This study was conducted following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines and was based on a registered protocol (PROSPERO: CRD42023386780 from 14/01/2023). Institutional review board (IRB) approval was unnecessary for this type of study. The inclusion criteria for this systematic review and meta-analysis were RCTs involving patients with advanced HCC treated with HAIC, either with or without systemic chemotherapy. Trials in which the patients received HAIC as adjuvant therapy following surgical resection or in combination with RFA or TACE were excluded.

The primary outcomes of this study were overall survival (OS) and progression-free survival (PFS). The secondary outcomes included adverse events (AEs), objective response rate (ORR), and disease control rate (DCR). ORR was defined as the sum of complete responses (CRs) and partial responses (PRs), and DCR was defined as the sum of CRs, PRs, and stable diseases. In all the included RCTs (10–15, 19), the modified Response Evaluation Criteria in Solid Tumors for HCC (mRECIST) was used to assess the treatment responses (20). AEs were assessed with the National Cancer Institute’s Common Terminology Criteria for Adverse Events (CTCAE).

An experienced information specialist conducted the electric searches of multiple databases, including Medline via Ovid, Embase via Elsevier, the Cochrane Central Register of Controlled Trials via Wiley, Web of Science, ClinicalTrials.gov, WHO International Clinical Trials Registry Platform (ICTRP), Koreamed, and Kmbase. The search was performed from the inception of the databases to December 2022 without any language or publication status restrictions.

Two review authors (HJK, SHL) used the Covidence software platform (www.covidence.org) to assess all potentially relevant records and select eligible studies. Any discrepancies in the assessment of eligibility were resolved through discussions with a third member of the review team. After the eligible studies were selected, two review authors independently extracted data from each study in duplicate. The extracted data included study design, duration, setting, country, sample size, patient characteristics, interventions, clinical outcomes such as OS, PFS, DCR, and OCR, and adverse events. A PRISMA flow diagram is presented to illustrate the study selection process. The Cochrane Risk of bias tool was used to evaluate the risk of bias in each study (21). Each bias item was classified as ‘low risk,’ ‘high risk,’ or ‘unclear’.

Dichotomous data such as ORR, DCR, and adverse events are presented as risk ratios (RRs) with 95% confidence intervals (CIs). Time-to-events data, including OS and PFS, are expressed as hazard ratios (HRs) with 95% CIs. Heterogeneity (inconsistency) was identified by the I ² statistics and interpreted according to the Cochrane Handbook (22). A visual inspection with Forest plots was used to assess the overlap of CIs. When any heterogeneities were detected, it was attempted to determine the possible reasons by evaluating the characteristics of each study and subgroup. A random-effects model was employed to summarize the data and interpret the results. The certainty of evidence (CoE) was rated on an outcome-specific basis using the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) approach (23).

To assess the overall quality of evidence, the GRADE framework was used. This framework considers criteria pertaining to validities, such as the risk of bias, imprecision, inconsistency, publication bias, and indirectness of results (23). Each author independently assigned a rating of high, moderate, low, or very low to the quality of evidence for each outcome in every comparison. Any inconsistencies were addressed through a consensus or, if necessary, by engaging other review authors for arbitration using GRADEpro (24). Subsequently, “Summary of findings” tables were generated that presented crucial information about the number of participants and studies, the certainty of the evidence, the estimated relative effects, and the anticipated absolute effects of each treatment strategy for each clinical outcome (25, 26). We used the GRADE guidance to describe the certainty of the evidence and the magnitude of the effect size (27).

We planned to conduct subgroup analyses of the Child–Pugh score, level of tumor markers, and portal vein thrombosis and described them in the protocol. Additionally, subgroup analyses based on chemotherapy regimen were added to compare possible differences in the outcomes of each chemotherapeutic used for HAIC.

Figure 1 is the PRISMA flow diagram that presents the process of selecting eligible studies. A total of 1,076 records from databases and 214 records from registers were identified. After removing duplicates, 916 records were screened by the review authors. One record was excluded by automation tools, and 915 reports were sought for retrieval. Among them, 887 reports were not retrieved because of animal studies or exclusion criteria. Twenty-eight reports were assessed for eligibility. Five reports were excluded because they were ongoing studies and five other reports were excluded due to a wrong comparator (n = 2), wrong study design (n = 2), or wrong measured outcomes (n = 1). Finally, seven RCTs were included in this review (10–15, 19).

