To assess the cost-effectiveness of CAR-T cell therapy generated by the AIDPATH platform compared to SoC and manually produced, centrally distributed CAR-T cell therapy (cilta-cel; Johnson & Johnson), a partitioned survival model was developed as part of this early health technology assessment (HTA). The partitioned survival model consists of health states that a patient can occupy at a given time point, which reflect the clinical trajectory of patients [4].

The study population comprised adult patients with MM in Germany who had not responded to or had experienced a relapse/refractory (R/R) after at least 3 prior lines of cancer treatment. These treatments include an immunomodulator, a proteasome inhibitor, and a cluster of differentiation 38 antibody. In cases where the disease no longer responds to agents in each of the 3 classes, it is referred to as “triple-class refractory” MM [5]. This patient group is currently eligible for CAR-T cell therapy within the German health care system [6].

Given that social, environmental, and other contextual factors influence disease progression [7], this study focuses on a single country. As the first AIDPATH platform is being implemented at the University Hospital Würzburg, a leading center for CAR-T cell research and treatment, Germany was selected as the setting for this analysis.

The partitioned survival model used in this study consists of 3 health states: progression-free disease (PFD), progressive disease (PD), and death (Figure 1). For the cohort receiving cilta-cel, the starting point was “apheresis” on the decision tree. Hereafter, patients were either not infused and streamed into the partitioned survival model or started a bridging period while they waited for their CAR-T therapy to be produced. After the bridging period, the infused cohort streamed into the partitioned survival model. All cohorts start in the PFD health state. Patients enter the model in the PFD health state and receive either CAR-T cell therapy (derived from AIDPATH or cilta-cel) or SoC. Patients enter the model at 61 years old, based on the selected patient group’s average age of 61.7 years in the cilta-cel clinical trial [8]. A cycle length of 1 week was used, and the time horizon for the analysis was 40 years (ie, 2080 cycles). This is in line with the Dutch Health Care Institute (Zorginstituut Nederland [ZIN]) cilta-cel study and represents a lifelong time horizon [5].

The number of patients in the PD health state at a certain time equals the overall survival (OS) proportion minus the progression-free survival (PFS) proportion. The number of patients in the death state equals 1–OS. OS and PFS rates for SoC were based on German real-world clinical population data. Here, time to next treatment was used as a proxy for OS [9]. The OS and PFS rates for the CAR-T therapy arms were based on Kaplan-Meier curves (median OS was not reached at the data cutoff; 27.7-month PFS rate: 54.9%) digitized from published clinical trial data [8-10] and reconstructed into pseudoindividual patient data to enable parametric survival modeling. Parametric survival models (including exponential, Weibull, Gompertz, gamma, log-logistic, log-normal, and generalized gamma distributions) were fitted to the reconstructed data to enable extrapolation beyond the observed trial period. Model selection was based on statistical goodness of fit (Akaike information criterion [AIC]), visual inspection of fitted vs observed Kaplan-Meier curves, and assessment of hazard functions. The analyses were then used to extrapolate the survival data beyond the time frame covered by clinical trials. Detailed methodology for extrapolation, data fitting, and model selection can be found in Section A in Multimedia Appendix 1 [8,11-41].

The mortality rate for the PFD health state equals background mortality, retrieved from German life tables, obtained from the World Health Organization [42]. The mortality rate for the PD health state is cancer-specific mortality. The life tables used were from the most recent year available (2019) and for ages 61 to 100 years.

Data on health state utilities were derived from the CARTITUDE study [8] and the Dutch Health Care Institute (ZIN) cilta-cel study [5]. In these studies, the EQ-5D-5L questionnaire was used to assess utility scores, and mixed-effects models were used to estimate utilities for the PD state when data were missing [5]. Utilities were not age-adjusted. Data on efficacy and safety from the SoC therapies and cilta-cel CAR-T therapy were collected from publicly available trial results. Data from the AIDPATH platform will be collected once it is set up. When data are not available from the AIDPATH platform, estimates are based on cilta-cel.

Assuming similar clinical outcomes (PFS and OS), adverse event rates, and manufacturing success rates for AIDPATH and cilta-cel was an intentionally conservative modeling choice. In the base case, the manufacturing failure rate for AIDPATH was assumed to be equal to that of cilta-cel (3%) [43]. Manufacturing failures were not modeled as separate clinical events; instead, their impact was incorporated through increased production costs, reflecting the need for repeated manufacturing attempts to achieve a successful infusion. This assumption was made despite the AIDPATH platform incorporating a highly automated and standardized manufacturing process designed to reduce operator-dependent variability and production errors.

Consequently, assuming equivalent performance likely underestimates the potential clinical and operational advantages of the AIDPATH platform.

