FormalPara Key Summary Points

Why carry out this study?

Pembrolizumab in combination with chemotherapy as neoadjuvant treatment and then continued as a single-agent adjuvant treatment after surgery has been approved in the USA for high-risk early-stage triple-negative breast cancer (eTNBC) on the basis of results from the randomized phase III KEYNOTE-522 trial, demonstrating that addition of pembrolizumab to neoadjuvant chemotherapy provided significantly improved event-free survival and a favorable trend in overall survival in high-risk eTNBC.

This study aimed to evaluate the cost-effectiveness of pembrolizumab in combination with chemotherapy as neoadjuvant treatment and continued as a single-agent adjuvant treatment after surgery in patients with high-risk eTNBC from a US third-party payer perspective.

What was learned from this study?

The results show that neoadjuvant pembrolizumab + chemotherapy followed by adjuvant pembrolizumab is expected to yield substantial improvements in quality-adjusted life years (QALYs) (by 2.90) and LYs (by 3.37) relative to treatment with neoadjuvant chemotherapy alone.

The incremental cost-effectiveness ratio (ICER) was estimated at $27,285/QALY in the base case, which is lower than all commonly cited willingness-to-pay thresholds in the USA.

Thus, pembrolizumab can be considered a cost-effective treatment option for patients with high-risk eTNBC in the USA, in addition to its higher efficacy versus chemotherapy as demonstrated in KEYNOTE-522.

Introduction

In 2021, there were an estimated 281,550 new cases of breast cancer (BC) and 43,600 BC-related deaths among women in the USA [1]. Approximately 10–15% of cases are triple-negative BC (TNBC), a heterogenous subtype lacking expression of the estrogen receptor, progesterone receptor, and human epidermal growth factor receptor 2 (also known as ERBB2) [2,3,4]. These receptors are common therapeutic targets in BC, making TNBC more difficult to treat and associated with worse outcomes and higher recurrence rates than other BC subtypes [2, 5,6,7]. Additionally, molecular subtyping can identify patients with high-risk forms of TNBC [8, 9].

Definitive surgical resection is the mainstay approach for the treatment of early-stage BC. Neoadjuvant (pre-surgical) chemotherapy has been the standard of care for high-risk early stage TNBC (eTNBC) [10, 11], while adjuvant (post-surgical) chemotherapy may or may not be used in this population [12]. However, high risks of recurrence and death remain for these patients, pointing to the urgent, unmet need for novel therapies that can augment the effectiveness of chemotherapy [13].

Pembrolizumab is a humanized monoclonal programmed cell death-1 (PD-1) antibody that inhibits the interaction between PD-1 and programmed death ligand 1 (PD-L1)/PD-L2 receptors [14]. Pembrolizumab for the treatment of high-risk eTNBC in combination with chemotherapy as neoadjuvant treatment and then continued as a single-agent adjuvant treatment after surgery was approved by the US Food and Drug Administration in July 2021 and the European Medicines Agency in May 2022 [15,16,17,18]. The regulatory approval was based on the results of the phase III KEYNOTE-522 trial (Clinicaltrials.gov identifier NCT03036488) comparing the efficacy and safety of neoadjuvant pembrolizumab + chemotherapy followed by adjuvant pembrolizumab after surgery (referred to henceforth as pembrolizumab + chemotherapy/pembrolizumab) with that of neoadjuvant placebo + chemotherapy followed by adjuvant placebo after surgery (referred to as chemotherapy henceforth) in treatment-naïve patients with high-risk stage II or III TNBC [16]. The first interim analysis of KEYNOTE-522 (median follow-up 15.5 months) indicated that pathological complete response (i.e., no invasive cancer in the breast and negative nodes) was significantly higher among patients receiving neoadjuvant pembrolizumab + chemotherapy vs. chemotherapy (64.8% vs. 51.2%, p < 0.001) [14]. The fourth interim analysis of KEYNOTE-522 (median follow-up 39.1 months) indicated a significant and clinically meaningful improvement in event-free survival (EFS) among the pembrolizumab + chemotherapy/pembrolizumab group vs. the chemotherapy group (36-month EFS rate 84.3% vs. 76.2%; hazard ratio [95% confidence interval (CI)] 0.63 [0.48, 0.82]; p < 0.001) [19]. Numerical improvement on overall survival (OS) was also observed (36-month OS rate 89.7% vs. 86.9%; HR 0.72; 95% CI [0.51, 1.02]). Furthermore, the addition of pembrolizumab to neoadjuvant chemotherapy was generally well tolerated, with no new safety concerns [19].

