Updated Network Meta-Analysis of First-Line Systemic Treatments for Advanced HCC: Consistent Role of TACE Open Access

The oral multiple tyrosine kinase inhibitor (TKI) sorafenib (Sora) was previously the mainstay of primary therapy for advanced hepatocellular carcinoma (HCC) [1, 2] until its clinical benefits were superseded by the emergence of a combination of the programmed death ligand-1 (PD-L1) inhibitor atezolizumab (Atez) and the anti-vascular endothelial growth factor (VEGF) inhibitor bevacizumab (Beva), which established a new standard of care in the front-line setting [3, 4]. Thereafter, recent years have seen the advance of other effective immune-oncology (IO) and targeted agents, which have demonstrated improvements in overall survival (OS) [5], as well as positive outcomes in combination with systemic and interventional therapies [6, 7].

In accord with the evolving treatment landscape for advanced HCC, we had conducted a network meta-analysis (NMA) of randomized controlled trials (RCTs) comparing various mono or combined regimens of interventional and targeted therapies with Sora, the old standard of care. This demonstrated that the Atez + Beva regimen was the best systemic modality, being significantly superior to Sora in both OS and progression-free survival (PFS) outcomes. Our subgroup analysis for different tumor settings also showed that addition of transarterial chemoembolization (TACE) to Sora may improve the outcomes of patients with portal vein tumor thrombosis (PVTT) [8].

Since the publication of these results in 2022, several NMAs have explored the survival ranks of competing systemic treatments with targeted and immune agents [9‒14]. However, there are few comparative analyses that have included interventional therapies as an important part of the treatment strategies. In addition, along with the approval of lenvatinib (Lenv) after demonstration of its non-inferiority to Sora [15], completed or ongoing trials have also evaluated different treatment strategies with Lenv as a control arm, which necessitates the inclusion of RCTs with such design in a further NMA. Therefore, this study forms an update to our previous NMA [8], evaluating the efficacy and safety data of RCTs comparing interventional and targeted therapies with standard TKI controls to identify the optimal first-line treatment for various tumor settings in patients with advanced HCC.

This systematic review and NMA was conducted in accordance with the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) extension statement for systematic reviews incorporating NMAs [16]. The protocol is registered in PROSPERO (Prospective Register of Systematic Reviews) as No. CRD42024545453. The Institutional Review Board of Asan Medical Center approved this trial-level NMA and waived the requirement for informed consent from individual patients (Institutional Review Board No. 2024-0684).

We aimed to identify all phase 2 or 3 RCTs that compared single or combination systemic regimens with Sora or Lenv as the standard treatment control in patients with advanced (unresectable or metastatic) HCC. In this updated NMA, we especially focused on RCTs published since 2018, a landmark year when Lenv demonstrated non-inferior efficacy compared to Sora, marking a pivotal shift toward diversified standard therapeutic options for HCC. Studies were excluded if they were (1) trials testing interventional therapies alone without combined systemic agents, specifically in locally advanced HCCs without extrahepatic metastasis (EHM); (2) trials without Sora or Lenv as a control arm; (3) trials not published in the form of an original article (e.g., conference abstracts, posters, and letters); and (4) non-English language articles.

Two separate literature searches were conducted to identify studies relevant to this NMA. Our previous literature search was conducted using the following databases: PubMed, Embase, Cochrane Library, CINAHL, and Web of Science. The search was limited to articles published in English from January 1, 2008, to February 28, 2021 [8]. The updated search was conducted on February 29, 2024, using the same search algorithm with MeSH terms and text words in combination with an RCT filter to identify any newly published studies that met the inclusion criteria. The detailed search strategy is presented in Table S1 (for all online suppl. material, see https://doi.org/10.1159/000546697). The list of titles and abstracts were independently screened by two reviewers (Y.R. Kim and J. An) to include relevant studies using a Cochrane online systematic review tool (Covidence). After excluding irrelevant and duplicated studies, the remaining studies were further screened by reviewing the full text. Any disagreement was resolved by discussion. If more than one report of the same trial was found, only the most up-to-date publication was included in the analysis.

The primary outcome was OS, defined as the time from the date of randomization to death from any cause, measured in the intent-to-treat population. Secondary outcomes included PFS, defined as the interval from random assignment to either disease progression or death, whichever occurred first. Radiographic tumor assessments were determined according to Response Evaluation Criteria in Solid Tumors (RECIST) 1.1 in 6 studies, according to modified RECIST in 4 studies, and according to both criteria in 5 studies. When results for both criteria were available, RECIST was prioritized for PFS analysis. In trials where PFS was not reported, time from randomization to objective tumor progression (TTP) was substituted as a surrogate endpoint [17]. Safety data were evaluated according to the proportion of patients with serious adverse events (SAEs) of grade 3 or higher classified according to the National Cancer Institute Common Terminology Criteria for Adverse Events.

