Trial design and end points in hepatocellular carcinoma: an EASL–AASLD–ILCA consensus statement

Panel appointment and development of the statements

The consensus panel was appointed by the four societies and comprised physicians, surgeons, patient association representatives and regulatory agency representatives, all with expertise in the care of patients with HCC and/or who were recognized as leaders in trial design and clinical or translational research. The steering committee comprised five members: J.M.L (Chair, EASL representative), A.G.S. (AASLD representative), R.S.F. (ASCO representative), T.M. (ILCA representative) and E.M. (Secretary). The expert panel comprised 14 members who were formally nominated by their respective scientific societies based on recognized scientific expertise and leadership in HCC clinical research and trial methodology. Considerations of topic coverage, geographical representation and gender diversity were complementary to these merit-based criteria (Box 1).

A modified Delphi methodology, combining systematic appraisal of the literature with expert opinion, was used to reach consensus¹³. A search of PubMed and MEDLINE was conducted to identify relevant studies involving patients with HCC and related conditions, using predefined keywords. Articles published before 1 May 2025 were reviewed by the steering committee to define the scope of discussion and draft preliminary statements. Given the paucity of evidence on some emerging domains, including HCC trial design and end points, a narrative rather than a systematic review was performed. Eight domains were identified as fundamental for discussion: (1) overview of end points and trial design in cancer and HCC; (2) trial design in surveillance studies for HCC; (3) trial design in early-stage HCC; (4) trial design in intermediate-stage HCC; (5) trials exploring downstaging to liver transplantation; (6) trial design in advanced-stage HCC; (7) regulatory requirements; and (8) emerging topics in HCC trial methodology. Following the final evaluation, 102 statements across these domains were agreed upon and are emphasized according to the recommendations provided in each section. These recommendations reflect the current state of evidence and are intended as a dynamic framework subject to periodic revision as new data emerge.

Delphi rounds and data collection

A total of 19 experts were invited to participate in the Delphi process, among whom 17 representing the four societies were convened to vote on the proposed recommendations, of which 102 were approved. Statements that received nine or more votes (>50%) were approved, and the rest were excluded from the final recommendations. The strength of the consensus was categorized as follows:

• Strong consensus: 16–17 votes (>95% agreement)

• Consensus: 13–15 votes (>75–95% agreement)

• Majority agreement: 9–12 votes (50–75% agreement)

• No consensus: ≤8 votes (<50% agreement)

Two Delphi rounds were conducted electronically. The first-round survey (R1) was conducted between 21 April and 5 May 2025. The panellists were encouraged to provide comments and suggest revisions for each statement or suggest new proposals, which were discussed during the consensus conference convened at the EASL Congress 2025 (Amsterdam, Netherlands), which included 104 attendees from 30 countries. A second-round survey (R2, final) capturing the amendments discussed at the conference was conducted between 7 May and 21 May 2025.

A minority of statements did not reach the predefined agreement threshold (>50%) during the initial voting rounds. In most cases, a lack of consensus was related to clarity of scope or wording, rather than substantive disagreement. These statements were revised, discussed during the consensus meeting and re-voted on when appropriate. Supplementary Table 1 details the original statements that did not reach consensus, including their level of agreement and final disposition.

Trial design

Clinical trials are the foundation for evaluating new cancer therapeutics, procedures and/or devices, and provide high-quality evidence on both safety and efficacy⁴. Unlike case series and real-world data, clinical trials are prospective studies designed to minimize bias and generate controlled data. Trials have become essential for testing molecular targeted therapies and immunotherapies, and also for providing data contributing to biomarker validation and generating hypotheses for future strategies¹⁴.

Drug development follows a structured, four-phase pathway, with each stage designed to address specific objectives. Phase I trials focus on defining the safety profile and pharmacokinetics, as well as defining the maximum tolerated dose and recommended dose of an investigational agent. These studies typically involve small cohorts of 30–40 patients and might include patients with multiple tumour types. Extended phase I–II trials including ~100 participants are increasingly being conducted in order to confirm efficacy signals for molecularly targeted therapies or immunotherapies in biomarker-enriched populations and eventually to provide data to support applications for accelerated approval in the USA^4,15–17.

Phase II trials, with either a single-arm or randomized design, aim to assess safety and seek signals of efficacy using several end points such as ORR, PFS and OS¹⁸. The outcomes of phase II trials are crucial for sponsors and investigators to decide whether to proceed to definitive phase III development¹⁹. In this context, the sample size and magnitude of benefit are considered indicative. The final decision to proceed to phase III is generally supported by a comprehensive interpretation of the outcomes of more than one end point.

Phase III trials are the pivotal stage for regulatory approval. These RCTs test investigational drugs against standard-of-care therapies in well-defined populations, typically stratifying patients by key prognostic variables. Trial design is based on clearly defined primary end point(s), usually OS in patients with advanced-stage and/or metastatic cancers, and RFS or DFS in patients with earlier stage disease²⁰. In certain contexts, such as cancers with longer natural histories or when multiple subsequent therapies might influence OS outcomes, surrogate end points such as PFS might be accepted^4,21. Over the past decade, critical appraisals of data from phase III trials have emphasized the importance of biomarker-driven stratification, prospective subgroup analysis and validation of molecular targets to ensure clinical and regulatory relevance^17,22. Phase IV trials are conducted after regulatory approval to assess real-world effectiveness, long-term safety and broader generalizability to diverse patient populations. These postmarketing studies might detect rare toxicities, confirm benefit across certain clinical subgroups and provide valuable data enabling refinements to use in routine clinical practice²³.

Across all phases, the definition of primary end points is one of the most critical aspects of trial design. End points must be prespecified, clinically meaningful, and supported by a robust statistical plan to preserve type I error and avoid post hoc reinterpretations. Ambiguity in end point selection or α allocation can undermine interpretability, weaken regulatory confidence and limit clinical applicability²⁴. The growing complexity of contemporary oncology, driven by targeted therapies, immunotherapy and biomarker integration, has made the adoption of standardized rules for end point definition and interpretation increasingly important.

In this context, the following definitions are proposed to provide clarity and consistency in trial design and reporting:

• Single primary end point: trials are designed with a single primary end point, allocating the entire prespecified α to evaluation of this hypothesis and determining superiority or non-inferiority compared with the control arm.

  • Positive trial definition: the trial is positive if the primary end point reaches statistical significance at the allocated α level.

• Dual primary end point: trials are designed to assess two distinct primary end points, which will be independently tested by split α allocation, adjusting the α to preserve the overall type I error rate.

  • Positive trial definition: the trial is positive if only one primary end point independently reaches statistical significance.

• Co-primary end point: trials are designed to use multiple end points (two or more), with no specific α allocation to each end point. This design is mostly used in non-oncological settings in which clinical benefit needs to be captured using several measurements. For example, in trials involving patients with Alzheimer disease, both cognition and functional outcomes are designated as co-primary end points, since only improvement in both domains is considered clinically meaningful.

  • Positive trial definition: the trial is considered positive exclusively when all co-primary end points reach statistical significance.

• Hierarchical adaptive primary end point with α reallocation: trials are designed with two end points. The surrogate end point (for example, PFS) is initially tested with prespecified hierarchical conditions for α reallocation. Upon reaching statistical significance, the remaining α is reallocated to another definitive primary end point (such as OS).

  • Positive trial definition: the trial is considered positive if the surrogate end point initially reaches significance (potentially leading to accelerated approval), and subsequently, the definitive clinical end point reaches statistical significance using the reallocated α. If the surrogate end point does not reach significance initially, the trial continues to explore the definitive end point without α transfer.

End points in cancer and HCC

Clinical trials rely on a variety of end points to determine the efficacy, safety and overall value of novel cancer therapies. The choice of end point must balance scientific robustness, clinical relevance and feasibility, as this decision will ultimately shape regulatory approval and clinical adoption (Table 1).

Efficacy end points.

The best-established efficacy end point is OS, defined as the time from treatment initiation until death from any cause. OS remains the gold standard across most tumour types because extending OS provides an unequivocal demonstration of therapeutic benefit^2,25,26. However, OS requires prolonged follow-up and might be confounded by use of post-progression therapies²⁶. Alternative end points can provide earlier signals of efficacy. ORR, defined as the proportion of patients with tumour shrinkage defined according to standardized criteria such as Response Evaluation Criteria in Solid Tumors (RECIST) 1.1 (ref. 27) or modified RECIST (mRECIST)²⁸, captures antitumour activity. The value of ORR as an end point increases when complemented by duration of response (DOR)²⁵. PFS is time to disease progression or death, and TTP is the time to progression with censoring of deaths before progression²⁵. Both PFS and TTP are frequently used in phase II trials and/or to inform accelerated approval decisions, but require validation as surrogates of OS^7,24. In the curative-intent setting, DFS, EFS or RFS are frequently used in adjuvant or neoadjuvant trials designed to capture the ability of treatment to prevent relapse and enable prolonged remission²⁵. DFS measures the time from randomization, usually after curative-intent therapy, and with proof of an absence of residual disease on imaging, to disease recurrence or death. EFS is generally used in trials testing neoadjuvant therapies, and also measures time elapsed between randomization and disease recurrence or death. In trials testing neoadjuvant therapies, randomization occurs months before curative-intent therapy, and EFS therefore also captures preoperative events, such as disease progression or the inability to undergo planned surgery, thus providing a broader assessment of treatment efficacy across the perioperative setting.

