Skip to content

Hepatocellular carcinoma — biomarkers

TL;DR — AFP is the only broadly deployed HCC biomarker, useful as an adjunct to surveillance, a tumor-biology marker, and a longitudinal response signal, but neither sensitive nor specific enough to diagnose HCC alone (Tzartzeva 2018, PMID 29425931). GALAD combines age, sex, AFP, AFP-L3, and DCP and often outperforms individual markers in case-control and multicentre studies, yet prospective surveillance utility remains unproven (Yang 2019, PMID 30464023; Huang 2022, PMID 34679250). AFP ≥400 ng/mL is a validated predictive enrichment marker for ramucirumab after sorafenib in REACH-2 (Zhu 2019, PMID 30665869). PD-L1, tumor mutational burden, Wnt/CTNNB1 status, circulating tumor DNA, and immune signatures are promising but do not currently select routine checkpoint therapy (Zhou 2023, PMID 36852452). Every candidate must be tied to a use—risk, detection, diagnosis, prognosis, response, or treatment selection—because high discrimination in one setting does not establish utility in another.

Biomarker-use taxonomy

Use Question Required study design
Risk Who will develop HCC? Longitudinal at-risk cohort
Early detection Is preclinical HCC present now? Prospective repeated-sampling surveillance cohort
Diagnosis Is an observed lesion HCC? Blinded comparison with pathology/validated imaging
Prognosis What outcome occurs regardless of therapy? Representative treated/untreated cohort
Predictive selection Does marker modify relative treatment benefit? Randomized treatment-by-marker interaction
Pharmacodynamic Is the drug hitting biology? Serial paired samples
Minimal residual disease Is microscopic cancer present after cure? Post-treatment longitudinal cohort with lead-time analysis

Reusing a case-control diagnostic AUC as proof of surveillance benefit is a category error.

AFP

AFP rises in some HCCs but also in active hepatitis, regeneration, pregnancy, and germ-cell tumors. Many early HCCs do not produce AFP.

In surveillance meta-analysis, ultrasound alone had approximately 47% sensitivity for early HCC, increasing to about 63% when AFP was added, with reduced specificity (Tzartzeva 2018, PMID 29425931).

AFP use Strength Limitation
Surveillance adjunct Improves sensitivity over ultrasound alone More false-positive diagnostic work-up
Longitudinal trend Patient acts as own baseline No universally validated slope rule
Prognosis/transplant Higher values reflect adverse biology in many cohorts Threshold and etiology dependence
Treatment response Rapid fall can support response Non-secretory tumors; liver inflammation
Ramucirumab selection AFP ≥400 ng/mL prospectively selected in REACH-2 Applies to specific drug/setting (Zhu 2019, PMID 30665869)

AFP should not overrule definitive imaging in an at-risk patient or prove HCC in a non-at-risk patient.

AFP-L3 and DCP/PIVKA-II

AFP-L3 is a glycoform fraction associated with malignant hepatocyte biology; DCP/PIVKA-II reflects abnormal prothrombin production. Their availability and integration differ by region.

Marker Potential role Confounder/boundary
AFP-L3 Early detection and prognosis Requires measurable AFP/assay-specific interpretation
DCP/PIVKA-II Detection, vascular invasion, prognosis Vitamin K status and anticoagulation
Combined panels Complementary sensitivity Threshold, cost, and false positives

These markers contribute to GALAD but do not independently establish population benefit.

GALAD and GALADUS

GALAD combines Gender, Age, AFP-L3, AFP, and DCP. GALADUS adds ultrasound.

In a US/German study, GALAD discriminated HCC better than ultrasound and the combined GALADUS model was proposed (Yang 2019, PMID 30464023). A Chinese multicentre study included 602 HCC cases and 923 controls, with 34.1% of cases at BCLC 0–A, and evaluated cross-sectional and longitudinal performance (Huang 2022, PMID 34679250).

Translation issues include:

  • case-control spectrum inflating apparent accuracy;
  • differing HBV, HCV, alcohol, and MASLD mix;
  • assay platform and cutoff;
  • whether the score precedes imaging detectability;
  • how a positive result triggers imaging;
  • cumulative false positives during repeated testing;
  • cost and assay access.

Detection panels and network comparisons

A network meta-analysis compared a multitarget blood test with ultrasound and AFP, suggesting potential performance gains but relying on indirect comparisons and heterogeneous study designs (Singal 2022, PMID 35945907). Such evidence ranks diagnostic accuracy, not mortality benefit.

An ideal prospective study repeatedly collects samples before diagnosis, keeps laboratory readers blind, adjudicates interval cancers, and reports stage shift, diagnostic procedures, overdiagnosis, and mortality—not only AUC.

