Skip to content

Hepatocellular carcinoma — systemic therapy

TL;DR — Sorafenib established systemic survival benefit in 2008; first-line therapy is now dominated by immunotherapy combinations, especially atezolizumab-bevacizumab and tremelimumab-durvalumab, with nivolumab-ipilimumab supported by the global CheckMate 9DW phase 3 report (Llovet 2008, PMID 18650514; Finn 2020, PMID 32402160; Abou-Alfa 2022, PMID 38319892; Yau 2025, PMID 40349714). No head-to-head trial ranks these regimens, and selection is driven by bleeding/variceal risk, autoimmune or transplant context, cardiovascular/renal risk, liver reserve, access, and patient priorities rather than a validated tumor biomarker. Most pivotal trials enrolled Child-Pugh A patients with preserved performance status, so evidence thins rapidly in decompensated cirrhosis. Second-line evidence was largely generated after sorafenib; comparative post-immunotherapy data are now available but remain non-randomized. AFP ≥400 ng/mL is the rare validated treatment-selection biomarker, identifying the REACH-2 population benefiting from ramucirumab (Zhu 2019, PMID 30665869).

Genealogy of systemic therapy

Era/trial Regimen/comparator Headline result Role
SHARP Sorafenib vs placebo OS 10.7 vs 7.9 months; HR 0.69 First proven systemic survival benefit (Llovet 2008, PMID 18650514)
REFLECT Lenvatinib vs sorafenib OS 13.6 vs 12.3 months; HR 0.92; non-inferior First-line TKI alternative (Kudo 2018, PMID 29433850)
IMbrave150 Atezolizumab-bevacizumab vs sorafenib Superior OS and PFS First-line IO/VEGF standard (Finn 2020, PMID 32402160)
HIMALAYA STRIDE tremelimumab-durvalumab vs sorafenib Superior OS for STRIDE First-line dual-IO option (Abou-Alfa 2022, PMID 38319892)
CheckMate 9DW Nivolumab-ipilimumab vs lenvatinib/sorafenib Median OS 23.7 vs 20.6 months; HR 0.79 (95% CI 0.65–0.96) First-line dual-IO option (Yau 2025, PMID 40349714)
RATIONALE-301 Tislelizumab vs sorafenib Non-inferior OS First-line monotherapy option in relevant jurisdictions (Qin 2023, PMID 37796513)
CARES-310 Camrelizumab-rivoceranib vs sorafenib Improved OS and PFS First-line combination in relevant jurisdictions (Qin 2023, PMID 37499670)

Availability and regulatory status differ by country and date; this page summarizes published efficacy, not local approval.

Atezolizumab plus bevacizumab

IMbrave150 randomized untreated unresectable HCC to PD-L1 blockade plus VEGF inhibition or sorafenib. The combination improved overall and progression-free survival and changed the first-line standard (Finn 2020, PMID 32402160).

Bevacizumab creates regimen-specific preparation:

  • assess varices and recent bleeding;
  • treat high-risk varices before therapy when feasible;
  • control hypertension;
  • evaluate proteinuria and renal function;
  • consider thrombosis and wound-healing risk;
  • coordinate procedures and surgery.

Real-world multicentre comparison focused on bleeding and thromboembolic outcomes after atezolizumab-bevacizumab versus lenvatinib, reinforcing that trial eligibility and endoscopic preparation do not eliminate risk (Ben Khaled 2024, PMID 38798717).

Tremelimumab plus durvalumab

HIMALAYA used one priming dose of CTLA-4 antibody tremelimumab with ongoing PD-L1 antibody durvalumab (STRIDE). It improved overall survival over sorafenib without a VEGF inhibitor (Abou-Alfa 2022, PMID 38319892).

Potential reasons to favor a non-VEGF doublet include major bleeding risk, uncontrolled hypertension, proteinuria, arterial events, or recent surgery. Dual immune activation can increase severe immune toxicity compared with PD-(L)1 monotherapy.

Nivolumab plus ipilimumab

CheckMate 9DW randomized 668 patients to nivolumab-ipilimumab or investigator choice of lenvatinib/sorafenib. At 35.2 months' median follow-up, median OS was 23.7 versus 20.6 months (HR 0.79, 95% CI 0.65–0.96). The survival curves crossed: mortality was higher with dual IO during the first six months (HR 1.65, 95% CI 1.12–2.43) and lower thereafter (HR 0.61, 95% CI 0.48–0.77). Grade 3–4 treatment-related adverse events occurred in 41% versus 42%, while 12 versus three deaths were attributed to treatment (Yau 2025, PMID 40349714).

