Cost-effectiveness of sorafenib versus SBRT for unresectable advanced hepatocellular carcinoma

A systematic literature search of the PubMed database was performed to identify all randomized controlled trials (RCT) of sorafenib or SBRT for unresectable hepatocellular carcinoma from January 1, 1999, to March 31, 2016. The search strategy was based on combinations of (“unresectable” or “inoperable” or “advanced” or “metastatic” hepatocellular carcinoma” [Mesh]) and (SBRT or sorafenib [Mesh] (“randomized controlled trials” or” clinical trials” [Mesh]). We also searched cost-effectiveness studies using the medical subject headings or text key words: quality-adjusted, QALY, life-year gained (LYG) and cost-effectiveness. The appropriate full text was reviewed and evaluated by 2 reviewers (AC and HL) independently, using the same inclusion and exclusion criteria. RCT or clinical studies published in English that evaluated sorafenib or SBRT for mHCC were included. Letters to the editor, case reports, non randomized trials, animal studies, editorials and posters were excluded. Any discrepancies in inclusion were resolved by consensus. We finally selected one RCT and one clinical trial of sorafenib and SBRT for advanced hepatocellular carcinoma as the clinical data source of the model.

The decision-analytic Markov model comparing the cost-effectiveness of two different regimens over a 5-year time horizon was programmed in TreeAge Pro 2014 Suite (R1.0 Released; TreeAge Inc., Williamstown, MA). The clinical data of population modeled, adverse events and treatment protocol were based on the SHARP and Phase I/II clinical trials [12, 16]. As the experts’ opinion and referred to the published literature, three health states were considered in the model: stable disease, progression disease and death for inoperable advanced HCC (Fig. 1). A patient in the model was considered to be in one of three health states at any time. All patients began from the stable stage and transit from one state to another on the basis of the transition probabilities and received either sorafenib or SBRT according to the treatment regimen. All parameters were detailed in Table 1. Our model did not include deaths from natural causes occurring in any health state. Death from cancer was assumed to happen after disease progression. The model perspective was based on that of the National Health Insurance (NHI) in Taiwan, with a 1-month cycle length and adjusted to half-cycle in each health state process. The model time horizon was set to 5 years to avoid the exclusion of long-term survivors. All costs and health outcomes were discounted at a real annual rate of 3 % to adjust for the relative value of the Taiwan dollar at present. We used the definition of willingness to pay (WTP) threshold suggested by the World Health Organization (WHO): 3 times the per capita gross domestic product (GDP) [19, 20]. The Taiwan per capita GDP in 2015 was $ NT$737,715 (US$22,355) [21]; thus, a WTP threshold was considered as $ NT2,213,145/QALY. The outcomes of the analysis are expressed as incremental costs effectiveness ratio (ICER) which was the costs spent to gain a quality adjusted life-years (QALY). Model parameters were described in Table 1. A panel of local experts (blood oncologist and radiation oncologist and one expert in economic evaluations) was consulted to ensure that assumptions taken into consideration in the model reflected routine clinical practice.

The outcome data from the SHARP and Phase I/II clinical trial were used [12, 16]. As expert opinions, the total radiation dose, sorafenib dose and schedule, patterns of treatment failure and survival of the regimen were assumed to be the same as those in the trials. According to the SHARP trial, patients were randomized to receive continuous sorafenib oral treatment with either 400 mg of sorafenib (consisting of two 200-mg tablets) twice daily for 6 week cycles. Tumor measurements were performed at screening, every 6 weeks during treatment (within 10 days before the end of each cycle), and at the end of treatment by computed tomography or magnetic resonance imaging. Patients visited the clinic every 3 weeks and at the end of treatment for assessment of compliance, safety, and determination of side effects. The SBRT treatment regimen was according to the phase I/II trial, doses of 30 to 54 Gy (24 to 54 Gy in Trial 1) in six fractions every other day over 2 weeks were delivered to the planning target volume (PTV). The dose to tumor vascular thrombosis plus PTV margin could be limited to 30 Gy. Detailed description was in the trial [16].

Key probabilities for this model were included in Table 1. The trial did not collect probabilities and utility weights, therefore, the utilities were necessary to retrieve from the published literature [17, 18]. Transition probabilities of health states were estimated as follows: P(1 month) = 1- (0.5) ^{(1/median PFS)}; this equation was derived from the following equations: P = 1 − e^− RandR = − ln 〔0.5 〕/1 − (0.5)^{(1/median time)/number of treatment cycles)} [22, 23]. The efficacies of sorafenib and SBRT treatment were extracted by Kaplan-Meier survival curves of the patients whose data were collected from two clinical trials.

