An economic evaluation of vagus nerve stimulation as an adjunctive treatment to anti-seizure medications for the treatment of drug resistant epilepsy in the United States

Abstract

Introduction

People with recurrent epileptic seizures are typically treated with anti-seizure medications (ASMs). Around a third of epilepsy patients fail to achieve an adequate response to ASMs and may be eligible to receive vagus nerve stimulation (VNS) therapy for their drug-resistant epilepsy (DRE) if they are unsuited to surgery. VNS received approval from the United States (US) Food and Drug Administration agency. However, there has to date been no comprehensive cost effectiveness evaluation of VNS within the US setting. This study was designed, using a US Medicare perspective, to estimate costs and quality-adjusted life years (QALYs) associated with VNS as an adjunct to ongoing ASM therapy, compared to ASMs alone.

Methods

We developed a cohort state transition model in Microsoft Excel, with four health states defined by different percentage reductions in seizure frequency, with a 3-month cycle and transition probabilities derived from published clinical trials and registry data. Sensitivity analyses were conducted to understand the impact of parameter uncertainty. Costs included the VNS device, placement, programming, battery changes, and removal; ASM therapy; adverse events associated with VNS (dyspnea, hoarseness, and cough); and costs associated with seizure burden (i.e. hospitalizations, emergency department visits, neurologist visits).

Results

Under base case assumptions, treatment with VNS was associated with a 0.385 QALY gain and a $109,678 saving per patient, when compared with ASM therapy alone. The incremental net monetary benefit (iNMB) was $128,903 at a threshold of $50,000 per QALY, with the positive iNMB indicating that VNS is a highly cost effective treatment. This result is explained by the modeled reduction in relative seizure frequency and associated reduction in healthcare resource use that the VNS group experienced. Sensitivity analyses supported this conclusion.

Conclusions

VNS was evaluated as a cost effective addition to the current standard of care in the treatment of DRE in the US Medicare context.

PLAIN LANGUAGE SUMMARY

Anti-seizure medications (ASMs) are drugs commonly prescribed to people with epilepsy to help prevent seizures from reoccurring. But these drugs do not work for all people: around a third keep having seizures despite taking the medication—a condition called drug-resistant epilepsy (DRE). For such people, their main options involve trying different combinations of ASMs, having brain surgery, or having a medical device implanted. In the United States (US), vagus nerve stimulation (VNS) therapy is approved by the Food and Drug Administration agency for DRE patients who are still having focal onset seizures despite trying ASM therapy. Using methods defined by the US evaluation body, the Institute for Clinical and Economic Review, we made an economic model to assess how cost effective VNS would be as an add-on to ASM therapy. The evaluation utilizes a previously published model, which was updated to use costs, health-related quality of life, and mortality estimates relevant to the US Medicare setting. The analysis demonstrated that VNS could generate cost savings when used as an add-on ASM treatment in the US Medicare setting. VNS may reduce the number of seizures, and subsequently improve patient quality of life and result in substantially lower costs for Medicare (e.g. in emergency and hospital care for a person having a seizure). We tested uncertainties in our model using standard methods—these additional analyses allow us to conclude that VNS is highly likely to be a cost effective addition to the current standard of care for DRE treatment from a US perspective.

Keywords:

• Cost effectiveness analysis
• Markov model
• neuromodulation
• refractory epilepsy
• anti-seizure medications
• vagus nerve stimulation

JEL CLASSIFICATION CODES:

• C50
• C5
• C
• I00
• I
• C60
• C6

Introduction

Around a third of people with epilepsy fail to achieve an adequate response to treatment with anti-seizure medications (ASMs)¹. Drug-resistant epilepsy (DRE) is a serious and potentially life-limiting condition² with significant health system burden³. The prevalence of DRE varies geographically and is dependent on the use of the definition of drug resistance^4,5 but is universally found to be associated with more frequent hospitalization and emergency department visits when compared to patients responding to ASMs^6,7. In the United States (US), the estimated financial burden of DRE is $4 billion per annum in direct and indirect costs⁸. Patients with DRE have significantly poorer quality of life⁷, with depression⁹ and cognitive impairment¹⁰ most commonly reported.

Vagus nerve stimulation (VNS) has been an important treatment option in DRE for three decades, with over 125,000 patients having received it since 1989¹¹. In the US, VNS is approved for use by the Food and Drug Administration (FDA) agency as an adjunctive treatment in reducing the frequency of seizures in patients aged four or older with focal onset seizures, who are not responsive to ASMs¹². As an already established treatment for DRE, there is abundant evidence demonstrating the clinical efficacy of VNS^13–15. A recent systematic review and meta-analysis of 30 studies in people with DRE¹⁶ observed similar efficacy for three modalities of neurostimulation during initial blinded phases of randomized controlled trials (RCTs), albeit with different eligibility criteria.

