Numerous studies employ meta-analysis and cost-effectiveness evaluation to compare non-vitamin K oral anticoagulants (NOACs) versus warfarin in patients with nonvalvular atrial fibrillation (AF; [1,2,3,4,5,6]). However, critical gaps in knowledge remain; little information is available comparing these agents when anticoagulation treatment occurs together with antiplatelet therapy. Such evaluation is necessary for three reasons. (1) This combination therapy is appropriate for patients with AF who develop coronary artery disease and require stent placement. The converse event sequence also occurs; patients on antiplatelet therapy who then develop AF and require anticoagulation. For example, combination therapy with clopidogrel or aspirin (ASA) and an oral anticoagulant is recommended with a Class IIaA indication (after an initial phase of triple therapy) for up to 12 months after acute coronary syndrome [7]. (2) Combination therapy is relatively common [8,9,10]; for example, ~40% of participants in the Randomized Evaluation of Long-Term Anticoagulation Therapy (RE-LY) trial of dabigatran versus warfarin used ASA at the start of the study [9]. (3) The prevalence of combination therapy will likely increase; for example, anticoagulation therapy has increased with the introduction of NOACs [11]. In addition, the incidence of AF is projected to increase [12], while risk factors for other cardiovascular diseases that may require antiplatelet therapy are also on the rise [13]. Consequently, combination therapy will be increasingly encountered and required.

Combination therapy raises the prospect of enhanced protection against thromboembolic complications, but with the potential for increased bleeding. Sufficient data are now available to provide estimates of these events. We recently conducted a meta-analysis to assess efficacy and risk in a subgroup of patients with AF on ASA enrolled in randomized clinical trials (RCTs) comparing NOACs with warfarin. Our analysis revealed NOACs plus ASA were more effective in reducing major thromboembolic complications than warfarin plus ASA and were as safe as the latter in terms of major bleeding [14]. We now aim to extend this investigation to evaluate the cost–utility of combination therapy.

Methods

We developed a decision tree for the two patient groups with nonvalvular AF; NOAC + ASA and VKA + ASA. In our model, potential clinical states and outcomes were assigned specific probabilities based on our meta-analysis and other published data. The first branches in the decision tree contained the following clinical states: no change, myocardial infarction (MI), stroke or systemic emboli, major bleeding, and noncardiovascular death (Fig. 1). Potential sequelae were included at the subsequent branch-level for each of these states (Fig. 1). We defined “disabling myocardial infarction” as development of heart failure after MI and “disabling stroke” as a modified Rankin Score of 3–6. After major bleeding, we assumed two potential states: death or good functional status.

Fig. 1
figure 1

Decision tree. ASA aspirin, CV cardiovascular, NOAC non-vitamin K oral anticoagulants, NVAF nonvalvular atrial fibrillation, SE systemic embolism, VKA vitamin K anticoagulants

Full size image

Recently published guidelines recommended that combination therapy should be discontinued in stabilized event-free patients 1 year after stenting [7]. Therefore, we restricted our economic analysis to a 1-year time horizon.

We used the model outputs to calculate the incremental cost-effectiveness ratio (ICER) of NOAC + ASA versus VKA + ASA. As a measure of utility, quality-adjusted life-years (QALYs) were estimated. Costs incorporated in our model reflect the German health-care payers’ perspective.

Strategy for identifying data sources

Data from our prior meta-analysis [14] of subgroups from four RCTs (RE-LY [8], Rivaroxaban Versus Warfarin in Nonvalvular Atrial Fibrillation, ROCKET-AF, [9], Apixaban Versus Warfarin in Nonvalvular Atrial Fibrillation, ARISTOTLE, [15], and Edoxaban Versus Warfarin in Nonvalvular Atrial Fibrillation, ENGAGE, [16]) were used to estimate the probabilities of first branch-level events in the model. The probabilities for subsequent outcomes were extracted from registries and published studies (Table 1). We assumed, if not otherwise indicated, subsequent outcome probabilities for disability, death, and good functional status were independent of treatment group. By contrast, clinical states at the first branch-level were dependent upon treatment.

