Interventional therapy combined with tyrosine kinase inhibitors with or without immune checkpoint inhibitors as initial treatment for hepatocellular carcinoma with portal vein tumor thrombosis: a systematic review and meta-analysis

We searched databases, including PubMed, Web of Science, EMBASE, Scopus, and the Cochrane Library database, for clinical studies comparing the triple therapy to the dual therapy for HCC-PVTT. The search terms included portal vein tumor thrombosis, interventional therapy, tyrosine kinase inhibitor, immune checkpoint inhibitor, etc. The search period is from the database built until April 24, 2024. Specific terms and keywords are shown in Supplementary Table 1.

Inclusion Criteria: (1) population: patients with clinical or pathological diagnosis of HCC-PVTT, and no previous relevant local or systemic anti-tumor treatments including interventional therapy, radiotherapy and systemic therapy, but recurrence after single surgery is accepted; (2) intervention: the triple therapy regimen in a combination of IT-TKI-ICI; types of interventional therapy include transarterial chemoembolization (TACE), hepatic arterial infusion chemotherapy (HAIC), portal vein stent and iodine-125 seed strand(PVS-I125), etc.; types of TKIs and ICIs are not restricted; (3) comparison: the dual therapy regimen with IT-TKI; (4) outcome: at least one major outcome indicator and we can directly or indirectly obtain effect measures; major outcome indicators include progression-free survival (PFS) and overall survival (OS); secondary outcome indicators include the number of complete response (CR), partial response (PR), stable disease (SD), progressive disease (PD), overall response rate (ORR), disease control rate (DCR), adverse events (AEs), and downstaging surgery rate (DSR). DSR refers to the proportion of HCC-PVTT patients who can successfully achieve tumor downstaging and undergo salvage surgery after the treatments.

Exclusion Criteria: (1) literature such as reviews, systematic reviews, conference abstracts, comments, letters, editorials, guidelines, animal experiments, etc.; (2) duplicate publications or literature without full-text access; (3) studies that included HCC with and without PVTT, but PVTT data from the two groups could not be separated; (4) literature that cannot directly or indirectly extract outcome indicators.

After the initial search, two authors (Changjie Du and Jiajun Yuan) independently screened the titles and abstracts of the articles to identify potentially relevant studies. Subsequently, the full-text articles were independently screened and reviewed by the former authors based on the inclusion and exclusion criteria, and literature data as well as quality were extracted. Disagreements were resolved through discussion with a third author (Hongyu Wu) to reach a consensus. Data extraction included the following: author, country, publication date, study design type, basic characteristics of patients (gender, age, liver function, overall condition, etc.), tumor characteristics (tumor size, number of tumors, PVTT classification, distant metastasis), treatment regimens, outcome indicators (PFS, OS, CR, PR, SD, PD, AEs, DSR), etc. For the randomized controlled trial (RCT) data, the Cochrane collaboration tool was used to assess the risk of bias. For non-randomised cohort studies, the Risk Of Bias In Non-randomised Studies—of Interventions (ROBINS-I) tool²⁵ and the Newcastle–Ottawa Scale (NOS)²⁶ were used to evaluate the quality of included studies. For the results of NOS, a score of 7–9 was considered high-quality research, 4–6 was considered medium-quality research, and less than 4 was considered low-quality research. We included studies of medium to high quality and excluded low-quality studies.

Data analysis of outcome indicators in the included studies was performed using RevMan 5.4 software. For meta-analysis of tumor response, DSR, and AEs, risk ratios (RRs) were our preferred outcome measure. For meta-analysis of PFS and OS, we preferred hazard ratios (HRs) and mean difference (MD), because HRs can provide time-to-event information and MD can quantify the time of survival. When HRs were not directly available, We would contacted the corresponding authors for them, or we performed secondary data analysis using Kaplan–Meier curves, P values, and median OS and PFS values indirectly^27–29. Cochrane's Q-test and I² statistics were used to assess heterogeneity³⁰. If the P value of Cochrane's Q-test was less than 0.01 and I² statistics were greater than 50%, indicating substantial heterogeneity, a random-effects model would be chosen³¹. Otherwise, heterogeneity was considered acceptable, and a fixed-effects model would be used. Publication bias was assessed using a funnel plot and Egger's test³². And the sensitivity analysis was assessed through the meta-analysis ignoring each study in turn³³. All positive outcomes were re-evaluated using trial sequential analysis (TSA) to further ensure outcome stability and reduce the possibility of false positive results. TSA can provide a threshold for a statistically significant treatment effect, if the cumulative test statistic curve (Z-curve) intersects with the the TSA adjusted significance threshold, it is considered to have a statistical significance and a stable result³⁴. If not, it is considered that false positive results may exist, and an extra study population size will be provided to achieve statistical significance.Trial sequential analysis (TSA) was performed with the TSA software (TSA-0.9.5.10-Beta).

In the initial retrieval strategy, a total of 332 relevant studies have been identified. Following screening based on inclusion and exclusion criteria, as well as quality assessment, we selected 6 retrospective cohort studies for quantitative analysis^{15,18,20–22,35}. Moreover, the ROBINS-I tool and the NOS were used to assess the quality of the studies, there was no serious or critical risk of bias observed in ROBINS-I tool (Supplementary Table 2) and all of which were rated as high quality with the NOS (Supplementary Table 3). No studies were excluded from the analysis after quality assessment and ultimately those studies were selected for our meta-analysis. The detailed process is illustrated in Fig. 1.

Six retrospective studies reported on the comparison between the triple therapy and the dual therapy for patients with HCC-PVTT. As some literature underwent propensity score matching (PSM), we opted for the inclusion of matched cohort data to reduce intergroup differences. A total of 568 patients were included, with 259 (45.6%) receiving the triple therapy and 309 (54.40%) receiving the dual therapy. The included literature was formally published in 2023 and 2024, and the study populations were all Chinese, with sample sizes ranging from 18 to 90 patients. The average age of patients ranged from 47.9 to 60 years, with the majority having a background of hepatitis B, consistent with the Chinese context. All patients had acceptable overall and liver function, theoretically tolerating combined therapy. Detailed characteristics are summarized in Table 1.

We collected 10 measurable outcomes, divided into 3 categories to compare the efficacy and safety of the triple therapy and the dual therapy.

We stratified the analysis based on PVTT and assessed whether triple therapy had survival benefits with main trunk PVTT (MPVTT).

Two studies reported the efficacy of triple therapy in MPVTT, while other studies did not provide subgroup reports based on PVTT subtyping. No substantial heterogeneity was found in Q-test and I² statistics, and fixed effect model was adopted. Compared to the dual therapy group, MPVTT patients in the triple therapy group exhibited significantly increased OS (HR = 0.65, 95% CI 0.49, 0.85, P = 0.002; MD = 6.07 months, 95% CI 3.45, 8.69, P < 0.001) and PFS (HR = 0.51, 95% CI 0.35, 0.74, P = 0.0004; MD = 3.16 months, 95% CI 0.84, 5.48, P = 0.008). Detailed results are depicted in Fig. 4A, B, C, D.