Cost-effectiveness analysis of seven treatments vs sorafenib as first-line therapy for advanced hepatocellular carcinoma in China

Abstract

BACKGROUND

Hepatocellular carcinoma (HCC) is the most common type of primary liver cancer and represents a significant health burden in China, where selecting cost-effective first-line therapies for advanced disease remains critical.

AIM

To evaluate the cost-effectiveness of seven high-efficacy regimens for advanced HCC from the perspective of the Chinese healthcare system.

METHODS

A systematic network meta-analysis was conducted on 23 first-line treatment regimens from 22 phase III randomized controlled trials. Seven high-efficacy regimens were then subjected to cost-effectiveness evaluation using a Markov model. Hazard ratios for progression-free survival and overall survival specific to Chinese subgroups were preferentially employed to derive survival parameters; when unavailable, Asian subgroup data were substituted to maintain population relevance. The analysis incorporated survival data, adverse event profiles, and direct medical costs.

RESULTS

The network meta-analysis ranked lenvatinib plus transarterial chemoembolization (L + T) highest for both progression-free survival and overall survival [surface under the cumulative ranking curve: (1) 100%; and (2) 99.4%, respectively]. The cost-effectiveness analysis revealed that L + T [incremental cost-effectiveness ratio = $37753.45 per quality-adjusted life year (QALY)] and rivoceranib plus camrelizumab (incremental cost-effectiveness ratio = $37198.84/QALY) were cost-effective options, both below the willingness-to-pay threshold of $37669/QALY. L + T showed the highest probability (67.9%) of being the optimal treatment option.

CONCLUSION

L + T emerges as the optimal first-line therapy for advanced HCC in China, offering an effective balance between clinical outcomes and cost-effectiveness.

Key Words: Hepatocellular carcinoma; Cost-effectiveness; Network meta-analysis; First-line treatment; China

Core Tip: The economic evaluation of various first-line treatment regimens for advanced hepatocellular carcinoma is crucial, yet such analyses remain scarce. This study conducted a network meta-analysis of 23 first-line regimens for advanced hepatocellular carcinoma and subsequently performed a cost-effectiveness analysis on seven highly efficacious regimens from the perspective of the Chinese healthcare system. The results indicated that lenvatinib combined with transarterial chemoembolization emerged as the most favorable first-line treatment under current economic conditions in China, achieving a 67.9% probability of being the most cost-effective option. This study contributes a novel perspective to inform clinical decision-making for healthcare professionals.

INTRODUCTION

Hepatocellular carcinoma (HCC) is the most common form of primary liver cancer and represents a significant global health challenge, causing hundreds of thousands of new cases and deaths each year[1]. Epidemiological evidence consistently shows that Asia – particular China – bears a disproportionate share of this burden, driven by a combination of chronic hepatitis B virus infection and increasing prevalence of metabolic risk factors in urban populations[1,2]. The prognosis for HCC remains poor, largely because most patients are diagnosed at advanced stages, greatly limiting opportunities for curative treatment[3-6]. Although these developments have expanded therapeutic choices, they have also made clinical decision-making more complex. Existing clinical evaluations often emphasize survival outcomes while overlooking cost considerations[7,8]. This gap highlights the urgent need for systematic cost-effectiveness analyses to guide treatment selection that is both clinically effective and economically viable.

Network meta-analysis (NMA) has become the gold standard for comparing multiple therapeutic regimens, yet most studies have focused primarily on survival endpoints like progression-free survival (PFS) and overall survival (OS)[6,9,10]. For instance, a recent NMA by Fulgenzi et al[6] reported that sintilimab plus a bevacizumab biosimilar IBI308 (S + I) and rivoceranib plus camrelizumab (R + C) offered the greatest OS benefit, whereas pembrolizumab plus lenvatinib provided the best PFS outcome. However, survival outcomes alone do not fully capture the clinical and economic implications of different treatment strategies. With the growing emphasis on value-based oncology, cost-effectiveness analyses are essential for identifying therapies that are both clinically beneficial and financially sustainable[11-14].

MATERIALS AND METHODS

Ethical considerations

This NMA was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines[15]. Informed consent and Institutional Review Board approval were waived, as all data were obtained from publicly available sources.

Data sources and search strategy

A comprehensive literature search was performed in PubMed, EMBASE, the Cochrane Library, and Web of Science up to October 30, 2024, without applying a lower date limit to ensure inclusion of all relevant studies. Only English-language publications were considered. The key search terms were: (1) Advanced; (2) Hepatocellular carcinoma; (3) First-line therapy; and (4) Survival. The detailed search strategy is presented in Supplementary Table 1. Abstracts and posters from major oncology meetings, including the Chinese Society of Clinical Oncology, European Society for Medical Oncology, Asian Pacific Association for the Study of the Liver, and American Society of Clinical Oncology, were also reviewed.

Studies were included if they met the following criteria: (1) Randomized phase III trials; (2) Advanced or metastatic HCC patients unsuitable for curative surgical resection; (3) First-line therapy; (4) Head-to-head comparisons with placebo, sorafenib, or lenvatinib as controls; and (5) Reporting of OS, PFS, and their hazard ratios (HRs) with 95%CI. Exclusion criteria comprised: (1) Pilot studies, post hoc analyses, real-world analyses, case reports, protocols, reviews, meta-analyses, retrospective studies, and cost-effectiveness analyses; (2) Studies limited solely to patients defined by the presence or absence of portal vein tumor thrombosis; (3) Duplicate versions of the same trial; and (4) Exploratory analyses of randomized controlled trials (RCTs). Two independent investigators screened the titles, abstracts, and full texts to determine eligibility.

Data extraction and quality assessment

Two independent investigators extracted data into a predefined spreadsheet, and discrepancies were resolved by discussion. Extracted items included basic study characteristics (publication year, trial name, authors, geographic region, sample size, and treatment regimens in the experimental and control arms) and outcomes of interest [HRs for OS and PFS, and incidence rates of adverse events (AEs) of any grade and grade ≥ 3].

