A cost-effectiveness analysis of capecitabine maintenance therapy versus routine follow-up for early-stage triple-negative breast cancer patients after standard treatment from a perspective of Chinese society

Patients in the capecitabine maintenance group yielded 9.30 QALYs compared with 8.01 QALYs for patients in the observational group. The total cost was $9106.67 in the capecitabine maintenance group and $5714.96 in the observational group, respectively. The capecitabine maintenance therapy provided an additional 1.29 QALYs at an additional cost of $3391.70. The ICER was $2630.53 (95% CI: $1159.81–$5090.12) per QALY gained, which was considerably lower than the recommended WTP threshold (i.e., $28,130 per QALY), indicating that 1-year capecitabine maintenance therapy as adjuvant treatment was highly cost-effective for early-stage TNBC patients after standard treatment, compared to routine follow-up. A sensitivity analysis was conducted to test the uncertainty of ICER according to alternative survival functions (i.e., Exponential, Weibull, Log-logistic, and Gamma). The results showed that the ICERs were consistently lower than the recommended WTP with a range from $2176.91 per QALY gained on Gamma distribution to $2876.78 per QALY gained on exponential distribution, indicating the robustness of ICER results on survival functions (Additional file 1: Table S5).

Results of the one-way sensitivity analysis are presented in the tornado diagram (Fig. 1). The parameters that considerably impacted ICER estimates included the annual discount rate, monthly drug cost, and treatment cost after relapse. Within the range of each parameter specified in Table 2, the ICER was consistently below the recommended WTP threshold.

The cost-effectiveness acceptability curve indicated that 1-year capecitabine maintenance therapy was more likely to be cost-effective compared with routine follow-up when the WTP threshold was above $2610.00 per QALY gained (Fig. 2). The probability that capecitabine maintenance therapy was cost-effective was 100%, indicating that capecitabine maintenance therapy was dominantly cost-effective, compared with routine follow-up, in improving the prognosis of early-stage TNBC patients after standard treatment. If the monthly capecitabine cost was 50% and 10% in the base-case model, the WTP threshold that capecitabine maintenance therapy could be treated as cost-effective was decreased to $1406.2 and $482.7 per QALY gain, respectively (Fig. 2A).

The absolute risk reduction of relapse rate at the 1, 2, and 5 years since the date of the randomization in the capecitabine maintenance group, compared with routine follow-up, were 3.4%, 6.0%, and 8.4%, and the NNTs to prevent one additional relapse event was 29.2 (95% CI: 13.2–196.6), 16.7 (95% CI: 8.4–111.6), and 12.0 (95% CI: 5.7–82.6), respectively (Fig. 3A). The additional costs of capecitabine maintenance therapy, compared with routine follow-up, to prevent one additional relapse were $138,361.75, $79,131.55, and $56,860.99 at 1, 2, and 5 years, respectively. However, compared with routine follow-up, the NNT to prevent one additional death of capecitabine maintenance therapy was not statistically significant (Fig. 3B).

The target population of the present study was from the SYSUCC-001 trial conducted in 13 academic centers and clinical sites in mainland China [14]. Briefly, eligible patients were women who had pathologically confirmed invasive breast ductal carcinoma, were hormone receptor and ERBB2 negative, were early stage with T1b-3N0-3cM0, and completed standard adjuvant chemotherapy. Eligible patients were randomly assigned (1:1) into either the capecitabine maintenance (interventional) group or the observational (control) group within 4 weeks after completion of standard adjuvant chemotherapy. In the observational group, patients were routinely followed up according to the clinical guidance. In the capecitabine maintenance group, patients received oral capecitabine at 650 mg/m² twice daily continuously for 1 year without interruption. Capecitabine was given to patients via monthly prescriptions at the hospital. Patients in the capecitabine maintenance group had their blood, liver, and renal function tests taken monthly to monitor the safety of the medication. In both groups, patients were followed up by physical examination, assessment of menopausal status, breast ultrasound, and abdominal ultrasound by trained oncologists every 3 months during years 1 to 2, every 6 months during years 3 to 5, and yearly thereafter. Mammography and chest x-ray were performed yearly in both groups.

