Cost-Effectiveness of Percutaneous Closure of the Left Atrial Appendage in Atrial Fibrillation Based on Results from PROTECT AF vs. PREVAIL

Abstract

Background

Randomized trials of left atrial appendage (LAA) closure with the Watchman device have shown varying results, and its cost-effectiveness compared to anticoagulation has not been evaluated using all available contemporary trial data.

Methods and Results

We used a Markov decision model to estimate lifetime quality-adjusted survival, costs, and cost-effectiveness of LAA closure with Watchman, compared directly with warfarin and indirectly with dabigatran, using data from the long-term (mean 3.8 year) follow-up of PROTECT AF and PREVAIL randomized trials. Using data from PROTECT AF, the incremental cost-effectiveness ratios (ICER) compared to warfarin and dabigatran were $20,486 and $23,422 per quality adjusted life year (QALY), respectively. Using data from PREVAIL, LAA closure was dominated by warfarin and dabigatran, meaning that it was less effective (8.44, 8.54, and 8.59 QALYs, respectively) and more costly. At a willingness-to-pay-threshold of $50,000 per QALY, LAA closure was cost-effective 90% and 9% of the time under PROTECT AF and PREVAIL assumptions, respectively. These results were sensitive to the rates of ischemic stroke and intracranial hemorrhage for LAA closure and medical anticoagulation.

Conclusions

Using data from the PROTECT AF trial, LAA closure with the Watchman device was cost-effective; using PREVAIL trial data, Watchman was more costly and less effective than warfarin and dabigatran. PROTECT AF enrolled more patients and has substantially longer follow-up time, allowing greater statistical certainty with the cost-effectiveness results. However, longer term trial results and post-marketing surveillance of major adverse events will be vital to determining the value of the Watchman in clinical practice.

Introduction

Non-valvular atrial fibrillation (AF) currently affects between 2 and 4 million people in the United States and the prevalence is increasing.^{1, 2} AF causes ischemic stroke, and long-term anticoagulation with warfarin or the target-specific oral anticoagulants can safely reduce the risk of ischemic stroke.^3-8 Device-based alternatives to anticoagulation that attempt to exclude the left atrial appendage (LAA) from the systemic circulation have been developed.^9-15 The Watchman device (Boston Scientific, Plymouth, MN), a catheter-based, self-expanding LAA occluding device, has been studied in two relatively small randomized trials against warfarin.^16-18 The results of these trials were disparate, resulting in significant uncertainty in the clinical community about the value of this new treatment paradigm. This uncertainty was reflected in a protracted US Food and Drug Administration (FDA) evaluation of the Watchman device that ultimately resulted in approval in March 2015.

The PROTECT AF (Percutaneous Closure of the Left Atrial Appendage Versus Warfarin Therapy for Prevention of Stroke in Patients With Atrial Fibrillation) study was an international trial of 707 subjects with AF and a CHADS₂ score ≥ 1 who were randomly assigned to LAA occlusion with the Watchman device and planned warfarin discontinuation or warfarin therapy.¹⁶ At a mean follow-up of 18 months LAA occlusion was non-inferior to warfarin for the composite endpoint of stroke, systemic embolism, or cardiovascular death.¹⁶ At a mean follow-up of 3.8 years, patients in the device group had a significantly lower risk of the same composite endpoint and the independent endpoints of stroke or systemic embolism, cardiovascular mortality, and all-cause mortality.¹⁸ However, there was a substantial rate of periprocedural complications, including stroke and pericardial effusion, and implant failure.¹⁶

PREVAIL (Prospective Randomized Evaluation of the Watchman LAA Closure Device In Patients with Atrial Fibrillation) was a confirmatory, U.S.-based trial that randomly assigned 407 patients to LAA occlusion with Watchman or warfarin.¹⁷ The patients in this study were somewhat higher risk for stroke than with PROTECT AF with a higher mean age (74 vs 72 years in the device group), higher CHADS2 score (2.6 vs 2.2), and higher rate of previous stroke or transient ischemic attack (27.5% vs 17.7%). At 18 months, the rates of the composite endpoint of stroke, systemic embolism, and cardiovascular mortality were similar (0.064 in the device group versus 0.063 in the control group), but the device did not achieve the prespecified criteria for noninferiority due to large 95% confidence intervals. Periprocedural safety events with LAA closure were substantially lower in PREVAIL than in PROTECT AF. However, the rates of intracranial hemorrhage and major extracranial hemorrhage were substantially higher in the device arm of the study.

The primary cost-effectiveness study on LAA closure to date showed that the technology may be cost-effective but used only the 18-month PROTECT AF data for model assumptions and a Canadian health care system payer perspective.¹⁹ A follow up analysis incorporating longer-term PROTECT AF data similarly demonstrated that LAA closure may be cost-effective relative to medical anticoagulation, but this study excluded the conflicting data from PREVAIL.²⁰ We evaluated the quality-adjusted survival, costs, and cost-effectiveness of the Watchman device for LAA closure compared with warfarin using the most contemporary clinical trial data and a US payer perspective. Given the divergent results of PROTECT AF and PREVAIL with respect to safety and efficacy, and the marked difference in follow-up (3.8 year average follow-up vs 18 months), we did not perform meta-analysis or pooling of underlying data inputs, but rather presented our cost-effectiveness estimates for the two trials separately. In addition, given that the health care landscape now includes the alternative to treat AF patients with the target-specific oral anticoagulants, we compared LAA closure and warfarin to dabigatran as a broadly representative agent in this new class of drugs.

