Population and Patient Characteristics

In this analysis, patients with high-risk eTNBC were considered as the target population, consistent with the population in the KEYNOTE-522 trial [7]. The baseline characteristics of patients, including starting age, body weight, and body surface area (BSA), were obtained from Hong Kong Breast Cancer Study (HKBCS). Only female patients were considered in the model for analysis (Table 2). WTP willingness-to-pay, QALY quality-adjusted life year.

Model Structure and Analysis

The state transition cohort model was developed using Microsoft Excel® 2016. Figure 1 illustrates the health states and allowable transitions in the cost-effectiveness model. The model consists of four mutually exclusive health states (event-free [14], locoregional recurrence [LR], distant metastases [DM], and death) to track patients’ disease course and death over time.

Fig. 1

Model schematic. AEs adverse events

In the KEYNOTE-522 trial, patients were randomized to the pembrolizumab + chemotherapy arm or chemotherapy arm. In the neoadjuvant phase of the pembrolizumab + chemotherapy arm, pembrolizumab (200 mg administered once every 3 weeks (Q3W) on day 1 of cycles 1–8) in combination with chemotherapy (4 cycles of paclitaxel plus carboplatin followed by 4 cycles of doxorubicin or epirubicin plus cyclophosphamide) was administered to the patients. The patients then received definitive surgery after 3–6 weeks of neoadjuvant treatment completion. In the adjuvant phase, radiation therapy as indicated or pembrolizumab as a single agent was administered Q3W for 9 cycles. Chemotherapy is considered as a base case comparator to pembrolizumab + chemotherapy as a management strategy for high-risk eTNBC in the neoadjuvant phase to reflect local clinical practice in Hong Kong. The patients then received definitive surgery after 3–6 weeks of neoadjuvant treatment completion. In the adjuvant phase, radiation therapy as indicated or placebo was administered Q3W for 9 cycles [7]. The details of dosing schedule of different regimens in both the pembrolizumab + chemotherapy arm and chemotherapy arm, based on the KEYNOTE-522 trial, are provided in Table 3, whereas the details for adjuvant pembrolizumab are provided in Table 4.

Table 3 Neoadjuvant treatments: dosing schedule, relative dose intensity, and treatment allocationTable 4 Adjuvant treatments: dosing schedule, relative dose intensity, and treatment allocation

This economic evaluation was conducted from a Hong Kong third-party payer perspective. Therefore, only direct healthcare costs were considered, expressed in Hong Kong dollars (HKD). The original cost values were inflated to 2022 prices using the consumer price index (CPI) obtained from the Census and Statistics Department, Hong Kong [15].

A time horizon of 32 years was used in the analysis to capture all relevant costs and benefits. The cohort of patients were followed up until 88 years of age, which is the average life expectancy of female citizens in Hong Kong as per the World Bank data [16], from the starting age of 56 years of the patient cohort at the model entry based on HKBCS. The present model used a weekly cycle length to allow for precise calculation of drug acquisition and administration costs. Also, half-cycle correction was applied to costs and effectiveness for additional precision in the base case. Costs and effectiveness were discounted at 3% annually, as recommended by the annual report on CPI, Census and Statistics Department, Hong Kong [17]. The outcomes of the model included aggregated and disaggregated incremental costs, quality-adjusted life years (QALYs) and life years (LYs) estimated for each treatment arm, along with the incremental cost-effectiveness ratio (ICER) expressed as incremental cost per QALY gained and per LY gained.

A series of scenario and one-way sensitivity analyses (OWSA) were conducted to assess the robustness of results through changes in parameters and under alternative model settings. A probabilistic sensitivity analysis (PSA) was also conducted to estimate the probability of pembrolizumab + chemotherapy being cost-effective relative to chemotherapy, based on the willingness-to-pay (WTP) threshold of HKD 1,171,308 per QALY, which is three times of gross domestic product (GDP) per capita for Hong Kong as of June 2023 [18]. For the PSA, a Monte Carlo simulation with 1000 iterations was conducted, and in each iteration, the model inputs were randomly drawn from the specified distributions (Table 2). The standard errors of the selected distributions were obtained directly from the same data source that informed the mean value, if available. Otherwise, the standard error for each cost parameter was assumed to be equal to 50% of the mean value. Tables S1 and S2 in the Supplementary Materials provide more information on the scenario analyses, OWSA and PSA.

