Cost-effectiveness and budget impact of Luspatercept in Thai adults with transfusion-dependent beta-thalassaemia: a multi-state Markov model based on national claims data

ABSTRACT

Background

Luspatercept is a first-in-class erythroid maturation agent that reduces transfusion burden in adults with transfusion-dependent beta-thalassaemia (TDT). However, its cost-effectiveness has not been evaluated in middle-income settings. This study aimed to evaluate the cost-effectiveness and budget impact of Luspatercept for TDT in Thai adults.

Methods

A hybrid decision tree and lifetime Markov cohort model was developed to compare Luspatercept plus best supportive care (BSC) with BSC alone from a societal perspective. Age-specific 12-week transition probabilities across three transfusion burden states (low, medium, and high) and death were derived from a multi-state Markov survival model fitted to national administrative claims data. Treatment efficacy was based on the phase 3 BELIEVE trial. Three treatment eligibility scenarios were evaluated: universal (all patients), targeted (medium and high transfusion burden), and restricted (high transfusion burden only). Deterministic, probabilistic, threshold price, and budget impact analyses were performed.

Results

In the universal scenario, Luspatercept produced an incremental gain of 0.205 QALYs and an additional cost of 955,053 THB (29,048 US$) per patient, resulting in an incremental cost-effectiveness ratio (ICER) of 4,650,427 THB/QALY (141,442 US$/QALY). This value exceeds the Thai willingness-to-pay threshold of 160,000 THB/QALY by approximately 29-fold. Restricting eligibility to patients with higher transfusion burden did not improve cost-effectiveness; in the restricted scenario, Luspatercept was dominated. A price reduction of 90.5% would be required for Luspatercept to become cost-effective, which would also reduce the five-year cumulative budget impact from 2,505 million THB to 296 million THB.

Conclusion

Luspatercept is not cost-effective for TDT in Thailand at its current price across all scenarios evaluated. A substantial price reduction would be necessary, though this would bring the budget impact to a manageable level. These findings may help inform pricing and reimbursement decisions in Thailand and other middle-income countries.

Background

Beta-thalassaemia is an inherited condition characterised by impaired production of beta-globin chains, causing excess unpaired alpha-globin chains to accumulate. This imbalance disrupts normal red blood cell development, resulting in ineffective erythropoiesis, and chronic haemolytic anaemia (Kattamis et al., 2022; Taher et al., 2021). Patients with transfusion-dependent thalassaemia (TDT) require lifelong, regular red blood cell transfusions to maintain adequate haemoglobin levels (Kattamis et al., 2022; Taher et al., 2021). Although transfusion therapy is essential for survival, it leads to progressive iron accumulation. Without adequate chelation, this can result in cardiac, hepatic, endocrine, and renal complications, which are the main causes of morbidity and mortality in this population (Kattamis et al., 2022; Taher et al., 2021).

Advances in transfusion medicine, iron chelation, and supportive care have transformed beta-thalassaemia from a fatal childhood disease into a chronic condition with extended survival into adulthood. However, this improved longevity comes with accumulating comorbidities and persistent treatment burden (Betts et al., 2020; Kattamis et al., 2022). Globally, thalassaemia is most prevalent in East and Southeast Asia, where age-standardised prevalence rates exceed 40 per 100,000 population (Tuo et al., 2024). Thailand is among the most affected countries, with an estimated 30–40% of the population carrying at least one thalassaemia gene and approximately 23 per 100,000 individuals living with TDT (Health Intervention and Technology Assessment Program (HITAP), 2019; Paiboonsukwong et al., 2022). This represents a substantial clinical and economic burden on the Thai healthcare system.

Current treatment of TDT consists of best supportive care (BSC), including regular red blood cell transfusions and iron chelation therapy. Maintaining a pre-transfusion haemoglobin of 9–10.5 g/dL improves anaemia and suppresses ineffective erythropoiesis (Kattamis et al., 2022; Taher et al., 2021). Although this suppression is substantial, it is achieved indirectly and does not correct the underlying globin-chain imbalance. Complications of ineffective erythropoiesis can therefore persist when transfusions are delayed or suboptimal (Taher et al., 2021).

In Thailand, BSC is reimbursed under the Universal Coverage Scheme and remains the standard of care. Allogeneic haematopoietic stem cell transplantation is the only established curative option. Its use is limited by donor availability, patient age, and accumulated organ damage, and it is seldom undertaken in adults. The newer disease-directed therapies are not routinely available. Gene therapies such as Betibeglogene autotemcel and Exagamglogene autotemcel and the oral pyruvate kinase activator Mitapivat are not registered with the Thai Food and Drug Administration. The only remaining disease-directed option is Luspatercept, a first-in-class erythroid maturation agent that binds select ligands of the transforming growth factor-beta superfamily, promoting late-stage erythropoiesis and reducing transfusion requirements (Suragani et al., 2014). Evidence from the phase 3 BELIEVE trial showed that 77% of patients receiving Luspatercept achieved a ≥ 33% reduction in transfusion burden over any 12-week period, compared with 35% in the placebo group (Cappellini et al., 2025). These findings have led to regulatory approval of Luspatercept in several countries for treating anaemia in adults with TDT (Markham, 2020).