Methods

Decision Model

We used a Markov model ²¹ to perform a decision analysis comparing three treatment strategies for the prevention of thromboembolism in patients with AF: adjusted-dose warfarin with a target international normalized ratio (INR) of 2.0-3.0, dabigatran 150 mg twice daily, and LAA closure with planned discontinuation of anticoagulation. Our base case consisted of a hypothetical cohort of patients aged 70 years with AF at increased risk for stroke (i.e., CHADS₂ score ≥1) and no contraindications to anticoagulation.

Most health states in the model have been previously described²² and included healthy with AF, reversible ischemic neurologic deficit (RIND), ischemic stroke (mild or moderate to severe), intracranial hemorrhage (ICH) (mild or moderate to severe), combined stroke and ICH, myocardial infarction (MI), systemic embolism, and death (Figure 1).⁴ We applied utilities and costs to each outcome in 2-week increments over its expected duration, and discounted costs and benefits at 3% per year.²³

Figure 1.

Decision model. The schematic at the top represents the treatment arms: warfarin, dabigatran, and percutaneous closure of the left atrial appendage. M represents a Markov process with 10 health states for each of the treatment options. These potential health states (and transitions) are identical for each of the treatment options. All patients remain in the “Well” state until one of the five events occur: transient ischemic attack (TIA), stroke, intracranial hemorrhage (ICH),systemic embolism, myocardial infarction, extracranial hemorrhage, or death. The probabilities of these events depend upon the prescribed therapy. The triangles on the right indicate which health state the patient enters after an event. Reversible ischemic neurologic event (RIND), is the health state that patients enter after a TIA or a stroke without residual deficit. “Mild” represents a neurologic event that results in a neurologic deficit but no limitation performing activities of daily living, while “moderate to severe” represents a neurologic event that results in loss of independence for at least one activity of daily living. This figure does not show the perioperative events associated with the LAA closure, but these events occur prior to the Markov nodes and are enumerated in Table 1 under the section, “Periprocedural adverse events, LAA closure”.

For LAA occlusion, we created two sets of model inputs, based on PROTECT AF and PREVAIL, and compared both sets of model results to those of warfarin and dabigatran treatment. Risks for adverse events included in our model were generally derived from the event rates published in the PROTECT AF, PREVAIL and RE-LY (Randomized Evaluation of Long-Term Anticoagulation Therapy; trial of dabigatran vs. warfarin) trials unless stated otherwise.^{4, 5, 17, 18} Because PROTECT AF data has been published in several iterations, we used the periprocedural complication rates from the index publication¹⁶ and long-term event dates from the most recent publication based on mean 3.8 year follow-up data.¹⁸ For values not available in the published literature, we identified adverse event rates from the FDA filing.²⁴ For conditions not included in our model, such as cancer, we assumed that event rates were similar across all treatments. For each treatment, we quantified quality-adjusted survival, adverse event risk, and net costs to age 100 years or until death (whichever occurred earlier). Model building and analyses were performed using TreeAge Pro Suite (TreeAge Software, Williamstown, Massachusetts) and Microsoft Excel (Microsoft, Redmond, Washington).

Probability of Adverse Outcomes in the Decision Model

Treatment assumptions

In PROTECT AF, the occlusion device was implanted in 408 of 449 patients (88%) in whom it was attempted. Patients for whom device implantation was unsuccessful were treated with warfarin. For patients assigned to the device arm of PROTECT AF, we assumed that clinical adverse events could occur in the periprocedural period or during the long-term period thereafter. In PROTECT AF, for the first 45 days after device placement, patients were maintained on warfarin anticoagulation and aspirin 81-325 mg daily. Eighty-six percent of patients were then transitioned to clopidogrel 75 mg daily and aspirin for day 45 until 6 months, but 14% were maintained on warfarin and aspirin until 6 months due to incomplete appendage closure, based on blood flow around the device demonstrated on transesophageal echocardiography. After 6 months, 92% patients who had successful LAA closure were transitioned to aspirin therapy only, and the remaining 8% were maintained on warfarin indefinitely due to incomplete appendage closure.

In PREVAIL, the LAA closure device was implanted in 252 of 265 patients (95%) in whom it was attempted. The same medication treatment paradigms were used in PREVAIL and PROTECT AF as specified by the trial protocols. Ninety-two percent of patients were then transitioned to clopidogrel 75 mg daily and aspirin 81-325 mg daily for day 45 until 6 months, but 8% were maintained on warfarin until 6 months due to incomplete appendage closure, based on blood flow around the device demonstrated on transesophageal echocardiography. After 6 months, 99% of patients who had successful LAA closure were transitioned to aspirin therapy only, and the remaining 1% were maintained on warfarin indefinitely due to incomplete appendage closure.