The costs for cilta-cel are based on the list price of €420,000 (€1=US $1.13; 2025 annual average) [5]. These costs are applied once. As AIDPATH does not have marketing authorization (yet), costs were based on production costs plus an estimation of marketing costs, totaling €120,000. The total cost of AIDPATH treatment was estimated at €120,000 based on a bottom-up aggregation of key cost components informed by published literature. Manufacturing (production) costs were assumed to be approximately €60,000 per treatment [11,44]. A 30% markup (€20,000) was applied to account for sales and marketing expenses [45]. Hospital-related costs were estimated at €27,000 [46], and quality control costs were approximately €13,000 [47]. Preinfusion period costs consisted of apheresis, bridging therapy, and conditioning therapy; these applied to both types of CAR-T therapies. Refer to Table S1 in Section B of Multimedia Appendix 1 for the cost breakdown.

Hospital resource use was based on data from the CARTITUDE-1 study [8], ZIN Carvykti economic analysis [5], and the AIDPATH consortium. Hospital costs and medication costs were based on the German fee tariff schedule (2023) [12], the Kassenärtzliche Bundesvereinigung, and publicly available economic evaluations [5,6] (Table 1). All costs were indexed to 2025 euros based on German inflation [48].

SoC encompasses the physician’s choice of treatment for patients with triple-class refractory MM. These data were obtained from the German Therapy Monitor, a registry of real-world oncology data. It contains anonymized data on patients with approximately 4000 R/R MM, obtained from a representative sample of 108 German centers. Refer to Table S2 in Section C in Multimedia Appendix 1 for an overview of included drugs and Table S3 in Section C in Multimedia Appendix 1 for detailed calculations.

When patients enter the PD health state, costs for follow-up treatment were applied once for all treatments. The treatment given as fifth-line R/R MM treatment is one of the regimens given in the fourth line that the patient has not received before [9]. Therefore, the average fifth-line treatment price for SoC and CAR-T is the same as that for the fourth line. The average treatment duration in the fifth line is assumed to be 4 months [50]. Monthly costs of fifth-line treatment are €10,349.98. The percentage of patients with fourth-line R/R MM that receive fifth-line treatment is 0.22, according to attrition rates in a study by Dhanasiri [51]. This results in the applied 1-time follow-up treatment costs being €9108.00 (4×€10,349.98×0.22).

All baseline parameters are presented in Table 1.

The primary outcome of the analyses is the incremental cost-effectiveness ratio (ICER), which is calculated by dividing the difference in total costs (incremental cost) by the difference in quality-adjusted life years (QALYs; incremental effect) [53].

Secondary outcomes include the avoidance of adverse events. Adverse events were only included if they incurred health care costs (required physician input and/or hospitalization and/or medication) or had a large impact on patient quality of life (refer to Table S4 in Section D in Multimedia Appendix 1 for a list of included adverse events). As the occurrence rates of adverse events for AIDPATH are not yet available, they are assumed to be equal to those of cilta-cel.

Following the methodological guidelines of the German National Institute for Quality and Efficiency in Health Care (Institut für Qualität und Wirtschaftlichkeit im Gesundheitswesen [IQWiG]), the German institute responsible for evaluating health care interventions, outcomes, and costs was discounted at 3% [49].

The model was constructed using Excel (version 2608; Microsoft Corp).

A 1-way sensitivity analysis was performed to assess the model results for robustness and to identify the main cost drivers. Each parameter was varied by 20% [54]. Refer to Table 1 for the lower- and upper-limit values.

Various scenarios were performed to assess the economic implications of the competing treatment strategies and to test the robustness of the different assumptions made in the model.

In the first scenario analysis, the risk of adverse events following AIDPATH-produced CAR-T cell therapy was decreased by 10% and 20%. CAR-T–specific side effects after AIDPATH-produced CAR-T cell infusion are expected to decrease. Furthermore, the CARs are equipped with an epidermal growth factor receptor marker, which may prevent fatal outcomes of CAR-T cell–related toxicity [55].

In the second scenario analysis, the shipping and manufacturing time of CAR-T cells produced by the AIDPATH platform was lowered. In this scenario, we assumed the manufacturing time (14 days) plus shipping time (0 days) as the bridging period for AIDPATH (the average bridging therapy duration for cilta-cel is 32 days). A shorter bridging period is expected to reduce the incidence of bridging-related adverse events. It also affects the number of patients who remain alive and eligible to receive lymphodepleting therapy and the subsequent CAR-T infusion. Because bridging therapy is shorter, the direct costs of bridging treatment decrease, along with the costs of managing adverse events associated with bridging chemotherapy.

To explore uncertainty in important assumptions related to AIDPATH, additional scenario analyses were performed. In the third scenario, the PFS rate of AIDPATH was varied by ±10% to assess the impact of potential differences in treatment efficacy. In the fourth scenario, the production costs of AIDPATH were increased by 10% and 20% to reflect a higher manufacturing failure rate. An increase in manufacturing failure rate implies that more production attempts are required per successfully treated patient, thereby increasing the cost per treatment. This approach allows an indirect assessment of the impact of manufacturing failure on cost-effectiveness outcomes.

