The stakes are enormous. Cancer remains the second leading cause of death worldwide, claiming nearly 10 million lives each year, and annual new cases, currently around 19.3 million, are projected to exceed 35 million by 2050. Yet existing screening programs catch only a fraction of diagnoses: standardized protocols detect about 24 percent of cancers diagnosed annually in the United Kingdom, 28 percent in the United States, and 63 percent in China.

Tools such as low-dose computed tomography for lung cancer and the Pap smear for cervical cancer suffer from adherence problems, and computed tomography's high false-positive rate can trigger invasive follow-ups and patient anxiety. Liquid biopsy promises to overcome these limits by capturing tumor-derived molecules, cell-free DNA, cell-free RNA, proteins, circulating tumor cells, and extracellular vesicles, circulating in bodily fluids. But the review emphasizes that the choice of sample is far from trivial.

Blood, particularly plasma, dominates the field because it reaches virtually every tissue and serves as a common reservoir of tumor signals, making it the only biofluid used across all pan-cancer studies surveyed. Plasma, however, dilutes tumor signals massively: the body holds roughly 4 to 5 liters of it, tumor molecules reach only trace concentrations, and circulating tumor DNA has a half-life of roughly 16 minutes to 2.5 hours. Detection rates reflect these constraints, with one study reporting that ctDNA was found in 47 percent of stage I patients versus 82 percent of stage IV patients.

Anatomically proximal fluids can outperform blood by concentrating tumor signals near their source. In paired samples from early-stage colorectal cancer patients, urine-based trans-renal tumor DNA detected KRAS mutations with 37.7 percent sensitivity, significantly better than the 21.3 percent achieved by plasma ctDNA. In central nervous system lymphoma, cerebrospinal fluid reached a 100 percent detection rate compared with 78 percent in plasma.

The survey also revealed cancer-specific preferences: urine dominated bladder cancer studies, stool featured prominently in colorectal cancer work, and cervical exfoliated cells were the specimen of choice in cervical cancer research. The authors caution, however, that wide heterogeneity in cohort sizes, which ranged from 6 to nearly 9,000 participants, control designs, and biomarker types prevents any head-to-head ranking of biofluids, and they call for prospective studies comparing matched fluid panels directly. Once a sample is chosen, the molecular target matters enormously.

Somatic mutations in cell-free DNA provide direct evidence of driver alterations but are scarce in early disease and poorly localized to a tissue of origin. Methylation patterns, by contrast, are stable, tissue-specific, and arise earlier in carcinogenesis, which explains why they appear in nearly every cancer type surveyed and dominate pan-cancer test design. Fragmentomics, which reads the size and cleavage patterns of DNA fragments as a fingerprint of chromatin disorganization, requires no chemical conversion and suits low-pass whole-genome sequencing; five of seven lung cancer studies in the survey relied primarily on this approach.

Proteins are cheap to measure on mature platforms but lack early-stage sensitivity, while circulating tumor cells are too rare in early disease to be useful beyond late-stage prognosis. Extracellular vesicles, RNA, metabolites, and even microbiome DNA each add distinct biological information but suffer from standardization gaps. The review argues that effective multi-omics integration depends on biological complementarity rather than simply stacking markers.

A breast cancer study using the SPOT-MAS approach combined methylomics and fragmentomics, reaching an area under the curve of 0.91 with 65 percent sensitivity and 96 percent specificity in early-stage disease, outperforming single-omic methods hampered by low ctDNA shedding. But integration carries a specificity cost. In a prospective colorectal screening study of nearly 10,000 average-risk participants, a multitarget stool DNA test detected 92.3 percent of cancers versus 73.8 percent for the fecal immunochemical test alone, yet its specificity dropped from 94.9 percent to 86.6 percent, which extrapolates to roughly 760 additional false positives per 10,000 people screened.