The economic value of pembrolizumab in eTNBC has not been comprehensively assessed but is essential information for healthcare payers considering the costs and clinical benefits of a new therapy for reimbursement decisions. Thus, the objective of this study was to evaluate the cost-effectiveness of pembrolizumab + chemotherapy/pembrolizumab vs. neoadjuvant chemotherapy in patients with high-risk eTNBC from a US third-party payer perspective. The cost-effectiveness analysis (CEA) was informed by data from KEYNOTE-522.

Methods

Model Overview

A multistate transition model was developed in Microsoft® Excel 2016 to evaluate the economic value of the addition of pembrolizumab to neoadjuvant chemotherapy for high-risk eTNBC from a US third-party public healthcare payer perspective. Efficacy, safety, and utility data were derived from KEYNOTE-522 (data cutoff: March 23, 2021; median duration of follow-up: 39.1 months) [19]. The main outcome of the model was the incremental cost-effectiveness ratio (ICER) in terms of the incremental costs per quality-adjusted life year (QALY) and life year (LY) gained. A lifetime horizon (51 years) was used and based on an average starting age of 49 years from KEYNOTE-522. A weekly cycle length was used to accurately estimate clinical outcomes and treatment-related costs. Costs and effectiveness were discounted at 3% annually, as recommended by the Institute for Clinical and Economic Review [20].

Target Population and Interventions

The target population was patients with high-risk eTNBC, based on the population of KEYNOTE-522. Patients in KEYNOTE-522 were randomized to pembrolizumab + chemotherapy/pembrolizumab or placebo + chemotherapy/placebo arms. In the neoadjuvant phase, pembrolizumab (at a dose of 200 mg) or placebo was administered once every 3 weeks (q3w) up to eight cycles, and the chemotherapy consisted of four cycles of paclitaxel (80 mg/m2 once weekly [qw]) and carboplatin (area under the curve [AUC5]) q3w or AUC 1.5 qw in the first 12 weeks), followed by four cycles of doxorubicin (60 mg/m2) or epirubicin (90 mg/m2) plus cyclophosphamide (600 mg/m2 q3w in the subsequent 12 weeks). Each subject underwent definitive surgery 3–6 weeks after conclusion of the neoadjuvant treatment. In the adjuvant phase, pembrolizumab 200 mg or placebo was administered q3w for up to nine cycles. Postoperative radiation therapy was acceptable in accordance with the standard of care as applicable.

Model Structure

The model included four mutually exclusive health states to simulate patients’ disease course over time: event-free (EF), locoregional recurrence (LR), distant metastasis (DM), and death (Fig. 1). For each health state, costs and utility were assigned within each weekly cycle to estimate the cumulative costs and health outcomes over the modeled time horizon.

Fig. 1
figure 1

Model diagram. AE adverse event

Full size image

Effectiveness Inputs

Transition Probabilities (TPs) Between Health States

TPs between health states were estimated on the basis of the patient-level data from KEYNOTE-522, with extrapolation using fitted parametric functions. Parametric function fitting was carried out in line with the National Institute for Health and Care Excellence (NICE) Decision Support Unit guidelines for survival analysis [21]. Standard one- and two-piece models were explored using exponential, Weibull, log-normal, log-logistic, Gompertz, gamma, and generalized gamma distributions. Statistical tests based on the Akaike information criterion (AIC) and Bayesian information criterion (BIC) combined with visual inspection were used to select the best-fit parametric distributions for the base case. Finally, the clinical plausibility of the long-term extrapolations was evaluated for the selected distributions. Alternative parametric functions were tested in scenario analyses.