Despite its demonstrated clinical efficacy, the cost-effectiveness of Luspatercept in TDT remains uncertain, particularly in resource-constrained settings. The only published economic evaluation, conducted by the Canadian Agency for Drugs and Technologies in Health (CADTH), found that Luspatercept was not cost-effective at its submitted price (Canadian Agency for Drugs and Technologies in Health (CADTH), 2021). However, this analysis was conducted from a high-income country perspective and relied on expert-informed assumptions for disease progression rather than population-level data. No economic evaluation has been conducted from a middle-income country perspective, where willingness-to-pay (WTP) thresholds are lower and the disease burden is higher. This study therefore aimed to evaluate the cost-effectiveness and budget impact of Luspatercept in combination with BSC, compared with BSC alone, for the treatment of TDT in Thai adults.

Methods

Overall description

A hybrid decision tree and Markov cohort model was developed to project the long-term clinical and economic outcomes of Luspatercept in combination with BSC, compared with BSC alone, in adult patients with TDT. The model structure was adapted from a previously published economic evaluation prepared for the 2021 Canadian Agency for Drugs and Technologies in Health (CADTH) reimbursement review (Canadian Agency for Drugs and Technologies in Health (CADTH), 2021). This study followed the Consolidated Health Economic Evaluation Reporting Standards (CHEERS) 2022 statement (Husereau et al., 2022). Model input parameters are presented in Table 1.

Table 1.

Model input parameters and data sources.

Parameter	Base-case value	Source
Baseline cohort distribution of transfusion burden
LTB (≤ 5 units per 12 weeks)	0.515	NHSO database
MTB (> 5 to ≤ 7 units per 12 weeks)	0.266	NHSO database
HTB (> 7 units per 12 weeks)	0.218	NHSO database
Transition probabilities per 12-week cycle (age-dependent)
Transition probabilities were estimated from a multi-state Markov model (msm package in R) fitted to individual patient-level claims data from the NHSO database. Age was included as a covariate on all transition intensities. Full age-specific matrices (ages 18–100) are provided in the Supplemental Table 2. Illustrative matrices at ages 30, 50, and 70 are shown below.
Age 30
LTB → LTB, MTB, HTB, Death	0.749, 0.183, 0.052, 0.017	NHSO database
MTB → LTB, MTB, HTB, Death	0.342, 0.453, 0.189, 0.017	NHSO database
HTB → LTB, MTB, HTB, Death	0.121, 0.254, 0.607, 0.018	NHSO database
Age 50
LTB → LTB, MTB, HTB, Death	0.783, 0.137, 0.039, 0.041	NHSO database
MTB → LTB, MTB, HTB, Death	0.398, 0.381, 0.179, 0.041	NHSO database
HTB → LTB, MTB, HTB, Death	0.156, 0.208, 0.593, 0.043	NHSO database
Age 70
LTB → LTB, MTB, HTB, Death	0.776, 0.097, 0.027, 0.100	NHSO database
MTB → LTB, MTB, HTB, Death	0.442, 0.299, 0.158, 0.100	NHSO database
HTB → LTB, MTB, HTB, Death	0.207, 0.159, 0.534, 0.100	NHSO database
Response probability*
Luspatercept	0.772	Cappellini et al. (2025)
BSC	0.348	Cappellini et al. (2025)
Probability of complications per 12-week cycle
Cardiac complications
LTB	0.0033	NHSO database
MTB	0.0046	NHSO database
HTB	0.0072	NHSO database
Liver complications
LTB	0.0007	NHSO database
MTB	0.0014	NHSO database
HTB	0.0042	NHSO database
Endocrine complications
LTB	0.0072	NHSO database
MTB	0.0073	NHSO database
HTB	0.0086	NHSO database
Renal complications
LTB	0.0031	NHSO database
MTB	0.0018	NHSO database
HTB	0.0024	NHSO database
Probability of adverse drug reactions per 12-week cycle
Luspatercept + BSC
Upper respiratory tract infection	0.0546	Cappellini et al. (2025)
Back pain	0.0401	Cappellini et al. (2025)
Bone pain	0.0259	Cappellini et al. (2025)
BSC
Upper respiratory tract infection	0.0868	Cappellini et al. (2025)
Back pain	0.0662	Cappellini et al. (2025)
Bone pain	0.0151	Cappellini et al. (2025)
Direct medical costs per 12-week cycle (THB [US$])
OPD and IPD visits (excluding blood transfusion and iron chelation therapy)
LTB	11,435 [348]	Kongpakwattana et al. (2026) & NHSO database
MTB	15,509 [472]	Kongpakwattana et al. (2026) & NHSO database
HTB	24,552 [747]	Kongpakwattana et al. (2026) & NHSO database
Blood transfusion
LTB	3,753 [114]	Kongpakwattana et al. (2026) & NHSO database
MTB	7,101 [216]	Kongpakwattana et al. (2026) & NHSO database
HTB	21,466 [653]	Kongpakwattana et al. (2026) & NHSO database
Iron chelation therapy
LTB	3,721 [113]	Kongpakwattana et al. (2026) & NHSO database
MTB	4,762 [145]	Kongpakwattana et al. (2026) & NHSO database
HTB	6,855 [209]	Kongpakwattana et al. (2026) & NHSO database
Treatment
Luspatercept	94,160 [2,864]	Hospital’s procurement document
Iron overload complications
Cardiac complications	6,802 [207]	NHSO database
Liver complications	2,382 [73]	NHSO database
Endocrine complications	4,958 [151]	NHSO database
Renal complications	5,085 [155]	NHSO database
Adverse drug reactions
Upper respiratory tract infection	6,546 [199]	NHSO database
Back pain	5,200 [158]	NHSO database
Bone pain	6,243 [190]	NHSO database
Direct non-medical costs per 12-week cycle (THB [US$])
LTB	5,060 [154]	NHSO database
MTB	6,101 [186]	NHSO database
HTB	7,998 [243]	NHSO database
Health utility values
LTB	0.915	Kongpakwattana et al. (2026)
MTB	0.695	Kongpakwattana et al. (2026); Martin et al. (2023)
HTB	0.451	Kongpakwattana et al. (2026); Martin et al. (2023)
Disutility: 1 complication**	0.129	Shafie et al. (2021)
Disutility: 2 complications**	0.162	Shafie et al. (2021)
Disutility: 3 + complications**	0.250	Shafie et al. (2021)

