Advancing Systemic Treatment beyond First Line for Advanced Hepatocellular Carcinoma Authors: Piotr T. Wysocki, Paulina Chmiel, Jacek Jassem, Leszek Kraj Published: 2026 ABSTRACT Purpose of ReviewHepatocellular carcinoma (HCC) ranks among the most prevalent solid malignancies worldwide. A substantial proportion of patients are diagnosed at advanced stages, making systemic therapies the primary treatment option. In recent years, treatment strategies have shifted from single-agent targeted therapies, such as sorafenib and lenvatinib, to combination regimens incorporating immune checkpoint inhibitors, with or without antiangiogenic agents. This evolving treatment paradigm has led to an increasing number of patients receiving second-line and subsequent therapies. However, clinical guidelines for treatment selection remain lacking.Recent FindingsCurrently, available salvage therapies include various molecularly targeted agents, immune checkpoint inhibitors, and combination regimens. Nonetheless, evidence supporting their efficacy and patient selection for specific strategies in this setting remains scarce.SummaryThis review summarizes recent clinical advancements and explores emerging strategies to optimize systemic treatment options beyond first-line therapy in the increasingly complex landscape of advanced HCC. Introduction In 2022, approximately 865,000 new hepatocellular carcinoma (HCC) cases and 664,000 deaths were reported worldwide [1]. HCC accounts for 75–85% of primary liver cancer cases and remains the sixth most common cancer and the third leading cause of cancer-related mortality globally. The disease burden is particularly significant in low- and middle-income countries, where viral hepatitis remains the primary risk factor. However, its incidence is increasing in some highly developed regions, largely driven by alcohol-related cirrhosis and the rising prevalence of metabolic risk factors such as obesity and diabetes. The therapeutic landscape for HCC is complex, with a diverse array of treatment options depending on the tumor stage. The Barcelona Clinic Liver Cancer (BCLC) staging system considers intra- and extrahepatic tumor burden, liver function, and the patient’s performance status (PS) [2]. Patients with BCLC stage C, characterized by extrahepatic spread or macrovascular invasion, good PS, and preserved liver function, are candidates for systemic treatments. Additionally, systemic therapy may be considered in the intermediate (BCLC-B) disease when the intrahepatic burden precludes locoregional management. Until 2020, first-line treatment options for advanced hepatocellular carcinoma (aHCC) were limited to two tyrosine kinase inhibitors (TKIs), sorafenib and lenvatinib (Fig. 1). However, the advent of immunotherapy, particularly immune checkpoint inhibitors (ICIs), has transformed the management of this malignancy. This paradigm shift has redefined the role of molecularly targeted agents (MTAs), which are now primarily utilized following disease progression on ICIs or in patients ineligible for immunotherapy. However, a major challenge in this evolving treatment landscape is the limited high-quality evidence supporting the efficacy of MTAs in the post-immunotherapy setting. Moreover, advancements in treatment strategies have significantly improved patient outcomes, expanding the number of patients eligible for second-line and subsequent therapies.Fig. 1Timeline of advancements in systemic treatment for advanced hepatocellular carcinoma. 1L- first line Timeline of advancements in systemic treatment for advanced hepatocellular carcinoma. 1L- first line This review synthesizes the current evidence supporting systemic therapies beyond first-line treatment in aHCC, with a particular focus on the post-immunotherapy setting. Additionally, we address the alignment of these therapies with clinical guidelines and highlight key gaps in knowledge and evidence that need to be addressed. First-line Systemic Management – The Evolving Paradigm Until recently, therapeutic options for aHCC were remarkably limited. Conventional systemic chemotherapy conferred abysmal response rates and had a questionable impact on overall survival (OS) [3]. As a result, early clinical efforts focused on molecular vulnerabilities, leading to the approval of several molecularly targeted agents (MTAs). Hepatocarcinogenesis is driven by the progressive acquisition of molecular alterations, including telomerase (TERT) promoter mutations, TP53 inactivation, Wnt/β-catenin pathway activation (via CTNNB1, AXIN1, and APC mutations), and dysregulation of the Ras/Raf/MAPK and PI3K/AKT-mTOR signaling. While most of these alterations are not directly druggable, they drive pathway rewiring, neoangiogenesis, and tumor growth dependent on receptor tyrosine kinases (RTKs). HCC is further characterized by excessive vascular endothelial growth factor A (VEGF-A) production, which promotes angiogenesis and hepatocyte growth factor (HGF) secretion, thereby stimulating tumor proliferation via MET receptor. Furthermore, elevated VEGF levels contribute to immunotolerant, pro-tumorigenic microenvironment. Finally, a subset of tumors exhibits amplifications of CCND1, FGF19, MYC, or MET, leading to overexpression of RTKs or their ligands [4, 5]. Consequently, the first approved systemic treatments for aHCC were multi-TKIs, such as sorafenib and lenvatinib. Sorafenib targets VEGF receptors (VEGFR1, VEGFR2, VEGFR3), platelet-derived growth factor receptors (PDGFRs), RET, c-Kit, and RAF kinases (RAF1, BRAF), while lenvatinib inhibits VEGFR1-3, fibroblast growth factor receptors (FGFR1-4), PDGFRs, RET, and c-Kit. Sorafenib, approved in 2008 for first-line treatment of aHCC, marked a major breakthrough in aHCC therapy. Its efficacy was established through the phase III SHARP and ASIA–PACIFIC trials, which demonstrated significant improvements in time-to-progression and OS compared to placebo [6, 7]. The REFLECT trial later established lenvatinib as a non-inferior alternative to sorafenib in terms of OS while offering superior overall response rate (ORR) and progression-free survival (PFS) [8]. With the advent of immunotherapy, the therapeutic landscape of aHCC has undergone a profound transformation. A considerable proportion of HCC cases exhibit immune tolerance, driven by a persistently inflamed and/or cirrhotic microenvironment. This immunosuppressive milieu is shaped by multiple factors, including elevated transforming growth factor β (TGFβ) and VEGF signaling, dysfunctional CD8 + T cells, increased number of regulatory T (Treg) cells, and the overexpression of immune checkpoints [4, 9–11]. These mechanisms collectively contribute to immune evasion, underscoring the rationale for targeting immune checkpoint pathways in HCC. Pivotal phase III trials published in 2020–2021 established immunotherapy (IO)-based regimens as superior to TKIs for first-line treatment of aHCC. In the IMbrave150 trial, the combination of atezolizumab (anti-PD-L1 antibody) plus bevacizumab (anti-VEGF-A antibody) improved ORR (30% vs 11%), PFS (median 6.8 vs 4.3 months), OS (median 19.2 vs 13.4 months; hazard ratio [HR] 0.58) and quality of life compared with sorafenib [12, 13]. Similar benefits were observed in the ORIENT-32 trial where sintilimab (anti–PD-1) combined with a bevacizumab biosimilar (IBI305) demonstrated improved outcomes in Chinese population [14]. Furthermore, the HIMALAYA trial showed that durvalumab (anti-PD-L1 antibody) with a single priming dose of tremelimumab (anti-CTLA4 antibody), the STRIDE regimen, resulted in superior ORR and OS compared with sorafenib (median 16.4 vs 13.8 months, HR 0.78) [15, 16]. ICI monotherapy with durvalumab or tislelizumab (anti-PD-1 antibody) demonstrated non-inferiority to sorafenib, offering an alternative for patients unsuitable for combination therapy [15, 17]. Finally, in the CheckMate-9DW, trial nivolumab (anti-PD1 antibody) plus ipilimumab (anti-CTLA4), achieved longer OS and higher ORR compared with physician’s choice of lenvatinib or sorafenib (median OS 23.7 vs. 20.6 months, HR 0.79; ORR 36% vs. 13%) [18]. Interestingly, most ICI–TKI combinations have not improved survival in phase III trials. Specifically, in the COSMIC-312 trial, atezolizumab plus cabozantinib failed to improve OS compared to sorafenib, and in the LEAP-002 trial, pembrolizumab (anti-PD1 antibody) plus lenvatinib did not show OS superiority over lenvatinib plus placebo [19, 20]. An exception is the predominantly Asia-based CARES-310 trial, in which camrelizumab (anti–PD-1 antibody) plus rivoceranib (a TKI, also known as apatinib) improved OS versus sorafenib but was associated with substantial toxicity (81% rate of grade ≥ 3 adverse events, AEs) [21]. As a result, IO-based regimens—atezolizumab-bevacizumab, durvalumab-tremelimumab (STRIDE) and nivolumab-ipilimumab—are now the preferred first-line treatments for aHCC, recommended by major clinical societies, including the European Society of Medical Oncology (ESMO), the American Society of Clinical Oncology (ASCO), the National Comprehensive Cancer Network (NCCN), the European Association for the Study of the Liver (EASL), and the American Association for the Study of Liver Diseases (AASLD) [2, 22–26]. Other first-line treatment options include camrelizumab–rivoceranib, tislelizumab, and durvalumab monotherapy. MTAs are now primarily considered second-line treatments following immunotherapy failure, except when ICIs are contraindicated. In such cases, lenvatinib and sorafenib remain the preferred first-line options as per current guidelines. Second-line Agents: Evidence from the Pre-First-Line Immunotherapy Era For patients who do not respond to first-line TKIs, second-line treatment options primarily consist of three MTAs: regorafenib, cabozantinib, and ramucirumab. These therapies were approved based on randomized, placebo-controlled phase III trials, and target distinct molecular pathways. Additionally, IO-based treatments may also be considered in the second-line setting, though supporting evidence for their efficacy remains less robust compared to MTAs. The mechanisms of action of agents used in the systemic treatment of HCC are illustrated in Fig. 2.Fig. 2A simplified scheme of therapeutic agents in advanced hepatocellular carcinoma. Abbreviations: FGFR-fibroblast growth factor receptor; GzmB-granzyme B; IFNγ-interferon-gamma; KIT-receptor tyrosine kinase; MET- mesenchymal-epithelial transition factor receptor; MHC- major histocompatibility complex; PD-1-programmed cell death protein 1; PD-L1-programmed cell death-ligand 1; PDGFR-platelet-derived growth factor receptors; PFN-perforin; TCR-T-cell receptor; TNFα-Tumor necrosis factor-alpha; VEGF-vascular endothelial growth factor; VEGFR-vascular endothelial growth factor receptor A simplified scheme of therapeutic agents in advanced hepatocellular carcinoma. Abbreviations: FGFR-fibroblast growth factor receptor; GzmB-granzyme B; IFNγ-interferon-gamma; KIT-receptor tyrosine kinase; MET- mesenchymal-epithelial transition factor receptor; MHC- major histocompatibility complex; PD-1-programmed cell death protein 1; PD-L1-programmed cell death-ligand 1; PDGFR-platelet-derived growth factor receptors; PFN-perforin; TCR-T-cell receptor; TNFα-Tumor necrosis factor-alpha; VEGF-vascular endothelial growth factor; VEGFR-vascular endothelial growth factor receptor Regorafenib Regorafenib is a structurally related but more potent multikinase inhibitor than sorafenib that targets VEGFR1–3, TIE2, PDGFR-β, FGFRs, c-Kit, RET, RAF1, and BRAF, thus retaining inhibition of core sorafenib targets while providing broader and stronger blockade of proliferation, angiogenesis, anti-apoptotic, and immune-modulatory pathways [5]. Evidence from preclinical in vitro and in vivo models indicates that regorafenib outperforms sorafenib in inhibiting HCC growth and tumor angiogenesis [27]. Regorafenib has been endorsed for the treatment of aHCC in patients with disease progression on sorafenib. Its approval is based on the results of a placebo-controlled RESORCE phase III trial [28]. Given the molecular similarity and comparable toxicity profile of sorafenib and regorafenib, eligibility criteria mandated good prior sorafenib tolerance. Among study participants, 98% had Child–Pugh class A liver function, 29% demonstrated macrovascular invasion, and 70% had extrahepatic spread. Compared to placebo, regorafenib improved OS (median 10.6 vs. 7.8 months; HR, 0.63; 95% CI, 0.50–0.79), PFS (median 3.1 vs. 1.5 months; HR 0.46; 95% CI, 0.37–0.56), and ORR per modified RECIST for HCC (mRECIST) criteria (11% vs. 4%). Grade 3 or 4 AEs occurred in 67% of regorafenib-treated patients, compared to 39% in the placebo group. The most common AEs included hypertension, palmar-plantar erythrodysesthesia, fatigue, and diarrhea. Real-world data reveal mOS varying from 6.5 to 17.3 months for patients receiving regorafenib as the post-TKI therapy, with outcomes influenced by liver function, performance status, and access to subsequent therapy [29–34]. The prospective, observational REFINE study further evaluated the efficacy of regorafenib in 1,005 patients, of whom 62% had BCLC-C stage disease and 61% had Child–Pugh A liver function [35]. Nearly all participants (96%) had previously received sorafenib, with regorafenib used as a second-line therapy in 84% and as a third-line or later therapy in 14%. The mOS was 13.2 months in the whole cohort and 13.9 months in the second-line group. Treatment tolerance aligned with findings from the pivotal RESORCE trial. Cabozantinib Cabozantinib, another TKI approved for sorafenib-refractory aHCC, inhibits VEGFR2, MET, AXL, RET, ROS1, TYRO3, and c-Kit. Utilization of cabozantinib in the post-TKI setting is supported by preclinical evidence demonstrating that HGF/c-MET axis promotes resistance to sorafenib and lenvatinib in HCC [36–38]. By blocking the hepatocyte growth factor (HGF)-stimulated MET pathway, cabozantinib inhibits HCC cell migration and invasion, preventing lung and liver metastases in murine models [39]. Additionally, HCC cells frequently exhibit upregulated AXL signaling, where the AXL ligand growth arrest-specific 6 (Gas6) enhances cell migration, invasiveness, and epithelial-mesenchymal transition (EMT) [40]. Notably, AXL activation has been associated with resistance to VEGFR inhibitors (sunitinib, sorafenib, lenvatinib) in renal carcinoma and HCC [41–43]. Cabozantinib's simultaneous inhibition of MET, VEGFR, and AXL kinases may thus overcome resistance to more selective VEGFR inhibitors. The phase III CELESTIAL trial evaluated cabozantinib in patients who had previously received one (71%) or two (27%) lines of systemic therapy, including sorafenib [44]. Among 707 patients randomized to receive 60 mg of cabozantinib once daily or a placebo, all had preserved liver function (Child–Pugh A), with 78% exhibiting extrahepatic spread, and 30% having macrovascular invasion,. Cabozantinib significantly improved OS, the primary endpoint (median 10.2 months vs. 8.0 months for cabozantinib and placebo, respectively; HR 0.76; 95%, CI 0.63–0.92) and PFS (median 5.2 months vs. 1.9 months; HR 0.44; 95% CI, 0.36–0.52). The ORR was also higher in the cabozantinib group (4% vs. 1%). Grade 3–4 AEs occurred in 68% of patients receiving cabozantinib, including hand-foot skin reactions (17%), hypertension (16%), elevated AST (12%), fatigue, and diarrhea. Treatment with cabozantinib caused a transient deterioration of the quality of life in the first few weeks of therapy [45]. The benefit of cabozantinib was more pronounced in the second than in the