Two independent investigators (Y.R. Kim and E. Kim) completed a standardized spreadsheet recording the following information from all included trials: (1) characteristics of the study design including interventional and control treatments; (2) baseline patient and tumor characteristics including age, sex, follow-up duration, Eastern Cooperative Oncology Group performance status, race/region, Child-Pugh score, albumin-bilirubin (ALBI) grade, hepatitis B virus (HBV) infection, hepatitis C virus (HCV) infection, Barcelona Clinic Liver Cancer (BCLC) stage, presence of PVTT and EHM, and serum alpha-fetoprotein (AFP) level at baseline; and (3) primary and secondary outcomes. Hazard ratios (HRs) and associated 95% confidence intervals (CIs) for OS, PFS, and TTP were extracted. The proportions (%) of SAEs were used to calculate their estimated odds ratios (ORs). In addition, the proportion of patients experiencing adverse events, frequency of sequential treatment implementation, and information on post-progression survival evaluation were also collected. We evaluated the main text and online supplementary materials to implement an extensive and detailed data extraction. If some data were not reported in the original study, we extracted information using post hoc analysis or follow-up abstracts derived from relevant trials. All data used were publicly available or computable from the included studies. Any disagreement was resolved by discussion.

The risk of bias was assessed for each included RCT using the Cochrane Risk of Bias Tool [18], for which individual aspects in the following six domains were assigned a judgment (high, low, or unclear): random sequence generation and allocation concealment (both within the domain of selection bias), blinding (performance bias), blinding of outcome assessors (detection bias), incomplete outcome data (attrition bias), selective reporting (reporting bias), and an auxiliary domain (other bias). Two reviewers (Y.R. Kim and H.I. Kim) independently evaluated the quality of the trials and resolved disagreements by discussion.

The NMA comparing interventions with the standard treatment control (i.e., Sora or Lenv) was conducted using performed least squares regression [19]. Pooled effect sizes were estimated with fixed-effects models and the treatment effects were quantified using HRs for survival outcomes and ORs for SAEs. We visualized the results through forest plots. Network plots were constructed to summarize the studies and illustrate the relationships between treatments and clinical endpoints. To rank the treatments and assess their relative efficacy, we calculated the P score for each intervention, representing the degree to which a treatment outperforms others. We also generated ranking curves derived from rankograms and calculated surface under the cumulative ranking curve (SUCRA) values, where higher SUCRA scores, approaching 1, indicate greater efficacy. We presented forest plots from the meta-analysis model for OS, using Lenv, Atez + Beva, and tremelimumab + durvalumab (Trem + Durv) as the respective reference groups. In addition, subgroup analyses for OS were performed separately based on the presence of PVTT, presence/absence of EHM, BCLC stage (stage B vs. stage C), etiology of liver disease (HBV vs. HCV vs. nonviral etiology), serum AFP level at baseline (<400 vs. ≥400 ng/mL), age (<65 vs. ≥65 years), race of study participants, study region (Asian region vs. the rest of the world), liver function assessed by Child-Pugh score (5 vs. 6), and ALBI grade (1 vs. 2). We assessed the transitivity assumption to ensure the comparability of treatments across studies. Descriptive statistics for patient characteristics, including age, sex, Child-Pugh score, HBV status, HCV status, BCLC stage, PVTT, and EHM, were compared across studies using the one-sample Wilcoxon test. All statistical analyses were conducted using R software (version 4.3.0), with the netmeta and gemtc packages employed for the main NMA, as well as for the calculation of ranking probabilities and SUCRA values.

