To perform the budget impact analysis (BIA), a three-state partitioned survival model (PSM) was used, which classifies patients into states of PFS, progressive disease (PD), and death (D). All patients were assumed to have started in the PFS state and either remained progression free (PF), progressed, or died in subsequent cycles. Time to progression was derived from the difference between the areas under the PFS and overall survival (OS) curves. After disease progression, up to three subsequent therapy lines were considered (Fig. 1). To reflect the dosing schedules for the included drug regimens, a 28-day cycle was used in the model. Treatment duration was based on the median PFS observed in the clinical trials.
The budget impact analysis was performed from the perspective of the German SHI over a 1-year and 3-year time horizon. A dynamic cohort model to account for the prevalent cohort in the starting year 2019 as well as the incident cohorts in the following years was applied. Additionally, the yearly cohort size was divided by 13 (equal to the number of 28-day cycles per year) to reflect a constant patient-inflow into the model.
The budget impact analysis was conducted using Microsoft Excel 2016 (Microsoft Coporation).
Regimens of interest
The focus of the current analysis was to evaluate the four triplet therapy regimens recently approved for the treatment of rrMM. These were carfilzomib plus lenalidomide plus dexamethasone (KRd), elotuzumab plus lenalidomide plus dexamethasone (ERd), daratumumab plus lenalidomide plus dexamethasone (DRd), which were referred to as intravenous therapy regimens. In addition, ixazomib plus lenalidomide plus dexamethasone (IRd), referred to as oral therapy regimen, was considered. Lenalidomide in these regimens is normally given until progression. Patients progressing under treatment with these triplet regimens are considered to be lenalidomide refractory. Therefore, lenalidomide-free regimens are preferred after lenalidomide-based treatment . Hence, three lenalidomide-free therapy regimens were considered after the progression on the four initial triplets mentioned above. These additional three therapy regimens were pomalidomide plus dexamethasone (Pd), daratumumab plus bortezomib plus dexamethasone (DVd), and panobinostat plus bortezomib plus dexamethasone (FVd).
The analysis was limited to the listed therapies above as well as to the order and frequency in which the therapies were prescribed. It was assumed that patients start treatment on one of the four lenalidomide-based therapies in second line, namely, KRd, ERd, DRd or IRd. Once those patients experienced a disease progression, lenalidomide-free therapies (Pd, DVd or FVd) were prescribed in third treatment line, only. Patients with a repeated disease progression were eligible for any of the considered therapies in fourth and fifth treatment line, excepting those previously received. This setting allowed to reflect 312 different treatment combinations across the four therapy lines.
The baseline target population of the model was defined as all adult patients with MM who have received at least one prior therapy and who may initiate a second-line therapy and more. The determination of the prevalent target population was based on the estimates from the German Centre for Cancer Registry Data (ZfKD) which regularly reports data on incidence and prevalence of cancer in Germany for the entire population as well as for population stratified by age and gender . As the ZfKD only reports data for the overall multiple myeloma population, additional criteria were applied to extract the relevant data on rrMM patients [11, 12]. For the derivation of the incidence population, the same approach as for the baseline target population was applied. These resulted in a prevalent target population of 10,262 in 2019. The incident populations in 2020 and 2021 were estimated at 2337 and 2417 patients, respectively. The distribution of the prevalent target population among the considered therapy regimens was based on the results from a nationwide, multi-institutional survey on treatment of multiple myeloma patients in Germany (TherapyMonitor) . Previous analyses revealed that the database from the TherapyMonitor has a good external validity to the German population regarding the distribution of treated patients with multiple myeloma .