third line setting (HR for OS, 0.74; 95% CI, 0.59–0.92; and 0.90; 95%, 0.63–1.29, respectively). Among patients who received only sorafenib as the prior treatment, cabozantinib and placebo yielded mOS of 11.3 and 7.2 months, respectively (HR, 0.70; 95% CI, 0.55–0.88) [46]. A Japanese prospective phase II trial reported exceptionally long mOS of 19.3 months in a cohort of 20 patients treated with cabozantinib after progressing on sorafenib, likely due to more favorable clinical characteristics than those in the CELESTIAL trial [47]. Cabozantinib was also evaluated in the first-line COSMIC-312 phase III, multicenter trial including 837 patients who were randomized to receive cabozantinib (40 mg daily) plus atezolizumab (1200 mg every 3 weeks), sorafenib (400 mg twice daily), or cabozantinib alone (60 mg daily) [19]. In a secondary endpoint analysis, cabozantinib monotherapy demonstrated an mPFS of 5.8 months vs. 4.3 months with sorafenib (HR 0.78; 95% CI, 0.56–0.97). ORR was numerically higher for cabozantinib plus atezolizumab (11%) than for cabozantinib alone (6.4%) and sorafenib (3.7%). Grade 3–4 AEs occurred in 66%, 57%, and 48% of patients in the combination regimen, cabozantinib monotherapy, and sorafenib arms, respectively. In routine clinical practice, cabozantinib is frequently utilized as a third-line treatment. Real-world data indicate that mOS with cabozantinib in the post-TKI setting varies between 6.9 and 12.1 months, reflecting geographic and clinical differences [48–52]. Ramucirumab Ramucirumab, the third agent approved in the post-sorafenib setting, is a monoclonal antibody targeting VEGFR2. Elevated VEGFR2 expression in HCC, a central mediator of angiogenesis, has been associated with poor outcomes in aHCC treated with sorafenib, thereby supporting VEGFR2 blockade as a strategy to overcome resistance [53]. While the initial REACH clinical trial did not demonstrate a survival benefit over placebo, subgroup analysis suggested potential efficacy in patients with high baseline serum alpha-fetoprotein (AFP) levels [54]. Subsequently, the phase III REACH-2 trial confirmed this finding, demonstrating significantly improved OS and PFS in patients with baseline AFP ≥ 400 ng/mL treated with ramucirumab compared to placebo (mOS 8.5 vs. 7.3 months, HR 0.71; 95% CI, 0.53–0.95; mPFS 2.8 vs. 1.6 months, HR 0.45; 95% CI, 0.34–0.60) [55]. ORR per RECIST 1.1 criteria were 5% vs. 1%, respectively. Serious AEs occurred in 35% of ramucirumab-treated patients vs. 29% in the placebo group. The most common grade 3–4 AEs in the experimental arm included hypertension (13%), hyponatremia (6%), and elevated AST (3%). A Japanese real-world study reported similar efficacy, with a mOS of 10.3 months in the second-line setting and 10.4 months in later lines among patients with high serum AFP levels [56]. An open-label expansion cohort of REACH-2 evaluated ramucirumab specifically in patients with AFP ≥ 400 ng/mL following failure of non-sorafenib therapies, showing comparable safety and efficacy to the original trial (mPFS 1.7 months, mOS 8.7 months) [57]. Other TKIs Rivoceranib (apatinib), a potent VEGFR2 inhibitor, is the latest TKI demonstrating efficacy in previously treated aHCC. The phase III AHELP trial, conducted in China, enrolled patients who had progressed on or were intolerant to at least one prior line of chemotherapy or targeted therapy [58]. OS and PFS were significantly longer with rivoceranib compared to placebo, with HRs of 0.79 (95% CI, 0.617–0.998) and 0.47 (95% CI, 0.37–0.60), respectively. The most common grade 3 or 4 treatment-related AEs in the rivoceranib arm included hypertension (28% vs. 2% in the placebo arm), hand–foot syndrome (18% vs. 0%), and thrombocytopenia (13% vs. 1%). Given that the AHELP trial was conducted exclusively in China, further studies are needed to validate its efficacy and safety in non-Asian populations. Another TKI, brivanib, a dual VEGF/FGFR inhibitor, improved ORR and PFS but failed to confer an OS benefit in a phase III trial in patients with sorafenib-refractory or sorafenib-intolerant HCC [59]. Lenvatinib, though approved as a first-line alternative to sorafenib, has also shown efficacy after prior sorafenib treatment [60–62]. A Japanese retrospective study involving 132 Child–Pugh A patients treated with lenvatinib showed comparable mPFS (5.2 vs. 4.8 months) and mOS (13.3 months vs. not reached) between first-line and subsequent-line use [61]. Conflicting data exist on the ORR of later-line lenvatinib, with some studies showing comparable rates between sorafenib-naïve and sorafenib-experienced patients (17% vs. 14% per RECIST; 44% vs. 34% per mRECIST, respectively) [60, 61], while others report a lower ORR in sorafenib-experienced group (26% vs. 62% per mRECIST) [62]. Second-line Immunotherapy Second-line options after TKIs include nivolumab monotherapy and nivolumab-ipilimumab, evaluated in the phase I/II CheckMate 040 trial [63–65]. In sorafenib-pretreated patients, nivolumab alone produced an ORR of 14% (per RECIST 1.1), mOS of 15.1 months, and a five-year OS of 12%. The combination, particularly at 1 mg/kg nivolumab plus 3 mg/kg ipilimumab, achieved superior outcomes (ORR of 34%, mOS 22.2 months, five-year OS 29%) [66]. NCCN guidelines endorse this combination for MTA-refractory aHCC [24]. Pembrolizumab results have been inconsistent: while KEYNOTE-240 did not show significant PFS or OS improvements,, the Asian KEYNOTE-394 trial demonstrated superior efficacy over placebo (mOS 14.6 vs. 13.0 months, HR 0.781; 95% CI, 0.611–0.998) [67, 68]. The standard first-line atezolizumab-bevacizumab regimen has occasionally been utilized after MTA failure. A retrospective French study including 50 patients after one to four lines of TKI reported an ORR of 14%, disease control rate (DCR) of 56%, mPFS of 8.0 months, and mOS of 17.1 months [69]. Survival outcomes were comparable regardless of the treatment line, with an mOS of 14.8 months in patients receiving one prior TKI and 17.1 months in those after two or more prior TKI lines. Retrospective Japanese studies also support the efficacy of atezolizumab-bevacizumab in post-TKI settings [70–72]. Comparisons of Second-line Agents in the Post-TKI Setting The lack of direct comparisons complicates optimal second-line selection after sorafenib or lenvatinib. Regorafenib, cabozantinib, and ramucirumab require prior sorafenib use per European Medicines Agency (EMA) and the United States Food and Drug Administration (FDA) guidelines. While these agents are sometimes utilized post-lenvatinib, the efficacy of this sequence remains insufficiently documented. Consequently, in many countries, sorafenib is the only approved option for patients who fail lenvatinib, and despite limited supporting data, ESMO guidelines recommend considering sorafenib after first-line lenvatinib [22, 73, 74]. The RESORCE trial, which led to the approval of regorafenib, mandated prior sorafenib tolerance (≥ 400 mg/day for at least 20 of the last 28 treatment days), and accordingly, ASCO guidelines explicitly recommend regorafenib for patients who tolerated sorafenib [23, 28]. Ramucirumab is indicated for patients with serum AFP ≥ 400 ng/mL, as regulatory guidelines are based on clinical evidence showing its effectiveness exclusively in individuals with high AFP levels [22–24, 55]. Matching-adjusted indirect comparisons (MAICs) suggest similar OS for cabozantinib and regorafenib (mOS 11.3 vs. 11.1 months, respectively; HR 0.83; 95% CI, 0.62–1.09), though cabozantinib exhibits superior PFS (5.5 vs. 3.0 months; HR 0.50; 95% CI, 0.41–0.62) [75]. Another MAIC reported consistent findings [76]. A meta-analysis of the CELESTIAL and RESORCE trials found no significant ORR, PFS or OS differences between both agents [77]. Similarly, real-world data indicate no significant PFS or DCR differences [78]. The duration of prior sorafenib exposure may influence the efficacy of subsequent regorafenib and cabozantinib. A MAIC incorporating data from the RESORCE and CELESTIAL trials showed that regorafenib, relative to cabozantinib, provides comparable PFS and shows a trend toward longer OS (HR for OS, 0.80, 95% CI, 0.54–1.20) in patients with ≥ 3 months of prior sorafenib exposure [79]. Conversely, cabozantinib may provide longer OS than regorafenib in patients who progress within six months of starting sorafenib [75]. Consistent findings from other studies indicated that patients experiencing early sorafenib failure tend to respond poorly to regorafenib but retain sensitivity to cabozantinib [31, 50]. Cabozantinib and ramucirumab were compared through a MAIC utilizing data from the CELESTIAL and REACH-2 trials in patients previously treated with sorafenib who had elevated serum AFP (≥ 400 ng/mL) [80]. The analysis revealed comparable OS (mOS 10.6 months for cabozantinib vs. 8.7 months for ramucirumab; nonsignificant), but superior PFS for cabozantinib (mPFS, 5.5 months vs. 2.8 months, respectively). Network meta-analyses (NMAs) suggest that regorafenib, cabozantinib, brivanib, ramucirumab, and pembrolizumab improve PFS, but only regorafenib and cabozantinib provide an OS benefit over placebo, with cabozantinib ranking highest for PFS and regorafenib for OS [81]. Another NMA found comparable outcomes for regorafenib, cabozantinib, and ramucirumab among patients with AFP ≥ 400 ng/mL [82]. However, in patients with AFP < 400 ng/ml, regorafenib was the only agent to show a superior OS compared to placebo (HR, 0.67; 95% CI, 0.50–0.90). In a propensity-score weighted analysis comparing nivolumab (1 mg/kg) plus ipilimumab (3 mg/kg) against regorafenib and cabozantinib, the median ORR per RECIST 1.1 was 31% for dual immunotherapy vs. 4% for cabozantinib and 5% for regorafenib [83]. The HR for OS with nivolumab-ipilimumab was 0.46 relative to cabozantinib and 0.56 relative to regorafenib. As lenvatinib received approval for aHCC much later than sorafenib and remains globally less accessible, there are few retrospective studies that have compared treatment options in the post-lenvatinib setting. In the retrospective RELEVANT study, the longest OS was achieved with subsequent immunotherapy (mOS 17.2 months), followed by ramucirumab (10.3 months), transarterial chemoembolization (TACE; 8.7 months), sorafenib (8.6 months), regorafenib (5.0 months), and best supportive care (4.5 months) [84]. Conversely, another report showed that in aHCC patients who experienced failure on lenvatinib, sorafenib conferred longer OS than nivolumab (8.7 vs. 3.0 months) [85]. A Japanese analysis of 53 patients who received sorafenib as second-line therapy after lenvatinib reported poor outcomes, with an mPFS of 1.8 months, an ORR of 1.8%, and a DCR of 20.8%. These results were significantly worse than those typically reported for sorafenib in other settings and were numerically inferior to outcomes with regorafenib (mPFS 3.8 months, ORR 17.6%, DCR 41.2%) [74]. Additionally, in a recent study of patients who had experienced treatment failure with lenvatinib, mOS was 15.8 months with subsequent sorafenib (HR 1.0), 20.8 months with non-sorafenib MTAs, such as regorafenib, cabozantinib, or ramucirumab (HR 0.85; 95% CI, 0.52–1.38), and not reached with immunotherapies (HR 0.69; 95% CI, 0.45–1.05) [73]. Accordingly, the optimal post-lenvatinib agent remains undetermined. MAIC studies suggested that cabozantinib and regorafenib generally exhibit comparable safety profiles [76, 77, 79]. However, cabozantinib shows slightly higher rates of certain high-grade toxicities, including ~ 5% absolute increase in grade 3–4 diarrhea and fatigue, along with a greater frequency of dose reductions and treatment discontinuations. In a meta-analysis, regorafenib was associated with a lower risk of serious AEs (SAE) compared to cabozantinib (odds ratio, OR 0.71; 95% CI, 0.54–0.92) and ramucirumab (OR 0.73; 95% CI, 0.56–0.95) [82]. However, these results should be interpreted with caution, as the analyses were based on data from the REFLECT trial, where only patients able to tolerate prior sorafenib were selected for regorafenib treatment. Additionally, a real-world analysis showed similar tolerability of regorafenib and cabozantinib [78]. A meta-analysis of phase III trials reported a lower risk of high-grade liver transaminase elevation with ramucirumab than with cabozantinib (OR 0.54; 95% CI, 0.30–0.99) and hypertension of any grade compared to both regorafenib (OR 0.49; 95% CI, 0.25–0.96) and cabozantinib (OR 0.49; 95% CI, 0.26–0.93) [82]. Lower rates of high-grade fatigue, increased AST, and hypertension with ramucirumab compared to cabozantinib were also demonstrated in a MAIC [80]. Notably, the occurrence of certain AEs during treatment with MTAs does not necessarily negatively impact patient outcomes. First, despite differences in the frequency of high-grade toxicities between ramucirumab and cabozantinib, the rates of AE-related treatment discontinuation are comparable [80]. Second, subgroup analyses indicate that some low-grade, treatment-related AEs—such as hypertension with cabozantinib, lenvatinib and ramucirumab, or hand-foot syndrome with regorafenib—are associated with improved survival [4, 31, 86]. Third, while concerns about AEs may lead to dose reductions and consequently lower relative dose intensity, no significant differences in ORR, DCR, or PFS were observed between patients receiving full versus reduced doses of cabozantinib [87, 88]. This observation is particularly relevant, as cabozantinib is frequently prescribed at reduced doses in clinical settings, especially in patients with compromised liver function [49, 50, 87, 89]. Similarly, initiating regorafenib at a reduced dose appears to have no adverse impact on treatment outcomes [31]. In summary, selecting the optimal second-line agent after the failure of upfront TKI therapy in aHCC remains challenging, primarily due to the lack of validated biomarkers to guide decision-making (aside from serum AFP ≥ 400 ng/mL for ramucirumab). Regorafenib is best suited for patients who previously tolerated sorafenib well and experienced clinical benefit from its use. Cabozantinib has demonstrated longer PFS than other agents in some studies and may be preferred for patients who experienced early progression on sorafenib. Ramucirumab, given its favorable safety profile, is an appropriate option for patients at higher risk of AEs, particularly the elderly or those with more advanced liver cirrhosis. Meanwhile, post-TKI immunotherapy represents a promising option for patients with aHCC related to viral hepatitis [90]. However, given the lack of evidence supporting OS superiority of any specific MTA, the final choice of therapy should be guided by patient and physician preferences, as well as medication availability. Second-line Treatment in the Post-immunotherapy Setting: Rationale As immunotherapy has become the frontline treatment for aHCC, determining the optimal therapeutic strategy for patients who experience ICI failure has become increasingly critical. Notably, many patients who progress on ICIs maintain a favorable overall condition. Retrospective data suggest that approximately 70% of patients who progressed on atezolizumab-bevacizumab retained a Child–Pugh A status, indicating sufficient liver function to qualify for subsequent therapies [91]. Sequencing of MTAs following IO-failure is supported by theoretical and preclinical evidence. First, ICIs exhibit prolonged persistence in the body, for instance, anti-PD-1 antibodies remain bound to circulating lymphocytes for over 20 weeks following infusion [92]. This sustained receptor occupancy may