Full Text Article

Introduction

Atrial fibrillation (AF) affects 350,000 Canadians and its prevalence is rising due to the ageing population [1]. AF is a cause of considerable morbidity and mortality; among its most serious consequences is stroke [2-5]. Strokes caused by AF have poorer prognosis than those that are non-AF related [6-10]. Evidence-based guidelines recommend anticoagulant therapy for the prevention of stroke in AF patients with additional stroke risk factors [11]. Newer oral anticoagulants (NOACs), which have been shown to be as effective and safer than vitamin K antagonists (VKAs), have been approved for use in NVAF patients at moderate to high risk of stroke, as an alternative to VKAs [11].

In recent years, AF has also been linked to cognitive decline and dementia [12]. Several mechanisms have been proposed to explain this relationship including stroke [13-15], chronic cerebral hypoperfusion [14-17], silent cerebral ischemia due to micro-embolization [18,19], as well as underlying cardiovascular risk factors [12,14]. However, new evidence suggests that cognitive decline may be the result of AF independently of stroke and underscores the importance of an appropriate therapeutic management of AF to prevent related cognitive decline and dementia [12,20]. These studies showed treatment with anticoagulants to be effective in reducing the incidence of cerebral ischemic events and new-onset dementia [20,21].

For patients at low risk of stroke, current treatment guidelines [11], recommend NOAC or anti-platelet drugs depending on the presence of additional risk factors and no antithrombotic therapy for those at the lowest risk in this category. However, limited information exists on the benefits of treating this patient’s group with anticoagulants to prevent cognitive decline. The ongoing “Brain-AF” trial, “Blinded Randomized trial of Anticoagulation to prevent Ischemic stroke and Neurocognitive impairment in Atrial Fibrillation” (NCT02387229), assesses the efficacy of rivaroxaban to lower the risk of stroke and cognitive decline in low stroke risk patients. Adding anticoagulation therapy to the treatment regimen of low stroke risk patients may nevertheless impact the healthcare costs. Here, we investigate the cost-effectiveness of rivaroxaban compared to placebo in a hypothetical cohort of patients with NVAF at low stroke risk. The results from the BRAIN-AF trial will inform future economic evaluations with observed event rates in this patient population.

Methods

This cost-utility study was conducted from a single-payer perspective, relating to the Quebec Ministry of Health, where only direct medical costs were considered.

Model Design

A multistate Markov model [22] was used to evaluate two treatment strategies for the prevention of stroke and cognitive decline in NVAF patients at low stroke risk: (1) rivaroxaban 15 mg QD, and (2) placebo. The model (Figure 1) simulated the progression of a hypothetical cohort of patients, aged 30 – 62 years with no other risk factors i.e., a CHADS2 (congestive heart failure, hypertension, age, diabetes mellitus, and stroke/TIA) score of 0, and no contraindications to anti-coagulation, as they moved in 3-month cycles through a series of health states. The health states included: NVAF without complications (healthy), systemic embolism (SE), ischemic stroke (IS) (transient attack [< 24h], minor, major (moderate to severe), fatal), intra-cranial hemorrhage (ICH) (minor, major (moderate to severe), fatal), myocardial infarction (MI), major bleed, clinically relevant non-major bleed, cognitive decline (mild, mod-erate, severe, fatal), and all cause deaths. The patient cohort was assumed to start in the NVAF healthy state and allowed to move exclusively to one of the health states or die during each cycle. The health states were either permanent, indicating that patients remain in them until death, or transient, meaning that patients only spend some time in that health state before returning to the NVAF healthy state. Transition probabilities (risk of experiencing an event), based on the published literature [23-25], were built into the model and applied to the cohort during each cycle to calculate how the patients would be distributed between the health states at the end of the cycle (Table 1). The model assumed full compliance with the medication regimen and no change in the assigned medication for the whole follow-up period. Health benefits (in terms of QALYs) and costs were assigned to each health state over a lifetime (60 years). Future costs and benefits were discounted at a rate of 1.5% annually [26].