Safety end points.

Efficacy must always be interpreted in the context of safety. Adverse events are graded according to standardized scales such as the Common Terminology Criteria for Adverse Events (CTCAE) grading system, ranging from mild events (grade 1) to death (grade 5)²⁹. Treatment-related adverse events (TRAEs) are the most widely used readout of drug toxicity, and are defined as any adverse events for which causal relationships with the study drug cannot be ruled out by the investigator³⁰. The threshold of acceptable and manageable TRAEs for a given drug or combination regimen varies according to the clinical setting. For example, assessments of clinical benefit from interventions administered in the adjuvant setting — with an absence of visual radiological disease — will consider clinically relevant TRAEs in a distinctly different manner from when they occur in the advanced-stage setting. Clinically serious adverse events include those with major consequences, such as hospitalization, disability or death. In the immunotherapy era, adverse events of special interest, such as immune-related adverse events (irAEs), are particularly relevant and are often characterized by the proportion of patients requiring systemic corticosteroids or other immunosuppressive therapies for management³⁰.

QOL and PROs.

QOL end points provide crucial insights into patients’ overall well-being, and encompass physical, emotional and function-related aspects. Validated tools, such as the European Organisation for Research and Treatment of Cancer (EORTC) QLQ-C30, ensure the consistency of assessments across studies, while disease-specific instruments such as Functional Assessment of Cancer Therapy (FACT)–Hepatobiliary and EORTC QLQ-HCC18 address the unique burden of HCC³¹.

Integration and interpretation.

No single end point adequately captures the full spectrum of therapeutic benefit. OS provides definitive unbiased evidence, although intermediate potential surrogate end points such as PFS, ORR or RFS all offer more timely signals that can accelerate development and support regulatory decision-making. Safety end points contextualize benefit by quantifying risk, and QOL and PROs integrate the patient perspective. A comprehensive and hierarchical interpretation of these end points is therefore required to establish the true clinical value of novel therapies.

Surrogate end points

The validation and use of surrogate end points in oncology trials remains an ongoing area of debate. By definition, a surrogate end point is a substitute for a direct clinical outcome (such as OS or QOL) that must reliably predict the treatment benefit on that outcome²⁶. For a surrogate end point to be accepted in regulatory or clinical decision-making, it must demonstrate biological plausibility, consistency across trials and statistical robustness at the trial level³². According to commonly accepted thresholds, an R value of 0.7–0.84 indicates moderate surrogacy, while an R value of ≥0.85 (R² ~0.7) denotes strong surrogacy²⁶. However, data from systematic evaluations suggest that only a minority of surrogate end points are strongly correlated with OS³³. An FDA review of surrogate validation studies conducted between 2005 and 2022 revealed that only one out of the 15 analyses demonstrated a high correlation (R ≥ 0.85) between the surrogate and OS⁷. Conversely, in patients with newly diagnosed acute myeloid leukaemia, for example, a robust correlation between EFS and OS was demonstrated (R = 0.93)³⁴. Similarly, evidence demonstrates that most surrogate measures used for cancer drug approval lack robust predictive value at the trial level, especially in the context of use of immunotherapies or prolonged post-progression survival³⁵.

This uncertain surrogacy underscores the need for caution in interpreting putative surrogate end points, particularly in advanced-stage disease settings in which survival after progression is short, thus limiting the window for meaningful OS improvements. Regulatory frameworks have responded accordingly: while accelerated approvals might rely on surrogates deemed ‘reasonably likely’ to predict clinical benefit, traditional approvals require a validated association with true clinical outcomes, such as prolonged survival²⁶. Consequently, early consultation with regulatory bodies to confirm the appropriateness of surrogate end points and ensure alignment in a confirmatory trial is an essential step. Ultimately, a reliance solely on surrogate end points must be justified by compelling validation and always interpreted within the clinical context that such end points are intended to represent. Otherwise, the risk of approving ineffective therapies that are incorrectly deemed to be effective remains unacceptably high.

Trial design in HCC

The panel of experts agreed in defining distinct trial design approaches according to the stages of the disease adopted in numerous clinical practice guidelines^9–12 (Fig. 1). For each stage, a summary of the target population, primary end points, stratification factors and control arms are summarized (Fig. 3). This structure has been replicated in all sections. Information on assessing expected outcomes and the magnitude of clinical benefit is also provided (Table 2).

Surveillance studies

For surveillance trials, a summary of target populations, expected outcomes and magnitude of clinical benefit is provided in Table 3. Semi-annual surveillance using ultrasonography plus serum α-fetoprotein (AFP) testing is recommended for at-risk individuals to improve the detection of early-stage HCC and enable curative treatment^9–11,36. HCC surveillance is cost-effective when the annual incidence of HCC is ≥1% in patients with cirrhosis and ≥0.2% in patients with HBV infection without cirrhosis at high risk of HCC, although not in groups with a very low incidence, such as those with non-cirrhotic metabolic dysfunction-associated steatotic liver disease (MASLD), in which the annual incidence is ~0.008 per 100 person-years¹. The target population for HCC surveillance trials should therefore include patients with cirrhosis of any aetiology and those with high-risk chronic HBV infection, ideally refined by validated risk stratification models to enable precision surveillance. Surveillance is not recommended for individuals with advanced liver dysfunction, severe comorbidities or an otherwise low risk of HCC (for example, individuals with non-cirrhotic MASLD, stages F1–F3 HCV-related fibrosis) given competing mortality risks, limited cost-effectiveness and a limited incidence in these groups^37,38. Increasing recognition of the limitations of ultrasonography (including suboptimal sensitivity (~63%) and poor adherence to semi-annual appointments (17.8% in the USA, 43.2% in Europe and 34.6% in Asia)) has prompted evaluations of alternative imaging-based and blood-based surveillance strategies^39–41.

The panel endorsed the Early Detection Research Network’s five phases of validation for emerging surveillance strategies, which has been adapted to HCC given specific nuances compared with cancers of other histologies^42,43 (Fig. 2). Biomarker validation for surveillance progresses through five phases: preclinical discovery (phase I), case–control studies of sensitivity and specificity (phase II), retrospective validation using prospectively collected samples (phase III), prospective cohorts to assess clinical utility (phase IV), and randomized trials to confirm mortality reduction and an acceptable balance of benefit to harm (phase V)^42,43.

The primary end point for HCC surveillance (phase V) studies should be cancer-related mortality³⁷. However, this end point requires both long-term follow-up and accurate discrimination between cancer-related death and liver dysfunction-related death, a feature that can be difficult to distinguish in some scenarios⁹. As a result, the panel proposed the identification of early-stage HCC that is amenable for curative therapies as the primary end point of phase IV studies along with assessments of harm resulting from false-positive referrals³⁷. Evaluations of physical, financial and psychological harm remain essential to defining the net benefit of surveillance⁴⁴.

After phase I exploratory studies have identified a promising surveillance strategy, phase II case–control studies are recommended to confirm an optimal threshold to define abnormal levels and characterize sensitivity and specificity estimates. Case–control studies can overestimate the performance of surveillance tests given that such studies can be affected by selection and spectrum biases; validation of test performance should therefore be performed in phase III retrospective cohort studies. Analyses are performed using a prospective specimen collection, retrospective blinded evaluation (PRoBE) design, in which samples are obtained prospectively, and the surveillance test is then retrospectively assayed among patients with and without cancer using banked samples or data. The primary outcome of phase III surveillance studies should be sensitivity and specificity of the test for preclinical cancer. The sensitivity of abdominal ultrasonography (63% for patients with early-stage HCC) establishes a threshold that a new surveillance strategy must achieve or surpass in a phase III study³⁹. Patients and providers report that they would be willing to accept a lower level of test specificity to facilitate higher sensitivity for early-stage HCC detection^44,45. Although other cancer screening programmes typically strive for a false-positive rate of ≤10%, a specificity of 80% might be acceptable given the high annual incidence of HCC in the setting of cirrhosis as well as the non-invasive recall pathway for HCC diagnosis^1,9. Regulatory approval of a surveillance strategy can be considered after successful phase III evaluation, although strategies that demonstrate sufficient test performance in phase III should still undergo prospective validation in phase IV and phase V studies to demonstrate clinical utility. The recommended primary outcome of phase IV studies is the early detection of HCC, which is presumably an accurate surrogate for reduced HCC-related mortality³⁶. Phase IV or V RCTs demonstrating superiority over semi-annual ultrasonography plus serum AFP testing would be expected to change clinical practice.

Trial design for early-stage HCC

Surgery, encompassing liver resection, liver transplantation and ablation, remains the cornerstone of the curative treatment of patients with HCC, and can lead to 5-year OS rates of up to 70–80% in selected cohorts² (Fig. 1). In the case of transplantation for early-stage disease (according to Milan criteria), the indication and extent of benefit are well established, making the need to address fundamental clinical questions regarding primary treatments, but not neoadjuvant approaches, through trials largely unnecessary^3,46. However, >50% of patients undergoing surgery will have disease recurrence within 5 years⁴⁷. Pattern, timing, anatomical location and treatability of recurrence can all affect the prognosis of patients with HCC, underscoring the urgent need for perioperative therapies to improve long-term outcomes^48–51.

