A cost-effectiveness analysis17 was carried out by adapting the economic model proposed by Dorian et al18 to the Spanish health care setting. This model covers the life-long course (with a life expectancy of 80.4 years for both sexes) of a cohort of 1000 patients with NVAF18,19 whose characteristics were those of the patients in the ARISTOTLE trial.16 It is a Markov model (Figures 1 and 2), with 6-week cycles and 10 major health states. The entire patient cohort was in a state of NVAF and, after each cycle, they remained in NVAF or passed to other states, in accordance with certain transition probabilities.

Figure 1.

Markov economic model of stroke prevention in patients with nonvalvular atrial fibrillation. Course of a patient with nonvalvular atrial fibrillation. Patients who progress to the state of nonvalvular atrial fibrillation without anticoagulant, return to nonvalvular atrial fibrillation. Patients who progress to states of myocardial infarction or systemic embolism can die a posteriori as a consequence of these complications. AC, anticoagulant; CRNM, clinically relevant nonmajor; ICH, intracranial hemorrhage; MI, myocardial infarction; NVAF, nonvalvular atrial fibrillation; SE, systemic embolism. *Change of treatment to acetylsalicylic acid.

(0.38MB).

Figure 2.

Markov economic model of stroke prevention in patients with nonvalvular atrial fibrillation. Course of a patient who has had a stroke (ischemic or hemorrhagic).

(0.31MB).

These probabilities of transition from one health state to another (stroke, systemic embolism, bleeding, death) for the drugs under comparison were obtained from the ARISTOTLE clinical trial16 and from other sources when necessary.20–27 According to the model, each patient can experience only 1 complication per cycle and only 1 recurrence of stroke (ischemic or hemorrhagic) is permitted. The simulation of avoided episodes was modeled throughout the patient's lifespan, based on the intention-to-treat analysis in the ARISTOTLE trial. Table 1 shows the rate of complications/100 person-years and the hazard ratio of these complications with acenocoumarol vs apixaban, adjusted according to the CHADS2 (congestive heart failure, hypertension, age, diabetes, stroke [doubled]) score, and the international normalized ratio (INR) quality control index according to the time in therapeutic range (TTR).

Overall mortality in women and men older and younger than 75 years was obtained by adjusting published Spanish mortality data19 to a Gompertz function, for the purpose of predicting all-cause mortality and excluding that due to stroke and hemorrhage in order to avoid counting them twice. Treatment with acetylsalicylic acid was considered after interruption of previous anticoagulation therapy (acenocoumarol or apixaban) due to major bleeding or other causes.22

For example, with respect to the method of extrapolating the results for acute myocardial infarction from the ARISTOTLE clinical trial16 to the patient's entire lifespan, calculations were based on the risks of having an infarction with apixaban and with warfarin obtained in the clinical trial, with a rate of 100 person-years of 0.53 and 0.61, respectively. These annual risks were transformed into a risk model for a 6-week period (the duration of the cycles of the Markov model). Thus, every 6 weeks, a percentage of patients with NVAF from a hypothetical cohort of 1000 will have an infarction. Some of these patients will die as a consequence of this complication and the remainder will survive. Those patients who have had an infarction complete the simulation at this time and are counted as failures in prevention, with the costs and utilities associated with the state of acute myocardial infarction.

Differences in the effectiveness of the treatments were measured in life years gained and in quality-adjusted life years (QALY) gained. The utilities of the different health states used to calculate the QALY were obtained from a study performed in the United Kingdom based on the EQ-5D (EuroQol 5D) questionnaire in patients with AF28,29 (Table 2). Determination of the loss of quality of life associated with the need for monitoring required by the use of acenocoumarol was based on a study by Gage et al30 (Table 2).

The per unit use and cost of resources were obtained from Spanish sources (in 2012 euros). The analysis was carried out from the perspective of the Spanish National Health System (considering only the direct health care costs) and of society (including direct nonhealth care costs, as well). The cost of the medications (RRP [recommended retail price] + VAT [value added tax], with a 7.5% reduction)31 was calculated by considering 10 mg/day (5-mg doses twice daily) of apixaban32 and 5mg/day of acenocoumarol (the daily dose recommended by the World Health Organization). The cost of INR monitoring was estimated to be € 320 a year, according to a study by De Solà-Morales Serra and Elorza Ricart.33 It was considered that all the patients, regardless of their treatment, visited their primary care physician every 3 months for routine follow-up of the AF. The cost of the complications of NVAF during the acute phase was obtained by taking the mean of the retail prices corresponding to diagnosis-related groups in the Spanish autonomous communities; the cost of treatment after the acute phase of the complications was obtained from other Spanish sources7,34–38 (Table 3). The direct nonhealth care costs included in the analysis from the societal perspective consisted of those derived from informal care (aid to the dependent patient and home adaptations) of patients who have had an ischemic or hemorrhagic stroke7 (Table 3). An annual discount rate of 3.5% was applied for health care costs and results.