A flowchart of the study selection process is shown in Figure 1. Through the literature search, a total of 5,045 studies were initially screened using Covidence, of which 123 studies fulfilled the eligibility criteria and were assessed by full-text screening. After review, 16 studies were finally selected for the NMA, with these being composed of 2 phase 2 and 12 phase 3 RCTs with Sora as a control and 2 phase 3 RCTs with Lenv as a control in the first-line setting for advanced HCC. The inclusion and exclusion criteria of the included studies are presented in online supplementary Table S2. Studies that made comparisons with Sora as the control arm evaluated 10 molecularly targeted regimen mono- or dual therapies in 6,854 patients, 3 transarterial treatments in combination with Sora in 1,037 patients, and 1 intratumoral therapeutic approach in combination with Sora in 459 patients. These treatments included Atez + Beva, Lenv, cabozantinib + Atez (Cabo + Atez), nivolumab (Nivo), camrelizumab + rivoceranib (Cam + Rivo), sintilimab + IBI305 (a bevacizumab biosimilar) (Sin + IBI305), donafenib (Dona), tisrelizumab (Tis), Trem + Durv, and Durv; hepatic arterial infusion chemotherapy + Sora (HAIC + Sora), TACE + Sora, and selective internal radiation therapy + Sora (SIRT + Sora); and pexa-vec + Sora (Pexa + Sora), respectively [3, 15, 20‒31]. For the Atez + Beva combination, the analysis incorporated updated data from Cheng et al. [3] (2022), while the previous NMA utilized the IMbrave150 trial data reported by Finn et al. [4] (2020). Two studies with Lenv as the control arm included 1,132 patients, comparing Lenv with TACE + Lenv [6] and Lenv with Lenv + pembrolizumab (Lenv + Pem) [32]. All RCTs consisted of two treatment arms, except for one three-arm trial that compared Sora with Trem + Durv and Durv monotherapy. A network diagram of treatment comparisons is presented in online supplementary Figure S1. The circle size represents the mean number of participants per trial, while the line width reflects the number of direct comparisons. The absence of a line connecting two treatments indicates that there was no direct comparison.

The characteristics of the included studies are outlined in Table 1. The inclusion criteria were generally consistent across all trials: all patients had preserved liver function according to Child-Pugh classification and had not received any previous systemic therapy for HCC. Additionally, all patients had either advanced HCC or HCC not amenable to curative or locoregional therapies, among which the proportion of patients with BCLC stages B and C ranged from 12% to 40% and from 64.1% to 88%, respectively. While most trials required measurable disease based on RECIST or mRECIST criteria, both the Park 2019 [30] and Peng 2023 trials [6] that incorporated TACE as part of the therapeutic regimen specifically excluded diffuse infiltrative HCC and a tumor burden exceeding 50%. Median age was relatively lower in the Peng et al. [6] (2023), Ren et al. [23] (2021), and Qin et al. [24] (2021) trials, which reported values of 54 (range 46–64), 53 (21–82), and 53 (46–62) years of age, respectively. The rate of viral etiology was the highest in the Peng et al. [6] (2023), Ren et al. [23] (2021), and Qin et al. [24] (2021) trials, reflecting the higher prevalence of HBV infection in Asian countries. The incidence of PVTT was mostly lower in studies that excluded patients with main portal vein invasion [15, 22, 23, 25, 27, 30, 32], with the lowest values reported as 16%, 15%, and 14.8% in the Llovet et al. [32] (2023), Qin et al. [22] (2023)(1), and Qin et al. [25] (2023)(2) trials, respectively. Efficacy and safety outcomes, as well as the median follow-up period, are summarized in online supplementary Table S3, with all included trials reporting OS as the primary endpoint. Information regarding the availability of subgroup analysis data from each included study is presented in online supplementary Table S4 [33‒36]. Details on sequential treatment implementation and post-progression survival evaluations from each study are presented in online supplementary Tables S5, S6, respectively.

The risk of bias assessed using the Cochrane risk of bias assessment tool demonstrated that all trials reported a “low risk” of bias in the majority of the seven domains of interest (online suppl. Fig. S2). However, because of the difficulty of double blinding among trials using interventions with different approaches, 15 trials showed a high risk of bias in blinding of participants and personnel. The absence of blinding in outcome assessment was found to be another reason for detection bias among 7 trials. Selective reporting was identified as a potential risk of bias in the Ricke 2019 trial [31] because of the absence of PFS/TTP data in the reported results.

Using Sora as the reference for the NMA, the HR estimates showed that TACE + Lenv (HR [95% CI], 0.41 [0.30–0.58]), Sin + IBI305 (0.57 [0.43–0.75]), Cam + Rivo (0.62 [0.48–0.80]), Atez + Beva (0.66 [0.51–0.85]), Lenv + Pem (0.77 [0.62–0.97]), Trem + Durv (0.78 [0.64–0.95]), and Dona (0.83 [0.70–0.99]) showed significantly improved OS in comparison with Sora (Fig. 2a). Among these therapies, TACE + Lenv had the highest probability of being the most effective according to the treatment ranking analysis (SUCRA and P score, 0.993 and 0.992). Despite the observed survival benefit of TACE + Lenv over Sora, the combination of transarterial therapies with Sora did not significantly improve survival in any of the three modalities: TACE + Sora (0.91 [0.68–1.21]), SIRT + Sora (1.01 [0.82–1.25]), and HAIC + Sora (1.03 [0.79–1.34]). Additionally, although the outcomes were not statistically significant, IO monotherapies, specifically Tis, Nivo, and Durv, showed similar reductions in mortality risk to Sora, with HRs of 0.85 [0.71–1.02], 0.85 [0.71–1.02], and 0.86 [0.71–1.05], respectively. Pexa + Sora again had the highest probability (64.0%) of being the lowest ranked in the rankogram (online suppl. Table S7A; Fig. S3A).