Mortality rates were adjusted for age (beginning at age 70 years). The presence of AF and antithrombotic therapy were accounted for in our model by the clinical event rates of the intention-to-treat analyses of PROTECT AF, PREVAIL and RE-LY.^{3, 4, 18, 25-29}

Periprocedural Adverse Events

The PROTECT AF model assumed that in the periprocedural period after LAA closure, 1.5% of patients had a serious pericardial effusion, 1.1% of patients had an ischemic stroke, 0.6% of patients had a device embolization, and 0.2% had an esophageal tear.¹⁶ The PREVAIL model assumed that in the periprocedural period after LAA closure, 0.4% of patients had a serious pericardial effusion, 0.7% of patients had an ischemic stroke, 0.7% of patients had a device embolization, and 0.2% had an esophageal tear.¹⁷

Ischemic Stroke

In the PROTECT AF base case, the annual rate of ischemic stroke was 1.1%/year for warfarin and the relative risk (RR) for ischemic stroke with dabigatran was 0.76 compared with warfarin.^{4, 5, 16, 18, 30} In the PREVAIL base case, the annual rate of ischemic stroke was 0.7%/year for warfarin.¹⁷ For LAA closure, the long-term annual rate of ischemic stroke (excluding periprocedural events) was 1.4% per year in PROTECT AF and 1.9% per year in PREVAIL (Table 1).^{17, 18, 30} In the base case, we assumed that ischemic strokes after LAA closure resulted in discontinuation of aspirin therapy and replacement with warfarin.

Table 1. Base-Case Values and Ranges Used in Sensitivity Analyses^*.

Variables	PROTECT AF Base-Case Value (Range)	PREVAIL Base-Case Value (Range)	Reference
Stroke
Annual rate of ischemic stroke,%
Warfarin	1.1 (0.5-1.7)	0.7 (0-3.9)	16-18, 30
Dabigatran, relative risk vs warfarin	0.76 (0.6-0.98)	0.76 (0.6-0.98)	4, 5
Aspirin	3.2 (2.0-5.0)	3.2 (2.0-5.0)	25, 31-33
LAA closure	1.1(0.7-1.7)	1.1 (0.2-3.2)	16-18, 30
Ischemic strokes with all therapies, %
Fatal (within 30 days)	8.2 (8.2-10.1)	8.2 (8.2-10.1)	3, 33-41
Major (nonfatal)	40.2 (40.2-41.7)	40.2 (40.2-41.7)	3, 33-41
Minor	42.5 (34.8-42.5)	42.5 (34.8-42.5)	3, 33-41
No residual	9.1 (9.1-13.3)	9.1 (9.1-13.3)	3, 33-41
Hemorrhage
Annual rate of ICH, %
Warfarin	1.1 (0.5-1.8)	0 (0-3.7)	16-18, 30, 42-44
Dabigatran, relative risk vs warfarin	0.41 (0.28-0.60)	0.41 (0.28-0.60)	4, 5
LAA closure	0.2 (0-0.4)	0.4 (0-2.2)	16-18, 30
Annual rate of major hemorrhage excluding ICH
Warfarin	2.0 (1.2-3.2)	2.1 (0.4-5.4)	16, 18, 24, 37, 44-47
Dabigatran, relative risk vs warfarin	1.07 (0.94-1.21)	1.07 (0.94-1.21)	4, 5
LAA closure	1.3 (0.8-2.0)	4.2 (2.1-7.5)	16, 18, 24
Myocardial infarction
Annual rate of MI, %
Warfarin	0.64 (0.4-0.9)	0.64 (0.4-0.9)	4, 5
Dabigatran, relative risk vs warfarin	1.27 (0.94-1.71)	1.27 (0.94-1.71)	4, 5
LAA closure	0.9 (0.5-1.4)	0.9 (0.5-1.4)	4
Systemic embolism
Annual rate of systemic embolism, %
Warfarin	0 (0-3.7)	0 (0-3.7)	4, 5, 17, 18, 30
Dabigatran, relative risk vs warfarin	0.75 (0.60-0.90)	0.75 (0.60-0.90)	4, 5
LAA closure	0.2 (0-0.4)	0.4 (0-2.2)	17, 18, 30
Mortality
Relative risk reduction	0.66 (0.45-0.98)	1.0	17, 18
Periprocedural adverse events, LAA closure
Rate of adverse event, %/procedure
Ischemic stroke	1.1 (0.4-2.6)	0.7 (0-3.9)	16, 17
Serious pericardial effusion	1.5 (0.6-3.1)	0.4 (0-3.6)	16, 17
Device embolization	0.6 (0.1-1.2)	0.7 (0.1-2.5)	16, 17
Esophageal tear	0.2 (0-1.1)	0.2 (0-1.1)	16
Age at start of 30 year interval, years	70 (65-100)	70 (65-100)	Assumption
Quality-of-life estimates (utility)
Healthy
Warfarin	0.987 (0.953-1.0)		48
Dabigatran	0.994 (0.975-1.0)		29
Aspirin	0.998 (0.994-1.0)		48
LAA closure	0.998 (0.994-0.996)		48
Neurologic event with residua
Mild	0.75 (0-1.0)		48
Moderate to severe	0.39 (0-1.0)		48
Recurrent	0.12 (0-1.0)		48
MI	0.84 (0-1.0)		49
Systemic embolism	0.73 (0-1.0)		49
Temporary states, days of perfect health lost
Major hemorrhage other than ICH	3 (0-5)		50, 51
Serious pericardial effusion	5 (0-10))		52
Device embolization	15 (0-30)		52
Esophageal tear	5 (0-10)		52
Costs
Daily cost of medication, $
Warfarin (not including INR monitoring)	0.16 (0.10-1.5)		53
Dabigatran	4.86 (3-10)		5, 53
Aspirin	.02 (0.005-0.20)		53
Plavix	0.35 (0.25-1.5)		53
Cost of INR laboratory, $	6.15 (4-10)		54
Cost of LAA closure with imaging, $	24,011.22 (10,000-30,000)		12, 55, Estimate
One-time cost of ischemic neurologic event
Moderate to severe	20,719.26 (12,000-27,000)		29, 55-58
Mild	14,857.85 (7000-20000)		29, 55-58
Transient ischemic attack or RIND	6453.44 (3000-12000)		29, 55-58
One-time cost of ICH	16,427.67 (10,000-65,000)		29, 55-58
Monthly cost of ischemic neurologic event
Moderate to severe	5716 (3000-9000)		29, 55-60
Mild	2400 (1000-4000)		29, 55-60
Monthly cost of ICH
Moderate to severe	5716 (3000-10000)		29, 55-59, 61
Mild	2400 (1000-4000)		29, 55-60
Monthly cost of ischemic neurologic event and ICH	6800 (3000-13,000)		29, 55-60
One-time cost of MI	20,016.33 (15,000-25,000)		55
Monthly cost of MI	325 (125-580)		62
One-time cost of systemic embolism	27,558.82 (15,000-35,000)		55
Monthly cost of systemic embolism	2389 (1000-5000)		63, 64
Other one-time costs
Major hemorrhage	9164.43 (5000-15,000)		55
Serious pericardial effusion	22,092.86 (10,000-30,000)		55
Device embolization	57,006.20 (35,000-70,000)		55
Esophageal tear	11,095.22 (5000-35,000)		55
Cost discounting rate, %	3 (0-5)		21

