The FINE-CKD model was used to calculate the cost-effectiveness ratio (ICER) of finerenone in combination with SoC compared to SoC in patient that represent the FIDELIO DKD trial. The FINE-CKD is a Markov model developed in Microsoft Excel 2016 (Redmond WA, USA) [15, 16].

Patient population

The FIDELIO-DKD trial predominantly included patients with CKD stage 3 or 4 with moderately or severely elevated albuminuria associated with T2D (Table 1) [13]. Dutch patients with T2D are 68.5 years old on average when diagnosed with CKD, which is comparable with the average age in the FIDELIO trial [1, 13].

Table 1 Patient characteristics of the FIDELIO-DKD ITT population [13]Interventions

Finerenone was added to SoC and compared to SoC alone. The SoC was based on the weighted average of background treatment over the time horizon in the FIDELIO-DKD trial (Additional file 1). Patients used a mix of ACEIs, ARBs, Beta-blockers, diuretics, calcium antagonists, and glucose-lowering therapies. As seen in the FIDELIO-DKD trial, patients were anticipated to carry a discontinuation risk of 0.03 per cycle for finerenone treatment. In addition, after initiation of RRT, all patients were assumed to discontinue finerenone treatment, and 25% of patients to discontinue SoC treatment due to the increased risk for hyperkalemia.

Model structure

The modelled discrete health states were defined in accordance with the CKD stage and history of CV events (Fig. 1). The model outcomes are validated to adequately reflect the clinical data and outcomes of other models [16]. Four stages of CKD health state progression were considered: CKD 1/2, CKD 3a/b, CKD 4, and CKD 5 without renal replacement therapy (RRT). Two health states for CKD5 or ESKD patients with RRT were considered: CKD 5 and dialysis, and CKD 5 and transplantation. In the absence of differentiated costs and outcomes, distinguishing between CKD 1 and CKD 2 or CKD 3a and 3b patients proved impossible. Patients resembled the trial population and entered the model in one of the CKD stages without CV events [13]. Patients remained in the same CKD stage for a cycle duration of four months or moved to either a more or a less advanced CKD stage while, at the same time, experiencing their first modelled CV event (non-fatal MI, non-fatal stroke, hospitalization for HF, or death). Transition probabilities were based on patient-level data of the FIDELIO-DKD trial; therefore, the model was not limited to CKD deterioration, and patients could move through multiple CKD stages per cycle. Once a CV event occurred, patients moved to the acute CV event health state for one cycle to account for the short-term impact of the CV event and then moved to the post-acute CV event health state for the rest of the model duration.

Fig. 1

Model structure Key: This figure depicts the modelled CKD in detail. The left panel depicts the CKD health states in which patients start the model. Patients can experience CKD progression and/or experience a CV event. When patients experience a CV event, they move to the acute CV event panel for one cycle and to the post-acute CV panel for the rest of the model duration. OHEs can occur in any depicted health state. A patient can move to the death health state from every health state. aCV events include a non-fatal stroke, non-fatal MI, and hospitalization for HF. bOHEs include a subsequent CV event, hyperkalemia leading to hospitalization, hyperkalemia not leading to hospitalization, and a new onset of atrial fibrillation/atrial flutter. CV cardiovascular, CKD chronic kidney disease, HF heart failure, MI myocardial infarction, OHE other health event, RRT renal replacement therapy

Apart from CKD progression and CV events, other health events (OHEs) were included in the model to account for additional relevant clinically meaningful outcomes seen in the FIDELIO-DKD trial (Additional file 2) [13]. OHEs included a subsequent CV event, hyperkalemia leading to hospitalization, hyperkalemia not leading to hospitalization, and a new onset of atrial fibrillation/atrial flutter. The risk of OHEs was dependent on the history of CV events; however, to reduce model complexity, OHEs were not modelled as discrete health states and did not affect downstream risks. Instead, their only impact was on costs and utility for one cycle length. Similar to acute CV events, OHEs were modelled for one cycle length.

In line with the observed mortality in the FIDELIO-DKD trial, three different reasons for death were implemented in the model (i.e., CV, renal, and mortality from other causes [not CV or renal related]) [13]. The 4 month cycle length was consistent with the primary endpoints in the FIDELIO-DKD trial and was supported by the fact that CKD associated with T2D is a chronic disease [13]. The Dutch guideline for economic evaluations in healthcare recommends a lifetime time horizon with a 100 year maximum age [18].