Imaging biomarkers

Imaging already contains biomarker-like information:

  • arterial enhancement and washout;
  • capsule and threshold growth;
  • diffusion restriction;
  • hepatobiliary-phase signal;
  • vascular invasion;
  • radiographic response;
  • ultrasound visualization quality.

Radiomics and artificial intelligence can quantify texture or spatial heterogeneity, but scanner/protocol effects and dataset shift threaten reproducibility. A radiomic score is not clinically valid until locked and externally tested.

Tissue biomarkers

Candidate Biological rationale Current status
PD-L1 Checkpoint pathway engagement Inconsistent assay/cutoff; not routine selector
Wnt/CTNNB1 Immune exclusion and anti-PD-1 resistance mechanism Retrospective/preclinical; not exclusion rule
Immune class Integrates infiltrate and signaling Taxonomy and assay not standardized
TMB Neoantigen proxy Generally low in HCC; no validated cutoff
Gene-expression signatures Proliferation or immune state Tissue and platform dependence
TERT/TP53/CTNNB1 Common drivers/classes Prognostic/mechanistic, not routine drug selectors

A systematic review/meta-analysis of PD-L1 in HCC found heterogeneous methods and associations, insufficient for a universal predictive cutoff (Zhou 2023, PMID 36852452).

Wnt/CTNNB1 as a treatment biomarker

β-catenin activation caused immune escape and anti-PD-1 resistance in experimental HCC, with supportive human correlates (Ruiz de Galarreta 2019, PMID 31186238). The evidence does not yet show that CTNNB1-mutant patients derive no benefit from every modern IO combination.

Validation requires:

  1. prespecified assay and cutoff;
  2. randomized therapy comparison;
  3. treatment-by-marker interaction;
  4. independent replication;
  5. demonstration that using the marker improves decisions.

AFP as a predictive marker: REACH-2

REACH-2 is the clearest HCC example of biomarker-selected systemic therapy. It enrolled patients with AFP ≥400 ng/mL after sorafenib and showed an overall-survival benefit for ramucirumab over placebo (Zhu 2019, PMID 30665869).

This result proves efficacy in an enriched population; it does not establish that AFP is a direct mechanistic measure of VEGFR2 dependence. The label “predictive” is clinically operational but mechanistic interpretation remains open.

Circulating tumor DNA

ctDNA can encode mutations, copy-number changes, fragmentation, and methylation. Potential HCC uses include:

  • early detection;
  • non-invasive molecular profiling;
  • response kinetics;
  • minimal residual disease after resection/ablation;
  • recurrence detection;
  • resistance evolution.
Stage ctDNA challenge
Early localized HCC Low tumor fraction
Cirrhosis surveillance Clonal/background cfDNA and repeated false positives
Advanced disease Easier detection but heterogeneity across sites
Post-treatment MRD Need to distinguish transient clearance from cure
Transplant Donor/recipient cfDNA and rejection confounding

Contemporary reviews find MRD signals promising but studies remain small, assay-diverse, and mostly observational (Galli 2025, PMID 40058162).

Dynamic biomarkers

Longitudinal change can outperform a single value when baseline varies by patient. Candidate dynamics include:

  • AFP velocity or nadir;
  • ctDNA clearance;
  • radiographic viable-tumor volume;
  • liver-function trajectory;
  • immune-cell or cytokine changes.

Dynamic markers are vulnerable to guarantee-time bias: a patient must survive and remain on treatment long enough to show a change.

Assay-readiness checklist

Criterion Question
Analytical validity Is the analyte measured reproducibly?
Clinical validity Does it predict the intended state/outcome?
Calibration Are probabilities correct in the target population?
Incremental value Does it improve on existing clinical/imaging variables?
Clinical utility Does acting on it improve outcomes?
Equity Does performance hold across etiologies, regions, sex, ancestry, and access?
Feasibility Can it return in time and at acceptable cost?

Phase matters more than AUC

Biomarker development should progress from case-control separation to prospective-specimen, blinded evaluation in the intended-use population and then to clinical utility. Case-control studies inflate performance because established cancers are biologically louder and controls are cleaner than real surveillance populations. Phase-3 studies now show that threshold choice and time before diagnosis materially change results.