The PFS outcomes of individual regimens in comparison with Sora are presented in Figure 2b. This analysis included 15 studies evaluating 15 treatment modalities. Among these, 14 studies involving 8,990 patients provided HRs for PFS, while Kondo et al. [28] (2019) trial involving 68 patients used TTP outcomes instead of PFS. However, the Ricke et al. [31] (2019) trial did not report either PFS or TTP outcomes. Compared with Sora, TACE + Lenv showed the most substantial reduction in the risk of PFS events (0.28 [0.21–0.38]) and had the highest probability of being the most effective treatment for improving PFS according to the rankogram, SUCRA (1.0), and P score (1.0) (online suppl. Table S7B; Fig. S3B). Following TACE + Lenv, all IO-based combination therapies, except for Trem + Durv, demonstrated a similar trend of significantly improved PFS. Lenv (0.65 [0.57–0.77]) and TACE + Sora (0.73 [0.59–0.91]) also showed significant PFS extension, whereas Pexa + Sora was associated with poorer PFS outcomes (1.22 [0.97–1.53]), consistent with our OS outcome. Stratified analysis based on imaging response criteria, encompassing 11 studies utilizing RECIST and 9 studies utilizing mRECIST, demonstrated similar trends in treatment outcomes (online suppl. Fig. S4).

The NMA for OS using Lenv as the reference revealed similar outcomes: TACE + Lenv (0.45 [0.33–0.61]) and the three ICI + TKI combinations of Sin+ IBI305 (0.62 [0.45–0.84]), Cam + Rivo (0.67 [0.50–0.90]), and Atez + Beva (0.72 [0.54–0.96]), were associated with a significantly higher reduction in mortality (online suppl. Fig. S5A). Correspondingly, TACE + Lenv ranked the highest in efficacy based on SUCRA (0.993) and P score (0.992) analyses. However, Lenv + Pem (0.84 [0.71–1.00]), Trem + Durv (0.85 [0.66–1.09]), and Dona (0.90 [0.72–1.13]), which showed significant OS benefits compared with Sora (Fig. 2a), did not demonstrate such significantly better efficacy in comparison with Lenv.

In comparison with Atez + Beva as the control, TACE + Lenv showed significantly improved OS (0.63 [0.41–0.95]) and had the highest probability of being the most effective for reducing the risk of death according to both the P score and SUCRA value of 0.992 and 0.993 (online suppl. Fig. S5B). The IO-based combinations of Sin + IBI305, Cam + Rivo, Lenv + Pem, and Trem + Durv demonstrated similar survival outcomes without achieving statistical significance, whereas Cabo + Atez was the only IO combination regimen associated with poorer OS in comparison with Atez + Beva (1.48 [1.05–2.10]). When we conducted analyses restricted to immunotherapy-based regimens only, similar trends were observed (online suppl. Fig. S6). Similarly, compared with Trem + Durv as the control, TACE + Lenv showed the most substantial improvement in OS, with a HR of 0.53 (95% CI: 0.36–0.79), while the efficacy outcomes of IO-based combinations were not significantly different (online suppl. Fig. S5C).

Data on the SAEs of ≥grade 3 were extracted from 13 RCTs for a total of 14 different treatments in 8,870 patients. The two trials that incorporated HAIC + Sora were excluded because they did not report safety data [28, 29]. Although the Peng 2023 trial provided safety information on individual adverse events [6], the proportion of overall incidences of SAEs ≥grade 3 was not available, and this study was therefore excluded from the analysis.