The quality of each RCT was assessed using the Cochrane Risk of Bias Tool, and the level of bias was rated as low, high, or unclear for each domain. Publication bias was evaluated visually using comparison-adjusted funnel plots and quantitatively using Egger’s regression test, with P < 0.05 was considered statistically significant. Any disagreements were resolved through consultation with a third reviewer.

Statistical integration and validation

The NMA was conducted within a Bayesian framework using JAGS and the gemtc package in R (version 4.0.3). Fixed-effects and random-effects models were fitted and compared using the Deviance Information Criterion (DIC); a fixed-effect model was adopted if the DIC difference was less than 5, otherwise, the model with the lower DIC value was selected[16]. Markov Chain Monte Carlo simulations were performed with 5000 burn-in iterations followed by 20000 sampling iterations. Treatment rankings were obtained by calculating the surface under the cumulative ranking curve (SUCRA) values, ranging from 0 to 1, with higher values indicating superior efficacy or a more favorable AE profile. Pairwise comparisons were illustrated as forest plots.

Cost-effectiveness analysis

Survival analysis and parameter estimation: Seven treatment regimens with the highest mean SUCRA values for PFS and OS were selected for cost-effectiveness evaluation from the perspective of the Chinese healthcare system. Sorafenib served as the reference treatment. Its survival curves were digitized from the original trial data using Graph Digitizer (version 2.24) and pooled with the MetaSurv package. The pooled survival curves were fitted with various distributions – including exponential, gamma, generalized-gamma, Gompertz, Weibull, Weibull proportional hazards, log-logistic, and log-normal – using the SurvHE package. The model with the lowest Akaike Information Criterion was considered the best fit.

Progression and death probabilities for each of the seven regimens were derived by multiplying the corresponding probabilities for sorafenib by the HRs (treatment vs sorafenib) obtained specifically from the Chinese subgroup of the NMA[17]. When Chinese subgroup data were unavailable, data from the Asian subgroup were substituted. Survival distribution parameters for each regimen were then determined. To verify the validity of this substitution, meta-regression and difference test were performed to assess potential heterogeneity between the Asian-substituted subgroup and the Chinese subgroup for both PFS and OS endpoints.

Cost and utility estimates: Costs were estimated from the perspective of the Chinese healthcare system, considering only direct medical expenses, including medications, routine follow-up tests (laboratory and imaging), inpatient care, management of grade ≥ 3 AEs with an incidence greater than 2%, best supportive care, and hospice care.

Drug prices were sourced from the most recent local public bid-winning prices as of October 30, 2024, and patient assistance programs were factored into the analysis when applicable. Costs for transarterial chemoembolization (TACE) and AEs were estimated according to the China Diagnosis-Related Group system for 2024. Specifically, the cost of AEs for specific regimen was calculated as: Total AE cost = sum (incident rate × expenditure).

where incident rate represents the estimated incidence of each AE, and expenditure denotes the management cost associated with the corresponding Diagnosis-Related Group item.

All costs were expressed in United States dollars, converted at an exchange rate of $1 = ¥7.12 (October 2024). Health utility values for the progression-free and progressive disease states were set at 0.76 and 0.63, respectively, based on previous consensus studies[18]. Disutilities for grade ≥ 3 AEs with an incidence > 2%, represented as negative “utility decrements”, were also incorporated.

The following assumptions were made: (1) Drug dosages and administration schedules followed the respective clinical trial protocols; (2) The average patient weight was 65 kg; (3) Second-line treatment costs after progression were assumed equivalent, based on regorafenib use as the standard second-line therapy for HCC; and (4) AE profiles were considered comparable between the overall and the Chinese populations.

Model construction and cost-effectiveness analysis: A partitioned survival model was developed using TreeAge Pro 2022, integrating the survival, cost, and utility parameters. The model employed a 10-year time horizon with annual cycles, and applied a 5% discount rate was applied to both costs and health outcomes. The primary outcomes were total cost, quality-adjusted life years (QALYs), and the incremental cost-effectiveness ratio (ICER), calculated as: (Cost_Treatment - Cost_Sorafenib)/(QALY_Treatment - QALY_Sorafenib).

A willingness-to-pay (WTP) threshold of three times China’s 2023 per-capita GDP ($37669 per QALY gained) was used to determine cost-effectiveness. Treatments with ICERs below this threshold were considered cost-effective, while those yielding higher QALYs at lower costs were deemed dominant; regimens with lower QALYs at higher costs were regarded as inferior.

Statistical analysis

One-way sensitivity analyses and probabilistic sensitivity analyses (PSA) were conducted to assess the robustness of model outcomes and address parameter uncertainty.

In the one-way sensitivity analysis, HRs for PFS and OS were varied according to their 95%CI from the NMA, the discount rate was varied between 0% and 8%, and other parameters by ± 20%. Results were presented as tornado diagrams.

In the PSA, model parameters were assigned pre-specified statistical distributions, and a Monte Carlo simulation with 1000 iterations was performed. The results were presented as scatter plots and cost-effectiveness acceptability curves, illustrating the probability of each regimen being cost-effective relative to sorafenib at the WTP threshold of $37669/QALY, as well as its likelihood of being the most cost-effective option across varying WTP thresholds.

RESULTS

NMA

Search results and study characteristics: The study selection process is outlined in Figure 1, which presents a flowchart detailing each step. The initial literature search identified 2088 articles. After removing 238 duplicates, 1850 records remained. Screening of title and abstract excluded 1777 records, leaving 73 studies for full-text review. Ultimately, 22 RCTs comprising 13965 patients met the inclusion criteria for the NMA[19-40]. Additionally, two conference abstracts containing survival data for the Chinese subgroup were identified through a manual search to facilitate subgroup analysis.

Figure 1 Flowchart of the literature search and selection process. The REFLECT and LEAP-002 trials reported Chinese subgroup data at Chinese Society of Clinical Oncology 2017 and Asian Pacific Association for the Study of the Liver 2024, respectively. HCC: Hepatocellular carcinoma; RCT: Randomized controlled trial; WoS: Web of Science.