Clinical efficacy data
Clinical efficacy data in terms of PFS and OS curves for each regimen of interest were retrieved from clinical trials [3,4,5,6, 15,16,17]. ERd, DVd and IRd data on OS curves were obtained from the EMA’s assessment reports [18,19,20]. Since the observed survival distributions for PFS and OS were limited by the time of follow-up in published sources, it was necessary to extrapolate them over a lifetime horizon. This was achieved by extracting individual data points from the published Kaplan–Meier (KM) curves for OS and PFS using the WebPlotDigitizer developed by Rohatgi . In addition, the number of patients at risk for each arm at regular time intervals during the follow-up was extracted. This information, usually known as the risk table, was presented beneath the published KM curves. By incorporating the information provided in the risk table, the accuracy of the approximated time-to-event data was improved . Then the extracted data was reconstructed using an algorithm (ipdfc) developed by Wei and Royston  for use in STATA. The algorithm estimates the number of censoring, the number of events, the censoring time, and event time. In addition, it corrects also for monotonicity violators, a situation where a pair of adjacent times and corresponding survival probabilities is inappropriately ordered. Table 1 shows the obtained outputs for treatment and control arms, respectively.
The reconstructed PFS and OS curves, respectively were then fitted to a variety of common parametric distributions, using the maximum likelihood methodology. The distributions that were tested included the exponential, Weibull, Gompertz, log-normal, and log-logistic. The final distributions were chosen based on the following criteria: (1) comparison of Akaike and Bayesian information criteria (AIC/BIC); (2) clinically plausible long-term projections; (3) a comparison of predicted median PFS time and the published figures; (4) visual inspection of the fit to the observed data over the available follow-up time; and (5) residuals against a 45° line (Cox-Snell residual analysis). For PFS, a Gompertz distribution was selected for KRd, DVd, and FVd; a Weibull accelerated failure time (AFT) distribution for ERd, DRd, and Pd; and a log-normal distribution for IRd. For OS, a Gompertz distribution was selected for IRd, KRd, ERd, and FVd; a Weibull distribution for DVd; and a Weibull AFT distribution for DRd and Pd.
The analysis included both direct medical costs and direct non-medical costs (i.e., transportation costs) which are covered by the German SHI. The direct medical costs included the initial and subsequent-line drug acquisition costs, comedication costs, drug administration costs including administration, patient monitoring and laboratory tests, as well as costs for management of adverse events (AE). Drug administration time, dosing, clinical examinations before treatment initiation, and comedications were based on the prescribing information for each agent [24,25,26,27,28,29]. The calculation of drug dosing was based on the relative dose intensity per mean body surface area (BSA) or mean body weight (BW) for carfilzomib and bortezomib or dexamethasone and elotuzumab, respectively. According to the German microcensus, an average German has a BSA of 1.89 m2 and a BW of 76.3 kg. Thus, the estimated drug costs correspond to the number of drug packages/vials used to meet the previously calculated dose intensity. Doses and prices per package were retrieved from a public price list (Lauer-Taxe) . The drug administration costs were retrieved from the physicians’ fee schedule (Einheitlicher Bewertungsmaßstab, EBM).
The incidence of AEs associated with each therapy regimen was obtained from pivotal trials, and the costs of AEs were obtained from the morbidity-oriented risk structure compensation scheme (Morbi-RSA) in Germany . AEs of grade 3 or higher, which were consistently defined across clinical trials, occurred in at least 5% patients, and their costs were listed in the Morbi-RSA only were considered for all regimens. These criteria resulted in 2 AEs that were included in the model: anemia and neutropenia. For each AE, the incidence rate per cycle was multiplied by the respective cost to obtain the AE-associated costs.
The considered non-medical costs included transportation costs, which are routinely covered by SHI for oncological intravenous treatments. These costs commonly refer to roundtrips per drug administration and were estimated according to the mode of transport and distance travelled. The average distance travelled for an intravenous rrMM treatment is 37.2 km in Germany . Due to the poor health condition of rrMM patients, it was assumed that half of the patients used a private mode of transportation driven by a relative and the other half by taxi.
