncancer controls. Validation involved an internal cohort of 1,746 participants and an independent cohort of 1,465 participants. An ongoing prospective analysis enrolled 3,724 asymptomatic individuals at two Chinese medical centres.

Researchers analysed plasma-derived cell-free DNA using low-coverage whole-genome sequencing.

Samples were processed under a double-blind protocol in which data analysts and clinical teams were separately blinded to clinical outcomes and molecular results. Sequencing data were input into two supervised machine learning classifiers: one trained to identify the presence of a cancer signal and the other to infer the tissue of origin.

Both models drew on multidimensional fragmentomics features, including cfDNA fragment size, copy number variation, nucleosome positioning, and inferred methylation profiles.

Bioinformatic pipelines were standardised across all phases. Model training and calibration occurred before validation began, using a fixed algorithm to prevent performance drift. All sequencing and analysis steps were conducted using the same laboratory procedures regardless of cohort or disease status.

In the independent validation cohort, the test achieved a sensitivity of 87.4% and a specificity of 97.8%. Sensitivity for early-stage cancers was 79.3% for stage I and 86.9% for stage II. Sensitivity reached 100% for liver and bile duct cancers, 94.5% for lung, 90.5% for ovarian, and 82.3% for colorectal. Sensitivity for pancreatic cancer was 76.9%, including 58.3% for stage I.

Tissue-of-origin prediction in the same cohort was accurate in 83.5% of cases based on the top-ranked prediction and 91.7% when the two most likely predictions were considered.

In the prospective Jinling cohort of 3,724 asymptomatic individuals, 43 cancer cases were identified within one year. The test detected 53.5% of these, with specificity at 98.1%. Sensitivity for the 13 targeted cancer types was 62.1%. Positive predictive value was 25%, and negative predictive value was 99.4%. Tissue-of-origin accuracy in this cohort was 63.2% for the top-ranked prediction and 89.5% when considering the top two.

Nearly half of the cancers detected by the test were not identified through standard screening or physical examination. High sensitivity for cancers typically identified late in the disease course, such as liver, ovarian, and pancreatic cancers, is extremely compelling, and the prediction of tissue origin adds further clinical relevance for early treatment.

Investigators conclude that the validation findings "... Indicate that the MCED test has strong potential to improve early cancer detection and support clinical decision-making."