LAA=left atrial appendage, ICH=intracranial hemorrhage, MI=myocardial infarction, NVAF=nonvalvular atrial fibrillation, INR=international normalized ratio

^*Costs are in 2014 US Dollars.

Intracranial Hemorrhage

For the base case, the annual rate of ICH was 1.1%/year in PROTECT AF and 0% in PREVAIL on warfarin.^{17, 18, 30} The relative risk for ICH on dabigatran was 0.26 compared with warfarin (0.4%/year), based on RE-LY.^{4, 5, 16} The rate of intracranial hemorrhage after LAA closure was 0.2%/year in PROTECT AF and 0.4%/year in PREVAIL.^{17, 18, 30}

Stroke and Hemorrhage Severity

We classified initial ischemic stroke into four categories: fatal, moderate to severe neurologic sequelae, mild neurologic sequelae, and no residual neurologic deficit.^{3, 29, 33-41} We assumed that a second mild ischemic stroke resulted in a moderate to severe ischemic stroke. We classified hemorrhage into 4 categories: fatal, ICH with moderate to severe neurologic sequelae, ICH with mild neurologic sequelae, and nonfatal extracerebral major hemorrhage.^{4, 31, 37, 42-47}

Quality-of-Life Estimates

We calculated quality-adjusted survival by multiplying the probabilities of adverse events by quality-of-life estimates (utilities).⁴⁸ We adjusted baseline quality of life for age.⁴⁹ We obtained the utility for warfarin without complications from published data on patients with AF that were based on patient ratings of their quality of life while receiving warfarin, including prothrombin time monitoring and changes in diet or lifestyle. The mean utility was 0.987 for warfarin,^{29, 48} and 0.998 for aspirin.^{29, 48}

We estimated the utility for dabigatran as 0.994 as in previous studies,²² which was derived from a survey of anticoagulation physicians for an older direct thrombin inhibitor, ximelagatran.^{4, 5, 29}

We accounted for the temporary decrease in quality of life associated with the LAA closure procedure by reducing quality of life by 30% for 10 days. After 10 days, patients were assigned the utility for warfarin while on that medication and the utility for aspirin thereafter.⁴⁸

Any ischemic or hemorrhagic neurologic event with residual deficit, MI, and systemic embolism were associated with permanent decreases in quality of life. We assigned temporary decrements in quality of life (utility) for ischemic stroke with no residual neurologic deficit, TIA, nonfatal extracerebral major hemorrhage, pericardial effusion, esophageal tear, and device embolization.

Costs

Costs, expressed in 2014 US Dollars, reflected the perspective of an ideal insurer that covered inpatient and outpatient medical care and prescription costs. For each treatment, we projected costs over 35 years; future costs and life-years were discounted at 3% per year. We included age-adjusted average health care expenditures for each patient and then added the costs associated with the three treatment strategies.⁶⁵ This analysis excluded indirect costs.