In SYSUCC-001 [14], a total of 434 patients (median age: 46, range: 24–70) were included in the analysis, including 221 in the capecitabine maintenance group and 213 in the observational group; the median age was 45 for the capecitabine maintenance group and 48 for the observational group. After a median follow-up of 61 months, there were 93 recurrence events and 72 death events. The 5-year DFS was 82.8% in the capecitabine group vs. 73.0% in the observational group, and the 5-year OS was 85.5% in the capecitabine group vs. 81.3% in the observational group, respectively.

A Markov state-transition model, using the TreeAge Pro 2005 software (TreeAge Software Inc, Williamstown, MA), was developed to evaluate the costs and effectiveness of 1-year capecitabine maintenance therapy for early-stage TNBC patients who had completed standard adjuvant chemotherapy compared with routine follow-up from the perspective of Chinese society. The three mutually exclusive health states of DFS, relapse, and death were included in the state-transition model (Fig. 4A). The abbreviated decision tree and Markov model are presented in Fig. 4B. All simulated patients started from the DFS state and could stay in the DFS state or move to the relapse or death state at the next cycle length due to corresponding transition probabilities; patients who progressed to the relapse state could only stay in the relapse state or move to the death state. Death was referred to as the absorbing state. The Markov model parameters were collected from the SYSUCC-001 trial and literature. Considering that the age peak of breast cancer incidence was around 45–49 years in China [34], and the median age of the capecitabine maintenance group was 45 in the SYSUCC-001 trial [14], the simulation was conducted from the age of 45 in a monthly cycle. The 1-month cycle length was used in line with the drug schedule. A 30-year time horizon was selected to capture the lifetime impacts of capecitabine maintenance therapy for early-stage TNBC patients [35]. During the 30-year period, the majority of patients would die. Transition probabilities were estimated based on individual-level data from the SYSUCC-001 trial. The study complied with the Consolidated Health Economic Evaluation Reporting Standards 2022 (CHEERS 2022) statement (Additional file 1: Table S1) [36].

The time horizon in the Markov model is beyond the period of the SYSUCC-001 trial. It is generally preferred to apply a fitted parametric distribution over using raw survival data in cost-effectiveness studies when the simulated time horizon is much longer than the trial. To optimally extrapolate the survival probabilities over a 30-year time horizon, five common parametric distributions with Exponential, Weibull, Log-normal, Log-logistic, and Gamma were fit to the original individual-level data from the SYSUCC-001 trial (Additional file 1: Table S2). The selection of optimal parametric distribution for survival function was based on the recommended criteria [37, 38], including the lower values of Akaike information criterion (AIC), Bayesian information criterion (BIC), absolute of –2log-likelihood, and the sum of the squared errors (SSE) of predictions of survival curve over the observed time span, along with the visual comparison of modeled against Kaplan–Meier survival curves. The goodness-of-fit results indicated that log-normal distribution had a better fit for individual-level data of both DFS and OS (Additional file 1: Table S3, and Figure S1) and was finally selected for the cost-effectiveness analysis. The internal validation of the Log-normal distribution was confirmed by the agreement of modeled clinical outcomes with trial data in terms of the 5-year DFS and OS (Fig. 5). The estimated parameters of Log-normal distribution in Table 1 were applied to fit the 30-year time horizon survival curves of the two groups (Fig. 5). Time-dependent state-transition probabilities in each cycle were calculated using the formula \(P\left(t\to 1\right)=1\frac{S\left(t+1\right)}{S\left(t\right)}\) at a given cycle t in the Markov model, where \(S\left(t\right)=1-\Phi \left[\frac{\mathrm{ln}t-\mu }{\sigma }\right]\) represents the survival probability at time t using parameters (μ, σ) of Log-normal in Table 1 [37, 39, 40]. The transition probability from DFS to death was set as the age-specific natural mortality rate of the general Chinese women population in 2020 [41], according to previous literature [42, 43]. A time-dependent transition probability matrix sample is presented in Table S4 (Additional file 1).