Table 1 Event probabilities
Full size table

We were unable to find utility values for oral anticoagulation therapy with concomitant antiplatelet use. However, the utility for VKA/NOAC alone was 0.987/0.998 and for ASA alone, 0.998 [17]. Therefore, we assumed the additional use of an antiplatelet would not change these utilities in a material way and we summed the values. Consequently, we extracted values for the final health states (Table 2) from publications describing utility values in patients with AF on NOACs or VKA alone [17,18,19]. VKA-related utilities were derived from patients for whom time trade-off and standard gamble methods were used to estimate quality of life; contributing factors included international normalized ratio (INR) measurements, diet, and lifestyle changes [17]. A utility value of “1” represents full health and a value of “0” denotes death. We assumed patients who made a complete recovery after MI, stroke/systemic embolism, and major bleeding had the same utility as patients with stable health status.

Table 2 Model utilities
Full size table

Costs were expressed in euro (Table 3). Average prices of antiplatelet, NOACs, and VKAs reflect German costs and were derived from the Red List [20]. For VKA treatment, we also included costs associated with required monitoring of the INR. Routine care costs for both NOACs and VKA patients associated with general practitioner visits were included [21]. The one-time costs or outcomes were taken from the Institute for Payment Regulations in German Hospitals (Institut für Entgeldsystem im Krankenhaus, InEK), according to German diagnosis related groups (G-DRGs; [22]). Patients who survived MI/stroke/systemic embolism were assumed to participate in inpatient rehabilitation programs for as long as 4 weeks. These costs were based on expert opinion. After stroke/systemic embolism, or MI with subsequent disability, additional long-term health-maintenance costs were attributed on the basis of educated estimates; we assumed one hospitalization per year together with four visits to outpatient specialists [21, 22].

Table 3 Model costs
Full size table

Sensitivity analysis

Deterministic sensitivity analyses were conducted to evaluate the effects of uncertainty in key input parameters and to test assumptions made in the calculation of the final cost-effectiveness.

We performed a one-way sensitivity analysis on the cost of NOACs. Their relatively high, but likely decreasing, costs would be expected to exert considerable influence. Conversely, the cost of ASA is sufficiently low (~0.03 €/day) to be neglected and thus was omitted from sensitivity analyses.

Additional one-way sensitivity analyses were conducted to examine potential variation in treatment costs for stroke/systemic embolism, MI, and major bleeding. We assumed, in the initial calculation, an average price for acute therapy independent of event severity. However, the DRG system permits differential billing based on severity; therefore, we examined a range of costs.

The type and severity class of AF influence utility values for AF [4, 23,24,25,26]. Therefore, we also ran a two-way sensitivity analysis to explore the effect of such changes; absolute utility values for AF ranged from 0.59 to 0.85, which were corrected for antiplatelet and anticoagulant use. Corresponding differences in values between NOAC and VKA utilities ranged from 0.007 to 0.013 [4, 23, 24] and were also incorporated into the sensitivity analysis.

Results

The expected 1‑year costs were ~50% lower for VKA + ASA versus NOAC + ASA (969.12 € versus 1914.58 €). NOAC + ASA treatment was estimated to provide an incremental 0.0718 QALY increase within our 1‑year time horizon. Hence, the resulting ICER was 13,168.50 € per QALY gained (Table 4).

Table 4 ICER calculation
Full size table

Medication costs exerted a considerable effect on ICER estimation (Table 5). By contrast, the cost of (one-time) acute stroke therapy had a less pronounced effect (Table 6). Our range was from mild cases (cost = 6000 €) to severe cases that required complex intensive care therapy (22,000 €). Similarly, we used a range of costs for MI to reflect event severity; from early discharge after uncomplicated MI (6000 €) to prolonged hospitalization associated with the development of cardiogenic shock (18,000 €). This range also had a limited effect on ICER; ICER ranged between 13,112.79 €/QALY and 13,279.93 €/QALY. Likewise, a fourfold increase in costs for major bleeding (2000–8000 €) produced a limited change in ICER; 11,921.05 €/QALY to 14,415.95 €/QALY.

Table 5 Effect of NOAC cost on ICER
Full size table
Table 6 Effect of one-time stroke therapy on ICER
Full size table

Two-way sensitivity analysis with different utilities for the baseline state in AF patients revealed remarkable differences in the ICER (Table 7). Patients treated with NOACs do not require INR measurements and have fewer dietary restrictions than VKA-treated patients. It therefore seems likely NOAC-associated utility values will always be higher than those associated with VKA treatment.