The regimen adds another first-line dual-IO option but does not resolve which immune combination is best.

TKI monotherapy remains relevant

Agent Target profile Evidence niche Characteristic concerns
Sorafenib Multikinase/RAF/VEGFR First-line when combinations unsuitable; historical comparator Hand-foot reaction, diarrhea, hypertension (Llovet 2008, PMID 18650514)
Lenvatinib VEGFR/FGFR and others Non-inferior first-line alternative Hypertension, proteinuria, appetite/weight loss (Kudo 2018, PMID 29433850)
Regorafenib Multikinase After tolerated sorafenib progression Similar class toxicity; requires prior tolerance (Bruix 2017, PMID 27932229)
Cabozantinib MET/AXL/VEGFR and others Later line after prior sorafenib Hypertension, hand-foot reaction, diarrhea (Abou-Alfa 2018, PMID 29972759)

TKIs are important for patients with contraindications to checkpoint therapy and as later-line options after IO, although prospective sequence-specific evidence is limited.

Second line: evidence and mismatch

Trial/agent Enriched or prior population Main finding
RESORCE/regorafenib Progressed after and tolerated sorafenib OS 10.6 vs 7.8 months; HR 0.63 (Bruix 2017, PMID 27932229)
CELESTIAL/cabozantinib Previously treated, largely post-sorafenib Improved OS over placebo (Abou-Alfa 2018, PMID 29972759)
REACH-2/ramucirumab Post-sorafenib; AFP ≥400 ng/mL Improved OS; biomarker-selected phase 3 (Zhu 2019, PMID 30665869)
Pembrolizumab studies Previously treated populations Activity with region/trial-specific statistical interpretation

The evidence mismatch is fundamental: modern patients often progress after atezolizumab-bevacizumab or dual IO, while registration trials tested post-sorafenib sequences. Cross-trial ordering is not a substitute for randomized sequencing evidence.

A failed combination is informative

COSMIC-312 compared cabozantinib-atezolizumab with sorafenib. The combination improved PFS but did not improve overall survival in final analysis, demonstrating that combining an active TKI and checkpoint inhibitor does not guarantee an OS advantage (Yau 2024, PMID 38364832).

This guards against a simple class rule that “IO + anti-angiogenic” combinations are interchangeable. Agent, dose, toxicity, subsequent treatment, and biology matter.

Treatment selection matrix

Clinical feature Avoid/deprioritize Consideration
Untreated high-risk varices/recent GI bleed Bevacizumab-containing therapy Dual IO or TKI after stabilization
Solid-organ transplant Checkpoint inhibitors TKI/anti-VEGF strategies; rejection risk is major
Active severe autoimmune disease Checkpoint combinations Individualized immunology/oncology assessment
Uncontrolled hypertension/proteinuria VEGF-pathway therapy Control risk or choose non-VEGF option
Severe frailty/decompensation Most pivotal-trial regimens Optimize liver disease; goals and evidence boundary
AFP ≥400 after prior therapy — Ramucirumab has biomarker-selected evidence (Zhu 2019, PMID 30665869)
Rapid need for response Regimen with low response probability Compare response rates, burden, and toxicity
Oral-adherence barrier TKI Infusion logistics may be preferable, or vice versa

Liver function is an effect modifier and safety boundary

Pivotal trials predominantly enrolled Child-Pugh A disease. Child-Pugh B is heterogeneous: a patient with mild laboratory impairment differs from one with active ascites or encephalopathy. Observational data cannot fully separate drug toxicity from the natural history of liver failure.

Before and during therapy track:

  • bilirubin, albumin, INR, creatinine, sodium;
  • ascites and encephalopathy;
  • performance status and frailty;
  • variceal bleeding and infection;
  • whether progression is tumor, liver, or both.

Stopping therapy for “progression” and stopping for decompensation are different clinical events and should be reported separately.

Response assessment

RECIST size criteria remain trial-standard; viable-enhancement approaches can add information in liver-dominant disease. Immunotherapy can produce delayed or mixed responses, but true rapid progression is more common than pseudoprogression.