Costs were calculated from the perspective of Chinese society. Direct medical costs included capecitabine, imaging, and laboratory tests for monitoring treatment safety and routine follow-up, managing severe adverse events (grade 3/4), and subsequent treatment after relapse. Since capecitabine (500 mg/piece) was only obtained by overseas importation to China at the beginning of the SYSUCC-001 trial (2010) and its generic was available in China at the end of 2013; Thus, its cost in the base-case model was therefore estimated on the imported unit price. Capecitabine has now entered the centralized drug procurement list in China; its price range was obtained from the Chinese Drug Bidding Database [44]. The unit costs of imaging, laboratory tests, and routine follow-up were based on the price charged at the Sun Yat-sen University Cancer Center (Guangzhou, China) in 2020. As the grade 1/2 adverse events were manageable with standard monitoring, we only involved the costs of grade 3 or above adverse events related to capecitabine maintenance therapy with a significant group difference in incidence rate, including hand-foot syndrome and diarrhea [11]. The incidence rates of severe adverse events were estimated from the SYSUCC-001 [14] and two relevant RCTs [12, 13]. The costs of managing severe adverse events were derived from the Sun Yat-sen University Cancer Center (for hand-foot syndrome) and literature (for diarrhea) [45]. For relapsed patients, chemotherapy, targeted therapy, immunotherapy, or surgery were given at the discretion of the oncologists. The average monthly treatment cost after relapse was estimated for individual total treatment cost by dividing his/her survival months after relapse, based on 16 dead cases (9 patients in the observation group and 7 patients in the capecitabine maintenance group) that occurred in Sun Yat-sen University Cancer Center. The reasonability of the estimated monthly treatment cost after relapse was confirmed by consulting an oncologist. Given that capecitabine was orally administered, we only involved the travel fee for drug collection and/or routine follow-up. Other direct non-medical costs were not considered in this study. The travel fee was set as $12.37 per hospital visit based on literature [46]. In addition, time costs (i.e., those incurred due to absenteeism) were considered in both groups due to drug collection and/or routine follow-up, estimated at $49.97 per day based on the average monthly salary from the Chinese National Bureau of statistics in 2020 [21]. The time cost was involved in the Markov model until 55 years old, which is the mandatory age for retirement in China. The costs for each input variable are presented in Table 2. All costs in the Markov model were converted to US dollars based on an exchange rate of $1 = ¥6.4665 (as of 9 July 2021). In the model, a 5% annual discount rate was considered for all costs and benefits [47].

Effectiveness was measured in quality-adjusted life years (QALYs), calculated as the survival time of one patient in a certain health state multiplied by the healthy utility value (quality of life weight) during the same period. According to a previous report [48], the utility values in the DFS, relapse, and death state were set to 0.8, 0.73, and 0 in this study, respectively. Disutility values due to grade 3/4 adverse events were set to 0.12 for hand-foot syndrome and 0.10 for diarrhea, respectively [49]. All adverse events were assumed to have been incurred in the first cycle [43].

The cost-effectiveness of capecitabine maintenance therapy versus routine follow-up was assessed by the incremental cost-effectiveness ratio (ICER), which is expressed as the incremental cost between two groups per QALY gained. Capecitabine maintenance therapy was considered cost-effective if the ICER was less than a willingness-to-pay (WTP) of $28,130 per QALY gained [50], which was three times that of China’s per-capita gross domestic product (GDP) according to the WHO guideline [20, 51].

One-way and probabilistic sensitivity analyses were performed to evaluate the uncertainty of model parameters on ICER. In one-way sensitivity analysis, the parameters varied once at a time by their ranges specified in Table 2. A tornado chart was present to rank-order the parameters based on their potential impact on ICER. In probabilistic sensitivity analysis (PSA), a Monte Carlo simulation with 10,000 iterations was conducted by simultaneously sampling the model parameters from the distributions of each parameter. The sampling methods were set as gamma distribution for costs parameters, beta distribution for the incidence rate of severe adverse events, utility and disutility parameters, and uniform distribution for time-dependent state-transition probabilities parameters (Table 2). The PSA results were expressed as incremental cost-effectiveness scatter plots and cost-effectiveness acceptability curves.

The number needed to treat (NNT) to prevent one additional relapse or death at a specific time point was calculated by fitting Cox proportional hazard model using the individual-level data of the SYSUCC-001 trial [52, 53], after adjustment of age, menopausal status, tumor size, node status, Ki67 index, surgery type, and treatment group. The NNT could be used to evaluate the treatment benefits of certain therapy in an absolute manner. A lower NNT indicated a higher clinical impact as fewer patients received the treatment without deriving a survival benefit from it. The costs to prevent one additional relapse were calculated based on the cost differences between the capecitabine maintenance and observational group.