Table 7 Two-way sensitivity analysis: impact of different utility values for baseline AF state on ICER (expressed in €/QALY)
Full size table

Table 8 shows a two-way sensitivity analysis designed to examine the influence of the probability of stroke incidence on ICER (€/QALY).

Table 8 Two-way sensitivity analysis: impact of different probabilities for stroke/SE on the ICER (expressed in €/QALY)
Full size table

Discussion

We found NOAC + ASA to be a cost-effective alternative to treatment with VKA + ASA. The calculated ICER was 13,168.50 € per gained QALY for a time horizon of 1 year. These benefits coincide with a lower incidence of stroke and systemic embolism and a generally better safety profile [14].

Sensitivity analysis indicated that NOAC costs exerted the greatest effect on ICER, reducing the cost per QALY gained. This is important because NOAC cost might be anticipated to decrease when patent protection expires; dabigatran in 2018, apixaban in 2019, rivaroxaban in 2021, and edoxaban in 2031. However, although it appears intuitively reasonable to assume prices decrease with patent expiration, the reality is more complex. Studies in Canada, the United States, and Europe demonstrated that after patent expiration, prices for original drugs remained constant or even increased [27]. Government policies, the use of other “clinically substitutable” in-patent agents, the influence of prescribers and pharmacists, and even consumer brand loyalty could all play roles in maintaining or increasing prices [28,29,30]—the so-called generics paradox [31]. However, generics, if they become available, are cheaper and thereby reduce the total cost burden from the health-care payer perspective. In Europe, countries where the overall market share for generic medicines is high appear to have larger post-patent price decreases than countries with low generic market share [32]. Consequently, Germany, which has long provided an environment conducive to promotion of generic medicines, may see larger price declines than countries with more recent adoption of such policies [33].

In contrast to NOAC costs, because probabilities for stroke and MI were small, the cost range used in the model for these outcomes had relatively little influence on ICER. However, as anticipated, we found ICER decreased slightly with higher stroke treatment costs. Similarly, for MI, ICER decreased slightly as treatment costs increased. Conversely, for major bleeding, ICER increased with rising treatment costs.

Other model parameters with potential for significant variation are the probabilities used for stroke or systemic emboli and major bleeding. The risk for such events primarily depends, for VKA therapy, on the degree of anticoagulation control and, for NOAC therapy, on medication adherence. Because both could differ from the values used in the model, they merit consideration.

One index used to assess anticoagulation control with VKA therapy is the time in therapeutic range (TTR; i. e., what proportion of time is spent with an INR within the specified target range). Adverse events occur most often when INR is outside this range. If INR is below target, the risk of thrombosis increases, and if above target, the risk of bleeding increases. Consequently, high TTR values are associated with reduced likelihood of stroke and bleeding [34, 35]. Anticoagulation control varies widely and appears to depend, at least in part, upon practice setting. A meta-analysis, conducted on studies from the United States, revealed higher TTR values in anticoagulation clinics (63%) than those achieved by community-based management (51%; [36]). Country-related differences were found in a study that compared INR control in France, Germany, Italy, and the United Kingdom [37]. The proportion of patients with good control, defined as TTR >70%, ranged between 44% in Germany and 65% in the United Kingdom, while mean TTR ranged from 65 to 73% in the same countries. However, these differences were attributed to different approaches to INR monitoring—the use of specialized anticoagulation clinics in the United Kingdom versus general practitioner management in the other countries. The mean or median TTR in all four of the RCTs we used to construct the model was below 70% [16, 38,39,40]; i. e., below what would be considered an indicator of “good control”. It should be emphasized that TTRs ≥70% can be achieved. Thus, there may be circumstances in which our model overestimates the probability of VKA-associated adverse events because INR control in the RCTs was “suboptimal.” Similarly, there may be circumstances in which our model underestimates the probability of NOAC-associated adverse events because of potentially reduced medication adherence in real-world therapy [41,42,43]. In fact, a cost-effectiveness study of NOACs versus warfarin (as sole therapies) for stroke prevention in patients with AF using Slovenian cost data found that if TTR was above 70%, then warfarin was more cost-effective [44]. It remains to be determined whether the same conclusion is reached when combination therapy is used or if the results apply to other countries.