BSC = best supportive care, HTB = high transfusion burden, ICT = iron chelation therapy, LTB = low transfusion burden, MTB = medium transfusion burden, NHSO = National Health Security Office, OPD = outpatient department, IPD = inpatient department, THB = Thai baht, US$ = United States dollar

* Response defined as ≥ 33% reduction in RBC transfusion burden from baseline during any 12-week interval.

** Disutility applied as absolute decrement from the transfusion burden state utility, based on cumulative number of active iron overload complications (cardiac, liver, endocrine, renal)

The model consisted of two consecutive phases (Figure 1). In the initial decision tree phase, patients entered the simulation at week 0 and were assigned to receive either Luspatercept plus BSC or BSC alone. Treatment response was evaluated in the Luspatercept group at week 48. Patients who achieved a response continued Luspatercept therapy in the subsequent Markov phase, whereas those who did not respond were switched to BSC alone.

Figure 1.

Hybrid decision tree and Markov model structure.

After response assessment, all patients moved into a lifetime Markov model. Health states were defined according to transfusion burden, categorised by the number of packed red blood cell (RBC) units received within each 12-week cycle: Low transfusion burden (LTB: ≤ 5 units), Medium transfusion burden (MTB: > 5 and ≤ 7 units), and High transfusion burden (HTB: > 7 units). An absorbing death state was also included.

Total costs, quality-adjusted life years (QALYs), and incremental cost-effectiveness ratios (ICERs) were estimated over a lifetime horizon. Consistent with the Thai Health Technology Assessment (HTA) guidelines (Health Intervention and Technology Assessment Program (HITAP), 2021), both costs and health outcomes were discounted at an annual rate of 3%. The analysis was conducted from a societal perspective.

Patient population and baseline characteristics

The simulated cohort consisted of Thai adults with TDT. Baseline characteristics, including age, sex, and the distribution of patients across transfusion burden distribution, were obtained from a national administrative claims database maintained by the National Health Security Office (NHSO) of Thailand. Details of the NHSO database processing are provided in the Supplemental Appendix. Briefly, the database included all healthcare encounters with a principal or secondary diagnosis of thalassaemia (ICD-10: D56.1) and at least one procedure code for blood transfusion (ICD-9-CM: 99.04). The study population was restricted to individuals aged 18 years or older who had at least 24 weeks of follow-up and no transfusion gap longer than 12 weeks. This corresponds closely to the best-performing administrative case definition evaluated by Snyder et al. (2026), which required at least two encounters with a thalassaemia code and a transfusion code on the same encounters and achieved a positive predictive value of 82–96% against clinician- and laboratory-confirmed diagnoses (Snyder et al., 2026). The additional restrictions applied here make our definition more specific for ongoing transfusion dependence. The mean age at model entry was 31.1 years. A body weight of 50 kg was assumed for Luspatercept dosing calculations.

Transition probabilities between transfusion burden states and mortality

Age-specific 12-week transition probabilities across the three transfusion burden states (LTB, MTB, HTB) and death were estimated using data from the NHSO database. A multi-state Markov survival model was fitted to derive these transitions. The analysis was conducted using the msm package in R, incorporating patient age as a time-varying covariate (see Supplemental Appendix for model specification and parameterisation details).

In the BSC arm, patients transitioned between states at each 12-week cycle according to these age-specific transition probability matrices. In the Luspatercept arm, responders who continued therapy were assumed to remain in their attained transfusion burden state until treatment discontinuation, while still being subject to the age-specific mortality risk of their current state. This reflects the pattern observed in BELIEVE, in which the reduction in transfusion burden was sustained without regression towards baseline: mean decreases from baseline were 4.8, 5.6, 6.2, and 6.4 RBC units during weeks 1–48, 49–96, 97–144, and 145–192, respectively, while the placebo group increased from baseline (Cappellini et al., 2025). These data span the three-year treatment duration modelled here, and holding responders in their attained state is conservative relative to the continued reduction observed after week 48. Non-responders followed BSC transition probabilities from the point of treatment failure onward.

Luspatercept-related clinical parameters

Treatment response was defined as a ≥ 33% reduction in RBC transfusion burden during any 12-week interval relative to the baseline period according to the BELIEVE trial (Cappellini et al., 2025). Over the full treatment period, 77% (173/224) of patients receiving Luspatercept met this criterion, compared with 35% (39/112) receiving placebo. Response rates were applied at week 48 using an intention-to-treat approach. The transfusion burden state attained by responders was derived from the observed distribution of RBC units in the NHSO cohort (n = 751) rather than assumed. A 33% reduction was applied to each patient's baseline number of units per 12-week cycle, and patients were then reclassified using the same state definitions. Responders in LTB remained in LTB, and all responders in MTB moved to LTB. Among responders in HTB, those receiving 8–10 units (75%) moved to MTB and those receiving more than 10 units (25%) remained in HTB. Responders were therefore distributed as 78.2% LTB, 16.4% MTB, and 5.5% HTB at week 48, while non-responders retained the baseline distribution. Response status was assessed once, at week 48, and was not reassessed thereafter. Responders were assumed to continue therapy until the end of the treatment duration, while non-responders switched to BSC and incurred no further Luspatercept costs.