TPs from EF State

The modeled cohorts started from the EF state and were subject to the risks of three competing events: LR, DM, or death. As of the trial data cutoff for KEYNOTE-522, 15.7% of patients in the pembrolizumab + chemotherapy/pembrolizumab arm and 23.8% in the chemotherapy arm had at least one documented event. For each model cycle, the cause-specific probability of each transition (EF → LR, EF → DM, or EF → death) was calculated on the basis of the estimated probability of an EFS event, along with the probabilities of LR, DM, or death being the first event in each treatment arm. The probability of the EFS event being LR, DM, or death was estimated using the KEYNOTE-522 trial data. Competing risks among the three events were considered [22].

To estimate the EFS probability over the modeled time horizon, parametric functions were used to extrapolate the EFS data beyond the trial period. First, standard parametric models were fitted to the patient-level data in the KEYNOTE-522 trial for each treatment arm. However, they were not able to provide a good fit to the observed data. Therefore, two-phase parametric models were conducted. The Kaplan–Meier (KM) curves, hazard plots, and cumulative hazard plots of the EFS data were inspected to identify the appropriate cutoff point for the two-phase fittings. Results suggested week 50 as an optimal turning point to test in the economic model. This is because it provided plausible visual fit and a good balance of robust KM data to be used directly in the first phase and enough remaining data to be used to fit a parametric curve in the second phase. Other cutoff points were tested as scenario analyses. The area under the KM curve was used directly for the first phase of extrapolation until week 50, and parametric functions were fitted thereafter. The generalized-gamma and log-normal distribution were selected as the best-fitting distributions and were used in the model base case for the pembrolizumab + chemotherapy/pembrolizumab and the chemotherapy arm, respectively. The choice was based on lowest AIC/BIC statistics, visual inspections against the observed data, and clinical plausibility of the extrapolations. The KM curves and the base case extrapolated curves of EFS are shown in Fig. 2. Alternative parametric functions and the base case curves are presented in Fig. S1 in the supplementary material.

Fig. 2
figure 2

Modeled EFS curves. EFS event-free survival, KM Kaplan–Meier. Modeled curve for pembrolizumab + chemotherapy/pembrolizumab: KM curve followed by generalized gamma distribution from week 50 and onwards; modeled curve for chemotherapy: KM curve followed by log-normal distribution from week 50 and onwards

Full size image

The estimated values of cumulative incidence of the three transitions EF → LR, EF → DM, and EF → death, as well as the cumulative incidence directly from the KEYNOTE-522 trial data, are shown in Fig. 3.

Fig. 3
figure 3

Modeled three competing transitions from EF state. EF event-free, LR locoregional recurrence, DM distant metastasis

Full size image
TPs from LR State

Parametric models were fitted to the data of time to DM or death following LR from the KEYNOTE-522 trial. As a result of the limited number of patients with documented LR in the trial, pooled data from both treatment arms were used. The exponential distribution was found to be the best fit. Therefore, the probability of LR → DM or death was estimated to be constant over time. For each model cycle, the transition probabilities of LR → DM and LR → death were calculated on the basis of the estimated probability of LR → DM or death, and the proportions of DM and death, respectively.

TPs from DM State

The TP from DM to death was estimated on the basis of the survival time following DM for patients who had documented DM in the KEYNOTE-522 trial. Parametric models were fitted to the patient-level data for each treatment arm and exponential distribution was found to be the best fit. Therefore, the time-constant TP of DM → death was estimated on the basis of the fitted exponential distribution.

In addition, age-specific annual probabilities of death for the general population were obtained from the US National Vital Statistics Report [23] and all individuals in the model were subject to general mortality [24].