Clinical InputsTransition Probabilities

The transition probabilities in the Markov model were based on patient-level data from the KEYNOTE-522 trial, with extrapolation using fitted parametric functions. Patients enter the model in the EF state, and at the end of each cycle, patients may remain in the EF state or transition to LR, DM, or death state. Also, patients who are in the LR state may stay in the same state or transition to DM, or death state, but could not transition back to the EF state. Patients in the DM state may stay in the DM or transition to death state but could not transition back to the EF or LR state.

The survival curve fitting for different parametric functions was carried out in line with the NICE Decision Support Unit guidelines [19]. Standard one-piece and two-piece parametric models were extrapolated using exponential, Weibull, log-normal, log-logistic, Gompertz, gamma, and generalized gamma distributions. For the base case, Akaike information criterion (AIC) and Bayesian information criterion (BIC) of statistical tests combined with visual inspection were used to obtain the best fit parametric distributions. The clinical plausibility of the long-term extrapolations was also evaluated for the selected distributions. Alternative parametric functions at different cutoff points were tested in scenario analyses. Table S3 in the Supplementary Materials presents detailed information regarding all health state transitions and data sources.

Transition Probabilities from EF State

The estimated probability of an EFS event and the probabilities of first EFS event being LR, DM, or death in each treatment arm were used to compute the cause-specific probability of each transition from the EF state (EF → LR, EF → DM, or EF → death) for each model cycle. Data from the KEYNOTE-522 trial were used to estimate the probability of the first EFS event being LR, DM, or death whereas the time to LR, time to DM, and time to death were analyzed using Gray’s method for competing risks [20]. The probability of the EFS event was capped by the all-cause natural mortality based on Hong Kong-specific life tables [21].

Different parametric models were fitted to the patient-level data from the KEYNOTE-522 trial to extrapolate the EFS throughout the modelled time horizon [22]. The base case parametric survival models were selected on the basis of the statistical fit, visual inspection, and clinical plausibility of the extrapolated model for both the pembrolizumab + chemotherapy and chemotherapy arms. Week 50 was selected as a suitable cutoff point for two-piece models. Generalized gamma and log-normal distributions were chosen as the best-fit for base case analysis for the pembrolizumab + chemotherapy arm and the chemotherapy arm, respectively [23]. Alternative parametric functions (log-normal and generalized gamma) at different cutoff points (week 43, 50, and 68) were tested in scenario analyses (Table 2). Figure 2a presents the Kaplan–Meier curves and the base case extrapolated curves for EFS. The base case curves and alternative parametric functions are shown in Fig. S1 in the Supplementary Materials and the estimated cumulative incidence of the three transitions EF → LR, EF → DM, and EF → death along with the actual cumulative incidence derived from the KEYNOTE-522 trial data is presented in Fig. S2 in the Supplementary Materials.

Fig. 2

Source for observed survival: Schmid P, Cortes J, Dent R, et al. Event-free survival with pembrolizumab in early triple-negative breast cancer. New England Journal of Medicine. 2022 Feb [11]

Predicted long-term outcomes under base case parametric distribution assumptions versus observed data from the KEYNOTE-522 trial. a EFS. b OS. EFS event-free survival, KM Kaplan–Meier, OS overall survival.

Transition Probabilities from LR State

The KEYNOTE-522 trial data for time from LR to DM or death were fitted using parametric models, and the best fit was an exponential distribution [22]. As a result of the limited number of patients with confirmed LR in the trial, pooled data from both the treatment arms were used to estimate transition probabilities of LR → DM or death. The transition probabilities of LR → DM and LR → death were obtained for each model cycle using estimated probability of LR → DM or death, and the proportions of DM and death, respectively, obtained from the KEYNOTE-522 trial.

Transition Probabilities from DM State

The transition probabilities from DM to death were calculated using the duration of survival for patients with documented DM in the KEYNOTE-522 trial [22]. Parametric models were fitted on patient-level data for each treatment arm, providing an exponential distribution as the best fit. As a result, the fitted exponential distribution was used to estimate the time-constant transition probability of DM → death.