Drug Treatment Costs

We estimated medication costs for warfarin, dabigatran, and aspirin based on the median consumer prices in the US.^{53, 66} For warfarin anticoagulation, we combined the annual medication cost with the cost for 14 INR tests and the Center for Medicare and Medicaid Services (CMS) reimbursement for a period of 90 days of anticoagulation management (Current Procedural Terminology [CPT] code 99364). In sensitivity analyses, we allowed for patients initiating warfarin to have up to 8 additional INR tests and for CMS reimbursement to be at the higher rate allowed for anticoagulation initiation for a 90 day period (CPT 99363).^{12, 67}

LAA Closure Costs

The Watchman device was recently approved by the Food and Drug Administration, and the Center for Medicare and Medicaid Services made a final determination to cover LAA closure on February 8, 2016 under DRG 273 (percutaneous intracardiac procedure with major complication or comorbidity) at a rate of $20,961 and DRG 274 (percutaneous intracardiac procedure without major complication or comorbidity) at a rate of $14,288. Because the device is only indicated in patients with AF and a CHADS2 score ≥ 2 or a CHA2DS2-VASc score ≥ 3, we estimated that the great majority of procedures will be reimbursed at the higher rate. We also included the cost of one transesophageal echocardiogram at the time of the procedure and physician fee reimbursement, for a total cost of $24,010.^{55, 68}

Ancillary costs of imaging and clinical follow-up were modeled according to the PROTECT AF trial protocol, including transesophageal echocardiography at 45 days, 6 months and 12 months after implantation and 4 established care outpatient physician visits.¹⁶

Adverse Event Costs

We estimated the one-time costs of ischemic stroke, transient ischemic attack (TIA), ICH, and MI based on the costs of a hospitalization for the DRG or the International Classification of Diseases, 9^th Revision (ICD-9) code published by AHRQ HCUPnet.⁵⁵ We estimated monthly costs of care for each of these complications based on previously published cost estimates using CMS reimbursement for the DRG, adjusted to 2014 US dollars using the GDP deflator.^{29, 56-58, 62, 69-71} We estimated the one-time cost of a major extracranial hemorrhage based on the hospital cost for the ICD-9 codes associated with gastrointestinal hemorrhage (530.21, 578.1, 578.9 and 772.4) published by AHRQ HCUPnet. We estimated the acute care costs of systemic embolization based on costs for the ICD-9 code 444.09 (Other arterial embolism and thrombosis of abdominal aorta). We estimated monthly costs of care for patients after systemic embolization based on previously published cost estimates using CMS reimbursement for the DRG, adjusted to 2014 US Dollars using the GDP deflator.^{28, 54-60} We estimated the one-time costs of a serious pericardial effusion based on the hospital costs for the ICD-9 codes 423.0 and 423.3 published by AHRQ HCUPnet. We estimated the one-time costs of an esophageal tear using the hospital cost for DRG 368 (major esophageal disorder) published by AHRQ HCUPnet.

Sensitivity analyses

We performed one-way sensitivity analyses of the variables in the decision model over their plausible ranges (Table 1). Ranges for clinical events were derived from 95% confidence intervals for event rates from the PROTECT AF, PREVAIL and RE-LY trials as well as other published literature.^{4, 16, 17, 24, 30} When 95% confidence intervals were not published, a Poisson distribution was assumed to calculate the confidence interval. Medication costs for aspirin, warfarin, and dabigatran included the range of discount and retail costs.⁵³

We conducted a sensitivity analysis in which we varied the baseline risk of ischemic stroke or ICH for all three treatment strategies by the same ratio to simulate the cost-effectiveness ratios for patients at varying risk for these adverse events (e.g., low to high CHADS₂).

Because procedural success and safety can be considerably worse with inexperienced operators outside of clinical trials, we also conducted a sensitivity analysis in which we decreased the rate of successful device implantation and increased the rates of each of the periprocedural adverse events from those reported in PROTECT AF and assessed how this impacted the cost-effectiveness of the device.

Probabilistic sensitivity analysis

Probabilistic sensitivity analysis takes into account all sources of uncertainty in the data inputs and generates a best estimate of which therapy is preferred at different cost-effectiveness thresholds. We performed second-order Monte Carlo simulations^{22, 72}, randomly sampling (with replacement) a distribution of all variables 10,000 times and then simulating outcomes. For event rates, we generally used a normal distribution, except for the mutually exclusive sub-categorization of stroke, for which we used a Dirichlet distribution. We used a beta distribution for utilities and a gamma distribution for costs.

Results

Base Case Analyses

Under base case conditions with mean 3.8 year follow-up data from PROTECT AF, the quality-adjusted life expectancy was 7.96 quality adjusted life years (QALYs) with warfarin, 8.28 QALYs with dabigatran, and 9.94 QALYs with LAA closure (Table 2, Supplemental Figure 1). Total costs were $92,190 for warfarin, $94,072 for dabigatran, and $132,844 for LAA closure. Comparing LAA closure with warfarin the incremental cost-effectiveness ratio (ICER) was $20,486 per QALY and comparing LAA closure with dabigatran the ICER was $23,422 per QALY. In a base case cohort of 10,000 patients followed over their lifetime, LAA closure was associated with an ischemic stroke rate lower than warfarin and similar to that of dabigatran and it was associated with a markedly lower rate of ICH compared with both anticoagulants (Table 3). LAA closure and dabigatran were also associated with higher rates of MI than warfarin.

Table 2. Projected Costs and QALYs for Patients with Nonvalvular Atrial Fibrillation for PROTECT AF and PREVAIL^*.

