Hepatocellular carcinoma — open questions¶

Last curated: 2026-08-30

Stable IDs are retained when questions are revised. Tier 1 questions could change practice and are designable now; Tier 2 questions require more assay, cohort, or mechanistic development.

Tier 1 — practice-changing and designable today¶

OQ-1 — Does contemporary HCC surveillance reduce mortality outside chronic HBV?¶

The sole large positive randomized trial enrolled 18,816 HBV-infected Shanghai adults and reported a mortality rate ratio of 0.63 despite 58% adherence (Zhang 2004, PMID 15042359). Observational meta-analysis supports early detection, curative treatment, and survival but remains vulnerable to bias and heterogeneity (Singal 2022, PMID 35139400), while systematic review rated mortality evidence very low strength (Kansagara 2014, PMID 24934699). A pragmatic randomized strategy trial in nonviral cirrhosis remains designable.

OQ-2 — Which non-cirrhotic MASLD subgroup should undergo surveillance?¶

Average incidence (~0.1–1.3 per 1,000 patient-years) is too low for blanket surveillance, but risk concentrates by age, sex, diabetes, and fibrosis (Huang 2021, PMID 33349658). In 30,414 veterans, HCC risk rose 18% per 5-kPa liver-stiffness increment, while an individual-patient/aggregate meta-analysis of 379,194 non-cirrhotic patients found that non-invasive tests enrich risk without yet validating an actionable universal threshold (John 2026, PMID 40810411; Siriwong 2025, PMID 39919008). The decisive study is a prospective two-stage design: lock/calibrate an eligibility rule, then randomize surveillance within the above-threshold stratum and measure late-stage HCC, false-positive procedures, cost, and mortality.

OQ-3 — Should poor ultrasound visualization trigger abbreviated MRI?¶

Ultrasound detects only ~47% of early HCC in meta-analysis (Tzartzeva 2018, PMID 29425931). In paired NAFLD-cirrhosis data, severe visualization limitation occurred in 35% with ultrasound versus 19% with abbreviated MRI, and obesity increased the odds of severe ultrasound limitation (OR 5.1, 95% CI 1.1–23.1) (Huang 2022, PMID 35229334). Active MRI-versus-ultrasound studies (NCT06658782; NCT07010588) should prespecify a VIS-C-triggered strategy, interval cancers, false referrals, decompensation, and cost—not sensitivity alone.

OQ-4 — Can blood panels improve surveillance outcomes, not only AUC?¶

Case-control performance is no longer enough. In phase-3 cohorts, HES V2.0 exceeded GALAD true-positive rate by 7.2% at a fixed 10% false-positive rate, yet external HEDS validation gave both algorithms AUROC 0.79 with different threshold profiles (El-Serag 2025, PMID 38899967; El-Serag 2026, PMID 41043723). A randomized replacement/augmentation trial should fix the false-positive budget, specify recall imaging, and measure late-stage incidence and treatment eligibility.

OQ-5 — At what post-HCV-cure risk can surveillance safely stop?¶

After sustained response, pooled HCC occurrence remains 2.47 per 100 person-years with cirrhosis versus 0.85 without (Lv 2024, PMID 38965190). A dynamic model incorporating fibrosis regression, age, diabetes, and competing death could support randomized discontinuation/non-inferiority testing.

OQ-6 — Which BCLC B phenotype should receive systemic therapy before TACE?¶

BCLC 2022 acknowledges heterogeneous transplantable, TACE-suitable, and diffuse intermediate disease (Reig 2022, PMID 34801630). EMERALD-1 and LEAP-012 improve PFS with TACE-systemic combinations, but direct comparison with systemic therapy alone and mature OS are lacking (Sangro 2025, PMID 39798579; Kudo 2025, PMID 39798578). A three-arm platform—selective TACE, contemporary systemic therapy, and combination—should stratify by up-to-seven burden, ALBI, distribution, and planned selectivity, with OS, time to decompensation, and quality-adjusted survival as co-key outcomes.