At each assessment distinguish:

  • target-lesion change;
  • new intrahepatic versus extrahepatic disease;
  • macrovascular progression;
  • liver-function trajectory;
  • symptomatic benefit or decline;
  • immune or VEGF/TKI toxicity.

Treatment beyond radiographic progression requires a credible clinical rationale, not reflex.

Checkpoint therapy can inflame liver, bowel, lung, endocrine organs, skin, nervous system, heart, kidney, and other tissues. In HCC, immune hepatitis must be distinguished from tumor progression, viral reactivation, biliary obstruction, ischemia, and cirrhotic fluctuation.

High-risk settings include:

  • prior liver or other solid-organ transplant;
  • active autoimmune disease;
  • baseline corticosteroid/immunosuppressive therapy;
  • viral hepatitis requiring monitoring;
  • poor liver reserve, where small injury causes decompensation.

Anti-VEGF and TKI safety

Toxicity Monitoring/mitigation
Hypertension Baseline and repeated blood pressure
Proteinuria Urinalysis/protein quantification
Bleeding Variceal assessment, history, medications
Thrombosis Symptoms and vascular risk
Wound healing Coordinate surgery/procedures and drug hold
Hand-foot reaction Early skin care and dose adjustment
Diarrhea/anorexia Hydration, nutrition, dose management

Trial ledger: positive, negative, and jurisdiction-specific signals

Trial/regimen Quantified outcome What it establishes—and does not
CheckMate 459: nivolumab versus sorafenib OS did not cross the prespecified significance boundary despite durable responses Single-agent PD-1 activity does not equal phase-3 first-line superiority (Yau 2022, PMID 34914889)
Donafenib versus sorafenib Median OS 12.1 versus 10.3 months; HR 0.831 (95% CI 0.699–0.988) Positive China-only phase 2/3 evidence; availability and external transportability vary (Qin 2021, PMID 34185551)
KEYNOTE-240 OS 13.9 versus 10.6 months; HR 0.781 (95% CI 0.611–0.998), but p=0.0238 missed multiplicity-adjusted p=0.0174 Clinically compatible benefit but formally negative statistical hierarchy (Finn 2020, PMID 31790344)
CheckMate 040 nivo-ipi At five years ORR 34%, 27%, and 29% across schedules; median response duration 51.2 months in the most active arm Durable minority responses; randomized dosing comparison, not phase-3 comparator evidence (Melero 2024, PMID 38844309)
HIMALAYA five-year update STRIDE OS HR 0.76 (95% CI 0.65–0.89); five-year OS 19.6% versus 9.4% Confirms a durable survival tail; does not directly compare with atezolizumab-bevacizumab (Rimassa 2025, PMID 40222621)
REFLECT response analysis RECIST ORR 18.8% versus mRECIST ORR 40.6%; OS 13.6 versus 12.3 months, HR 0.92 Response rate depends heavily on criteria; noninferiority does not mean identical toxicity or response depth (Finn 2024, PMID 39435271)

Important populations at the evidence boundary

Vp4 portal-vein invasion carries a poor prognosis. In an exploratory IMbrave150 subgroup, median OS was 7.6 versus 5.5 months and HR 0.62 (95% CI 0.34–1.11) for atezolizumab-bevacizumab versus sorafenib; the wide interval and descriptive p value preclude certainty despite a directionally consistent effect (Finn 2024, PMID 39687036). Baseline ALBI grade remained strongly prognostic within IMbrave150, and benefit was observed across ALBI strata represented in this Child-Pugh A trial; this does not validate use in decompensated disease (Kudo 2023, PMID 37901766).

BCLC B systemic-treatment evidence is also post hoc. Only 74 of 501 IMbrave150 participants were recorded as BCLC B, so apparent benefit in TACE-unsuitable intermediate disease supports migration but cannot define the exact TACE-unsuitable boundary (Kudo 2023, PMID 37767068). Child-Pugh B, recent decompensation, severe portal hypertension, organ transplantation, and uncontrolled autoimmune disease remain underrepresented across pivotal trials.

Sequencing after first-line immunotherapy

Second-line approvals were built after sorafenib, not after modern combinations. Regorafenib required prior sorafenib tolerance; cabozantinib and ramucirumab populations were mostly post-sorafenib; KEYNOTE-240 likewise enrolled previously treated disease (Bruix 2017, PMID 27932229; Abou-Alfa 2018, PMID 29972759; Zhu 2019, PMID 30665869; Finn 2020, PMID 31790344).

