2.1 Overview

We built a decision-analytic microsimulation model and analysed the decision problem for the cost year 2022. A willingness to pay of EUR 50,000 per quality-adjusted life year (QALY) was assumed for the base-case analysis, based on a prior analysis by the Ständige Impfkommission (STIKO) [20], though no official threshold exists for Germany. An annual discount rate of 3% was chosen for costs and outcomes as recommended by the German Institut für Qualität und Wirtschaftlichkeit (IQWiG) [21]. Outcomes included life years, QALYs, the net monetary benefit (NMB), and the incremental cost-effectiveness.

For ED, we assumed that patients would be treated a single time and that re-treatment would not be possible, as viral vectors trigger an immune response [22]. It was assumed that each patient would receive exactly one dose of ED, even in the case of treatment failure. Patients switched treatment to EHL-FIX prophylaxis if the effectiveness of ED waned sufficiently. Patients receiving EHL-FIX prophylaxis were assumed to receive regular FIX infusions for their entire life.

2.2 Model Structure

The model was implemented in the statistical software R. The R code is provided via GitHub (https://github.com/NiklausMeier/HB_CE_microsimulation). The use of a microsimulation is particularly relevant for haemophilia due to the accumulation of chronic symptoms over time for a heterogenous patient population and the potential for treatment effectiveness to wane over the lifetime of patients treated with ED. The microsimulation used a 3-month cycle length and a lifetime horizon, assuming that treatment assignment could only change every 3 months. The 3-month cycle length was chosen as a compromise between computational speed and the precision of the simulation in regard to time-sensitive events. The 3-month bleed rate served as a key model parameter that was influenced by patient characteristics and treatment. The cumulative number of joint bleeds at each point in time was calculated as a patient’s historic number of bleeds at model baseline plus the cumulative number of joint bleeds since model baseline. Associated clinical outcomes such as hospitalizations, arthropathy, and mortality were modelled for every patient as functions of either 3-month or cumulative joint bleed rates, and patient characteristics. The technical aspects of the model are described in greater detail in the appendix (see the Electronic Supplementary Material [ESM]).

Bleeds were classified either as joint bleeds or as any other kind of bleed. The accumulation of joint bleeds led to worsening arthropathy over time, quantified via the Pettersson score (PS). The total number of bleeds increased the probability of dying in each cycle. The frequency of bleeds in the model was defined via the ABR, meaning the expected number of bleeds per year for a given patient. At population generation, each patient in the model was assigned an individual untreated ABR, drawn from a distribution to reflect between-patient heterogeneity, and a historic ABR. Untreated ABR was defined as the average number of annual bleeds a patient would experience if they did not receive any ABR-lowering treatment. The specific untreated ABR of each individual in the model was lowered by the relative bleed reduction of their treatment (ED or EHL-FIX prophylaxis), ensuring that ABR was reduced in a proportional manner. Historic ABR was defined as the average number of annual bleeds prior to the start of the model.

Survival status in each cycle was implemented as a survival probability based on the probabilities of death in all previous cycles. It was adjusted via the life-table method [23] as a form of half-cycle correction, assuming that on average patients would die mid-cycle. As the model did not use any state transitions, the probability of death was the only aspect that required an active half-cycle correction. The correction was carried forward to all other outcomes, as they were calculated as functions of survival and treatment assignment.

Quality of life was modelled for each 3-month cycle via a baseline utility that was modified based on sex, age, arthropathy, the total number of bleeds during the cycle, and the burden of prophylactic infusions.

For both deterministic and probabilistic analysis, heterogeneous populations were generated, varying in terms of age, sex, untreated ABR, and historic ABR. Age was varied at the patient level, but not as a probabilistic parameter. Sex and untreated ABR were varied both probabilistically and as patient-level attributes, described in further detail below. The historic ABR of each patient was the product of their individual untreated ABR and the mean of the probabilistic historic ABR. In all analyses, the generated patients were duplicated and treated with either ED or EHL-FIX prophylaxis, creating perfectly matched populations for both the deterministic and probabilistic analysis.