Treatment duration was set at three years, corresponding to the median treatment exposure of 153.6 weeks reported in the BELIEVE trial (Cappellini et al., 2025). Following treatment discontinuation, a gradual waning of effect over two years was assumed (Cappellini et al., 2025). During this waning period, responders progressively transitioned from maintaining their attained state to following the age-specific transition probabilities estimated under BSC. After the waning period, all former responders followed BSC transition probabilities for the remainder of the time horizon.

Adverse drug reactions, including upper respiratory tract infection, back pain, and bone pain, were incorporated based on safety data from the BELIEVE trial (Cappellini et al., 2025).

Complications from iron overload

Chronic transfusion therapy leads to progressive iron accumulation and long-term organ damage. The model incorporated four categories of iron overload complications: cardiac, liver, endocrine, and renal events. These complications were modelled as cumulative and irreversible. Once a complication occurred, associated costs and utility decrements were applied from the cycle of onset until death or the end of the time horizon. First-onset incidence rates, stratified by transfusion burden state, and incremental per-cycle costs for each complication category were estimated from the NHSO database (see Supplemental Appendix for methodological details).

Cost parameters

All cost inputs were estimated from Thai sources. In accordance with Thai HTA guidelines (Health Intervention and Technology Assessment Program (HITAP), 2021), indirect costs were excluded because the use of QALYs already captures morbidity and mortality effects. Only direct medical and direct non-medical costs were included.

Direct medical costs comprised: (1) routine care costs (outpatient visits, inpatient admissions, blood transfusions, and iron chelation therapy), (2) Luspatercept acquisition price, (3) costs of iron overload complications, and (4) costs of managing adverse drug reactions. Direct non-medical costs included transportation, meals, accommodation, and caregiving time.

Routine care and direct non-medical costs, stratified by transfusion burden state, were estimated using predictive equations from a cross-sectional cost-of-illness study among Thai adults with TDT (Kongpakwattana et al., 2026) (see Supplemental Appendix for predictive equations details). Luspatercept acquisition price was obtained from a publicly available hospital procurement document (Siriraj Piyamaharajkarun Hospital, 2024), assuming a dose of 1.0 mg/kg subcutaneously every three weeks per the product label. All costs were expressed in 2025 Thai baht (THB) and United States dollar (US$), converted at 1 US$ = 32.8788 THB (Bank of Thailand, 2025).

Health utility data

Health utility values were derived from a cross-sectional study of Thai adults with TDT (Kongpakwattana et al., 2026), which employed the EQ-5D-5L instrument. Utility estimates did not differ significantly across transfusion burden categories, likely reflecting a ceiling effect in this chronically adapted population. Therefore, the LTB utility value was adopted as the baseline for all patients. To differentiate between transfusion burden states, utility ratios from Martin et al. (2023), an international time trade-off based valuation study, were applied to the Thai baseline value to derive MTB and HTB utilities. For iron overload complications, disutilities reported by Shafie et al. (2021) were subtracted from the transfusion burden state baseline utility at each cycle, with adjustments accounting for the proportion of patients experiencing multiple concurrent complications (see Supplemental Appendix for implementation details).

Analyses

Base-case analyses: three scenarios stratified by baseline transfusion burden

To explore how baseline disease severity affects cost-effectiveness, three treatment eligibility scenarios were evaluated: (1) a universal treatment scenario, reflecting the real-world distribution of transfusion burden (51.5% LTB, 26.6% MTB, 21.8% HTB) observed in the NHSO database, (2) a targeted treatment scenario, in which eligibility included patients with MTB (54.9%) and HTB (45.1%), and (3) a restricted treatment scenario, in which eligibility was limited to patients with HTB only. These scenarios were designed to inform policy decisions regarding the optimal target population for reimbursement.

Cost-effectiveness was assessed against the WTP threshold of 160,000 THB per QALY gained (approximately 4,866 US$ per QALY), as recommended by the Thai Health Economic Working Group (HEWG) for the National List of Essential Medicines (NLEM) (Thavorncharoensap et al., 2013).

Sensitivity analyses

Deterministic one-way sensitivity analyses (OWSA) were performed to assess robustness and identify parameters with the greatest influence on the ICER. Each parameter was varied individually across its plausible range (±20% or published confidence intervals), with results summarised in a tornado diagram.

A structural sensitivity analysis was also performed to examine the assumption that responders remain in their attained transfusion burden state during treatment. In this scenario, treated patients were subject to the same age-specific transition probabilities as the BSC arm, so that response determined the transfusion burden state attained at week 48 but conferred no protection against subsequent transitions. A second structural sensitivity analysis applied the observed Thai utility values directly (0.915, 0.932, and 0.849 for LTB, MTB, and HTB, respectively), without differentiation by the Martin et al. (2023) ratios, to examine the consequence of taking the absence of significant utility differences across transfusion burden states at face value.

Probabilistic sensitivity analyses (PSA) used second-order Monte Carlo simulation to jointly characterise parameter uncertainty. Beta distributions were assigned to probabilities and utility values, and gamma distributions to cost parameters. Results were presented as a cost-effectiveness acceptability curve (CEAC) and a scatterplot on the cost-effectiveness plane.