The direct medical costs in this study were extracted from the National Health Insurance research database (NHIRD) and converted to 2015 NT dollar (1US$ = NT$ 33.0).The direct medical cost in this study included drug costs, laboratory test, physician visits, pharmacy dispensing fee, administration and nursing care fee. The drug costs for sorafenib and the costs of treatments for grade 3/4 adverse events (AE) were also included (Table 2). As a policy issued by the NHI, the reimbursement cost of SBRT is about NT$210,000 as a treatment package for early hepatocellular carcinoma and lung cancer, we therefore assumed this reimbursement cost for advanced HCC. We also assumed that the patient who survived by additional 1 month may produce the additional costs of visiting outpatient monthly.

One-way sensitivity analysis was conducted to determine the potential impact of different parameters on this analysis; the result was presented as a tornado diagram. We hypothesized that the parameters varied over a range of ± 30 % in relation to its base-case value (Table 1).

The probabilistic sensitivity analysis using a Monte Carlo simulation was conducted to assess the impact of the uncertainty around the key parameters of the model on the ICER. That is, distributions for each parameter with the probabilistic sensitivity analysis were modeled. Log-normal distributions were adopted for all costs and beta distributions were adopted for probabilities, utilities and toxicity. The probabilistic sensitivity analysis was based on 10,000 samples, and the results were presented as a cost-effectiveness acceptability curve.

We considered only grade 3–4 toxicity associated with sorafenib or SBRT as the two clinical trials. The rate of drug-related severe toxicity caused permanent treatment and discontinuation of the drug was 11 % for sorafenib and 24.8 % for SBRT [12, 16].

The baseline characteristics of the patients treated with sorafenib or SBRT were well balanced and no significant differences were noted with respect to sex, age, ECOG performance status, plasma levels of α-fetoprotein (AFP), Barcelona Clinic Liver Cancer staging B and C. However, the percentage of Child-Pugh class A liver function, extrahepatic spread and vascular invasion were higher in SHARP trial compared to the Phase I/II trial (Table 3).

The direct medical costs of the drugs, laboratory test, physician visits, pharmacy dispensing fee, administration, nursing care fee and grade 3/4 AE treatments were shown in Table 2.

The median overall survival (mOS) and median time to progression (mTTP) were 10.7 and 4.1 months in the sorafenib group of the SHARP trial, whereas, the mOS and mTPP were 17 and 6.0 months in the SBRT group from the Phase I/II trials, as shown in the Kaplan-Meier survival curves (Fig. 2). According to the equation described above, the monthly transition probability for PFS state was 0.153, from PFS to progression disease (PD) state was 0.0627, from PD state to death was 0.1184 for the sorafenib group. For SBRT, the monthly transition probability for PFS state was 0.086, from PFS to progression disease (PD) state was 0.109, from PD state to death was 0.0399 (Table 1). Utility scores of the health states were retrieved from two previously published studies (sorafenib group: 0.68 for PD, 0.76 for PFS; SBRT group: 0.63 for PD₁, 0.72 for PFS₁) [24, 25].

The cost-effectiveness analysis demonstrated that the ICER for sorafenib compared to SBRT was NT$ 3,788,238 per quality-adjusted life year gained (cost/QALY) in the base scenario. Based on the WTP threshold of NT$ 2,213,145/QALY, the sorafenib was not cost-effective versus SBRT under the defined WTP threshold in this analysis (Table 4).

One-way sensitivity analysis was done for SBRT and sorafenib with a hypothesized variation of ±30 % in patients treated with two regimens by using the tornado diagram. The results showed the important parameters driving the model. The broader the horizontal bar in a tornado diagram, the more impact the input parameter has on the model. Analyses showed that the results of the model were highly sensitive to an assumption on the first five parameters, which were utility of progression disease for the sorafenib treatment, utility of progression free survival for SBRT, utility of progression free survival for sorafenib, utility of PFS to progression disease for SBRT and transition probability of progression disease to dead for SBRT (Fig. 3).

Varying all variables simultaneously by ± 30 % in the Monte Carlo simulation, the results demonstrated that the probability of cost-effectiveness at a willingness to pay threshold of NT$ 2,213,145 per QALY was 100 % and 0 % chance for SBRT and sorafenib (Fig. 4).

This work did not require any written patient consent. The local ethics committee of Centre Jean Perrin approved this work.

Not applicable.

The datasets supporting the conclusions of this article are included within the article.