OQ-7 — What is the optimal stopping rule for repeated TACE?¶

Repeated incomplete response and liver-function decline predict diminishing benefit, yet “TACE refractoriness” definitions vary. Japanese criteria use at least two consecutive inadequate responses after technical reassessment, while European “untreatable progression” also emphasizes invasion, spread, performance, and liver reserve (Kudo 2014, PMID 25427730; Raoul 2014, PMID 24945002). A prospective stopping rule should be tested against clinician judgment and jointly model viable-tumor response, selectivity, bilirubin/ALBI change, and the opportunity cost of delaying systemic therapy.

OQ-8 — Which first-line immunotherapy combination should be chosen?¶

Atezolizumab-bevacizumab, tremelimumab-durvalumab, nivolumab-ipilimumab, and regionally available combinations were tested against sorafenib or TKIs, not one another (Finn 2020, PMID 32402160; Abou-Alfa 2022, PMID 38319892; Yau 2025, PMID 40349714; Qin 2023, PMID 37499670). A targeted PubMed search on 2026-08-30 found no randomized direct comparison among the active combinations. A pragmatic head-to-head trial stratified by bleeding risk and etiology would change practice.

OQ-9 — What is the best sequence after first-line IO combination failure?¶

Regorafenib, cabozantinib, and ramucirumab phase 3 evidence was largely generated after sorafenib (Bruix 2017, PMID 27932229; Abou-Alfa 2018, PMID 29972759; Zhu 2019, PMID 30665869). New post-IO comparisons remain observational: LEVIATHAN reported adjusted advantages for lenvatinib over sorafenib, and a 1,663-patient reconstructed-data meta-analysis found heterogeneous TKI outcomes with residual confounding (Lombardi 2025, PMID 41321927; Akkus 2026, PMID 42128139). The 2026-08-30 search identified no prospective randomized post-IO sequence trial.

OQ-10 — Does MASLD/MASH etiology modify checkpoint benefit?¶

Mechanistic and pooled subgroup work suggests impaired antitumor surveillance in NASH-HCC (Pfister 2021, PMID 33762733), but “nonviral” is heterogeneous and subgroup comparisons are underpowered. Prospective individual-patient randomized interaction testing is needed before etiology guides treatment.

OQ-11 — Does adjuvant immunotherapy improve cure rather than delay recurrence?¶

IMbrave050 initially improved recurrence-free survival after high-risk resection/ablation, but the 2026 update found that benefit was not sustained (RFS HR 0.90, 95% CI 0.72–1.12); OS remained immature and numerically unfavorable (HR 1.26, 95% CI 0.85–1.87), so the benefit–risk profile did not support adjuvant atezolizumab-bevacizumab (Qin 2023, PMID 37871608; Yopp 2026, PMID 41580093). A 332-patient phase-3 adjuvant-TACE trial was negative overall (RFS HR 0.88, 95% CI 0.62–1.24), whereas a selected 280-patient HBV/high-risk trial was positive (HR 0.68), showing that “high risk” is biologically heterogeneous (Ma 2025, PMID 39808820; Wang 2018, PMID 29420221). Future trials should distinguish early metastatic recurrence from late new primaries and require durable RFS plus mature OS or patient-important net benefit.

OQ-12 — How should transplant selection balance morphology, biology, and equity?¶

Milan provides transparency; Metroticket 2.0 improves continuous prediction using AFP, number, and size (Mazzaferro 1996, PMID 8594428; Mazzaferro 2018, PMID 28989060). Allocation simulations should quantify graft utility and disparate access before replacing categorical rules.

OQ-13 — How far beyond standard limits can successful downstaging select transplant candidates?¶

XXL randomized evidence and MERITS-LT support transplant after sustained response in selected patients, but upper entry limits and required observation time vary (Mazzaferro 2020, PMID 32615109; Mehta 2021, PMID 34331914).