Most treatments had significantly higher risks of SAEs than Sora, with Lenv + Pem having the highest risk of an incident (OR [95% CI], 3.08 [1.98–4.79]), followed by Pexa + Sora (2.11 [1.45–3.07]) and TACE + Sora (2.02 [1.23–3.32]). The results are presented in Figure 3. Of note, while Atez + Beva was not associated with a significant increase in SAEs in our previous NMA [8], the present analysis using the results of the Cheng 2022 trial with updated safety data revealed that the combination had a significantly higher rate of SAEs compared with Sora, with an OR of 1.95 (95% CI: [1.32–2.88]). Overall, IO monotherapies showed a similar risk for SAEs as found with Sora, while IO-based combination therapies were generally associated with worse safety profiles. Online supplementary Table S8 summarizes the frequency of both immune-related and treatment-related adverse events observed across all 9 IO-based regimens evaluated in this analysis. Although Dona (0.82 [0.55–1.22]) did not show a significant reduction in the rate of SAEs, it achieved the highest rank in safety in the rankogram, with a probability of 62.5% (online suppl. Table S7C). In terms of Lenv, the relative dose intensity and proportion of dose alterations in Lenv and Lenv-based therapies were comparable, as shown in online supplementary Table S9.

Among 13 studies that reported HRs for a subset of patients with PVTT, TACE + Lenv showed the greatest reduction in risk of death (HR [95% CI], 0.34 [0.21–0.55]) in comparison with Sora, followed by Sin + IBI305 (0.57 [0.37–0.87]) and Atez + Beva (0.68 [0.47–0.98]) in order of statistical significance. Although TACE + Sora (0.52 [0.27–1.02]) and Cam + Rivo (0.56 [0.28–1.13]) demonstrated borderline superiority over Sora in respect to reducing mortality, the differences were not statistically significant (Fig. 4a). Twelve studies reported OS outcomes in patients with EHM. While TACE + Lenv provided the greatest improvement in OS (0.50 [0.33–0.76]), all included IO-based combination therapies, except for Cabo + Atez, were associated with significant OS benefits in comparison with Sora (Fig. 4b). Based on 13 studies, TACE + Lenv was also ranked highest in OS among patients without EHM (Fig. 4c). Twelve studies were included in the subgroup analysis based on BCLC stage (online suppl. Fig. S7). Several immunotherapy-based combinations demonstrated significant benefits in BCLC stage C patients compared to Sora, while only Sin + IBI305 (0.25 [0.09–0.67]) was significant in BCLC stage B patients.

Subgroup analyses stratified by serum AFP levels at baseline are presented in online supplementary Figure S8. In patients with low AFP (<400 ng/mL), several treatments demonstrated significantly improved OS compared to Sora, with Cam + Rivo (0.40 [0.26–0.60]), TACE + Lenv (0.45 [0.28–0.73]), and Sin + IBI305 (0.54 [0.35–0.83]) showing the most substantial benefits, supported by high P scores and SUCRA values. In contrast, in the high AFP (≥400 ng/mL) subgroup, no treatment demonstrated statistically significant superiority over Sora, although TACE + Lenv showed the most pronounced treatment effect estimate with a HR of 0.30 (0.06–1.64; P score 0.993).

The subgroup analyses of OS according to etiology included 12 studies with HRs for HBV-related HCC, 10 studies for HCV-related HCC, and 9 studies for non-viral HCC (Fig. 5 a–c). With the exception of Cabo + Atez, all combinations with IO agents significantly improved OS compared to Sora in HBV-infected HCC patients. TACE + Lenv showed the most significant reduction in mortality risk for HBV-related HCC (0.39 [0.27–0.57]); however, TACE + Lenv data for HCV and nonviral HCC subsets were unavailable and thus not included in the subgroup analysis. None of the treatment modalities were associated with improved OS compared to Sora in patients with nonviral HCC (Fig. 5c).

Next, we performed subgroup analyses stratified by patient age (online suppl. Fig. S9), region where study was conducted (online suppl. Fig. S10), and race of participants (online suppl. Fig. S11). In the elderly cohort (≥65 years) based on 10 studies, TACE + Lenv exhibited superior OS with a HR of 0.44 (0.24–0.79) compared to Sora. This was followed by Cam + Rivo (0.49 [0.29–0.83]) and Lenv+Pem (0.63 [0.45–0.88]). Among younger patients (<65 years), TACE + Lenv similarly demonstrated the most favorable survival outcomes (0.39 [0.26–0.60]). According to study region, the overall results of the Asian region (14 studies) remained similar to the primary analysis, whereas subgroup analysis of the rest of the world (9 studies), which did not include TACE + Lenv, showed that only Atez + Beva (0.68 [0.50–0.93]) demonstrated a significant survival benefit compared to Sora.