The key characteristics of the included trials are summarized in Supplementary Table 2[19-40]. The risk-of-bias assessment, presented in Supplementary Figure 1[19-40], revealed that several studies had a high risk of bias in the “blinding of participants and personnel” domain due to open-label designs. Three studies exhibited unclear risk in the “blinding of outcome assessment” domain, while the remaining domains were assessed as low risk overall. Comparison-adjusted funnel plots for PFS, OS, and AEs did not show substantial asymmetry (Supplementary Figure 2), suggesting minimal publication bias. The P values for Egger’s test were 0.413 (PFS), 0.827 (OS), 0.162 (AEs), and 0.769 (AEs ≥ grade 3), further supporting the absence of significant publication bias.

PFS and OS: PFS and OS are critical clinical endpoints in HCC that guide treatment decisions. A total of 22 studies evaluating 23 treatment regimens were included in the analysis, with network structures shown in Supplementary Figure 3A and B. Based on DIC values, fixed-effects models were applied (Supplementary Table 3).

The SUCRA values and rankings of all 23 treatment regimens are summarized in Table 1. For PFS, lenvatinib plus TACE (L + T) exhibited the highest efficacy (SUCRA: 100%), with a 99.9% likelihood of being the most effective regimen. Other high-ranking regimens included R + C, lenvatinib plus pembrolizumab (L + P), and S + I (Table 1). Pairwise comparisons (Supplementary Figure 4) showed that all treatment regimens significantly prolonged PFS compared to placebo (HR < 1, P < 0.05), with L + T achieving the lowest HR (HR: 0.159, 95%CI: 0.113-0.224). For OS, L + T again ranked highest (SUCRA: 99.4%) with an 89.9% likelihood of providing the greatest survival benefit, followed by S + I and R + C (Table 1). All regimens except sunitinib significantly improved OS over placebo, with L + T achieving the lowest HR (HR: 0.285; 95%CI: 0.193-0.421; Supplementary Figure 5).

Table 1 Bayesian surface under the cumulative ranking curve and ranking profiles for efficacy endpoints (progression-free survival and overall survival).

Treatment	PFS			OS			Mean (PFS and OS)
	SUCRA (%)	Rank	PrBest (%)	SUCRA (%)	Rank	PrBest (%)	SUCRA (%)	Rank
Placebo	0	23	0	1	22.8	0	0.50	22.90
Sorafenib	28.2	16.8	0	28.5	16.7	0	28.35	16.75
Lenvatinib	75.5	6.4	0	43.5	13.4	0	59.50	9.90
Donafenib	38.4	14.6	0	59.1	10	0	48.75	12.30
Sintilimab plus IBI305	85.7	4.2	0	92.3	2.7	6.8	89.00	3.45
Rivoceranib plus camrelizumab	89.7	3.3	0.1	88.9	3.5	2.2	89.30	3.40
Tislelizumab	15.3	19.6	0	57.4	10.4	0	36.35	15.00
Avastin plus tecentriq	74	6.7	0	85.2	4.3	0.9	79.60	5.50
Cabozantinib plus atezolizumab	75.4	6.4	0	48.6	12.3	0	62.00	9.35
Single Tremelimumab Regular Interval Durvalumab	44.4	13.2	0	71	7.4	0	57.70	10.30
Durvalumab	25	17.5	0	55.5	10.8	0	40.25	14.15
Lenvatinib plus pembrolizumab	89.1	3.4	0	70.1	7.6	0	79.60	5.50
Lenvatinib plus TACE	100	1	99.9	99.4	1.1	89.9	99.70	1.05
Sorafenib plus TACE	63.4	9	0	45.6	13	0	54.50	11.00
Nivolumab	38.7	14.5	0	57.4	10.4	0	48.05	12.45
Sorafenib plus erlotinib	12.2	20.3	0	42.1	13.7	0	27.15	17.00
Linifanib	60.9	9.6	0	22.4	18.1	0	41.65	13.85
Sunitinib	12.1	20.3	0	4.9	21.9	0	8.50	21.10
Brivanib	26.8	17.1	0	19.2	18.8	0	23.00	17.95
Sorafenib plus pravastatin	28.5	16.7	0	31.3	16.1	0	29.90	16.40
Sorafenib plus doxorubicin	37.9	14.7	0	24.3	17.7	0	31.10	16.20
Sorafenib plus hepatic arterial infusion chemotherapy	60.2	9.8	0	31.4	16.1	0	45.80	12.95
Avastin plus toripalimab	68.6	7.9	0	70.9	7.4	0.1	69.75	7.65

OS: Overall survival; PFS: Progression-free survival; PrBest: Probability of being best; SUCRA: Surface under the cumulative ranking curve; TACE: Transarterial chemoembolization.

Based on their efficacy profiles, the top eight regimens, ranked by mean SUCRA values for PFS and OS (Table 1), were as follows: (1) L + T (mean SUCRA: 99.7%; mean rank: 1.05); (2) R + C (mean SUCRA: 89.3%; mean rank: 3.40); (3) S + I (mean SUCRA: 89.0%; mean rank: 3.45); (4) L + P (mean SUCRA: 79.6%; mean rank: 5.50); (5) Avastin plus tecentriq (A + T) (mean SUCRA: 79.6%; mean rank: 5.50); (6) Avastin plus toripalimab (mean SUCRA: 69.7%; mean rank: 7.65); (7) Cabozantinib plus atezolizumab (C + A) (mean SUCRA: 62.0 %; mean rank: 9.35); and (8) Lenvatinib (mean SUCRA: 59.5%; mean rank: 9.90). These regimens were selected for cost-effectiveness analysis, except for C + A due to the commercial unavailability of cabozantinib in China.