Perioperative trial designs should include baseline biopsy sampling, followed by systemic therapy and collection of tumour specimens at surgery to enable analysis of biomarker dynamics⁴⁷ (Table 4). Neoadjuvant immunotherapy is an investigational strategy under active evaluation for early-stage HCC⁴⁷. Phase I–II trials have evaluated immune checkpoint inhibitor (ICI)-based regimens, either alone or in combination, with encouraging early efficacy and safety signals. These trials found major pathological response (MPR) rates of 20–40% and pathological complete response (pCR) rates of ~10%, with minimal surgical delays or dropouts and a manageable incidence of grade ≥3 toxicities at ~15% of patients, indicating potential clinical benefit^47,52–57. Ongoing trials, such as those evaluating atezolizumab plus bevacizumab (NCT07018947), pembrolizumab and lenvatinib (NCT05185739) and durvalumab plus tremelimumab (NCT05440864), aim to validate these findings. Notably, the phase II–III CARES-009 trial testing perioperative camrelizumab plus rivoceranib versus surgery alone was the first RCT to demonstrate a positive EFS result for perioperative therapy in patients with HCC (median 42.1 months versus 19.4 months; HR 0.59, 95% CI 0.41–0.85; P = 0.004)⁵⁸. However, this trial, conducted exclusively in Chinese patients with HBV-related HCC, was limited by the inclusion of patients with early-stage, intermediate-stage and advanced-stage HCC (portal vein invasion grade Vp1 to Vp3) undergoing resection, which resulted in a median EFS of ~19 months in the control arm as opposed to the expected 33–44 months found in recent global studies which predominantly enrolled patients with earlier-stage disease^59,60.

Regarding adjuvant phase III trials assessing ICIs, only two (IMbrave050 (testing atezolizumab plus bevacizumab) and KEYNOTE-937 (testing pembrolizumab)) have data available, and both failed to meet the primary end point^{47,48,61–63}. The updated analysis of the IMbrave050 trial did not confirm a sustained RFS benefit with adjuvant atezolizumab–bevacizumab, as found in the first interim analysis^61–63. Further analyses of these data emphasize that the interpretation of interim efficacy signals must account for potential non-proportional hazards (NPHs), which might lead to positive signals that are not confirmed over longer follow-up. Thus, prespecified analytical strategies and adequate data maturity are both essential to the accurate determination of durable clinical benefit^63,64.

Ongoing studies examining the efficacy of ICIs either alone or in combination with bevacizumab or other targeted agents are aiming to clarify the role of these agents or combinations in the adjuvant setting. Meanwhile, consensus on the resectability of HCC in patients with cirrhosis remains elusive, and is often evaluated within specific domains of liver function, extent of resection and remnant liver volume^9,11. Optimal HCC resection requires the availability of surgeons proficient in minimally invasive approaches (laparoscopic and robotic) when feasible, in order to reduce the incidence of complications, improve early outcomes and favour timely inception of postoperative adjuvant therapies.

Target population.

Eligibility for trials testing neoadjuvant therapies should be determined by preoperative factors, whereas postoperative variables are more appropriate for adjuvant studies (Fig. 3). All clinical guidelines consistently emphasize that both clinical and pathological factors can strongly influence recurrence risk and therefore should guide eligibility for neoadjuvant trials^9–11. Accordingly, as well as technical considerations, risk factors for recurrence include tumour diameter >5 cm, multifocality, microvascular or macrovascular invasion, poor differentiation, satellite lesions and elevated serum AFP levels of >400 ng/ml^{47,51,65–67}. Including patients with an intermediate risk of recurrence (one risk factor) to high risk of recurrence (two or more risk factors) ensures sufficient event rates to test experimental strategies, and avoids ceiling effects (when nearly all patients have disease recurrence, obscuring benefit) or floor effects (when recurrence is too rare owing to the over-representation of patients with low-risk disease).

End points.

The design of neoadjuvant trials enrolling patients with HCC requires end points that are both clinically meaningful and feasible within the unique biology of HCC. The high recurrence rates after curative-intent resection or ablation underscore the need for reliable surrogate end points of treatment efficacy⁴. In early-phase studies, the primary objective is to establish feasibility and safety, particularly in the context of ICIs with which irAEs might compromise surgical timing and eligibility. Phase I trials should therefore focus on toxicities and preliminary efficacy signals such as those provided by ORR and MPR, while also integrating biomarker analyses to understand mechanisms and identify predictive signatures. For phase II trials, pathological end points are increasingly being adopted. MPR, typically defined as a finding of ≥70–90% necrosis on quantification of residual viable tumour in the surgical specimen, has emerged as a surrogate of activity in other solid tumours^47,68,69. However, the correlation between pCR, RFS and OS in patients with HCC remains unproven, and current studies lack sufficient statistical power to confirm surrogacy⁵⁴. Harmonization of histological methodology is key to ensuring reproducibility^4,51,70. Imaging and classification according to radiological response criteria, such as RECIST 1.1 and mRECIST, might underestimate certain effects related to immunotherapies and LRTs, such as fibrosis or necrosis, making pathological assessments more sensitive but also more heterogeneous⁵⁴.

In phase III trials, time-to-event end points are recommended⁴. RFS remains the most appropriate primary end point for adjuvant studies because it directly reflects the ability of an intervention to delay recurrence after curative-intent therapy⁴⁷. In the neoadjuvant setting, EFS defined as time from randomization to progression, recurrence, death or toxicity precluding surgery, is better suited, as it captures both preoperative and postoperative events^26,58. OS should be consistently reported as a secondary end point, although interpretation of this end point is complicated by prolonged survival durations after recurrence and treatment crossover⁴. Complementary secondary end points should include patient-centred measures such as safety, QOL and functional recovery^4,5,47.

The biology of HCC recurrence is heterogeneous^50,65. Early recurrences (<2 years) usually reflect the presence of residual micrometastatic disease, and are the main target of systemic anticancer therapies, whereas late recurrences (>2 years) are typically de novo tumours arising in a carcinogenic cirrhotic background^47,48,50. Adjuvant trials are mostly targeting true metastases, although de novo disease continues to account for ~30% of recurrent HCC within the first 2 years after resection. This bimodal pattern supports the tailoring of treatment strategies according to risk: high-risk patients (with multiple adverse features) might benefit most from systemic neoadjuvant therapies, whereas lower-risk groups might be more appropriately managed with interventions addressing the underlying liver injury (for example, with antiviral or antifibrotic agents), either alone or in combination⁴⁷.

Stratification factors.

Defining high-risk features associated with recurrence after curative-intent therapy remains inconsistent across studies⁴⁷. To enhance comparability and external validity, risk stratification should include a maximum of four factors from among the following: morphometric (tumour diameter >5 cm, multifocality), pathological findings (microvascular invasion, satellite nodules, poor degree of differentiation), laboratory features (preoperative serum AFP >400 ng/ml) and type of intervention (resection or ablation)^47,51. These variables are incorporated into validated post-resection risk scores and might also inform the design of studies investigating liver ablation and/or transplantation^9,11,47,71.

Control arms.

The choice of an appropriate control arm is essential for the interpretability and regulatory acceptability of neoadjuvant trials involving patients with HCC^4,5. The current international guidelines recommend active surveillance after curative-intent resection or ablation as the standard of care, given that no adjuvant systemic therapy has yet demonstrated an RFS benefit^9–11. Thus, active surveillance remains the most suitable comparator in phase III trials testing perioperative studies (including neoadjuvant and/or adjuvant interventions)^4,47. The expected median RFS durations in the control arms has been 33 months⁵⁹, 36 months⁶² and ~44 months⁶⁰ (Table 2) in global studies testing adjuvant therapies^7,24,72.

Surveillance strategies must be standardized across trial arms to avoid detection bias and ensure homogeneity in recurrence ascertainment⁴. Guidelines recommend structured follow-up with contrast-enhanced CT or MRI and serum AFP measurements at regular intervals, given the complementary role of these methods in detecting intrahepatic and extrahepatic recurrence^9–11. The highest recurrence risk occurs within the first 2 years after curative-intent treatment, largely owing to micrometastatic residual disease, warranting intensive monitoring every 3 months during this period⁵⁰. Beyond 2 years, recurrence risk decreases and surveillance can reasonably be spaced to every 6 months, consistent with standard clinical practice⁴⁷. Extrahepatic recurrence, particularly pulmonary metastases, is not uncommon in patients with HCC⁶⁵. Guidelines emphasize chest imaging as part of post-treatment surveillance, especially in the first 2–3 years, when the risk of metastatic recurrence is highest^9,11. Incorporating 6-monthly chest CT during this interval ensures early detection of lung metastases and provides uniform outcome assessments across study populations.

Overall benefit assessment.

The overall extent of benefit from neoadjuvant ICIs in patients with HCC remains under investigation^10,11,73. Data from the phase II–III CARES-009 trial indicate that perioperative (neoadjuvant plus adjuvant) camrelizumab plus rivoceranib improves EFS compared with surgery alone (HR 0.59, P = 0.004), although the short follow-up and treatment compliance, open-label design, and exclusively Chinese population with HBV-related HCC make these results difficult to generalize⁵⁸. Furthermore, in light of the previous experience in the IMbrave050 trial, interim findings must be interpreted cautiously, and only mature data, defined by adequate follow-up and a sufficient number of events, can establish durable benefit. Advanced statistical approaches (such as MaxCombo testing, restricted mean survival time (RMST) or piecewise hazard ratios) are essential tools for statistically robust evaluations of outcomes in the presence of NPHs⁶⁴.