Transition probabilities

The transition of patients between health states was dependent on the probability of CKD progression, CV events, death from renal and CV causes, and OHEs (Table 2) [13]. These probabilities were derived from patient-level data of the FIDELIO-DKD-ITT population [13]. While the FIDELIO-DKD trial was designed to assess composite renal and cardiovascular outcomes, demonstrating a significant positive effect of finerenone, different HRs were used to incorporate the influence of finerenone on the model treatment pathways to prevent double counting. The HRs for CV events, dialysis, CV, and renal death of the FIDELIO-DKD trial were used to adjust for the effect of finerenone on CKD progression and CV events (Table 3) [13]. Additionally, we used the HR for the sustained eGFR decrease < 15 mL/min/1.73 m2 to adjust for the effect of finerenone on progression to CKD 5. As the median follow-up duration in the FIDELIO-DKD trial was 2.6 years, HRs to account for the longer-term risk of a CV event (Additional file 3), increased risk of renal and CV mortality, and occurrence of the first modelled CV event (Additional file 3) were derived from literature [19,20,21]. Mortality from causes other than renal and CV events was retrieved from Dutch Central Bureau of Statistics data and adjusted for the proportion of deaths caused by CV events and CKD (Additional file 4) [22].

Table 2 Transition probabilities: CKD progression and first modelled CV event and OHE probabilities [13]Table 3 HRs used to reflect the effectiveness of finerenone [13]Utilities

Quality-adjusted life years (QALYs) were calculated to express the effect of finerenone on life years gained corrected for quality of life. Utility values measure the health-related quality based on preference values attached to the patient’s health status. Utility values were scaled between 0 (equal to death) and 1 (equal to perfect health). Utility values were derived from the FIDELIO-DKD trial for the health states and OHEs with the EuroQol five-dimension questionnaire (EQ-5D, specifically the five-level EQ-5D-5L) using a multilevel mixed repeated measurements model and the Dutch EQ-5D value set [13, 23]. As the number of patients who experienced RRT in the FIDELIO-DKD trial was low, a systematic literature review (SLR) was undertaken to estimate the disutility during dialysis and transplantation [24]. Utility values were adjusted for age, using the population norms for the Netherlands [25]. Additional file 5 presents the baseline utility and utility decrements for the different health states used in the model. All utilities were discounted at 1.5% per year, following Dutch guidelines for economic evaluations in healthcare [18].

Costs

Costs within the healthcare system (i.e., medical costs) and costs for patients and caregivers (i.e., indirect non-medical costs) were included in the model in accordance with Dutch guidelines [18]. They were mostly based on Dutch literature and inflated to March 2023 prices [26], with an applied discount rate of 4% per year [18].

Costs within the healthcare system

Drug costs for finerenone and the current SoC were based on list prices per defined daily dose (Table 4, additional file 1) [13, 27]. We assumed that finerenone treatment discontinuation impacted both costs and effects by considering the same transition probabilities as patients treated with SoC. Treatment discontinuation of SoC only impacted costs. All treatment discontinuation assumptions were validated by clinical experts.

Table 4 Overview of model inputs for costs within the healthcare system

Health state costs associated with CKD 1/2, CKD 3a/b, CKD 4, and CKD 5 were calculated using a bottom-up approach validated by clinical experts. Resource use was based on CKD progression as well as the NHG and NIV guidelines, incorporating visits to the GP, outpatient visits, eGFR and albuminuria assessments, treatment with an ACE inhibitor, an influenza vaccine, and risk for a hospital admission unrelated to CV outcomes [5] (Additional file 6). Costs of ESKD with dialysis or transplantation, CV events, and OHEs were based on the literature and adjusted for the four-month cycle [28,29,30,31]. For dialysis and transplantation, both direct and indirect medical costs were considered based on Dutch health insurance claims [28]. Indirect costs included healthcare, medication, medical devices, and transportation (Additional file 7). It was assumed that patients with mild hyperkalemia (not leading to hospitalisation) were treated with either calcium polystyrene sulfonate for 40 days, sodium polystyrene sulfonate for 40 days, sodium zirconium cyclosilicate for 106 days, or patiromer for a full cycle, according to the Dutch Medicine and Resource Information Project (Genees- en hulpmiddelen Informatie Project, GIP) databank [32]. It was further assumed that patients with severe hyperkalaemia were admitted to an intensive care unit for one day and then transferred to a general ward for a two-day admission (i.e., 20% of all admitted cases). Additionally, 10% of all patients experienced acute dialysis. The other 80% who require hospitalization were admitted to a general ward for three days on average.