HRs specific to Chinese or Asian subgroups compared with sorafenib were illustrated in Supplementary Figure 6. Including or excluding C + A resulted in consistently low and stable heterogeneity (I² for PFS: 12%-14%; I² for OS: 13%-15%). To evaluate potential bias from substituting Asian for Chinese subgroup data, we performed additional analyses. Meta-regression using population type (Asian-substituted vs Chinese) as a covariate yielded a coefficient of -0.11 (95%CI: -0.49 to 0.27) for PFS (P = 0.50) and -0.09 (95%CI: -0.56 to 0.38) for OS (P = 0.65), indicating negligible population bias. Similarly, Q tests showed no significant differences between the two population types (P = 0.37 for PFS and P = 0.55 for OS).

AEs: Treatment-related toxicity is a crucial factor in balancing efficacy and safety, as AEs affect patient adherence and quality of life. Fixed-effects models were used for AE analyses based on DIC values (Supplementary Table 3).

For any-grade AEs, 15 studies involving 16 treatment regimens were included (Supplementary Figure 3C). The ranking of treatment-related toxicity, from mildest to most severe (Table 2), indicated that durvalumab exhibited the lowest toxicity (SUCRA: 96.8%; rank: 1.5), followed closely by placebo (SUCRA: 96.5%; rank: 1.5), nivolumab (SUCRA: 85.9%; rank: 3.1), and tislelizumab (SUCRA: 80.7%; rank: 3.9). Single Tremelimumab Regular Interval Durvalumab ranked fifth (SUCRA: 70.9%; rank: 5.4), followed by A + T (SUCRA: 68.5%; rank: 5.7), donafenib (SUCRA: 53.3%; rank: 8.0), and lenvatinib (SUCRA: 46.2%; rank: 9.1). The highest toxicity was observed in R + C (SUCRA: 4.0%; rank: 15.4), followed by C + A (SUCRA: 11.1%; rank: 14.3) and linifanib (SUCRA: 16.3%; rank: 13.6).

Table 2 Bayesian surface under the cumulative ranking curve and ranking profiles for safety endpoints (adverse events).

Treatment	AEs of any grade			Grade ≥ 3 AEs
	SUCRA (%)	Rank	PrBest (%)	SUCRA (%)	Rank	PrBest (%)
Placebo	96.5	1.5	48.3	77.8	5.0	0
Sorafenib	32.9	11.1	0	55.9	8.9	0
Lenvatinib	46.2	9.1	0	28	14	0
Donafenib	53.3	8.0	0	74.8	5.5	0
Sintilimab plus IBI305	20.7	12.9	0	31.5	13.3	0
Rivoceranib plus camrelizumab	4	15.4	0	4.4	18.2	0
Tislelizumab	80.7	3.9	0	93.9	2.1	16.1
Avastin plus tecentriq	68.5	5.7	0	55.9	8.9	0
Cabozantinib plus atezolizumab	11.1	14.3	0	13.3	16.6	0
Single Tremelimumab Regular Interval Durvalumab	70.9	5.4	0	78.4	4.9	0
Durvalumab	96.8	1.5	51.7	98.6	1.3	79.5
Lenvatinib plus pembrolizumab	37.5	10.4	0	16.9	16.0	0
Lenvatinib plus TACE	NA	NA	NA	NA	NA	NA
Sorafenib plus TACE	NA	NA	NA	3.2	18.4	0
Nivolumab	85.9	3.1	0	90.8	2.7	4.5
Sorafenib plus erlotinib	35.2	10.7	0	47.8	10.4	0
Linifanib	16.3	13.6	0	29.9	13.6	0
Sunitinib	NA	NA	NA	35.1	12.7	0
Brivanib	43.6	9.5	0	48.6	10.2	0
Sorafenib plus pravastatin	NA	NA	NA	NA	NA	NA
Sorafenib plus doxorubicin	NA	NA	NA	NA	NA	NA
Sorafenib plus hepatic arterial infusion chemotherapy	NA	NA	NA	NA	NA	NA
Avastin plus toripalimab	NA	NA	NA	65	7.3	NA

AEs: Adverse events; NA: Not available; PrBest: Probability of being best; SUCRA: Surface under the cumulative ranking curve; TACE: Transarterial chemoembolization.

For grade ≥ 3 AEs, 18 studies analyzing 19 treatment regimens ranked toxicity from lowest to highest (Supplementary Figure 3D). Durvalumab again demonstrated the most favorable safety profile (SUCRA: 98.6%; rank: 1.3), followed by tislelizumab (SUCRA: 93.9%; rank: 2.1), nivolumab (SUCRA: 90.8%; rank: 2.7), Single Tremelimumab Regular Interval Durvalumab (SUCRA: 78.4%; rank: 4.9), placebo (SUCRA: 77.8%; rank: 5.0), and donafenib (SUCRA: 74.8%; rank: 5.5) (Table 2). Lower safety rankings included avastin plus toripalimab (SUCRA: 65.0%; rank: 7.3), sorafenib (SUCRA: 55.9%; Rank: 8.9), A + T (SUCRA: 55.9%; rank: 8.9), brivanib (SUCRA: 48.6%; rank: 10.2), and sorafenib + erlotinib (SUCRA: 47.8%; rank: 10.4). The highest toxicity was observed in L + P (SUCRA: 16.9%; rank: 16.0), C + A (SUCRA: 13.3%; rank: 16.6), R + C (SUCRA: 4.4%; rank: 18.2), and sorafenib plus TACE (SUCRA: 3.2%; rank: 18.4), indicating a substantial increase in severe AEs with these regimens.

Overall, immune checkpoint inhibitors such as durvalumab, nivolumab, and tislelizumab exhibited the most favorable safety profiles, with the lowest AE rates. In contrast, combination regimens like R + C, L + P, and sorafenib plus TACE demonstrated stronger efficacy but higher toxicity, necessitating careful clinical consideration.

