The characteristics of the population used for the base-case analysis were the same as in the ENGAGE AF-TIMI 48 trial,19 namely, a diagnosis of NVAF, a moderate to high risk of stroke (CHADS2 stroke risk score ≥ 2), a higher proportion of men than women (62.29%), and a mean age of 71 years.

We also compared high-dose edoxaban with VKA therapy,19 but instead of warfarin we analyzed acenocoumarol, the most widely used VKA in Spain,4 and assumed that the 2 drugs had identical safety and efficacy.

We estimated the mean per-patient cost-effectiveness of each treatment, calculated the incremental cost-effectiveness ratio of edoxaban vs VKA therapy, and determined the additional cost required to gain 1 quality-adjusted life year (QALY) with edoxaban.

The costs were analyzed from the perspective of the Spanish national health system and included direct medical costs (calculated in 2017 euros). The time horizon contemplated in the analyses was the patient's full lifetime. An annual discount of 3% was applied for costs and future effects.20

A group of experts selected for their experience and familiarity with clinical practice was formed to help adapt the necessary aspects of the study to Spain. The members of this group filled in individual questionnaires and were then presented with a base case for evaluation.

A Markov model was used to simulate the course of events in patients with NVAF and a moderate to high risk of stroke through the analysis of different clinical situations (health states) associated with the use of different resources and specific clinical outcomes. A simplified version of this model is shown in Figure 1. The transition between health states and the resulting clinical/economic outcomes were evaluated in monthly cycles. The probability of a patient progressing from a milder to a more serious state, depending on the case, was derived from the clinical outcomes of the treatments.

Figure 1.

Markov model showing clinical events in patients with NVAF. AMI, acute myocardial infarction; NVAF, nonvalvular atrial fibrillation; TIA, transient ischemic attack.

(0.14MB).

At the start of the analysis, the patients had a stable NVAF status, and during each cycle they transitioned between the following health states: hemorrhagic or ischemic stroke (mild, moderate, or severe), systemic embolism, and acute myocardial infarction. These events were associated with an initial impact (acute event) and a long-term impact (postevent)

Other events (complications) considered were bleeding (intracranial, major extracranial, and clinically relevant nonmajor bleeding) and transient ischemic attacks. These were possible in any of the health states.

A stroke could be followed by new events (recurrence), complications, and temporary or permanent interruption of treatment (with a treatment switch).

Risk of death varied according to the demographic characteristics of each patient (age and sex) and their health status.

The clinical data for the model were mainly obtained from the phase III ENGAGE AF-TIMI 48 trial19 and available scientific evidence.

The monthly transition probabilities are shown in Table 1. It was assumed that the risk of an event increased with age.

The monthly risk of stroke recurrence is 0.25% for ischemic stroke and 0.26% for hemorrhagic stroke.25 This risk increases to 0.44% following an acute myocardial infarction.25

Data on treatment interruption and switching following a stroke were based on the opinions of the expert group (Table 2).

Three types of mortality were considered for the analysis: overall mortality26 (adjusted for the risk associated with AF, hazard ratio [HR] = 1.3427), mortality due to an acute event, and mortality after surviving an acute event (Table 3).19,28–32

Quality-of-life utility scores in relation to health states fell on a scale of 0 (death) to 1 (perfect health). The utilities for each state and the loss of utility due to complications were taken from the literature (Table 4).33–38

Despite having a stable INR, patients being treated with a VKA may experience a slight deterioration in quality of life due to anxiety about remaining within the therapeutic range and potential interactions between the drug and other medication or food.39 For the base case, we considered a loss of utility of 0.012 in these patients.40

The direct medical costs considered were pharmacologic costs, VKA monitoring costs, and the treatment costs for each event.

Daily pharmacologic costs were calculated according to the established treatment regimens and retail prices,41 with application of the corresponding deduction.42 The resulting estimates were a daily cost of €2.58 for edoxaban (60mg/d) and €0.07 for acenocoumarol (2.75mg/d).

VKA therapy requires regular INR monitoring (14 tests a year43) and possible dose adjustments. According to the expert group, 50% of these tests would be performed in an in-hospital setting and 50% in an outpatient setting. The estimated costs were €20.15 and €30.89, respectively.44 The annual cost of INR monitoring was therefore set at €357.28.

The costs per event type were calculated using data from the relevant literature and opinions expressed by the expert group (Table 5).

To characterize the uncertainty surrounding the study input parameters and their potential influence on outcomes, we performed a probabilistic sensitivity analysis using second-order Monte Carlo simulation with a hypothetical cohort of 1000 patients.20 The main parameters from the base-case analysis were varied simultaneously and assigned the following distributions: normal distribution for age of onset; Dirichlet distribution for stroke severity; beta distribution for mortality due to an acute event, utilities, and probabilities (of events and recurrences); log-normal distribution for HRs, mortality following an event, adjustment factors for overall mortality, and risk for each decade of life; and gamma distribution for costs.

The results of the analysis were plotted on a cost-effectiveness plane, where each point corresponded to 1 simulation run according to the stipulated parameters and distributions. The results were also plotted on an acceptability curve.