Threshold price analysis

A threshold price analysis was conducted to determine the maximum Luspatercept acquisition price at which the ICER would equal the Thai WTP threshold in the treatment scenario resulting in the most favourable ICER. The Luspatercept acquisition price was reduced until the ICER reached the threshold value, and the corresponding percentage reduction from the current price was reported.

Budget impact analysis

A budget impact analysis (BIA) estimated the financial consequences of introducing Luspatercept for TDT in Thailand over five years from the healthcare payer perspective, without discounting, following the ISPOR Principles of Good Practice (Sullivan et al., 2014). In contrast to the societal perspective of the cost-effectiveness analysis, the BIA included only direct medical costs borne by the payer because its purpose is to inform affordability from the budget holder's standpoint. The eligible population was calculated by applying a TDT prevalence estimate (23 per 100,000 population) (Health Intervention and Technology Assessment Program (HITAP), 2019) to the Thai population (National Statistical Office. Ministry of Digital Economy and Society, 2025), then restricting to adults (78.8%) (National Statistical Office. Ministry of Digital Economy and Society, 2025) with at least one transfusion-related complication (35.7%) (Teawtrakul et al., 2022). This restriction was applied as a conservative assumption about the population likely to be prioritised for reimbursement rather than as a clinical eligibility criterion. Because patients without a history of complications may also elect treatment, a sensitivity analysis was conducted in which the restriction was removed and all Thai adults with TDT were considered eligible. The analysis compared the current scenario (all patients on BSC) with a new scenario in which eligible patients progressively adopted Luspatercept plus BSC, assuming conservative market uptake increasing from 5% to 25% over five years. Annual per-patient costs for each arm were derived from the hybrid decision tree and Markov model. Results were reported as annual and cumulative five-year incremental budget impact in THB and US$.

Model validation

Internal validation assessed the goodness of fit of the multi-state model, from which transition probabilities were derived. Observed and model-predicted state prevalences were compared over the follow-up period, marginalised over the observed covariate distribution in the NHSO cohort, for all four states (LTB, MTB, HTB, and death). External validation compared the Markov model's predicted survival curve against published Kaplan-Meier estimates from an independent Thai thalassaemia registry (n = 2,570) (Viprakasit & Jansutjawan, 2015), after conditioning registry survival on being alive at the model entry age of 31.1 years to ensure comparability. This comparison was undertaken specifically to assess whether combining beta-thalassaemia subtypes within a single cohort was appropriate, given that the NHSO database does not distinguish them. Details of internal and external model validation are provided in the Supplemental Appendix.

Results

Base-case cost-effectiveness analysis

The base-case results for all three treatment eligibility scenarios are presented in Table 2. In the universal treatment scenario, Luspatercept plus BSC resulted in total discounted costs of 2,219,288 THB (67,499 US$) and 6.075 QALYs per patient, compared with 1,264,235 THB (38,451 US$) and 5.869 QALYs for BSC alone. The resulting ICER was 4,650,427 THB per QALY gained (141,442 US$/QALY).

Table 2.

Base-case cost-effectiveness results by transfusion burden distribution scenario.

Base-case result per patient	Luspatercept + BSC	BSC
1. Universal treatment scenario (all patients: LTB, MTB, and HTB)
Total costs (THB) [US$]	2,219,288 [67,499]	1,264,235 [38,451]
Quality of life (QALYs)	6.075	5.869
ICER (THB/QALY) [US$/QALY]	4,650,427 [141,442]
2. Targeted treatment scenario (MTB and HTB patients only)
Total costs (THB) [US$]	2,305,673 [70,126]	1,326,585 [40,348]
Quality of life (QALYs)	5.750	5.650
ICER (THB/QALY) [US$/QALY]	9,777,840 [297,390]
3. Restricted treatment scenario (HTB patients only)
Total costs (THB) [US$]	2,457,018 [74,730]	1,408,599 [42,842]
Quality of life (QALYs)	5.255	5.454
ICER (THB/QALY) [US$/QALY]	Dominated*

BSC = best supportive care, ICER = incremental cost-effectiveness ratio, QALYs = quality-adjusted life years, THB = Thai baht, US$ = United States dollar

* Luspatercept plus BSC incurred higher costs and produced fewer QALYs than BSC alone (incremental costs: +1,048,419 THB, incremental QALYs: −0.199) and is therefore dominated. The corresponding ICER of −5,269,828 THB/QALY (−160,280 US$/QALY) is reported for completeness only. The negative sign arises from negative incremental QALYs with positive incremental costs (northwest quadrant of the cost-effectiveness plane) and should not be compared with the willingness-to-pay threshold.

In the targeted treatment scenario, restricting eligibility to patients with MTB and HTB worsened the ICER to 9,777,840 THB per QALY gained (297,390 US$/QALY), with incremental QALYs of only 0.100. In the restricted treatment scenario, limiting eligibility to HTB patients only, Luspatercept was dominated, producing higher costs and fewer QALYs than BSC alone (incremental QALYs: −0.199; ICER: −5,269,828 THB/QALY).

Deterministic sensitivity analysis

The tornado diagram from the OWSA of the universal treatment scenario is shown in Figure 2. The ICER was most sensitive to health utility values. The LTB utility was the single most influential parameter, with the ICER ranging from approximately 3.2 million to 8.7 million THB/QALY across its plausible range. Luspatercept acquisition price was the second most influential parameter, followed by HTB utility, MTB utility, response probability, and gradual fade of effectiveness (waning duration). No parameter variation within its plausible range brought the ICER below the Thai WTP threshold.

Figure 2.

Tornado diagram of one-way deterministic sensitivity analysis of the universal treatment scenario.

