The cost effectiveness of zonisamide as adjunctive therapy in adult partial seizure epilepsy

Summary

Many patients with epilepsy continue to have partial seizures requiring add-on antiepileptic drugs (AEDs). Zonisamide, a broad-spectrum AED, has recently entered the European market and must be compared with the current standard of care. The objective of this study was to determine the cost effectiveness of zonisamide as treatment for adults with uncontrolled partial epilepsy in Scotland. A Markov decision model was developed from the perspective of the Scottish National Health Service. Outcome measures included quality-adjusted life-years (QALYs) and costs in GBP.

Patients treated with zonisamide had a 15-year incremental cost of £20 and 0.026 additional QALYs compared with patients treated with levetiracetam. These modest differences result in an incremental cost-effectiveness ratio for zonisamide of £761/QALY.

We conclude that zonisamide is a cost-effective treatment for adult patients with refractory partial epilepsy being treated according to Scottish treatment patterns and costs, assuming a willingness to pay of £20,000/QALY.

Keywords: :

• cost effectiveness
• epilepsy
• quality-adjusted life-years
• partial seizures
• utilities

Introduction

Epilepsy is a common, chronic neurological disorder requiring significant utilisation of medical resources and causing substantial economic burden to individuals and society. In an annual review by Epilepsy Scotland1 in 2004, the prevalence of epilepsy was estimated to be 38,000, with 13 new cases of epilepsy each day. Of these it is estimated that 62% have partial epilepsy2, of which 32% are uncontrolled on monotherapy3. The social and health consequences associated with epilepsy are often magnified among patients with uncontrolled epilepsy. Epilepsy affects patients’ employment, education and personal relationships, and many feel socially excluded and stigmatised.

Over the past several years there has been an effort to research and quantify the cost of epilepsy and the economic benefits of treatment, including evaluations of new antiepileptic drugs (AEDs) on direct healthcare costs4–9.

To explore further the potential impact of new AEDs on costs, benefits and resource utilisation, we adapted a previously developed economic model to evaluate the cost and benefit of care associated with two newer AEDs, zonisamide and levetiracetam, for the treatment of patients with partial epilepsy that is refractory to treatment. This study used published and unpublished reports of treatment of epilepsy as well as expert opinion to gather information on effectiveness, benefits and resource utilisation.

Methods

To assess the comparative clinical and economic impact of zonisamide on refractory epilepsy, a cost-utility analysis was conducted using a Markov modelling approach.

Most of the data were obtained from published studies and standard national cost references; resource use data specific to Scotland were unavailable from the published literature and were obtained by surveying clinicians. These data were used to develop a clinical decision-analytic model describing the experience of patients with refractory epilepsy following initiation of adjunctive AED therapy.

Model design

Owing to the chronic nature of epilepsy and the recommendations of clinical experts, the model is based on 15 years of treatment with AEDs in the base-case analysis to capture long-term costs and benefits related to the alternative treatment options. A Markov-type model was developed, building on earlier work10, which compares the expected costs and utilities associated with treatment alternatives for patients with epilepsy refractory to treatment. Markov models are well suited to model the progression of chronic disease because the disease is divided into distinct health states, and transition probabilities are assigned for movement between these health states over a discrete time period known as a cycle10. The costs and outcomes associated with the disease over the long-term can be estimated by adding resource use and health outcomes to the health states and running the model over a number of cycles.

The model simulates 3-month cycles over a 15-year period (Figure 1). The simulated patients are designed to reflect those evaluated in the clinical trials. Although age inclusion criteria differed between studies, the average for each was approximately 35 years. The population tested in clinical trials of zonisamide were younger and had more females than the levetiracetam trials. All patients were refractory to monotherapy and required adjunctive treatment. An important concept for the model is response to treatment, which is defined as a reduction in seizure frequency of at least 50% compared with baseline. This was chosen for the following reasons: the refractory nature of this patient population; the low incidence of seizure freedom observed with adjunctive therapy; and the fact that this parameter is reported as an endpoint (often the primary endpoint) in all of the included trials (reflecting regulatory requirements for pivotal trials in epilepsy).

Figure 1. Markov model of treatment of adult partial seizure epilepsy with antiepileptic drugs, simulating 3-month cycles over a 15-year period. Response to treatment is defined as a reduction in seizure frequency of at least 50% compared with baseline.

The patient cohort enters the model on a baseline AED and starts therapy with an adjunctive AED. The initial treatment period (first 3 months) is separated from the rest of the model because resource use is greater during the treatment initiation period owing to close monitoring of AED dose titration and because the management of patients who do not respond to treatment differs from that in the maintenance period. Patients who respond to treatment are classified in the ‘response’ health state; they continue taking treatment and enter the maintenance period. In the initial period, all patients who do not have a response are switched to a new AED, as are patients who experience a treatment-limiting adverse event (AE). Those patients who switch treatment are classified in the ‘treatment-limiting event’ health state, a tunnel state that leads to a new treatment initiation period with a new adjunctive AED.