Whilst several international studies have assessed the cost effectiveness of VNS and have found VNS a cost effective strategy to manage DRE in adults in countries, such as the Netherlands, China, and England^17–19, cost effectiveness studies of VNS in the US remain scarce¹⁶. Recent literature reviews searching for cost effectiveness analyses found no US cost effectiveness analyses of VNS in adult DRE, with only studies analyzing budget impact or cost consequence, pediatric populations, or responsive neurostimulation (RNS) devices found^15,16. A recent study found that RNS was cost effective vs. ASM alone, with an incremental cost effectiveness ratio (ICER) of $28,825 per quality-adjusted life year (QALY)²⁰. The objective of this study was to develop the first comprehensive cost-utility analysis (CUA) of VNS as an adjunct therapy to ongoing ASM in adult DRE patients. As per a previously published cost effectiveness study of neurostimulation in DRE patients, the perspective adopted was of US Medicare²⁰. This is in line with the Institute for Clinical and Economic Review reference case, which recommends the inclusion of costs paid by third-party payer²¹. Moreover, DRE patients are eligible for Medicare reimbursement²², with ∼50% of VNS patients in the US covered by Medicare²³.

Methods

The main model design elements and methods, including the health state structure and effectiveness parameters, are based on a published English National Health Service (NHS) cost effectiveness model for VNS in DRE¹⁹. Since the publication for the English NHS analysis contains a full account of the model development, the Methods section for our US analysis provides only a brief recapitulation of the main model setup, followed by full details regarding the steps taken to adapt this model for a de novo analysis applicable to the US setting.

Design and population

Our CUA complies with the US Institute for Clinical and Economic Review reference case²¹ to estimate the costs and benefits of two DRE management strategies:

1. ASM regimen as typically administered in DRE management, with no VNS.

2. VNS administered as an adjunct to the ASM regimen.

Other management strategies, such as RNS and deep brain stimulation (DBS), were considered as comparators; however, they were not included in the analysis due to a scarcity of comparative evidence with VNS.

Health benefits were estimated as QALYs, and costs took a US Medicare perspective. To account for time preference, costs and benefits were discounted at an annual rate of 3%, as specified in the US reference case²¹. A time horizon of 10 years was used to capture the most relevant costs and benefits relating to treatment with DRE, and to reflect the extent of available evidence. The ICER, calculated as incremental costs over incremental QALYs, is evaluated against a threshold; if it falls below the threshold, the intervention of interest can be considered cost effective. In this analysis, the US reference case range of $50,000 per QALY to $150,000 per QALY was used²¹.

Two pivotal RCTs (named E03 and E05) evaluating the efficacy of high vs. low stimulation VNS therapy given adjunctively to ASMs were used to inform the model cohort population characteristics and are summarized in the Technical Supplement^24,25.

Model structure

We used the same cohort state transition model as the English NHS analysis¹⁹. Patients could be in one of four health states defined by percentage reductions in seizure frequency achieved from baseline, with the lowest category (<50% reduction) corresponding to “no response” and the highest (100% reduction) representing “seizure-free”. Death was also modeled as an absorbing state. A graphical summary is provided in Figures 1 and 2.

Figure 1. Model structure for VNS + ASM group. ASM: anti-seizure medication; VNS: vagus nerve stimulation.

Figure 2. Model structure for ASM only group. ASM: anti-seizure medication; VNS: vagus nerve stimulation.

Although not directly quantifying epilepsy severity, these health-state categories are meaningful with respect to therapeutic goals and have been used in other published health economic analyses in this field^26–28.

Transitions occurred on a 3-month cycle basis. All patients started in the “no response” health state, and patients in the VNS arm could transition between health states any time up to two years after implantation, consistent with published long-term observational data²⁹. After this, their health state remained fixed (subject to death or discontinuation). By contrast, ASM only patients moved between health states in the first 3-month cycle, whereafter their response remained static.

Clinical evidence—usage and assumptions

Efficacy of VNS and health state transitions

The sourcing and application of efficacy parameters were identical to the English NHS model¹⁹. Individual patient-level data from the international trials of VNS E03 and E05 informed efficacy estimates, both for the VNS and “ASM only” groups, for the first three months of the model^24,25. Longer-term effectiveness of VNS was based on estimates from a systematic review of VNS studies²⁹, validated and cross-checked against real-world registry data and open-label trial extension data³⁰.