We analyzed a base case with the mean values of all the parameters. To confirm the stability of the results, simple univariate sensitivity analyses were performed (in each sensitivity analysis, the baseline value of a variable was modified, applying its extreme values) for the main variables of the model (Table 1). One of those analyzed was the percentage of TTR of the study population. According to the ARISTOTLE trial,16 in the base case of the analysis, 25% of the patients were equally distributed among the 4 established TTR intervals (< 52.38%; from ≥ 52.38% to < 66.02%; from ≥ 66.02% to < 76.51%, and ≥ 76.51%). In this respect, sensitivity analysis of the extreme values considering that 100% of the patients were in the worst (< 52.38% in TTR) or best (≥ 76.51% in TTR) of the intervals was carried out. Sensitivity analyses were also performed for acenocoumarol and apixaban, taking the extreme values of variables such as risk of stroke and intracranial hemorrhage, the probability of hospital admission, the cost of each visit for INR monitoring for acenocoumarol, and the loss of utilities associated with the use of acenocoumarol or apixaban (Figure 3). As the reduction in the risk of myocardial infarction vs warfarin was not statistically significant, a sensitivity analysis was carried out considering that apixaban did not reduce the number of myocardial infarctions vs warfarin, that is, taking the same rate of infarctions (0.530/100 person-years) for apixaban and warfarin. Finally, a probabilistic analysis was performed (Monte Carlo simulation with 2000 simulations in the 1000-patient cohort) which included, at the same time, all the variables of the analysis, adjusted for the appropriate statistical distributions in each case (beta for the transition probabilities and the utilities and gamma for costs). When the model simulated the course of a hypothetical cohort of 1000 patients with NVAF, the course of each patient in that cohort was different, since the modifications in all the variables of each patient (eg, the hazard ratio of stroke, myocardial infarction, systemic embolism, intracranial hemorrhage, costs, utilities, etc) occurred randomly in accordance with the previously defined statistical distributions. Thus, an attempt was made to imitate the real clinical course of the patients.

Figure 3.

Deterministic sensitivity analysis from the perspective of the Spanish National Health System. Tornado diagram (the extreme values used are indicated to one side). Ratio of incremental cost-effectiveness (euros per quality-adjusted life years gained with apixaban vs acenocoumarol) (in 2012 euros). The figure shows the values of the base case and the extreme values used in the sensitivity analysis. The risks are expressed as rate/100 patient-years. Mortality is the rate of death during the clinical trial. CV, cardiovascular; ICER, incremental cost-effectiveness ratio; ICH, intracranial hemorrhage; TTR, time in therapeutic range.

(0.38MB).

According to the results produced by the model applied to a cohort of 1000 patients with NVAF, apixaban would prevent a considerable number of complications compared with acenocoumarol: 18 strokes (4 ischemic and 14 hemorrhagic), 71 hemorrhages (12 of which would be intracranial and 16 would be cases of major bleeding), 1 systemic embolism, and 2 myocardial infarctions. In contrast, apixaban would prevent 23 deaths related to thromboembolic or hemorrhagic episodes (Table 4).

Consequently, for each patient treated with apixaban rather than acenocoumarol we would obtain more life years (0.187 life years gained) and an increase in QALY (0.194 gained). The incremental costs per patient treated with apixaban would be € 2488 for the Spanish National Health System and € 1826 for society. This difference in costs is mainly due to the more marked decrease in mortality obtained with apixaban and its higher purchase price. The cost per life years gained for the Spanish National Health System and for society was € 13 305 and € 9765, respectively. The cost per QALY gained was € 12 825 and € 9412, respectively (Table 5). These results indicate that, compared with acenocoumarol, apixaban is cost-effective in the prevention of stroke in patients with NVAF in Spain, as the costs per life years gained or for QALY gained are below € 30 000, which is the threshold that is generally accepted in Spain.39

The sensitivity analyses confirmed that apixaban is a cost-effective treatment compared with acenocoumarol. The changes in the most sensitive parameters of the study, such as the risks of complications with one treatment or the other, the loss of utilities associated with INR monitoring with acenocoumarol, and the costs of follow-up visits, do not affect the results of the analysis since, in every case, the costs of gaining 1 life year and the cost of gaining 1 QALY with the most effective treatment (apixaban) were below the threshold of € 30 000 (Figure 3).