Time on Treatment (TOT)

TOT data from KEYNOTE-522 were used to separately model the durations of neoadjuvant and adjuvant treatment, which accounted for early treatment discontinuation. No more patients remained on study interventions as of the data cutoff; therefore, the observed TOT data were used directly in the model, avoiding the need for extrapolation (Fig. S2 in the supplementary material).

Adverse Event (AE) Rates and Duration

The model considered all-cause grade 3+ AEs with an incidence of at least 5% from the combined neoadjuvant and adjuvant phases in either arm. Incidence rates, mean durations, and hospitalization rates of AEs were obtained from KEYNOTE-522 (Table 1).

Table 1 Base case inputs
Full size table

Utility Inputs

Health state utility inputs were derived from the primary analyses of the EuroQoL-five-dimension-five level (EQ-5D-5L) data collected in KEYNOTE-522. The EQ-5D-5L questionnaire was administered at treatment cycle 1 of treatment 1 and cycle 1 and 4 of treatment 2 in the neoadjuvant phase, and treatment cycle 1, 5 and 9 in the adjuvant treatment phase, as well as the early discontinuation visit and long-term follow-up visits. Generic health status assessed with the EQ-5D-5L questionnaire was converted to a population-based utility value using a US scoring algorithm based on time trade-off techniques. The generic health status assessed from the EQ-5D questionnaires was converted to population-based utility values using published algorithms (US 5L value set [25] for the base case and crosswalk from 5L to three-level (3L) [26] along with US 3L value set [27] for the scenario analyses). Utilities were analyzed by health state, treatment status, and AE status. Pooled data were used as there were no significant differences between treatment arms (Table 1).

Healthcare Resource Use (HCRU) and Cost Inputs

The cost inputs used in the model, including costs of drug acquisition and administration, disease management, terminal care, and AE management were based on the US clinical practice and are listed in Table 1. It was assumed that 80% of healthcare costs would be paid by healthcare payers. All costs were updated to 2021 US dollars (USD) using the Consumer Price Index, medical care component [28].

Neoadjuvant and Adjuvant Treatment Costs

Pembrolizumab is available in single-use 100-mg vials and administered at a fixed dose of 200 mg q3w on day 1 of cycles 1–8. The list price for pembrolizumab is $5134 per 100-mg vial [29]. The average number of vials per administration for regimens which were body surface area (BSA)-based or body weight-based was estimated using a log-normal distribution of the BSA or body weight distribution, respectively, of patients in KEYNOTE-522 (mean [standard deviation], BSA 1.8 [0.2] m2; body weight 69.7 [16.3] kg). An optimal vial mix algorithm was used to minimize vial wastage in the base case. Administration costs (i.e., intravenous infusion) were derived from the Centers for Medicare and Medicaid Services Physician Fee Schedule [30].

Neoadjuvant and adjuvant treatment costs were calculated using the drug acquisition and administration costs per administration, respective TOT and relative dose intensity (RDI). The neoadjuvant and adjuvant TOT were estimated using the observed KM curves from KEYNOTE-522. The number of treatment cycles in the model was calculated on the basis of the estimated TOT and was adjusted using the RDI observed in KEYNOTE-522 to account for any delays or interruptions in administration which could be due to AEs, non-compliance, or other reasons.

Surgery and Radiation Costs

According to the KEYNOTE-522 protocol, patients who completed neoadjuvant treatment would receive definitive surgery 3 to 6 weeks later. The surgery was followed by radiation therapy, as indicated, and adjuvant treatments. Surgery and radiation costs were applied as one-time costs (Table 1). The proportion of patients receiving surgery (pembrolizumab + chemotherapy/pembrolizumab vs. chemotherapy 98% vs. 97.7%) and radiation therapy (75.9% vs. 78.5%, respectively) were obtained from KEYNOTE-522.

Disease Management and Terminal Care Costs

The average weekly costs of disease management in the EF, LR, and DM states were estimated on the basis of an Optum Health Reporting and Insights claims database study by Sieluk et al. [31]. For patients who remained EF for longer than 10 years, the disease management costs were assumed zero.