For “ASM only” patients, we found limited evidence with which to model longer-term changes in response. Therefore, we took the assumption that modeled “ASM only” patients would maintain in perpetuity the distribution of response categories achieved by “low stimulation” arm patients in the E03 and E05 studies. The simplicity of the assumption matches the data scarcity and allows for transparent sensitivity analysis.

Adverse events

Three adverse events commonly observed in the VNS clinical trials were included in the analysis as annual risks: hoarseness, cough, and dyspnea³⁰. Adverse effects associated with ASMs were not modeled since the base case assumed that both modeled cohorts received the same stable ASM regimens throughout the time horizon.

The most common complication of VNS implantation is surgical site infection, which has been estimated to occur in 1.3% of patients³¹. As per the previously published model¹⁹, any health-related quality of life (HRQoL) impact of surgical site infection was assumed to be negligible in the context of the time horizon, given the infrequent and one-off nature of the event³¹. Further, the Hospital-Acquired Condition Reduction program financially disincentivizes infections, with some costs of infection falling outside the Medicare perspective³². A sensitivity analysis, whereby the cost of implementing VNS was increased substantially, was conducted to explore uncertainty of this parameter (i.e. to understand if conclusions would change even if infection rates were high and included in addition to the Medicare reimbursement cost).

Device explantation or replacement due to battery depletion

Patients receiving VNS could discontinue treatment due to non-response (leading to the deactivation of the device) or if explantation (for any cause) was needed, with the latter involving an additional procedure cost. Replacement of VNS devices due to battery depletion was associated with a respective cost, but patients retained the health benefits of VNS. Time to device explantation and battery replacement were modeled using survival models fit to Kaplan-Meier estimates from real-world registered implant data from the US, detailing the use of VNS (AspireSR 106) in 26,261 patients for up to 7.14 years³³.

Health-related quality of life

A quality-of-life utility score for each health state was obtained using data from a prospective study of 81 patients by Messori et al.³⁴, which estimated utilities using the time trade-off method³⁵ for a set of seizure frequency categories. These categories, which reflected absolute numbers of seizures, could be mapped with reasonable similarity onto the percentage seizure reduction categories. The utilities used for each health state are summarized in Table 1.

Table 1. Health-related quality of life values used in model.

Parameter	Corresponding seizure frequency group reported in Messori et al.^34	Mean utility^a	Range for DSA
0–50% reduction (no response)	10+ per month	0.66	0.65–0.67
50–74% reduction (responder)	2–9 per month	0.79	0.78–0.80
75–99% reduction (responder)	0–1 per month	0.91	0.90–0.92
100% reduction (seizure free)	0 seizures per year	0.96	0.95–0.97
Baseline for population^b		0.86^b	–

DSA: deterministic sensitivity analysis.

^aMean values displayed are as reported by or derived from the literature. These are used to understand the relative difference between model health states.

^bThis value represents the age- and gender ratio-adjusted baseline for “seizure free” patients to which the relative utility differences between health states should be applied³⁶.

Health state utilities were adjusted over time to reflect the expected decline in utility as patients age. It was assumed that seizure-free patients would have a utility equal to the US general population over time. To adjust the utilities for each health state over time, we first calculated the relative difference between the health state utilities and the utility of the US general population adjusted for the baseline mean age and gender mix of the modeled DRE population, i.e. 0.96³⁶. We then applied these relative differences to the US baseline population utility value corresponding to the mean age of patients at each cycle in the model to calculate the adjusted health state utilities.

Consistent with the English NHS model, the three types of VNS adverse events were modeled to have an HRQoL decrement¹⁹.

Costs and resource use

Health state costs

Healthcare resource unit costs, summarized in the Technical Supplement, were obtained from publicly available sources, in line with the US reference case²¹. Primary care costs were not included as care for DRE patients was assumed to take place in a secondary care setting. Secondary care (hospital) episodes were costed using two statistical briefs of Healthcare Cost and Utilization Project data published by the Agency for Healthcare Research and Quality, which reflects Medicare costs^37,38. Costs were inflated to 2021 using medical care inflation rates from the Bureau of Labor Statistics³⁹.

The cost of outpatient visits associated with each health state in the model is based on the national payment amount of $103.82 for service in a non-facility setting (current procedural terminology: 99,214), assuming a Medicare reimbursement of 80% of the visit⁴⁰.

