This study was based on a previously conducted randomized, open-label phase 3 study and does not contain any new data evaluating human participants or animals. Eli Lilly and Company Ltd. approve the use of these data.

Model Structure

The model is a five-state cohort transition model (Fig. 1) populated with data from monarchE OS IA3 and the literature, with guidance and validation provided by expert opinion. Within the model a maximum of two years of treatment with abemaciclib in combination with a minimum of five years of physician’s choice adjuvant ET is compared to ET alone among patients with HR+, HER2–, node-positive EBC at high-risk of disease recurrence. For the purposes of the monarchE trial, and the CE model, the intention-to-treat (ITT) population comprised:

Cohort 1 (91%) defined by clinical and pathologic features as ≥ 4 positive axillary lymph nodes (pALN) OR 1–3 pALN and at least one of the following: tumor size ≥ 5 cm or histologic grade 3

Cohort 2 (9%) defined by 1–3 pALN and central Ki-67 (protein associated with cancer proliferation) ≥ 20% AND tumor size < 5 cm and no grade 3 disease [12].

Fig. 1

Model overview. ET endocrine therapy, IDFS invasive disease-free survival. aET-resistant: Disease recurrence while receiving or within 12 months of completing prior adjuvant ET.bET-sensitive: Disease recurrence at least 12 months after completing prior adjuvant ET. cA fixed payoff approach was used to accrue life years, QALYs and costs

The five health states were IDFS, non-metastatic recurrence (NMR), remission, metastatic recurrence (MR), and death. All patients entered the model in the IDFS health state where they receive treatment. From here, patients could: (1) die, (2) experience a non-metastatic disease recurrence and transition to NMR, (3) experience a distant disease recurrence and transition to MR, or (4) complete adjuvant ET and remain in IDFS without further treatment until recurrence or death.

Patients moving to NMR were split depending upon the disease recurrence event: (1) second primary neoplasm or (2) locoregional/contralateral recurrence. Those experiencing second primary neoplasm exited the model to follow the clinical care pathway appropriate for their new diagnosis (not shown). Those with locoregional/contralateral disease recurrence transitioned through a tunnel state, spanning up to 12 months of treatment dictated by the type/location of the disease recurrence. Patients could die at any point while in this tunnel state or experience disease recurrence and move to MR. Those who do not die or experience metastatic disease recurrence, exit the tunnel state and transition to remission. Patients remained in remission until experiencing a distant disease recurrence or die from any cause.

At the time of this analysis, data from monarchE were not available to model the pathway for patients who experienced a metastatic disease recurrence. A systematic literature review (SLR) of clinical observational studies also failed to identify suitable data to model the metastatic setting in greater detail. The MR health state was informed from previously developed CE models evaluating the use of abemaciclib plus ET compared with alternative treatment regimens for the management of locally advanced (non-curative) and metastatic BC. To avoid additional complexity and model burden, the MR health state was modeled as an absorbing health state with fixed payoffs for costs, life years (LYs), and quality-adjusted life years (QALYs). The specific treatment pathway for patients with MR is dependent on the timing of the disease recurrence. Patients who experienced a distant disease recurrence while receiving, or within 12 months of completing, adjuvant ET, were considered to have ET-resistant (ETR) disease. Those who experienced a distant disease recurrence more than 12 months after completing adjuvant ET were considered to have ET-sensitive (ETS) disease; this was in accordance with ESO-ESMO international consensus guidelines for advanced BC [13]. ETR patients received payoffs based on an existing CE model, which evaluated abemaciclib plus fulvestrant compared with alternative treatment options for patients who had disease recurrence and were considered resistant to prior ET. ETS patients received payoffs based on an existing CE model, which evaluated abemaciclib plus a nonsteroidal aromatase inhibitor (NSAI) compared with alternative treatment options for patients whose cancer was sensitive to prior ET. Both CE models have been used to support access and reimbursement but neither have been published [14, 15].

