2.1 Health Economic Analysis Plan

The hypothetical target population was adult patients with RRMM, with a median age of 64.5 years, consistent with the eligibility criteria of the DREAMM-7 trial [16]. This required patients to have progressed after at least one prior line of therapy, while excluding those refractory to anti-CD38 therapy or anti-BCMA therapy. Patients in the BVd group were assumed to receive belantamab mafodotin (2.5 mg/kg intravenously every 3 weeks), while the DVd group received daratumumab (16 mg/kg intravenously every week for cycles 1–3, every 3 weeks for cycles 4–8, and every 4 weeks thereafter). Both groups also received bortezomib (1.3 mg/m2 subcutaneously on days 1, 4, 8, and 11 of 21-day cycles) and dexamethasone (20 mg on the day of and day after each bortezomib administration) during the first 8 cycles.

From the Chinese healthcare system perspective, a partitioned survival model (PSM) was developed in R to project direct medical costs, quality-adjusted life years (QALYs), and aggregate life years (LYs) associated with BVd compared with DVd. A discount rate of 4.5% per year is recommended for the base-case analysis and to be applied to both costs and health outcomes, per the 2025 edition of the China Guidelines for Pharmacoeconomic Evaluation [20, 21]. The model adopted a lifetime horizon of 40 years. In line with the treatment cycles in DREAMM-7, the cycle length was set as 3 weeks (21 days) for BVd and 3 weeks (for the first 8 cycles) followed by 4 weeks (from the 9th cycle) for DVd. According to the 2025 edition of the China Guidelines for Pharmacoeconomic Evaluation, it is generally recommended to use no more than twice the national per capita gross domestic product (GDP) as the willingness-to-pay (WTP) threshold for each QALY gained [20, 21]. Accordingly, this study estimated the threshold price of belantamab based on a WTP threshold of twice China’s 2025 GDP per capita (99,665 CNY ≈US$14,722), which was US$29,444 per QALY [22]. The reporting of the study followed the Consolidated Health Economic Evaluation Reporting Standard recommendations [23].

2.2 Model Structure

A three-state PSM was constructed including progression-free state, progressed disease (PD) state, and death (Fig. 1). The entire cohort entered the model in the progression-free state, from where patients could transition to the PD state or death state. Those with progressed disease could remain in the PD state or transition to the death state. Patients could not return to the progression-free state after disease progression, reflecting the progressive nature of multiple myeloma. All the patients eventually entered the absorbing death state. The model utilized the underlying parametric curves fitted to the progression-free survival (PFS) and the updated overall survival (OS) data in DREAMM-7 to estimate the proportion of patients in each health state over time. Outcomes were measured in aggregate LYs and QALYs.

Fig. 1

Three-state partitioned survival model

2.3 Model Assumptions

Several assumptions were employed in this analysis. First, patients were assumed to continue the assigned treatment until disease progression or death, whichever came first, and they would initiate a new anti-myeloma therapy as subsequent treatment upon entering the PD state. Second, as patient-reported outcome analysis of DREAMM-7 showed that only a small proportion of patients were very much bothered by adverse events [24], only serious adverse events (SAEs) reported in at least 2% of patients in either treatment arm in DREAMM-7 were assumed to significantly impact quality of life and incur additional treatment costs. These SAEs, including pneumonia, pyrexia, COVID-19, COVID-19 pneumonia, thrombocytopenia, sepsis, lower respiratory tract infection, anemia, syncope, respiratory failure, orthostatic hypotension, and infusion-related reaction, were assumed to incur one-time treatment charges upon occurrence. The model's extrapolation method was validated using data from the trial and external data.

2.4 Model Inputs

The key model inputs are presented in Table 1.