Healthcare resource assumptions

The method for estimating healthcare resource use (HCRU) in each health state was identical to the published English NHS model¹⁹, except that estimates for the 50–74% seizure reduction health state were obtained from a published US study⁴¹. The annual health state resource use frequencies were multiplied by unit costs to give annual costs as shown in Table 2, with frequencies varied by the ranges shown for sensitivity analysis.

Table 2. Summary of annual healthcare resource costs and utilization.

	Health state costs
Health care resource item	100% seizure freedom^a	75–99% seizure reduction^b	50–74% seizure reduction^c	<50% seizure reduction^d
Base level of seizures	0	243	287	506
ED visits	$0 (0.00)	$446 (0.61 [0.52–0.70])	$527 (0.72 [0.62–0.83])	$1,610 (2.21 [1.6–2.92])
Inpatient visits	$0 (0.00)	$15,401 (1.13 [0.96–1.30])	$18,206 (1.33 [1.13–1.53])	$75,309 (5.51 [3.98–7.29])
Outpatient neurologist visits	$156 (1.50 [1.28–1.73])	$1,087 (10.47 [8.90–12.04])	$1,285 (12.38 [10.52–14.23])	$2,337 (22.51 [16.26–29.77])
Total HCRU	$156	$16,933	$20,018	$79,256

ED: emergency department; HCRU: healthcare resource use.

^aAs reported by Kristian et al.²⁷.

^bInformed by the mean number of seizures of each response group from E03 and E05 (at 3 months follow up for response states, i.e. a 15% reduction in resource use from the 50–74% seizure reduction health state is assumed for each HCRU.

^cInformed by: Bernstein et al.⁴¹.

^dInformed by applying the lower estimate of the relative difference expected between “uncontrolled” and “controlled” epilepsy reported by a US study⁴². To note, Manjunath et al.⁴² did not define “hospitalization” and the ratio is applied uniformly to estimates in-patient admissions derived using data from Bernstein et al.⁴¹.

Anti-seizure medication costs

No direct, publicly available evidence on the schedules of ASMs administered to DRE patients was identified for the US context. The model used an annual ASM cost estimate of $21,772 (inflated cost: $24,021.91) reported in a US cost effectiveness study of surgery for drug-resistant temporal lobe epilepsy⁴³. This analysis assumes that the costs and effects of ASM are the same for the modeled VNS and ASM only groups.

VNS costs—implantation, explantation, and battery replacement

An overview of the values used to estimate the cost of VNS implantation, as well as explantation and replacement where applicable, is given in Table 3. Implantation procedure costs include the costs of full system implant in hospital settings and full system placement in facility settings; the cost reflects that ∼93% of VNS procedures are expected to take place in an outpatient or ambulatory surgical care setting⁴⁶.

Table 3. Resource use and costs for VNS.

Parameter	Cost for base case	Source
Implantation procedure	$30,691	Includes Full system implant or replacement (Hospital) (CPT code: 64568; APC code: 5465) and Full system placement or replacement in facility settings (Surgery) (CPT code: 64568)^44.
VNS therapy device	$37,078	Cost of SenTiva generator^45. Cost applied in a scenario analysis.
Pre- and post-operative neurologist review	$54.33	Complex programming includes adjustment of more than three parameters in non-facility settings (CPT code: 95977)^44. $362 was applied for six outpatient reviews.
Explant procedure	$11,895	Surgical removal of generator or receiver in facility settings (CPT code 61888; APC code: 5463)^44. Costs associated with removal of generator or receiver in hospital and surgery setting (CPT code 61888)^44.
Replacement procedure	$21,455	Inclusive of surgical and hospitalization costs of Replacement of generator or receiver with connection to a single electrode array (CPT code 61885)^44.

APC: ambulatory payment classification; CPT: current procedural terminology; VNS: vagus nerve stimulation.

Adverse event costs

The expected resource use and associated cost for a patient experiencing hoarseness, cough, or dyspnea over the course of 1 year were estimated at $156. This corresponds to 1.5 neurologist visits to reflect the general treatment and management of the adverse event, including potential changes in device programming⁴⁰. The impact of the changing of the programming of the device on seizure reduction following an adverse event was not specifically modeled due to a lack of available data, in line with the previously published model¹⁹. The annual cost was converted to a mean cost per cycle ($39) and applied proportionally to the percentage of the cohort experiencing an event in any given cycle.