When treated patients were subject to the same transition probabilities as the BSC arm, the ICER in the universal scenario rose from 4,650,427 to 49,813,232 THB/QALY, reflecting a fall in incremental QALYs from 0.205 to 0.020 with incremental costs largely unchanged. The targeted and restricted scenarios resulted in 23,998,529 and 23,578,814 THB/QALY, respectively. The ordering of the eligibility scenarios inverted, and Luspatercept was no longer dominated in the restricted scenario. Nonetheless, all scenarios remained far above the WTP threshold.

When the observed Thai utility values were applied directly, the ICER was 26,168,583 THB/QALY in the universal scenario, 30,021,810 THB/QALY in the targeted scenario, and 56,366,642 THB/QALY in the restricted scenario. These changes reflect the much smaller utility differences between transfusion burden states. Still, all scenarios remained far above the WTP threshold.

Probabilistic sensitivity analysis

On the cost-effectiveness plane (Figure 3), all 1,000 iterations fell well above the WTP threshold line, with the scatter concentrated in the northeast quadrant. Four of the 1,000 iterations (0.4%) produced negative incremental QALYs. As none produced cost savings, these iterations fell in the northwest quadrant. The probability that Luspatercept produces net health loss under parameter uncertainty is therefore low, at approximately 0.4%. The CEAC (Figure 4) showed 0% probability of Luspatercept being cost-effective at the Thai WTP threshold. The probability of cost-effectiveness reached 50% only at a WTP of approximately 4–5 million THB/QALY.

Figure 3.

Cost-effectiveness plane from probabilistic sensitivity analysis of the universal treatment scenario (1,000 iterations).

Figure 4.

Cost-effectiveness acceptability curve from probabilistic sensitivity analysis of the universal treatment scenario.

Threshold price analysis

A 90.5% reduction in Luspatercept acquisition price (from 94,160 THB to approximately 8,942 THB per 12-week cycle, or from 2,864 US$ to 272 US$) would be required for Luspatercept to achieve cost-effectiveness at the Thai WTP threshold in the universal treatment scenario. This analysis was not performed for the other scenarios. In the restricted scenario, Luspatercept was dominated in the base case, and as a price reduction cannot alter health outcomes, no acquisition price would achieve cost-effectiveness. In the targeted scenario, the less favourable ICER implies that an even larger price reduction would be required. Hence, the universal threshold price therefore represents the upper bound relevant to price negotiation.

Budget impact analysis

The budget impact analysis results are presented in Table 3. Under the base-case scenario with full Luspatercept acquisition price, the five-year cumulative incremental budget impact was 2,505 million THB (76.2 million US$), representing a 29.9% increase over the current BSC-only budget. At the threshold price (90.5% reduction in Luspatercept acquisition price), the five-year cumulative incremental budget impact decreased to 296 million THB (9.0 million US$), a 3.5% increase over the current budget. With the complication restriction removed, the eligible population increased from 4,267 to 11,953 patients. At the full acquisition price, the five-year cumulative incremental budget impact increased from 2,505 to 7,011 million THB (213.2 million US$), an 83.8% increase over the current budget. At the threshold price it increased from 296 to 829 million THB (25.2 million US$), a 9.9% increase.

Table 3.

Budget impact analysis over a 5-year time horizon (healthcare perspective).

Budget impact analysis	Year 1	Year 2	Year 3	Year 4	Year 5	5-Year Total
Target population
Estimated TDT patients in Thailand (n)	15,169	15,169	15,169	15,169	15,169	–
Estimated eligible patients for Luspatercept (adult patients with at least one transfusion-related complication)	4,267	4,267	4,267	4,267	4,267	–
Luspatercept uptake rate*	5%	10%	15%	20%	25%	–
Patients initiating Luspatercept (n)	214	427	641	854	1,067	–
Cumulative patients ever treated with Luspatercept (n)	214	641	1,282	2,136	3,203	–
Patients on BSC only (n)	14,995	14,528	13,887	13,033	11,966	–
Base-case budget impact analysis (Universal treatment scenario with full Luspatercept acquisition price)
Total costs: BSC only (million THB) [million US$]	1,942.2 [59.1]	1,667.0 [50.7]	1,965.5** [59.8]	1,449.1 [44.1]	1,340.4 [40.8]	8,364.2 [254.4]
Total costs: Luspatercept + BSC (million THB) [million US$]	2,038.0 [62.0]	1,922.0 [58.5]	2,418.6 [73.6]	2,174.9 [66.1]	2,315.4 [70.4]	10,868.9 [330.6]
Incremental budget impact (million THB) [million US$]	95.9 [2.9]	255.0 [7.8]	453.1 [13.8]	725.8 [22.1]	975.0 [29.7]	2,504.7 [76.2]
% increase over current budget	4.9%	15.3%	23.1%	50.1%	72.7%	29.9%
Threshold price analysis (Universal treatment scenario with 90.5% reduction in Luspatercept acquisition price)
Total costs: BSC only (million THB) [million US$]	1,942.2 [59.1]	1,667.0 [50.7]	1,965.5** [59.8]	1,449.1 [44.1]	1,340.4 [40.8]	8,364.2 [254.4]
Total costs: Luspatercept + BSC (million THB) [million US$]	1,951.0 [59.3]	1,694.6 [51.5]	1,989.0 [60.5]	1,543.5 [46.9]	1,482.2 [45.1]	8,660.2 [263.4]
Incremental budget impact (million THB) [million US$]	8.8 [0.3]	27.5 [0.8]	23.4 [0.7]	94.4 [2.9]	141.8 [4.3]	296.0 [9.0]
% increase over current budget	0.5%	1.7%	1.2%	6.5%	10.6%	3.5%

BSC = best supportive care, TDT = transfusion-dependent beta-thalassaemia, THB = Thai Baht, US$ = United States dollar

* Conservative market uptake assumes gradual adoption at 5% annual increments.

