Study overview

This study adopts a healthcare system perspective to evaluate the cost-effectiveness of allo-HSCT procedures in AL patients. The comparative analysis focuses on two clinically established blood cell separators: Spectra Optia (intervention) versus COM.TEC (control), with particular emphasis on CE as the primary determinant of CD34+ cell yield. Building upon established post-transplant Markov models for AL (Reference Chen, Chen, He, Guo, Liu and Zheng20–Reference Maziarz, Devine and Garrison21), we developed a decision–Markov model using Microsoft Excel 2016 to simulate longitudinal clinical trajectories, which consists of two distinct components:

Decision tree stage: Focused on the short term (within 30 days post-transplant), simulating hematopoietic stem cell engraftment. It categorized patients into successful implantation or implantation failure based on neutrophil and platelet recovery.

Markov model stage: Simulated long-term post–implantation trajectories over 5 years with monthly cycles, incorporating five health states: (i) implantation failure; (ii) post–implantation relapse; and (iii) graft-versus-host disease (GVHD), including acute GVHD (aGVHD, occurring within 100 days post-transplantation) and chronic GVHD (cGVHD, occurring after 100 days post-transplantation); (iv) GVHD and relapse-free survival (GRFS); and (v) an absorbing state (death). Values of transition probabilities were derived from the literature and remained constant across cycles.

Key model parameters include resource utilization during peri-implantation care, long-term complication management costs, health state utilities, and survival probabilities. All data were obtained from publicly published clinical studies and national health economic reports, which exempted this study from requiring approval from an institutional review board.

Our research process is as follows: (i) Through probabilistic simulation modeling, we generated CD34+ cell yield distributions for 10,000 virtual patients by incorporating variability in critical biological parameters including donor and patient characteristics, mobilization efficiency, and device CE. (ii) To enhance clinical relevance, we established two secondary collection scenarios – clinically acceptable minimum thresholds versus optimal therapeutic targets (fewer collection times versus better patient prognosis) – to reflect real-world clinical decision-making processes. (iii) To assess comparative effectiveness and cost-effectiveness, we employed a decision tree–Markov model to simulate long- and short-term prognosis in 10,000 patients by varying CD34+ cell thresholds, with differential engraftment kinetics as the basis for evaluating comparative effectiveness. Meanwhile, the economic evaluation within the model incorporated both short-term in-hospital implantation treatment costs and long-term complication management costs to compare economic differences between the two systems.

CD34+ cells and scenario setting based on collection efficiency

The model input parameter is the number of CD34+ cells infused per patient, measured in cells/kg body weight. Based on prediction algorithm research (Reference Bojanic, Besson, Vidovic and Cepulic22), we developed a stochastic simulation model to calculate the counts of CD34+ cells obtained by two apheresis systems (Spectra Optia and COM.TEC) in 10,000 simulated AL patient–donor pairs. The outcome was defined as the CD34+ cell yield per kilogram of recipient body weight, with thresholds set according to clinical guidelines: optimal requirement (≥4 × 10⁶ CD34+ cells/kg) and minimum requirement (≥2 × 10⁶ CD34+ cells/kg). The CD34+ yield was calculated according to the following function:

$$ {\displaystyle \begin{array}{c}\mathrm{CD}34+\mathrm{yield}\left(\mathrm{cells}/\mathrm{kg}\right)=\\ {}\frac{\mathrm{CE}\left(\%\right)\times \mathrm{Pre}-\mathrm{CD}34+\mathrm{count}\left(\frac{\mathrm{cells}}{\unicode{x03BC} \mathrm{L}}\right)\times \mathrm{Processed}\ \mathrm{blood}\ \mathrm{volume}\left(\mathrm{mL}\right)}{\mathrm{Recipient}\ \mathrm{weight}\left(\mathrm{kg}\right)}\end{array}} $$

The CE for Spectra Optia is 54.6 percent and 50.1 percent for COM.TEC, derived from a registry of healthy donors at a single tertiary medical center in China (Reference Jiang, Xiang, Wei, Gao and Liu23).

The pre-CD34+ cell counts for the 10,000 simulated donors were generated using SPSS version 24.0 (IBM Corp., Armonk, NY) to fit a normal distribution (μ = 34.10/μL and σ = 18.98/μL), with parameters derived from a prediction algorithm study (Reference Leberfinger, Badman, Roig and Loos24). Parameter estimation was performed via maximum-likelihood estimation, and distribution normality was confirmed through the Kolmogorov–Smirnov testing (p > 0.05). All 10,000 donors included in the simulation met the clinical requirement for collection eligibility, with precollection peripheral blood CD34+ cell counts ≥20 cells/μL (25).

Recipient weights were stratified into six categories (35–45 kg, 45–55 kg, 55–65 kg, 65–75 kg, 75–85 kg, and 85–95 kg) based on the normal distribution of body weights in China (Reference Yuejiao26). Using standardized normal distribution intervals (z-scores), the probabilities for each weight category were calculated as 4.2 percent, 15.6 percent, 29.7 percent, 29.7 percent, 15.6 percent, and 4.2 percent, respectively.