Sensitivity and scenario analyses

Deterministic sensitivity analysis

To identify the key drivers of model outcomes, single parameters or groups thereof were varied as appropriate - by a plausible minimum and maximum range reported from the source, or by ±15% where ranges were not available, keeping all other inputs constant.

Probabilistic sensitivity analysis

Parameter uncertainty was assessed by probabilistic sensitivity analysis (PSA) using 5,000 simulations and the likelihood of being cost effective at various thresholds evaluated. The parameter distributions used in PSA are provided in the Technical Supplement.

Other sensitivity analyses

A set of threshold and scenario analyses were designed to test the model’s sensitivity to the following concepts:

• mapping HCRU to health states

• assumptions regarding longer-term treatment effect

• expected inpatient and elective cost burden for a DRE patient, designed to reflect and explore the impact of potential discrepancies between Medicare reimbursement and the total cost of provision of VNS, or regional differences in cost within the US of healthcare provision more generally (this latter scenario was run via a two-way sensitivity analysis that varied hospitalization and additional device costs).

All scenario analyses and their results are reported in detail in the Technical Supplement.

Results

Base case

The cost effectiveness analysis estimates that the total cost over a 10-year time horizon is $689,959 for the VNS group and $799,637 for the ASM only group. The total QALYs are estimated at 6.051 for the VNS group and 5.666 for the ASM only group. The incremental total cost difference is $109,678 favoring VNS, with an estimated incremental QALY gain of 0.385. Therefore, VNS is expected to be cost effective and dominate ASM only (i.e. cost saving and conferring more QALYs than the comparator). A breakdown of the results is provided in Table 4.

Table 4. Base-case cost effectiveness results.

Group	VNS + ASM	ASM only	Increment
Costs
Device cost sub-total	$39,824	$0	$39,824
Implantation	$31,017	$0	$31,017
Explantation	$210	$0	$210
Replacement	$8,596	$0	$8,596
ASM acquisition	$204,256	$204,256	$0
TRAEs costs	$141	$0	$141
Health state costs sub-total	$445,738	$595,380	-$149,643
Acute care	$9,429	$12,250	-$2,821
Inpatient	$420,958	$564,679	-$143,720
Outpatient	$15,350	$18,452	-$3,101
Total cost	$689,959	$799,637	-$109,678
Health outcomes
Life years	8.503	8.503	0
QALYs	6.051	5.666	0.385
Incremental outcomes
ICER (per QALY)	VNS dominates
Incremental net monetary benefit ($50,000 threshold)	$128,903
Incremental net health benefit ($50,000 threshold)	2.58

ASM: anti-seizure medication; ICER: incremental cost effectiveness ratio; QALY: quality-adjusted life year; TRAE: treatment-related adverse event; VNS: vagus nerve stimulation.

Deterministic sensitivity analysis

As Table 5 shows, no input tested had the potential to change conclusions regarding cost effectiveness when considering the ICER threshold range of $50,000–$150,000 per QALY (noting that a positive incremental net monetary benefit [iNMB] indicates cost effectiveness at the threshold tested). A tornado plot is provided in Figure 3.

Figure 3. Deterministic sensitivity analysis—tornado diagram. ASM: anti-seizure medication; ED: Emergency Department; NMB: net monetary benefit; QALY: quality-adjusted life year; SMR: standardized mortality ratio; TRAE: treatment-related adverse event; VNS: vagus nerve stimulation. Notes. Where a parameter sensitivity analysis involved change to a single value, lower and upper bound values are shown on the diagram; ^¥lower or upper bound applied to the group of inputs informing the parameter simultaneously.

Table 5. Top 15 most sensitive parameters.

	iNMB
Parameter varied	Lower bound value	Upper bound value	Delta
Base-case	$128,903		N/A
Inpatient resource use^¥	$84,793	$180,973	$96,180
Probability of <50% seizure reduction for VNS in first cycle (0.74 [0.85–0.63])	$81,143	$176,663	$95,520
Explantation probabilities^¥	$131,810	$66,806	$65,004
Long term probabilities^¥	$106,732	$155,098	$48,366
VNS and ASMs clinical trial probabilities^¥	$151,421	$106,385	$45,036
Inpatient admission unit costs ($13,668 [$11,618–$15,719])	$107,345	$150,461	$43,116
ASM reduction with VNS (1.00 [0.85–1.15])	$146,934	$110,872	$36,061
Total implant-related cost ($31,017 [$26,365–$35,670])	$133,556	$124,250	$9,305
Health state utilities^¥	$130,162	$127,644	$2,519
Replacement cost ($21,781 [$18,514–$25,048])	$130,192	$127,614	$2,579
Outpatient resource use^¥	$127,738	$130,315	$2,577
ED resource use^¥	$127,998	$129,979	$1,981
TRAE probabilities^¥	$129,725	$127,941	$1,784
Neurologist outpatient visit unit cost ($104 [$88–$119])	$128,459	$129,347	$888
ED unit cost ($728 [$619–$837])	$128,480	$129,326	$846

ASM: anti-seizure medication; ED: emergency department; iNMB: incremental net monetary benefit; N/A: non-applicable; TRAE: treatment-related adverse event; VNS: vagus nerve stimulation.