Comparison with other studies

As far as we are aware, this is the first cost–utility study to examine combination therapy and thus direct comparison with other studies cannot be made. Also, because combination therapy has yet to be prospectively examined in patients with AF, there is insufficient information available to derive event estimates beyond the follow-up periods conducted in the RCTs: ~2–3 years. Furthermore, for most cases, antiplatelet therapy will be prescribed for a finite duration [45]. Therefore, we decided to limit our time horizon to 1 year (we calculated events per year averaged over the follow-up) and omit Markov-chain analysis.

There are, however, numerous publications that examined the cost–utility of NOACs versus warfarin without addition of antiplatelet agents. Although comparisons should be made with caution because of different cost models and different methodologies, it is interesting to note that, for example, recent cost–utility evaluations of apixaban versus warfarin reported estimated ICERs/QALY of ~10,500–14,500 € in The Netherlands, France, and Greece [23, 46, 47]. That these ICER values are similar to ours could indicate the net effect of adding ASA therapy has limited effect on cost–utility.

Limitations

Our study examined aggregated estimates obtained from four different trials that compared NOACs with VKA. Therefore, we were unable to compare the different NOACs with each other or with VKA. Second, we only analyzed data from subgroups of the original RCTs; this could have introduced selection bias. Third, we focused only on ASA. Comparative studies comparing combination therapy of NOAC versus warfarin with a single P2Y12 inhibitor and comparable dosages of NOAC as we did in our meta-analysis have yet to be published. Studies such as PIONEER AF-PCI [48] and RE-DUAL PCI [49] could not be included because they enrolled a different patient population. Specifically, they included only patients undergoing percutaneous coronary intervention with the urgent need for a P2Y12 inhibitor, while we focused on ASA therapy. Fourth, country-specific cost estimates were used, which limits generalizability. Moreover, these costs may not reflect real costs because the German DRG-based health-care system allows services to be subsidized if necessary. However, the costs do reflect those paid by the German health insurance system. We did not include indirect costs, such as productivity loss, because our analysis was based on a 65-year-old patient with AF (i. e., retired).

The health-state probabilities used were extracted from the four NOAC RCTs, observational studies, and international registries. We further assumed that all extracted estimates are applicable to Germany. The choice of time horizon can exert considerable influence on results and thereby on the interpretation [50]. Short time horizons can potentially mislead if there are high initial costs for therapies that provide benefit over extended periods. In our analysis, this was not the case. In addition, although temporal changes in benefit-to-risk ratios likely occur (cumulative adverse event risk increases and drug costs recur), the relative change over prolonged periods remains unexamined. Hence, for these and the clinical reasons described earlier, we limited our analysis to a 1-year time horizon.

Implications for practice and conclusion

In conclusion, NOACs used in combination with ASA therapy are cost-effective from a German public health-care insurance perspective with an ICER of 13,168.50 €/QALY. Even though the German health-care system imposes no thresholds on ICER, the ICER we calculated would be considered an acceptable level for those countries that do impose thresholds. The cost–utility analysis allows for a comparison across different health programs and policies because it uses a common unit of measure (cost/QALYs gained).

The cost, morbidity, and mortality of adverse events associated with AF and its treatment mean that selecting effective and safe therapies is of paramount importance. In the RE-LY, ROCKET-AF, ARISTOTLE, and ENGAGE trials, NOACs were shown to be a valuable alternative to VKAs in terms of efficacy and safety. However, “real-life,” long-term benefits of NOACs when used together with antiplatelet agents await confirmation.

Our findings should not be regarded as a universal endorsement of superior cost–utility for combination NOAC + ASA versus VKA + ASA. There may be circumstances when ICER values shift in favor of VKA + ASA; for example, when this therapy’s probability of stroke decreases or as the probability of stroke with NOAC + ASA increases. These two constructs can occur with improved TTR for the former and poor medication adherence for the latter. In addition, there are country-specific considerations that should be taken into account. Instead, our findings should be considered as a starting point for decision-making and calculation.