The maintenance period of the model encompasses the successive 3-month cycles after the initial period on an AED. In the maintenance period, patients are again classified into health states according to their response to AED therapy. There are four health states: ‘response-stay on treatment’; ‘no response-stay on treatment’; ‘treatment-limiting event’; and ‘death’. Patients who no longer experience a 50% reduction in seizure frequency with treatment may or may not switch to a new AED. This reflects clinical reality: the refractory nature of epilepsy and the fact that a seizure reduction of less than 50% still has tangible clinical benefit. Accordingly, patients in the ‘no response-stay on treatment’ are those who experience less than a 50% reduction in seizure frequency but continue treatment. The ‘treatment-limiting event’ health state represents those patients who do not experience a response or who experience a treatment-limiting AE and switch to a new AED. Again, the ‘treatment-limiting event’ state is a tunnel state leading to the start of a new AED.

The first treatment switch to a new adjunctive AED in the model is to a specific drug. If patients need to switch from the second adjunctive AED, they are assumed to receive some unspecified form of treatment (referred to as third-line treatment) and acquire treatment costs and outcomes reflecting average values in the epilepsy population until either death or the end of the simulation period. Both future costs and future benefits have been discounted at 3.5% over the duration of the model, in accordance with requirements of the Scottish Medicines Consortium (SMC).

Choice of comparators

This economic analysis compares zonisamide, which has recently been introduced to the Scottish market as adjunctive therapy, with levetiracetam, the leading AED for adjunctive therapy in Scotland for adults with partial seizures. The choice of comparative therapies and dosages was based on use in clinical practice in Scotland and was validated by two key opinion leaders in the treatment of epilepsy.

A dose range of 300–500 mg/day of zonisamide has been studied for adjunctive treatment of partial seizures. A daily dose of 300 mg has been demonstrated in a pivotal trial to achieve a good balance between efficacy and tolerability11. Sales data from the USA and preliminary naturalistic audit data from Europe for zonisamide suggest that 300 mg is the average daily dose within this indication12. Based on this information, we expect 300 mg will be the average daily dose in Europe and is thus the daily dose assumed for zonisamide in this analysis.

The data for levetiracetam were pooled from three published clinical trials reporting its use across a range of doses (from 1,000 mg to 4,000 mg/day; mean dose 2,000 mg) as an adjunctive therapy for partial seizures13–15. Data were pooled using weighted averages based on the sample size in each study. The average patient in this group was actually taking 1,907 mg of levetiracetam. To aid in the interpretation of results, a common second-choice treatment was used in all comparisons. Lamotrigine is the leading new-generation AED in Scotland, but the majority of its usage is in monotherapy and in seizure types other than partial seizures, so it was considered suitable for a switch rather than the main comparator. Within the model, a daily dose of 300 mg of lamotrigine was used16.

Data sources

Efficacy

The distribution of patients within the health states during the initial period, as well as occurrence of treatment-limiting AEs and side effects, are based on rates reported in the selected clinical trials as shown in Table 1. The efficacy data were sourced from a literature search of medical databases (Medline, EMBASE) of clinical randomised controlled trials (RCTs) of zonisamide, levetiracetam and lamotrigine. The search was restricted through manual searching to those trials conducted in patients with partial (localisation-related) epilepsy refractory to two or three other AEDs, which utilised placebo as a comparator and which employed the doses selected for the modelled comparisons. Accordingly, all trials were in the adjunctive treatment setting.

Table 1. Clinical data sources and model variables

AED	Source	Variable
		Effectiveness during initial period	%	Side effects	%
Lamotrigine 300 mg	Matsuo et al^16	Response *	18	Weight loss	0
		No response	82	Concentration problems	0
		Switch due to AEs	4	Rash	17
				Vomiting/nausea	18
Levetiracetam 2,000 mg	Pooled from Shorvon et al^15, Cereghino et al^14 and Betts et al^13	Response	33	Weight loss	0
		No response	67	Concentration problems	0
		Switch due to AEs	11	Rash	0
				Vomiting/nausea	8
Zonisamide 300 mg	Brodie et al^11	Response	42	Weight loss	4
		No response	58	Concentration problems	0
		Switch due to AEs	18	Rash	3
				Vomiting/nausea	11
AED, antiepileptic drug; AEs, adverse events.
* Response to treatment is defined as a reduction in seizure frequency of at least 50% compared with baseline.