** Annual costs were aggregated from 12-week Markov cycles. Because a year does not contain a whole number of cycles, Year 3 spans five cycles (60 weeks) whereas adjacent years span four (48 weeks), accounting for its higher value. However, the 5-year totals are unaffected.

Discussion

This study presents the first cost-effectiveness and budget impact analysis of Luspatercept for transfusion-dependent beta-thalassaemia conducted from a Thai perspective and, to our knowledge, the first in any middle-income country. The analysis used a hybrid decision tree and lifetime Markov model populated with age-specific transition probabilities derived from a national administrative claims database, providing locally calibrated natural history estimates.

Luspatercept was not cost-effective at its current acquisition price in any of the three eligibility treatment scenarios evaluated. In the universal treatment scenario, the ICER of 4,650,427 THB/QALY exceeded the Thai WTP threshold by approximately 29-fold. The modest incremental QALY gain of 0.205 reflects the mechanism of action of Luspatercept, which reduces transfusion burden but does not cure TDT or alter mortality. This finding aligns with the conclusion of the CADTH reimbursement review, which also found Luspatercept not cost-effective (695,395 CA$/QALY, against the WTP threshold of 50,000 CA$/QALY) without 85% price reduction (Canadian Agency for Drugs and Technologies in Health (CADTH), 2021). The Thai context amplifies this conclusion given the lower WTP threshold relative to high-income settings.

The finding that narrowing eligibility to patients with a higher disease burden reduced, rather than improved, cost-effectiveness deserves further discussion. In the model, Luspatercept responders are assumed to maintain their attained transfusion burden state during treatment, which means that treated LTB patients are protected from transitioning to lower-utility states over time. Because LTB patients have the highest baseline utility (0.915) and thus the largest potential utility loss from disease progression, preventing their deterioration generates substantial QALY gains. When these patients are excluded from the targeted scenario, this protective effect is lost, and the analysis is left with only the more modest gains achievable in MTB and HTB patients, where the utility difference between adjacent states is smaller. In the restricted HTB-only scenario, Luspatercept was dominated because the response definition (≥ 33% reduction in transfusion burden) moves HTB responders to MTB rather than to LTB, resulting in a modest utility gain (0.451–0.695). This effect is further influenced by the age-specific transition probabilities from the NHSO data, which show a general shift towards lower transfusion burden over time under BSC. As a result, some untreated patients may reach LTB and accumulate greater utility gains than responders maintained in the MTB state during treatment. In addition, approximately 23% of patients do not respond but still incur treatment costs, contributing to negative incremental QALYs overall. This ordering, however, is contingent on the maintained-state assumption. When treated patients were instead subject to the same transition probabilities as the BSC arm, the ordering inverted and the restricted scenario became the most favourable. The comparison across eligibility scenarios therefore does not by itself support targeting Luspatercept to any particular transfusion burden group. What is robust to this structural choice is that no scenario approaches the Thai WTP threshold at the current price.

Sensitivity analyses supported the stability of the cost-effectiveness findings. All 1,000 PSA iterations fell above the WTP threshold, and the CEAC showed no probability of cost-effectiveness at 160,000 THB/QALY. The deterministic sensitivity analysis identified health utility values, particularly the magnitude of the differences between transfusion burden states, as the dominant driver of results. The counterintuitive direction of influence for HTB and MTB utilities, where increasing either value worsened the ICER, can be explained by the model’s underlying structure. Luspatercept generates QALY gains primarily by protecting patients from transitioning to lower-utility states. When the utility gap between states is reduced, the protective effect diminishes, thereby lowering the incremental QALY gain. This structural feature suggests that the precision of utility differentiation across transfusion burden categories is the most critical evidence gap influencing the cost-effectiveness outcome. Thai EQ-5D-5L data showed compressed utility values across transfusion burden states, which may reflect patient adaptation to chronic disease and potential ceiling effects of the EQ-5D-5L instrument in this population. The structural sensitivity analysis applying these compressed values directly raised the ICER approximately six-fold in the universal scenario, confirming that the utility gap between states is the principal source of Luspatercept's QALY gain. Because the differentiated base case gives Luspatercept its most favourable structure, the conclusion that it is not cost-effective at the current price is conservative and holds under both utility structures. Direct valuation methods applied to condition-specific health state descriptions may better capture clinically meaningful utility differences than generic preference-based instruments in this context.

The threshold price analysis suggested that a 90.5% reduction in Luspatercept acquisition price would be required to achieve cost-effectiveness in the universal scenario. While substantial, such reductions are not unprecedented in Thai pharmaceutical negotiations, particularly for high-cost specialty drugs. At the threshold price, the five-year cumulative budget impact decreased from 2,505 million THB to 296 million THB, representing only a 3.5% increase compared with current expenditure levels. This level of impact is likely to be manageable within existing reimbursement structures. However, this conclusion depends on how many patients are considered eligible. In a sensitivity analysis that removed the requirement for a history of transfusion-related complications, the eligible population nearly tripled and the budget impact rose in proportion, reaching 9.9% of current expenditure levels even at the threshold price. The true eligible population is likely to fall between these bounds, and the budget impact is therefore best interpreted as a range rather than a point estimate. Taken together, these findings suggest that price negotiation alone may not fully resolve the affordability question, and that a managed entry agreement (MEA) combining a negotiated price with explicit eligibility criteria or a volume cap would offer the Thai healthcare system greater budgetary predictability.