All analyses were performed using TreeAge Pro modeling software (TreeAge Pro 2021, R1. TreeAge Software, Williamstown, MA).

Model Inputs

Event Rates

The clinical event rates (Table 1) for patients taking placebo were obtained from a retrospective cohort study [27] assessing the impact of antithrombotic therapy (aspirin, warfarin and no treatment) among 39,400 NVAF patients at low stroke risk. We used the rates per 100 person-years of stroke, SE, IS, bleeding, ICH, and death, from the “no treatment” and “no risk factors” arm (ITT analysis) in our model. The rates of MI and TIA were obtained from the placebo arm in the “Stroke Prevention in Atrial Prevention Study” (SPAF) [23]. This study compared ASA or warfarin to placebo for the prevention of ischemic stroke and systemic embolism in patients with NVAF. Although the population of “SPAF” was different from our hypothetical cohort of patients, in terms of age and comorbidities, no adjustments were made to the clinical event rates. The confidence around the rates was evaluated in sensitivity analyses.

The clinical event rates for patients on rivaroxaban (Table 1) were obtained from XANTUS [28,29], a prospective, international, observational, post-authorization, noninterventional study designed to collect safety and efficacy data on the use of rivaroxaban for stroke prevention in patients with NVAF, in routine clinical practice. The study enrolled 11,121 patients of whom 26% were below the age of 65, 73% received rivaroxaban 20 mg, and 40% had a CHADS₂ score equal 0 - 1. The rates of IS, SE, major bleed and all-cause death were obtained from a subgroup analysis in patients with CHADS₂ score equal 0 [29,30]. Other clinical event rates from XANTUS were applied to our model with no adjustment. The confidence around those rates was evaluated in sensitivity analyses.

The rates of dementia were obtained from Friberg [31], who assessed the incidence of new dementia in patients with AF taking NOACs compared to those not on NOACs. The probabilities of new dementia observed in each group of patients were adjusted using the reported hazard ratio of dementia in the stroke low-risk group (CHADS2-VASC = 0 – 1) [31]. Transition probabilities, based on Spackman, et al. [24], were applied to account for changes in the severity of dementia (Table 1).

An increased risk of recurrence after a first event was applied in the model to the event rate of ischemic stroke (2.20), myocardial infarction (2.04) and bleeding (2.66), based on long-term prognosis and survival data from epidemiological studies [32-34].

Utilities

Utility values were sourced from published studies in similar patients’ population [35,36] and applied whenever a patient experienced an event in a cycle (Table 2). Given that patients had no history of stroke, a starting utility value of 0.81 was used, derived from the utility value for AF [35]. The analysis assumed no difference in utility values between the treatments. The utility values were assumed to apply from the cycle in which the event occurred until the end of the follow-up period or death. Upper and lower utility values (95% CI) (Table 2) were applied in the sensitivity analyses [35-41].

Costs

One-time event and long-term costs of medical care and hospitalization were derived from the literature [42-45] and were based on the Ontario Case Costing Initiative, the Ontario Drug Benefit Formulary (ODBF), and the Canadian Institute for Health Information. The cost of rivaroxaban was obtained from the “Régie de l’Assurance Maladie du Québec” price list [46] with inclusion of an $8.50 dispensing fee and an 8% pharmacist’s markup (Table 3). This cost assumed daily use and no discontinuation. All costs were updated to 2020 Canadian dollars by using the Bank of Canada inflation calculator which is based on the Consumer Price Index (CPI) [47].

Analyses

The relative clinical and economic benefits of rivaroxaban compared to placebo were assessed using the incremental cost-effectiveness ratio (ICER), which was estimated based on the additional costs (Canadian dollars) per additional quality adjusted life-years (QALYs). The ICER was then compared with the commonly accepted Canadian payers’ willingness-to-pay threshold of $50K for each QALY gained.