Post-atezolizumab-bevacizumab evidence has since become more informative without becoming randomized. LEVIATHAN compared 125 lenvatinib-treated with 105 sorafenib-treated patients and reported median OS 11.9 versus 7.4 months (HR 0.67, p=0.018); propensity adjustment cannot eliminate treatment-selection bias (Lombardi 2025, PMID 41321927). A reconstructed individual-patient-data meta-analysis included 1,663 patients across 16 predominantly retrospective studies; median OS was 9.8 months overall, with exploratory RMST advantages for lenvatinib and regorafenib over sorafenib, but residual confounding precludes a regimen ranking (Akkus 2026, PMID 42128139).

Treatment choice should therefore preserve future options. After anti-VEGF/PD-L1, a TKI avoids immediate re-use of the same biology; after dual IO, a VEGF-directed regimen is mechanistically non-cross-resistant. Rechallenge, switching between checkpoint combinations, and the optimal placement of biomarker-selected ramucirumab have no definitive randomized sequence trials.

Efficacy is not the only endpoint

Response depth and durability matter when disease threatens portal flow or symptoms, while time to deterioration and health-related quality of life may distinguish regimens with similar OS. LEAP-002 incorporated prespecified patient-reported outcomes even though lenvatinib-pembrolizumab did not establish a new efficacy standard, illustrating why a negative superiority trial can still inform toxicity and lived benefit (Finn 2025, PMID 40435872). Five-year immunotherapy updates demonstrate a survival tail, but most patients still die within five years and immune toxicity can be permanent (Rimassa 2025, PMID 40222621).

Controversies

  • Atezolizumab-bevacizumab versus dual IO. A PubMed search updated 2026-08-30 found no randomized direct comparison among active first-line combinations; each pivotal regimen used sorafenib or a TKI control. Bleeding risk, autoimmune context, response probability, dosing schedule, and local access—not indirect ranking—should drive selection (Finn 2020, PMID 32402160; Abou-Alfa 2022, PMID 38319892; Yau 2025, PMID 40349714).
  • Statistical versus clinical negativity. KEYNOTE-240 and CheckMate 459 showed signals without meeting prespecified thresholds; they support biological activity but not rewriting the primary endpoint (Finn 2020, PMID 31790344; Yau 2022, PMID 34914889).
  • mRECIST versus RECIST. Vascular devitalization can inflate mRECIST response relative to size response; response criteria should be prespecified and not treated as interchangeable (Finn 2024, PMID 39435271).
  • Child-Pugh B treatment. The need is high and evidence sparse. Observed outcomes confound liver death, cancer death, selection, dose intensity, and immortal time; prospective stratified trials remain necessary.
  • Post-IO sequencing. Comparative observational evidence now favors some TKIs over others, but no prospective randomized sequence trial was identified in the 2026-08-30 PubMed search; current ordering remains an extrapolation rather than a comparative fact (Lombardi 2025, PMID 41321927; Akkus 2026, PMID 42128139).

Open questions

  • Which first-line IO combination is superior or safest in direct comparison?
  • What is the optimal second-line sequence after atezolizumab-bevacizumab or dual IO?
  • Can CTNNB1, immune class, circulating DNA, or microbiome prospectively predict treatment interaction?
  • How should systemic therapy be studied in Child-Pugh B without conflating drug effect and liver mortality?
  • Does AFP select ramucirumab because it is predictive, or partly because it defines adverse angiogenic biology (Zhu 2019, PMID 30665869)?

