A Markov model of cost-effectiveness and cost-utility was used, with a six-week cycle length, the minimum period in which changes in health (or disease) state would be expected, following a cohort of 1000 patients over a lifetime horizon. The model, details of which were recently published by Lip et al.10, was programmed in Excel using Visual Basic for Applications (Figure 1).

ICH: intracranial hemorrhage; IS: ischemic stroke; NVAF: non-valvular atrial fibrillation; NVAF subsequent ASA: NVAF patients on second-line aspirin. Reproduced from Lip et al.10'> ICH: intracranial hemorrhage; IS: ischemic stroke; NVAF: non-valvular atrial fibrillation; NVAF subsequent ASA: NVAF patients on second-line aspirin. Reproduced from Lip et al.10' title='Markov model decision tree. AC: anticoagulants; ASA: aspirin; CRNM: clinically relevant non-major; HS: hemorrhagic stroke; ICH: intracranial hemorrhage; IS: ischemic stroke; NVAF: non-valvular atrial fibrillation; NVAF subsequent ASA: NVAF patients on second-line aspirin. Reproduced from Lip et al.10'/>

Figure 1.

Markov model decision tree. AC: anticoagulants; ASA: aspirin; CRNM: clinically relevant non-major; HS: hemorrhagic stroke; ICH: intracranial hemorrhage; IS: ischemic stroke; NVAF: non-valvular atrial fibrillation; NVAF subsequent ASA: NVAF patients on second-line aspirin. Reproduced from Lip et al.10

(0.28MB).

In the model, the natural history of the disease is represented by 11 mutually exclusive health states: non-valvular AF; mild, moderate or severe non-fatal ischemic stroke; mild, moderate or severe non-fatal hemorrhagic stroke; systemic embolism; myocardial infarction (MI); non-valvular AF with discontinued first-line anticoagulation; and death. After six weeks the patient can enter, remain in, or transition to another state according to the corresponding transition probability, defined as the likelihood of an event occurring within that period.

The risk of ischemic stroke is calculated according to the patient's CHADS2 score11 (the method for estimating thromboembolic risk in use at the time of the clinical trials of the drugs under analysis) and the level of anticoagulation for patients treated with warfarin as determined by time in therapeutic range (TTR) of the international normalized ratio (INR). The likelihood of stroke, MI, intracranial bleeding and other major and non-major bleeding increases with age. The model also considers the long-term impact of MI and systemic embolism on mortality, reflected in higher hazard ratios (HR). For patients in the state of non-valvular AF who discontinue first-line anticoagulation, the model structure remains the same but the transition probabilities differ.

Severity of stroke (ischemic or hemorrhagic) is classified according to the modified Rankin scale (mRS)12: mild, 0–2; moderate, 3–4; severe, 5; and fatal, 6. All patients with fatal stroke transition to the state of death in the following cycle, while non-fatal stroke is modeled as a tunnel state from which patients can only transition to recurrent stroke or death. Patients can only experience one recurrent stroke in the model, from which the transition is to stroke of the same or greater severity. The model does not allow recurrent MI or systemic embolism, patients either remaining in the same health state or transitioning to death.

At the end of each cycle health care costs, quality-adjusted life years (QALYs) and life years gained are calculated. Levels of health-related quality of life (utilities), clinical outcomes and mortality rates vary according to stroke severity. In accordance with the Portuguese Ministry of Health's guidelines for economic evaluation studies of drugs,13 published by Infarmed, costs and utilities are discounted at an annual rate of 5%.

In the model's base-case scenario, the characteristics of the population are those of patients enrolled in trials of apixaban, specifically ARISTOTLE,14 in terms of median age (70 years), gender (64.7% male), and distribution of CHADS2 scores (1–2: 69%; 3–4: 27%; and 5–6: 4%).

The results of treatment with apixaban 2.5–5 mg twice daily are compared with (1) dabigatran 150 mg twice daily in patients aged ≤80 years and 110 mg twice daily in patients aged >80 years with high bleeding risk and those treated with verapamil (the dabigatran group) and (2) rivaroxaban 15–20 mg once daily.

Economic evaluations of new health technologies such as drugs analyze their effectiveness and the associated costs compared to existing options. Assessment of the relative effectiveness of the NOACs is thus one of the central aims of this study. There have to date been no head-to-head studies between the NOACs, so their effectiveness in AF must be estimated by indirect analysis using a common comparator, in this case warfarin.

