Model Summary
The CREDENCE Economic Model of DKD (CREDEM-DKD) model uses the microsimulation approach to estimate cost-effectiveness of treatment interventions. It is implemented with discrete event simulation (DES) methods, which are well suited to situations where there are competing clinical events and accelerating risks (e.g. of reaching ESRD), and relatively long-term “time-to-event” data are available to enable estimation of risk prediction equations [11, 20], as is the case with DKD and the CREDENCE trial. First- and second-order uncertainty are captured explicitly, and the time horizon and discount rates are user-definable. CREDEM-DKD is implemented in Microsoft® Excel (Microsoft Corporation, Redmond, Washington, USA) and Visual Basic for Applications. CREDEM-DKD was designed, constructed, and populated with patient-level data from CREDENCE in accordance with the recommendations of the International Society of Pharmacoeconomic and Outcomes Research–Society of Medical Decision Making (ISPOR-SMDM) Modeling Good Research Practices Task Force-2 [21]. A pragmatic and parsimonious design was chosen to preserve the information in the CREDENCE trial without creating unnecessary complexity or introducing undue risk for overfitting of the model to the CREDENCE study. This approach included limiting time-varying covariates in the risk equations to cases where ignoring a temporal relationship could not reasonably be avoided (e.g. age) and omitting treatment intensification algorithms. The model structure is presented in Fig. 1.
The model is described in detail elsewhere [19]. Briefly, though, simulated patients are characterised at baseline, and at each time point thereafter by a DKD health state based on National Kidney Foundation–Kidney Disease Outcomes Quality Initiative (NKF-KDOQI) stages (1, 2, 3A, 3B, 4, and 5 prior to dialysis), or a “receiving dialysis” or “post renal transplant” health state. Each of these eight renal health states are mutually exclusive [22, 23]. The renal health states are supplemented with additional health states representing history of MI, stroke, and/or heart failure (HF); and by age, smoking status, and projected eGFR and urine albumin-to-creatinine ratio (UACR) biomarker values. The hypothetical patients experience DKD progression in terms of eGFR decline and UACR increase, based on linear mixed model regression equations from the CREDENCE trial [11] or the occurrence of dialysis start or kidney transplant [24]. Start of maintenance dialysis and/or kidney transplantation can optionally be initiated automatically at a user-defined eGFR threshold. The patients also face risks of all-cause mortality (ACM), nonfatal MI, nonfatal stroke, and HHF, all based on parsimonious parametric time to event risk equations (see Willis et al. [19]). CV can be assigned as cause of death using a logistic regression equation estimated with patient-level CREDENCE data. Treatment can affect eGFR and UACR biomarkers as well as directly modify the risk for dialysis, CV events, and ACM. AEs are defined by an event rate, periodicity (e.g. one-time, repeated, or permanent), duration (e.g. 1 week), and a cost and quality-adjusted life-year (QALY) disutility weight. The user can optionally set a maximum duration after which treatment effects of canagliflozin versus SoC diminish, either immediately with reversion to SoC or linear waning over a user-defined time horizon until the hazard ratios (HRs) between treatment arms return to 1.0.
CREDEM-DKD was validated in accordance with the recommendations of the ISPOR-SMDM Modeling Good Research Practices Task Force-2 [21], which consisted of model verification (i.e. ‘stress testing’), internal validation (i.e. replicating the CREDENCE study results), and limited external validation based on replicating results of the subgroup of CANagliflozin cardioVascular Assessment Study (CANVAS) Program [25] subjects that met CREDENCE study recruitment criteria. External validation will be reassessed when other renal outcomes trials report out. Validation is summarised in an Assessment of the Validation Status of Health-Economic Decision Models evaluation (AdViSHE) [26] (see electronic supplementary material [ESM] 2).