References

  1. Llovet JM, et al. Sorafenib in advanced hepatocellular carcinoma. N Engl J Med. 2008;359:378-390. PMID 18650514
  2. Kudo M, et al. Lenvatinib versus sorafenib in first-line treatment of unresectable hepatocellular carcinoma. Lancet. 2018;391:1163-1173. PMID 29433850
  3. Finn RS, et al. Atezolizumab plus bevacizumab in unresectable hepatocellular carcinoma. N Engl J Med. 2020;382:1894-1905. PMID 32402160
  4. Abou-Alfa GK, et al. Tremelimumab plus durvalumab in unresectable hepatocellular carcinoma. NEJM Evid. 2022;1:EVIDoa2100070. PMID 38319892
  5. Yau T, et al. Nivolumab plus ipilimumab versus lenvatinib or sorafenib as first-line treatment for unresectable hepatocellular carcinoma (CheckMate 9DW): an open-label, randomised, phase 3 trial. Lancet. 2025;405:1851-1864. PMID 40349714
  6. Qin S, et al. Tislelizumab vs sorafenib as first-line treatment for unresectable hepatocellular carcinoma. JAMA Oncol. 2023;9:1651-1659. PMID 37796513
  7. Qin S, et al. Camrelizumab plus rivoceranib versus sorafenib as first-line therapy for unresectable hepatocellular carcinoma (CARES-310). Lancet. 2023;402:1133-1146. PMID 37499670
  8. Bruix J, et al. Regorafenib after sorafenib treatment (RESORCE). Lancet. 2017;389:56-66. PMID 27932229
  9. Abou-Alfa GK, et al. Cabozantinib in patients with advanced and progressing hepatocellular carcinoma. N Engl J Med. 2018;379:54-63. PMID 29972759
  10. Zhu AX, et al. Ramucirumab after sorafenib in advanced HCC and increased AFP (REACH-2). Lancet Oncol. 2019;20:282-296. PMID 30665869
  11. Ben Khaled N, et al. Atezolizumab/bevacizumab or lenvatinib in HCC: real-world study focused on bleeding and thromboembolic events. JHEP Rep. 2024. PMID 38798717
  12. Yau T, et al. Cabozantinib plus atezolizumab versus sorafenib for advanced HCC (COSMIC-312): final results. Lancet Gastroenterol Hepatol. 2024;9:310-322. PMID 38364832
  13. Yau T, et al. Nivolumab versus sorafenib in advanced HCC: CheckMate 459 phase 3 trial. Lancet Oncol. 2022;23:77-90. PMID 34914889
  14. Yau T, et al. Nivolumab plus ipilimumab after sorafenib: CheckMate 040 randomized cohort. JAMA Oncol. 2020;6:e204564. PMID 33001135
  15. Qin S, et al. Donafenib versus sorafenib in unresectable or metastatic HCC. J Clin Oncol. 2021;39:3002-3011. PMID 34185551
  16. Finn RS, et al. Atezolizumab-bevacizumab in HCC with Vp4 portal-vein invasion. Liver Cancer. 2024;13:655-668. PMID 39687036
  17. Kudo M, et al. ALBI analyses of atezolizumab-bevacizumab versus sorafenib in IMbrave150. Liver Cancer. 2023;12:479-493. PMID 37901766
  18. Melero I, et al. Five-year nivolumab-ipilimumab results from CheckMate 040. Ann Oncol. 2024;35:537-548. PMID 38844309
  19. Kelley RK, et al. Tremelimumab plus durvalumab in unresectable HCC: randomized phase 1/2 expansion. J Clin Oncol. 2021;39:2991-3001. PMID 34292792
  20. Finn RS, et al. Pembrolizumab as second-line therapy in KEYNOTE-240. J Clin Oncol. 2020;38:193-202. PMID 31790344
  21. Rimassa L, et al. Five-year OS update from HIMALAYA. J Hepatol. 2025;83:899-908. PMID 40222621
  22. Roy A, et al. Updated efficacy and safety data from IMbrave150. J Clin Exp Hepatol. 2022;12:1575-1576. PMID 36340313
  23. Kudo M, et al. IMbrave150 BCLC-B exploratory analysis. Liver Cancer. 2023;12:238-250. PMID 37767068
  24. Finn RS, et al. Characterization of lenvatinib tumor responses in REFLECT. Liver Cancer. 2024;13:537-547. PMID 39435271
  25. Finn RS, et al. Health-related quality of life in LEAP-002. ESMO Open. 2025;10:105065. PMID 40435872
  26. Lombardi P, et al. Lenvatinib vs. sorafenib as second-line treatment post atezolizumab plus bevacizumab for hepatocellular carcinoma: the LEVIATHAN study. JHEP Rep. 2025;7:101595. PMID 41321927
  27. Akkus E, et al. Second-line TKI after first-line immunotherapy-based treatment in advanced HCC: reconstructed IPD meta-analysis. JHEP Rep. 2026;8:101893. PMID 42128139