allow subsequently administered MTAs to synergize with residual ICI activity, creating a "pseudo-combination" effect that could help overcome immunoresistance. Second, TKIs possess anti-proliferative and cytotoxic properties, leading to the release of cancer neoantigens and potentially rekindling an immune response in patients with inadequate initial immune activation [92]. Third, given the central role of angiogenesis in the pathophysiology of HCC, inhibition of angiogenic signaling with TKIs represents a logical strategy, especially in patients who progress on dual immunotherapy and are naïve to antiangiogenic treatment. Additionally, even in patients previously exposed to bevacizumab, TKIs permit a more comprehensive blockade of tumor angiogenesis, given that bevacizumab inhibits only VEGF-A. Resistance to VEGF-A blockade is multifactorial and is proposed to arise from compensatory activation of redundant pathways, including other VEGF family members, angiopoietins, and fibroblast growth factors [93]. Elevated plasma levels and intratumoral expression of VEGF-D, an alternative ligand of VEGFR2, have been implicated in resistance to bevacizumab in colorectal cancer and glioma, and increased serum levels of VEGF-D and ANG-2 have been reported in patients with HCC who develop resistance to atezolizumab-bevacizumab [94–97]. All approved TKI for HCC inhibit VEGFRs, thereby enable broad suppression of signaling mediated by redundant VEGF ligands. Beyond angiogenesis, VEGFR signaling contributes to an immunosuppressive tumor microenvironment by promoting myeloid cell–mediated immune suppression, impairing dendritic cell maturation, driving CD8⁺ T-cell exhaustion, and facilitating regulatory T-cell expansion [98–100]. Moreover, inhibition of VEGFR2 has been demonstrated to suppress HCC growth in syngeneic models previously treated with combined anti-PD-L1 and anti-VEGF-A therapy. This effect appears to be mediated, at least in part, through the elimination of cancer stem cells and the disruption of cell–cell interactions between AFP-positive HCC cells and VEGFR2-expressing endothelial cells [101]. In addition to targeting VEGFR signaling, several TKIs (sorafenib, lenvatinib, and regorafenib) also inhibit PDGFRs. These oncogenic drivers are frequently overexpressed in HCC, their upregulation associates with poor prognosis, and contributes to tumor proliferation, metastasis, neoangiogenesis, and drug resistance. Accordingly, PDGFR inhibition may be another valuable strategy to suppress disease progression in immunotherapy-refractory HCC [102]. Recent retrospective studies indicate that post-IO therapy improves OS compared to no further systemic treatment, with a consistent HR of approximately 0.40 across multiple studies, even after adjusting for confounding factors [103–106]. Furthermore, real-world evidence suggests that the efficacy of TKIs in the post-IO setting is comparable to, or even superior to, their efficacy in treatment-naïve or TKI-pretreated HCC patients [31, 50, 87, 89, 107–113]. Predictors of prolonged survival during second-line TKI therapy include a favorable response to prior immunotherapy, longer ICI treatment duration, and the occurrence of low-grade immune-related AEs [91, 104, 114–116]. Early transition to second-line TKIs may be particularly advantageous in non-responders to ICI-only regimens. Cross-trial comparisons of first-line phase III trials reveal numerically higher rates of early progression in patients treated with durvalumab-tremelimumab in the HIMALAYA trial and nivolumab-ipilimumab in the CheckMate 9DW trial, compared to regimens employing antiangiogenic agents, including atezolizumab-bevacizumab in IMbrave150 and lenvatinib-pembrolizumab in the LEAP-002 trial [13, 16, 18, 20]. These observations suggest the importance of early identification of progressors on the ICI-only regimens, as this could facilitate a prompt transition to antiangiogenic therapies. Post-immunotherapy Systemic Treatment: Recommendations and Evidence Clinical guidelines from major oncological and hepatological societies advocate for second-line systemic treatment for aHCC patients with preserved liver function and adequate performance status following disease progression or intolerance to first-line immunotherapy. Recommended options include traditional first-line TKIs (sorafenib or lenvatinib), second-line MTAs (cabozantinib, regorafenib, or ramucirumab), and alternative IO-based regimens [22–24, 26, 117–119]. Nevertheless, due to scarce evidence, selecting an optimal treatment strategy remains challenging. Firstly, all second-line phase III clinical trials were conducted in patients progressing on sorafenib. Secondly, the efficacy of MTAs following immunotherapy has only been assessed in a few small, single-arm, prospective trials. Thirdly, the absence of head-to-head comparisons between agents and the inherent limitations of available retrospective comparative studies further complicate the decision-making process. Further, no validated biomarkers or clinical predictors are available to guide treatment selection. Finally, evidence on the efficacy of alternative IO-based regimens following the failure of a primary ICI treatment is sparse. Consequently, the absence of robust evidence has led international guidelines to assign the lowest level of recommendation to all available agents in the post-IO setting and suggest that the choice of second-line treatment should be based on patient and clinician’s preferences, comorbidities, and overall patient health status. In many countries, reimbursement status may also be an issue, leaving sorafenib as the only reimbursed treatment option [4]. The available data (mostly from retrospective studies) on the efficacy of particular MTAs in the post-IO setting are summarized in Fig. 3 and Table 1.Fig. 3Efficacy of various agents in patients discontinuing immunotherapy for advanced HCC in prospective and retrospective studies (studies with ≥ 10 patients; details in Table 1). A Objective response rates. B Median overall survival. Abbreviations: ICI – Immune checkpoint inhibitors; ORR – Objective response rate; mOS – median overall survival; RECIST 1.1 – Response Evaluation Criteria in Solid Tumors, version 1.1Table 1Prospective and retrospective studies on systemic treatment outcomes for advanced HCC after failure of immunotherapyPublication{Author,yeear,reference)CountryAgent(s)Patient number & type of prior IOTreat-ment lineBCLC stageEHSMVIChild–PughViralORR(RECIST 1.1 unless stated otherwise)PFS(median with 95% CI, unless stated otherwise)OS(median with 95% CI, unless stated otherwise)High grade AE rateYau 2019 (abstract), retrospective [120]Hong KongLenv (11)Sora (10)Rego (6)Axitinib (3)30 post-ICI (mono anti-PD-(L)1, anti-CTLA4)2L: 30% ≥ 3L: 70%EHS + MVI: 83%A: 80%B: 20%HBV: 83%10% (all PR)19.8 mo (4.1-NE)Yau 2021 (abstract), retrospective [121]China, Hong Kong, SingaporeTKI mono (96; mostly sora/lenv)TKI + other (52, incl. 39 with ICI)148 post-ICI ≥ 2LA: 78%HBV: 76%TKI alone: 14.6%TKI + other: 25.0%TTP:TKI mono: 3.9 mo (3.3–4.5)TKI + other: 3.5 mo (1.7–5.2)TKI mono: 8.6 mo (5.8–11.4)TKI + other: 15.1 mo (5.7–24.5)TKI + ICI: 15.1 mo (6.7–23.5)G3 + : 15.5%TKI monoG3 + : 15.6%Yoo 2021, retrospective [110]Korea, Singapore, Hong KongSora (29)Lenv (19)Cabo (1)49 post atezo + bev2LC: 100%39%A: 100%HBV: 84%HCV: 6%All: 6.1%Sora: 0%,Lenv: 15.8%All: 3.4 mo (1.8–4.9)Sora: 2.5 mo (1.3–3.8)Lenv: 6.1 mo (1.6–10.5)All: 14.7 mo (8.1–21.2)Sora: 11.2 mo (2.7–19.6)Lenv:16.6 mo (3.6–29.6)TRAE –G3: 6.3% G4: 0%Chen CT 2022, retrospective [91]TaiwanSora (19)Lenv (9)30 post atezo + bev2L67%80%A: 73%B: 13%C: 13%HBV: 87.0%HCV: 10.0%Sora: 0%Lenv: 11.1%All: 2.5 moSora: 2.6 moLenv: 2.0 moPost-1L survival –All: 8.0 moSora: 8.3 moLenv: 3.8 moSharma 2022, retrospective [103]USA, UK, Taiwan, Germany, Italy, JapanMTA (109):- Sora (49)- Lenv (31)- Cabo (13)- Rego (33)- Ramu (6)Other: RFA/MVA (4), TACE/TARE (19), EBRT (28), CTx (9), surgery (6), other (23)163 post ICI (nivo, pembro, nivo + ipi, STRIDE, atezo, other)2L: 68%3L: 19% ≥ 4L: 8%A: 82%HBV: 36%HCV: 35%All: 12.2 mo (9.3–15.0)Sora: 6.8 mo (6.0–8.4)Lenv: 10.0 mo (8.6–12.8)Cabo: 12.2 mo (6.8–17.6)Rego: 6.2 (5.0–8.0)Liver-directed therapy/surgery: 11.4 mo (8.8–14.8)Hiraoka 2023, retrospective [111]JapanLenv (101)Other MTA (29):Sora (14)Ramu (11)Cabo (4)130 post atezo + bev2LLenv:A: 1%B: 35%C: 60%D: 4%Other MTA:A: 3%B: 48%C: 45%D: 3%Lenv: 46%Other MTA:31%31%Lenv:A: 82%B: 18%C: 1%Other MTA:A: 76%B: 24%Lenv: HBV 16%HCV21%Other MTA: HBV: 24%HCV: 24%Lenv:RECIST: 15.4% (CR 1%)mRECIST: 23.9% (CR 3%)Other MTA:RECIST: 0%Lenv: 4.4 mo (3.3–5.4)Other MTA: 3.5 mo (2.7–10.0)-Sora: 4.0 mo (1.3–15.5)-Ramu: 2.8 mo (1.5-NE)-Cabo: NE (NE-NE)Lenv: 15.7 mo (9.6-NE)Other MTA: 13.6 mo (8.5-NE)-Sora: NE (6.9-NE)-Ramu: 10.0 mo (1.5-NE)-Cabo: NE (NE-NE)Persano 2023 and 2025, retrospective [73, 138]Italy, Japan, Germany, South Korea, PortugalSoraf (51)Lenv (86)Cabo (23)Rego (3)Ramu (14)Other ICI (6)Unspecified (33)TACE (27)243 post atezo + bev2L53%Sora: 14.3 mo (8.8–15.8)Lenv: 18.9 mo (14.8–18.9)Cabo: 12.4 mo (7.2–13.4)Other: NETACE: 15.9 mo (14.6–16.3)Armstrong 2023, retrospective [104]USASora (6)Lenv (12)Rego (6)Cabo (4)Ramu (1)30 post 1L ICI:nivo (12), nivo + ipi (2), atezo + bev (10), cemi (4)durva (1) ≥ 2LHBV: 23%HCV: 47%5.0 mo (3.0–15.0)19.0 mo (13.0–48.0)Aldawoodi 2023 (abstract), retrospective [122]USASora (29)Lenv (52)81 post atezo + bev2LSora: 4.5%Lenv: 8.9%1-year PFS –Sora 21.4%Lenv: 28.3%1-year OS –Sora 75.0%Lenv: 70.2%Asaoka Y 2023 (abstract), retrospective [123]JapanSora (8)Lenv (72)Ramu (6)Cabo (3)89 post atezo + bev2LSora: 0%Lenv: 14.9%Ramu: 0%Cabo: 0%Falette-Puisieux 2023, retrospective [124]FranceSora (41)Lenv (8)Cabo (4)Rego (29)82 post atezo + bev2LB: 7%C: 93%81%38%A: 83%B: 17%HBV: 27%HCV: 9%Sora: 2.6 mo (2.2–3.4)Lenv: 4.4 mo (1.9–5.7)Cabo: 2.8 mo (1.2–9.2)Rego: 2.6 mo (1.9–4.0)Sora: 7.0 mo (4.4–11.6)Lenv: 5.7 mo (4.7–9.1)Cabo: 9.2 mo (1.2–9.2)Rego: 15.8 mo (10.5-NE)RegoG3: 17.2% G4: 0%Other TKI G3: 28.3%G4: 0%Talbot 2023, retrospective [105]UK, Italy, Austria, Germany, USA, Taiwan, JapanICI ± TKI beyond PD (64)TKI mono (108; sora, rego, lenv, cabo)Other: EBRT, TACE (27)199 post-ICI:ICI mono (157),dual ICI (9),ICI + TKI (7),atezo + bev (8)2L: 47% ≥ 3L: 53%58%Post-1L survival –ICI beyond PD: 5.6 mo (1.9–18.4)TKI mono: 10.9 mo (5.0–14.6)ICI beyond PD, then TKI: 9.7 mo (1.4–22.0)Chon 2024, retrospective{Chon, 2024 #98KoreaSora (86)Lenv (40)126 post atezo + bev2LB: 28%C: 87%69%46%A: 72%B: 28%73%Sora: 8.3% (all PR)Lenv: 5.6% (all PR)Sora: 1.8 mo (IQR 1.6–2.3)Lenv: 3.5 mo (IQR 3.0–4.2)Sora: 7.5 mo (IQR 4.7–9.0)Lenv: 10.3 mo (IQR 6.8-NE)TRAE –SoraG3/4: 38.4%LenvG3/4 35.0%Cheon 2025, retrospective [139]Korea, JapanSora (67)Lenv (56)123 post atezo + bev2L87%18%A: 100%74%Sora:RECIST 4.5%mRECIST 8.9%Lenv:RECIST 10.7%mRECIST 30.4%Sora: 3.3 mo (2.5–4.4)Lenv: 4.9 mo (3.7–6.2)Sora: 11.5 mo (7.8–17.1)Lenv: 13.2 mo (7.4–23.3)SoraG3 + : 22.4%LenvG3 + : 26.8%Lombardi 2025, observation prospective [140]Italy, Korea, Taiwan Germany, Austria, USASora (105)Lenv (125)230 post atezo + bev2LA: 7%B: 22%C: 55%ND: 16%54%PVT 77%A: 100%68%Sora: 2.6 moLenv: 5.5 moSora: 7.4 moLenv: 11.9 moWu 2025, retrospective [106]Global- TKI (155):Sora (75),Lenv (71),Cabo (9)- ICI-based (45):Nivo + ipi (8),ICI + TKI (8),ICI mono (8),Atezo + bev beyond PD (6),Unspecified (15)- Other (22)222 post atezo + bev2LA: 3%B: 25%C: 72%47%PVT 33%A: 79%B: 20%C: 1%HBV: 32%HCV: 28%Post-1L survival –All: 9.7 mo (7.3–11.1)TKI: 8.4 moICI-based: 14.9 mo(HR 1.37, p = 0.26)Lenv: 10.2 moSora: 7.0 moLee 2025, retrospective [116]Korea, Hong Kong, Taiwan, Thailand, SingaporeSora (324)Lenv (154)Rego (36)Cabo (12)TKI + ICI (50)605 post atezo + bev2LA: 1%B: 10%C: 74%D: 0.5%ND: 14%21%ND: 20%52%ND: 21%A: 80%B + C: 16%(ND: 4%)HBC: 72%HCV: 6%Sora: 4.5% (all PR)Lenv: 5.9% (all PR)Rego: 11.4% (all PR)Cabo: 0%TKI + ICI: 13% (all PR)Sora: 2.3 mo (2.0–2.6)Lenv: 4.0 mo (3.5–4.9)Rego: 3.6 mo (2.7–5.4)Cabo: 5.4 mo (3.1–6.8)TKI + ICI: 5.4 mo (3.0–8.8)Sora: 6.3 mo (5.3–7.8)Lenv: 8.0 mo (7.0–10.9)Regor: 9.7 mo (15.2-NE)Cabo: 11.2 mo (6.9–15.8)TKI + ICI: 12.6 mo (9.8-NE)Yoo 2026KCSG HB23-04, prospective [125, 157]KoreaLenv50 post atezo + bev2LA: 12%B: 12%C: 76%78%24%A: 100%HBV: 66%HCV: 2%Both: 4%14% (all PR)5.4 mo (4.2–7.1)9.8 mo (8.1-NE)Aoki 2020, retrospective [108]JapanLenv36 post anti-PD1-(L)1 mono2L: 61.1%3-4L: 38.9%B: 25%,C: 75%72%14%A: 100%HBV: 17%HCV: 28%RECIST 1.1: 22.2% (all PR)mRECIST: 55.6% (CR 2.8%)10.0 mo (8.3–11.8)15.8 mo (8.5–23.2)G3/4: 46%Muto 2023, retrospective [107]JapanLenv20 post atezo + bev2L: 95%3L: 5%B: 30%C: 70%55%45%A: 80%B: 20%HBV: 15%HCV: 15%RECIST: 25% (all PR)mRECIST 60% (all PR)6.0 mo (3.0–7.8)10.5 mo (6.9-NE)Yano 2023, retrospective [126]JapanLenv24 post atezo + bev2LB: 33%C: 67%54%33%HBV: 21%HCV: 46%RECIST 1.1: 33.3% (all PR)mRECIST: 54.2% (CR 4.2%)4.0 mo (2.5–6.4)15.3 mo (10.5-NE)Chen 2023, retrospective [127]TaiwanLenv14 post atezo + bev2LC: 100%43%21.4%(all PR)4.2 mo8.3 moTrojan 2020 (abstract), CELESTIAL (subgroup), retrospective [128]GlobalCabo14 post ICI:nivo (7), nivo + ipi (2), pembro (4), durva (1)3LA: 100%3.7 mo (1.9–5.6)7.9 mo (5.1-NE)G3/4: 64.3%Bang 2022, retrospective [50]KoreaCabo93/110 post ICI:nivo (82), atezo + bev (10)durva (2)pembro (3)2L: 1.8%3L: 6.4% ≥ 4L: 81.8%C: 100%95%46%A: 80%B: 20%HBV: 90%HCV: 1.8%All: 3.6% (all PR)All: 3.7 mo (3.1–4.9)Prior ICI (yes vs no): HR 1.26 (0.71–2.25)All: 7.5 mo (5.5–9.5)Prior ICI (yes vs no): HR 1.16 (0.62–2.16)G3/4: 16.4%Storandt 2022, retrospective [112]USACabo26 post ICI:atezo + bev (13)nivo (12)durva (1)2L: 15%3L: 69%4L: 15%A: 72%B: 28%HCV: 42%4% (all PR)2.1 mo (1.3–3.9)7.7 mo (5.3–14.9)G3 + : 26.9%Ahn 2025, retrospective [109, 160]USACabo148 post ICI:- ICI + non-ICI (54, incl. 91.9% atezo + bev),- ICI only, mono/dual (54, incl. 73.3% nivo mono)66 post TKI (incl. 61.8% lenv)2L49%HBV: 7%HCV: 52%Both: 6%TTD –All: 3.2 mo (2.7–4.3)Post-ICI + non-ICI: 2.6 mo (2.1–4.4)Post-ICI only: 3.6 mo (1.9–5.8)Post-TKI:3.8 mo (2.9–4.6)TNTD –All: 7.6 mo (6.3–10.0)Post-ICI + non-ICI: 7.3 mo (5.6-NE)Post-ICI only: 9.1 mo (7.6-NE)Post-TKI: 6.5 mo (5.5–9.5)12-mo OS –All: 62%Post-ICI + non-ICI: 63%Post-ICI only: 60%Post-TKI: 61%Kanzaki 2023, retrospective [49]JapanCabo13/23 post atezo + bev2L: 22%3L: 26% ≥ 4L: 53%B: 30%C: 70%65%22%A: 78%B: 22%HBV: 26%HCV: 30%Whole cohort:4.4% (all PR)All: 3.7 mo (1.5–8.9)All: 4.3 mo (2.1–8.8)AllG3 + : 56.5%Kuzuya 2023; retrospective [89]JapanCabo19 post atezo + bev3L: 47%4L: 37%5L: 11%6L: 5%B: 21%C: 79%58%PVT 37%A: 63%B: 37%HBV: 32%HCV: 21%5.3% (all PR)2.1 mo (1.1–3.9)6.8 mo (1.8–13.4)Nouso 2023; retrospective [87]JapanCabo32/54 post atezo + bev2: 11%3L: 52%4L: 37%B: 35%C: 65%56%15%A: 80%HBV: 22%HCV: 19%All: 3.7% (all PR)All: 4.4 moAll: 6.9 moAllG3/4: 37.0%Chan 2024, prospective [129]Hong Kong, KoreaCabo47 post ICI (26 post atezo + bev)2L: 57%3L: 43%C: 94%A: 100%HBV: 72%HCV: 4%6.4% (all PR)All: 4.1 mo (3.3–5.3)2L: 4.3 mo (3.3–6.7)3L: 4.0 mo (1.4–6.6)All: 9.9 mo (7.3–14.4)2L: 14.3 mo (8.9-NR)3L: 6.6 mo (5.1–12.2)Lai 2025, retrospective [141]TaiwanCabo:mono (30)w/anti-PD-L1 (6), w/CTx (14)37/50 post ICI2-4L: 42% ≥ 5L: 42%B: 48%C: 38%D: 4%48%A: 34%B: 52%C: 14%HBV: 74%HCV: 8%Both: 4%All: 0%All: 3.3 mo (2.6–4.0)All: 6.1 mo (3.1–9.1)Kim YJ 2024, REFINE, observational [35]GlobalRego97 post ICI (mostly anti-PD1 mono)2L: 19% ≥ 3L: 81%3.6 mo (3.0–4.3)9.9 mo (7.5–15.0)El-Khoueiry 2024 (abstract), prospective [130]France, Italy, Korea, USARego + pembro95 post ICI:- Cohort 1: post atezo + bev (68);- Cohort 2: post durva (8), nivo (8), ipi (6), pembro (5)2LB: 22%C: 78%63%28%A: 100%All: 7.4%Cohort 1: 5.9%Cohort 2: 11.1%Cohort 1: 2.8 