The robustness of the model’s base case results was assessed in 1-way sensitivity analyses, which consisted of varying each model parameter, using its low and high values, while keeping all others constant. The base case value of clinical events and their costs were varied by ± 25%, the cost of rivaroxaban by ± 15% and utility decrements using either the literature reported 95% CI or a ± 25% variation in the base case value. Also, in these analyses, we used a discount rate of 0% or 3%, a cycle-length of 1 or 6-months and a time horizon of 10, 20 or 40 years.

Additionally, probabilistic sensitivity analyses (PSA), which allow all model parameters to be varied simultaneously, were performed using Monte Carlo simulations [48]. The analysis was run for 10,000 iterations where the value of each model parameter was randomly sampled from a probability distribution uniquely determined for each type of model parameter. The results of the probabilistic analysis were used to generate a scatter-diagram representing the additional gains in QALYs with rivaroxaban compared to placebo (x-axis) against the additional costs of the drug (y-axis). The results of these analyses were also used to derive cost-effectiveness acceptability curves (CEACs) representing the proportion of simulations for which each treatment was the optimal strategy at a given willingness-to-pay threshold.

Results

Base Case Analysis

For a cohort of 1,000 patients followed over their lifetime, treatment with rivaroxaban rather than placebo was predicted to result in fewer cases of ischemic stroke (43 first and 39 recurrent), SE (30), MI (79 first and 7 recurrent), cognitive decline (36) and death (225). However, compared with placebo, rivaroxaban increased the number of ICH (93 first and 25 recurrent) and major bleed (326 first and 614 recurrent). The treatment with rivaroxaban 15 mg led to a net increment in total cost over a lifetime of $128,543 while generating a net increment in total QALY of 11.26. This resulted in an ICER of $11,411 per QALY gained (Table 4). This ratio is below the commonly accepted willingness-to-pay threshold [26] of $50K per QALY gained, indicating that rivaroxaban is cost-effective when compared to placebo in this hypothetical cohort of patients (Figure 2). There was no dominance in the model given that the less costly treatment also had less QALYs.

Deterministic Sensitivity Analyses

Figure 3 presents the results from the one-way sensitivity analyses for the top 20 parameters that had the largest effect on the ICER, in the order of their respective influence.

Regardless of the variation in costs, utility/disutility values or clinical event rates, the ICERs remained below the WTP threshold of $50K (Appendix 1). The lowest ($7,972) and highest ($14,600) ICER values were observed with a ± 25% variation in the base case value of major bleed with rivaroxaban. The next lowest and highest ICER values were obtained with a ± 25% variation in the cost of major bleed ($8,701 - $14,122).

Similarly, varying the discount rate, the cycle length and the time horizon resulted in ICERs below the WTP threshold of $50K. Nevertheless, greater variations were observed with: 1) a 6-month cycle ($22,769) and 2) a time horizon of 10 years. In this latter analysis, the ICER was negative ($-1,730) with absolute dominance of placebo (by rivaroxaban) due to a higher cost and less QALY.

Probabilistic Sensitivity Analysis

The probabilistic sensitivity analyses demonstrated that rivaroxaban was more effective at a small additional cost versus placebo, over a lifetime horizon. The results of the probabilistic analyses are shown in Figure 4 (A) and in Figure 4 (B). The CEACs indicated that rivaroxaban was an optimal treatment choice representing a maximum net benefit over placebo, at a WTP of $50K per QALY gained. The scatter diagram showed ICERs below the WTP threshold of $50K per QALY gained in 100% of the simulations comparing the two agents. The results remained constant regardless of the time horizon.

Discussion

The study assessed the cost-effectiveness of rivaroxaban in the prevention of stroke and cognitive decline in a hypothetical cohort of patients with NVAF at low stroke risk, compared with placebo. Patients on rivaroxaban were predicted to have fewer strokes, systemic embolisms, myocardial infarctions, cognitive impairment and death compared to those on placebo. Nevertheless, hemorrhagic events were predicted to be more likely with rivaroxaban, an expected result with anticoagulant medications, particularly, when compared to placebo or no drug.