Safety is equally crucial⁵. High-grade TRAEs, particularly life-threatening or fatal events, might offset efficacy. Given that surgery is intended to be curative, neoadjuvant therapy must not compromise operative safety and/or delay or preclude resection⁷⁴. The use of contemporary minimally invasive liver surgery reduces surgical morbidity and should be maintained as the standard approach whenever possible¹¹. Therefore, all trials testing neoadjuvant therapies must incorporate stringent safety monitoring, neoadjuvant outcome reporting and recovery assessments to ensure that systemic therapy does not harm patients who might otherwise be cured with surgery alone.

Trial design when testing LRT as the primary treatment

Trial design is a major challenge for patients with intermediate-stage HCC (Barcelona Clinic Liver Cancer (BCLC) stage B (BCLC-B))^2,4,75. This population includes patients with multinodular disease, preserved performance status and relatively maintained liver function (Child–Pugh class A or B7), although variability in tumour burden, biological characteristics and treatment response can all complicate trial comparability and clinical decision-making^2,3 (Table 5).

Currently, patients with intermediate-stage HCC can receive either liver transplantation after downstaging, LRTs (mainly transarterial chemoembolization (TACE) or transarterial radioembolization (TARE)) or even systemic therapies^2,75. TACE is the established standard-of-care treatment for these patients, as demonstrated by two phase III trials that revealed a survival advantage over best supportive care or less-effective treatments^76,77 (Fig. 1). This conclusion was corroborated by a subsequent meta-analysis⁷⁸. Approximately half of all patients with BCLC-B HCC can undergo TACE, with the other half being ineligible owing to issues such as liver dysfunction, technical limitations (such as impaired portal vein flow) or a large tumour burden (largest tumour >10 cm in diameter)⁷⁵. The AASLD and EASL guidelines also recommend TACE for patients with early-stage HCC who are unsuitable for curative approaches (surgery or ablation)^9,11. Although trials testing TACE combined with a tyrosine kinase inhibitor (TKI) did not demonstrate significant improvements in OS, these studies helped establish the current benchmarks for median OS (~26–30 months), PFS (~8 months) and objective response rate (~50–60% according to mRECIST)⁷⁵ (Table 2). Two trials that combined TACE with durvalumab plus bevacizumab (EMERALD-1) or lenvatinib plus pembrolizumab (LEAP-012) demonstrated significant improvements in PFS, and another trial (TALENTACE) showed improvements in the modified PFS end point of TACE-PFS (defined as time from randomization to the development of TACE-untreatable disease)^{9–11,79–81}.

In parallel to TACE, TARE with ⁹⁰Y microspheres has been evaluated in patients with intermediate-stage HCC. Unlike TACE, which induces ischaemic necrosis via embolization, TARE delivers targeted intra-arterial radiation with only a minimal embolic effect, potentially preserving vascular patency and reducing the extent of decompensation after the procedure. Both retrospective and prospective data suggest that TARE might result in a longer time to disease progression and have favourable tolerability in selected patients, particularly those with a high tumour burden, large lesions or prior TACE refractoriness^9,82. However, randomized trials comparing TARE with TACE have yielded mixed results, partly owing to differences in patient selection and procedural standardization (Figs. 1 and 3).

External beam radiotherapy (EBRT), including stereotactic body radiotherapy (SBRT), has been acknowledged as part of the spectrum of LRTs available for patients with early-stage disease by current guidelines based on data from cohort studies and phase II trials (Fig. 1). This approach has also been tested in the neoadjuvant and advanced-stage disease settings, particularly in patients with portal vein tumour thrombosis^3,83,84.

Target population.

More stringent definitions of clinical trial eligibility have been proposed in an attempt to address the heterogeneity of patients with intermediate-stage HCC. Patients with tumours confined to the liver without vascular invasion, preserved hepatic reserve (Child–Pugh class B7 or lower, albumin–bilirubin (ALBI) grade 1 or 2), and an Eastern Cooperative Oncology Group (ECOG) performance status of 0, are those most likely to benefit from LRT (TACE or TARE)⁴. Conversely, LRT has shown limited benefit in patients with ALBI grade 3, ECOG performance status ≥1, or features indicating unsuitability for supraselective embolization, such as very large tumours (diameter ≥10 cm for TACE), bilobar involvement affecting >50% of the liver, infiltrative growth patterns or macrovascular invasion^9,11.

For TACE-unsuitable populations, who would otherwise meet the criteria for systemic therapy, randomized trials evaluating local–systemic combinations (such as EBRT or TARE plus ICI-based regimens) should incorporate a systemic therapy control arm in order to appropriately test incremental benefit and avoid confounding owing to treatment stage migration³. Importantly, in the setting of TARE, ~10–13% of patients might be deemed ineligible after pretreatment angiographic evaluation, which could have implications for clinical trial design, including estimates of statistical power⁷⁵.

Finally, baseline tumour biopsy is increasingly advocated not only to confirm the diagnosis in patients with a non-cirrhotic liver or atypical imaging features but also to provide tissue for translational analyses, biomarker discovery and molecular characterization, including in patients who otherwise meet AASLD non-invasive diagnostic criteria⁸⁵. Together, these elements show that precise population definition is essential not only to optimize clinical benefit and trial interpretability, but also to permit meaningful cross-trial comparisons and more reliable effect size estimation.

End points.

OS remains the regulatory gold standard and the most definitive end point in phase III trials involving patients with intermediate-stage HCC⁴. As a result of previous guideline and consensus recommendations, the panel endorsed a threshold hazard ratio of ≤0.8 or a minimum 3-month absolute improvement in median OS as signals of clinical benefit considering a standard baseline median OS of ~30 months^5,86,87. Nevertheless, interpreting OS is often complicated by the lengthy follow-up needed to reach maturity, the effects of crossover and subsequent treatments, and the competing risks associated with cirrhosis-related mortality⁴. These limitations have prompted increasing interest in surrogate end points, particularly PFS, which might support accelerated approvals when accompanied by consistent improvements in secondary measures⁷². PFS might serve as an appropriate primary end point when supported by mature data demonstrating signals of OS benefit, although distinct hazard ratio thresholds apply for different purposes. In cohorts enrolling patients with advanced-stage HCC, PFS surrogacy for OS requires a magnitude of benefit represented by a hazard ratio of ≤0.6 based on a meta-analysis of data from 21 phase III trials evaluating systemic therapies⁸⁶, whereas data establishing this threshold in patients with intermediate-stage HCC are lacking⁷⁵. To overcome this limitation, an RCT with a positive PFS end point and a trend towards improvements in OS (HR ≤0.8) would be required to support a change in clinical practice. Other end points, such as TACE–PFS⁸¹, and time to treatment failure⁵⁰ have not been validated and currently lack sufficient robustness to support the panel’s recommendations.

Secondary end points typically include ORR, TTP, PROs and QOL, all of which provide complementary perspectives on the effects of an intervention beyond survival⁴. Radiological assessments have a central role in defining several of these outcomes and must be based on standardized and reproducible methodologies^28,75,88.

Conducting a centralized, blinded review at regular intervals (~3 months) using validated criteria is crucial to minimizing the extent of interobserver variability and ensuring consistency across trial sites^4,9,10. Evidence strongly supports the use of centralized mRECIST-based evaluations, particularly for patients with intermediate-stage HCC receiving LRT, in whom assessments of tumour viability (arterial enhancement) are critical^28,88, although phase III trials have also used RECIST 1.1. A meta-analysis of data from seven studies has demonstrated a robust correlation between mRECIST response and OS (HR 0.39, 95% CI 0.26–0.61) for LRT⁸⁸. Response assessment after >3 months is recommended when testing the combination of LRT and ICIs, owing to the delayed responses typically associated with the latter^4,80,85. Additionally, the Response Evaluation Criteria in Cancer of the Liver (RECICL) criteria⁸⁹, which are used in some trials conducted in Asian centres (such as TALENTACE⁸¹ and CARES-005 (ref. 90)), define investigator-assessed TACE–PFS as a composite end point including either time to progression untreatable by TACE, treatment failure, actual progression or death. However, this end point requires validation and is not endorsed by the current consensus panel. Likewise, immunotherapy-based RECIST (iRECIST)⁹¹ is rarely applied in trials involving patients with HCC and is also not currently endorsed by this consensus panel.

An analysis published in September 2025 indicates that ~20% of pivotal phase III trials encounter NPHs, and particularly those testing ICI-containing regimens⁶⁴. In such scenarios, treatment efficacy should be assessed only after data are sufficiently mature and using appropriate statistical methods⁶⁴.

Stratification factors.

The interpretation of treatment effects in patients with intermediate-stage HCC is strongly influenced by baseline disease characteristics, making appropriate stratification essential for reliable trial outcomes⁴. Tumour burden, captured by systems such as the up-to-seven⁹² or six-and-twelve criteria⁹³, has consistently been shown to correlate with prognosis and treatment response^9,11. The six-and-twelve score, based on the sum of HCC nodules and the largest tumour diameter, demonstrates clear prognostic stratification: patients with scores of ≤6 have a median OS of 49.1 months versus 32.0 months for those with scores of >6 to ≤12, and 15.8 months for scores of >12 (ref. 93). The up-to-seven criteria, utilizing a similar methodology of adding tumour number and maximum diameter, are designed to identify patients with better tolerability and responsiveness to LRT, with a median OS of 32.3 months in those with scores within the up-to-seven criteria and 28.3 months in those with scores beyond the criteria^92,94. Applying broader inclusion criteria might dilute treatment effects by incorporating biologically heterogeneous populations, whereas overly restrictive criteria might limit the generalizability of the conclusions. Nonetheless, both scoring systems require formal validation before application as primary stratification factors in prospective trials.