2.4.1 Clinical Data Inputs

The PFS and OS of patients in the BVd group and DVd group were based on the results of DREAMM-7 [16, 17]. Data points from the published survival curves of the trial were extracted using Engauge Digitizer, Version 12.1. The extracted data points, the number of patients at risks, and the number of events at various time points from the published trial were used to reconstruct individual patient-level data using the algorithm developed by Guyot et al. [27]. For each endpoint (progression-free survival and overall survival) in each treatment group, the reconstructed individual patient-level data were fitted to parametric survival models for the commonly used distributions, including the exponential, the Weibull, the gamma, the lognormal, the log-logistic, and the generalized gamma. The best-fitting survival model was selected based on Akaike and Bayesian information criterion (AIC/BIC), visual inspection of the fitted curves, and examination of log-cumulative hazard plots. The fitted OS and PFS curves are presented in Supplementary Figures 1 and 2, with log-cumulative hazard plots shown in Supplementary Figures 3 and 4, and AIC/BIC values are reported in Supplementary Table 1 (see electronic supplementary material [ESM]).

The lognormal distribution was identified as the best‑fitting distribution for OS in both treatment groups. To ensure biological plausibility, we imposed a general population mortality (GPM) constraint on the OS extrapolations with the GPM-clamped approach. A simulated cohort of 5000 patients, matched to the target population in age (median age 64.5 years, IQR 57–71) and sex (55% male), was generated using the age- and sex-specific mortality rates for the Chinese general population from the lifetable in the United Nations 2024 Revision of World Population Prospects [28]. The lifetables used were from 2024 to 2100 and for ages 0 to 100 years, and we extended them to age 130 years using Gompertz model. At each month, the all-cause mortality hazard was set as the maximum of the lognormal-projected hazard and the corresponding GPM rate, thereby preventing unrealistic long-term survival from the heavy tail of the lognormal distribution. To assess the robustness of our results to the choice of OS extrapolation method, we used an excess hazard (EH) model, which decomposed the all-cause mortality hazard into the sum of the background GPM rate and an excess hazard attributable to RRMM, as scenario analysis [29]. For PFS, both the log-logistic and lognormal distributions captured the characteristic long tail typically observed in RRMM patients. Therefore, we selected log-logistic distribution, which yielded smaller AIC/BIC values, to extrapolate PFS curves in base case analysis, while retaining the lognormal model as a scenario analysis to assess the impact of this structural assumption on our results. The proportions of patients in the progression-free state were estimated by the area under the reconstructed PFS curves, and the proportions of patients in the PD state were estimated by the difference between the modeled OS and PFS. The proportions of patients in the death state were estimated as the complement of the OS.

2.4.2 Utility Inputs

An on-site questionnaire survey was conducted in Zhejiang, China to assess the health utility in RRMM [26]. This survey employed a convenience sampling method for distributing the questionnaires to both outpatients and inpatients, and collected a total of 558 valid questionnaires from several general hospitals. In this survey, the utility values of disease remission and progression for different lines of treatment were calculated using the Chinese version of the EQ-5D-5L. We extracted the utility values from this survey to calculate the health utility for the progression-free state and the PD state (Table 1). The disutility and the duration of each serious adverse event were derived from literature and are presented in Supplementary Table 2 (ESM) [25, 30,31,32,33]. Disutilities associated with SAEs were multiplied by the proportions of patients reporting each event in each treatment arm. The resulting QALY decrements were summed up and subtracted from the total QALYs in each group.

2.4.3 Cost Inputs

The costs considered included the drug acquisition and administration cost, subsequent treatment cost, monitoring cost for post progression, cost of disease evaluation and laboratory test, and SAEs management costs. All costs were adjusted for inflation based on the Chinese Consumer Price Index released in April 2026 and reported in USD with a currency conversion ratio of 1 USD equivalent to 6.77 CNY.