A one-time terminal care cost, assumed to include HCRU in the last 30 days of life, was applied upon patients’ death. The cost was estimated by a US Surveillance, Epidemiology, and End Results (SEER) Medicare study amongst patients with metastatic TNBC [32]. Cost elements included inpatient, emergency care, outpatient, skilled nursing facility, and hospice visits. The resultant cost updated to 2021 USD was $15,779.

Subsequent Therapy Costs

Approximately 62.5% of patients in the pembrolizumab + chemotherapy/pembrolizumab arm and 70.3% in the chemotherapy arm received systemic anticancer therapy following documented DM in KEYNOTE-522. Drug acquisition and administration costs associated with first- to fourth-line metastatic TNBC treatments were applied as a one-time cost upon entry into the DM state. The distribution and duration of subsequent treatments from the trial were used in the model. The average cost of subsequent therapy upon DM was estimated at $59,467 per patient in the pembrolizumab + chemotherapy/pembrolizumab arm and $47,804 per patient in the chemotherapy arm (Table 1).

AE Management Costs

The total average cost per patient for managing AEs for each treatment arm was calculated on the basis of AE incidence rates, rate of hospitalization, and unit costs of medical management for each AE episode in the inpatient or outpatient setting. Incidence and costs of selected grade 3–5 AEs are summarized in Table 1.

Sensitivity Analyses

Various scenario analyses were conducted to assess model uncertainty related to key model settings and assumptions. One such analysis explored the lifetime impact of chronic immune-related AEs, including hypothyroidism and adrenal insufficiency for pembrolizumab-treated patients. Another scenario analysis assumed that patients who had not experienced locoregional or metastatic recurrence 8 years after treatment were “cured”, where these patients were assumed no longer at risk for events and only subject to natural mortality. In addition, a scenario analysis was conducted under the assumption that the relative treatment effect of pembrolizumab would diminish (i.e., patients who received pembrolizumab + chemotherapy/pembrolizumab would experience the same hazard of events and death as those who received chemotherapy alone) after 8 years of treatment. The inclusion criteria of chronic AEs and the assumption of “cure” were based on input from an advisory panel review of the model. Waning of treatment effect for pembrolizumab was not supported by clinical evidence nor prompted by the advisory panel; however, this assumption was explored as it was often discussed in reimbursement reviews. Additional scenario-based sensitivity analyses adjusted the model time horizon, the discounting of costs and outcomes, the EQ-5D-5L mapping algorithm, RDI, as well as the application of alternative plausible parametric functions to EFS.

In addition, one-way deterministic sensitivity analyses (DSA) and probabilistic sensitivity analyses (PSA) were conducted to test the robustness of the model results (Table S1 in the supplementary material).

Model Validation

This model and other analogous versions were evaluated by multiple health economists and breast cancer specialists during the model development process. In addition, an expert panel consisting of eight medical oncologists and two health economists provided detailed review and feedback. Model validation covered the conceptual model as well as the model assumptions, methods, inputs, and results. The panel also provided verification to the model structure, analysis plans, clinical plausibility of clinical inputs, duration of treatment effects and modelled outcomes such as the projections of EFS and OS. Where consensus was not achieved, the panel provided guidance on the scope of sensitivity analyses.

The modeled OS was first validated with observed data from the KEYNOTE-522 trial. As shown in Fig. S3 in the supplementary material, the projected OS curve matches closely with the observed KM curve for both arms. Then the projected OS curve for chemotherapy was compared with external data presented by Walsh et al. [33] and Sikov et al. [34] (Fig. S3). The clinical plausibility of the projected long-term outcomes was also verified by independent clinical experts.

No external data are currently available to validate the modeled OS for the pembrolizumab-treated patients with eTNBC. The plausibility of the long-term OS projection for the pembrolizumab + chemotherapy/pembrolizumab arm was supported by feedback from a clinician advisory panel review of the projected treatment effects of pembrolizumab.