OQ-14 — Can checkpoint therapy be integrated safely before liver transplantation?¶

Checkpoint exposure can precipitate graft rejection. An individual-patient meta-analysis found 24 rejections among 91 exposed recipients (26.4%), while an international 119-recipient cohort found rejection in 20.2%; washout <30 days had OR 21.3 and 30–50 days OR 9.48 versus >50 days (Rezaee-Zavareh 2025, PMID 38996924; Moeckli 2025, PMID 40042053). A prospective protocol must record agent, dose, last exposure, receptor-occupancy/immune assays, donor type, immunosuppression, rejection histology, graft loss, recurrence, and the denominator who never reaches transplant.

OQ-15 — How should Child-Pugh B systemic therapy be tested?¶

Pivotal trials largely enrolled Child-Pugh A. Child-Pugh B combines mildly abnormal laboratories, prior or active decompensation, and heterogeneous portal hypertension, making uncontrolled outcomes hard to interpret. A dedicated platform should stratify B7 versus B8/9, active versus historical ascites/encephalopathy, ALBI, and tumor burden; use starting-dose rules; and jointly adjudicate cancer progression, treatment toxicity, decompensation, non-cancer death, symptoms, and quality of life.

OQ-16 — Which intervention closes the surveillance-to-curative-treatment equity gap?¶

US meta-analysis found lower early-stage detection and worse survival for Black patients; SEER-Medicare showed lower curative-treatment receipt, especially in high-poverty neighborhoods (Rich 2022, PMID 33387668; Wagle 2022, PMID 34796703). Cluster-randomized navigation/registry interventions should measure stage and curative therapy, not orders alone.

OQ-17 — Does early integrated palliative care improve HCC outcomes?¶

Qualitative studies show symptom, information, and caregiver burden, while intervention evidence remains thin (Hansen 2015, PMID 25122134; Hansen 2017, PMID 28820518). A pilot randomized intervention enrolled 57 of 109 approached patients and retained 52, establishing feasibility but not definitive benefit (Verma 2023, PMID 36149682). A multicentre trial should trigger at advanced disease or first decompensation and test symptom burden, quality of life, goal-concordant treatment, acute-care use, caregiver outcomes, and survival without treating lower anticancer intensity as failure.

Tier 2 — enabling biology, tools, and future designs¶

OQ-18 — Can dominant HCC drivers be exploited therapeutically?¶

TERT, CTNNB1, and TP53 dominate the exome landscape but remain undrugged; “actionable” alterations rarely translate into HCC matches (Schulze 2015, PMID 25822088; Harding 2019, PMID 30373752). Synthetic-lethal and state-directed screens need liver-specific toxicity constraints.

OQ-19 — Is Wnt/CTNNB1 a predictive biomarker for specific IO regimens?¶

β-catenin activation drives immune exclusion and anti-PD-1 resistance in models and retrospective correlates (Ruiz de Galarreta 2019, PMID 31186238). Prospective randomized treatment-by-marker interaction is needed; mutation alone may not equal pathway state.

OQ-20 — Can spatial immune architecture be reduced to a deployable assay?¶

Single-cell work identifies heterogeneous malignant, macrophage, T-cell, and tertiary-lymphoid states (Lu 2022, PMID 35933472). A clinically useful assay must survive sampling, platform, spatial, and temporal heterogeneity.

OQ-21 — Can ctDNA define minimal residual disease after local cure?¶

MRD studies are assay-diverse and observational (Galli 2025, PMID 40058162). A locked assay should stratify recurrence early enough to support randomized ctDNA-guided adjuvant intervention (NCT07350824).

OQ-22 — What causes early versus late post-resection recurrence?¶

Micrometastatic recurrence and new primary HCC require different prevention. Multi-region tumor/adjacent-liver phylogenetics plus longitudinal ctDNA could distinguish them and refine adjuvant endpoints.