Test/study Quantified performance Boundary
Longitudinal GALAD At 90% specificity, sensitivity 66.7% versus AFP 40.5%; early-HCC sensitivity 69.2% 397-patient cohort with 42 HCC; needs program-level validation (Singal 2022, PMID 34618932)
GALAD phase 3 Seven-center prospective-specimen validation Performance was lower than phase-2 case-control reports, demonstrating spectrum effects (Marsh 2025, PMID 39293548)
HES V2.0 versus GALAD In 2,331 patients/125 HCC, HES V2.0 TPR exceeded GALAD by 7.2% overall at fixed 10% FPR Increment depends on longitudinal gradients and assay availability (El-Serag 2025, PMID 38899967)
External HEDS validation 1,485 patients/119 HCC; HES V2.0 and GALAD AUROC both 0.79 Similar AUC can conceal different sensitivity at a chosen false-positive rate (El-Serag 2026, PMID 41043723)
AFP+AFP-L3+DCP Meta-analysis of 13 studies Combination improves discrimination but heterogeneity and diagnostic—not surveillance—sampling limit inference (Wang 2020, PMID 33015170)
NAFLD early HCC DCP sensitivity/AUC 0.60/0.74 versus AFP 0.34/0.59 Etiology changes marker performance; GALAD/DCP comparisons remain heterogeneous (Li 2023, PMID 37929312)
cfDNA overall 38-study meta-analysis: sensitivity 0.54, specificity 0.90, AUC 0.82 versus healthy controls Healthy-control comparison is not the clinical surveillance differential (Zhang 2021, PMID 33470842)
PD-L1 ORR OR 1.86 (95% CI 1.35–2.55) for PD-L1 positive versus negative across 11 studies/1,330 patients Enrichment is insufficient for exclusion because responses occur in PD-L1-negative tumors (Yang 2023, PMID 36965092)

Cell-free DNA and minimal residual disease

A blinded 247-person validation of a multi-analyte cfDNA/protein/clinical test reported AUROC 0.944, but included 122 HCC cases and 125 chronic-liver-disease controls rather than a longitudinal surveillance cohort (Lin 2022, PMID 35244350). A 649-person prospective case-control methylation study likewise supports analytical promise while retaining case-control spectrum bias (Rammohan 2025, PMID 40485822). Pooled ctDNA studies combine methylation, mutation, fragmentation, and concentration assays, so a single summary sensitivity masks materially different technologies (Li 2022, PMID 36348665).

MRD is a different intended use. A personalized tumor-informed assay in 88 resected patients associated postoperative ctDNA with recurrence risk, while a tumor-naïve 32-patient prospective study combined copy-number, fragment, and AFP signals (Hu 2025, PMID 39526910; Ren 2024, PMID 39704423). Neither establishes that acting on molecular recurrence improves survival. The required trial randomizes an intervention at molecular positivity, not merely reports a prognostic hazard.

Imaging biomarkers and treatment selection

Radiomics studies report c-statistics of 0.66–0.95 for differentiating HCC from other lesions and promising microvascular-invasion prediction, but a 54-study review found inconsistent reporting and methodological standardization (Harding-Theobald 2021, PMID 34390014). For histological grade, 11 studies/2,245 patients included no prospective designs and only two external test cohorts (Wang 2023, PMID 37541183). These models are candidate biomarkers, not clinical-grade substitutes for pathology.

AFP is the only validated treatment-selection biomarker in advanced HCC through REACH-2, where AFP ≥400 ng/mL defined ramucirumab eligibility (Zhu 2019, PMID 30665869). PD-L1, tumor mutational burden, Wnt status, etiology, radiomics, and ctDNA remain exploratory. Osteopontin and other protein markers may outperform AFP in selected diagnostic meta-analyses, but lack standardized prospective surveillance thresholds (Wan 2014, PMID 25034355).

Controversies

  • Replacement versus augmentation. Blood panels could overcome poor ultrasound visualization, but a non-localizing positive test still triggers imaging; utility depends on incremental sensitivity at an acceptable false-positive rate (Marsh 2025, PMID 39293548; El-Serag 2026, PMID 41043723).
  • Fixed versus longitudinal thresholds. Within-person change may detect cancer earlier and reduce between-person noise, but requires reliable prior samples and complicates implementation (Singal 2022, PMID 34618932).
  • Prognostic versus predictive. A marker associated with poor survival is not necessarily a marker of differential treatment benefit. AFP for ramucirumab is predictive-enrichment evidence; PD-L1 is not yet exclusionary (Zhu 2019, PMID 30665869; Yang 2023, PMID 36965092).
  • MRD actionability. ctDNA can stratify recurrence, but no mature randomized evidence shows improved outcomes from treating molecular positivity before radiologic recurrence (Hu 2025, PMID 39526910).

Open questions

  • Can a blood panel improve early-stage detection and mortality in prospective cirrhosis surveillance rather than only case-control AUC (Huang 2022, PMID 34679250)?
  • Is CTNNB1 predictive for specific IO regimens after randomized interaction testing (Ruiz de Galarreta 2019, PMID 31186238)?
  • Can post-resection ctDNA-guided adjuvant therapy improve survival (Galli 2025, PMID 40058162)?
  • What longitudinal AFP rule balances early detection and false positives?
  • Why does AFP enrich ramucirumab benefit, and can a more mechanistic marker improve selection (Zhu 2019, PMID 30665869)?