Processed blood volume: Chinese Technical Management Specification for Nonrelated Hematopoietic Stem Cell Collection Technology Management Standards (No. 253,2006) stipulates that the processed blood volume per time for allo-HSCT cell collection should not exceed 15,000 mL, with a maximum of twice the collection (27). A processed blood volume of 6,000 mL is considered safe for most donors. Based on these guidelines, we hypothesize each donor starts with 6,000 mL. The volume then increases incrementally by 500 mL until reaching the upper limit of 15,000 mL. If a single collection reaches the maximum processed blood volume but fails to meet the required optimal cell yield (≥4 × 10⁶ cells/kg), a secondary collection will be performed. The CD34+ cell counts from the second collection are assumed to demonstrate no significant difference compared to the first. We assume that up to two collections can meet the minimum transfusion requirement of ≥2 × 10⁶ CD34+ cells/kg.

Guided by the CD34+ cell thresholds (minimum: ≥2 × 10⁶ cells/kg; optimal: ≥4 × 10⁶ cells/kg) established in HSCT guidelines, we categorized single-collection outcomes into three groups: <2 × 10⁶ cells/kg, 2–4 × 10⁶ cells/kg, and ≥ 4 × 10⁶ cells/kg. In consideration of transplantation efficacy (permitting lower cellular thresholds), and institutional resource allocation priorities, two collection scenarios were modeled.

Scenario 1 (lower resource consumption and donor impact): Adopted secondary collection only when processed blood volume reached 15,000 mL and CD34+ yield <2 × 10⁶ cells/kg, with reclassified yields upgraded to 2–4 × 10⁶ cells/kg.

Scenario 2 (clinical efficacy priority): Based on scenario 1, additionally triggered secondary collection for CD34+ yield 2–4 × 10⁶ cells/kg, with reclassified yields upgraded to ≥4 × 10⁶ cells/kg.

Model structure

This model consisted of a decision tree model and a Markov model. The model structure is shown in Figure 1.

Figure 1.

Decision tree–Markov model structure. N, neutrophils; P, platelets; GVHD, graft-versus-host disease; GRFS, GVHD-free and relapse-free survival.

Decision tree model structure

It takes about 14–21 days for hematopoietic stem cells to generate new blood cells, and most patients will complete the implantation within 30 days, and the probability of successful implantation after more than 30 days is very low (28). Clinically, 0–30 days after HSCT is set as the implantation period of blood cells, and the whole procedure to successful implantation needs to be performed in the transplant warehouse with drug use and medical care. Patients will undergo successful implantation if both the neutrophil recovery and the platelet recovery meet the required lines (criteria for successful implantation were neutrophil counts >0.5 × 10⁹/L for 3 consecutive days and platelets >20 × 10⁹/L for 3 consecutive days and without platelet transfusion for 7 consecutive days), and otherwise, patients will experience implantation failure. The decision tree model was mostly used for the evaluation of short-time study, and this study applied it to the stage of short-term patient implantation, namely, 30 days after allo-HSCT.

Markov model structure

Following successful implantation, the most significant adverse events impacting patients’ quality of survival are disease relapse and GVHD (Reference Gratwohl, Brand and Frassoni29). The aGVHD has a rapid onset and high mortality (2-year OS 60.7 percent), whereas cGVHD responds well to treatment (5-year OS 72.5 percent) (Reference Xing, Qian and Zhao30–Reference Dongyang31). The model defines that aGVHD occurs within the first three cycles post-transplantation (approximately 100 days), while cGVHD occurs between 3 and 60 cycles post-transplantation. Achieving a 5-year period free of both relapse and GVHD post-transplantation indicates clinical stability, with substantially reduced risks of subsequent adverse events and mortality. We employed a Markov model to evaluate the 5-year health status of AL patients after allo-HSCT, utilizing a 5-year time horizon with 1-month cycle intervals. Within the 5-year simulation framework, patients with successful implantation initially entered the GRFS state, with transition probabilities to relapse or GVHD stratified by CD34+ cell dose categories (2–4 × 10⁶ cells/kg and ≥ 4 × 10⁶ cells/kg). Patients experiencing graft failure remained exclusively in the terminal state of death without further disease progression.

Model parameters

The model parameters in this study were classified as transition probabilities, cost, and utility (Table 1).

Table 1.