Compliance with Ethics Guidelines

The data used in this analysis were based on previously conducted trials and published literature; no studies with human participants or animals were performed by any of the authors. Ethics committee approval is not required.

Results

Base Case

The base case results (Table 2) projected that pembrolizumab + chemotherapy/pembrolizumab would result in a longer EFS (1.62 years) and LYs (3.37) and greater QALYs (2.90) compared to chemotherapy over the lifetime horizon. The addition of pembrolizumab was also projected to be associated with higher overall medical costs ($79,046), mainly driven by drug acquisition costs. The resultant incremental cost per QALY gained with pembrolizumab + chemotherapy/pembrolizumab vs. chemotherapy was $27,285; the incremental cost per LY gained was $23,489.

Table 2 Base case analysis results
Full size table

Scenario Analyses

The results of the scenario analyses are presented in Table S2 in the supplementary material. The majority of the scenarios were consistent with the base case results. However, the results were sensitive to a change in the time horizon, where a 10-year horizon resulted in an ICER of $153,574/QALY, and the consideration of “cure” after 8 years (ICER of $40,126/QALY). Assuming waning of treatment effect after 8 years of pembrolizumab treatment led to an ICER of $45,273/QALY. Alternative plausible parametric distributions for EFS for pembrolizumab + chemotherapy/pembrolizumab increased the ICER, ranging from $34,664/QALY to $57,569/QALY.

DSA and PSA

The impact of parameter variation on ICERs derived by the one-way DSA is shown in the tornado diagram in Fig. 4. The ICERs ranged from $16,384 to $75,674 per QALY gained and the results were most sensitive to variations in the parametric estimates of EFS for pembrolizumab + chemotherapy/pembrolizumab or chemotherapy.

Fig. 4
figure 4

One-way sensitivity analyses tornado diagram of ICER (Δcost/ΔQALY): Pembrolizumab + chemotherapy/pembrolizumab vs. chemotherapy. AE adverse event, DM distant metastases, EF event-free, EFS event-free survival, ICER incremental cost-effectiveness ratio, L lower bound confidence interval of the base case estimate, LR locoregional recurrence, QALY quality-adjusted life years, U upper bound confidence interval of the base case estimate

Full size image

The PSA yielded an average ICER of $30,950/QALY (Fig. 5a). The cost-effectiveness acceptability curve (Fig. 5b) showed an 83% probability that ICER would be below $50,000/QALY and an 99% probability that ICER would be below $150,000/QALY.

Fig. 5
figure 5

Probabilistic sensitivity analysis: a incremental cost and effectiveness plane, b cost-effectiveness acceptability curve. PSA probabilistic sensitivity analysis, QALY quality-adjusted life years, USD United States dollars

Full size image

Discussion

Over the modeled time horizon, neoadjuvant pembrolizumab + chemotherapy followed by adjuvant pembrolizumab for high-risk eTNBC is expected to yield substantial improvements in QALYs (by 2.90) and LYs (by 3.37) relative to treatment with neoadjuvant chemotherapy alone. The ICER for pembrolizumab + chemotherapy/pembrolizumab vs. chemotherapy was estimated at $27,285/QALY in the base case, which falls under all published willingness-to-pay (WTP) thresholds, including the thresholds recommended by the Institute of Clinical and Economic Review (i.e., $50,000 to $150,000 per QALY) [35, 36] and the World Health Organization (WHO) criteria of three times the gross domestic product (GDP) per capita [37]. The results were primarily driven by prevented or delayed cancer recurrence and progression with the addition of pembrolizumab in both neoadjuvant and adjuvant phases, increasing the LYs and QALYs attributed to the pembrolizumab + chemotherapy/pembrolizumab arm. The results of the scenario and sensitivity analyses supported the base case findings, attesting to their robustness. When considering the WTP threshold of $150,000/QALY, the PSA demonstrated that pembrolizumab + chemotherapy/pembrolizumab had a 99% probability of being cost-effective versus chemotherapy in the USA.