Considering that 100% of the patients were in the lowest percentage of TTR (< 52.38%), the cost of gaining 1 QALY would be € 7054. In the opposite case (≥ 76.51% in TTR), the cost per QALY gained would increase to € 12 404, in both cases, from the perspective of the Spanish National Health System (Figure 3). Should apixaban fail to reduce the number of myocardial infarctions compared with warfarin, the cost of gaining 1 QALY would be € 14 150.

According to the probabilistic sensitivity analysis, the probability of apixaban being cost-effective vs acenocoumarol was 87% (Figure 4).

Figure 4.

Probabilistic sensitivity analysis. The probability that apixaban is cost-effective (cost per quality-adjusted life year gained < € 30 000) compared with acenocoumarol is 87%.

(0.39MB).

In the assessment of these results, we must take into account the fact that this is a theoretical model that is, by definition, a simplified simulation of reality. Pharmacoeconomic models enable the creation of economic simulations of complex drug-related health care processes. They are especially useful when the intention is to simulate the course of a disease beyond the period of clinical tests.17 Markov models with Monte Carlo simulation are preferred over deterministic models when dealing with chronic diseases like NVAF, as they are better at simulating the course of the disease over the long term. The Markov deterministic model is based on the course of a hypothetical patient cohort. However, the Monte Carlo probabilistic analysis simulates the course of individual patients, based on the values of randomly chosen variables, which more closely resembles clinical reality. The data on efficacy and on adverse effects utilized in the model were obtained from a randomized clinical trial involving 18 201 patients with NVAF,16 providing a high level of evidence.

One of the main suppositions of the model is the therapeutic equivalence between warfarin and acenocoumarol. Thus, we adopted the effectiveness values obtained for acenocoumarol in the ARISTOTLE clinical trial, which compared it with warfarin.16 This assumption can be considered a limitation of the study, given that, although the 2 drugs could have similar effectiveness in clinical practice,12 an observational study found no significant differences in the quality of INR control, but did report a longer duration of the TTR in the patients treated with acenocoumarol than in those treated with warfarin (37.6% vs 35.7%; P = .0002).13 Nevertheless, in other previously published Spanish economic studies, acenocoumarol and warfarin were considered to be perfect substitutes for each other, both in efficacy and safety and the in the resource use associated with each.45,46

The costs assumed for the acute phase of the strokes and hemorrhages in the model were obtained from the diagnosis-related groups published by the Spanish autonomous communities, prices that could differ from the real direct costs of the processes.

The cost of the drugs was calculated considering a dose of 5mg/d of acenocoumarol, the daily dose recommended by the World Health Organization. Nevertheless, it should be considered that this premise serves only for the calculation of the cost of treatment with acenocoumarol, a variable that does not significantly affect the result because of the low price of the drug.

The utilities of the model were obtained from a study performed in the United Kingdom based on the EQ-5D questionnaire in patients with AF.28,29 Concerning the validity of these utility data corresponding to the Spanish population, a study based on 83 000 evaluations of 44 health states, with the EQ-5D applied in 6 European countries (including Spain) observed wider variability between individuals than between countries.47

Despite the aforementioned limitations of the model, the results obtained concur, as the result favoring apixaban is maintained both in deterministic sensitivity analyses and in the Monte Carlo analysis, with a high probability (87%) that apixaban is more cost-effective than acenocoumarol.

The present report is the first analysis of the cost-effectiveness of apixaban in NVAF performed in Spain. Two analyses comparing the cost-effectiveness of apixaban and warfarin in NVAF have recently been published in the United States.48,49 These studies found that apixaban would be cost-effective in 98% and 62% of the simulations for primary and secondary stroke prevention, respectively.

Two economic analyses of oral anticoagulants vs acenocoumarol in the prevention of stroke in patients with NVAF in Spain have recently been published.45,46 From the perspective of the Spanish National Health System, the cost per QALY gained would be € 17 581 with dabigatran5 and € 11 274 with rivaroxaban.46 The results of the economic models of apixaban, dabigatran, and rivaroxaban are not directly comparable, since they differ in the structure of the Markov models used; however, they reflect the fact that the oral anticoagulants available in Spain are cost-effective options in the current Spanish health care context.