The probabilistic distributions are described in Table 1. Beta distributions were used for probabilities and ratios. Gamma distributions were used for continuous, positive parameters. Given the lack of randomized data comparing ED with untreated patients, the ABRs of untreated patients, patients with ED, and patients with EHL-FIX, were sampled independently in the probabilistic analysis, rather than as relative risk reductions. Uniform distributions were chosen for the properties of ED for which there was no data (described further in Sect. 2.4, “Etranacogene Dezaparvovec”).

Table 1 Overview of model inputs

Probabilistic analysis considered both the heterogeneity of patients (first-order uncertainty) as well as statistical uncertainty of input parameters (second-order uncertainty) in a single analysis. More detail on the distributions and their parameters is included in the ESM appendix (Sect. 1, “Probabilistic distribution”).

The probabilistic analysis was used to generate our main results. 2000 simulations were run with random draws of the probabilistic parameters. Per simulation, 100 patients were generated, and treated with either ED or EHL-FIX, creating a perfectly matched population of 200 patients with 50:50 treatment assignment. 2000 simulations with 200 patients each resulted in a total of 400,000 simulated individuals. The deterministic analysis was used as a reference for the probabilistic analysis, as well as for univariate sensitivity analyses and scenario analyses, since using the probabilistic analysis for these analyses would not be computationally feasible. For the deterministic sensitivity analysis, 10,000 patients were generated, and treated with either ED or EHL-FIX, creating a perfectly matched population of 20,000 patients with 50:50 treatment assignment.

2.3 Population

The population was designed to be representative of adult patients with moderate-to-severe HB in Germany in 2022 and included patients at different stages of their disease history. 4.7% of patients in the model were female [24]. The average age of patients at the start of the model was 36.3 years based on the average age of HB patients in the Socioeconomic Survey (CHESS) study [25], and the minimum age was 18. The weight of patients was determined via the mean weight of German individuals, stratified by age and sex, and was updated as patients aged in the model at the start of every 3-month cycle [26]. As the relation between EHL-FIX dosage and weight is completely linear, and as only EHL-FIX dosage depends on weight in the model, weight was not varied between patients of the same sex and age, or as a probabilistic parameter.

The mean untreated ABR was 32.9 and was obtained from a clinical study of patients with moderate-to-severe HB receiving on-demand treatment [27]. As patients with moderate-to-severe HB in Europe all receive at least on-demand treatment [25], it was assumed that the ABR under on-demand treatment would be a satisfactory proxy for the ABR when receiving neither EHL-FIX prophylaxis nor gene therapy, and that bleed reductions could be calculated relative to the on-demand ABR. The mean historic ABR in the model was 4.6 and was based on a study of real-world outcomes of HB patients in Europe [25], who received a mix (44% on-demand and 56% prophylaxis) of different treatment strategies. Based on the mean historic ABR of 4.6 and the mean age of 36.3, the average patient in the model had 130.68 cumulative bleeds at the start of the model.

2.4 Etranacogene Dezaparvovec

The effectiveness of ED was drawn primarily from the publication of the phase 3 HOPE-B clinical trial [16]. 52 out of 54 patients were able to discontinue EHL-FIX prophylaxis after ED treatment. This observation was implemented as a 96.3% treatment success probability. Success and failure were drawn randomly for each patient in the simulation. A mean ABR of 1.51 was achieved during a period of 18 months after being treated with ED.Footnote 1 The mean ABR was transformed into a relative bleed reduction via the ratio of the trial ABR to the assumed untreated ABR of 32.9, yielding a relative bleed reduction of 95.4%.

In the absence of evidence on long-term effectiveness of ED in the HOPE-B clinical trials, it was assumed that the relative bleed reduction of 95.4% would be maintained for 10 years [17, 18]. Both animal [17] and human [18] studies showed transgene expression to last for up to 10 years and potentially longer for haemophilia. It was further assumed that after 10 years, the relative bleed reduction would decline gradually by 10 percentage points per year. There is no clinical evidence available for this assumption, but given the biological mechanisms, it did not seem plausible that the effectiveness of ED would wane completely from one cycle to the next. A clinical expert for haemophilia within the group of authors considered this to be a plausible assumption, given the lack of evidence. Patients who received ED switched to EHL-FIX prophylaxis when their ABR exceeded a value of 4, ensuring that patients would not be under-treated after loss of treatment benefit. The threshold of 4 was chosen as it is the closest integer to the 4.19 ABR of FIX prophylaxis in the HOPE-B trial, and in the absence of further evidence, a simplifying assumption seemed most reasonable [16].