Similarly, liver function assessed using the ALBI score grading system refines risk prediction even in patients with Child–Pugh class A disease, whereas serum AFP ≥400 ng/ml has repeatedly emerged as a marker of poor survival outcomes^9–11,95. Geographical region further contributes to heterogeneity, reflecting differences in both disease aetiology and standards of care⁴. Consideration of prior therapies and the underlying cause of the liver disease adds an additional level of granularity to trial design, ensuring comparability between study arms and enhancing the external validity of the results⁴.

Control arms.

The choice of control arm has direct implications for both clinical relevance and the regulatory acceptability of trial outcomes⁴. In patients with intermediate-stage HCC, TACE has historically served as the standard comparator following an RCT demonstrating improvements in survival over supportive care⁷⁵. An expected median OS of 26–30 months in well-selected patients establishes a therapeutic benchmark for experimental interventions^75–77. However, outcomes following TACE vary widely across studies owing to differences in technique, retreatment schedules and criteria for defining disease progression or treatment failure. Standardization of these elements within trial protocols is therefore essential to avoiding bias and ensuring interpretability^9,10. Both conventional TACE and drug-eluting bead TACE have demonstrated comparable efficacy profiles, supporting the acceptability of either modality as control arms when appropriate stratification is applied^9–11. In the context of trials testing combination therapies, the use of placebo has followed established oncology standards. Landmark studies such as EMERALD-1 and LEAP-012 have used placebo plus TACE rather than TACE monotherapy as the control, thereby maintaining appropriate blinding while ensuring that the control arm receives standard-of-care therapy^79,80.

Discordant regional guidelines require accommodation within global trial designs. For example, the AASLD guidelines provide level 2 evidence supporting the use of TARE therapy in patients with intermediate-stage HCC, an approach commonly used in the USA, whereas the EASL guidelines maintain reservations regarding this approach, pending phase III validation^9,11.

TARE has also been explored as an alternative first-line LRT approach for patients with intermediate-stage HCC^9,10. TARE offers several theoretical advantages, particularly in patients with larger-diameter tumours or those who are refractory to TACE⁹. Nevertheless, trial design is complicated by the fact that a proportion of patients might be excluded after angiographic assessments^75,96. This post-randomization exclusion can lead to a reduction in statistical power, requiring larger sample sizes as well as careful planning and study design.

Overall benefit assessment.

Therapeutic benefit extends beyond OS and PFS in patients with intermediate-stage HCC. Other clinically relevant dimensions of therapeutic benefit include treatment-related toxicity, preservation of liver function, durability of response and the potential to reduce tumour burden to an extent that enables the administration of subsequent curative-intent therapies (conversion therapy). Randomized trials combining LRT with systemic therapies, such as EMERALD-1 and LEAP-012, have demonstrated significant improvements in PFS (HR 0.77, 95% CI 0.61–0.98, and HR 0.66, 95% CI 0.51–0.84, respectively). OS outcomes from EMERALD-1 and LEAP-012 have not yet been reported, although such outcomes extending beyond the expected median durations of ~30 months (Box 2) might make similar RCTs difficult to conduct with OS as the sole primary end point, owing to the implicit need for long-term follow-up of large numbers of patients. Thus, the current panel advocates claiming efficacy signals when a positive PFS result is demonstrated along with mature trends suggesting OS benefit, as a minimum.

Rationale and target populations.

Liver transplantation remains the only treatment capable of curing both HCC and the underlying liver disease^9–11. While Milan criteria remain the standard for early stage disease, downstaging strategies designed to bring patients within these criteria, and thus eligible for transplantation, are now widely applied, supported by data showing acceptable post-transplantation outcomes in successfully downstaged patients^{9–11,97–99} (Fig. 1). Importantly, successful downstaging to comply with the Milan criteria not only enables access to liver transplantation but also serves as a biological selection tool, identifying tumours with favourable disease characteristics that are likely to yield adequate long-term outcomes⁴⁶. When designing studies to assess the efficacy of novel therapies leading to downstaging to Milan criteria and the long-term effects of subsequent liver transplantation (defined by RFS and/or OS), three scenarios should be taken into account in terms of target populations, which justify three distinct trial designs (Table 6):

1. Patients within United Network for Organ Sharing (UNOS) criteria (Milan and UNOS-downstaging (DS)). These patients are currently considered eligible for transplantation and meet the Milan criteria (1a), either at the point of staging or after successful downstaging (UNOS-DS, 1b). Stratification between these two target populations is recommended. These patients usually receive bridging with LRT while on the waiting list, mainly to avoid disease progression and consequent ineligibility for transplantation^9,11,100. Exploratory neoadjuvant therapies in this setting should be compared with LRT (TACE or TARE).

2. Patients with intermediate-stage disease with extended indi-cations beyond UNOS-DS criteria. Patients beyond UNOS-DS criteria with intermediate-stage HCC can be considered for liver transplantation under downstaging circumstances. Thus, in these cases the study should be designed first to downstage patients to meet Milan criteria, which can be considered as conversion therapy. Patients should subsequently receive bridging neoadjuvant therapies while on the waiting list.

   The recommended active treatment arm for this target population should be LRT with or without an ICI versus LRT (TACE or TARE) alone in the control arm, although other approaches can be explored in the setting of appropriately controlled research studies^9,11,100. In these studies, both the efficacy of downstaging and dropout rates are key end points in phase II trials designed to provide the rationale for phase III investigations.

3. Patients with extended indications including advanced-stage disease (ECOG PS 1 or 2 and/or macrovascular invasion (grade Vp1 or Vp2)). In this controversial scenario, a clinical trial is aimed at exploring the feasibility of potential conversion to transplantation for candidates with advanced-stage HCC. This approach is conceptually appealing and has attracted interest in preliminary studies, although it pertains to a study population that has not previously been formally investigated as comprising potential candidates for transplantation, and is an approach that is certainly not endorsed by current clinical practice guidelines¹⁰¹. Under these specific conditions, comparing the active treatment administered during the downstaging process for conversion to transplant eligibility with a control treatment that includes ICI-based regimens is imperative. Comprehensive phase II trials are recommended to evaluate safety, feasibility, efficacy and dropout rates. A sustained objective response (according to RECIST and mRECIST) exceeding 3 months in patients who have been successfully downstaged, along with a concurrent and equally sustained reduction in serum AFP levels, is associated with favourable post-transplantation outcomes and transplant eligibility, and is advised in these patients⁴⁶.

Regimens combining LRT with immunotherapies are the suggested active arm for all RCTs testing downstaging strategies^9,10. In the scenarios of conversion and downstaging to Milan criteria, several specific end points can provide signals of efficacy in phase II trials, such as efficacy of downstaging and prevention of dropout, as well as safety. Ultimately, however, clinical benefit should be evaluated primarily using EFS or RFS and a 5-year OS of at least 60% by the intention-to-treat analysis⁴⁶. Finally, capturing post-transplantation complications and irAEs leading to rejection is a mandatory secondary end point when assessing the efficacy of immunotherapies administered prior to liver transplantation⁴⁷.

Trial design in this area must consider the scarcity of donor organs as futile transplantation in patients with aggressive disease is an unacceptable use of this scarce resource⁴⁶. Only patients with sustained responses defined by accepted criteria should proceed to liver transplantation, and study protocols must align with local organ allocation systems to ensure equity and generalizability (Fig. 3).

End points.

In the context of combined LRT and systemic therapy strategies, the design of trials testing liver transplantation requires cautious and precise end point definition. Phase II trials should focus on safety, downstaging efficacy, dropout rates and transplantation feasibility. Phase III trials must evaluate EFS and OS on an intention-to-treat basis, thus accounting for waiting list dropout⁴⁶. Secondary outcomes should capture the extent of benefit from transplantation, feasibility and complications from intensified immunosuppression, including cardiovascular events, chronic kidney disease and de novo malignancies⁴⁶. Acute cellular rejection (ACR) after pre-transplantation ICIs remains a key concern when designing such studies¹⁰². Early studies suggested ACR rates of <20%, with preliminary data showing 3-year OS exceeding 70–80% and disease recurrence in <10% of patients^103,104. Washout periods of about two half-lives (~60 days) of immunotherapy agents appear to reduce ACR risk, although this approach requires validation in prospective trials. Evidence is needed to establish whether regimens combining ICIs with LRT can safely downstage patients for liver transplantation and deliver durable survival without excess waiting-list mortality¹⁰⁵.

Stratification factors.

Risk stratification in trials involving patients with HCC requiring liver transplantation depends on tumour burden, liver function, tumour biology and response to prior treatments⁴⁶. Prognostic models such as up-to-seven, Metroticket 2.0 and the AFP model integrate tumour metrics and serum AFP levels to predict post-transplantation outcomes^92,106,107. Liver function assessments are crucial for this purpose. The Child–Pugh score remains the standard approach, whereas the ALBI grade provides a more objective alternative that does not include subjective variables such as ascites, and has better prognostic value, especially in patients with Child–Pugh class A disease, in whom ALBI grade ≥2 predicts inferior survival outcomes¹⁰⁸. Incorporating ALBI grade together with criteria related to tumour burden and serum AFP ≥400 ng/ml enables improved trial stratification, ensures more balanced arms, and improves the interpretability of outcomes¹⁰⁹.