In the base-case analysis, over a lifetime horizon (60 years), the net benefits in terms of reduction of clinical events with rivaroxaban yielded an incremental QALY at an incremental cost with a resulting ICER of $11,411, a value below the WTP of $50K, deeming rivaroxaban to be cost-effective compared to placebo. The results were insensitive to the variation in the model’s inputs with ICERs constantly remaining below the WTP of $50K.

Our study is the first to compare the cost-effectiveness of rivaroxaban 15 mg daily versus placebo in preventing both stroke and cognitive decline among NVAF patients at low stroke risk. Existing cost-effectiveness studies of NOACs in NVAF patients at moderate to high stroke risk have consistently found these agents to be cost-effective. The higher up-front cost of NOACs compared to an alternative treatment such as a vitamin K antagonist (VKA) was offset by a reduction in clinical events and an increase in patients’ quality of life [38,42,49-55]. In studies [53-55] assessing solely the cost-effectiveness of rivaroxaban in comparison to warfarin, the former was shown to be cost-effective with ICERs (US$27,498, €15,207, €8,809) below the country respective WTP threshold. All three studies used clinical event data observed in the pivotal randomized controlled trial “ROCKET-AF” (Rivaroxaban Once Daily Oral Direct Factor Xa Inhibition Compared with Vitamin K Antagonism for Prevention of Stroke and Embolism Trial in Atrial Fibrillation) which assessed the efficacy and tolerability of rivaroxaban in moderate to high stroke risk patients. A recent real-world cost-effectiveness study of rivaroxaban and apixaban each in comparison to VKA [56], showed both agents to be cost-effective in moderate to high stroke risk patients, based on real-world evidence (RWE). The incremental cost-effectiveness ratio was £14,437 for rivaroxaban compared with VKA [56]. Comparisons between our results and those of previous rivaroxaban economic evaluation studies were not feasible given important differences in the model design, drug dosage and patient population characteristics.

Our analysis has some limitations. Given the lack of data in low stroke risk NVAF patients, certain clinical event rates as well as the probabilities of transitioning between events in our model were obtained from studies with moderate to high stroke risk patients. The SPAF (Stroke Prevention in Atrial Prevention Study) study [23], from which we obtained the MI and TIA rates for the placebo arm, included patients with co-morbidities such as hypertension (52%), diabetes (19%) and congestive heart failure (20%). Similarly, the XANTUS [28,29] main study, from which we obtained the ICH, MI and CRNMB rates for the rivaroxaban arm, included a majority of moderate to high stroke risk patients treated with rivaroxaban 20mg. Those rates are higher than what could have been expected in patients at low stroke risk treated with rivaroxaban 15 mg and likely could have impacted the ICER estimation. Nevertheless, the sensitivity analyses showed the ICER values to be insensitive to a variation of ± 25% in those rates and rivaroxaban to remain cost-effective in 100% of the simulations. Also, our model assumed patients’ full persistence/compliance with the treatments which may have led to an over or under estimation of the ICER. Additionally, the patients’ age, dosage regimen, risk score, and costs used in the model as well as the Canadian public health care perspective make the results not necessarily generalizable to other settings. Finally, given the lack of data, the model allowed patients to have one event per cycle and one type of event over lifetime. Data on the incidence of different combination of clinical events in patients with NVAF at low risk of stroke would have certainly contributed to more precise calculations of the overall cost and benefits of rivaroxaban treatment. The results from the BRAIN-AF trial, once available, will inform the analysis with observed event rates in this patient population.

Conclusion

This economic evaluation predicted that rivaroxaban would be a cost-effective alternative to placebo for the prevention of stroke and cognitive decline in NVAF patients at low stroke risk. The “no-treatment” therapeutic recommendation for this patients’ group is based on the low stroke risk but does not take into consideration another potential risk as debilitating and life-threatening as cognitive impairment and dementia. Future economic evaluations of NOACs for the treatment of NVAF that account for the impact of the disease on patients’ mental health would better estimate the patients’ overall health benefits and the costs of these drugs. Our analysis of a hypothetical cohort of NVAF patients at low stroke risk has shown that rivaroxaban could offer health benefits for a marginal increase in healthcare costs.

Appendix 1