moCohort 2: 4.2 moG3: 56%G4: 5%TRAE –G3: 37%G4: 3%Cheon 2025, prospective [86]KoreaRego40 post atezo + bev2LB: 3%C: 98%85%25%A: 100%HBV: 77%HCV: 10%10.0% (all PR)3.5 mo (3.0–3.9)10.5 mo (7.1–13.8)Shimose 2023, retrospective [131]JapanRamu12 post atezo + bev2-6L33.3% (all PR)3.9 mo7.6 moG3 + : 41.6%Kuzuya 2022, retrospective [132]JapanRamu13 post atezo + bev2L: 15%3L: 77%5L: 8%B: 31%C: 69%62%PVT 23%A: 69%B: 31%15.4% (all PR)TTP: 3.0 mo (range 1.3–7.7)4.8 mo (range 1.6–12.4)G3: 46.2%G4/5: 0%Finn 2022, REACH-2, prospective [57]USA, Taiwan, China, Hong Kong, Germany, SwitzerlandRamu24/47 post ICI:nivo (6)atezo + bev (4) pembro + lenv (3)nivo + ipi (2)STRIDE (2)durva (2)2L: 87% ≥ 3L: 11%B: 9% C: 92%EHS /MVI 85%A: 100%HBV: 55%HCV: 26%All: 10.6%Post ICI: 16.7%All: 1.7 mo (1.5–4.1)All: 8.7 mo (4.6–12.2)AllG3 + : 57%Kobayahi 2024,R-EVOLUTION, prospective [133]JapanRamu10/17 post ICI:atezo + bev (7)atezo + bev-lenv sequence (3)2: 82%3L: 18%C: 82%53%47%A: 100%HBV: 47%HCV: 24%0%All: 3.7 mo (1.2–4.6)Post atezo + bev:2.7 mo (1.1–4.3)All: 12.0 mo (4.9-NE)Post atezo + bev:NRAllG3 + : 70.6%Post atezo + bevG3 + : 57.1%Wong 2021, retrospective[134]Hong KongNivo + ipi (12)Pembro (13)25 post ICI:nivo (18) nivo + cabo (1) pembro (5) atezo + bev (1)2L: 60%3L: 28%4L: 12%B: 16%,C: 76%D: 8%80%12%A: 68%B: 24%C: 8%HBV: 76%HCV: 8%16% (CR 12%, PR 4%)TTP – 2.96 mo (1.61–4.31)10.9 mo (4.0–17.8)TRAE –G3-4: 8%,G5: 4%Roessler 2023, retrospective [135]GermanyNivo + ipi10 post ICI: atezo + bev (7),nivo + lenv (3)2L: 50%3L: 10% ≥ 4L: 40%C: 80%B: 10%D: 10%70%30%A: 80%B: 10%C: 10%HBV: 20%HCV: 10%RECIST 1.1: 30% (all PR)mRECIST: 30% (CR 10%)2.9 mo(range 0.7–15.8)7.4 mo(range 0.7–18.6)irAE –G3 + : 10%Non-irAE – G3: 30%Lai 2024, retrospective[142]TaiwanDurva + lenv (41) + rego (24)65 post ICI:pembro (56)nivo (9) ≥ 2L69%PVT 43%A: 77%B: 18%HBV: 43%HCV: 3%13.8%(CR 4.6%)5.4 mo (3.1–7.7)9.6 mo (5.5–13.3)Miura R 2024, retrospective [136]JapanSTRIDE16 post atezo + bev2LA: 19%B: 25%C: 56%25%38%A: 50%B: 31%C: 19%38%0%TRAE –G3: 62.5%,G4/5: 0%Mori N, 2024, prospective [143]JapanSTRIDE58 post atezo + bev;11 post TKI ≥ 2LA: 1%B: 43%C: 55%42%16%A: 74%B: 26%HBV: 19%HCV: 32%mRECISTpost atezo + bev: 5%post TKI: 55%Post atezo + bev: 2.9 moPost-TKI: NR6-mo OSPost atezo + bev 80%Post TKI: 67%Sho 2024, retrospective [144]JapanSTRIDE25 post atezo + bev;19 post lenv3 post sora2LB: 53%C: 47%27%PVT 18%A: 93%B: 7%HBV: 27%HCV: 18%Post atezo + bev: 14.3% (all PR)Post TKI: 22.2% (CR 5.6%)Post atezo + bev: 4.0 mo (2.3–7.2)Post TKI: 7.8 mo (2.3-NR)G3 + : 33.3%Yonemoto 2025, retrospective [145]JapanSTRIDE33 post atezo + bev ≥ 2LC: 56%50%22%A: 91%HBV: 21%HCV: 29%9.1%2.0 mo (1.7–4.7)EL-Khoueiry 2026, prospective [146]USAbotensilimab + balstilimab19 post ICI (incl. 11 post atezo + bev, 1 post ipi)2L: 74% ≥ 3L: 26%68%ALBIG1: 53%G2: 47%HBV: 42%HCV: 21%17% (CR 5.6%)4.4 mo (1.4–6.9)12.3 mo (8.4–21.4)TRAE –G3/4: 58%Scheiner 2023, retrospective [137]Germany, Austria, Switzerland, Italy, UK, USA- ICI mono (4)- Dual ICI (12)- ICI + anti-VEGF/TKI (42)Used post atezo + bev:nivo + ipi (8)atezo + cabo (3)pembro + lenv (1)atezo + lenv (1)other (4)53 post ICI:- ICI mono (26);- Dual ICI (1);- ICI + anti-VEGF/TKI (31, incl. 17 atezo + bev) ≥ 2LB: 17%C: 83%59%36%A: 86%B: 14%41%All: 26% (CR 2%)ICI mono: 50% (CR 25%)Dual ICI: 33.3% (all PR)ICI + anti-VEGF/TKI: 19.0% (all PR)TTP –5.2 mo (3.3–7.0)12.0 mo (7.5–16.5)TRAE –G3/4: 17%Chen 2026, retrospective [147]ChinaVarious ICI (tislelizumab, camrelizumab, sintilimab, envafolimab, pembro, atezo, cadonolimab, toripalimab, adebrelimab, penpulimab272 post ICI (camrelizumab, sintilimab, tislelizumab, toripalimab, penpulimab, atezo, pembro):- mono (20%)-w/TKI (68%)-w/bev (12%)2LB: 28%C: 72%39%A: 82%B: 18%B: 88%RECIST 1.1: 18.4% (CR 0.4%)mRECIST: 23.9% (CR 3.3%)EFS: 4.4 (3.8–5.0)14.3 mo (11.7–16.9)Abbreviations: AE Adverse event; Atezo Atezolizumab; Bev Bevacizumab; Cabo Cabozantinib; CR Complete Response; DCR Disease Control Rate; Durva Durvalumab; EBRT External Beam Radiotherapy; EHS Extrahepatic Spread; G Grade; HR Hazard Ratio; Ipi Ipilimumab; irAE Immune-related adverse event; IQR Interquartile Range; L Line of therapy; Lenv Lenvatinib; MTA Molecularly Targeted Agent; MVA Microwave Ablation; MVI Macrovascular Invasion; NA Not Applicable; ND Not Data; NE Not Evaluable; Nivo Nivolumab; OS Overall Survival; Pembro Pembrolizumab; PFS Progression-free survival; PR Partial Response; PVT Portal Vein Thrombosis; Ramu Ramucirumab; Rego Regorafenib; RFA Radiofrequency Ablation; Sora Sorafenib; STRIDE Single Tremelimumab Regular Interval Durvalumab; TACE Transarterial Chemoembolization; TARE Transarterial radioembolization; Treme Tremelimumab; TNTD Time to Next Treatment or Death; TTD Time to discontinuation; TTP Time to progression.[35, 49, 50, 57, 73, 86, 87, 89, 91, 103–105, 107–112, 115, 116, 120–147] Efficacy of various agents in patients discontinuing immunotherapy for advanced HCC in prospective and retrospective studies (studies with ≥ 10 patients; details in Table 1). A Objective response rates. B Median overall survival. Abbreviations: ICI – Immune checkpoint inhibitors; ORR – Objective response rate; mOS – median overall survival; RECIST 1.1 – Response Evaluation Criteria in Solid Tumors, version 1.1 Table 1Prospective and retrospective studies on systemic treatment outcomes for advanced HCC after failure of immunotherapyPublication{Author,yeear,reference)CountryAgent(s)Patient number & type of prior IOTreat-ment lineBCLC stageEHSMVIChild–PughViralORR(RECIST 1.1 unless stated otherwise)PFS(median with 95% CI, unless stated otherwise)OS(median with 95% CI, unless stated otherwise)High grade AE rateYau 2019 (abstract), retrospective [120]Hong KongLenv (11)Sora (10)Rego (6)Axitinib (3)30 post-ICI (mono anti-PD-(L)1, anti-CTLA4)2L: 30% ≥ 3L: 70%EHS + MVI: 83%A: 80%B: 20%HBV: 83%10% (all PR)19.8 mo (4.1-NE)Yau 2021 (abstract), retrospective [121]China, Hong Kong, SingaporeTKI mono (96; mostly sora/lenv)TKI + other (52, incl. 39 with ICI)148 post-ICI ≥ 2LA: 78%HBV: 76%TKI alone: 14.6%TKI + other: 25.0%TTP:TKI mono: 3.9 mo (3.3–4.5)TKI + other: 3.5 mo (1.7–5.2)TKI mono: 8.6 mo (5.8–11.4)TKI + other: 15.1 mo (5.7–24.5)TKI + ICI: 15.1 mo (6.7–23.5)G3 + : 15.5%TKI monoG3 + : 15.6%Yoo 2021, retrospective [110]Korea, Singapore, Hong KongSora (29)Lenv (19)Cabo (1)49 post atezo + bev2LC: 100%39%A: 100%HBV: 84%HCV: 6%All: 6.1%Sora: 0%,Lenv: 15.8%All: 3.4 mo (1.8–4.9)Sora: 2.5 mo (1.3–3.8)Lenv: 6.1 mo (1.6–10.5)All: 14.7 mo (8.1–21.2)Sora: 11.2 mo (2.7–19.6)Lenv:16.6 mo (3.6–29.6)TRAE –G3: 6.3% G4: 0%Chen CT 2022, retrospective [91]TaiwanSora (19)Lenv (9)30 post atezo + bev2L67%80%A: 73%B: 13%C: 13%HBV: 87.0%HCV: 10.0%Sora: 0%Lenv: 11.1%All: 2.5 moSora: 2.6 moLenv: 2.0 moPost-1L survival –All: 8.0 moSora: 8.3 moLenv: 3.8 moSharma 2022, retrospective [103]USA, UK, Taiwan, Germany, Italy, JapanMTA (109):- Sora (49)- Lenv (31)- Cabo (13)- Rego (33)- Ramu (6)Other: RFA/MVA (4), TACE/TARE (19), EBRT (28), CTx (9), surgery (6), other (23)163 post ICI (nivo, pembro, nivo + ipi, STRIDE, atezo, other)2L: 68%3L: 19% ≥ 4L: 8%A: 82%HBV: 36%HCV: 35%All: 12.2 mo (9.3–15.0)Sora: 6.8 mo (6.0–8.4)Lenv: 10.0 mo (8.6–12.8)Cabo: 12.2 mo (6.8–17.6)Rego: 6.2 (5.0–8.0)Liver-directed therapy/surgery: 11.4 mo (8.8–14.8)Hiraoka 2023, retrospective [111]JapanLenv (101)Other MTA (29):Sora (14)Ramu (11)Cabo (4)130 post atezo + bev2LLenv:A: 1%B: 35%C: 60%D: 4%Other MTA:A: 3%B: 48%C: 45%D: 3%Lenv: 46%Other MTA:31%31%Lenv:A: 82%B: 18%C: 1%Other MTA:A: 76%B: 24%Lenv: HBV 16%HCV21%Other MTA: HBV: 24%HCV: 24%Lenv:RECIST: 15.4% (CR 1%)mRECIST: 23.9% (CR 3%)Other MTA:RECIST: 0%Lenv: 4.4 mo (3.3–5.4)Other MTA: 3.5 mo (2.7–10.0)-Sora: 4.0 mo (1.3–15.5)-Ramu: 2.8 mo (1.5-NE)-Cabo: NE (NE-NE)Lenv: 15.7 mo (9.6-NE)Other MTA: 13.6 mo (8.5-NE)-Sora: NE (6.9-NE)-Ramu: 10.0 mo (1.5-NE)-Cabo: NE (NE-NE)Persano 2023 and 2025, retrospective [73, 138]Italy, Japan, Germany, South Korea, PortugalSoraf (51)Lenv (86)Cabo (23)Rego (3)Ramu (14)Other ICI (6)Unspecified (33)TACE (27)243 post atezo + bev2L53%Sora: 14.3 mo (8.8–15.8)Lenv: 18.9 mo (14.8–18.9)Cabo: 12.4 mo (7.2–13.4)Other: NETACE: 15.9 mo (14.6–16.3)Armstrong 2023, retrospective [104]USASora (6)Lenv (12)Rego (6)Cabo (4)Ramu (1)30 post 1L ICI:nivo (12), nivo + ipi (2), atezo + bev (10), cemi (4)durva (1) ≥ 2LHBV: 23%HCV: 47%5.0 mo (3.0–15.0)19.0 mo (13.0–48.0)Aldawoodi 2023 (abstract), retrospective [122]USASora (29)Lenv (52)81 post atezo + bev2LSora: 4.5%Lenv: 8.9%1-year PFS –Sora 21.4%Lenv: 28.3%1-year OS –Sora 75.0%Lenv: 70.2%Asaoka Y 2023 (abstract), retrospective [123]JapanSora (8)Lenv (72)Ramu (6)Cabo (3)89 post atezo + bev2LSora: 0%Lenv: 14.9%Ramu: 0%Cabo: 0%Falette-Puisieux 2023, retrospective [124]FranceSora (41)Lenv (8)Cabo (4)Rego (29)82 post atezo + bev2LB: 7%C: 93%81%38%A: 83%B: 17%HBV: 27%HCV: 9%Sora: 2.6 mo (2.2–3.4)Lenv: 4.4 mo (1.9–5.7)Cabo: 2.8 mo (1.2–9.2)Rego: 2.6 mo (1.9–4.0)Sora: 7.0 mo (4.4–11.6)Lenv: 5.7 mo (4.7–9.1)Cabo: 9.2 mo (1.2–9.2)Rego: 15.8 mo (10.5-NE)RegoG3: 17.2% G4: 0%Other TKI G3: 28.3%G4: 0%Talbot 2023, retrospective [105]UK, Italy, Austria, Germany, USA, Taiwan, JapanICI ± TKI beyond PD (64)TKI mono (108; sora, rego, lenv, cabo)Other: EBRT, TACE (27)199 post-ICI:ICI mono (157),dual ICI (9),ICI + TKI (7),atezo + bev (8)2L: 47% ≥ 3L: 53%58%Post-1L survival –ICI beyond PD: 5.6 mo (1.9–18.4)TKI mono: 10.9 mo (5.0–14.6)ICI beyond PD, then TKI: 9.7 mo (1.4–22.0)Chon 2024, retrospective{Chon, 2024 #98KoreaSora (86)Lenv (40)126 post atezo + bev2LB: 28%C: 87%69%46%A: 72%B: 28%73%Sora: 8.3% (all PR)Lenv: 5.6% (all PR)Sora: 1.8 mo (IQR 1.6–2.3)Lenv: 3.5 mo (IQR 3.0–4.2)Sora: 7.5 mo (IQR 4.7–9.0)Lenv: 10.3 mo (IQR 6.8-NE)TRAE –SoraG3/4: 38.4%LenvG3/4 35.0%Cheon 2025, retrospective [139]Korea, JapanSora (67)Lenv (56)123 post atezo + bev2L87%18%A: 100%74%Sora:RECIST 4.5%mRECIST 8.9%Lenv:RECIST 10.7%mRECIST 30.4%Sora: 3.3 mo (2.5–4.4)Lenv: 4.9 mo (3.7–6.2)Sora: 11.5 mo (7.8–17.1)Lenv: 13.2 mo (7.4–23.3)SoraG3 + : 22.4%LenvG3 + : 26.8%Lombardi 2025, observation prospective [140]Italy, Korea, Taiwan Germany, Austria, USASora (105)Lenv (125)230 post atezo + bev2LA: 7%B: 22%C: 55%ND: 16%54%PVT 77%A: 100%68%Sora: 2.6 moLenv: 5.5 moSora: 7.4 moLenv: 11.9 moWu 2025, retrospective [106]Global- TKI (155):Sora (75),Lenv (71),Cabo (9)- ICI-based (45):Nivo + ipi (8),ICI + TKI (8),ICI mono (8),Atezo + bev beyond PD (6),Unspecified (15)- Other (22)222 post atezo + bev2LA: 3%B: 25%C: 72%47%PVT 33%A: 79%B: 20%C: 1%HBV: 32%HCV: 28%Post-1L survival –All: 9.7 mo (7.3–11.1)TKI: 8.4 moICI-based: 14.9 mo(HR 1.37, p = 0.26)Lenv: 10.2 moSora: 7.0 moLee 2025, retrospective [116]Korea, Hong Kong, Taiwan, Thailand, SingaporeSora (324)Lenv (154)Rego (36)Cabo (12)TKI + ICI (50)605 post atezo + bev2LA: 1%B: 10%C: 74%D: 0.5%ND: 14%21%ND: 20%52%ND: 21%A: 80%B + C: 16%(ND: 4%)HBC: 72%HCV: 6%Sora: 4.5% (all PR)Lenv: 5.9% (all PR)Rego: 11.4% (all PR)Cabo: 0%TKI + ICI: 13% (all PR)Sora: 2.3 mo (2.0–2.6)Lenv: 4.0 mo (3.5–4.9)Rego: 3.6 mo (2.7–5.4)Cabo: 5.4 mo (3.1–6.8)TKI + ICI: 5.4 mo (3.0–8.8)Sora: 6.3 mo (5.3–7.8)Lenv: 8.0 mo (7.0–10.9)Regor: 9.7 mo (15.2-NE)Cabo: 11.2 mo (6.9–15.8)TKI + ICI: 12.6 mo (9.8-NE)Yoo 2026KCSG HB23-04, prospective [125, 157]KoreaLenv50 post atezo + bev2LA: 12%B: 12%C: 76%78%24%A: 100%HBV: 66%HCV: 2%Both: 4%14% (all PR)5.4 mo (4.2–7.1)9.8 mo (8.1-NE)Aoki 2020, retrospective [108]JapanLenv36 post anti-PD1-(L)1 mono2L: 61.1%3-4L: 38.9%B: 25%,C: 75%72%14%A: 100%HBV: 17%HCV: 28%RECIST 1.1: 22.2% (all PR)mRECIST: 55.6% (CR 2.8%)10.0 mo (8.3–11.8)15.8 mo (8.5–23.2)G3/4: 46%Muto 2023, retrospective [107]JapanLenv20 post atezo + bev2L: 95%3L: 5%B: 30%C: 70%55%45%A: 80%B: 20%HBV: 15%HCV: 15%RECIST: 25% (all PR)mRECIST 60% (all PR)6.0 mo (3.0–7.8)10.5 mo (6.9-NE)Yano 2023, retrospective [126]JapanLenv24 post atezo + bev2LB: 33%C: 67%54%33%HBV: 21%HCV: 46%RECIST 1.1: 33.3% (all PR)mRECIST: 54.2% (CR 4.2%)4.0 mo (2.5–6.4)15.3 mo (10.5-NE)Chen 2023, retrospective [127]TaiwanLenv14 post atezo + bev2LC: 100%43%21.4%(all PR)4.2 mo8.3 moTrojan 2020 (abstract), CELESTIAL (subgroup), retrospective [128]GlobalCabo14 post ICI:nivo (7), nivo + ipi (2), pembro (4), durva (1)3LA: 100%3.7 mo (1.9–5.6)7.9 mo (5.1-NE)G3/4: 64.3%Bang 2022, retrospective [50]KoreaCabo93/110 post ICI:nivo (82), atezo + bev (10)durva (2)pembro (3)2L: 1.8%3L: 6.4% ≥ 4L: 81.8%C: 100%95%46%A: 80%B: 20%HBV: 90%HCV: 1.8%All: 3.6% (all PR)All: 3.7 mo (3.1–4.9)Prior ICI (yes vs no): HR 1.26 (0.71–2.25)All: 7.5 mo (5.5–9.5)Prior ICI (yes vs no): HR 1.16 (0.62–2.16)G3/4: 16.4%Storandt 2022, retrospective [112]USACabo26 post ICI:atezo + bev (13)nivo (12)durva (1)2L: 15%3L: 69%4L: 15%A: 72%B: 28%HCV: 42%4% (all PR)2.1 mo (1.3–3.9)7.7 mo (5.3–14.9)G3 + : 26.9%Ahn 2025, retrospective [109, 160]USACabo148 post ICI:- ICI + non-ICI (54, incl. 91.9% atezo + bev),- ICI only, mono/dual (54, incl. 73.3% nivo mono)66 post TKI (incl. 61.8% lenv)2L49%HBV: 7%HCV: 52%Both: 6%TTD –All: 3.2 mo (2.7–4.3)Post-ICI + non-ICI: 2.6 mo (2.1–4.4)Post-ICI only: 3.6 mo (1.9–5.8)Post-TKI:3.8 mo (2.9–4.6)TNTD –All: 7.6 mo (6.3–10.0)Post-ICI + non-ICI: 7.3 mo (5.6-NE)Post-ICI only: 9.1 mo (7.6-NE)Post-TKI: 6.5 mo (5.5–9.5)12-mo OS –All: 62%Post-ICI + non-ICI: 63%Post-ICI only: 60%Post-TKI: 61%Kanzaki 2023, retrospective [49]JapanCabo13/23 post atezo + bev2L: 22%3L: 26% ≥ 4L: 53%B: 30%C: 70%65%22%A: 78%B: 22%HBV: 26%HCV: 30%Whole cohort:4.4% (all PR)All: 3.7 mo (1.5–8.9)All: 4.3 mo (2.1–8.8)AllG3 + : 56.5%Kuzuya 2023; retrospective [89]JapanCabo19 post atezo + bev3L: 47%4L: 37%5L: 11%6L: 5%B: 21%C: 79%58%PVT 37%A: 63%B: 37%HBV: 32%HCV: 21%5.3% (all PR)2.1 mo (1.1–3.9)6.8 mo (1.8–13.4)Nouso 2023; retrospective [87]JapanCabo32/54 post atezo + bev2: 11%3L: 52%4L: 37%B: 35%C: 65%56%15%A: 80%HBV: 22%HCV: 19%All: 3.7% (all PR)All: 4.4 moAll: 6.9 moAllG3/4: 37.0%Chan 2024, prospective [129]Hong Kong, KoreaCabo47 post ICI (26 post atezo + bev)2L: 57%3L: 43%C: 94%A: 100%HBV: 72%HCV: 4%6.4% (all PR)All: 4.1 mo (3.3–5.3)2L: 4.3 mo (3.3–6.7)3L: 4.0 mo (1.4–6.6)All: 9.9 mo (7.3–14.4)2L: 14.3 mo (8.9-NR)3L: 6.6 mo (5.1–12.2)Lai 2025, retrospective [141]TaiwanCabo:mono (30)w/anti-PD-L1 (6), w/CTx (14)37/50 post ICI2-4L: 42% ≥ 5L: 42%B: 48%C: 38%D: 4%48%A: 34%B: 52%C: 14%HBV: 74%HCV: 8%Both: 4%All: 0%All: 3.3 mo (2.6–4.0)All: 6.1 mo (3.1–9.1)Kim YJ 2024, REFINE, observational [35]GlobalRego97 post ICI (mostly anti-PD1 mono)2L: 19% ≥ 3L: 81%3.6 mo (3.0–4.3)9.9 mo (7.5–15.0)El-Khoueiry 2024 (abstract), prospective [130]France, Italy, Korea, USARego + pembro95 post ICI:- Cohort 1: post atezo + bev (68);- Cohort 2: post durva (8), nivo (8), ipi (6), pembro (5)2LB: 22%C: 78%63%28%A: 100%All: 7.4%Cohort 1: 5.9%Cohort 2: 11.1%Cohort 1: 2.8 moCohort 2: 4.2 moG3: 56%G4: 5%TRAE –G3: 37%G4: 3%Cheon 2025, prospective [86]KoreaRego40 post atezo + bev2LB: 3%C: 98%85%25%A: 100%HBV: 77%HCV: 10%10.0% (all PR)3.5 mo (3.0–3.9)10.5 mo (7.1–13.8)Shimose 2023, retrospective [131]JapanRamu12 post atezo + bev2-6L33.3% (all PR)3.9 mo7.6 moG3 + : 41.6%Kuzuya 2022, retrospective [132]JapanRamu13 post atezo + bev2L: 15%3L: 77%5L: 8%B: 31%C: 69%62%PVT 23%A: 69%B: 31%15.4% (all PR)TTP: 3.0 mo (range 1.3–7.7)4.8 mo (range 1.6–12.4)G3: 46.2%G4/5: 0%Finn 2022, REACH-2, prospective [57]USA, Taiwan, China, Hong Kong, Germany, SwitzerlandRamu24/47 post ICI:nivo (6)atezo + bev (4) pembro + lenv (3)nivo + ipi (2)STRIDE (2)durva (2)2L: 87% ≥ 3L: 11%B: 9% C: 92%EHS /MVI 85%A: 100%HBV: 55%HCV: 26%All: 10.6%Post ICI: 16.7%All: 1.7 mo (1.5–4.1)All: 8.7 mo (4.6–12.2)AllG3 + : 57%Kobayahi 2024,R-EVOLUTION, prospective [133]JapanRamu10/17 post ICI:atezo + bev (7)atezo + bev-lenv sequence (3)2: 82%3L: 18%C: 82%53%47%A: 100%HBV: 47%HCV: 24%0%All: 3.7 mo (1.2–4.6)Post atezo + bev:2.7 