Annual Stroke and ICH Rate With Warfarin, %	Therapy	Cost, $	QALYs	Marginal Cost per QALY for LAA closure vs. dabigatran, $	Marginal Cost per QALY for LAA closure vs. warfarin, $
Base Case Trial
PROTECT AF	Warfarin	92,190	7.96
	Dabigatran	94,072	8.28
	LAA closure	132,844	9.94	23,422	20,486
PREVAIL	Warfarin	73,077	8.54
	Dabigatran	83,746	8.59
	LAA closure	120,977	8.44	Dominated	Dominated

QALY= quality-adjusted life-year; ICH = intracranial hemorrhage; LAA= left atrial appendage.

^*Costs are in 2014 U.S. dollars. QALYs in the table are rounded values, but marginal costs per QALY were calculated using non-rounded data.

Table 3.

Major adverse events for a hypothetical cohort of 10,000 patients with AF treated with each therapy and followed over their lifetime starting at age 70 years

Trial Data Source	Therapy	Ischemic Stroke	Intracranial Hemorrhage	Myocardial Infarction
PROTECT AF	Warfarin	1,786	1,234	762
	Dabigatran	1,557	592	949
	LAA Closure	1,703	292	1,311
PREVAIL	Warfarin	1,332	105	799
	Dabigatran	1,187	104	977
	LAA Closure	1,362	472	1,062

AF= atrial fibrillation, LAA= left atrial appendage

Using data from PREVAIL for the base case data inputs, the quality-adjusted life expectancy was 8.54 QALYs with warfarin, 8.59 QALYs with dabigatran, and 8.44 QALYs with LAA closure (Table 2, Supplemental Figure 2). Total costs were $73,077 for warfarin, $83,746 for dabigatran, and $120,977 for LAA closure. LAA closure was dominated by both warfarin and dabigatran, meaning that it was both less effective (lower QALYs) and more costly than both medical alternatives. In a base case cohort of 10,000 patients followed over their lifetime, LAA closure was associated with an ischemic stroke rate similar to warfarin and higher than dabigatran and a markedly higher rate of ICH (Table 3).

Sensitivity Analyses

One-way sensitivity analyses for the comparison of LAA closure to warfarin showed that the cost-effectiveness of LAA closure was most sensitive to changes in the rates of stroke and intracranial hemorrhage (Supplemental Figure 3 and 4). The model was also sensitive to the cost of the Watchman, the utility with warfarin, the cost of stroke, the rate of systemic embolism, the cost of intracranial hemorrhage, and the rate of MI, and the utility associated with major adverse events (stroke, ICH, and MI). Using the PROTECT AF data inputs, the cost-effectiveness estimates were only moderately sensitive to these parameters, and the ICER for LAA closure compared with both medical therapies remained below $35,000 per QALY for all values of each variable tested. Using PREVAIL data inputs, the cost-effectiveness estimates were insensitive to these parameters, and the ICER for LAA closure compared with both medical therapies remained above $100,000 per QALY for all value ranges of each variable tested.

Ischemic Stroke

The cost-effectiveness of LAA closure could be sensitive to changes in ischemic stroke rates for patients with LAA closure. Using data from PROTECT AF, one-way sensitivity analysis on the rate of stroke after LAA closure over the plausible range from the trial showed that the incremental cost-effectiveness ratio of LAA closure remained below $50,000 per QALY in comparison to both warfarin and dabigatran (Figure 2). Using data from PREVAIL, if the long-term rate of stroke with LAA closure dropped below 0.6 strokes per 100 person-years, LAA closure would no longer be dominated, but the incremental cost-effectiveness ratio of LAA closure would remain above $100,000 per QALY even if the long-term stroke rate dropped as low as 0.2 per 100 person-years (Figure 3).

Figure 2.

One-way sensitivity analysis showing the effect of varying the long-term ischemic stroke rate of percutaneous closure of the LAA on the incremental cost-effectiveness of LAA closure compared with warfarin and dabigatran in PROTECT AF. The incremental cost-effectiveness ratio of LAA closure remained below $50,000 per QALY in comparison to both warfarin and dabigatran across the range of stroke rates. The vertical dotted line shows the base case.

Figure 3.

One-way sensitivity analysis showing the effect of varying the long-term ischemic stroke rate of percutaneous closure of the LAA on the incremental cost-effectiveness of LAA closure compared with warfarin and dabigatran in PREVAIL. If the long-term rate of stroke with LAA closure dropped below 0.6 strokes per 100 person-years, LAA closure would no longer be dominated, but the incremental cost-effectiveness ratio of LAA closure would remain above $100,000 per QALY across the range of stroke rates. The vertical dotted line shows the base case.

Intracranial Hemorrhage

The cost-effectiveness of LAA closure could be sensitive to changes in ICH rates for patients with LAA closure (Supplemental Figure 5 and 6). Using data from PROTECT AF, one-way sensitivity analysis on the rate of ICH after LAA closure over the plausible range from the trial showed that the incremental cost-effectiveness ratio of LAA closure remained below $30,000 per QALY in comparison to both warfarin and dabigatran. Using data from PREVAIL, if the rate of ICH with LAA closure dropped below 0.1% ICHs per 100 person-years, LAA closure would no longer be dominated but the incremental cost-effectiveness ratio would remain above $500,000 per QALY for both warfarin and dabigatran.