Subgroup analyses according to liver function were conducted using ALBI grade and Child-Pugh score. For ALBI grade-based analysis, data were available from five studies (online suppl. Fig. S12). In patients with ALBI grade 1, TACE + Lenv (0.40 [0.22–0.72]) and Atez + Beva (0.50 [0.35–0.72]) demonstrated significantly improved survival compared to Sora. In contrast, among patients with ALBI grade 2, only TACE + Lenv showed a significant survival benefit (0.42 [0.27–0.66]). For Child-Pugh score-based subgroup analyses, 7 studies contributed data for patients with Child-Pugh score A5 and 6 studies for those with Child-Pugh score A6, without available data for TACE + Lenv, Atez + Beva, and Trem + Durv (online suppl. Fig. S13). Only HAIC + Sora was associated with significantly worse survival compared to Sora (1.91 [1.08–3.40]) in patients with Child-Pugh score A6 liver function.

Overall, the transitivity assumption was not challenged, and its assessment did not reveal significant differences in the baseline parameters (online suppl. Fig. S14).

Since our previous NMA on first-line treatments for advanced HCC conducted in 2022 [8], multiple RCTs have evaluated the efficacy of different IO-based regimens or TACE combined with TKI against standard treatments, among which the STRIDE regimen Trem + Durv recently demonstrated significant survival improvement over Sora [27], and is now approved by the US Food and Drug Administration as another primary option for unresectable disease [37, 38]. However, the majority of these trials were limited to comparisons with Sora as the sole control, and to date, head-to-head comparisons among these novel therapies are lacking. Consequently, we have updated our systematic review and NMA of various IOs or TKIs, and their combinations with transarterial interventions, in the first-line setting for advanced HCC, thereby providing a comprehensive comparison of up-to-date systemic therapies with old and new standard controls to aid in decision-making in different tumor settings and patient indications.

Among the 16 treatments evaluated in our NMA of RCTs, in comparison with Sora, TACE + Lenv was unequivocally associated with the greatest improvements in the OS and PFS of patients with advanced HCC. This finding was also consistent when comparing OS with different standards as the control: Lenv, Atez + Beva, and Trem + Durv. In terms of pharmacologic strategies, combinations of IOs, TKIs, and VEGF inhibitors provided significant OS benefits over Sora, with the exception of Cabo + Atez, which demonstrated a significant increase in PFS in comparison with Sora, but did not show a statistically significant improvement in OS in the original RCT [20]. However, the observed outcomes regarding safety profiles showed contrary findings: IO monotherapies showed a trend of similar safety profiles to Sora, whereas combinations of an IO agent with TKI, anti-VEGF, or another IO generally reported significantly higher risks of SAEs, possibly due to enhanced immune activation and overlapping toxicities. Interestingly, emerging evidence suggests that immune-related adverse events may serve as positive prognostic biomarkers in IO-treated advanced HCC, with studies documenting improved outcomes in patients who developed these reactions [39‒42]. This paradoxical relationship underscores the complex interplay between immune activation, toxicity, and therapeutic response. Consequently, a thorough risk-benefit assessment remains a crucial component of clinical decision-making when selecting among the expanding therapeutic options for advanced HCC.

Among drug regimens, the IO-based combinations, namely, Sin + IBI305 followed by Cam + Rivo, Atez + Beva, Lenv + Pem, and Trem + Durv in order showed significantly higher OS than Sora. While the efficacy of both Atez + Beva and Trem + Durv is well established, our study revealed that in comparison with Sora, Sin + IBI305 and Cam + Rivo were associated with the highest probabilities of improving OS, rather than the two currently accepted combination regimens. The increased benefits of Sin + IBI305 and Cam + Rivo may be attributed to the relatively younger populations (median age of 53 and 58 years, respectively, vs. 64 and 65 years in Atez + Beva and Trem + Durv trials) and the exclusion of patients with main PVTT in the respective studies [22, 23], as well as the potential advantage from the mechanism of Sin and Cam as anti-PD-1 antibodies, over anti-PD-L1 antibodies such as Atez and Durv. Previous meta-analyses showed that anti-PD-1-based combination therapy was associated with significantly higher OS than anti-PD-L1-based therapy in advanced lung and head and neck cancers [43, 44]. This difference might be due to the PD-1 binding of T lymphocytes, which elicits a systemic effect and a potentially stronger immune response. The unique binding epitope of Cam, distinct from that of other anti-PD-1 antibodies, as described in the original trial, could also contribute to its increased efficacy [22]. Additionally, both the Sin + IBI305 and Cam + Rivo trials included higher proportions of patients with HBV (94% and 76%, respectively), and furthermore, the two regimens also ranked highest among the drugs analyzed in our HBV subseries. Growing evidence suggests that patients with viral HCC may have an enhanced response to IO-mediated immune reactivation in the context of the immune-cold tumor microenvironment created by chronic viral inflammation [45‒48]. Therefore, differences in disease etiology may also explain the observed survival benefits. However, when compared with Atez + Beva or Trem + Durv as the control arm, neither Sin + IBI305 nor Cam + Rivo showed a significant OS advantage. These findings suggest that the former existing IO regimens, as well as the latter newer combinations, are likely to offer comparably favorable survival outcomes, with their use depending on availability or clinical preference. Considering that the Sin + IBI305 trial was conducted exclusively in China, its use may be particularly advocated in that region. Additionally, the demonstrated OS benefits of Sin + IBI305 and Cam + Rivo in a selected subset of HBV-associated HCC could potentially provide new promising options for treating these patients, while the standard treatment may still be recommended in the non-HBV setting.