Model construction introduces its own traps. The authors stress that control group design must match the intended clinical use: pan-cancer screening in asymptomatic populations demands controls that reflect real-world complexity, as GRAIL's CCGA program did by including cancer-free participants and women undergoing routine mammography, while single-cancer triage tests must include the specific confounders that create diagnostic ambiguity, such as smokers with inflammatory lung nodules, patients with hematuria from stones, or, in prostate cancer, indolent Gleason 3+3 lesions assigned deliberately to the control group. On the algorithmic side, high-throughput platforms generate millions of candidate features, inviting overfitting.

The CCGA2 study distilled roughly 30 million genomic CpG sites down to about 1 million highly informative sites, enabling deeper sequencing and stable classification across more than 50 cancer types. The authors also recount a cautionary tale: a 2002 Lancet study reporting near-perfect ovarian cancer detection from serum proteomic patterns collapsed on independent reanalysis because its model had learned batch effects rather than tumor biology, a vivid illustration of why data leakage prevention and blind validation are foundational. The steepest gradient in the translational pathway is the drop in performance from case-control studies to real-world screening.

GRAIL's CCGA3 case-control study reported 51.5 percent overall sensitivity, but this fell to approximately 29 percent in the prospective PATHFINDER cohort of asymptomatic adults. The MERCURY-based CanScan test achieved 87.4 percent sensitivity in its validation set yet only 53.5 percent in the real-world prospective JinLing cohort. Even strong real-world performance does not guarantee benefit: the review cites South Korea's ultrasound screening program for thyroid cancer, which produced a 15-fold increase in incidence with no corresponding mortality decline, a warning that early detection without randomized trial evidence can fuel overdiagnosis.

Currently, no single-cancer liquid biopsy test has randomized controlled trial evidence, and the only large-scale RCT, the NHS-Galleri trial for multi-cancer detection, has not yet reported its primary endpoint. Progress nonetheless varies by cancer type. Colorectal cancer leads in clinical maturity: the ECLIPSE trial compared a blood-based test against colonoscopy in more than 20,000 participants, met prespecified performance criteria, and the Shield test was approved by the FDA in 2024, while the PREVENT trial validated the stool RNA test ColoSense in nearly 9,000 people.

Liver cancer benefits from well-defined high-risk groups, enabling the aMAP-2 prospective study of over 13,000 participants, and lung cancer research leverages existing low-dose CT screening cohorts, as the DELFI assay has done. Beyond laboratory science, the review identifies four systemic bottlenecks that determine whether validated tests reach patients: cost-effectiveness, with a US analysis of the Galleri test estimating roughly $66,000 per quality-adjusted life year but flagging unmodeled budget impacts; assay standardization, where reference materials lag behind rapidly proliferating biomarker platforms; fragmented regulatory pathways across jurisdictions; and health-system capacity, illustrated starkly by a Chinese program in which only about 22 percent of people with positive stool screens completed follow-up colonoscopy. The authors conclude that closing the gap between technological innovation and clinically impactful screening will require coordinated action among developers, regulators, payers, and providers, extending well beyond the laboratory into health economics and policy.

Subject of Research: Translational pathway of liquid biopsy for early cancer screening, from biofluid selection and biomarker modeling to clinical validation

Article Title: Clinical translation of liquid biopsy in early cancer screening: biofluid selection, model construction, and clinical validation

Article References: Cui, P., Xu, S., Lin, X., Song, S., Huang, B., & Xu, Z. (2026). Clinical translation of liquid biopsy in early cancer screening: biofluid selection, model construction, and clinical validation. Frontiers in Oncology, 16, Article 1946352. https://doi.org/10.3389/fonc.2026.1946352

Image Credits: AI Generated

DOI: 10.3389/fonc.2026.1946352

Keywords: liquid biopsy, early cancer detection, multi-cancer early detection, cell-free DNA, methylation, fragmentomics, clinical validation, biomarkers, plasma, screening trials, overdiagnosis, health economics