References

  1. Tzartzeva K, et al. Surveillance imaging and alpha fetoprotein for early detection of HCC in patients with cirrhosis: a meta-analysis. Gastroenterology. 2018;154:1706-1718.e1. PMID 29425931
  2. Yang JD, et al. GALAD score for HCC detection in comparison with liver ultrasound and proposal of GALADUS. Cancer Epidemiol Biomarkers Prev. 2019;28:531-538. PMID 30464023
  3. Huang C, et al. Validation of the GALAD model for early diagnosis and monitoring of HCC in a Chinese multicenter study. Liver Int. 2022;42:210-223. PMID 34679250
  4. Singal AG, et al. Comparison of a multitarget blood test to ultrasound and AFP for HCC surveillance: network meta-analysis. Hepatol Commun. 2022;6:2925-2936. PMID 35945907
  5. Zhu AX, et al. Ramucirumab after sorafenib in advanced HCC and increased AFP (REACH-2). Lancet Oncol. 2019;20:282-296. PMID 30665869
  6. Zhou X, et al. Evaluation of PD-L1 as a biomarker for immunotherapy for HCC: systematic review and meta-analysis. Liver Cancer. 2023. PMID 36852452
  7. Ruiz de Galarreta M, et al. β-Catenin activation promotes immune escape and resistance to anti-PD-1 therapy in HCC. Cancer Discov. 2019;9:1124-1141. PMID 31186238
  8. Galli E, et al. Circulating blood biomarkers for minimal residual disease in hepatocellular carcinoma: a systematic review. Cancer Treat Rev. 2025;135:102908. PMID 40058162
  9. Wang X, et al. Combined AFP, AFP-L3, and DCP for HCC diagnosis: a meta-analysis. Biomed Res Int. 2020;2020:5087643. PMID 33015170
  10. Marsh TL, et al. Phase-3 biomarker validation of GALAD for HCC detection in cirrhosis. Gastroenterology. 2025;168:316-326.e6. PMID 39293548
  11. El-Serag HB, et al. HES V2.0 outperforms GALAD for HCC detection. Hepatology. 2025;81:465-475. PMID 38899967
  12. Tayob N, et al. Phase-3 performance of AFP, AFP-L3, and DCP for HCC detection. Clin Gastroenterol Hepatol. 2023;21:415-423.e4. PMID 35124267
  13. El-Serag HB, et al. HES V2.0 validation compared with GALAD and ASAP in HEDS. J Hepatol. 2026;84:578-586. PMID 41043723
  14. Singal AG, et al. GALAD in a prospective cirrhosis surveillance cohort. Hepatology. 2022;75:541-549. PMID 34618932
  15. Lin N, et al. A multi-analyte cell-free-DNA blood test for early HCC detection. Hepatol Commun. 2022;6:1753-1763. PMID 35244350
  16. Rammohan A, et al. Multiplex hypermethylation blood test to detect HCC. J Clin Exp Hepatol. 2025;15:102578. PMID 40485822
  17. Hu J, et al. Personalized perisurgical ctDNA MRD assessment in HCC. Clin Cancer Res. 2025;31:1047-1056. PMID 39526910
  18. Ren Q, et al. Tumor-naïve MRD detection using ctDNA features and AFP. Cancer Med. 2024;13:e70511. PMID 39704423
  19. Li JJ, et al. Diagnostic performance of ctDNA for HCC: a meta-analysis. PeerJ. 2022;10:e14303. PMID 36348665
  20. Zhang J, et al. Diagnostic performance of circulating cell-free DNA for HCC: a meta-analysis. Biomark Med. 2021;15:219-239. PMID 33470842
  21. Yang Y, et al. Predictive value of PD-L1 expression with PD-1/PD-L1 inhibitors in HCC. Cancer Med. 2023;12:9282-9292. PMID 36965092
  22. Harding-Theobald E, et al. Radiomics for diagnosis and prognosis of HCC: a systematic review. Aliment Pharmacol Ther. 2021;54:890-901. PMID 34390014
  23. Wang Q, et al. Radiomics for preoperative histopathological grade prediction in HCC. Eur J Radiol. 2023;166:111015. PMID 37541183
  24. Li H, et al. GALAD and serum biomarkers for NAFLD-related HCC: network meta-analysis. Expert Rev Gastroenterol Hepatol. 2023;17:1159-1167. PMID 37929312
  25. Wan HG, et al. Osteopontin versus AFP for HCC diagnosis: a meta-analysis. Clin Res Hepatol Gastroenterol. 2014;38:706-714. PMID 25034355