Input Data
Baseline patient characteristics were sourced from CREDENCE [11, 20] and are presented in Table 1. Treatment effects for the canagliflozin 100 mg + SoC arm were also sourced from CREDENCE [11, 20] and the details are presented in ESM Table 1. Between-arm separation in the Kaplan–Meier plots for dialysis, stroke, and mortality first occurred visually after 1 year; so to ensure a conservative approach, treatment effects for these endpoints were not applied during the first year. SoC alone is associated with clinical progression defined by the risk prediction equations in CREDEM-DKD, which are based on the placebo arm in CREDENCE. AEs were included only for the canagliflozin arm (loaded as between-arm difference in CREDENCE) and consist of urinary tract infection, genital mycotic infection, diabetic ketoacidosis, and lower extremity amputation (LEA). The event rates are presented in ESM Table 1. Canagliflozin was discontinued when dialysis is started, kidney transplant occurs, or diabetic ketoacidosis or LEA is simulated to occur, and affected hypothetical patients were reverted to SoC treatment. The price of canagliflozin 100 mg was set at £476.93 annually, which corresponds to 100% patient compliance at UK list price [27]. The price of SoC was calculated on the basis of use of background therapy in CREDENCE, separately by arm (£259.40 annually for canagliflozin and £259.18 annually for the SoC alone arm). (See ESM Table 2 for description of SoC unit costs).
Table 1 Baseline patient characteristics The default risk equations were used to model the risks of starting dialysis, CV events, and death. However, as a result of the short follow-up period in the CREDENCE trial, the event rates for initiation of dialysis or renal transplant were relatively low, with the majority of ESRD events defined via a persistent eGFR of below 15 mL/min/1.73 m2. To enhance realism by overriding scenarios in which hypothetical patients could reach implausibly low eGFR without starting RRT, a minimum eGFR threshold of 6 mL/min/1.73 m2 was set, at which all hypothetical patients would immediately be assigned to start dialysis. The threshold of 6 mL/min/1.73 m2 was selected conservatively in line with both clinical guidance on initiation of dialysis [28] and actual retrospective UK data showing that the mean eGFR at start of RRT is 7.4 mL/min/1.73 m2 [29]. For patients simulated to reach stage 5 DKD or to start dialysis, a 7.5% annual probability of receiving a kidney transplant was assumed [24].
Unit costs (inflated to £2019) and QALY weights were sourced from the literature (Table 2). The widely used UKPDS cost estimates [30] were used for individual CV complications. Kerr and colleagues [1] was used for cost of dialysis and kidney transplant. Costs for DKD stages were sourced from a previous National Institute for Health and Care Excellence (NICE) Single Technology Appraisal [24]. Health-related quality of life data were not collected in CREDENCE, and utility values used to inform kidney disease stages in the aforementioned appraisal [24] were critiqued for their lack of generalisability to the UK. A targeted literature review was therefore conducted to identify utilities to inform the DKD stages 1–5 and RRT health states that would be reflective of UK patients (see ESM 1 for description of the search strategy). Jesky and colleagues [31] was selected as the most relevant and valid source of QALY weights from a UK perspective for DKD stages 1–5, whereas Lee and colleagues [32] was selected as the most relevant and valid source for dialysis and kidney transplant. Costs and QALYs were discounted at 3.5% annually. All analyses consisted of simulating 500 cohorts of 500 hypothetical patients and the time horizon in the base case was 10 years (capturing much of the economic and health consequences of treatment for this patient population but not engendering undue uncertainty by extrapolating too far beyond the duration of the CREDENCE trial).
Table 2 Unit costs and disutility weights Sensitivity Analysis
Eight carefully chosen sensitivity analyses were conducted to evaluate the robustness of model results to changes in essential model parameters and assumptions (see details in Table 3):
-
Longer decision-making horizons of 20 years and 40 years (for this patient group, closely reflecting the NICE preferred lifetime horizon) and a shorter time horizon of 5 years
-
Waning canagliflozin direct treatment effects (HRs) on renal, CV, and mortality outcomes starting at year 6 (consistent with the longest demonstrated duration of effects in the maximum follow-up of the CANVAS trial [25, 39]) and continuing linearly to end of simulation
-
Including 1st year treatment effects (HRs) for stroke, dialysis, and mortality
-
Removal of the assumed eGFR fail-safe floor of 6 mL/min/1.73 m2, below which dialysis is started (if not already ongoing)
-
Assuming same eGFR trajectory for canagliflozin 100 mg and SoC arms
Table 3 Economic outcomes for canagliflozin 100 mg + SoC versus SoC alone over 10 years Compliance with Ethics Guidelines
This article is based on previously conducted studies and does not contain any studies with human participants or animals performed by any of the authors.