Cost-effectiveness analysis

Base-case analysis: To evaluate the cost-effectiveness of different treatment regimens, a model incorporating three mutually exclusive health states – PFS, progressive disease, and death – was constructed (Figure 2). Seven treatment regimens were included in the analysis, and clinical input details are presented in Table 3[18,30,41-53]. Some variables were sourced from previous literatures[18,30,41-53]. Sorafenib was selected as the reference treatment, with survival curves pooled from the sorafenib arms of the RECIFIC, CARES-310, IMbrave-150, ORIENT-32, and HEPAORCH trials (Supplementary Figure 7). Based on model fitting, the log-normal distribution (μ = -1.079, σ = 0.855 for PFS; μ = 0.115, σ = 0.992 for OS) provided the best representation of survival data.

Figure 2 Partitioned survival model. HCC: Hepatocellular carcinoma; TACE: Transarterial chemoembolization.

Table 3 Model parameters.

Model variables	Base line value (range)		Minimum	Maximum	Distribution	Source
Survival distribution
PFS of sorafenib	μ = -1.079, σ = 0.855		Fixed	Fixed	Log-normal	Model fitting
OS of sorafenib	μ = 0.115, σ = 0.992		Fixed	Fixed	Log-normal	Model fitting
PFS of lenvatinib	μ = -0.515, σ = 1.100		μ = -0.218, σ = 1.219	μ = -0.785, σ = 0.990	Log-normal	NMA
OS of lenvatinib	μ = 0.407, σ = 1.075		μ = 0.689, σ = 1.171	μ = 0.144, σ = 1.003	Log-normal	NMA
PFS of camrelizumab plus rivoceranib	μ = -0.582, σ = 1.043		μ = -0.339, σ = 1.135	μ = -0.791, σ = 0.961	Log-normal	NMA
OS of camrelizumab plus rivoceranib	μ = 0.534, σ = 1.129		μ = 811, σ = 1.207	μ = 0.256, σ = 1.044	Log-normal	NMA
PFS of sintilimab plus IBI305	μ = -0.592, σ = 1.016		μ = -0.376, σ = 1.093	μ = -0.782, σ = 0.947	Log-normal	NMA
OS of sintilimab plus IBI305	μ = 0.592, σ = 1.108		μ = 0.858, σ = 1.176	μ = 0.339, σ = 1.039	Log-normal	NMA
PFS of L + P	μ = -0.350, σ = 1.166		μ = 0.179, σ = 1.360	μ = -0.809, σ = 0.971	Log-normal	NMA
OS of L + P	μ = 0.692, σ = 1.157		μ = 1.240, σ = 1.311	μ = 0.205, σ = 1.042	Log-normal	NMA
PFS of atezolizumab plus bevacizumab	μ = -0.649, σ = 1.007		μ = -0.232, σ = 1.160	μ = -1.009, σ = 0.861	Log-normal	NMA
OS of atezolizumab plus bevacizumab	μ = 0.878, σ = 1.199		μ = 1.508, σ = 1.360	μ = 0.328, σ = 1.042	Log-normal	NMA
PFS of L + T	μ = 0.477, σ = 1.467		μ = 0.930, σ = 1.6380	μ = 0.023, σ = 1.3025	Log-normal	NMA
OS of L + T	μ = 1.267, σ = 1.337		μ = 1.771, σ = 1.483	μ = 0.824, σ = 1.228	Log-normal	NMA
Treatment cost ($)
Sorafenib; Bayer (per plate)	12.07		Fixed	Fixed	Gamma	Bid-winning price
Lenvatinib; Eisai (per plate)	15.17		12.14	18.2	Gamma	Bid-winning price
Sintilimab; Xinda (per cycle)	303.37		242.7	364.04	Gamma	Bid-winning price
IBI305; Xinda (per kg)	23.51		18.81	28.21	Gamma	Bid-winning price
Apatinib; Hengrui (per plate)	14.69		11.75	17.63	Gamma	Bid-winning price
Camrelizumab; Hengrui (per cycle)	361.89		289.51	434.27	Gamma	Bid-winning price
Atezolizumab; Roche (per cycle)	4606.74		3685.39	5528.09	Gamma	Bid-winning price
Bevacizumab; Roche (per 100 mg)	210.67		168.54	252.8	Gamma	Bid-winning price
Pembrolizumab; Merck (per cycle)	5033.15		4026.52	6039.78	Gamma	Bid-winning price
Regofenib, Bayer (per plate)	24.16		19.33	28.99	Gamma	Bid-winning price
TACE (per cycle)	2986.24		2388.99	3583.49	Gamma	Bid-winning price
Best supportive care (per year)	3180		2544	3816	Gamma
Bed charge (per cycle)	58.57		46.86	70.28	Gamma	Bid-winning price
Follow-up and monitoring (per year)	659.58		527.66	791.5	Gamma
Hospice care (per person)	1839		1471.2	2206.8	Gamma
Cost of adverse events ($)
Proteinuria	562.64		450.11	675.17	Gamma	DRG
Hypertension	520.51		416.41	624.61	Gamma	DRG
Asthenia	0		0	0	Gamma	DRG
Rash	245.79		196.63	294.95	Gamma	DRG
Hand-foot syndrome	245.79		196.63	294.95	Gamma	DRG
RCCEP	577.11		461.69	692.53	Gamma	DRG
Diarrhea	630.2		504.16	756.24	Gamma	DRG
Increased AST	474.72		379.78	569.66	Gamma	DRG
Increased blood bilirubin	439.75		351.8	527.7	Gamma	DRG
Neutropenia	586.8		469.44	704.16	Gamma	DRG
Decreased platelet count	582.02		465.62	698.42	Gamma	DRG
Anemia	685.67		548.54	822.8	Gamma	DRG
Sorafenib	174.72		139.78	209.66	Gamma	Calculation
Lenvatinib	234.41		187.53	281.29	Gamma	Calculation
Camrelizumab plus rivoceranib	505.2		404.16	606.24	Gamma	Calculation
Sintilimab plus IBI308	177.39		141.91	212.87	Gamma	Calculation
Atezolizumab plus bevacizumab	126.12		100.9	151.34	Gamma	Calculation
L + P	262.36		209.89	314.83	Gamma	Calculation
L + T	329.78		263.82	395.74	Gamma	Calculation
After progression	283.15		226.52	339.78	Gamma	Calculation
Other parameters
Weight	65		52	78	Normal
Discount rate	0.05		0	0.08	Uniform
Cycle of TACE	3		1	6	Uniform
Exchange rate	7.12		-	-	-	Fixed
Utility
PFS	0.76		0.608	0.912	Beta
Progressive disease	0.68		0.544	0.816	Beta
Disutiliy of toxicities
Proteinuria	0		0	0	Beta
Hypertension	0.08		0.064	0.096	Beta
Asthenia	0.1		0.08	0.12	Beta
Rash	0.1		0.08	0.12	Beta
Hand-foot syndrome	0.116		0.093	0.139	Beta
RCCEP	0.1		0.08	0.12	Beta
Diarrhea	0.05		0.04	0.06	Beta
Increased AST	0		0	0	Beta
Increased blood bilirubin	0		0	0	Beta
Neutropenia	0.2		0.16	0.24
Decreased platelet count	0.11		0.088	0.132	Beta
Anemia	0.07		0.056	0.084	Beta
Sorafenib	0.036		0.029	0.043	Beta	Calculation
Lenvatinib	0.034		0.027	0.041	Beta	Calculation
Camrelizumab plus rivoceranib	0.08		0.064	0.096	Beta	Calculation
Sintilimab plus IBI305	0.018		0.014	0.022	Beta	Calculation
Atezolizumab plus bevacizumab	0.01		0.008	0.012	Beta	Calculation
L + P	0.032		0.026	0.038	Beta	Calculation
L + T	0.031		0.0248	0.0372	Beta	Calculation
After progression	0.044		0.0352	0.0528	Beta	Calculation