Control arms.

The control strategy in trials investigating liver transplantation must suit the target population, which should be LRT for target populations 1 and 2 (with early-stage or intermediate-stage disease) and ICI-based systemic therapy regimens for target population 3 (with advanced-stage disease)¹¹⁰. Over 90% of patients receive at least one form of LRT, with treatment choices based on tumour diameter and anatomical location. Historically, TACE has been the most widely used method worldwide for intermediate-stage HCC, although use of TARE has been increasing¹¹¹. Comparative data are limited, although studies suggest that both approaches are similarly effective^82,112. For this population, either TACE or TARE, with standardized dosing and retreatment protocols, is a suitable therapy for patients in control arms. By contrast, for exploratory studies involving patients with advanced-stage disease (defined as ECOG performance status 1 or 2 or with segmental portal vein invasion (grade Vp1 or Vp2)), the standard control arm for trials testing conversion therapy should be systemic therapy^9–12.

Overall benefit assessment.

In the transplantation setting, therapeutic benefit must be interpreted within the context of the ethics of organ allocation. The primary role of systemic therapies or LRT combination regimens before transplantation is to achieve and maintain eligibility for transplantation rather than preventing disease recurrence, therefore trials should capture oncological outcomes and graft long-term viability. The available evidence indicating 5-year OS of ≥60% in the intention-to-treat populations of various studies provides a benchmark for expanding liver transplantation criteria without compromising allocation fairness^46,100,113 (Table 2). A study in which most patients (75.6%) had transplantation-ineligible disease at baseline (according to Milan criteria) and received an ICI plus LRT, downstaging was successful in most patients (75%), with 3-year post-transplantation OS of >80%, and manageable rejection risk¹¹⁴. If confirmed prospectively and if non-HCC mortality remains stable, these results might change clinical practice. Randomized phase III trials testing LRT–ICI combinations are recommended across downstaging strategies, with a benefit threshold of ≥60% 5-year OS in ITT analyses^46,113 (Box 2).

Trial design for advanced-stage HCC

The therapeutic landscape for advanced-stage HCC has evolved rapidly in the past years, driven by ICI-based combinations that have outperformed sorafenib and lenvatinib in phase III studies^10–12. Regimens such as atezolizumab–bevacizumab in IMbrave150 (ref. 115), tremelimumab–durvalumab in HIMALAYA¹¹⁶ and nivolumab–ipilimumab in CheckMate 9DW¹¹⁷ have established ICI-based regimens as preferred first-line options, with median OS durations of 16–24 months, whereas sorafenib and lenvatinib^118,119 remain essential for those with contraindications to ICIs, with median OS durations of 14–20 months^9–12 (Fig. 1). The combination of camrelizumab and rivoceranib also led to survival benefits versus sorafenib, although this combination was found to have a significantly worse TRAE profile in a nested meta-analysis compared with all other regimens (relative risk 1.59, 95% CI 1.25–2.03; P < 0.001)¹²⁰. Furthermore, the lack of FDA and EMA approval has prevented a clear endorsement in clinical practice guidelines^9–12, thus supporting the current consensus. Similarly, other therapies with positive or non-inferior results versus sorafenib have obtained approval in China¹²¹ (Fig. 3). Finally, the LEAP-002 trial (lenvatinib plus pembrolizumab versus lenvatinib) epitomizes a scenario in which, despite a nominally significant P value for OS (P = 0.023), the trial was deemed negative because the prespecified, more stringent significance threshold required for the dual primary end points (P < 0.019 for OS and P < 0.002 for PFS) was not met, as the overall α had been split between both analyses. This changing landscape has required a reappraisal of how new trials involving patients with advanced-stage HCC are designed, including the definition of the target population, selection of end points, stratification strategies, choice of control arms, and criteria for interpreting overall benefit (Table 7).

Following disease progression on first-line ICI-based combinations, the evidence guiding subsequent therapeutic strategies for patients with advanced-stage HCC remains limited. Only small-cohort phase II trials testing cabozantinib, regorafenib or lenvatinib — with median OS durations of ~10 months^122–124 — and data from retrospective studies also suggest potential benefits following atezolizumab plus bevacizumab¹²⁵. TKIs such as lenvatinib, sorafenib and cabozantinib, and regorafenib are recommended after progression on ICI-based regimens according to clinical guidelines and supported by efficacy data as viable treatment options^9–12. Ramucirumab has also been shown to prolong OS in patients with serum AFP levels ≥400 ng/ml following treatment with sorafenib¹²⁶. In this setting, second-line pembrolizumab demonstrated improved OS compared with placebo in an Asian population, whereas the combination of nivolumab plus ipilimumab demonstrated response rates of approximately 32% after sorafenib^125,127. Prospective data defining the optimal sequencing of therapies following specific first-line ICI-based combinations remain limited and should be considered a key area for future clinical trial development. The principles outlined in this framework are intended to apply across treatment lines, including post-progression settings. In the absence of definitive evidence establishing the optimal therapeutic sequence after first-line ICI-based regimens, the designs of trials in this setting should remain adaptive, with careful consideration of prior treatment exposure, mechanisms of resistance and evolving standards of care.

Target population.

Defining the appropriate target population has been a persistent challenge in trials involving patients with advanced-stage HCC⁴. Simply applying the highly heterogeneous and broad category of ‘unresectable disease’ is discouraged. Instead the target populations for trials testing systemic therapies is defined by both patients with intermediate-stage (BCLC-B) disease who are unsuitable for or have had disease progression after LRT, and those with advanced-stage HCC (BCLC-C)⁴. Within this spectrum, the inclusion of advanced-stage HCC with main portal vein invasion (grade Vp4) was considered particularly important, as such patients have historically been included in only two RCTs (SHARP and IMbrave150), and represent a target population with an unmet need^115,118,128.

In patients with advanced-stage HCC with a predominantly intrahepatic tumour burden and grade Vp3 or Vp4 portal vein tumour thrombi, randomized trials evaluating systemic therapy combined with LRT, particularly TARE or EBRT, versus systemic therapy alone can be considered^83,129. Such designs allow formal testing of the hypothesis that improved intrahepatic disease control might delay liver decompensation driven by tumour progression. and ultimately translate into survival benefit⁸³.

Accurate assessments of liver function are a key component of trial design⁴. The ALBI score has emerged as a reproducible and objective tool that provides more granular prognostic discrimination than the Child–Pugh score, particularly in patients with Child–Pugh class A liver function¹⁰⁸. The inclusion of patients with Child–Pugh class B liver function is not recommended in pivotal trials, and specific studies focused on this population are recommended^130–132. In fact, dedicated studies enrolling patients with Child–Pugh class B disease are needed, acknowledging the inferior prognosis and increased susceptibility to treatment-related toxicities of this subgroup^130,133. Finally, across all populations, baseline tumour biopsy sampling is advocated, not only to confirm diagnosis but also to provide appropriate material for biomarker development and translational research^9,10.

End points.

OS has long been regarded as the main end point in trials involving patients with advanced-stage HCC when testing either first-line or second-line therapies. This end point has the advantages of both objectivity and clinical relevance, and remains aligned with the priorities of most patients⁴. In the first-line setting, all pivotal phase III trials leading to drug approval have used OS as either the primary or a co-primary end point, with those testing ICI-based regimens consistently having hazard ratios of <0.80 and absolute improvements in median OS of at least 2–3 months compared with TKIs^115–117 (Fig. 4). Improving OS continues to be the ultimate goal of systemic therapy in patients with advanced-stage HCC^9–12 (Table 2). However, a reliance on OS alone comes with certain limitations in the current therapeutic landscape. With frontline ICI-based combinations now extending median OS to nearly 2 years, both the cohort sizes and follow-up times required to detect significant differences in OS have increased substantially. Moreover, OS might be confounded by crossover and access to multiple post-progression therapies, which can involve ~50% of patients, and can dilute the apparent treatment effect^6,86. These challenges have led to a renewed interest in PFS as a potential surrogate end point. In trials testing non-ICI systemic therapies, strong correlations between PFS and OS have been observed (R = 0.84), and data from a meta-analysis suggest that a PFS hazard ratio of ≤0.60 is a reliable predictor of OS benefit⁸⁶. For ICI-based regimens, this correlation is weaker (R² ~ 0.5), reflecting the possibility that OS gains might occur in the absence of a PFS advantage owing to delayed but durable responses to these agents¹³⁴.

Hepatic decompensation is a clinically significant outcome in patients with advanced-stage HCC¹³⁵. While preventing decompensation is crucial in managing cirrhosis and HCC, application of decompensation as a trial end point is challenging. Decompensation events such as ascites, bleeding or encephalopathy are confounded by tumour progression, treatment toxicities and complications related to the natural history of cirrhosis and portal hypertension. Current analytical approaches face methodological limitations, and despite recommendations published in August 2025 to use hepatic decompensation or time to decompensation as an end point¹³⁵, the current panel estimates that available data are insufficiently mature and that further research is needed to refine the best approach to capture this event.