Notes. Incremental net monetary benefit calculated using a threshold of $50,000 ((incremental benefit × threshold)−incremental cost).

^¥Lower or upper bound applied to the group of inputs informing the parameter simultaneously (please see supplement for additional information).

The model is most sensitive to the resource estimates and unit cost applied for hospitalization. The model is also sensitive to the probability of <50% seizure reduction for VNS in the first cycle, mostly because HCRU costs are directly linked in our model to seizure reduction. This echoes findings in the Raspin et al. study, which gives further explanation to the relationships between the parameters and the results displayed here¹⁹.

An increase in the aggregate cost of the VNS device and implantation of just over $75,000 was the threshold at which VNS no longer dominated ASM only (full tabulation is provided in the Technical Supplement). This provides reassurance that should additional costs (i.e. for infection) need reimbursement, VNS would still be considered a cost effective strategy.

Scenario analyses

Mapping healthcare resource use to health states

Healthcare resource consumption is a key driver of cost effectiveness in this therapy area, yet the methods and assumptions used in this model are subject to uncertainty. Therefore, a range of alternative scenarios was tested and written up in the Technical Supplement. In all scenarios tested, VNS remained a dominant strategy compared to ASM alone.

Long-term clinical effectiveness

Exploratory scenarios were tested in which the long-term effectiveness of VNS, or its modeled rate of explantation, were varied. In all scenarios, VSM remained cost saving or cost effective compared to ASM alone.

Expected inpatient and elective cost burden for a DRE patient

Hospitalization is the most costly form of HCRU in the model, so a set of scenarios around the cost and usage of inpatient care was explored and is reported in the Technical Supplement. VNS continued to dominate vs. ASMs alone for hospitalization costs over $4,000 and remained cost effective at a $50,000 threshold with a hospitalization cost of over $2,000.

The analysis was also rerun using the procedure associated with the CPT code 99215 for outpatient visits ($183.07 in facility settings), which is reimbursed up to $146.50 by Medicare. This change did not impact on the conclusion of dominance of VNS when compared to ASM therapy alone.

Probabilistic sensitivity analysis

In PSA, VNS was dominant in 100% of the 5,000 iterations, with a probabilistic iNMB of $129,337 (95% CI: $64,823–$203,833) using a threshold of $50,000. A PSA scatter plot is provided in Figure 4.

Figure 4. Probabilistic sensitivity analysis—scatter plot. ASM: anti-seizure medication; QALY: quality-adjusted life year; VNS: vagus nerve stimulation.

Discussion

Overview

This novel analysis evaluates the cost effectiveness of VNS in the US Medicare setting, using contemporary cost estimates and granular patient-level efficacy data in a detailed long-term model. The base case results suggest that VNS is a cost saving intervention in the US setting, with a dominant ICER value ($109,678 cost savings and 0.385 QALYs gained over a ten-year horizon). The majority of savings seen in the VNS group are realized through hospitalizations avoided compared to the amount expected in the ASM only group. The higher the expected cost of hospitalization, the greater the saving that can be realized through reduced seizure burden.

In the US, DRE patients have access to several non-pharmacological interventions, including VNS, RNS, DBS, and surgery (if suitable). Although there have been no prior US-based cost effectiveness analyses of VNS in an adult population, published analyses for RNS and surgery provide informative comparisons of methodology.

A recent study by Youngerman et al. estimated the cost effectiveness of RNS vs. ASM alone²⁰, using comparable methods to link HCRU and utility values to defined health states. The study found RNS was associated with an incremental cost vs. ASM, which was considered cost effective given the incremental health benefits. The relative difference in HRQoL between health states for the RNS model was greater than for this study; with Youngerman et al. using utilities elicited from the general population in South Korea for hypothetical focal onset seizure health states, and our analysis using utilities elicited from patients with medically refractory epilepsy in Italy. Had the Youngerman et al. source been used in our study, a higher incremental QALY gain would be achieved in favor of VNS. Another key difference between study methodology is the modeling of adverse events, with our study opting to model the risk of battery replacement and explantation of the device—as well as the potential for adverse events, such as cough, dyspnea, and hoarseness.