mo (1.1–4.3)All: 12.0 mo (4.9-NE)Post atezo + bev:NRAllG3 + : 70.6%Post atezo + bevG3 + : 57.1%Wong 2021, retrospective[134]Hong KongNivo + ipi (12)Pembro (13)25 post ICI:nivo (18) nivo + cabo (1) pembro (5) atezo + bev (1)2L: 60%3L: 28%4L: 12%B: 16%,C: 76%D: 8%80%12%A: 68%B: 24%C: 8%HBV: 76%HCV: 8%16% (CR 12%, PR 4%)TTP – 2.96 mo (1.61–4.31)10.9 mo (4.0–17.8)TRAE –G3-4: 8%,G5: 4%Roessler 2023, retrospective [135]GermanyNivo + ipi10 post ICI: atezo + bev (7),nivo + lenv (3)2L: 50%3L: 10% ≥ 4L: 40%C: 80%B: 10%D: 10%70%30%A: 80%B: 10%C: 10%HBV: 20%HCV: 10%RECIST 1.1: 30% (all PR)mRECIST: 30% (CR 10%)2.9 mo(range 0.7–15.8)7.4 mo(range 0.7–18.6)irAE –G3 + : 10%Non-irAE – G3: 30%Lai 2024, retrospective[142]TaiwanDurva + lenv (41) + rego (24)65 post ICI:pembro (56)nivo (9) ≥ 2L69%PVT 43%A: 77%B: 18%HBV: 43%HCV: 3%13.8%(CR 4.6%)5.4 mo (3.1–7.7)9.6 mo (5.5–13.3)Miura R 2024, retrospective [136]JapanSTRIDE16 post atezo + bev2LA: 19%B: 25%C: 56%25%38%A: 50%B: 31%C: 19%38%0%TRAE –G3: 62.5%,G4/5: 0%Mori N, 2024, prospective [143]JapanSTRIDE58 post atezo + bev;11 post TKI ≥ 2LA: 1%B: 43%C: 55%42%16%A: 74%B: 26%HBV: 19%HCV: 32%mRECISTpost atezo + bev: 5%post TKI: 55%Post atezo + bev: 2.9 moPost-TKI: NR6-mo OSPost atezo + bev 80%Post TKI: 67%Sho 2024, retrospective [144]JapanSTRIDE25 post atezo + bev;19 post lenv3 post sora2LB: 53%C: 47%27%PVT 18%A: 93%B: 7%HBV: 27%HCV: 18%Post atezo + bev: 14.3% (all PR)Post TKI: 22.2% (CR 5.6%)Post atezo + bev: 4.0 mo (2.3–7.2)Post TKI: 7.8 mo (2.3-NR)G3 + : 33.3%Yonemoto 2025, retrospective [145]JapanSTRIDE33 post atezo + bev ≥ 2LC: 56%50%22%A: 91%HBV: 21%HCV: 29%9.1%2.0 mo (1.7–4.7)EL-Khoueiry 2026, prospective [146]USAbotensilimab + balstilimab19 post ICI (incl. 11 post atezo + bev, 1 post ipi)2L: 74% ≥ 3L: 26%68%ALBIG1: 53%G2: 47%HBV: 42%HCV: 21%17% (CR 5.6%)4.4 mo (1.4–6.9)12.3 mo (8.4–21.4)TRAE –G3/4: 58%Scheiner 2023, retrospective [137]Germany, Austria, Switzerland, Italy, UK, USA- ICI mono (4)- Dual ICI (12)- ICI + anti-VEGF/TKI (42)Used post atezo + bev:nivo + ipi (8)atezo + cabo (3)pembro + lenv (1)atezo + lenv (1)other (4)53 post ICI:- ICI mono (26);- Dual ICI (1);- ICI + anti-VEGF/TKI (31, incl. 17 atezo + bev) ≥ 2LB: 17%C: 83%59%36%A: 86%B: 14%41%All: 26% (CR 2%)ICI mono: 50% (CR 25%)Dual ICI: 33.3% (all PR)ICI + anti-VEGF/TKI: 19.0% (all PR)TTP –5.2 mo (3.3–7.0)12.0 mo (7.5–16.5)TRAE –G3/4: 17%Chen 2026, retrospective [147]ChinaVarious ICI (tislelizumab, camrelizumab, sintilimab, envafolimab, pembro, atezo, cadonolimab, toripalimab, adebrelimab, penpulimab272 post ICI (camrelizumab, sintilimab, tislelizumab, toripalimab, penpulimab, atezo, pembro):- mono (20%)-w/TKI (68%)-w/bev (12%)2LB: 28%C: 72%39%A: 82%B: 18%B: 88%RECIST 1.1: 18.4% (CR 0.4%)mRECIST: 23.9% (CR 3.3%)EFS: 4.4 (3.8–5.0)14.3 mo (11.7–16.9)Abbreviations: AE Adverse event; Atezo Atezolizumab; Bev Bevacizumab; Cabo Cabozantinib; CR Complete Response; DCR Disease Control Rate; Durva Durvalumab; EBRT External Beam Radiotherapy; EHS Extrahepatic Spread; G Grade; HR Hazard Ratio; Ipi Ipilimumab; irAE Immune-related adverse event; IQR Interquartile Range; L Line of therapy; Lenv Lenvatinib; MTA Molecularly Targeted Agent; MVA Microwave Ablation; MVI Macrovascular Invasion; NA Not Applicable; ND Not Data; NE Not Evaluable; Nivo Nivolumab; OS Overall Survival; Pembro Pembrolizumab; PFS Progression-free survival; PR Partial Response; PVT Portal Vein Thrombosis; Ramu Ramucirumab; Rego Regorafenib; RFA Radiofrequency Ablation; Sora Sorafenib; STRIDE Single Tremelimumab Regular Interval Durvalumab; TACE Transarterial Chemoembolization; TARE Transarterial radioembolization; Treme Tremelimumab; TNTD Time to Next Treatment or Death; TTD Time to discontinuation; TTP Time to progression.[35, 49, 50, 57, 73, 86, 87, 89, 91, 103–105, 107–112, 115, 116, 120–147] Prospective and retrospective studies on systemic treatment outcomes for advanced HCC after failure of immunotherapy Lenv (11) Sora (10) Rego (6) Axitinib (3) 2L: 30% ≥ 3L: 70% A: 80% B: 20% TKI mono (96; mostly sora/lenv) TKI + other (52, incl. 39 with ICI) TKI alone: 14.6% TKI + other: 25.0% TTP: TKI mono: 3.9 mo (3.3–4.5) TKI + other: 3.5 mo (1.7–5.2) TKI mono: 8.6 mo (5.8–11.4) TKI + other: 15.1 mo (5.7–24.5) TKI + ICI: 15.1 mo (6.7–23.5) G3 + : 15.5% TKI mono G3 + : 15.6% Sora (29) Lenv (19) Cabo (1) HBV: 84% HCV: 6% All: 6.1% Sora: 0%, Lenv: 15.8% All: 3.4 mo (1.8–4.9) Sora: 2.5 mo (1.3–3.8) Lenv: 6.1 mo (1.6–10.5) All: 14.7 mo (8.1–21.2) Sora: 11.2 mo (2.7–19.6) Lenv: 16.6 mo (3.6–29.6) TRAE – G3: 6.3% G4: 0% Sora (19) Lenv (9) A: 73% B: 13% C: 13% HBV: 87.0% HCV: 10.0% Sora: 0% Lenv: 11.1% All: 2.5 mo Sora: 2.6 mo Lenv: 2.0 mo Post-1L survival – All: 8.0 mo Sora: 8.3 mo Lenv: 3.8 mo MTA (109): - Sora (49) - Lenv (31) - Cabo (13) - Rego (33) - Ramu (6) Other: RFA/MVA (4), TACE/TARE (19), EBRT (28), CTx (9), surgery (6), other (23) 2L: 68% 3L: 19% ≥ 4L: 8% HBV: 36% HCV: 35% All: 12.2 mo (9.3–15.0) Sora: 6.8 mo (6.0–8.4) Lenv: 10.0 mo (8.6–12.8) Cabo: 12.2 mo (6.8–17.6) Rego: 6.2 (5.0–8.0) Liver-directed therapy/surgery: 11.4 mo (8.8–14.8) Lenv (101) Other MTA (29): Sora (14) Ramu (11) Cabo (4) Lenv: A: 1% B: 35% C: 60% D: 4% Other MTA: A: 3% B: 48% C: 45% D: 3% Lenv: 46% Other MTA: 31% Lenv: A: 82% B: 18% C: 1% Other MTA: A: 76% B: 24% Lenv: HBV 16% HCV 21% Other MTA: HBV: 24% HCV: 24% Lenv: RECIST: 15.4% (CR 1%) mRECIST: 23.9% (CR 3%) Other MTA: RECIST: 0% Lenv: 4.4 mo (3.3–5.4) Other MTA: 3.5 mo (2.7–10.0) -Sora: 4.0 mo (1.3–15.5) -Ramu: 2.8 mo (1.5-NE) -Cabo: NE (NE-NE) Lenv: 15.7 mo (9.6-NE) Other MTA: 13.6 mo (8.5-NE) -Sora: NE (6.9-NE) -Ramu: 10.0 mo (1.5-NE) -Cabo: NE (NE-NE) Soraf (51) Lenv (86) Cabo (23) Rego (3) Ramu (14) Other ICI (6) Unspecified (33) TACE (27) Sora: 14.3 mo (8.8–15.8) Lenv: 18.9 mo (14.8–18.9) Cabo: 12.4 mo (7.2–13.4) Other: NE TACE: 15.9 mo (14.6–16.3) Sora (6) Lenv (12) Rego (6) Cabo (4) Ramu (1) 30 post 1L ICI: nivo (12), nivo + ipi (2), atezo + bev (10), cemi (4) durva (1) HBV: 23% HCV: 47% Sora (29) Lenv (52) Sora: 4.5% Lenv: 8.9% 1-year PFS – Sora 21.4% Lenv: 28.3% 1-year OS – Sora 75.0% Lenv: 70.2% Sora (8) Lenv (72) Ramu (6) Cabo (3) Sora: 0% Lenv: 14.9% Ramu: 0% Cabo: 0% Sora (41) Lenv (8) Cabo (4) Rego (29) B: 7% C: 93% A: 83% B: 17% HBV: 27% HCV: 9% Sora: 2.6 mo (2.2–3.4) Lenv: 4.4 mo (1.9–5.7) Cabo: 2.8 mo (1.2–9.2) Rego: 2.6 mo (1.9–4.0) Sora: 7.0 mo (4.4–11.6) Lenv: 5.7 mo (4.7–9.1) Cabo: 9.2 mo (1.2–9.2) Rego: 15.8 mo (10.5-NE) Rego G3: 17.2% G4: 0% Other TKI G3: 28.3% G4: 0% ICI ± TKI beyond PD (64) TKI mono (108; sora, rego, lenv, cabo) Other: EBRT, TACE (27) 199 post-ICI: ICI mono (157), dual ICI (9), ICI + TKI (7), atezo + bev (8) 2L: 47% ≥ 3L: 53% Post-1L survival – ICI beyond PD: 5.6 mo (1.9–18.4) TKI mono: 10.9 mo (5.0–14.6) ICI beyond PD, then TKI: 9.7 mo (1.4–22.0) Sora (86) Lenv (40) B: 28% C: 87% A: 72% B: 28% Sora: 8.3% (all PR) Lenv: 5.6% (all PR) Sora: 1.8 mo (IQR 1.6–2.3) Lenv: 3.5 mo (IQR 3.0–4.2) Sora: 7.5 mo (IQR 4.7–9.0) Lenv: 10.3 mo (IQR 6.8-NE) TRAE – Sora G3/4: 38.4% Lenv G3/4 35.0% Sora (67) Lenv (56) Sora: RECIST 4.5% mRECIST 8.9% Lenv: RECIST 10.7% mRECIST 30.4% Sora: 3.3 mo (2.5–4.4) Lenv: 4.9 mo (3.7–6.2) Sora: 11.5 mo (7.8–17.1) Lenv: 13.2 mo (7.4–23.3) Sora G3 + : 22.4% Lenv G3 + : 26.8% Sora (105) Lenv (125) A: 7% B: 22% C: 55% ND: 16% Sora: 2.6 mo Lenv: 5.5 mo Sora: 7.4 mo Lenv: 11.9 mo - TKI (155): Sora (75), Lenv (71), Cabo (9) - ICI-based (45): Nivo + ipi (8), ICI + TKI (8), ICI mono (8), Atezo + bev beyond PD (6), Unspecified (15) - Other (22) A: 3% B: 25% C: 72% A: 79% B: 20% C: 1% HBV: 32% HCV: 28% Post-1L survival – All: 9.7 mo (7.3–11.1) TKI: 8.4 mo ICI-based: 14.9 mo (HR 1.37, p = 0.26) Lenv: 10.2 mo Sora: 7.0 mo Sora (324) Lenv (154) Rego (36) Cabo (12) TKI + ICI (50) A: 1% B: 10% C: 74% D: 0.5% ND: 14% 21% ND: 20% 52% ND: 21% A: 80% B + C: 16% (ND: 4%) HBC: 72% HCV: 6% Sora: 4.5% (all PR) Lenv: 5.9% (all PR) Rego: 11.4% (all PR) Cabo: 0% TKI + ICI: 13% (all PR) Sora: 2.3 mo (2.0–2.6) Lenv: 4.0 mo (3.5–4.9) Rego: 3.6 mo (2.7–5.4) Cabo: 5.4 mo (3.1–6.8) TKI + ICI: 5.4 mo (3.0–8.8) Sora: 6.3 mo (5.3–7.8) Lenv: 8.0 mo (7.0–10.9) Regor: 9.7 mo (15.2-NE) Cabo: 11.2 mo (6.9–15.8) TKI + ICI: 12.6 mo (9.8-NE) Yoo 2026 KCSG HB23-04, prospective [125, 157] A: 12% B: 12% C: 76% HBV: 66% HCV: 2% Both: 4% 2L: 61.1% 3-4L: 38.9% B: 25%, C: 75% HBV: 17% HCV: 28% RECIST 1.1: 22.2% (all PR) mRECIST: 55.6% (CR 2.8%) 2L: 95% 3L: 5% B: 30% C: 70% A: 80% B: 20% HBV: 15% HCV: 15% RECIST: 25% (all PR) mRECIST 60% (all PR) B: 33% C: 67% HBV: 21% HCV: 46% RECIST 1.1: 33.3% (all PR) mRECIST: 54.2% (CR 4.2%) 21.4% (all PR) 14 post ICI: nivo (7), nivo + ipi (2), pembro (4), durva (1) 93/110 post ICI: nivo (82), atezo + bev (10) durva (2) pembro (3) 2L: 1.8% 3L: 6.4% ≥ 4L: 81.8% A: 80% B: 20% HBV: 90% HCV: 1.8% All: 3.7 mo (3.1–4.9) Prior ICI (yes vs no): HR 1.26 (0.71–2.25) All: 7.5 mo (5.5–9.5) Prior ICI (yes vs no): HR 1.16 (0.62–2.16) 26 post ICI: atezo + bev (13) nivo (12) durva (1) 2L: 15% 3L: 69% 4L: 15% A: 72% B: 28% 148 post ICI: - ICI + non-ICI (54, incl. 91.9% atezo + bev), - ICI only, mono/dual (54, incl. 73.3% nivo mono) 66 post TKI (incl. 61.8% lenv) HBV: 7% HCV: 52% Both: 6% TTD – All: 3.2 mo (2.7–4.3) Post-ICI + non-ICI: 2.6 mo (2.1–4.4) Post-ICI only: 3.6 mo (1.9–5.8) Post-TKI: 3.8 mo (2.9–4.6) TNTD – All: 7.6 mo (6.3–10.0) Post-ICI + non-ICI: 7.3 mo (5.6-NE) Post-ICI only: 9.1 mo (7.6-NE) Post-TKI: 6.5 mo (5.5–9.5) 12-mo OS – All: 62% Post-ICI + non-ICI: 63% Post-ICI only: 60% Post-TKI: 61% 2L: 22% 3L: 26% ≥ 4L: 53% B: 30% C: 70% A: 78% B: 22% HBV: 26% HCV: 30% Whole cohort: 4.4% (all PR) All G3 + : 56.5% 3L: 47% 4L: 37% 5L: 11% 6L: 5% B: 21% C: 79% A: 63% B: 37% HBV: 32% HCV: 21% 2: 11% 3L: 52% 4L: 37% B: 35% C: 65% HBV: 22% HCV: 19% All G3/4: 37.0% 2L: 57% 3L: 43% HBV: 72% HCV: 4% All: 4.1 mo (3.3–5.3) 2L: 4.3 mo (3.3–6.7) 3L: 4.0 mo (1.4–6.6) All: 9.9 mo (7.3–14.4) 2L: 14.3 mo (8.9-NR) 3L: 6.6 mo (5.1–12.2) Cabo: mono (30) w/anti-PD-L1 (6), w/CTx (14) 2-4L: 42% ≥ 5L: 42% B: 48% C: 38% D: 4% A: 34% B: 52% C: 14% HBV: 74% HCV: 8% Both: 4% 2L: 19% ≥ 3L: 81% 95 post ICI: - Cohort 1: post atezo + bev (68); - Cohort 2: post durva (8), nivo (8), ipi (6), pembro (5) B: 22% C: 78% All: 7.4% Cohort 1: 5.9% Cohort 2: 11.1% Cohort 1: 2.8 mo Cohort 2: 4.2 mo G3: 56% G4: 5% TRAE – G3: 37% G4: 3% B: 3% C: 98% HBV: 77% HCV: 10% 2L: 15% 3L: 77% 5L: 8% B: 31% C: 69% A: 69% B: 31% G3: 46.2% G4/5: 0% 24/47 post ICI: nivo (6) atezo + bev (4) pembro + lenv (3) nivo + ipi (2) STRIDE (2) durva (2) HBV: 55% HCV: 26% All: 10.6% Post ICI: 16.7% All G3 + : 57% Kobayahi 2024, R-EVOLUTION, prospective [133] 10/17 post ICI: atezo + bev (7) atezo + bev-lenv sequence (3) 2: 82% 3L: 18% HBV: 47% HCV: 24% All: 3.7 mo (1.2–4.6) Post atezo + bev: 2.7 mo (1.1–4.3) All: 12.0 mo (4.9-NE) Post atezo + bev: NR All G3 + : 70.6% Post atezo + bev G3 + : 57.1% Nivo + ipi (12) Pembro (13) 25 post ICI: nivo (18) nivo + cabo (1) pembro (5) atezo + bev (1) 2L: 60% 3L: 28% 4L: 12% B: 16%, C: 76% D: 8% A: 68% B: 24% C: 8% HBV: 76% HCV: 8% TRAE – G3-4: 8%, G5: 4% 10 post ICI: atezo + bev (7), nivo + lenv (3) 2L: 50% 3L: 10% ≥ 4L: 40% C: 80% B: 10% D: 10% A: 80% B: 10% C: 10% HBV: 20% HCV: 10% RECIST 1.1: 30% (all PR) mRECIST: 30% (CR 10%) 2.9 mo (range 0.7–15.8) 7.4 mo (range 0.7–18.6) irAE – G3 + : 10% Non-irAE – G3: 30% Durva + lenv (41) + rego (24) 65 post ICI: pembro (56) nivo (9) A: 77% B: 18% HBV: 43% HCV: 3% 13.8% (CR 4.6%) A: 19% B: 25% C: 56% A: 50% B: 31% C: 19% TRAE – G3: 62.5%, G4/5: 0% 58 post atezo + bev; 11 post TKI A: 1% B: 43% C: 55% A: 74% B: 26% HBV: 19% HCV: 32% mRECIST post atezo + bev: 5% post TKI: 55% Post atezo + bev: 2.9 mo Post-TKI: NR 6-mo OS Post atezo + bev 80% Post TKI: 67% 25 post atezo + bev; 19 post lenv 3 post sora B: 53% C: 47% A: 93% B: 7% HBV: 27% HCV: 18% Post atezo + bev: 14.3% (all PR) Post TKI: 22.2% (CR 5.6%) Post atezo + bev: 4.0 mo (2.3–7.2) Post TKI: 7.8 mo (2.3-NR) HBV: 21% HCV: 29% 2L: 74% ≥ 3L: 26% ALBI G1: 53% G2: 47% HBV: 42% HCV: 21% TRAE – G3/4: 58% - ICI mono (4) - Dual ICI (12) - ICI + anti-VEGF/TKI (42) Used post atezo + bev: nivo + ipi (8) atezo + cabo (3) pembro + lenv (1) atezo + lenv (1) other (4) 53 post ICI: - ICI mono (26); - Dual ICI (1); - ICI + anti-VEGF/TKI (31, incl. 17 atezo + bev) B: 17% C: 83% A: 86% B: 14% All: 26% (CR 2%) ICI mono: 50% (CR 25%) Dual ICI: 33.3% (all PR) ICI + anti-VEGF/TKI: 19.0% (all PR) TTP – 5.2 mo (3.3–7.0) TRAE – G3/4: 17% 272 post ICI (camrelizumab, sintilimab, tislelizumab, toripalimab, penpulimab, atezo, pembro): - mono (20%) -w/TKI (68%) -w/bev (12%) B: 28% C: 72% A: 82% B: 18% RECIST 1.1: 18.4% (CR 0.4%) mRECIST: 23.9% (CR 3.3%) Abbreviations: AE Adverse event; Atezo Atezolizumab; Bev Bevacizumab; Cabo Cabozantinib; CR Complete Response; DCR Disease Control Rate; Durva Durvalumab; EBRT External Beam Radiotherapy; EHS Extrahepatic Spread; G Grade; HR Hazard Ratio; Ipi Ipilimumab; irAE Immune-related adverse event; IQR Interquartile Range; L Line of therapy; Lenv Lenvatinib; MTA Molecularly Targeted Agent; MVA Microwave Ablation; MVI Macrovascular Invasion; NA Not Applicable; ND Not Data; NE Not Evaluable; Nivo Nivolumab; OS Overall Survival; Pembro Pembrolizumab; PFS Progression-free survival; PR Partial Response; PVT Portal Vein Thrombosis; Ramu Ramucirumab; Rego Regorafenib; RFA Radiofrequency Ablation; Sora Sorafenib; STRIDE Single Tremelimumab Regular Interval Durvalumab; TACE Transarterial Chemoembolization; TARE Transarterial radioembolization; Treme Tremelimumab; TNTD Time to Next Treatment or Death; TTD Time to discontinuation; TTP Time to progression.