Periprocedural Adverse Events and Less Experienced Operator Sensitivity Analysis

One-way sensitivity analyses showed that the model was insensitive (incremental cost effectiveness varied by <$1000 per QALY) to each of the periprocedural adverse events, including pericardial effusion, ischemic stroke, device embolization and esophageal tear.

To test the possible effect of less experienced operators performing LAA closure, we decreased the rate of successful device implantation to 80% and increased each of the periprocedural adverse events rates to the highest value in their plausible ranges. For this scenario, using data from PROTECT AF the ICER for LAA closure was $29,090 per QALY compared with warfarin and $35,360 per QALY compared with dabigatran. For this scenario, using data from PREVAIL LAA closure remained dominated compared to warfarin and dabigatran, meaning that it was both less effective (lower QALYs) and more costly than both medical alternatives.

Probabilistic Sensitivity Analysis

In the Monte Carlo simulation varying all data inputs simultaneously based on data from PROTECT AF, warfarin was most likely to be cost-effective at a willingness-to-pay threshold <$6,000 per QALY, dabigatran was most likely to be cost-effective between $6,000 and $21,000 per QALY and LAA closure was most likely to be cost-effective above that. LAA closure was 89% and 98% likely to be cost-effective at willingness-to-pay thresholds of $50,000 and $100,000, respectively (Figure 4). Using data from PREVAIL, warfarin was most likely to be cost-effective at all willingness-to-pay thresholds. Percutaneous closure of the LAA was 9% and 14% likely to be cost-effective at willingness-to-pay thresholds of $50,000 and $100,000 per QALY, respectively (Figure 5). We generated scatter plots including all the values from these simulations for both trials, which showed relatively modest variability around our base- case estimates of cost and quality adjusted survival (Supplementary Figures 7 and 8).

Figure 4.

Cost-effectiveness acceptability curves representing the probability that each treatment strategy is cost-effective for a given willingness-to-pay threshold per QALY gained in PROTECT AF. Warfarin was most likely to be cost-effective at a willingness-to-pay threshold <$6,000/QALY, dabigatran was most likely to be cost-effective between $6,000 and $21,000 per QALY and LAA closure was most likely to be cost-effective above that. LAA closure was 89% and 98% likely to be cost-effective at willingness-to-pay thresholds of $50,000 and $100,000, respectively. LAA= left atrial appendage, QALY= quality-adjusted life-year.

Figure 5.

Cost-effectiveness acceptability curves representing the probability that each treatment strategy is cost-effective for a given willingness-to-pay threshold per QALY gained in PREVAIL. Warfarin is most likely to be cost-effective at allwillingness-to-pay thresholds. Percutaneous closure of the LAA was 9% and 14% likely to be cost-effective at willingness-to-pay thresholds of $50,000 and $100,000 per QALY, respectively. LAA= left atrial appendage, QALY= quality-adjusted life-year.

Discussion

We found that in patients aged 70 years with AF at an increased risk of stroke (CHADS₂ score ≥1), compared to oral anticoagulation, the quality-adjusted survival and cost-effectiveness of LAA closure with the Watchman device varied considerably based on which clinical trial was used for the base case data inputs and assumptions. Using long term follow-up (mean 3.8 years) data inputs from PROTECT AF, the incremental cost-effectiveness ratio of LAA closure compared with warfarin was $20,486 per QALY. However, using the 18-month follow-up data inputs from PREVAIL, LAA closure had lower quality-adjusted survival and was more costly (dominated in cost-effectiveness terms). At a willingness-to-pay-threshold of $50,000 per QALY, LAA closure was cost-effective 89% of the time under PROTECT AF assumptions and 9% of the time under PREVAIL assumptions. These results were sensitive to the rates of ischemic stroke and ICH for LAA closure and medical anticoagulation, and it was the marked variation in these rates across the trials that accounted for most of the variation in the effectiveness (quality adjusted survival) and cost-effectiveness of the different treatment modalities. These findings indicate that the cost-effectiveness of LAA closure will critically depend on whether rates of adverse events such as stroke and ICH in typical clinical practice mirror those of the clinical trials, and demonstrate the critical need for systematic post-market surveillance of LAA closure.

Our results utilizing longer-term follow-up data from PROTECT AF extend the results of a previous cost-effectiveness analysis using 18-month follow-up data from PROTECT AF and modeled in the Canadian health care system.¹⁹ That study showed an incremental cost-effectiveness ratio of $41,565 per QALY for LAA closure compared with warfarin. Over the 3.8 year follow-up of the PROTECT AF study, the rates of ischemic stroke and bleeding were very favorable for LAA closure relative to medical anticoagulation, leading to better quality adjusted survival and cost-effectiveness. The rate of ICH with LAA closure in particular was dramatically lower than with medical anticoagulation. Over the extended time frame of our base case model, the cumulative effect of ongoing higher relative risk of major adverse events, particularly bleeding, with medical anticoagulation outweighed the early risk of periprocedural adverse events with LAA closure, resulting in a low incremental cost-effectiveness ratio of $20,486 per QALY compared with warfarin.