While the combinations of Sora with different transarterial interventions did not show promising results, our updated NMA revealed that TACE + Lenv demonstrated the best survival outcomes among all modalities, with significantly greater reduction in mortality in comparison with all four standard controls. The benefits of TACE combined with systemic therapy in advanced-stage HCC have been implied in previous studies, highlighting the potential immunomodulatory effects of locoregional therapy and the prognostic value of intrahepatic tumor control [6, 7, 49]. The observed superiority of the efficacy of TACE + Lenv over not only TACE + Sora, but also HAIC or SIRT + Sora, might be due to the differences in mechanism of the two TKIs: while both are antiangiogenic, Lenv additionally targets both fibroblast growth factor receptors and hypoxia-inducible factors, effectively suppressing tumor cell proliferation in post-embolization hypoxic environments [50, 51]. Given that relative dose intensity (RDI) was comparable between Lenv and TACE + Lenv groups, the improvement in survival outcomes could be attributed to the combined effect with TACE, particularly in cases where RDI decreases with long-term administration of Lenv. Supporting this combination approach, the TACTICS-L trial provides robust evidence that TACE + Lenv confers significant efficacy with an acceptable safety profile in patients with intermediate-stage HCC [52]. Moreover, the recent LEAP-012 trial demonstrated statistically significant improvement in PFS with TACE plus Lenv plus Pem, compared to TACE plus placebo, in patients with mainly intermediate, TACE-amenable HCC [53]. The synergistic mechanism of this triple combination involves lenvatinib normalizing tumor vasculature and suppressing TACE-induced VEGF, while TACE promotes tumor antigen release and increases CD8+ T-cell infiltration, enhancing anti-PD-1 efficacy. Collectively, these data, including our findings, underscore the therapeutic potential of TACE + Lenv-based combination strategies in the management of HCC. Additional investigation of immunotherapy with TACE + Lenv in prospective clinical trials may help further optimize treatment approaches for advanced HCC.

However, despite the synergistic effects of TACE + Lenv in tumor modulation, its pronounced superiority for improving OS in comparison with other proven pharmacologic strategies should be interpreted with caution, with consideration of the specific characteristics of the trial. The patients included in the TACE + Lenv trial were generally younger with a median age of 54 years, had a tumor burden of less than 50%, and had a higher prevalence of HBV infection (87.1%), all of which may have contributed to a more favorable treatment response, especially in patients with intrahepatic tumor alone or with low levels of AFP as demonstrated in our subgroup analysis [54]. Notably, the study selectively included patients without contraindications to TACE, excluding those with portal-systemic shunting, hepatofugal flow, significant atherosclerosis, or primary bile duct invasion – factors associated with higher procedural risks and poorer outcomes [55]. Furthermore, 26 patients (15.3%) in the TACE + Lenv group, compared with 3 (1.8%) in the Lenv monotherapy group, underwent subsequent curative surgical resection [6]: the higher incidence of radical treatment in the combination group relative to the other trials could also have contributed to better survival outcomes. Collectively, the selective inclusion criteria for TACE eligibility in the Peng 2023 study might have some influence on the ranking in our NMA.