[35, 49, 50, 57, 73, 86, 87, 89, 91, 103–105, 107–112, 115, 116, 120–147] Sorafenib and Lenvatinib In clinical practice, sorafenib and lenvatinib are the most commonly used agents for patients in the post-IO setting [4, 73, 103, 110, 116, 121]. The advantages of traditional first-line TKIs include the potential for subsequent MTA use (e.g., cabozantinib, regorafenib, or ramucirumab) in the third-line setting and the low cost of sorafenib therapy. The main theoretical rationale for sequencing sorafenib or lenvatinib after immunotherapy lies in their potential to inhibit VEGFR signaling, thereby counteracting liver carcinogenesis, immune resistance, and VEGF-A–independent mechanisms of neoangiogenesis [93, 97–100]. Furthermore, preclinical evidence suggests that β-catenin–active HCC cells exhibit greater sensitivity to sorafenib than β-catenin wild-type cells [148]. Given that elevated WNT/β-catenin signaling in HCC is associated with resistance to ICI-based only regimens (i.e., without bevacizumab), it may be proposed that sorafenib may be particularly beneficial in CTNNB1-mutant HCC that has not been previously exposed to antiangiogenic agents. Moreover, high expression of neuropilin-1 (NRP1), the VEGFR and MET co-receptor, has been associated with inferior outcomes in HCC patients treated with atezolizumab plus bevacizumab, however, NRP1 expression did not impact the magnitude of benefit from sorafenib [149]. On the other hand, in other studies sustained NRP1 expression has been associated with lenvatinib resistance [150, 151]. Lenvatinib, in comparison to sorafenib, may offer potential advantages in IO–refractory tumors due to its substantially stronger inhibition of VEGFRs and potent activity against FGFRs [92]. Overexpression of FGFR2 and FGFR3 has been associated with more aggressive HCC biology, while FGF19 amplification or overexpression (observed in 14% and 50% of HCC cases, respectively) correlates with poorer prognosis [152]. Genomic amplification of the FGF/FGFR axis members has been identified as a predictive biomarker of immunotherapy resistance across multiple tumor types [153]. Furthermore, preclinical evidence suggests that FGFR1 signaling in HCC may drive resistance to ICIs by promoting macrophage infiltration and their polarization toward an immunosuppressive M2 phenotype [154]. Interactions between FGF19 and FGFR4 appear to play a particularly prominent role in HCC progression and resistance to TKIs. It has demonstrated that FGF19–FGFR4 signaling promotes upregulation of PD-L1 expression, thereby contributing to immune escape [102, 152, 155]. Clinically, outcomes with sorafenib after immunotherapy remain modest. In the retrospective studies, sorafenib following first-line immunotherapy provided ORR ranging from 0 to 8%, mPFS from 1.8 to 4.0 months, and mOS from 6.3 to 14.2 months [73, 106, 110, 111, 115, 116, 123, 124, 140]. These OS outcomes are broadly comparable to those reported for sorafenib in the first-line setting [6, 7, 19]. Lenvatinib is most frequently used following immunotherapy in East Asian patients, particularly in Japan and South Korea. Retrospective studies using this agent reported ORR ranging from 5.6% to 33.3%, mPFS from 3.5 to 10.0 months, and mOS from 8.0 to 18.9 months [60, 73, 106–108, 110, 115, 116, 121–124, 126, 127, 138–140, 156]. A recent prospective, single-arm, multicenter Korean phase II KCSG HB23-04 trial evaluated second-line lenvatinib following atezolizumab-bevacizumab in 50 patients. The study reported an ORR of 14% per RECIST 1.1 criteria, mPFS of 5.4 months (95% CI, 4.2–7.1), and mOS of 9.8 months (95% CI, 8.1–not assessable). Grade 3–4 AEs occurred in 46% of patients, and were numerically less frequent than in the first-line REFLECT trial (75%), possible due to higher dose reduction rate in the KCSG HB23-04 study [8, 157]. The efficacy of lenvatinib in this trial was not influenced by etiology (viral vs. non-viral) or by the duration of prior treatment with atezolizumab plus bevacizumab [157]. Retrospective comparisons of sorafenib and lenvatinib following immunotherapy have produced findings that are not entirely consistent. A small study using these agents, mostly in the third line or later, reported significantly better outcomes for lenvatinib, including a higher DCR (51.5% vs. 25.8%), longer median TTP (6.3 vs. 1.8 months), and improved mOS (12.0 vs. 5.9 months) [121]. In the second-line setting in Asia, specifically after progression on or intolerance to atezolizumab-bevacizumab, lenvatinib outperformed sorafenib in ORR (15.8% vs. 0%) and mPFS (6.1 vs. 2.5 months), though the difference in mOS (16.6 vs. 11.2 months) was insignificant [110]. Similar findings in the same setting were reported in 123 Korean and Japanese patients, where second-line lenvatinib improved mRECIST ORR and PFS but did not significantly improve OS after propensity-score matching (median 11.4 vs. 7.9 months; p = 0.197) [139]. In a large retrospective analysis involving 233 patients treated after the failure of atezolizumab-bevacizumab, relative to sorafenib, HRs for death with lenvatinib and cabozantinib were 0.50 (95% CI, 0.27–0.90) and 1.29 (95% CI, 0.55–3.01), respectively. In the update report, patients receiving lenvatinib achieved an mOS of 18.9 months compared to 14.3 months with sorafenib and 12.4 months with cabozantinib, with the superiority of lenvatinib sustained in multivariate analysis and all subgroups [73, 138]. Similarly, another international retrospective study found that, among patients who experienced progression on atezolizumab-bevacizumab, second-line treatment with lenvatinib was associated with a numerically longer survival compared to sorafenib (mOS 10.2 vs. 7.0 months; HR 0.61; 95% CI, 0.37–1.01) [106]. An Asia–Pacific study reported consistent findings, with lenvatinib showing longer mPFS (4.0 vs. 2.3 months; HR 0.53) and mOS (8.0 vs. 6.3 months; HR 0.75) than sorafenib after discontinuation of atezolizumab and bevacizumab, with the PFS benefit also confirmed in multivariable analysis [116]. A large multicentre observational LEVIATHAN study of Italian and Asian patients who progressed on atezolizumab plus bevacizumab showed that lenvatinib or sorafenib was superior as second-line therapy, improving DCR (70.2% vs. 35.7%), PFS (median 5.5 vs. 2.6 months; HR 0.41), and OS (median 11.9 vs. 7.4 months; HR 0.67). These benefits persisted in the propensity score–matched cohort and were also observed in patients with primary resistance to immunotherapy [140]. On the other hand, both a simulation derived from randomized phase III trial outcomes and an analysis of data from the US Veterans Health Administration database suggested that sorafenib and lenvatinib, administered after first-line atezolizumab-bevacizumab, yield comparable OS outcomes [122, 158]. Two other retrospective studies also found no significant differences in PFS and OS between lenvatinib and alternative TKIs (primarily sorafenib or regorafenib), albeit lenvatinib produced higher ORR [91, 111]. Finally, a small propensity-matched analysis showed that lenvatinib was superior to sorafenib in terms of DCR (66.7% vs. 22.2%) and PFS (mPFS 3.5 vs. 1.8 months), but not ORR (5.6% vs. 8.3%) and OS (mOS 10.3 vs. 7.5 months) [115]. Collectively, most studies indicate improved tumor response and progression-related outcomes with lenvatinib compared to sorafenib in the post-IO setting, although the suggested OS advantage remains to be definitively validated. Regarding treatment safety, similar rates of any grade and grade 3 or 4 AEs were reported for lenvatinib and sorafenib (80% vs. 87.2%, and 35.0% vs. 38.4%, respectively) [115]. However, sorafenib was associated with a slightly higher rate of toxicity-related treatment discontinuations and a significantly higher incidence of hand-foot syndrome, whereas lenvatinib more frequently caused low-grade fatigue and hypertension; nonetheless, high-grade events for both toxicities were more common with sorafenib [110, 122]. Cabozantinib Cabozantinib may represent a rational option following immunotherapy, owing to its potent VEGFR2 inhibition and additional activity against MET, a receptor frequently upregulated in aHCC [102]. Preclinical data suggest MET activation contributes to resistance to sorafenib and lenvatinib through redundant MAPK signaling [102]. However, the role HGF/c-MET axis inhibition in the IO-refractory setting remains unclear. In HCC cell lines and murine models, MET inhibition has been shown to promote stabilization and upregulation of PD-L1, potentially facilitating tumor immune evasion [159]. The clinical relevance of this finding remains uncertain, as the combination of cabozantinib with atezolizumab in the first-line treatment of aHCC did not demonstrate synergistic activity in the COSMIC-312 trial [19]. Moreover, AXL upregulation, common in sorafenib-resistant HCC, has been linked in preclinical HCC models to immunotherapy resistance via suppression of TNF-α and STING–type I interferon pathways, thereby supporting the use of cabozantinib in the post-IO setting [42]. The post-IO efficacy of cabozantinib in aHCC was recently evaluated in a prospective, single-arm, multicenter phase II trial involving 47 patients who had previously undergone ICI-based treatment (56% with atezolizumab-bevacizumab) and up to two lines of systemic therapy (57% with one and 43% with two lines) [129]. The mPFS was 4.1 months, mOS 9.9 months, and the one-year OS 45.3%. Objective responses were observed in 6.4% of patients, and the DCR was 76.6%. In patients treated with cabozantinib as second-line therapy, the mPFS and mOS were 4.3 months (95% CI, 3.3–6.7) and 14.3 months (95% CI, 8.9–NR), respectively, outperforming outcomes achieved in the third-line setting (HR for OS: 0.37; 95% CI, 0.15–0.90). Cabozantinib was relatively well-tolerated, with a toxicity profile consistent with prior studies. Dose reductions were required in 74.5% of patients, with a median daily dose of 40 mg. The clinical activity of cabozantinib following immunotherapy was also examined in a subgroup analysis of the registrational phase III CELESTIAL trial, which enrolled patients who had received up to two prior regimens (including sorafenib) [128]. Among 14 patients previously exposed to ICIs — mostly single-agent anti-PD-(L)1 antibodies — the mOS and mPFS were 7.9 months (95% CI, 5.1–NE) and 3.7 months (95% CI, 1.9–5.6), respectively. These oncological outcomes and the safety profile closely resembled those observed in the broader population of patients receiving cabozantinib in the third-line setting. In retrospective studies that evaluated the efficacy of cabozantinib after immunotherapy, the ORR ranged from 0 to 5.3%, DCR from 52.6 to 90.9%, mPFS from 2.1 to 5.4 months, and mOS from 6.8 to 12.4 months [50, 73, 80, 89, 112, 116]. In some studies, survival appeared to be inferior to that observed in the CELESTIAL trial (mOS of 10.2 months), likely reflecting the real-world practice where cabozantinib is frequently administered as a third-line or later therapy, often in patients with poorer liver function [112]. Time to subsequent treatment or death, a proxy for PFS, was not significantly influenced by the type of first-line therapy, whether TKI, ICI plus non-ICI combinations, or ICI-only regimens. Retrospective analyses confirmed the safety of cabozantinib across different lines of post-IO therapy [50, 109, 160]. Regorafenib Regorafenib represents an attractive therapeutic option in the post-IO setting, owing to its distinctive spectrum of molecular targets. Beyond inhibition of VEGFR and PDGFR, it also disrupts the angiopoietin–TIE2 axis, thereby providing a dual blockade of vascular proliferation and sprouting [161]. In addition, regorafenib uniquely targets colony-stimulating factor 1 receptor (CSF1R), inhibition of which promotes a shift towards pro-inflammatory M1-like polarization of tumor-associated macrophages [162, 163]. These effects suggest that regorafenib might bypass tumor immune evasion mechanism by comprehensively normalizing tumor vasculature, reprogramming the tumor microenvironment, and directly inhibiting multiple oncogenic signaling pathways. The prospective Korean phase II REGONEXT trial enrolled 40 patients who had progressed on atezolizumab-bevacizumab, administering second-line regorafenib until progression or intolerable toxicity. The primary endpoint, mPFS, was 3.5 months (95% CI, 3.0–4.0), while mOS was 9.7 months (95% CI, 8.3–11.1), with an ORR of 10.0% and DCR of 82.5%. Outcomes were significantly better in patients who had received at least four cycles of atezolizumab-bevacizumab compared to those with shorter immunotherapy duration [86]. The activity of regorafenib in the post-IO setting was also evaluated in the prospective observational REFINE study [35]. Among 1,005 patients who received regorafenib, nearly all previously treated with sorafenib, and 10% had prior exposure to immunotherapy (primarily single-agent ICIs). In this subset, mOS was 9.9 months (95% CI, 7.5–15.0), and mPFS was 3.6 months (95% CI, 3.0–4.3). Similar findings have been reported in retrospective studies. A Korean study of 440 patients previously treated with systemic therapies and subsequently receiving regorafenib showed an mPFS of 3.2 months and mOS of 12.1 months [31]. Among these patients, 26.1% had prior immunotherapy (primarily anti-PD-(L)1 monotherapy, 91.4%), with comparable survival outcomes between those with and without prior ICI therapy. A French retrospective study of patients treated beyond atezolizumab-bevacizumab suggested that regorafenib may offer superior outcomes than other approved TKIs, primarily sorafenib (the most commonly used comparator; other TKIs were poorly represented) [124]. While PFS was similar between regorafenib and alternative TKIs (HR 1.07; 95% CI, 0.61–1.86), OS was significantly longer in patients receiving regorafenib (HR 0.40; 95% CI, 0.20–0.79). This superiority was confirmed in multivariate analysis with an adjusted HR of 0.35. In a recent Asian study, regorafenib administered after atezolizumab-bevacizumab achieved an ORR of 11%, mPFS of 3.6 months, and mOS of 9.7 months [116]. Regorafenib in the post-IO setting has also been studied as part of a TKI plus ICI combination regimen. An international phase II trial investigated the efficacy of regorafenib combined with pembrolizumab in 95 patients who had progressed on one prior ICI regimen [130]. Patients were divided into two cohorts based on prior treatment: atezolizumab-bevacizumab (cohort 1) and other ICI regimens (monotherapy or in combination, cohort 2). After a median follow-up of 7.1 months, the activity of regorafenib-pembrolizumab was numerically inferior in cohort 1 compared to cohort 2 (ORR 5.9% vs. 11.1%; mPFS 2.8 vs. 4.2 months, respectively; mOS not reached in either cohort). The incidence of grade 3–4 AEs was higher in cohort 1 (61% vs. 41%). Overall, regorafenib plus pembrolizumab after the failure of first-line ICI-based regimens showed activity comparable to previously reported outcomes with single-agent regorafenib, with no apparent benefit from the addition of anti-PD-1 therapy. Ramucirumab Ramucirumab selectively inhibits VEGFR2 by blocking the binding of its ligands, VEGF-A as well as alternative ligands VEGF-C and VEGF-D, which, beyond their role in lymphangiogenesis, may also contribute to tumor angiogenesis [132, 161]. Accordingly, its utility in the post-IO setting applies to both angiogenesis inhibitor–naïve patients and those previously treated with bevacizumab. In colorectal cancer, VEGF-D has been associated with improved outcomes with the addition of ramucirumab to chemotherapy, whereas it has been linked to reduced benefit from bevacizumab [95, 164]. Furthermore, a study of aHCC patients treated with atezolizumab-bevacizumab demonstrated a significant rise in serum VEGF-D levels at the time of disease progression, suggesting a role in acquired resistance to the regimen [97]. Nevertheless, while ramucirumab provides selective VEGFR2 blockade, this narrow mode of action may be hypothetically less effective in tumors that rely on alternative signaling pathways, including other VEGFRs, PDGFR and MET. Clinically, ramucirumab was prospectively evaluated in aHCC following non-sorafenib systemic therapy in an expansion cohort of the REACH-2 trial [57]. This substudy enrolled 47 patients with baseline serum AFP of ≥ 400 ng/ml who had received one or two prior lines of systemic treatment, excluding sorafenib or chemotherapy. Among these, 42 patients (89%) received ramucirumab as a second-line treatment, of whom 30 had previously been treated with ICI-based regimens (single-agent ICI, dual ICI or ICI plus antiangiogenic therapy). The primary safety endpoint was consistent with findings from the original REACH-2 population, with treatment discontinuation due to AEs in 6% of patients. The mPFS and mOS were 1.7 and 8.7 months, respectively, with an ORR of 10.6% and a DCR of 46.8%. While the type of prior treatment did not influence survival, ORR was higher in patients with prior ICI exposure. The prospective R-evolution study enrolled 17 patients with high AFP levels, seven of whom had earlier received atezolizumab plus bevacizumab, seven lenvatinib, and three both [133]. The mPFS and mOS were 3.7 and 12.0 months, respectively, numerically better than those described in the REACH-2 trial. However, the incidence of AEs was higher than that in the REACH-2 study, with 29.4% of patients discontinuing treatment due to AEs. A Japanese retrospective study suggested that ramucirumab may exhibit greater activity after atezolizumab-bevacizumab than after prior treatment with TKIs (primarily lenvatinib) [131]. In