This study has several strengths. First, age-specific transition probabilities were derived from a national claims database using a multi-state Markov survival model, ensuring that the natural history estimates are firmly grounded in real-world data across the entire adult population. This approach reduces reliance on expert opinion or extrapolation from short-duration clinical trial data. Second, the model was validated through both internal goodness-of-fit assessment and external comparison with survival data from an independent Thai thalassaemia registry, which enhances the credibility of the model’s predictions. Third, the evaluation of multiple treatment eligibility scenarios allows the findings to inform a range of potential reimbursement policy options. Fourth, cost inputs were comprehensively derived from Thai data, including the use of predictive cost equations derived from a cross-sectional cost-of-illness study (Kongpakwattana et al., 2026), further supporting the contextual relevance of the analysis.

Several limitations should be considered. First, the NHSO database does not differentiate between beta-thalassaemia subtypes, meaning that the study cohort includes a combination of beta-thalassaemia major and beta-thalassaemia/haemoglobin E patients, who may follow distinct clinical pathways. Case identification also relied on an ICD-based algorithm that has not been formally validated against clinical records in Thailand. The algorithm nevertheless mirrors the highest-performing definition reported by Snyder et al. (2026) and applies additional restrictions. External validation was therefore undertaken specifically as a check on this limitation, and demonstrated that model-predicted survival fell between the registry curves for beta-thalassaemia major and beta-thalassaemia/haemoglobin E, as expected for a mixed cohort. Second, treatment efficacy for Luspatercept was based on data from the BELIEVE trial, which primarily involved non-Thai participants, and therefore the applicability of these response rates to the Thai population remains uncertain. The same response probability was also applied to patients in all three transfusion burden states. Response in BELIEVE was defined as a ≥ 33% reduction relative to each patient's own baseline transfusion burden during any rolling 12-week interval over the entire treatment period, and the trial reported response by baseline transfusion burden using the same cut-points as the health states in this model, at 66%, 84%, and 76% for the low, medium, and high burden groups against 77% overall. The trial did not report a statistical comparison across these subgroups, and the pooled rate was therefore retained. Although the probability of response was applied uniformly, the effect of responding was not, because the attained transfusion burden state was derived from the observed distribution of units in the Thai cohort. Nonetheless, whether response differs by transfusion burden in routine practice also remains uncertain. The assumption that responders maintain their attained transfusion burden state is supported by BELIEVE data through week 192, which span the modelled treatment duration, but the relative ranking of the eligibility scenarios is sensitive to it, as shown in the structural sensitivity analysis. Third, health utility values for MTB and HTB were derived by applying utility ratios from an international study to a Thai baseline value. This approach assumes proportional utility decrements are transferable across settings and may not fully represent Thai patient preferences. However, the structural sensitivity analysis using the observed Thai utilities directly produced higher ICERs in all three scenarios, indicating that the conclusion does not depend on the international ratios. Fourth, the model extrapolates to a lifetime horizon from an observation window of approximately ten years, and internal validation showed that model-predicted cumulative mortality fell progressively below the observed count after the first two to three years of follow-up (Supplemental Fig. S3). This divergence is consistent with unobserved heterogeneity and selective retention in the claims cohort, meaning the sparsely observed tail is itself an uncertain calibration target. Therefore, blending the extrapolation with the observed curve was not pursued, and the external registry comparison, in which predicted survival fell between the subtype-specific curves over the extrapolated horizon (Supplemental Fig. S4), was relied upon instead. Nevertheless, to the extent that late mortality is underestimated, survival is overestimated equally in both arms. Because the incremental benefit of Luspatercept accrues almost entirely during the treatment and waning period while incremental costs are concentrated in the treatment years, any resulting bias would inflate incremental QALYs and understate the ICER. Correcting it would therefore move the ICERs further above the WTP threshold and would not alter the conclusion. Finally, the model assumed a fixed three-year treatment duration based on the median exposure observed in the BELIEVE trial, whereas in real-world practice, treatment continuation is more likely to depend on individual patient response and clinical judgment. Relatedly, response status was determined by a single week-48 assessment and was not reassessed thereafter. The influence of this classification is confined to the treatment and waning periods, after which all patients follow BSC transition probabilities. The durability of response observed in BELIEVE through week 192 also suggests that reassessment would rarely have altered a patient's status. Where the simplification does introduce error, its direction favours Luspatercept, as late responders are credited with benefit from week 48 onward and responders are assumed never to lose response during treatment. The negative cost-effectiveness conclusion is therefore conservative with respect to this structural choice.

Conclusion

Luspatercept in combination with BSC is not cost-effective compared with BSC alone for the treatment of TDT in Thai adults at its current acquisition price, with the ICER exceeding the Thai WTP threshold by around 29-fold. This finding is consistent across all treatment eligibility scenarios, as well as in sensitivity and probabilistic analyses. An approximate 90.5% reduction in the acquisition price would be necessary for the intervention to become cost-effective, which would also lower the five-year budget impact to a more manageable level. These results may support pricing negotiations and reimbursement considerations for Luspatercept in Thailand, and may also be relevant to other middle-income countries facing similar healthcare financing constraints.

Funding Statement

This research project is supported by grants for development of a new faculty staff, Ratchadaphiseksomphot Fund, Chulalongkorn University.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.

Supplemental Material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/20523211.2026.2735077.