Adverse Events

The model accounted for medical costs and health disutilities associated with all-cause grade 3+ adverse events (AEs) with an incidence of at least 5% from the combined neoadjuvant and adjuvant phases in either arm. KEYNOTE-522 trial data were used to determine the incidence rates, mean durations (Table 2), and hospitalization rates of AEs (provided in Table S5 in the Supplementary Materials) [22].

Health Utilities

Utility inputs used in the base case and scenario analyses were derived through primary analyses of the EuroQoL-five-dimension questionnaire (EQ-5D-5L) data collected in the KEYNOTE-522 trial [22]. The generic health status assessed from the EQ-5D questionnaires were converted to population-based utility values using the UK algorithm [24]. Utilities based on the UK EQ-5D-5L value set were used in the base case analysis. The impact of the UK five-level (5L) crosswalk to three-level (3L) utility algorithm was tested in the scenario analysis. Utilities were examined by health state, treatment status (on- or off-treatment), and AE status. Pooled data were used because of insignificant differences between treatment arms. Furthermore, disutility related to aging was applied on the basis of a published model with mean health state utility values from the general population [25]. Coefficients used for age-related disutility in the model are presented in Table S4 in the Supplementary Materials.

Costs

Different cost categories considered in the analysis are presented in Fig. 3 and their corresponding costs are provided in Table 2. The costs were obtained from Hospital Authority, Hong Kong and were inflated to 2022 on the basis of CPI obtained from the Census and Statistics Department, Hong Kong [15].

Fig. 3

Different cost categories included in the analysis

Neoadjuvant and Adjuvant Drug Cost

Drug acquisition costs for neoadjuvant and adjuvant treatment were calculated as a function of the list price per drug unit (Table 5), defined dosing for the medication, relative dose intensity (RDI), and time on treatment (TOT) (Tables 3 and 4 for neoadjuvant and adjuvant treatment, respectively). The list price per drug unit were retrieved from Hospital Authority, Hong Kong. The dosing schedules for all drugs were consistent with the treatment protocol used in the KEYNOTE-522 trial [7]. The TOT for neoadjuvant and adjuvant treatments was estimated using observed Kaplan–Meier curves from the KEYNOTE-522 trial which were used to determine the number of treatment cycles in each arm (Fig. S3 in the Supplementary Materials depicts the TOT in neoadjuvant and adjuvant setting). The projected TOT was then adjusted according to the RDI observed in the KEYNOTE-522 trial, to account for any delay or interruptions in administration [22].

Table 5 Drug regimens and unit costs

Pembrolizumab is administered at a fixed dose of 200 mg, once Q3W on day 1 of cycles 1 through 8 and is available in 100-mg vials only. The unit price of pembrolizumab 100 mg vial (HKD 19,800) was obtained from Hospital Authority, Hong Kong. The average number of vials per administration for each BSA-based or body weight-based regimens was estimated using a log-normal distribution of the BSA (using mean as 1.92 m2 and SD as 0.23) or body weight distribution (using mean value as 76.37 kg and SD as 17.84), respectively, of patients in Hong Kong. An optimal vial mix algorithm was employed to reduce vial wastage in the base case, where the lowest cost per milligram vial size was selected.

A unit cost of HKD 2400.00 for chemotherapy preparation and administration, obtained from the Hospital Authority, List of Private Services, Hong Kong, was included as an administration cost for each treatment combination, and then applied to the weekly model cycles according to the dosing schedule of each individual drug (Table 2) [26].

Subsequent Therapy Cost

After the documented DM in the KEYNOTE-522 trial, about 62.5% of patients in the pembrolizumab + chemotherapy arm and 70.3% in the chemotherapy arm received systemic anticancer therapy. When patients entered the DM state, the expenses of drug acquisition and administration for first- to fourth-line metastatic TNBC treatments were included. The KEYNOTE-522 trial data were used for distribution and duration of subsequent treatments in the model [22]. The average cost for post-DM therapy was estimated to be HKD 308,030.97 per patient for the pembrolizumab + chemotherapy arm, whereas it was calculated to be HKD 206,854.46 per patient for the chemotherapy arm. Detailed dosing schedule, relative dose intensity, and treatment allocation in first-line metastatic treatment are provided in Table 6.