US-based studies evaluating the cost effectiveness of surgery have also been carried out for drug-resistant temporal lobe epilepsy⁴³ and Lennox-Gastaut Syndrome (LGS)⁴⁷. These studies used alternative model structures to our analysis to capture clinical risks and outcomes associated with surgery. In the temporal lobe epilepsy study, surgery was found to be cost saving and more effective—that is, dominant—compared to medical management in surgically eligible patients⁴³. In the LGS analysis, corpus callosotomy was to be less cost effective than VNS with an incremental cost of $451,952 per positive seizure outcome gained over a 1-year time horizon⁴⁷.

Despite limited evidence of cost effectiveness of VNS in DRE in the US, our US-based study supports the finding that VNS is considered a cost effective option managing DRE in adults, as found in other international cost effectiveness studies (such as a Dutch study comparing VNS to DBS and care usual¹⁷ and a Chinese evaluation of VNS vs. ASM¹⁸.

The most relevant international cost effectiveness study of VNS to consider is the 2021 English NHS model by Raspin et al.¹⁹, which used the same health state structure and clinical effectiveness data sources. Similar to the current study, the main driver of cost effectiveness in the English NHS model was inpatient costs—these were avoided in the VNS group in amounts that offset device and service costs, making VNS a cost effective intervention compared to ASM alone.

When modeled for the US Medicare perspective, threshold analysis results suggest that even if payers have to pay up to $75,000 in additional device costs (i.e. those potentially not covered by Medicare reimbursement), VNS as an adjunct to ASM therapy remains cost saving when compared to ASM therapy alone. Additional threshold analyses of our model found that VNS remained dominant as long as the hospitalization unit cost was over $4,000 and cost effective (at a $50,000/QALY threshold) as long as it was over $2,000.

With HCRU differences generating such large per-patient savings, alternative data sources and assumptions were tested in scenario analyses to inform the modeled HCRU estimates. Our healthcare resource utilization estimates came from a 2007 US-based observational study⁴¹ which is likely to be the most relevant for a US setting. The model continued to support the conclusion that VNS is cost effective across numerous other scenario analyses (see Technical Supplement).

Key assumptions considered in detail

Since the model structure in large part mirrors that of the English NHS analysis, the main considerations around model structure and assumptions are described in that publication¹⁹. Briefly, there are four main considerations. First, ASM seizure reduction rates were sourced from the control arms of the E03 and E05 trials. These control arms were not true placebo but rather “low stimulation” VNS, so the model may underestimate the true relative incremental benefit of VNS. Second, the “ASM only” group was assumed to permanently retain the level of response achieved by control arm patients in the E03 and E05 trials after three months (around 15.7% reached “responder” health states). This assumption is simplistic and captures neither long-term therapeutic decisions over time about ASM in DRE patients nor potential changes to seizure frequency. However, the assumption reflects the lack of long-term data available¹⁶ and potentially favors the “ASM only” group, in whom absolute response numbers have been observed to decline^48–50. Third, a constant rate of mortality was applied across all health states. However, sudden death in epilepsy risk is higher for people with ongoing seizures than for those who are seizure free⁵¹. Finally, the E03 and E05 trials used older devices, yet our model device costings were based on the currently available SenTiva model of VNS generator. The SenTiva device has an “AutoStim” feature that triggers stimulation when there is a change in heart rate, which may provide better seizure control than previous models^52,53, and have a different lifespan and likelihood of explantation.

There were also specific assumptions to our US-based model worth noting. Medicare reimbursement costs—rather than unit financial costs—were used. This means that, in general, the modeled costs of healthcare, in particular for hospitalization and inpatient care associated with seizures, may underestimate the true cost borne by the hospitals. In turn, this means that seizures avoided through VNS may in fact be associated with greater healthcare resource savings. Additionally, implantation procedures were also costed using appropriate Medicare reimbursement amounts. However, since this reflects a potential situation whereby hospital providers are not fully covered by Medicare for the costs of implanting and maintaining VNS devices, we also performed sensitivity analyses in which the implantation and maintenance costs were added to the direct up-front device cost as a single bundle. Even in this most pessimistic scenario, VNS remained dominant compared to using ASMs alone. Further analyses using cost estimates from alternative perspectives, such as Medicaid or private insurance could be conducted to further understand the impact of VNS on the US healthcare system.