Clinical OutcomesState Transition and Survival Model Parameters

Inputs for IDFS and time to treatment discontinuation (TTD) were extrapolated from monarchE individual patient level data (IPD). In accordance with NICE Decision Support Unit (DSU) technical support document (TSD) 14, seven parametric distributions (Weibull, loglogistic, generalized gamma, gamma, Gompertz, log-normal, and exponential) were evaluated when parameterizing the model along with both 1- and 2-knot hazard spline models [16]. Distributions were selected based on goodness of fit criteria [Akaike Information Criteria (AIC) and Bayesian Information Criteria (BIC)] as well as visual assessment of fit. External validation of the chosen curves was conducted using multiple sources: data identified through an internally conducted SLR of randomized controlled trials evaluating adjuvant ET-based regimens in patients with HR+, HER2- EBC; and external and internal clinical expert opinion [17].

For IDFS, the dependent loglogistic distribution provided the best statistical fit (supplementary Table A1). Figure 2 confirms the fit of the loglogistic extrapolation to the monarchE Kaplan–Meier (KM) curve. The 1-knot hazard spline model, the second-best statistical fit, was explored through scenario analysis. A visual assessment of fit for all curves is presented in supplementary Fig. A1 (abemaciclib plus ET) and Fig. A2 (ET alone).

Fig. 2

IDFS Kaplan–Meier curves and loglogistic extrapolation for abemaciclib plus ET and ET alone (OS IA3; ITT population). ET endocrine therapy, IDFS invasive disease-free survival, ITT intention-to-treat, KM Kaplan–Meier, OS IA3 overall survival interim analysis 3

Application of Treatment Effect Waning

The observed IDFS data of the monarchE trial did not indicate a treatment waning effect at 5-year follow-up, although abemaciclib treatment was stopped at 2 years. The ATAC (Arimidex, Tamoxifen, Alone or in Combination) trial explored long-term outcomes after a median follow-up of 10 years for women with early-stage BC treated for five years with adjuvant ET [18]. The authors of the paper demonstrated in their most recent publication the falling recurrence rates for HR+patients on anastrozole versus tamoxifen over time with ‘carryover benefit’ lasting up to eight years following which the treatment effect begins to wane. Based on these data and treatment practice, stopping rules were applied to abemaciclib (2 years) and adjuvant ET (5 years) with the assumption that the treatment effect would be maintained for eight years before waning. Waning was assumed until year 26, the point at which the IDFS hazard rate was equal to the general background mortality hazard rate in the United Kingdom (UK; Fig. 3) [19]. This approach aligned with that used, and accepted, by NICE in TA612 for neratinib [20]. The impact of including treatment waning was explored through scenario analysis.

Fig. 3

Crossing of IDFS hazard rate with general population mortality hazard rate for the UK. ET Endocrine therapy, IDFS Invasive disease-free survival, UK United Kingdom

Time to Treatment Discontinuation

Time to treatment discontinuation data were available for the maximum two years of abemaciclib, as such the observed KM curve was used. For ET, TTD, for both arms, was modeled using a 2-knot independent hazard spline, which had the best statistical fit.

Overall Survival Without Distant Disease Recurrence

There were limited follow-up data from which to model the transitions to death from IDFS, NMR, and remission. Given that metastatic disease recurrence was the key prognostic of death, an assumption was made that mortality would be equivalent for patients in each of these health states. As such, transitions to death from IDFS, NMR, and remission were modeled via OS without distant recurrence.

A dependent exponential model was selected, based on statistical fit and internal validity, to model OS without distant disease recurrence for abemaciclib plus ET and ET alone.

Non-metastatic Recurrence

Data from monarchE indicated that, among patients experiencing an IDFS event, other than death, 27% of those receiving abemaciclib plus ET experienced an NMR compared to 25% of those receiving ET alone.

The rate of transition to MR, from both NMR and remission, were assumed constant over time due to the limited evidence available. MonarchE IPD were used to inform an exponential distribution for the transition from NMR to MR. The transition from remission to MR was informed by the NICE TA for trastuzumab (TA632) [21].

Metastatic Recurrence

Data from monarchE indicated that, among all patients experiencing an IDFS event, other than death, 73% of those receiving abemaciclib plus ET experienced a metastatic disease recurrence compared to 75% of those receiving ET alone. Clinical data from models focused on MR and the wider advanced BC literature were used to model the MR health states. Outcomes (resource use, treatment pathways, and utilities) from existing CE models were used in the model to generate outcomes for patients whose disease recurs.