Table 6 First-line treatments for metastatic TNBC: dosing schedule, relative dose intensity, and treatment allocationSurgery and Radiation Cost

The KEYNOTE-522 protocol stated that when neoadjuvant therapy was completed, patients would undergo definitive surgery 3–6 weeks later [7]. Following the procedure, radiation therapy and adjuvant therapies were administered as prescribed. Costs for surgery were calculated on the basis of the weighted average unit costs of lumpectomy (HKD 24,000.00) and mastectomy (35,750.00) obtained from Queen Mary Hospital Private Clinic (Table 2) [27]. The unit cost of radiation, HKD 150,000.00 (taken as an average of HKD 100,000.00 and HKD 200,000.00, the end-points of the range of cost), was obtained from Queen Mary Hospital Private Clinic, which estimated the mean radiation costs among patients who had breast-conserving surgery [27]. The percentages of patients who underwent surgery and radiation therapy in the pembrolizumab + chemotherapy arm (98.00% and 75.90%, respectively) and chemotherapy arm (97.70% and 78.50%, respectively) were obtained from KEYNOTE-522 trial data [22].

Disease Management Cost

The disease management cost comprised recurring and one-off disease management costs. The recurring disease management costs in EF state across various time periods were based on the average cost of consultation for each subsequent follow-up, and were obtained from the Hospital Authority, List of Private Services, Hong Kong [26]. The frequency of use of these resources and the respective costs were used to estimate the average recurring disease management costs and these were then converted to costs per week for EF state, which are shown in Table 2. Recurring disease management costs for the LR and DM states include the cost of clinical visits and the cost of medical interventions (cost of tumor markers, positron emission tomography/computed tomography (PET/CT) whole body, magnetic resonance imaging (MRI) brain, and standard lab tests). Just like in the EF state, the average recurring disease management costs were estimated and further converted into costs per week (Table 2) for both the LR and DM state.

After entering LR or DM stage, a one-time disease management cost was included as an average cost for biopsy of recurrence sites. These costs were taken from the Hospital Authority’s List of Private Services, Hong Kong (Operations) (Table 2) [28].

Terminal Care Cost

Terminal care cost was applied as a one-time cost prior to death, under the assumption that the cost was equal for death from EF, LR, or DM states. It includes expenses in the last 6 months of life and was obtained from Wong et al. [29]. The cost was inflated to 2022 using CPI indices from the Census and Statistics Department, Hong Kong, and then converted to from USD to HKD using currency exchange rate to include in the CEA [21].

Adverse Event Management Cost

The one-time AE costs were computed as a function of the AE rates, the proportion hospitalized for each AE event, and the unit costs of medical management for each AE in the inpatient or outpatient setting, for each treatment arm.

The unit costs of AE management were calculated by multiplying the number of hospitalized days and outpatient visits required for each AE, obtained from Hospital Authority, Hong Kong, with the unit cost of hospitalization (HKD 6650) and outpatient visit (HKD 1990), respectively, obtained from the List of Private Services, Hospital Authority, Hong Kong [26]. Table 2 displays the medical management unit costs for each AE. It was assumed that the patients with neutropenia, neutrophil count decreased, and decreased white blood cell count did not require hospitalization, and thus the hospitalization cost per unit for the aforementioned AEs was zero.

Ethical Approval

This article is based on the data drawn from trials that had already been completed and from published literature and does not contain any studies with human participants or animals performed by any of the authors.

Model Validation

The observed OS from the KEYNOTE-522 trial serves as the initial validation for the modelled OS [22]. The projected OS curves for both the arms closely resemble the observed Kaplan–Meier curve, as seen in Fig. 2b. Following that, external data from Walsh et al. [30] and Sikov et al. [14] were used to compare the projected OS curve for chemotherapy (Fig. S4 in the Supplementary Materials). Independent clinical experts also confirmed the clinical plausibility of projected long-term outcomes. External data were not available to validate the modelled OS for the patients with eTNBC who received pembrolizumab treatment. The panel of key opinion leaders (KOLs), consisting of eight medical oncologists and two health economists from Europe, validated the intervention extrapolation in terms of the clinical plausibility and in comparison, with the comparator extrapolation in terms of the expected improvement with immuno-oncology therapies. The panel of KOLs supported the validity of the long-term OS estimate for the pembrolizumab + chemotherapy arm.