Strengths and limitations

Strengths

This study used individual data from 310 patients in E03 and E05 to inform efficacy. Long-term follow-up data came from registries collected in 3,182 patients over 24 months and 1,194 patients over 48 months across multiple countries. A systematic review combining data from 2,689 patients was also used, demonstrating that treatment response on VNS is sustained, if not enhanced, over time in real clinical practice²⁹. Furthermore, survival analysis of long-term data in a sample of 26,286 patients was used to estimate time to battery replacement and explantation for VNS over time.

Limitations

Using percentage reductions in seizure rates to define health states is considered reasonable, but it necessitates mapping estimates of HRQoL and HCRU from published studies, which typically report their values against absolute seizure rates. Where possible, we tied percentage reductions to commonly reported clinical outcomes in epilepsy, including “seizure freedom” (100% reduction), “non-responder” (<50% reduction), and “responder” (>50% reduction). However, this approach does not address potential differences in absolute baseline seizure rates between the referenced studies. Where possible, we matched and adjusted values to counter this type of variety, but some misalignment remains. It was also not possible to determine how well the severity of seizures aligns between publications. For example, we assumed that “refractory seizures” captured in the HRQoL publication by Messori et al. were not severe due to the relatively high quality of life reported by patients who were experiencing them. By contrast, 55.0–70.4% of patients in the E03 trial experienced focal to bilateral tonic clonic (referenced in the trial as secondary generalized) seizures, which can have a considerable impact on a patient’s HRQoL.

It was necessary to adapt and use HCRU data reported from multiple US and non-US healthcare settings^27,41,42 to complete the matrix of health state costs. This posed further challenges since each publication differed in its definition and categorization of response. There is also uncertainty in whether patients in the “<50% response” category may have more costly inpatient stays compared to patients who have achieved some level of seizure control due not only to possible differences in seizure severity, but also their pattern of long-term recurrence. The model uses the same unit cost per healthcare item, irrespective of which seizure reduction health state it is used in. Cost savings of VNS could be underestimated if the “<50% response” group ends up consuming more healthcare resources when hospitalized.

Finally, the model does not include the benefits of a relative reduction in seizure frequency for carers of an epilepsy patient or the societal costs of lost productivity, such as those reported by Gupta et al.⁵⁴. It is conceivable that incorporation of such benefits would increase the cost effectiveness of VNS.

Conclusion

Under base case assumptions and when taking a US Medicare perspective, VNS is considered a cost saving intervention as an adjunct to ASMs in patients with DRE who are not suitable for surgery. There is uncertainty in how VNS impacts on the reduction of disabling seizure and therefore its relationship with HRQoL, HCRU, and costs. The analyses presented should be considered exploratory until further data are forthcoming. If future research indicates a different ratio of HCRU between health states, then the conclusions of the analysis could change. This said, a wide range of potential scenarios, including a scenario whereby the cost of providing VNS was increased substantially, were modeled and results consistently indicated that VNS was more cost effective than ASM alone.

Transparency

Declaration of funding

This study was funded by LivaNova PLC.

Declaration of financial/other relationships

FB, JM, and VD have received support from LivaNova PLC as paid employees. CR, JA, and VP have served as paid consultants from Symmetron Ltd., they were contracted by LivaNova PLC to undertake the study. EF has been a member of scientific advisory boards for Biogen, Eisai, Neurelis, SK Life Science, Sage Pharmaceuticals, and the Centers for Disease Control and Prevention, and has received research support from UCB Pharma.

Author contributions

All authors contributed to the critical revision and final review of the manuscript. CR, JA, and VP developed the economic model and undertook the initial analysis and interpretation of results. FB, JM, and VD provided advice on the conceptual model and, FB, JM, VD, and EF assisted in the provision of data that informed the model. CR and JA drafted the manuscript, with kind medical writing assistance from Jana Tillotson. VP, EF, FB, JM, and VD provided a review of model structure, inputs and contributed to the interpretation of results. All authors contributed to the critical revision and final review of the manuscript.

Acknowledgements

The authors are grateful to Jana Tillotson, Medical Writer, for her kind assistance in preparing the draft manuscript. The authors are thankful to Šárka Sirůčková (Project Coordinator, Symmetron) for her kind assistance in the manuscript submission process.

Reviewer disclosures

Peer reviewers on this manuscript have received an honorarium from JME for their review work but have no other relevant financial relationships to disclose.