In contrast, over the 18-month follow-up of the PREVAIL study, LAA closure was associated with similar or higher rates of ischemic stroke compared with medical anticoagulation and higher rates of ICH, leading to lower quality adjusted survival and poor cost-effectiveness. This was driven, in large part, by the substantially lower risk of stroke and intracranial hemorrhage in the warfarin arm of PREVAIL. Importantly, these adverse event rates for warfarin were markedly lower than in PROTECT AF and in the control arms of several recent studies evaluating the novel anticoagulants (Supplemental Table 1).^{4, 6-8} In addition, if the trends in long term bleeding rates are consistent with PROTECT AF (lower in the device arm off anticoagulation), the extended follow-up results from PREVAIL may show an ongoing accrual of major bleeding events with medical anticoagulation that would improve the quality-adjusted survival and cost-effectiveness of LAA closure over time.

Our cost-effectiveness estimates for PROTECT AF and PREVAIL were disparate because the studies were small, limiting the statistical precision of their findings and causing substantial variation in adverse event rates. As such, we believe that our results for the two studies likely represent the range of possible cost-effectiveness estimates. However, given the larger patient experience of the PROTECT AF trial (707 patients compared with 407 in PREVAIL) and the much longer 3.8 year average follow-up time (compared with 18 months in PREVAIL), our estimates from PROTECT AF have less statistical uncertainty. In addition, a recently published study which included pooled data from the 4-year PROTECT AF follow-up, PREVAIL and the two Continued Access protocol (CAP) registries, showed that the major adverse event rates for the CAP registries (enrolling a combined 1145 patients) were consistent with those of the long-term PROTECT AF study, which corroborates those findings.⁷³ Finally, as noted above, the adverse event rates for warfarin in PROTECT AF were much more consistent with those in the control arms of several recent studies evaluating the target-specific anticoagulants (Supplemental Table 1).^{4, 6-8}

Notably, periprocedural complications did not substantially influence the effectiveness or cost-effectiveness of LAA closure in our study. Although periprocedural complications were more common in PROTECT AF than in PREVAIL, our analyses demonstrated that the model was relatively insensitive to changes in the risk of these events. These procedural complications occur acutely and over time their influence on quality adjusted survival and cost-effectiveness is overwhelmed by the effects of highly morbid and costly adverse events such as stroke and intracranial hemorrhage. Similarly, in a sensitivity analysis to simulate inexperienced operators in which we decreased the rate of successful LAA closure device implantation to 80% and increased each of the periprocedural adverse events rates to the highest value in their plausible ranges, we found that the quality adjusted survival and cost-effectiveness was only moderately affected.

For our study, several caveats should be considered. First, comparison of LAA closure to dabigatran is an indirect one, as they have not been studied in any head-to-head trials. However, secondary analyses from RE-LY demonstrated that the effectiveness of dabigatran appeared consistent across many strata of patients, suggesting that it is reasonable to assume consistent effectiveness in a comparison with the population from the Watchman trials.^{4, 5, 74} Second, cost-effectiveness may vary across the novel anticoagulants, but meta-analysis has demonstrated that major adverse events rates including ischemic stroke, intracranial hemorrhage, and death are similar across the target-specific anticoagulants. Third, the base case inputs used from PROTECT AF and PREVAIL had different follow-up times, and extended follow-up could lead to significant differences in long-term event rates. In the index PROTECT AF publication with a mean follow-up of 18 months, LAA closure was non-inferior to warfarin,¹⁶ whereas after a mean follow-up of almost four years, LAA closure was superior across several primary and secondary endpoints.¹⁸ Because PREVAIL has reported only 18-month follow-up so far,¹⁷ there may be greater uncertainty with PREVAIL-based estimates until longer-term data is reported. Fourth, clinical safety and effectiveness may depend on a number of factors not captured in these trials, including patient selection, operator or site factors, and concomitant use and duration of anticoagulation or antiplatelet therapy following LAA occlusion. If safety or effectiveness is different outside of the clinical trial setting, this could substantially affect value.⁷⁵ Finally, our study findings only apply to patients who are eligible for anticoagulation, as patients with contraindications to anticoagulation were excluded from both PROTECT AF and PREVAIL.

In conclusion, we found that the quality adjusted survival and cost-effectiveness of LAA closure with the Watchman device varies substantially depending on whether the clinical event rates reflect those of the PROTECT AF or PREVAIL clinical trials. The cost-effectiveness of LAA closure using PROTECT AF data was in a range generally considered to be cost-effective. Using data from PREVAIL, however, LAA closure was dominated by warfarin and dabigatran, meaning that it was less effective and more costly. PROTECT AF enrolled more patients and has substantially longer follow-up time, allowing greater statistical certainty with the cost-effectiveness results. Longer term trial results and post-marketing surveillance of major adverse events will be vital to determining the ultimate value of this novel treatment modality in clinical practice.

What is Known

• Randomized trials of left atrial appendage (LAA) closure with the Watchman device have shown varying results, and its cost-effectiveness compared to anticoagulation has not been evaluated using all available contemporary trial data.

What the Study Adds

• Using data from the PROTECT AF trial, LAA closure with the Watchman device was cost-effective

• Using PREVAIL trial data, Watchman was more costly and less effective than warfarin and dabigatran.

• PROTECT AF enrolled more patients and has substantially longer follow-up time, allowing greater statistical certainty with the cost-effectiveness results, but longer term trial results and post-marketing surveillance of major adverse events will be vital to determining the value of the Watchman in clinical practice.