Among the seven RCTs included in this study, three studies were conducted in Japan (10, 15, 19), three in China (12–14), and one in South Korea (11). Five trials were multicenter trials (10, 11, 13, 15, 19), and the remaining two were single-center trials (12, 14). The included studies were performed between 2010 and 2020, and their publication dates ranged from 2016 to 2022. The number of eligible participants ranged from 58 to 262 individuals. Among the included RCTs, five trials compared HAIC + sorafenib and sorafenib alone, encompassing a total of 695 patients (10, 13–15, 19). The remaining two trials compared HAIC and sorafenib, with a total of 320 patients included (11, 12). In all the studies, the diagnosis of HCC was confirmed either histologically or clinically, according to the American Association for the Study of Liver disease criteria (2). All eligible patients were adults who were not suitable for surgery or locoregional treatment such as ablation or TACE (10–15, 19). Four trials included patients with ECOG PS 0–1 (10, 11, 15, 19), while the remaining three included patients with ECOG PS 0–2 (12–14). Two trials included patients with Child-Pugh grade A (Child-Pugh score 5 or 6) (13, 14), whereas the other five trials also included Child-Pugh score 7 (10–12, 15, 19). In all the studies, sorafenib was initiated at a dose of 400 mg twice daily (10–15, 19). However, there were variations in the HAIC regimens applied. Ikeda et al. and Kondo et al. utilized a cisplatin monotherapy as a HAIC regimen (15, 19), while Kudo et al. and Choi et al. used a combination of cisplatin and fluorouracil (10, 11). He et al., Zheng et al., and Lyu et al. employed a combination of oxaliplatin, leucovorin, and fluorouracil (12–14). All of the studies allowed for the participation of patients with extrahepatic metastasis as long as it was determined that the metastatic lesions would not have an influence on their prognosis (10–15, 19). One study only enrolled patients with portal vein invasion (13), while two studies exclusively included patients with major portal vein tumor thrombosis (Vp3–4) (11, 14). Although two studies initially used RECIST1.1 and conducted post hoc analyses with mRECIST (13, 14), all the included studies assessed tumor response using mRECIST. Six studies reported funding sources; five studies were supported by government agencies (10, 12–14, 19), and one study was funded by a non-profit organization (15). The other study did not specify its funding source (11). Five studies reported no conflicts of interest (11–14, 19), while two provided details of their conflicts of interest (10, 15). The characteristics of the included studies are summarized in Supplement Table 1 .

The risks of bias in the included studies are summarized and presented in Figure 2 . Most of the included studies had a low risk of bias across numerous domains; however, all the included studies were considered to have an overall high risk of performance bias and detection bias for the subjective outcomes. In the HAIC groups, patients had to undergo arterial catheterization, while the sorafenib groups did not. As a result, blinding of participants was inevitably impossible.

Seven RCTs comprising 1,010 patients that compared sorafenib monotherapy with either HAIC + sorafenib or HAIC alone were identified (10–15, 19). The findings of this meta-analysis study suggest that HAIC, with or without sorafenib, may improve OS, PFS, and response rates in advanced HCC patients without significant differences in adverse events.

Most of the included RCTs demonstrated the survival benefits of HAIC treatments, while the studies conducted by Kondo et al. and Kudo et al. did not show such benefits (10, 15). However, it is imperative to approach the interpretation of their results with caution. Both trials utilized cisplatin monotherapy or low-dose cisplatin with fluorouracil. In the study by Kondo et al., the concept of “Clinical PD” was introduced based on the levels of alpha-fetoprotein (AFP) and des-gamma carboxyprothrombin (DCP). However, the authors interpreted that clinical PD is unlikely to be disadvantageous. Furthermore, in the sorafenib group, more than half of the patients who received subsequent treatments underwent HAIC, suggesting a potential crossover effect that should not be overlooked (15). In the study by Kudo et al., the calculated sample size for each group was 95 patients, and 102 patients were allocated to the HAIC + sorafenib group. However, 14 patients did not receive treatment, resulting in an underpowered sample size (10).

The initial expectation was that the combination of sorafenib and HAIC would yield better results compared to HAIC alone. Overall, HAIC plus sorafenib showed favorable OS, PFS, and ORRs compared with sorafenib alone; however, those of the HAIC alone groups were numerically better ( Tables 1 , 2 ). Due to significant heterogeneities observed in the analyses of HAIC plus sorafenib, conducting subgroup analyses based on the HAIC regimen provided valuable insights into the underlying reasons ( Supplement Figures 1, 3 ). Among the five RCTs that used HAIC plus sorafenib, three utilized cisplatin-based chemotherapy (10, 15, 19), while the other two employed oxaliplatin-based chemotherapy (13, 14). As depicted in Supplement Figures 1 and 3 , the subgroup receiving the oxaliplatin-based treatment demonstrated significantly better survival outcomes compared with the subgroup receiving the cisplatin-based treatment. The two RCTs investigating HAIC alone employed a higher dose of cisplatin (60 mg/m²/cycle) or oxaliplatin in combination with fluorouracil (11, 12). Additionally, notable differences were observed in the intervals between each HAIC cycle. In the subgroup receiving the cisplatin-based chemotherapy, the intervals ranged from 4 to 6 weeks (10, 15, 19), which were longer than the intervals in the other studies, which ranged from 3 to 4 weeks (11–14).