To our knowledge, this is the first study to examine the cost-effectiveness of pembrolizumab + chemotherapy/pembrolizumab vs. chemotherapy in high-risk eTNBC in the USA. A key strength of the current analysis is its use of data from KEYNOTE-522, a phase 3 randomized controlled trial comparing pembrolizumab + chemotherapy/pembrolizumab and chemotherapy in high-risk eTNBC. This permitted the use of the most accurate data for this patient population as inputs in the model, and validation of the estimated OS against the observed OS in KEYNOTE-522. EFS and OS extrapolations for the chemotherapy arm were validated against external studies [33, 34]. Additionally, an expert panel of seven medical oncologists and two health economists validated the model structure, assumptions, inputs, and outcomes.

A positive reimbursement recommendation was recently made by Canadian Agency for Drugs and Technologies in Health (CADTH) for pembrolizumab in high-risk eTNBC [38]. The recommendation highlighted that addition of pembrolizumab to neoadjuvant chemotherapy addressed unmet needs for this patient population in terms of reducing the risk of disease recurrence, controlling disease, and providing a manageable toxicity profile. The base case CEA proposed by CADTH resulted in an ICER of $81,408/QALY. As the CADTH analysis adopted different model assumptions and inputs, and reflected different clinical practices and payer systems compared to the present study, the results were not directly comparable.

The current findings are consistent with previously conducted CEAs comparing novel therapies with prior standard of care in other early-stage solid tumors in the USA. For example, a study by Bensimon et al. using data from KEYNOTE-054 found that adjuvant pembrolizumab was highly cost-effective versus observation alone following complete resection of high-risk stage III melanoma at a WTP threshold of $100,000 [39]. An ICER of $39,228/QALY (adjusted to 2021 USD) was reported by Garrison et al. for the addition of trastuzumab to chemotherapy as an adjuvant treatment in early-stage HER2-negative BC from a US payer perspective [40].

However, the results of this analysis should also be considered in the light of several limitations. First, the median follow-up of KEYNOTE-522 was 39 months at the data cutoff date for this analysis; extrapolations were necessary to model a lifetime horizon and there is inherent uncertainty in this process. To address this, multiple scenario analyses were performed to evaluate alternative extrapolation approaches for EFS, and parameter uncertainty associated with the base case extrapolation was tested in the sensitivity analyses. The results from the sensitivity analyses supported the robustness of the base case ICER. In addition, a rigorous validation process was carried out to validate the extrapolation of EFS and OS outcomes. However, longer-term follow-up of the clinical trial will be imperative to confirm the model results. Second, the model results can be sensitive to uncertainty in the duration of treatment effects for pembrolizumab. However, it is challenging to model this because of lack of evidence. Although a scenario analysis was conducted to explore the potential impact of treatment waning effect in this study, there is no evidence to indicate that the treatment effect with pembrolizumab would be lost at later follow-ups. Instead, the current KEYNOTE-522 trial shows clear separation of the survival curves after median follow-up of the trial, and the separation between the treatment arms grows more substantial over time [19]. In the absence of clinical rationale or evidence on the duration of treatment effect after stopping therapy, application of an arbitrary treatment waning effect may not be appropriate [41]. Third, the transition following the entry of the DM state was estimated from the KEYNOTE-522 data, which reflected the treatment pattern during the trial follow-up period. As TNBC care is changing rapidly, with immunotherapies and targeted therapies emerging as a new treatment paradigm, the predicted mean OS may be an underestimate compared to that in current real-world practice because more patients would be receiving more efficacious treatments in the metastatic setting. Finally, the costs of grade 3+ AEs with an incidence of less than 5% in both groups were not included in the model. Considering their low incidence, exclusion of these events was not expected to bias the results.

Conclusions

In this economic evaluation from a US third-party payer perspective, pembrolizumab in combination with chemotherapy as neoadjuvant treatment and continued as a single-agent adjuvant treatment after surgery improves patients’ life expectancy and QALYs compared with neoadjuvant chemotherapy and is projected to be a cost-effective treatment option for patients with high-risk eTNBC.