Despite these possible confounding factors, it is important to note that the LAUNCH trial included a substantial proportion of patients with PVTT (71.8%) and EHM (55.3%), and in our subgroup analysis of these patient subsets, TACE + Lenv consistently showed the first rank of OS superiority in comparison with Sora, with HRs of 0.34 (95% CI: 0.21–0.55) and 0.50 (0.33–0.76), respectively. Additionally, although TACE + Sora was previously identified as the leading treatment for improving OS in the PVTT subset in our prior NMA [8], the updated analysis presented here reveals that TACE + Lenv was associated with the best outcomes in such patients, followed by TACE + Sora, which showed borderline superiority over Sora. This finding underscores the importance of effective locoregional tumor control combined with systemic therapy upon the prognosis, especially in patients with PVTT [56‒58]. In contrast, the IO-based combinations that have shown promising treatment responses – Cam + Rivo, Sin + IBI305, Trem + Durv, and Lenv + Pem – were investigated in patient groups with a low incidence of PVTT because of the original exclusion of patients with main portal vein invasion, thereby limiting the comparison of their efficacy in this specific tumor condition. However, Atez + Beva continued to demonstrate a significant extension of OS in comparison with Sora, consistent with the results of our previous NMA. While further evidence for the use of newer IO-based combination therapies in these patients is warranted, our results suggest that in current practice, patients with PVTT supported by Child-Pugh class A liver function are likely to benefit most from TACE + Lenv and Atez + Beva in regard to prolonging their OS. Specifically, TACE + Lenv may be preferred over Atez + Beva in patients with intrahepatic tumor status suitable for TACE, such as a tumor burden of less than 50% and absence of main bile duct invasion, or in patients with autoimmune diseases or transplanted organs that limit the use of immunotherapies. Since the TACE + Lenv trial was conducted in China, its application is more likely to be adopted in Asians until its efficacy is confirmed in Western patients.

Our findings should be interpreted with consideration of the inherent limitations of NMA, with the results being derived from indirect comparisons between RCTs. To mitigate potential bias from cross-trial heterogeneity, we included only RCTs with a common comparator, Sora or Lenv, to ensure transitivity. However, variations in patient and tumor characteristics, including HCC etiology and geographical regions of enrollment, may limit the direct applicability of these results in clinical practice. In addition, we were unable to analyze post-progression survival and IO sequencing patterns due to insufficient reporting across included studies, highlighting the need for standardized reporting of these outcomes in future research [59]. Despite these acknowledged limitations, this study delivers a comprehensive analysis of the potential therapeutic options available, including combinations with transarterial interventions, in comparison with the current standard of care, thereby supporting clinical decision-making in heterogeneous populations with unresectable HCC.

In conclusion, our updated NMA suggests that in patients with advanced HCC, IO-based combination regimens, especially Sin + IBI305, Cam + Rivo, and Atez + Beva, can significantly improve survival outcomes compared with the first-line TKIs Sora and Lenv, irrespective of tumor status and etiology. For selected patients with PVTT and/or EHM, TACE + Lenv presents a promising therapeutic option and could serve as a viable alternative to Atez + Beva or Trem + Durv, especially if a patient has an adequate tumor profile for TACE or is susceptible to immunotherapy-related toxicities. Given these considerations, direct head-to-head comparisons of treatments with demonstrated efficacy, particularly among the various IO-based combination regimens sharing wide indications, are necessary to provide evidence-based practice guidance and to enable precision oncologic care in the front-line settings of advanced-stage HCC.

This trial-level network meta-analysis was approved by the Institutional Review Board of Asan Medical Center (Institutional Review Board No. 2024-0684), which waived the requirement for informed consent from individual patients due to the nature of the study.

The authors declare no competing interests.

This study was supported by grants from the National Research Foundation of Korea funded by the Ministry of Science and ICT (NRF-2022R1A2C3008956, NRF-2021R1A6A1A03040260, and RS-2022-00166674), Asan Institute for Life Sciences (Grant No. 2022IP0046), and the Elimination of Cancer Project Fund from the Asan Cancer Institute of Asan Medical Center. The grant source was not involved in study design, the collection, analysis, and interpretation of data, the writing of the report, and the decision to submit the paper for publication.

Study concept and design and data acquisition, analysis, and interpretation: Ye Rim Kim, Euichang Kim, Ha Il Kim, Jihyun An, and Ju Hyun Shim. Manuscript draft and verification of the underlying data: Ye Rim Kim, Euichang Kim, Ha Il Kim, Jihyun An, Ju Hyun Shim, and Seungbong Han. Critical revision of the manuscript for important intellectual content: Ye Rim Kim, Jihyun An, and Ju Hyun Shim. Statistical analysis: Seungbong Han.

Ye Rim Kim, Euichang Kim, and Ha Il Kim contributed equally to this work.

All data generated or analyzed during this study are included in this article and its online supplementary material files. Further inquiries can be directed to the corresponding author.