It is therefore essential to assess the reliability of the estimates of effectiveness used in the model. To this end we carried out a systematic review of the literature to identify indirect comparisons between NOACs that provide data on their effectiveness in AF, searching the MEDLINE and Cochrane Library databases in September 2014 using the search terms meta-analysis, indirect comparison, bayesian, network, apixaban, dabigatran, rivaroxaban and atrial fibrillation. Ten studies were identified, six frequentist10,15–19 and four Bayesian (network meta-analyses).6,20–22

Table 1 summarized the characteristics of each of these studies. As can be seen in Figure 2, the estimates for the various outcomes in these publications are consistent and are similar to those used in the base-case scenario in the economic model.10 Given the aim of the present study, Lip et al.10 (frequentist indirect comparison using the method of Bucher et al.23) and Mitchell et al.20 (Bayesian network meta-analysis) probably give the best estimates of the relative effectiveness of the three NOACs in AF, since they use only data from phase III clinical trials and establish associations using HRs, which takes the time factor into account and respects the primary statistical analysis of each trial. Supplementary Figure 3 (Annex) shows the evidence network used for these two indirect comparisons.

ITC: indirect treatment comparison; R: rivaroxaban. *The RE-LY study only presents results for minor bleeding, which were used as a proxy for clinically relevant non-major bleeding.'> ITC: indirect treatment comparison; R: rivaroxaban. *The RE-LY study only presents results for minor bleeding, which were used as a proxy for clinically relevant non-major bleeding.' title='Estimates of effectiveness of apixaban compared to other new oral anticoagulants in published indirect comparisons. A: apixaban; B: Bayesian network meta-analysis; D: dabigatran; ITC: indirect treatment comparison; R: rivaroxaban. *The RE-LY study only presents results for minor bleeding, which were used as a proxy for clinically relevant non-major bleeding.'/>

Figure 2.

Estimates of effectiveness of apixaban compared to other new oral anticoagulants in published indirect comparisons. A: apixaban; B: Bayesian network meta-analysis; D: dabigatran; ITC: indirect treatment comparison; R: rivaroxaban. *The RE-LY study only presents results for minor bleeding, which were used as a proxy for clinically relevant non-major bleeding.

(1.03MB).

The event rates in the base-case scenario are derived from the HRs reported by Lip et al.10 (Table 2). The distribution of stroke events by severity is presented in Supplementary Table 3.

As stated above, the risk of ischemic stroke and bleeding events associated with warfarin depends on the level of anticoagulation as determined by INR (Supplementary Table 4). The model classifies patients in four categories according to various cutoffs for median center time in therapeutic range (cTTR) based on the results of centers participating in the ARISTOTLE trial. The distribution is uniform, with 25% of patients in each category.

To parametrize the model to reflect the situation in Portugal, we used data from a convenience sample of patients attending the anticoagulation clinics at two hospitals in the Lisbon region, Centro Hospitalar Lisboa Central and Hospital Fernando da Fonseca. These data are from 2011 and 2012 and are on patients who underwent at least 10 INR measurements, with a total of 39630 measurements in 2850 patients, and were used to calculate patient median TTR. The median cTTR could not be estimated since the data are limited to two hospitals, but the above median TTR was considered a reasonable approximation to the cTTR defined in the model. The robustness of the results obtained from this hospital sample was checked by comparing them with a sample of 4470 outpatient INR measurements in 233 patients; no statistically significant differences were found.

On the basis of these data, anticoagulation levels in the Portuguese population are lower than considered in the model, since only 44.5% of Portuguese patients have TTR ≥52.38%, as opposed to 75% with ≥52.38% in the trials (Supplementary Table 5).

Treatment discontinuation rates (%/year) due to non-vascular causes were obtained from a secondary analysis of the ARISTOTLE trial results (13.2% with apixaban and 14.4% with warfarin), assuming constant rates over time. Supplementary Table 6 shows HRs for treatment discontinuation for reasons other than vascular events. Second-line treatment was assumed to be aspirin. Absolute risks associated with events per 100 patient/years are summarized in Supplementary Table 7.

The study adopts the perspective of the NHS and therefore does not analyze indirect costs. Three main types of costs are identified in the model: costs arising from vascular events, costs of anticoagulant therapy, and costs of monitoring and/or routine consultations. Costing is based on (1) Order in Council 20/201424 for unit costs of consultations, diagnostic exams and diagnosis-related groups (DRGs); (2) analysis of the database of NHS hospitalizations (DRGs) in 201325; (3) the Portuguese Ministry of Health's drug database (Infomed) for prices of medications, consulted on January 2, 201526; and (4) estimates outpatient care resource use by a geographically representative expert panel of various specialists. For the health states of non-fatal ischemic or hemorrhagic stroke, MI and systemic embolism, costs were divided into acute and long-term maintenance, the acute phase including the first two weeks of hospital stay and the first three months of rehabilitation. The model assumes that the maintenance stage will continue until death and according to the expert panel, includes costs of consultations, emergencies and transport, diagnostic exams, medication and technical assistance. It was not possible to estimate the costs of stroke according to severity (mild, moderate or severe), since there are no data on costs according to the mRS in Portugal. For the other health states only the costs of hospitalization (acute phase) were included.