Additional sensitivity analyses were conducted for the following scenarios:

• •

  To account for regional variability at INR testing sites, a scenario where 20% of tests are performed in a hospital and 80% in an outpatient setting and vice versa.

• •

  To account for possible overestimation of annual INR testing costs due to the use of potentially outdated data, a scenario contemplating a third of the annual base cost (€119.09).

• •

  A scenario with patients with CHADS2 ≥ 3.48.

• •

  A scenario with patients with a time in therapeutic range ≥ 60%.48

• •

  A scenario in which patients on VKA therapy had no loss of utility.

Time horizons and hence treatment durations are important considerations when evaluating treatment effectiveness. We therefore evaluated different time horizons for both the base case and the scenarios involving the subgroups of patients with a CHADS2 score of at least 3 and time in the therapeutic range of 60% or more.

Edoxaban was more effective than the VKA (0.34 QALYs and 0.30 life years gained) due to lower rates of stroke, intracranial and extracranial bleeding, systemic embolism, and acute myocardial infarction (Table 6).

Edoxaban was associated with an additional cost of €3916 compared with the VKA. This extra cost was mainly due to the higher price of the drug, but was partially offset by the lower costs of INR monitoring and treatment of events and complications (which are less common with edoxaban) (Table 6). Compared with the VKA, edoxaban was associated with an incremental cost-effectiveness ratio of €11 518 for each QALY gained, which is under the generally accepted threshold for cost-effectiveness (€25 000-€30 000 per QALY).49,50

The results of the probabilistic sensitivity analysis, plotted on the cost-effectiveness plane (Figure 2) and the acceptability curve (Figure 3), show that edoxaban was cost-effective in 73% to 81% of the simulations run.

Figure 2.

Probabilistic sensitivity analysis. QALYs, quality-adjusted life years.

(0.2MB).

Figure 3.

Acceptability curve *Percentage of simulations under the cost-effectiveness threshold analyzed.

(0.15MB).

It was also cost-effective in the different subgroup analyses (see scenario results in Table 7).

Finally, as indicated in Table 8, edoxaban becomes cost-effective after 3 to 5 years of treatment (2-3 years for patients with CHADS2 ≥ 3 and 6-9 years for those with a time in the therapeutic rage ≥ 60%) (Table 8).

The results presented in this study are based on mathematical modeling, and to run the necessary simulations, we had to extrapolate short-term data (collected over a period of 2.8 years in the ENGAGE AF-TIMI 48 trial19) to the lifetime horizon of the current study and assume that the safety and efficacy of edoxaban remained constant over time. This is an inherent limitation that applies to any studies that need to calculate long-term treatment effects based on short-term clinical evidence mainly from phase III trials. Data from phase IV studies would confer greater external validity to these studies. Risk of stroke assessed using the CHADS2 scale, for example, was higher in the ENGAGE AF-TIMI 48 trial than in clinical practice59 (mean score of 2.8 vs 2.3).

The external validity of the current study may also be limited, as CHAD2S2-VASc scores are now used to assess AF stroke risk1 (and not CHADS2 scores like in the ENGAGE AF-TIMI 48 trial).

Other limitations are related to the variability of the data used and assumptions made about missing data. The need to model parameters from different sources is an inherent aspect of studies of chronic disease,60 and data on utilities, for example, often need to be drawn from international studies because of a shortage of national data. That said, in our case these data were applied to the 2 drugs and the relevant result in the analyses is the incremental ratio. We also accounted for potential variations and influences by conducting a probabilistic sensitivity analysis.

We were unable to investigate the potential influence of age as there was insufficient evidence for subgroup analyses. Nonetheless, efficacy and safety outcomes for edoxaban vs warfarin appear to be independent of age.61

One of the main assumptions in our model was the therapeutic equivalence of warfarin and acenocoumarol. Other cost-effectiveness studies performed in Spain have made similar assumptions about efficacy, safety, and resource utilization.47,62,63

Although different criteria were used to account for the loss of utility (or not) associated with VKA therapy in the cost-effectiveness analyses in the above studies,47,62,63 omission of this loss did not affect the results of the analyses.

The efficacy of edoxaban in our study may be underestimated, as in clinical practice, many patients with NVAF are given aspirin to prevent stroke or are not treated.64 Consideration of other types of costs associated with the long-term disability costs of stroke (eg, indirect health costs or productivity loss) would strengthen the cost-effectiveness of edoxaban.

The cost-effectiveness of edoxaban vs VKA therapy decreases with improved anticoagulation control,57 and therefore, in line with current recommendations, edoxaban can be considered to be particularly cost-effective in patients on VKA therapy with poor control (time in therapeutic range <60%).10

Considering the absence of randomized controlled trials directly comparing DOACs in patients with NVAF, differences observed indirectly through network meta-analyses should be confirmed in clinical practice.51 In brief, there are no reliable data for guiding the optimal use of one DOAC or another,65 although all DOACs appear to be a safe and effective alternative to VKAs.17 Phase IV studies will help to evaluate the sustainability of the widespread use of DOACs in the industrialized world.