To address uncertainty in the AIDPATH cost estimate, a fifth scenario analysis was added to explore the impact of the total costs of an AIDPATH treatment. To do so, costs were varied to €200,000 and €300,000.

This study presents an early HTA based exclusively on secondary data obtained from previously published literature, publicly available sources, and aggregated data. No primary data were collected, and no individual-level patient data were accessed.

According to German and Dutch regulations governing research involving human participants, ethics approval from an ethics committee (Ethikkommission) is not required for studies that rely solely on anonymized, aggregated, or publicly available data and do not involve interventions with human participants [56,57]. Therefore, formal ethics approval and informed consent were not required for this study.

The results indicate that AIDPATH-produced CAR-T cell therapy is projected to be both more effective and less costly than current SoC and cilta-cel treatment. Specifically, patients receiving AIDPATH therapy are expected to gain more QALYs while incurring lower overall treatment costs. This improvement in cost-effectiveness is reflected in the ICERs, which are negative for both comparisons: –€3075.08 when comparing AIDPATH to SoC and –€1,986,526.56 when comparing AIDPATH to cilta-cel. A negative ICER indicates that AIDPATH therapy dominates the alternatives, meaning it provides better health outcomes at a lower cost. Detailed results are presented in Table 2.

Most costs of the CAR-T cell therapies can be attributed to the acquisition, followed by adverse events and monitoring (Table 3).

An overview of the model input parameters with the largest impact on the ICER when comparing AIDPATH-produced CAR-T cell therapy to SoC is shown in Figure S3 and Table S8 in Section F in Multimedia Appendix 1. The results show that the discount rate of outcomes and PFS rate have the largest impact on the ICER. This indicates that assumptions about the valuation of future health benefits and long-term patient outcomes play a critical role in determining the overall cost-effectiveness of AIDPATH therapy.

An overview of the model input parameters with the largest impact on the ICER, when comparing AIDPATH therapy to cilta-cel CAR-T cell therapy, is shown in Figure S4 and Table S9 in Section F in Multimedia Appendix 1. The results show that the discount rate of outcomes and the vein-to-vein time have the largest impact on the ICER. This suggests that both the assumptions regarding how future health benefits are valued over time and the time required for manufacturing and delivering the CAR-T cell product play a crucial role in determining the relative cost-effectiveness of AIDPATH therapy compared to cilta-cel.

The scenario analyses demonstrated that the results were generally robust to variations in key assumptions. Reducing adverse event risks (scenario 1) slightly lowered costs, with QALYs unchanged, and AIDPATH remained dominant vs both comparators. Shortening manufacturing and shipping times (scenario 2) increased QALYs with modest cost increases; AIDPATH remained dominant compared to cilta-cel but was less cost-saving compared to SoC.

Varying PFS rates (scenario 3) affected both costs and QALYs; however, AIDPATH remained dominant across all variations. Increasing production costs to reflect lower manufacturing success rates (scenario 4) raised total costs without affecting health outcomes. Despite these increases, AIDPATH remained cost-effective compared with cilta-cel and SoC. Increasing treatment prices (scenario 5) substantially increased costs (Table 4).

The objective of this study was to assess the cost-effectiveness of CAR-T cell therapy produced by the AIDPATH platform for patients with triple-class refractory MM in Germany, compared to manually produced and centrally distributed CAR-T cell therapy (cilta-cel) and SoC. In this model, given its assumptions and from a hospital perspective, AIDPATH CAR-T therapy was shown to be dominant over a period of 40 years, offering improved outcomes and lower overall costs than both comparators. Despite the inferiority, only a small difference was observed in QALY outcomes between the CAR-T treatments in the base case analysis. This is despite the expectation that shorter vein-to-vein time would increase the proportion of patients remaining eligible for infusion by limiting progression during the manufacturing period [2]. Due to limited evidence, equal eligibility was conservatively assumed, and thus, the observed QALY difference primarily reflects earlier treatment administration rather than differences in infusion eligibility.

The findings suggest that the AIDPATH platform could be considered for implementation within the German health care setting. The results may provide a basis for estimating the return on investment for relevant stakeholders. Importantly, the base case assumptions were intentionally conservative. Equivalent clinical outcomes (PFS and OS), adverse event rates, and manufacturing failure rates were assumed for AIDPATH and cilta-cel, despite the expectation that AIDPATH’s automated and standardized manufacturing process may reduce variability and production errors in practice. The scenario analyses, which included further conservative adjustments, such as increased costs and limited efficacy gains, demonstrated that AIDPATH remained cost-effective, supporting the robustness of the findings.