In all the included RCTs, patients treated with HAIC demonstrated higher response rates than patients receiving sorafenib (range, HAIC with or without sorafenib vs. sorafenib, 17.1–54.8% vs. 3.1–18.0%). Particularly, He et al. (13). reported that a significantly higher number of patients in the HAIC plus sorafenib group proceeded to curative surgery compared with the sorafenib group (HAIC + sorafenib vs. sorafenib, 16 [12.8%] vs. 1 [0.8%], P < 0.001). Similar results were reported by Lyu et al. (12). These findings suggest that HAIC may contribute to downstaging and enabling a switch to curative surgery, thereby improving survival outcomes. In the phase III trial of sorafenib conducted in Western countries, the ORR of sorafenib was 2% (7 of 299 patients), and in the Asia-Pacific region, the ORR was 3.3% (5 of 150 patients) (28, 29). In the updated report of the IMbrave150 study, the ORR based on mRECIST for atezolizumab plus bevacizumab was 30% (97 of 326 patients), while the ORR for sorafenib was 11% (18 of 159 patients) (30). Considering the relatively high ORR observed with HAIC and the possibility of crossover to curative resection or locoregional therapy, HAIC may have a crucial role in the treatment of HCC, even in the era of immune checkpoint inhibitors.

Major vascular invasion or portal vein thrombosis are significant adverse prognostic factors for HCC (3). In the subgroup analyses according to the portal vein invasion and thrombosis (PVTT), HAIC plus sorafenib showed a trend toward improved OS and PFS compared to sorafenib alone, especially in patients with PVTT. However, it did not reach statistical significance.

There have been several published systematic reviews on this topic that have presented the positive effect of HAIC on HCC treatments. Long et al. recently reported that sorafenib plus HAIC showed significantly better OS (HR 0.56 [95% CI 0.37–0.83]; P < 0.01), PFS (HR 0.44 [95% CI 0.27–0.72]; P < 0.01), and ORR (RR 3.77 [95% CI 1.87–7.58]; P < 0.01) than sorafenib alone (31). Zhang et al. also conducted a recent systematic review, comparing HAIC to sorafenib in advanced HCC with PVTT, and reported significant improvements in OS (HR 0.50, 95% CI 0.40–0.63, P < 0.05), PFS (HR = 0.49, 95% CI 0.35–0.67, P < 0.05), and ORR (RR 4.21, 95% CI 2.44–7.28, P < 0.000001) (32). Additionally, this study demonstrated the benefits of HAIC, whether with or without sorafenib.

However, prior reported systematic reviews have exhibited some methodologic errors. Two systematic reviews were published in 2019 and 2022. Although the authors assessed the quality of the studies, they did not evaluate the risk of bias in the included studies (16, 33). Zhang et al. evaluated the quality of the cohort studies using the Newcastle Ottawa scale and conducted a risk assessment according to the Cochrane Collaboration Network recommendations (32). However, they also neglected to include the risk of bias assessment. Long et al. used the GRADE method and presented the risk of bias but did not incorporate this into the interpretation of the review results (31). However, the most considerable error is that all the aforementioned systematic reviews conducted meta-analyses by combining retrospective or observational studies with prospective RCTs. These methodologic errors can introduce flaws in the results.

The rigorous methodology is a strength of this review. This systematic review was conducted based on a prospectively registered protocol, and an experienced information specialist performed the comprehensive literature search. Unlike most of the previous reviews, this study exclusively included RCTs for the meta-analysis. Furthermore, this review is unique in that the GRADE method was adopted, incorporating a certainty of evidence rating and presenting the absolute effect sizes within a clinical context.

However, this study also has some limitations. The combination of atezolizumab with bevacizumab is currently the first choice for advanced HCC, and other targeted agents, such as Lenvatinib, are also an acceptable alternative treatment for advanced HCC. However, since there were no RCTs comparing HAIC to these systemic therapies, we were unable to investigate the relative impacts of HAIC compared to the recently updated treatment for HCC. All of the included RCTs were performed in Asian countries, where the main cause of HCC is hepatitis B virus (HBV) infections. In contrast, patients with hepatitis C virus (HCV) infections, which are the primary cause of HCC in Western countries, comprised a small portion of this study. Therefore, it is necessary to conduct more RCTs in Western countries to obtain a comprehensive understanding of the effectiveness of HAIC for HCC treatment across different populations and etiologies. The small sample size of RCTs was another limitation of this study. Since the current body of evidence is moderate to very low, more robust randomized trials are needed to confirm the efficacy of HAIC.

This study provides evidence supporting the use of HAIC to treat advanced HCC. Furthermore, this study suggests that HAIC with an oxaliplatin-based regimen may contribute to a higher survival rate than HAIC with a cisplatin-based regimen.

HAIC represents a potential alternative for advanced HCC treatment, offering advantages over systemic therapies. Combinations of cytotoxic chemotherapy and targeted agents have emerged as a major trend in anti-cancer treatments, but their application in HCC has been limited due to the low efficacy of systemic administration of cytotoxic chemotherapy. HAIC offers the possibility of combining therapies in HCC while minimizing systemic adverse events. Additionally, HAIC may serve as a bridging therapy or induction therapy for curative resection or locoregional treatments.