2.5 EHL-FIX Prophylaxis

The effectiveness of EHL-FIX prophylaxis was drawn from the publication of the phase 3 HOPE-B clinical trial [16]. In this trial, a mean ABR of 4.19 was achieved during a lead-in period of at least 6 months, during which patients received FIX prophylaxis with a dose and product determined by their physician. The relative bleed reduction of 87.3% of EHL-FIX prophylaxis compared to on-demand treatment was calculated based on the 4.19 ABR under FIX prophylaxis and the untreated ABR of 32.9. The specific FIX products were not described in the study, and the market shares of EHL-FIX products could not be sourced for Germany. We thus modelled the costs of EHL-FIX based on the price of nonacog beta pegol. Out of the three EHL-FIX products included in the CHESS II study [12], nonacog beta pegol had the middle price, and as the choice of EHL-FIX product mainly affected costs, we deemed this to be the most representative, lacking further information. This choice affected the price, the dosage, and the treatment frequency with EHL-FIX.

2.6 Bleeding and Arthropathy

In the deterministic analysis, the fraction of bleeds which occurred in the joints was 61.7% of all bleeds, based on patterns observed in patients with severe haemophilia in a UK patient registry [28]. PS increased by 1 point whenever a patient accumulated 13 additional joint bleeds [29]. Joint surgery occurred in the first cycle in which PS reached the threshold for clinically relevant damage, which lies at 28 out of 78 points [30].

2.7 Mortality

The background mortality of the German general population was drawn from the Human Mortality Database [31]. The effects of bleeding on mortality were derived from an observational study of Dutch patients with haemophilia, which showed an age-adjusted standardized mortality ratio of 2.4 for patients with severe haemophilia [32]. It was assumed that this standardized mortality ratio would apply for patients with the historic ABR of 4.6. In each cycle, patients with a current ABR less than the historic ABR of 4.6 [25] were assumed to have a lower standardized mortality ratio and a higher probability to survive a cycle, and vice versa, relative to background mortality. The precise assumptions and mathematical functions required to model this relationship between bleeds and mortality are described in the ESM appendix (Sect. 2.5, “Mortality”).

2.8 Quality of Life

The non-disease specific utilities were based on a regression equation from a study on the Health Survey for England data [33], considering sex and age. The disutilities for arthropathy were derived from an analysis of multiple studies (which included X-rays and SF-36 questionnaires) for patients with moderate-to-severe haemophilia, which the authors used to calculate an association between PS and utility [2]. This study showed a disutility of 0.03 for a PS between 12 and 21, and an additional disutility of 0.07 for a PS of 22 or higher. The disutilities from bleeds were based on a phase 4 diary study (DOSE) in the United States, which analysed the daily quality of life of patients in relation to haemophilic bleeding events [34]. Each bleed led to a disutility of 0.2, which lasted for 1 day; the number of bleeds in a cycle was multiplied by the disutility of 0.2 and then divided by the cycle length (measured in days) to calculate the average bleed-disutility for that cycle, which was then applied to the entire cycle. The disutilities from joint surgeries were based on EQ-5D-5L questionnaire data collected in France and the UK to assess the quality of life of patients with haemophilia [35]. This comparison showed a utility difference of 0.18 between patients with and without joint surgery, which was implemented as a disutility that lasted for 27 days after surgery [38]. The disutilities from prophylactic FIX infusion were based on a vignette-based time trade-off study [36]. This time trade-off study showed that each coagulation factor infusion led to a one-time utility decrement of −0.0003, and since prophylactic nonacog beta pegol treatment requires 52 infusions per year, this is equivalent to a disutility of 0.0156 per life year. This 0.0156 disutility was subtracted from each patient’s quality of life in each cycle that they received EHL-FIX prophylaxis.