The sensitivity analyses illustrate the sensitivity of the results to changes in methodological choices. In both sensitivity analyses, only a few parameters had a large impact on the outcomes. These parameters are important to consider when modifying or scaling up the platform to safeguard its cost-effectiveness compared to alternative treatment strategies.

The results of this study can be used to guide discussions with German stakeholders, including investors and policymakers, regarding the incremental cost-effectiveness of CAR-T cell therapy and reimbursement decisions. Furthermore, the findings can be considered exemplary for countries with comparable health care systems. Finally, AIDPATH uses a patient-centric approach to CAR-T cell production, and based on the results, this approach shows potential.

To our knowledge, this is the first early HTA study to assess the cost-effectiveness of AI-supported CAR-T cell therapy production in Germany from the hospital perspective. This study is tailored to the German health care setting and provides a model that can be updated when new data become available.

Although this cost-effectiveness study provides valuable insights to support decision-making, it has several important limitations that must be acknowledged. First, by focusing narrowly on cost-effectiveness, the study may overlook the broader potential benefits of automated CAR-T therapy production. These may include system-level advantages associated with decentralized manufacturing, such as reduced logistical complexity, lower risks related to cold chain management and cross-border shipment, and improved access to treatment through reduced patient travel or expanded treatment capacity. In addition, automated production may improve treatment convenience and patient satisfaction and streamline clinical workflows. These qualitative outcomes can significantly impact health care delivery and patient experience but are often difficult to adequately quantify in economic evaluations [58]. Moreover, as with many applications of medical AI, there are pressing concerns about appropriate regulation and evaluation [59]. HTA has been recognized as a potentially key method for evaluating AI-driven innovations. However, this model does not fully address AI-specific challenges, such as the assessment of continuously learning systems, the explainability of AI decision-making, and interoperability with existing health care infrastructure. To address these gaps, Boverhof et al [59] proposed a comprehensive framework of 29 evaluation elements across 4 domains, developed through a systematic and targeted review, which could serve as a more nuanced approach for assessing the value of AI in medicine. This broader perspective is crucial for ensuring that automated CAR-T cell production and similar technologies are appraised not only for their economic impact but also for their systemic and societal implications.

Second, no specific data regarding the safety and effectiveness of CAR-T therapy generated by the AIDPATH platform are available yet. The platform is currently being tested, and it is important to update the model as soon as clinical trial data become available. In addition, survival extrapolation is an important source of uncertainty in this analysis. A mixture-cure log-normal model was selected based on statistical goodness-of-fit, visual inspection, and clinical plausibility. Alternative parametric models were explored during model selection; however, these showed inferior fit and/or less plausible long-term extrapolations.

Third, a limitation in achieving the objective of this study is the reliance on Dutch CARTITUDE trial data for key comparator inputs, specifically for the centrally manufactured CAR-T therapy cilta-cel [5]. Due to the lack of German-specific data, key cost and resource use parameters for cilta-cel were derived from a Dutch study, and these values were also used as a proxy for the AIDPATH-generated CAR-T cell therapy. However, this approach introduces uncertainty, as health care delivery, cost structures, and reimbursement mechanisms differ between the Dutch and German hospital systems. For example, hospital tariffs, staff costs, and care pathways in Germany may be higher or more complex, potentially leading to an underrepresentation of true costs from the German hospital perspective. This could result in a more favorable cost-effectiveness profile for CAR-T cell therapies than would be observed with German-specific data. Future studies should aim to validate these inputs with empirical data from the German context to strengthen the reliability of the cost-effectiveness conclusions. Because this approach was used for both CAR-T cell therapy arms, it is unlikely that an updated model would change the conclusions of the current study.

Despite these limitations, an HTA at this early stage is considered highly valuable because it provides critical insights into the potential clinical and economic impact of this new medical product. It supports informed decision-making during early development stages, helps prioritize investments, guides research efforts, and identifies key data gaps to address before implementation [60].

This study is the first to assess the cost-effectiveness of AI-supported, hospital-based CAR-T cell therapy produced by the AIDPATH platform for patients with MM in Germany from the hospital perspective. On the basis of the assumptions inherent in this model, AIDPATH therapy was found to be cost-effective and dominant compared with both cilta-cel and SoC from the hospital perspective over a 40-year time horizon. This was found in the base case analysis and was confirmed by all scenario analyses performed.

However, key inputs were partly based on Dutch data due to the lack of German-specific figures, introducing uncertainty, particularly in cost estimates. Additionally, no clinical data are yet available for AIDPATH-generated CAR-T cell therapy, and broader benefits, such as patient satisfaction and system efficiency, were not captured in the cost-effectiveness analysis.

Despite these challenges, this early HTA provides valuable insights to guide further research, inform reimbursement discussions, and support planning for future implementation in Germany and similar health care systems.