Owing to the sometimes delayed effects of ICIs, potential early severe toxicities and the potential for long-lasting responses, some methodological adaptations are required when evaluating clinical benefit from these agents. Primary end points for the target populations should reflect clinical benefit, and studies with primary end points based on landmark analysis do not suffice in this regard. In addition, NPHs should be considered, given this scenario can emerge in at least 20% of pivotal RCTs⁶⁴, an issue that should be addressed by reporting both RMST and long-term actuarial survival at 48 months and 60 months, provided that at least 10% of patients remain at risk at these time-points^6,72 (Fig. 3).

In the second-line and later-settings, in which survival durations remain shorter, OS remains the most appropriate and reliable end point¹⁰. The limited prospective evidence supporting any specific interventions following ICI-based combinations underscores the need for more robust data (Fig. 3). Secondary end points are also indispensable, and can potentially complement OS in both the first-line and second-line settings. ORR, TTP, DOR and PROs are all recommended in this regard because they have been consistently reported to capture important aspects of tumour dynamics and/or patient experience^4,5.

Stratification factors.

Balanced randomization is essential in trials enrolling patients with advanced-stage HCC owing to the need to ensure comparability between treatment arms as well as reliable interpretation of outcomes⁴. Historically, pivotal trials testing first-line therapies have incorporated BCLC stage (B versus C), ECOG performance status (0 versus 1), baseline serum AFP levels (≥400 ng/ml versus <400 ng/ml), presence versus absence of macrovascular invasion and/or extrahepatic spread, and geographical region as key stratification factors^115–117. The role of aetiology as a stratification factor remains debatable, particularly given that MASLD-related HCC and alcohol-associated liver disease-related HCC are typically not properly captured in phase III trials¹³⁶. Data suggesting that the MASLD-HCC subgroup of patients might be less responsive to immunotherapies^137,138 have thus far not been corroborated by analyses of differences between viral and non-viral aetiologies^71,139. Nonetheless, the current panel endorsed stratification by aetiology, particularly among patients with MASLD-related HCC, to enable the interpretability of subgroup analyses, which might inform future biomarker-driven treatment strategies.

In both the second-line and later-line settings, additional dimensions become relevant. The nature of prior systemic therapy (ICI-based regimens versus TKIs) and the best response or duration of benefit from earlier treatments might have implications for the efficacy of subsequent therapies as well as prognosis. Including these parameters in stratification ensures that the results of trials testing post-progression therapies reflect true treatment effects rather than imbalances in previous treatment exposures¹⁴⁰.

Control arm.

The rapid evolution of systemic therapy for patients with advanced-stage HCC has reshaped what constitutes an appropriate control arm. In the first-line setting, ICI-based regimens, comprising either an anti-PD-L1 antibody plus VEGF blockade (for example, atezolizumab–bevacizumab) or an anti-PD-(L)1 antibody plus an anti-CTLA4 antibody (for example, tremelimumab–durvalumab or nivolumab–ipilimumab), are now established comparators¹⁰. The choice of comparator should be dictated by the design of the experimental arm: when a novel compound is added to a specific approved regimen, that regimen should serve as the control arm, whereas for trials testing a completely new mechanism of action, any of the established ICI-based doublets can be used as an appropriate comparator. For patients with contraindications to ICIs, sorafenib or lenvatinib remain valid control options (median OS 13–15 months versus 19–20 months, respectively)^9–12.

In the second-line setting, the absence of agents specifically approved after ICI-based combinations creates additional complexity, and both clinical equipoise and ethical standards favour the use of active comparators¹⁴⁰. TKIs (including sorafenib, lenvatinib, regorafenib and cabozantinib) are the most pragmatic choices, with trial designs either mandating a specific agent or allowing individual physicians to select from these agents^9–12. ICI-based regimens (comprising an anti-PD-(L)1 plus an anti-CTLA4 antibody) with distinctly different mechanisms of action to those administered in the first line (such as atezolizumab–bevacizumab) or other combinations can be tested in the second line. Future trials should therefore prespecify prior treatment exposures and incorporate appropriate stratification to enhance interpretability and facilitate more robust cross-trial comparisons while also acknowledging the inherent limitations of indirect cross-trial evaluations. Double-blind, double-dummy designs remain the standard recommendation for trials assessing combination regimens in the active arm.

Overall benefit assessment

Interpretation of benefit in patients with advanced-stage HCC requires the integration of efficacy, safety and patient-centred outcomes^5,141. For trials with proportional hazards, a hazard ratio for OS ≤0.80 is considered a meaningful threshold for both regulatory purposes and clinical relevance. In the presence of NPHs (~20% of phase III trials involving patients with HCC), specific statistical tests (such as the MaxCombo test) need to be applied to assess superiority for the primary end point. When assessing a potentially positive trial result, RMST gains of at least 2 months at 24 months of follow-up provide an alternative benchmark for clinical benefit^6,64,72 (Fig. 3).

Crucially, trial interpretation must extend beyond OS to incorporate safety, with pre-eminent relevance to TRAEs of grade 3 or 4, PROs, duration of treatment exposure, access to post-progression therapies and biomarker-defined subgroups, all of which have implications for the generalizability and regulatory relevance of results. As multiple active agents become available, defining clinically meaningful benefit requires the balancing of efficacy against toxicity and resource implications, ensuring that advances in efficacy translate into real gains for patients^5,6.

Biomarkers for patient selection and stratification

Unlike many other tumour types, HCC remains largely without validated biomarkers to guide treatment decisions with the exception of elevated serum AFP for the use of ramucirumab¹²⁶ (Table 8). The development of biomarkers in HCC has been hindered by under-sampling (owing to a reliance on radiological rather than histological diagnosis), biased tissue acquisition, marked intertumoural and intratumoural heterogeneity, a paucity of actionable oncogenic drivers relative to many other solid tumours, and the confounding effects of underlying liver disease, which can have implications for prognosis, candidacy for therapy and the tumour immune microenvironment¹⁴². Efforts to develop targeted therapies based on specific driver mutations¹⁴³ or immune cell infiltration¹⁴⁴ have been largely limited to phase II trials that have thus far not led to pivotal positive phase III trials, thus precluding adoption by regulatory agencies. Several biomarkers predictive of survival outcomes generated based on data from pivotal phase III studies (such as gene signatures of response and blood-based cytokines) have not entered clinical practice^145–148. In this regard, aside from the established pathways for regulatory approval relying on data on predefined biomarkers from dedicated phase II–III trials, a biomarker development approach based on post hoc analysis has been proposed¹⁷.

The landscape of systemic therapies has expanded rapidly, mostly following trials enrolling ‘all comers’, or unselected patients. Despite ~25% of HCCs harbouring a potentially druggable alteration, no successful phase III trials enriched for such molecular aberrations have thus far been conducted^2,143. Currently, several investigational agents are being developed to target biomarker-defined subtypes of HCC, such as CTNNB1 mutations and FGF19 overexpresion¹⁴⁹. Glypican 3, a cell surface protein expressed in ~50–70% of HCCs that is largely absent in non-malignant tissues, is a particularly attractive target for novel modalities such as bispecific T cell engagers, chimeric antigen receptor T cells, antibody–drug conjugates and targeted radiopharmaceuticals¹⁵⁰. Histology-agnostic approvals, for example, larotrectinib–entrectinib for NTRK fusions, BRAF–MEK inhibitors for BRAF^V600E and T-DXd for HER2-positive tumours might benefit the <1% of patients with HCC who are eligible for these agents¹⁵¹.

Despite these advances, no biomarker-driven systemic therapy beyond serum AFP for ramucirumab has received regulatory approval for use in HCC¹²⁶. Most targeted approaches remain in the early stages of clinical development, and exploratory signals suggesting efficacy have not yet translated into positive phase III trials. Furthermore, regulatory frameworks are increasingly demanding robust biomarker validation to justify trial enrichment or stratification, underscoring the need for well-defined strategies to integrate biomarkers into clinical research¹⁷ (Fig. 2). Advancing effective biomarker-driven therapies in HCC will likely therefore require systematic incorporation of tumour and blood sampling alongside molecular profiling into pivotal trials for regulatory approval¹⁷. Such integration might accelerate discovery, validation and eventual clinical implementation, and enable more personalized and evidence-based treatment approaches.

Quality of life and patient-reported outcome assessments

PROs have the advantage of bringing the patient’s perspective into clinical research and can provide data that complements objective clinical end points¹⁵². PRO data can contribute to a drug’s benefit–risk assessment, help to define treatment effectiveness for regulatory purposes, support labelling claims, and inform future patient–physician communications^31,153,154 (Table 9).

Both the FDA and the EMA have developed guidance documents on the use of PROs in cancer trials, addressing aspects such as core PROs, instrument selection, trial design and labelling considerations^153–155. PROs incorporated into clinical trials should be prespecified with clearly defined objectives, clinically meaningful thresholds and timing of assessments. However, particular attention must be given to handling of missing data and sensitivity analyses, as differential levels of attrition might bias the longitudinal interpretation of health-related QOL, particularly in patients with advanced-stage HCC. Time-to-deterioration end points and sustained health-related QOL benefit might complement survival outcomes, particularly in non-inferiority settings or when therapies aim to preserve functional status and/or minimize toxicity^147,149,150.