Overall costs per event, treatment costs and costs of monitoring and routine care are shown in Table 3.

The probabilities of death associated with vascular events in the model are those observed in the trials of the NOACs, with the exception of the fatality rate in MI, which was obtained from Scarborough et al.27 The model assumes that these probabilities are independent of treatment. For the period corresponding to the duration of the ARISTOTLE trial, mortality from non-vascular causes is assumed to be the same for all three NOACs, and the figures in the ARISTOTLE trial (3.08% for apixaban and 3.34% for warfarin) was used in the model. Mortality after the period analyzed in the clinical trials was estimated on the basis of Portuguese life tables,28 multiplied by the HRs associated with the population with AF estimated by Friberg et al. to take into account the increased risk of this population.29 Specifically, the parameters of a Gompertz survival function were calculated by age-group (<75 years and ≥75 years) and by gender. The model includes adjustments to mortality risk to account for the increased mortality associated with AF and different degrees of stroke severity (Supplementary Table 9).

The mean values for utilities and disutilities associated with different health states were taken to be the same as those estimated for the UK population by Sullivan et al.30 There are also disutilities associated with warfarin therapy (unlike the NOACs) and with other vascular events. The model assumes that these disutilities are cumulative. Table 4 summarizes the utilities and disutilities used in the model.

One-way sensitivity analyses were performed to assess the robustness of the results in terms of the following parameters: (1) use of the HRs estimated by Mitchell et al. (Bayesian network meta-analysis) instead of those estimated by Lip et al.10; (2) anticoagulation levels as reported in the clinical trials, instead of those obtained in Portuguese patients; (3) duration of the acute phase of hospitalization taken to be six rather than two weeks; (4) different costs of stroke depending on severity, with weighting calculated on the basis of UK figures, instead of a uniform cost for stroke of any severity; (5) the same distribution of stroke of similar severity for all NOACs (based on the distribution in the case of apixaban); (6) the same treatment discontinuation rates for non-vascular causes for all comparators as for apixaban (13.2%/year) from the beginning of treatment, instead of the rates reported in the clinical trials; (7) mortality rates after the period covered by the trials taken to be the same as for the general population, thus underestimating mortality; (8) use of different utilities associated with each health state, as estimated in a previous study by Sullivan et al.,46 and used in other studies of the cost-effectiveness of NOACs32–34; and (9) a discount rate for costs and utilities of 0% or 3% instead of 5%.

A probabilistic sensitivity analysis using 2000 Monte Carlo simulations incorporating second-order uncertainty was also performed.35 The results are presented as the probability of apixaban being cost-effective compared to the other therapeutic options based on a threshold of €20000/QALY, the limit usually taken to be acceptable for funding new health technologies in Portugal.

Table 5 shows the number of vascular events associated with each anticoagulant in a cohort of 100000 patients according to the rates derived from the model. The number of vascular events and event-related deaths was lower with apixaban except for hemorrhagic stroke. The difference was greatest for ischemic stroke, other major bleeding, clinically relevant non-major bleeding and event-related deaths.

Table 6 and Figure 3 present the breakdown of costs associated with the different therapeutic options over a lifetime horizon. Warfarin has the lowest mean cost per patient and rivaroxaban the highest. The total mean cost of apixaban is between these two, with the lowest clinical costs (due to its low vascular event rate) and lowest costs of monitoring and routine care. Although the daily cost of apixaban is lower than dabigatran and rivaroxaban, lifetime costs are greater because the duration of treatment tends to be longer due to its lower discontinuation rate.

Figure 3.

Breakdown of mean total costs per patient for each therapeutic option over a lifetime horizon.

(0.13MB).

Table 7 and Figure 4 show the results of the cost-effectiveness analysis of apixaban compared to the other therapeutic options. As suggested in the literature36,37 for multiple comparisons, the results are presented as a graph in which the x-axis represents the differences in QALYs and the y-axis the differences in cost between the comparators and the reference therapy (in this case warfarin). The red line linking the points on the graph represents the efficient frontier. The frontier consist of three segments: its slope corresponds to €4367/QALY when it joins the points representing warfarin and dabigatran, €9163/QALY when it joins the points representing dabigatran and apixaban, and is vertical above apixaban because no therapy is more effective. Rivaroxaban is dominated because it is to the left of the cost-effectiveness frontier, presenting greater costs and fewer QALYs than other therapies on the frontier. Rivaroxaban is also dominated by apixaban when analyzed in isolation.