Model parameters

Transition probabilities

We divided the final CD34 + cells into two groups: 2–4 × 10⁶ cells/kg and ≥ 4 × 10⁶ cells/kg. We obtained the 30-day engraftment success rate of neutrophils and platelets in two groups of HSCT patients from a retrospective observational study (Reference Pulsipher, Chitphakdithai and Logan32), which was used in the short stage of model surgery to simulate patient engraftment success. In the long-term simulation at 5 years after surgery, transition probabilities between Markov states (GRFS, relapse, GVHD, implantation failure, and death) were derived from clinically reported event progression rates (Reference Dongyang31;Reference Remberger, Gronvold and Ali33–Reference Luxin, Yamin and Jimin36) and assumed constant per cycle. These probabilities were converted using the Weibull function: $ \mathrm{P}\left(\mathrm{t}\right)=1-{\left(1-\mathrm{P}\left(\unicode{x03B3} \right)\right)}^{\frac{1}{\unicode{x03B3}}} $ . Here, P(t) represents the transition probability within each cycle, P( $ \unicode{x03B3} $ ) denotes the incidence rate of the event, and $ \unicode{x03B3} $ stands for the total number of cycles.

Cost

The cost parameters of this study covered both short-term (0–30-day) HSCT cost and long-term treatment cost of each health state, with all monetary outcomes standardized to 2025 US dollars (USD). The costs were discounted at an annual rate of 5 percent (Reference Guoen37).

Short-term costs were derived through clinical workflow mapping of critical transplantation phases (Supplementary file 1), identifying key cost drivers including surgery, medical consumables, diagnostic procedures, and supportive therapies. HSCT cost parameters were derived from the average medical service fees of four representative provinces ranked 1st, 16th, 17th, and 34th in per capita gross domestic product (GDP) among China’s 34 provincial-level administrative regions (2024 pricing schedules) (38), ensuring geographical and socioeconomic representativeness in cost estimation. The specific cost of each province is shown in Supplementary file 2. The relevant consumables data come from the professional information service platform of the Chinese pharmaceutical industry. Preoperatively, typing, physical examination (once), and flow cytometry (twice) were performed for comorbidities. For donors, mobilizing agents (five doses) were administered. For allo-HSCT, collection was conducted 1–2 times according to the collection target. The patient’s bed fee and anticoagulant cost varied with the number of collections. Postoperatively, we set the inpatient days as 16 days in the successful implantation group and 30 days for patients in the failure group (Reference Zaucha, Gooley and Bensinger16). The consumables, testing, care, and average daily medication costs vary with hospitalization length. Due to the complexity of actual drugs and cost calculation in the warehouse, we investigated the average daily drug cost of patients in 13 large transplant centers in China to calculate the mean value (Supplementary file 3).

Long-term state-specific treatment costs were derived from internationally published economic burden studies of AL patients (Reference Pandya, Chen and Medeiros39–Reference Michonneau, Quignot and Jiang41). The economic outcomes were standardized through a two-step conversion protocol: non-US economic outcomes adjustment to USD using World Bank purchasing power parity (PPP) conversion factors, followed by healthcare-specific PPP conversion between China and the United States to generate Chinese yuan renminbi (CNY)-denominated cost estimates (Reference Changjiang42). Per-cycle treatment costs were calculated based on reported follow-up intervals and applied to corresponding Markov health states through model transitions.

Utility

Utilities were obtained from the published literature (Reference Maziarz, Devine and Garrison21). Implantation failure utility values were equated to relapse state utilities based on comparable health-related quality of life impairment and treatment regime.

Outcome measures

We report the results in terms of both effectiveness and cost-effectiveness. Effectiveness was measured by collection frequency, success rate of implantation, number of people per health status (implantation failure, relapse, GVHD, GRFS, death), and quality-adjusted life years (QALYs). QALYs were discounted at an annual rate of 5 percent (Reference Guoen37). Cost-effectiveness outcomes included short-term and long-term medical treatment costs and incremental net benefits (△QALY ×WTP-△Cost; WTP: willingness to pay). According to the recommendations of the World Health Organization (WHO) and the Guidelines for the Evaluation of Chinese Pharmacoeconomics, we used 1–3 times the per capita GDP of China reported in 2024 ($13,152.44–$39,457.32) as the WTP threshold (38).

Sensitivity analyses

One-way sensitivity analysis was used to test the influence of the variation of each model parameter on the results, and the results were represented as tornado plots. The estimated range for each parameter was ±5 percent. We also performed a probabilistic sensitivity analysis using Monte Carlo simulations for 10,000 iterations to evaluate the uncertainty of our input parameters and gave the results by a cost-effectiveness curve. The probability parameters and utility values between 0 and 1 were set to follow the beta distribution, and the cost parameters greater than 0 and positive bias were set to follow the gamma distribution.

Equipment price and “break-even” calculation

In fee for service, patients bear no device price differences, gaining net benefits from better equipment, with device costs excluded from direct medical costs due to incalculable single-transplant consumption. However, high equipment prices significantly impact healthcare systems. So, we collect public procurement prices of the two devices published on the China Government Procurement Network from 2020.1 to 2024.12 (49). From the healthcare system perspective, this study calculates the minimum total transplants for Spectra Optia to be more economical than COM.TEC via “price difference ÷ per-patient net benefit.”