AST: Aspartate aminotransferase; DRG: Diagnosis-Related Group; L + P: Lenvatinib plus pembrolizumab; L + T: Lenvatinib plus transarterial chemoembolization; NMA: Network meta-analysis; OS: Overall survival; PFS: Progression-free survival; RCCEP: Reactive cutaneous capillary endothelial proliferation; TACE: Transarterial chemoembolization.

As shown in Table 4, sorafenib yielded 1.07 QALYs at a total cost of $46493.12. All seven regimens achieved higher total QALYs than sorafenib, in ascending order: Avastin plus toripalimab (1.39 QALYs), lenvatinib (1.46 QALYs), R + C (1.57 QALYs), S + I (1.68 QALYs), L + P (1.85 QALYs), A + T (2.07 QALYs), and L + T (2.75 QALYs), corresponding to incremental QALY gains of 0.32, 0.39, 0.50, 0.61, 0.78, 1.00, and 1.68, respectively. Among them, L + T exhibited the highest QALY in the PFS state (1.90 QALYs), whereas A + T produced the highest QALY following disease progression (1.46 QALYs). In terms of costs, L + T incurred the highest expense during PFS ($68059.03), while A + T generated the highest post-progression cost ($70963.42). L + T also showed the highest total cost ($109918.91), followed by A + T ($106082.73) and L + P ($90995.46).

Table 4 Base-case analysis results.

Treatment	QALYs PFS	QALYs PD	QALYs total	Cost PFS ($)	Cost PD ($)	Cost total ($)	InCr cost ($)	InCr QALYs	Incremental cost-effectiveness ratio ($/QALY)	Acceptability at willingness-to-pay	Probability optimal
Sorafenib	0.35	0.72	1.07	10380.95	36112.17	46493.12	Reference	Reference	Reference	Reference	11.30%
Avastin plus toripalimab	0.5	0.89	1.39	31707.84	44091.81	75799.65	19276.99	0.32	91582.91	0.00%	0.00%
Lenvatinib	0.73	0.73	1.46	29365.81	36404.3	65770.11	18599.42	0.39	49428.18	48.50%	5.90%
Rivoceranib plus camrelizumab	0.61	0.96	1.57	18063.18	47029.36	65092.54	35766.97	0.5	37198.84	50.60%	14.90%
Sintilimab plus IBI305	0.64	1.04	1.68	31308.96	50951.13	82260.09	44502.34	0.61	58634.38	0.10%	0.00%
Lenvatinib plus pembrolizumab	0.88	0.97	1.85	43623.22	47372.24	90995.46	59589.61	0.78	57054.28	0.60%	0.00%
Avastin plus tecentriq	0.61	1.46	2.07	35119.31	70963.42	106082.73	63425.79	1	59589.61	0.00%	0.00%
Lenvatinib plus transarterial chemoembolization	1.9	0.85	2.75	68059.03	41859.88	109918.91	29306.53	1.68	37753.45	73.50%	67.90%

InCr: Increased; PD: Progressive disease state; PFS: Progression-free survival; QALY: Quality-adjusted life year.

The ICER rankings from lowest to highest were: (1) R + C ($37198.84); (2) L + T ($37753.45); (3) Lenvatinib ($49428.18); (4) L + P ($57054.28); (5) S + I ($58634.38); (6) A + T ($59589.61); and (7) Avastin plus toripalimab ($91582.91). Notably, R + C and L + T had ICER values below the WTP threshold, indicating that both are cost-effective options compared with sorafenib. A detailed pairwise comparison is provided in Supplementary Table 4.

One-way sensitivity analysis: To assess the robustness of the cost-effectiveness results, a one-way sensitivity analysis was conducted, with findings summarized in tornado diagrams (Figure 3, Supplementary Figure 8). The costs of hospitalization (bed charges), imaging, best supportive care, hospice care, AEs, and treatment-related disutilities had minimal impact on ICERs across all regimens. The most influential parameter for A + T, R + C, S + I, L + P, and avastin plus toripalimab was the HR for OS, derived from the NMA. In contrast, for L + T and lenvatinib, the HR for PFS exerted the greatest impact on ICERs. Additionally, the cost of lenvatinib was a key determinant of its cost-effectiveness. For A + T and S + I, variations in the weight and pricing of anti-vascular endothelial growth factor inhibitors (bevacizumab and the biosimilar IBI305) introduced substantial uncertainty. These findings highlight the parameters that most strongly affect cost-effectiveness, underscoring the need for accurate estimation of survival benefits and treatment costs.