The drug acquisition costs for daratumumab and bortezomib were estimated based on the average procurement price obtained from the Beijing Sunshine Medical Procurement Platform. The cost of dexamethasone was considered negligible in the analysis due to its low acquisition cost in China. In the absence of a listed price, the value-based unit price of belantamab was estimated using the model. Price details are shown in Supplementary Table 3 in the ESM. As the median age was 64.5 years among RRMM patients in DREAMM-7, a senior citizen (60–69 years of age) with a body surface area of 1.7 m2 and a weight of 64.3 kg, as per the Fifth National Physical Fitness and Health Report, was used to calculate the dosages and the drug acquisition cost per cycle.

Subsequent treatment costs were estimated based on the proportions of anti-myeloma therapies used as first subsequent therapy by more than 10% of patients receiving subsequent therapies in DREAMM-7 [16], the average procurement price of drugs from the Beijing Sunshine Medical Procurement Platform, and the dosing regimens from labels (see ESM, Supplementary Tables 3 and 4). To address the circularity arising from belantamab being also an option as subsequent treatment, we isolated its contribution from the total subsequent treatment cost as \(total subsequent treatment cost per cycle = other subsequent treatment cost per cycle + W\times P\), where W denotes the proportion of patients receiving belantamab as first subsequent therapy in the DREAMM-7 trial (1% in the BVd group and 12% in the DVd group) and P represents the threshold cost per cycle of belantamab.

The drug administration cost, the monitoring cost for post progression, and the cost of disease evaluation and laboratory testing were extracted from an economic evaluation of RRMM treatment in China (Table 1) [25].

The management cost for SAEs was calculated for each treatment arm. It was based on the proportions of patients experiencing each SAE, sourced from the updated results of the DREAMM-7 trial [16], and the estimated cost per SAE episode derived from published studies (ESM, Supplementary Table 2) [34,35,36,37,38,39,40].

2.5 Base Case Analysis

The threshold price of belantamab at which BVd would be considered cost effective relative to DVd was estimated using probabilistic sensitivity analysis (PSA) with 10,000 Monte Carlo simulations, simultaneously varying all model inputs over their uncertainty distributions. Unit costs of all the drugs, except belantamab, were obtained from the Beijing Sunshine Medical Procurement Platform, for which variance estimates were unavailable. Distributional parameters were also unavailable for treatment costs of SAEs. We therefore represented cost uncertainty using a gamma distribution centered on the point estimate with an assumed coefficient of variation of 20% (Table 1). Likewise, variance estimates were unavailable for the disutilities of some SAEs. We applied a gamma distribution to QALY decrement associated with SAEs, also with an assumed coefficient of variation of 20%. Other model inputs were assigned pre-specified distributions according to parameter type and the corresponding distributional parameters were obtained from published sources, as summarized in Table 1. The mean threshold price derived from PSA with its 95% credible interval (95% CrI) is reported. For model verification and transparency, the deterministic output from running the model with the point estimate of each input is also reported to facilitate comparison and to assess the degree of non-linearity in the model.

2.6 Sensitivity Analyses

To assess the robustness of the model results, probabilistic one-way sensitivity analyses were performed by varying each input individually across its 95% uncertainty interval. At each fixed value of an input, the conditional mean threshold price was calculated by PSA with 10,000 Monte Carlo simulations given uncertainty of all other inputs.

2.7 Scenario Analyses

Scenario analyses were conducted to assess the structural uncertainty associated with model specifications. We modeled scenarios by (i) limiting the time horizon to 25 years, (ii) applying other utility values for the progression-free state and PD state derived from another Chinese study on DVd treatment in RRMM [30], (iii) applying discount rates of 0% or 8%, (iv) applying lognormal distributions for PFS in both groups, and (v) replacing the GPM-clamped lognormal OS extrapolation with an excess hazard (EH) model for both treatment groups.

Given that higher WTP thresholds have been proposed for high-value innovative drugs, we further extended the analysis to a WTP threshold of up to US$73,610 per QALY (five times the 2025 Chinese GDP per capita) to estimate the acceptable price range of belantamab [30, 41].