particular, the ORR (33.3% vs. 0%), DCR (83.3% vs. 32.3%), and mPFS (3.9 vs. 1.9 months) were significantly better in patients previously treated after atezolizumab-bevacizumab. OS did not differ significantly (mOS 7.6 vs. 8.4 months), but ramucirumab was more frequently used as a third-line or later therapy in the atezolizumab/ bevacizumab-pretreated group. Other retrospective studies with small cohorts of ICI-pretreated patients reported an mOS of 10.0 months when ramucirumab was used as a second-line treatment, whereas its use as a third-line or later therapy was associated with an mOS of 4.8 months [111, 132]. Salvage Immunotherapy and Immunotherapy beyond Progression Alternative treatment strategies for aHCC patients experiencing disease progression on first-line immunotherapy-based regimens include either continuing ICI therapy or transitioning to a different ICI-based regimen. Continuing ICI therapy beyond disease progression may be particularly beneficial for patients experiencing limited disease progression or those with concerns regarding potential TKI-related toxicity. The decision to extend immunotherapy may also consider the pattern of progression. Patients with progression characterized by new intrahepatic lesions, extrahepatic growth or new extrahepatic lesions have notably better outcomes than those with alternative progression patterns, such as intrahepatic growth or new vascular invasion [105]. The latter progression types often require prompt therapeutic intervention due to a higher risk of liver function deterioration and limited future treatment options, whereas in select cases of limited progression of the former types, continuing immunotherapy beyond progression may offer potential benefits. Pivotal phase III trials permitted the continuation of first-line ICI therapy beyond radiological disease progression until the loss of clinical benefit. In the IMbrave 150 study, among 336 patients treated with the combination of atezolizumab and bevacizumab, 236 experienced disease progression [165]. Of these, 130 (56%), continued to receive atezolizumab (with or without bevacizumab), while 60 received other treatments, and 46 received no further therapy. Patients who continued first-line therapy beyond progression, received a median of five post-progression treatment cycles, achieving a post-progression mOS of 14.5 months, compared to 6.9 months in those who switched treatment and 2.0 months in those without subsequent therapy. Continuation of ICI following progression has also been adopted in routine clinical practice, with 17.6% of patients remaining on ICI therapy beyond progression [105]. Retrospective analyses suggest that ICI continuation is associated with improved post-progression survival compared to best supportive care alone. However, switching to next-line TKIs, either alone or combined with ICI, demonstrated numerically better survival outcomes [105]. The combination of a second-line TKI plus ICI, following failure of atezolizumab-bevacizumab, showed promising efficacy in a retrospective Asian study, with superior PFS and OS compared to sorafenib in multivariable analysis [116]. The ongoing Phase III IMbrave251 trial (NCT04770896) is investigating the continuation of atezolizumab beyond progression, while switching from bevacizumab to a TKI, either sorafenib or lenvatinib, after disease progression. Although the results are highly anticipated, it is worth noting that a similar strategy—continuing ICI therapy while transitioning to an alternative TKI after progression on an initial ICI-TKI combination—was recently evaluated in two phase III trials for metastatic clear-cell renal carcinoma. In both trials, CONTACT-03 (testing atezolizumab plus cabozantinib) and TiNivo-2 (nivolumab plus tivozanib), the addition of an ICI to another TKI failed to show superiority over TKI monotherapy [166, 167]. The use of alternative salvage immunotherapy regimens after prior ICI failure is uncommon in aHCC but has demonstrated clinical potential in retrospective studies [134, 135, 137, 147]. In particular, a large retrospective analysis found that among patients who discontinued atezolizumab-bevacizumab, subsequent immunotherapy was associated with numerically longer post-progression survival compared to TKIs (14.9 vs. 8.4 months, respectively) [106]. Salvage immunotherapy has also proven feasible in patients who experienced high-grade AEs or required steroid intervention during their initial ICI treatment [137]. However, data on clinical outcomes remain conflicting, with some studies suggesting comparable activity of first and subsequent exposure to ICI [137, 147], while others report much poorer outcomes upon rechallenge [143–145]. Several factors have been suggested to positively influence the efficacy of immunotherapy rechallenge, including an objective response to prior immunotherapy and switching to an alternative class of ICI (e.g., from anti-PD-1 to anti-PD-L1 or to bispecific antibodies) [142, 147]. Conversely, primary resistance to first-line immunotherapy is generally associated with poorer outcomes upon re-exposure [147], however, some studies have reported objective responses to a second ICI-based regimen even among patients whose best response to prior ICI treatment was disease progression [134, 135, 137]. In patients who fail single-pathway PD-1 blockade, a frequently explored salvage regimen combines anti-PD-1 and anti-CTLA-4 antibodies (e.g. ipilimumab). This approach utilizes CTLA-4 inhibition to enhance CD8 + lymphocyte priming while simultaneously reducing the population of immunosuppressive regulatory T cells, thereby counteracting the tumor's immune-suppressive microenvironment [134]. Additionally, in patients who develop resistance to single-agent PD-1 pathway blockade, alternative ICIs such as CTLA-4 may become upregulated. Notably, a higher baseline proportion of CTLA-4–positive cells in peripheral white blood cells and even more prominently, early on-treatment changes in CTLA-4–positive cell counts have been shown to distinguish responders from non-responders to atezolizumab-bevacizumab, and to exhibit predictive value for TTP and OS [168]. Ipilimumab has been shown to reverse acquired mechanism of immune resistance, promoting the persistence of memory T cells essential for durable therapeutic responses [134]. Moreover, atezolizumab, the most commonly used first-line ICI in aHCC, is more immunogenic than other anti-PD-(L)1 antibodies, with anti-drug antibodies (ADA) developing in approximately 30% of patients in the IMbrave 150 study [106, 169]. High ADA levels emerging early during atezolizumab-bevacizumab therapy have been linked to poor treatment outcomes in aHCC [170]. Consequently, it may be hypothesized that switching to a less immunogenic anti-PD-1 agent, such as nivolumab, may help reinvigorate the immune response in some patients who experience treatment failure. Additionally, because PD-1 engages both PD-L1 and PD-L2 ligands to suppress immune responses, the use of anti–PD-1 antibodies (such as nivolumab) in subsequent lines of therapy might hypothetically overcome the potential contribution of PD-L2 to resistance against anti–PD-L1 therapies. This is particularly relevant in light of emerging evidence indicating that PD-L2 expression in various malignancies can influence prognosis and responses to immunotherapy [171, 172]. Under these provisions, salvage therapy with nivolumab plus ipilimumab has demonstrated ORR ranging from 16 to 30%, with complete response (CR) observed in 3%–12% of cases, and an mOS of 7.4–10.9 months in both primary and secondary immune-resistant HCCs [134, 135, 173]. Case reports have described responses to durvalumab-tremelimumab (STRIDE) after progression on atezolizumab-bevacizumab following bevacizumab withdrawal [174], while retrospective data indicate disease control without objective responses in most patients previously treated with atezolizumab-bevacizumab [136]. Nevertheless, several other retrospective studies found that STRIDE used following atezolizumab-bevacizumab produces inferior ORR and PFS than in patients who received STRIDE after prior TKI treatment [143–145]. A recent early-phase trial explored a novel dual immunotherapy regimen consisting of balstilimab (an anti–PD-1 antibody) combined with botensilimab, an Fc-enhanced multifunctional anti–CTLA-4 antibody. Among 19 patients who had progressed on or after prior immunotherapy (mostly atezolizumab-bevacizumab), clinical activity was observed, with an ORR of 18%, mPFS of 4.4 months, and mOS of 12.3 months. TRAEs of grade 3 occurred in 37% of patients [146]. In conclusion, in the absence of head-to-head comparisons between immunotherapy rechallenge and TKIs, current evidence is insufficient to support immunotherapy rechallenge as a universally preferred option over TKIs in the second-line setting, particularly in TKI-naïve patients. However, in selected cases, rechallenge may be considered, particularly in patients who previously achieved a favorable response to immunotherapy, exhibit a less aggressive pattern of disease progression, maintain adequate liver function, and depending on the type of prior immunotherapy received. Locoregional Therapy While beyond the scope of this review, locoregional therapies, such as transarterial chemoembolization (TACE), radioembolization, and stereotactic body radiotherapy (SBRT), may play a role in selected patients with aHCC progressing on systemic treatment. In real-world practice, their use after systemic therapy discontinuation is common, with over 30% of patients receiving liver-directed treatments following immunotherapy [103]. Retrospective data suggest that TACE following first-line lenvatinib or atezolizumab–bevacizumab may offer better OS compared with second-line systemic therapy [73] SBRT achieves excellent local control and has been associated with improved survival when combined with systemic therapy. Radiotherapy may be particularly beneficial for patients with oligoprogressive disease or new portal vein tumor thrombosis, an unfavorable progression pattern associated with a high risk of hepatic decompensation [175–179]. Overall, locoregional treatments may help control selected patterns of progression and delay disease worsening in properly selected patients. Requirements for Treatment beyond the First Line The efficacy of second-line treatment in aHCC has been extensively documented in the aforementioned randomized placebo-controlled trials and in numerous real-world studies conducted in both post-TKI and post-ICI settings [33, 73, 91, 103–105, 180, 181]. However, patients who receive post-progression treatment constitute a distinct population compared to those who do not. These patients typically exhibit better baseline liver function, as reflected by a superior Child–Pugh score and ALBI (Albumin-Bilirubin) grade, along with a less advanced cancer stage (more often classified as BCLC-B, lower incidence of portal vein thrombosis or invasion, and lower AFP levels). They are also more likely to have a history of surgical resection, a lower neutrophil-to-lymphocyte ratio, and better performance status at both the initiation of first-line therapy and at the time of disease progression [73, 105, 181]. The PFS duration during first-line ICI does not appear to influence the likelihood of receiving subsequent treatment [105]. However, patients receiving frontline anti-PD-(L)1 plus TKI combinations are less likely to receive second-line therapy, possibly due to the limited availability of alternative drug classes after first-line TKI use or the increased toxicity associated with combination regimens [105]. Despite these confounding factors, multivariable analyses consistently demonstrate an OS benefit for patients who undergo second-line treatment [103, 104, 116, 180]. Historically, systemic treatment beyond the first-line was rarely utilized for aHCC. For example, in the REFLECT trial, only about one-third of participants received post-lenvatinib or post-sorafenib systemic therapy [8]. In the more recent IMbrave150 study, 52% and 36% of patients treated with sorafenib and atezolizumab-bevacizumab, respectively, received at least one subsequent line of systemic therapy [13]. Similarly, in the HIMALAYA trial, 42% of patients in the STRIDE arm received subsequent anticancer therapy [15, 16]. In real-world settings, where patient eligibility criteria are less stringent than in clinical trials, the rates of post-progression therapy currently reach 50% to 70% [73, 74, 84, 91, 103–106, 116, 181, 182]. Many patients receiving later-line agents deviate from the ideal eligibility profile for systemic treatment. A retrospective analysis from Canada found that only 13.1% of patients who failed to sorafenib met the strict clinical trial eligibility criteria for any MTA (10.1% for cabozantinib, 7.5% for regorafenib, and 5.8% for ramucirumab). When the criteria were relaxed to include patients with an ECOG performance status of 2 and/or Child–Pugh B7 classification, the eligibility rate increased to 31.7%. Nonetheless, patients meeting the liberal criteria had significantly worse OS than those meeting the stricter criteria [180]. Additionally, a retrospective analysis from Austria, Switzerland, and Germany found that only 42% of aHCC patients treated with cabozantinib would have met the inclusion and exclusion criteria of the CELESTIAL trial [51]. Importantly, patient selection for late-line systemic treatment of aHCC should be guided by evidence-based criteria. The primary candidates are those with preserved liver function and good performance status. Notably, all pivotal phase III trials leading to the approval of second-line agents predominantly enrolled patients with Child–Pugh A liver function [28, 44, 55, 63]. Consequently, data supporting a survival benefit in individuals with impaired liver function remain limited. The original REACH trial protocol allowed the enrollment of patients with both Child–Pugh A (88%) and Child–Pugh B (12%) disease. An OS benefit with ramucirumab, compared to placebo, was observed only in patients with a Child–Pugh score of 5 and 6, but not in those scoring 7 or 8. Additionally, Child–Pugh B patients experienced a higher incidence of grade ≥ 3 AEs across both treatment arms. In the Child–Pugh A subgroup, hypertension was the only high-grade AE reported more frequently with ramucirumab than with placebo. This suggests that most high-grade AEs in Child–Pugh B patients stem from underlying liver dysfunction rather than drug toxicity [183]. In line with these findings, a retrospective study found that Child–Pugh B patients who received regorafenib (mostly in the second line) had significantly worse PFS and OS compared to a matched cohort of Child–Pugh A patients (1.8 vs. 3.2 months for PFS, and 4.6 vs. 12.1 months for OS, respectively), along with a higher incidence of AEs [184]. Similarly, impaired baseline liver function (Child–Pugh B) was associated with poorer outcomes and a higher risk of AE-related early treatment discontinuations in patients treated with cabozantinib [49, 51, 78]. One cohort in the CheckMate-040 trial specifically recruited Child–Pugh B patients (scores of 7 or 8) for nivolumab monotherapy (either as upfront treatment or following sorafenib). The treatment conferred an ORR of 12%, a DCR of 55%, a median duration of response of 9.9 months, and an mOS of 7.6 months (9.8 and 7.4 months in the sorafenib-naïve and sorafenib-pretreated patients, respectively) [185]. The safety profile of nivolumab in Child–Pugh B was similar to that observed in Child–Pugh A patients. While