All costs were adjusted for a cycle length of 28 days. Costs were not discounted as recommended by the International Society for Pharmacoeconomics and Outcomes Research (ISPOR) Task Force for BIAs . Data are presented as rounded numbers, although model calculations were performed without rounding. The cycle length and regimen dosing in combination with the drug acquisition costs per cycle as well as other considered cost components are shown in Table 2.
To estimate the budget impact of rrMM patients starting second-line treatment with KRd, ERd, DRd or IRd, 6 different scenarios were considered within a 1-year and 3-year time horizon.
1-year budget impact
Using estimates from the TherapyMonitor Report, the current market share of the four considered therapy regimens was 20%, 14%, 5%, and 4% for KRd, DRd, ERd, and IRd, respectively, in Germany during the 1st quarter of 2019 . To allow for a meaningful comparison of budget impact between the four regimens of interest, these market shares were reweighted to sum up to 100% and assumed to hold true for the entire year 2019. Hence, the reweighted market shares in the reference scenario were estimated at 46%, 32%, 13% and 9% for KRd, DRd, ERd and IRd, respectively. The budget impact for the reference scenario was compared to several new market shares/penetration scenarios. These included the equivalence scenario, where all four triplets had an equal market share (25% each); the KRd-Scenario, where the market share of KRd was assumed at 100% and 0% for all other triplets; the DRd-Scenario, where the market share of DRd was assumed at 100% and 0% for all other triplets; the ERd-Scenario, where the market share of ERd was assumed at 100% and 0% for all other triplets, and the IRd-Scenario, where the market share of IRd was assumed at 100% and 0% for all other triplets. The flow of treatment of the progressed patients after the second line is described in Table 3. The market shares of Pd, DVd, FVd for the third treatment line were derived from the TherapyMonitor and reweighted to sum up to 100%. Once a patient progresses from the third line to the subsequent lines, that patient cannot receive the medication received in the second and third therapy line. For example, if a patient received KRd in the second line and Pd in the third line, this patient cannot receive KRd or Pd in the fourth line and the fifth line. To account for this assumption, patients not eligible for a particular therapy are redistributed among the remaining therapies in the fourth and fifth lines.
3-year budget impact
In all 3-year scenarios, the prevalent cohort (2019) in second line reflects the market share observed in the TherapyMonitor Report during the 1st quarter 2019. The treatment of the progressed patients after the second line followed a similar logic that was applied in the 1-year analysis (Table 4). The incident cohorts (2020 and 2021) in the second line vary with the chosen scenario and reflect the projected market shares in subsequent years. In reference scenario, the market share of KRd and ERd decreased from 46 to 18% and from 13 to 8% (2019–2021), respectively, and the market shares of DRd and IRd increased from 32 to 63% and from 9 to 11%, respectively. This scenario reflects a strong market penetration of DRd that was observed in 2018 . As the IRd was associated with the most-favorable safety profiles among the considered therapy regimens, a small uptake was assumed, especially due to the increased use in patients at advanced age affected by comorbid conditions. The strong decrease of KRd therapy regimen was related to the frequently reported cardiotoxicty problems in patients treated with KRd and decrease in market share observed during the previous year [13, 34]. In scenarios with 100% market share for each therapy regimen, it was assumed that in 2020 and 2021 the market is dominated by a specific therapy regimen. One exception is the scenario with equal market shares, where equal market shares were assigned to all drugs in the second and subsequent treatment lines.
The sensitivity of the model was assessed through a combination of one-way and scenario analyses for equivalence scenario and 1-year time horizon. First, deterministic sensitivity analyses were conducted to identify the most influential inputs on the total budget. Second, for scenario analyses, the most plausible alternative values were used. Scenario analyses performed included: (a) alternative body surface area (1.8–1.89 m2); (b) alternative body weight (70–76.3 kg); (c) replace the treatment duration based on median PFS by separately calculated duration of treatment (DoT) curves; (d) use an alternative parametric distribution such as the second-best fit curve; and (e) assume a 100% mortality rate on progression and thus, only consider second-line treatment costs.