OQ-23 — Can personalized TARE dosimetry be standardized?¶

DOSISPHERE-01 shows dose planning changes response and survival, while comparative TARE literature mixes dosimetry approaches (Garin 2021, PMID 33166497; Brown 2023, PMID 35943116). Cross-platform absorbed-dose standards and liver-toxicity models are needed.

OQ-24 — What biomarker separates immune hepatitis from decompensation or progression?¶

All can raise liver tests during checkpoint therapy. Serial immune, viral, imaging, and tissue studies need adjudicated diagnoses to prevent both delayed immunosuppression and inappropriate steroid exposure.

OQ-25 — Can risk scores predict net surveillance benefit rather than cancer incidence?¶

Existing HBV models vary in transportability, and most omit competing death and treatment eligibility (Wu 2021, PMID 33667678). Utility models should combine HCC risk, detection performance, curative access, and non-HCC mortality.

OQ-26 — How should patient preferences enter treatment allocation algorithms?¶

Interviews report clinician-led decisions despite variable goals; preference studies show tradeoffs among route, efficacy, and burden (Wörns 2024, PMID 39052152; Chiba 2019, PMID 31118587). A preference-elicitation tool needs evidence that it improves concordance without delaying care.

Dots not yet connected¶

#	Dot A	Dot B	The missing junction	Powers
D1	MASLD non-cirrhotic absolute-risk epidemiology	Ultrasound visualization failure in obesity	Risk model that simultaneously selects who and which test	OQ-2, OQ-3
D2	Post-HCV-cure dynamic fibrosis	Surveillance harm/cost models	Individual stopping rule with competing mortality	OQ-5, OQ-25
D3	Wnt/CTNNB1 immune exclusion	Multiple first-line IO combinations	Randomized interaction showing regimen-specific resistance	OQ-8, OQ-19
D4	Single-cell immune/ecologic classes	Routine blood/tissue biomarkers	Minimal assay that preserves treatment-predictive spatial information	OQ-20, OQ-4
D5	ctDNA MRD after local cure	Adjuvant IO toxicity and uncertain cure benefit	MRD-triggered randomized adjuvant treatment	OQ-11, OQ-21
D6	Personalized TARE tumor dose	BCLC B systemic migration	Dose-response rule that decides TARE versus systemic first	OQ-6, OQ-23
D7	Downstaging response as tumor biology	Continuous transplant recurrence models	Allocation model combining response trajectory with AFP/morphology	OQ-12, OQ-13
D8	Checkpoint immune memory	Transplant graft tolerance	Prospective immune/washout predictor of rejection	OQ-14, OQ-24
D9	Surveillance racial/SES disparities	Closed-loop navigation interventions	Causal pathway from outreach to curative treatment and survival	OQ-16, OQ-1
D10	Qualitative caregiver uncertainty	Acute decompensation/bleeding red flags	Caregiver intervention tested against emergency utilization	OQ-17, OQ-26
D11	MASLD-specific immune surveillance	Etiology-specific systemic outcomes	Proper randomized treatment-by-etiology interaction	OQ-10, OQ-19
D12	Portal-hypertension measurement	Modern minimally invasive resection	Physiologic threshold for postoperative decompensation	OQ-12, OQ-15
D13	Phase-3 blood-panel validation at fixed false-positive rates	Real-world closed-loop recall failures	End-to-end trial from positive blood test to timely curative treatment	OQ-4, OQ-16
D14	Dynamic liver stiffness after etiologic cure	Time-varying competing mortality	Prospective rule for surveillance entry, interval extension, and stopping	OQ-5, OQ-25
D15	Patient preference for preserved daily function	Trial endpoints dominated by PFS/OS	Preference-weighted estimand validated for treatment allocation	OQ-8, OQ-26

“Powers” names the open questions that become more tractable if the junction is studied; it does not imply that no relevant study exists anywhere, only that the two evidence bodies remain unintegrated in this build.