2.9 Resource Use

We considered medical resources with potential major cost implications. For patients treated with ED, treatment occurred immediately at the start of the simulation. For nonacog beta pegol, the dosage of EHL-FIX per treatment was 40 international units (IU) per kilogram of body weight [37]. All patients in the model were treated with this dosage after every bleed, and additionally once per week for patients receiving EHL-FIX prophylaxis. Perfect vial sharing with no wastage of EHL-FIX was assumed. The probability of hospitalization per bleed was 27.8% [12]. The length of stay at the hospital after a bleed was 1.5 days, and the length of stay after surgery was 27 days [38].

2.10 Unit Costs

For ED, a price of EUR 1,500,000 was assumed. In the United States, a list price of USD 3,500,000 has been announced; the price that will actually be paid by health care systems is expected to be substantially lower. We obtained costs of EHL-FIX (nonacog beta pegol) (EUR 1.70 per IU), intensive care (EUR 1469 per day), non-intensive inpatient care (EUR 597.20 per day), and joint surgery (EUR 866 per procedure, excluding costs for hospital days) from the CHESS I and II studies [12, 39]. All prices were adjusted for inflation to the cost year 2022 [40].

2.11 Uncertainty Analyses

Probabilistic analysis The probabilistic analysis was used for the main results and to evaluate the probability of ED being cost-effective over EHL-FIX.

Univariate sensitivity analysis The model was run using the lower and upper bound value of each uncertain parameter, one by one, holding all other parameters constant at their deterministic values. For the parameters with assigned second-order probability distributions based on published data, the lower and upper bounds were equivalent to the 2.5% and 97.5% percentiles of the respective distributions. For the parameters with uniform probabilistic distributions based on assumptions, the lower and upper values of those uniform distributions were used for the lower and upper bounds in the univariate analysis. Wide value ranges were chosen for the assumed and therefore highly uncertain properties of ED.

Scenario analyses The prices of ED and the EHL-FIX product were varied incrementally to see at which price levels ED would have a positive incremental NMB and therefore be cost-effective. Additionally, the following scenarios were run and compared with the probabilistic analysis.

Bleed disutility for 7 days: In the probabilistic analysis, it was assumed that the disutility from a bleed only lasted for 1 day. However, acute symptoms may last longer. Since the exact duration of the disutility is not known, a disutility per bleed lasting 7 days was examined.

5 × costs per day in intensive care unit (ICU): In the probabilistic analysis, the costs of treating bleeds were based on an analysis from the secondary literature, which may be conservative and not consider all potential costs due to bleed hospitalizations. To test an especially extreme hypothetical assumption not captured in the other uncertainty analyses, these costs were increased fivefold.

No treatment switching: In the probabilistic analysis, it was assumed that patients switch from ED to EHL-FIX prophylaxis when their ABR exceeded a threshold of 4. In this scenario, patients did not switch treatments.

No vial sharing: In the probabilistic analysis, it was assumed that vials of EHL-FIX were shared perfectly, with no waste. This assumption was relaxed, and EHL-FIX was instead consumed in discrete package sizes of 500 IU (the smallest available package).

Cost-minimization: In the probabilistic analysis, a willingness to pay of EUR 50,000/QALY was assumed. In this scenario, the willingness to pay was set to EUR 0/QALY, making incremental NMB equivalent to incremental costs. This analysis shows the lifetime per-patient cost impact of ED compared to EHL-FIX prophylaxis.

Age of 18/60: In the probabilistic analysis, the simulated population included patients of different ages. In two scenario analyses, the baseline ages of all patients were set to either 18 or 60 years to assess the impact of age on cost-effectiveness.

German non-disease specific utilities: In the probabilistic analysis, non-disease specific utilities were based on a study using English data [24]. In this scenario, alternative utilities from a quality-of-life study from Germany [41] were used for the baseline, sex, and age.

Duration of maximum bleed reduction 20/30 years: In the probabilistic analysis, the duration of the maximum bleed reduction ranged from 5 to 15 years. In this scenario, the impact of assuming even higher values, 20 or 30 years, is shown.

Bleed rate increase per year 20/30%: In the probabilistic analysis, the bleed rate increase per year ranged from 5% to 15%. In this scenario, the impact of assuming even higher values, 20% or 30%, is shown.