Costs for patients and caregivers

The base case analysis incorporated costs for informal care and productivity losses, both based on literature and expert opinion (Table 5). Due to the low estimated impact, travel costs were not factored in. Productivity losses and informal care were considered during CKD stages 3–5, acute and post-acute CV events, dialysis, and kidney transplants for patients below the Dutch retirement age (i.e., 67 years) [38,39,40]. To calculate productivity losses, the friction cost method was used, as outlined in the Dutch costing guideline [35]. All productivity losses were valued at the hourly rate and vacancy data stemming from 2022 [41]. After initiation of dialysis and transplantation, a certain percentage of patients was estimated to be on long-term sick leave [40]. Although the friction method indicated that a vacancy in 2023 should be filled in approximately 20 weeks, our model structure allowed us to incorporate a maximum duration of sick leave for a full cycle (i.e., 12 weeks) in acute dialysis and transplantation health states [35]. In addition to sick leave, short-term production losses were taken into account for each stage of CKD progression, dialysis and transplantation, CV events and OHEs [38, 40]. Short-term productivity losses were adjusted to the labour participation rate found in Dutch CKD patients [42]. The impact of informal care was estimated by valuing the hours of informal care a patient received per cycle at the home care replacement rate (i.e., hourly wage informal care) based on the Dutch costing manual [35]. In case of the absence of informal care data, we applied the same ratio for informal care during both the acute and post-acute states, as observed in productivity losses.

Table 5 Overview of the model inputs for costs incurred by patients and caregiversAnalysesBase case analysis

In the base case, the model calculates the ICER for a willingness-to-pay (WTP) threshold of €20,000 (Table 6). The Dutch guidelines for economic evaluations in healthcare recommend a WTP threshold of €20,000 for a proportional disease shortfall of 0.10–0.40 QALY and a WTP threshold of €50,000 for a proportional disease shortfall of 0.40–0.60 QALY [43]. An estimated disease shortfall of 0.47 QALY was determined and since this value is considered low within the €50,000 WTP threshold, a conservative approach was taken, and the €20,000 WTP threshold was applied [43]. In addition to the life time horizon, the incremental costs were calculated for a time horizon of 1 to 34 years. Our analysis was described using the CHEERS reporting guidance for health economic evaluations (Additional file 8) [17].

Table 6 Overview of base case model characteristicsSensitivity analysis

The deterministic sensitivity analysis (DSA) was performed to assess the impact of the individual input parameters on the ICER by varying them between the lower and upper bounds of their confidence intervals (CIs), which were set at 2.5% and 97.5%, respectively. In addition, a probabilistic sensitivity analysis (PSA) was conducted as extension on the base case analysis to assess the model’s robustness, given the uncertainty around input parameters. That is, the PSA provides a range of results reflecting model uncertainty, whereas the base case assumes certainty around all selected parameters. Input parameters were simultaneously varied across 1000 simulations within their respective 95% CIs. The parameters in the analysis were varied with their respective distributions (normal, beta, gamma, Dirichlet). A standard error of 25% from the deterministic value was applied when the standard error or 95% CI were not available. An overview of all the included parameters, along with their respective CIs and distributions, is presented in Additional file 9.

Scenario analyses

Scenario analyses were performed to establish the impact of several input and model assumptions (Additional file 10). The time horizon was set to ten years and discount rates were varied in line with the Dutch guidelines of economic evaluations. In addition, to assess the impact of patient’s age at baseline, scenarios were performed in which patients were respectively 45, 55, and 68.5 years (i.e., the latter the average age of patients with T2D at diagnosis for CKD) at model initiation. Moreover, the impact of different sources of utility data were separately analysed. The number of patients with more advanced CKD and RRT in the FIDELIO-DKD trial was low [13]; therefore, in the base case analysis, a combination of utility data derived from the FIDELIO-DKD trial data and literature was used to estimate the utility values of patients who experienced dialysis or transplantation. To account for the uncertainty in the utility data, three additional scenario analyses were performed. In the first scenario, solely data from the FIDELIO-DKD trial was considered [13]. Subsequently, a scenario with solely utility data retrieved from the systematic literature review was performed [24, 44, 45]. An additional scenario analysis was performed using utility data previously validated by the Dutch National Healthcare Institute [46]. Additional file 5 presents the utility data incorporated in each scenario.

Our model considered all (in) direct costs related to dialysis and transplantation (e.g., healthcare, transportation, and medication). As indirect costs are a major part of the total cost related to dialysis and transplantation (18–19%), a scenario was run where only direct dialysis and transplantation costs were considered to assess the impact of indirect costs. In the base case, productivity losses were estimated with by the use of various literature sources. To address potential uncertainties in the methodology, a scenario analysis was conducted using alternative literature sources [38, 39, 47].