Figure 3 Tornado diagrams of one-way sensitivity analyses, using sorafenib as the comparator, the top ten variables were showed. BSC: Best supportive care; HR: Hazard ratio; L + T: Lenvatinib plus transarterial chemoembolization; OS: Overall survival; PD: Progressive disease; PFS: Progression-free survival; R + C: Rivoceranib plus camrelizumab; Sora: Sorafenib; TACE: Transarterial chemoembolization; Tori: Toripalimab.

PSA: A PSA was performed to examine uncertainty in cost-effectiveness outcomes. In the ICER scatter plot (Figure 4), points below the WTP threshold indicate cost-effective treatments compared with sorafenib, whereas points above the threshold represent non-cost-effective options.

Figure 4 Probabilistic sensitivity analysis scatter plot for each treatment compared with sorafenib. QALY: Quality-adjusted life year; TACE: Transarterial chemoembolization.

The probabilities of being cost-effective relative to sorafenib were as follows: (1) Avastin plus toripalimab (0%); (2) Lenvatinib (48.5%); (3) R + C (50.6%); (4) S + I (0.1%); (5) L + P (0.6%); (6) A + T (0%); and (7) L + T (73.5%) (Table 4). Cost-effectiveness acceptability curves (Figure 5) demonstrated that when the WTP threshold was below $30135/QALY, sorafenib was the most cost-effective option. However, as the WTP exceeded $30135/QALY, L + T emerged as the optimal choice. Under the predefined WTP threshold of $37669, the probabilities of each regimen being the most cost-effective were: (1) Avastin plus toripalimab (0%); (2) Lenvatinib (5.9%); (3) R + C (14.9%); (4) S + I (0%); (5) L + P (0%); (6) A + T (0%); and (7) L + T (67.9%) (Table 3)[18,30,41-53]. These results indicate that L + T is the most cost-effective regimen under current economic conditions in China.

Figure 5 Cost-effectiveness acceptability curves under different willingness-to-pay thresholds. TACE: Transarterial chemoembolization.

Scenario analysis: In China, generic medications that have passed consistency evaluations are considered suitable substitutes for brand-name drugs and are typically offered at lower prices. Currently, seven generic lenvatinib formulations have been included in the national centralized purchasing catalog, with prices ranging from $0.51 to $4.63 per tablet. Scenario analysis revealed that as the cost of lenvatinib decreased, the ICERs for lenvatinib, L + P, and L + T declined proportionally. Despite price variations, L + T consistently maintained the highest probability of being the optimal regimen (Supplementary Table 5). These findings suggest that the cost-effectiveness superiority of L + T is robust.

DISCUSSION

This study evaluated the clinical efficacy and cost-effectiveness of first-line treatment regimens for advanced HCC in China through an NMA and cost-effectiveness analysis. NMA, a widely accepted approach for indirect treatment comparison, primarily evaluates efficacy but often overlooks financial toxicity[54,55]. Given that certain therapies provide survival benefits at substantial costs, economic evaluation is essential to guide clinical decision-making. Unlike previous cost-effectiveness studies that compared only two or a few regimens[56-59] – leading to inconsistent conclusions, such as conflicting findings regarding A + T's cost-effectiveness[56-58] – our study provides a comprehensive evaluation of all available first-line options. Both systemic and locoregional therapies were integrated into a unified framework, thereby offering greater value for informing clinical practice than earlier investigations.

Consistent with previous research, our findings confirm that most next-generation regimens offer PFS and OS advantage over sorafenib[6,9,10,57,58], potentially introducing bias. To enhance accuracy, our study employed HRs specific to the Chinese or Asian subgroups for each regimen relative to sorafenib to derive survival distribution parameters. As previously reported in both CARES-310 and HEPAORCH trials, the Asian subgroups consisted predominantly of Chinese patients. Moreover, our bias analysis indicated that heterogeneity between Chinese and Asian substituted subgroups was not statistically significant; thus, substituting Asian data for R + C and avastin plus toripalimab was methodologically acceptable.

Although C + A demonstrated efficacy comparable to lenvatinib in the NMA, it was excluded from the cost-effectiveness analysis because cabozantinib is not yet commercially available in China. Obtaining unmarketed pharmaceuticals from international sources involves complex legal, safety, and economic challenges, rendering C + A economically inaccessible within the Chinese healthcare context. In the COSMIC-312 trial, the Asian subgroup sample size was relatively small – 63 participants in the C + A group vs 33 in the sorafenib group – indicating limited statistical power. Furthermore, none of the seven included regimens designated C + A as a comparator in the original clinical trials. Inclusion or exclusion of C + A resulted in consistently low and stable heterogeneity in the NMA; therefore, this exclusion likely had minimal impact on our conclusions.

Subgroup analyses of AEs are often omitted in clinical trial reports. In our cost-effectiveness analysis, we assumed that AEs were comparable between the Chinese and overall population. Variations in drug toxicity across different races have been documented, mainly related to single nucleotide polymorphisms[60]. For anti-tumor agents, this phenomenon has been observed in cytotoxic agents like taxanes[61]. However, in HCC therapy – dominated by anti-angiogenic and immunotherapeutic agents – evidence of racial differences in toxicity is limited. Moreover, disutilities and cost associated with AEs exerted minimal influence on ICERs, as shown by sensitivity analysis. Therefore, this assumption is unlikely to affect our results or alter the overall conclusions.