The EORTC QLQ-C30 (ref. 156) and the FACT–General¹⁵⁷ questionnaires with their respective liver cancer-specific supplementary modules EORTC QLQ-HCC18 (ref. 158) and FACT–Hepatobiliary¹⁵⁹ are the most extensively validated PRO questionnaires in patients with HCC³¹. These questionnaires enable assessments of overall QOL, as well as individual concepts contributing to QOL, such as disease symptoms, and physical, role, emotional, and social functioning and well-being. Benefits assessed using these instruments can be associated with objective clinical end points, as has been the case with several trials testing first-line therapies¹⁶⁰. For example, time to deterioration in patient-reported QOL was significantly longer with atezolizumab plus bevacizumab than with sorafenib, consistent with the observed survival benefit¹¹⁵.

QOL outcomes can also lead to either an upgrade or a downgrade of the ESMO Magnitude of Clinical Benefit Scale (MCBS) score⁵. To be eligible for a QOL adjustment, QOL data need to fulfil three prerequisites (QOL being at least a secondary end point; use of a valid and reliable QOL instrument; and statistically and clinically significant improvements in QOL) and four additional items on a checklist¹⁶¹. While PROs provide essential insights into tolerability and the patient perspective, QOL cannot be assessed as an end point independent of OS, given that QOL measures are conditional on patients being alive. This intrinsic dependency explains why OS and QOL must be interpreted together, with joint evaluations providing a more accurate and patient-centred measure of clinical benefit (Fig. 4).

Symptomatic toxicities assessed by clinicians should be reported using the CTCAE scale²⁹, although in some instances these events are under-reported compared with patient reporting¹⁶². In an effort to standardize assessments of patient-reported symptomatic adverse events and to complement existing safety evaluations, the US National Cancer Institute developed the PROs version of the CTCAE (PRO-CTCAE)¹⁶³. The phase III HIMALAYA and CheckMate 9DW RCTs both included PRO-CTCAE as an exploratory end point in their protocols^116,117,164. In this context, a network meta-analysis including data from 6,425 patients with advanced-stage HCC from nine phase III RCTs found that atezolizumab–bevacizumab provided the most favourable balance between survival and QOL, and was the most-effective regimen in terms of delaying deterioration in global health (Fig. 4), pain, jaundice and abdominal swelling, thus underscoring the clinical value of integrating PROs into treatment evaluations¹⁶⁰.

Disparities in trial design

Given the substantial variations in host and hepatitis virus genetics, metabolic dysfunction and exposure to chronic environmental, dietary, economic, psychosocial and cultural stressors globally, which contribute to both the intertumoural and intratumoural heterogeneity of HCC, ensuring enrolment of patients across all continents and demographic groups in clinical trials is essential^165,166. The panel encouraged balanced regional distributions in enrolment for pivotal global phase III trials. RCTs enrolling patients from a single continent are discouraged, given that heterogeneity in patient aetiology and management, among others, likely prevents or limits the generalization of results (Table 10). In this regard, the initially positive SHARP trial, which provided pivotal evidence supporting the efficacy of sorafenib, enrolled patients in Europe, Australasia, the USA and South America^118,119, and had to be replicated in Asia (Asia–Pacific trial)¹⁶⁷ to gain international acceptance. Elsewhere, this limitation was addressed by a meta-analysis of pooled data from two phase III trials (KEYNOTE-240 and KEYNOTE-394) to collectively analyse the extent of benefit from pembrolizumab as second-line therapy¹⁶⁸. Conversely, the conclusions of positive trials conducted exclusively in Asia, such as ORIENT-32, which demonstrated the efficacy of sintilimab plus the VEGF biosimilar IBI305 (ref. 169), cannot be extrapolated to Western guidelines.

Although low-income and middle-income countries (LMICs), particularly in sub-Saharan Africa, account for a disproportionately high share of the global HCC burden (~40%), these locations remain markedly under-represented in so-called global clinical trials (typically providing <5% of patients)^1,170. To achieve adequate representation, greater efforts are required to build infrastructure and professional capacity enabling the administration of standard-of-care HCC therapies and supporting trial activities at institutions in LMICs¹⁶⁶. Prespecifying subgroup analyses by race and ethnicity, sex and liver disease aetiology, as well as standardizing the reporting of demographic, socioeconomic and clinical variables, including enrolment, dropout rates and outcomes, are crucial to evaluating trial representativeness.

Patients with HCC who are also HIV-positive require specific attention and should be considered for clinical trials when the disease is controlled, defined by viral loads of <200 copies/ml and CD4 counts of ≥200/μl^171,172. This consideration is particularly relevant in regions with a high incidence of HIV co-infections with chronic hepatitis B or hepatitis C¹⁷². Regulatory agencies have an important role in mandating equitable representation in pivotal clinical trials conducted for new drug and device registrations, or as a minimum, in requiring additional postmarketing studies or real-world evidence from LMICs, along with dissemination among health-care professionals in these regions^173,174.

The patient perspective

A diagnosis of HCC generates profound emotional distress for both patients and their families, underscoring the need for structured support and clear, honest communication¹¹. Providing accurate information on prognosis and treatment options is not only a cornerstone of quality care but also essential to facilitating enrolment in clinical trials¹¹. Beyond clinical parameters, social determinants of health can substantially influence outcomes, with strong family and social support consistently associated with improved QOL in patients with liver disease¹⁷⁵. Regardless of disease stage, all patients with HCC should be offered emotional and psychological support, nutritional support and pain relief where appropriate^3,176. Additionally, patients will need close monitoring to identify clinical or biochemical liver decompensation and begin appropriate treatments. This consideration is extremely important for patients presenting at a disease stage that precludes primary antitumour therapies; for example, those with advanced liver dysfunction or BCLC-D disease.

The management of HCC is inherently multimodal and requires coordination with a multidisciplinary team^9–12. Patients enter this process with expectations of therapeutic benefit, but also with fears regarding adverse effects and prognosis. Transparent and personalized counselling is therefore essential to support shared decision-making¹⁷⁷. In the context of clinical research, strategies such as structured educational initiatives, strengthened communication pathways and peer support networks have been shown to enhance the engagement and participation of patients and close family members¹⁷⁸. These considerations highlight the importance of embedding the perspective of patients from the earliest stages of trial design and conduct, ensuring that objectives, procedures and communication strategies align with patients’ priorities, expectations and real-world feasibility.

Regulatory requirements

Regulatory agencies such as the FDA and EMA have progressively refined their frameworks to ensure that new oncology therapies deliver meaningful clinical benefit. Accelerated approval pathways have facilitated earlier access to promising agents in the USA, although this mechanism requires that significant improvements observed in surrogate end points ultimately translate into improvements in key outcomes, such as OS^24,179,180.

Historically, many oncology trials have relied on surrogate measures such as PFS and ORR, often without robust validation of their associations with survival outcomes¹⁸¹. As a result, regulators have become increasingly cautious, stressing the need for mature OS data and/or confirmatory studies. FDA guidance from 2023 highlights that surrogate end points must be justified by biological plausibility and historical correlation with OS, whereas the EMA requires contextualization of effect size against available alternatives and disease stage^24,180. An analysis of 102 FDA accelerated approvals (1992–2022) that were further confirmed with phase III trials showed a direct balance between rapid access and evidence generation. The 77% of RCTs showing intermediate or higher levels of benefit according to the ESMO MCBS correlated with faster conversion, compared with the 23% of trials deemed to provide low levels of benefit^182–184.

The experience in HCC illustrates these challenges. Two phase III RCTs (EMERALD-1 and LEAP-012) met their primary end point of PFS, albeit ultimately with no significant difference in OS^79,80. Only the Chinese NMPA has adopted the LEAP-012 regimen (TACE plus pembrolizumab and lenvatinib), whereas Western agencies are reluctant to approve therapies based only on PFS and without clear evidence of OS benefit or other notable advantages such as a substantially improved safety profile. The role of PFS alone as an end point is currently under discussion. For example, signals of early efficacy (such as PFS) might be masked in trials assessing OS as the primary end point, owing to crossover and the availability of multiple subsequent therapies.

In parallel, an increasing recognition of the need to assess the magnitude of benefit beyond statistical significance is emerging. Tools such as the ESMO MCBS and the ASCO Value Framework provide structured approaches to support the contextualization of efficacy, although the formal adoption of such frameworks in regulatory decision-making remains limited^5,141. Nevertheless, both frameworks emphasize the importance of transparency in reporting of benefits, risks and uncertainties, which aligns with the FDA’s calls to make benefit–risk assessments more accessible and systematically incorporate such assessments into clinical decision-making.

Methodological complexities add another layer of difficulty. NPH, non-informative censoring and treatment crossover can all distort survival analyses, obscure long-term effects, and reduce the interpretability of standard measures such as hazard ratios or median OS^6,64. To address these issues, the FDA and EMA increasingly recommend incorporating alternative methods, including RMST, milestone analyses and prespecified sensitivity analyses, all of which can provide more robust and clinically interpretable estimates^72,185. Both the FDA and EMA now encourage early dialogue with sponsors to ensure alignment on end point selection and confirmatory strategies, especially when accelerated approval is considered.

Ultimately, regulatory acceptance of new therapies depends not only on statistical significance but also on the strength, consistency and durability of observed benefit. The balance between earlier access through accelerated approval and the assurance of true patient benefit remains delicate. This delicate balance underscores the need for trial designs that integrate validated end points, long-term follow-up, rigorous methodology and clear consensus on how to measure clinical benefit according to disease stage.