Costs and Resource Use

All costs were sourced for the year 2023 based on 2021/2022 National Health Service (NHS) Reference Costs, Unit Costs of Health and Social Care 2022 or inflated using the NHS cost inflation index (NHSCII) to 2023 [22,23,24]. Drug costs were, list price, taken from the England-based electronic market information tool (eMIT) and/or British National Formulary (BNF) databases [25, 26]. TTD curves were used alongside drug acquisition costs to determine estimated total cost of treatment.

Resource use in the IDFS health state was based on the length of time spent in the health state, aligned with the approach adopted in previous studies [20, 21, 27]. Resource use for best supportive care (BSC) was derived from monarchE trial data. Resource use in the NMR health state was informed by experts as well as NICE guidelines, and involved a mix of surgery, radiotherapy, chemotherapy, and/or adjuvant ET [28]. Terminal care costs were included for all deaths that occurred in the model and were guided by the previous metastatic BC models and NICE guidelines [29]. The costs applied to the MR health state were derived, as detailed above, from resource use in existing metastatic BC models combined with updated costs.

The proportions of patients experiencing Grade 3 or 4 adverse events in the IDFS health state were derived from monarchE [30]. Resource use associated with treatment of adverse events were counted as a one-off cost in the first model cycle.

Health-Related Quality of Life

Utility values were derived from clinical trials and the literature. For the IDFS health state, utility values were derived from monarchE IPD. The European Quality of Life 5 Dimensions 5 Level (EQ-5D-5L) values were cross-walked to the EQ-5D-3Level (3L) using the Van Hout algorithm, to which UK tariffs were applied [31]. As there were no significant differences between treatment arms, an overall utility was applied instead of a treat-specific utility. These values were age-adjusted, per NICE guidance, using the formula provided by Ara and Brazier (2011) [32]. Utility values for remission were assumed to equal those for IDFS. This approach aligned with that adopted in Technology Appraisal (TA) for trastuzumab emtansine (TA632) [21]. The weighted utility for NMR was assumed to consider a lower utility (obtained from the literature [33]) for the first 3 months of reinitiating intensive treatment with the remaining time equal to IDFS. For the MR health state, utility values were derived from the existing metastatic BC models. Utility values were tested through sensitivity analyses.

Disutility values were used to assess the impact of adverse events on QALYs. The disutility value per adverse event was multiplied by the mean adverse event duration to obtain a utility decrement that was applied to the proportion experiencing the event during the first model cycle. Disutility values and adverse event duration were informed by the literature and previous NICE TAs in early and metastatic BC [34,35,36,37].

Key Assumptions in the Model

Several assumptions were integrated into the model to estimate the cost-effectiveness of abemaciclib plus ET over time. Of particular importance are the assumptions regarding the consistent proportion of transitions from IDFS to NMR or MR, with the assumptions that the hazard of dying is the same across the IDFS, NMR, and remission health states, the treatment effect waning over 18 years, commencing from year eight, and the exclusion of rechallenge with a CDK4/6 inhibitor following distant disease recurrence.

Analysis

Costs and QALYs for both abemaciclib plus ET and ET alone were calculated over a lifetime horizon. In the base case, the time horizon was 50 years coinciding with the time point at which survival in both arms fell to < 1% in the extrapolations. The time horizon was varied through scenario analyses. The perspective was that of the UK NHS and Personal Social Services. In line with NICE guidance, both costs and outcomes were discounted at 3.5% [8]. The cycle length was 28 days, to reflect the treatment, and a half-cycle adjustment was made to correct for events not occurring at the beginning, or end, of the cycles.

Uncertainty Analyses

To test the robustness of the model, inputs and assumptions were varied and tested through deterministic sensitivity, scenario, and probabilistic analyses. For the deterministic sensitivity analyses, each parameter was varied between its upper and lower bound value. The probabilistic analysis involved assigning distributions to the parameters and performing 1000 simulations to assess the impact of the uncertainty across the parameters simultaneously. Different scenario analyses were conducted to explore the uncertainty related to assumptions and methodological choices made in the base case.

Table 1 details the settings and assumptions used in key base case and scenario analyses.

Table 1 Key model inputs: base case and scenario analyses