In the cost-effectiveness analysis, a 10-year time horizon was adopted, exceeding the typical life expectancy of patients with advanced HCC. This extended time frame was chosen to capture the potential long-term survival benefits associated with modern therapies, particularly the durable responses seen with immunotherapy. Consequently, QALY values ranked from highest to lowest as L + T, A + T, L + P, S + I, R + C, lenvatinib, and avastin plus toripalimab, aligning with the OS ranking from NMA. ICERs ranked from lowest to highest as R + C, L + T, lenvatinib, S + I, L + P, A + T, and avastin plus toripalimab. Only R + C and L + T had ICERs below the WTP threshold, indicating they are cost-effective first-line options compared with sorafenib, while the others were not cost-effective at current prices.

Despite numerous randomized trials, most locoregional-tyrosine kinase inhibitor (TKI) combination strategies have failed to outperform TKI monotherapy[31,62]. However, the LAUNCH trial demonstrated that L + T significantly improves PFS and OS compared with lenvatinib monotherapy, establishing it as a first-line treatment option for advanced HCC[30]. Cost-effectiveness analyses by Li and Wan[12] and He et al[63] also support L + T over lenvatinib alone. Our findings reinforce this evidence: Integrating NMA and PSA results shows that L + T ranks highest in PFS, OS, and QALYs. Although L + T has a slightly higher ICER than R + C, PSA indicates that it remains the most probable optimal choice at China’s current WTP threshold. The increasing availability of generic lenvatinib ($0.51-$4.63 per tablet) is expected to further improve the cost-effectiveness of L + T, as confirmed by our scenario analysis.

The CARES-310 trial confirmed the clinical superiority of R + C over sorafenib for advanced HCC[24]. However, in QALY rankings, R + C lagged behind L + T, A + T, L + P, and S + I, suggesting limited efficacy benefits for Chinese patients. Additionally, R + C’s higher disutility, driven by severe AEs such as hand-foot syndrome, hypertension, and thrombocytopenia (grade ≥ 3), indicates poorer tolerability. While its lower cost is advantageous, the impact of AEs on patient care must be considered. In our cost-effectiveness analysis, R + C had the lowest total cost and ICER among the seven regimens, ranking second only to L + T in the probability of being the optimal treatment at the WTP threshold. Given the economic disparities across China, R + C may represent a viable option in less-developed regions. For instance, in provinces with a per capita GDP below $10000 (e.g., Yunnan Province and Guizhou Province), its affordability may enhance accessibility despite higher medical resource utilization due to severe AEs. However, the increased incidence of severe AEs, such as an 18.2% rate of hand-foot syndrome, may necessitate additional hospitalization and supportive care, complicating its use in resource-limited settings. Thus, while R + C remains a cost-effective option in certain areas, its toxicity burden should be carefully evaluated, particularly where healthcare resources are constrained.

The IMbrave150 trial demonstrated the superiority of A + T over sorafenib, with Chinese patients deriving greater survival benefits than the global population[26]. Similarly, the LEAP-002 trial suggested a trend toward prolonged PFS and OS with L + P in the Chinese subgroup, though without statistical significance[29]. In our study, A + T and L + P exhibited comparable efficacy, yielding QALYs of 2.07 and 1.85, respectively, ranking just behind L + T. Notably, A + T achieved the highest post-progression QALY, likely reflecting the long-tail effect of programmed death ligand-1 inhibition, which enhances immune activation and reshapes the tumor microenvironment[64,65]. Additionally, A + T showed the lowest disutility value, indicating better tolerability and making it suitable for older or frail patients who are more prone to treatment-related AEs. However, both A + T and L + P impose substantial financial burdens, with ICERs exceeding the WTP threshold, rendering them non-cost-effective under current economic conditions in China. Price reductions could improve their cost-effectiveness. For instance, a 30% price reduction for L + P through negotiation could lower its ICER to $42241, approaching the WTP threshold and underscoring the critical role of pricing policies. Moreover, the Phase III HEPATORCH study, presented at Chinese Society of Clinical Oncology 2024, confirmed PFS and OS benefits of avastin plus toripalimab over sorafenib. Nevertheless, this regimen had the lowest QALY and the highest ICER among those evaluated, making it a less favorable choice for advanced HCC.

Furthermore, our scenario analysis highlights generic drug adoption as a crucial future strategy to improve treatment affordability. The availability of multiple generic lenvatinib formulations at substantially lower prices ($0.51-4.63 per tablet) demonstrates significant potential to enhance the cost-effectiveness of lenvatinib-based regimens. Small-molecule generics are chemically synthesized and follow streamlined approval processes, making them dominant in the global generics market[66]. Their introduction has profoundly reshaped the pharmaceutical landscape, providing cost-effective alternatives to branded drugs. We suggest that developing generics for cabozantinib – a multitargeted small-molecule TKI similar to lenvatinib – represents a promising approach to improve the future cost-effectiveness of the C + A regimen.

This cost-effectiveness analysis has several limitations that should be acknowledged. First, it relies on specific assumptions, which may result in discrepancies between model predictions and real-world outcomes. Although sensitivity analysis and PSA help address uncertainties, potential bias remains unavoidable. Real-world data are warranted to validate these findings. Second, the analysis is based on NMA, where wide confidence intervals for HRs – owing to its indirect nature – introduce a major source of uncertainty. Finally, our analysis only accounts for grade 3 or higher AEs, which is a common pragmatic approach in oncology for the sake of balancing complexity and precision[67]. This may underestimate the overall disutility and associated costs of AEs. However, because low-grade AEs exert limited impact on quality of life and incur relatively low healthcare expenditures, and given that sensitivity analysis demonstrated minimal influence of AE-related disutilities or costs on ICER outcomes, the robustness of our conclusions remains uncompromised.

CONCLUSION

In conclusion, L + T emerges as the most favorable first-line treatment for advanced HCC, balancing efficacy and cost-effectiveness and thus representing the preferred option in China. Although less effective, R + C remains a reasonable alternative in underdeveloped regions due to its lower cost and acceptable ICER. Nevertheless, the study’s reliance on indirect data and assumptions underscores the need for high-quality RCTs and real-world evidence to validate these findings, particularly in economically disadvantaged settings.