these outcomes compared favorably to those reported for sorafenib in this population (mOS of 2.5–5.4 months), this indirect comparison should be interpreted cautiously. Notably, in a meta-analysis, Child–Pugh B patients administered later-line immunotherapy showed lower ORR and DCR, and an adjusted HR for OS of 2.33 compared to Child–Pugh A patients [186]. While significant liver impairment, as assessed by the Child–Pugh score, correlates with inferior treatment outcomes, more nuanced evaluations using the ALBI score revealed that patients with minor baseline liver function impairments may still benefit from MTAs. In the CELESTIAL trial, which exclusively enrolled Child–Pugh A patients, cabozantinib demonstrated improved OS and PFS compared to placebo in both ALBI grade 1 and 2 populations, although the OS benefit over placebo was greater in the former (HR of 0.63 and 0.84, respectively) [187]. Similarly, a pooled analysis of the REACH and REACH-2 trials demonstrated that ramucirumab improved OS in both ALBI score 1 and 2 subgroups [188]. Transient deterioration of liver function during treatment with TKIs does not necessarily negate treatment benefits. A retrospective analysis of the CELESTIAL trial indicated that cabozantinib significantly improved OS compared to placebo, even in patients whose liver function deteriorated from baseline Child–Pugh A to Child–Pugh B status by study week 8 (mOS 8.5 vs. 3.8 months for cabozantinib and placebo, respectively; HR 0.32; 95% CI, 0.18–0.58). The tolerability of cabozantinib in this subgroup was consistent with the overall study population [189]. These findings show that early deterioration of liver function does not necessarily warrant therapy discontinuation. This is particularly significant given the frequent transient deteriorations in ALBI score observed during the initial weeks of treatment with cabozantinib or lenvatinib [89, 132, 190]. Regarding the impact of BCLC stage and extrahepatic spread, phase III trial subgroup analyses showed that ramucirumab benefited both BCLC B and C stages, regorafenib was effective regardless of disease spread, and cabozantinib was more beneficial in extrahepatic disease [28, 44, 55]. In summary, while second-line therapy is increasingly utilized in patients with impaired liver function, its benefits are often counterbalanced by reduced tolerance to treatment-related AEs and a heightened risk of premature mortality due to liver failure. Identifying suitable candidates for anticancer therapy is particularly challenging, as liver function exists on a continuum. Nonetheless, later-line systemic treatments should be avoided in patients with severe liver dysfunction, such as a Child–Pugh score of ≥ 8 or ALBI grade 3. Treatment Sequencing beyond the Second Line The therapeutic landscape for aHCC continues to evolve, now incorporating treatments beyond the second line setting. In the CELESTIAL trial, 27% of patients received cabozantinib as a third-line therapy. Compared to placebo, cabozantinib improved PFS (HR 0.58; 95% CI, 0.41–0.83), although mOS remained identical in both treatment arms (8.6 months) [44]. However, these findings must be interpreted in the context of modern clinical practice. At the time of the CELESTIAL trial, no standard second-line treatments were established, and most patients had received experimental therapies or chemotherapy rather than ICIs. Consequently, the trial outcomes may not fully reflect the efficacy of sequential cabozantinib within the current treatment landscape, where ICIs and other modern therapies are widely utilized. Retrospective studies consistently show that approximately half of patients who fail second-line therapy receive further oncological treatment [103, 110, 115, 116, 127, 131]. A variety of regimens have been employed in third- and later-line settings, including cabozantinib (often following two prior MTAs), regorafenib (frequently used after progression on the ICI—sorafenib sequence), sorafenib (especially after second-line lenvatinib discontinuation), lenvatinib, ramucirumab, and ICI-based regimens (e.g., nivolumab, nivolumab-ipilimumab, or atezolizumab-bevacizumab) [48, 51, 88, 115, 116, 123, 131, 191]. However, data on the efficacy of third-line regimens remain limited and inconclusive. For example, PFS and OS outcomes with third-line ramucirumab were comparable to those observed in the second-line setting. In contrast, third-line regorafenib and cabozantinib demonstrated similar PFS but shorter OS compared to second-line treatment [35, 50, 56, 78, 129, 191]. A propensity-matched analysis found no significant OS differences between patients who received third-line treatment and those who did not, among those previously treated with atezolizumab plus bevacizumab—lenvatinib or atezolizumab plus bevacizumab—sorafenib sequences [115]. Clinical guidelines recommend considering third-line therapy for patients with Child–Pugh class A liver function and good performance status. Guidelines recommend prioritizing agents with distinct mechanisms of action from previously administered treatments (ASCO, AASLD) or specifically recommend cabozantinib as the preferred agent (EASL) [2, 23, 118]. Some guidelines propose that patients progressing on ICI-based regimens should first receive sorafenib or lenvatinib, allowing for the third-line use of evidence-supported post-sorafenib therapy with cabozantinib, regorafenib, or ramucirumab [4, 117, 118]. Predictive Biomarkers for Treatment Selection As previously noted, selecting second-line agents for aHCC remains challenging due to the absence of clinically validated predictive biomarkers. The sole exception is a serum AFP level of ≥ 400 ng/mL, which serves as a prerequisite for ramucirumab therapy. The utility of CEA and CA19-9 in HCC remains unclear, as these markers are not typically associated with this tumor type. Elevated CA19-9 may indicate mixed cholangiocarcinoma-HCC histology or progenitor cell activation, and has been linked to worse outcomes and reduced response to systemic therapies, including immunotherapy [192, 193]. However, their predictive value remains largely unexplored. An exploratory analysis of the REFLECT trial revealed that elevated baseline serum levels of VEGFs, angiopoietin-2 (ANG2), and FGF21 were associated with shorter OS across the entire study population treated with sorafenib and lenvatinib [194]. Notably, lenvatinib conferred longer OS (HR 0.53) than sorafenib in patients with elevated baseline FGF21 levels, while high serum levels FGF19 predicted shorter OS in patients administered sorafenib, but not lenvatinib [195]. In another study, responders to lenvatinib exhibited a significant increase in FGF19 levels and a decrease in ANG2 levels by week 4 of treatment [196]. High tumoral FGFR4 expression has also been identified as an independent predictor of improved response and prolonged PFS with lenvatinib [197, 198]. Notably, the aggressive macrotrabecular-massive histological subtype of HCC (accounting 10–20% of cases) is characterized by VEGF overexpression and FGF19 amplifications, thereby suggesting a hypothetical role for VEGFR and FGFR targeting in this tumor subtype [199]. Together, these findings suggest that patients with FGF-enriched tumors might derive greater benefit from lenvatinib than from other TKIs, although further validation is needed before FGF related biomarkers can be incorporated into clinical decision making. For regorafenib, predictive biomarkers were explored in the RESORCE trial, where higher baseline plasma levels of ANG-1, cystatin B, latency-associated peptide of TGF-β1, LOX-1, and MIP-1α were associated with improved survival, however, these markers similarly require prospective evaluation [200]. The efficacy of systemic therapies in aHCC can also be influenced by specific genomic alterations. For instance, PI3K/mTOR pathway mutations have been linked to poor outcomes in sorafenib-treated patients. Alterations in the WNT/β-catenin pathway (in particular related to CTNNB1 or AXIN1 mutations and PTK2 overexpression), are linked to an immune-exclusion phenotype (20–30%) and reduced efficacy of ICI monotherapy, but do not affect sorafenib activity [201]. Still, combining ICIs with anti-VEGF agents has been shown to mitigate the negative effects of WNT/β-catenin pathway alterations on ICI monotherapy outcomes [148]. Furthermore, the negative impact of CTNNB1 mutations on response to immunotherapy may vary, as different mutation types confer varying degrees of pathway activation, with weaker alterations exerting less influence on immune cell infiltration [202]. Additional factors associated with reduced ICI efficacy include activation of stromal related signaling pathways such as TGF β and Notch. In contrast, enrichment of the tumor microenvironment with CD8-positive T-effector cells, pro-inflammatory CXCL10-positive macrophages, and gene expression signatures related to interferon-gamma signaling and antigen presentation has been associated with improved responses to immunotherapy [5, 149]. Impaired response to immunotherapy combined with bevacizumab has been observed in HCC with upregulated VEGF co-receptor NRP1, whereas sorafenib activity appears unaffected, suggesting that switching to subsequent TKI therapy may be a reasonable strategy in NRP1-high tumors [149]. The role of expression of MET and AXL in predicting response to cabozantinib remains unclear. A study of a 10-patient cohort of heavily pretreated patients with aHCC receiving cabozantinib suggested improved PFS and OS in those with AXL-positive tumors [141]. Exploratory analyses from the phase 3 METEOR trial in renal cancer suggest that the PFS benefit of cabozantinib over everolimus (an mTOR inhibitor) is more pronounced in patients with high AXL expression, highlighting AXL as a potential predictive biomarker to cabozantinib in clear cell renal cell carcinoma [203]. In a small cohort of aHCC patients, MET expression did not show predictive value for clinical outcomes with cabozantinib [141]. Consistently, the exploratory analysis from a metastatic renal cancer clinical trial found no association between MET levels and treatment outcomes [204]. On the other hand, case reports and exploratory analysis from clinical studies have described highly favorable responses to cabozantinib, tepotinib and crizotinib (i.e. TKIs with MET inhibitory activity) in aHCC with MET amplification, a rare alteration occurring in ca. 1.5% of cases [201, 205–207]. Ethnicity does not appear to be associated with the efficacy of TKIs, as shown in registrational trials in the post-sorafenib setting and in global retrospective studies conducted after immunotherapy [8, 28, 44, 55, 138, 140]. However, the etiology of HCC may influence responses to specific therapies. Sorafenib appears to be more beneficial in HCV-related HCC compared with other etiologies, whereas subgroup analyses from the CELESTIAL trial suggest that cabozantinib may be less effective in this population [44, 208]. No such association with viral factors has been observed for lenvatinib, regorafenib, or ramucirumab [8, 28, 55]. Immunotherapy demonstrates efficacy regardless of HCC etiology, although the most favorable outcomes have been reported in HBV-related disease, while tumors with a metabolic etiology may derive less benefit [90]. Alcohol-related HCC has been associated with resistance to sorafenib in both clinical and preclinical studies, likely mediated by alcohol-induced activation of the ERK signaling pathway. Similar in vitro findings with lenvatinib, regorafenib, and cabozantinib suggest a potential class-wide TKI resistance effect related to alcohol exposure [209]. Based on the currently available evidence, we propose a tentative framework to guide the selection of second-line therapies across different clinical scenarios (Fig. 4). It should be emphasized that, although numerous studies provide insights into potential molecular and clinical predictors of treatment response, many of these associations remain putative and require further validation before they can be reliably incorporated into routine clinical practice.Fig. 4Proposed framework to guide the selection of second-line therapies across diverse clinical scenarios in advanced HCC. The predictors and decision-making factors presented include both validated variables (underlined) and predominantly putative factors (italics) derived from exploratory analyses and/or retrospective studies, which are not yet sufficiently validated for routine clinical practice Proposed framework to guide the selection of second-line therapies across diverse clinical scenarios in advanced HCC. The predictors and decision-making factors presented include both validated variables (underlined) and predominantly putative factors (italics) derived from exploratory analyses and/or retrospective studies, which are not yet sufficiently validated for routine clinical practice Conclusions and Future Perspectives Significant advancements in aHCC treatment have established ICI-based combinations as the cornerstone of frontline systemic therapy, relegating TKIs to later treatment lines. The increased efficacy of first-line therapy has led to a broader adoption of subsequent treatments. However, while first-line strategies are well-established, the optimal approach for later-line therapy remains uncertain. The second-line agents with phase III trial evidence—regorafenib, cabozantinib, and ramucirumab—were evaluated in an era when sorafenib was the dominant first-line therapy. Consequently, their trial outcomes may not fully apply to the current treatment landscape, where ICIs or, in some cases, lenvatinib are preferred as initial therapy. Additionally, registrational trials often enrolled patients with clinical profiles that do not fully reflect the broader, more diverse population encountered in real-world practice. While retrospective and small prospective studies suggest that MTAs remain effective and safe in ICI pretreated patients, with some evidence indicating a potential advantage for lenvatinib, there is no clear guidance on the optimal therapy in the post-IO setting. A key question is whether second-line treatment should prioritize traditionally second-line MTAs (regorafenib, cabozantinib, or ramucirumab) or repurpose historically first-line TKIs (sorafenib or lenvatinib) to preserve the former agents for third-line use. Another critical consideration is identifying patients who may benefit from continuing ICI therapy beyond progression and evaluating the efficacy of ICI rechallenge strategies. These may include dual immunotherapy combinations, or ICI-based regimens paired with alternative anti-angiogenic agents or TKIs. Additionally, the therapeutic landscape for treatment-refractory aHCC is expected to expand further, introducing even more complexity to clinical decision-making. Some of the currently investigated approaches include FGFR4 and FGFR4/3 inhibitors (such as fisogatinib, irpagratinib, and TYRA-430) [210, 211], therapies targeting proteins overexpressed in HCC such as glypican-3, including chimeric antigen receptor (CAR) T-cell strategies [212], and Wnt inhibitors (e.g., ETN101, WNTinib), which may be particularly beneficial in immune-excluded β-catenin-mutated HCC [213, 214]. As the number of effective first-line therapies continues to grow, the need for prospective randomized trials in pretreated patients becomes increasingly urgent. Such trials, incorporating biomarker-driven approaches, are essential for defining evidence-based treatment sequencing strategies in the evolving therapeutic landscape of aHCC.