NOACs) relative to warfarin (represented by the coordinates 0,0). The red line represents the efficient frontier; the slope of each segment corresponds to the incremental cost-effectiveness ratio between the points defining that segment. NOACs with fewer incremental QALYs are to the left and those with greater incremental costs are higher. The incremental cost of apixaban is €1064 compared to warfarin but it is more cost-effective than the other therapeutic options. Points to the left of the line are dominated by therapies that are more effective than at the frontier, and so rivaroxaban, with fewer QALYs and greater costs, is dominated by apixaban. ICER: incremental cost-effectiveness ratio; QALYs: quality-adjusted life years.'> NOACs) relative to warfarin (represented by the coordinates 0,0). The red line represents the efficient frontier; the slope of each segment corresponds to the incremental cost-effectiveness ratio between the points defining that segment. NOACs with fewer incremental QALYs are to the left and those with greater incremental costs are higher. The incremental cost of apixaban is €1064 compared to warfarin but it is more cost-effective than the other therapeutic options. Points to the left of the line are dominated by therapies that are more effective than at the frontier, and so rivaroxaban, with fewer QALYs and greater costs, is dominated by apixaban. ICER: incremental cost-effectiveness ratio; QALYs: quality-adjusted life years.' title='Effectiveness (measured in quality-adjusted life years) and incremental costs of the new oral anticoagulants (NOACs) relative to warfarin (represented by the coordinates 0,0). The red line represents the efficient frontier; the slope of each segment corresponds to the incremental cost-effectiveness ratio between the points defining that segment. NOACs with fewer incremental QALYs are to the left and those with greater incremental costs are higher. The incremental cost of apixaban is €1064 compared to warfarin but it is more cost-effective than the other therapeutic options. Points to the left of the line are dominated by therapies that are more effective than at the frontier, and so rivaroxaban, with fewer QALYs and greater costs, is dominated by apixaban. ICER: incremental cost-effectiveness ratio; QALYs: quality-adjusted life years.'/>

Figure 4.

Effectiveness (measured in quality-adjusted life years) and incremental costs of the new oral anticoagulants (NOACs) relative to warfarin (represented by the coordinates 0,0). The red line represents the efficient frontier; the slope of each segment corresponds to the incremental cost-effectiveness ratio between the points defining that segment. NOACs with fewer incremental QALYs are to the left and those with greater incremental costs are higher. The incremental cost of apixaban is €1064 compared to warfarin but it is more cost-effective than the other therapeutic options. Points to the left of the line are dominated by therapies that are more effective than at the frontier, and so rivaroxaban, with fewer QALYs and greater costs, is dominated by apixaban. ICER: incremental cost-effectiveness ratio; QALYs: quality-adjusted life years.

(0.08MB).

The results of the one-way and probabilistic sensitivity analyses confirm the robustness of the study's findings. In the one-way analysis of the nine parameters specified in the Methods section, which reflect a range of alternate scenarios, apixaban is always dominant compared to rivaroxaban. Compared to the other therapeutic options, apixaban presents ICERs well below €20000/QALY, ranging between €4909 and €6741/QALY compared to warfarin and between €5162 and €12016/QALY compared to dabigatran. If it is assumed that discontinuation rates for non-vascular causes remain the same from the beginning of treatment, the costs of apixaban are less than either rivaroxaban or dabigatran. In this case, apixaban is dominant compared to rivaroxaban and, for a threshold of €20000/QALY, is cost-effective compared to warfarin and dabigatran. The results of the sensitivity analyses are summarized in Supplementary Table 14.

In the probabilistic sensitivity analysis, the probability of apixaban being cost-effective for a threshold of €20000/QALY is 96%, 87% and 95% compared to warfarin, dabigatran and rivaroxaban, respectively. If all the comparators are considered together (Figure 5), apixaban is the best alternative from a threshold of €8000/QALY. In this scenario, for a willingness to pay of €20000/QALY, the probability of apixaban being cost-effective is 70%.

Figure 5.

Cost-effectiveness acceptability curves showing the percentage of simulations for each willingness-to-pay value that are cost-effective for each treatment, enabling simultaneous comparison between all the therapeutic options. Apixaban is the best alternative from €8000/QALY. For a willingness to pay of €20000/QALY, the probability of apixaban being cost-effective compared to all the other alternatives is 70%.

(0.19MB).

The authors declare that no experiments were performed on humans or animals for this study.

The authors declare that they have followed the protocols of their work center on the publication of patient data.

The authors declare that no patient data appear in this article.