The search criteria produced trials that had all been submitted successfully for licensing purposes. Additionally, they conformed to regulatory guidelines for approval of such drugs in an adjunctive setting. In line with this, the inclusion criteria specifying baseline seizure frequency were similar at around four seizures per month, and the endpoints all specified 50% response rate, often as a primary variable. Although there were some differences between the included trials, they were relatively homogeneous in terms of their design, recruited population, concomitant medication, endpoints and length. Given the lack of comparative data, we considered this the best available approach. Where more than one trial was available, a pooled analysis using weighted averages was performed.

A relatively high proportion of patients switch AEDs after the initial treatment period owing either to lack of effectiveness or to side effects. Trials usually report only the overall number of treatment-limiting side effects, so it was assumed that the appearance of side effects was independent of response category.

Transition probabilities

Since long-term seizure control data are not available for each of these drugs as an adjunctive AED, the maintenance period is based on long-term seizure control from an observational study that followed patients from the time they started on one of a number of newer AEDs for up to 5 years17. These same data are applied, in the absence of treatment-specific information, to all drugs in the maintenance period. The resulting transition probabilities (Tables 2 and 3) are differentiated by whether or not the patients have switched therapy. From the transition probabilities in Table 2, there is an absorbing state ‘death’ and a tunnel state ‘treatment-limiting event’. Using a tunnel state allows those experiencing a treatment-limiting event to restart the initial period on another AED. After a switch, patients enter their second maintenance period using the transition probabilities from Table 3. Patients having a second treatment-limiting event are switched to the third-line treatment and remain on this unspecified treatment until death or the end of the simulation. The transition probabilities used in the maintenance period are not treatment-specific; AED treatments are differentiated only by their initial effectiveness and side-effect profiles.

Table 2. Three-month transition probabilities in maintenance period of epilepsy before switch

Transition from:	Transition to:
	Response-stay on Tx *	No response-stay on Tx	Tx-limiting event	Death
Response-stay on Tx *	0.95125	7.16512E–17	0.04090	0.00786
No response-stay on Tx	0.05070	0.83268	0.10870	0.00792
Tx-limiting event	0	0	1	0
Death	0	0	0	1
Tx, treatment.
* Response to treatment is defined as a reduction in seizure frequency of at least 50% compared with baseline.

Table 3. Three-month transition probabilities in maintenance period of epilepsy after switch

Transition from:	Transition to:
	Response-stay on Tx *	No response-stay on Tx	Second Tx-limiting event	Death
Response-stay on Tx *	0.95125	7.1651E–17	0.04090	0.00786
No response-stay on Tx	0.05070	0.83268	0.10870	0.00792
Second Tx-limiting event	0	0	0.99206	0.00794
Death	0	0	0	1
Tx, adjunctive antiepileptic drug treatment.
* Response to treatment is defined as a reduction in seizure frequency of at least 50% compared with baseline.

Seizure frequency

To assign resource use and death rates, it was necessary to understand the absolute frequency of seizures in patients. Using the number of seizures experienced by patients in a clinical trial for adjunctive therapy18, and assuming that treatment response is independent of baseline seizure frequency, it was assumed that 55% of responding patients have <1 seizure/month. Guberman et al18 reported that 16% of non-responding patients have <1 seizure/month.

Mortality rates

Mortality rates were estimated by combining estimates of death rates in the general population (11.6 per 1,000)19, the odds ratio of mortality associated with epilepsy (2.04)20 and the relative risk of sudden death from epilepsy (SUDEP) by seizure frequency (Table 4)21. The relative risk of overall mortality by seizure frequency was not available. Therefore, seizure frequency-specific mortality rates were calculated for SUDEP only, using the relative risk reported by Nilsson et al20. Mortality rates for seizure frequency categories were transformed to mortality rates for model health states using the same method described in the paragraph above, that is 55% of the group experiencing more than 50% reduction in seizure frequency and 16% of the non-responding group assumed to have the same mortality rate.

Table 4. Utilities and mortality in adult epilepsy model

	Response	No response	Unspecified form of treatment	Source
Annual death rate per 1,000	31.43	31.67	31.15	General Register Office for Scotland^19; Mohanraj et al^20; Nilsson et al^21
Utilities-initial period	0.9044	0.8288	N/A	Selai et al^17
Utilities-maintenance period	0.8942	0.8633	0.7961	Selai et al^17
N/A, ??

Utilities

To our knowledge, there has been no utility study conducted in epilepsy patients in Scotland. Therefore, health state utilities were estimated from a long-term study in which EQ-5D scores were collected from epilepsy patients in the UK who were not adequately controlled on one or more AEDs17. This was a prospective observational study at a single study centre among patients with refractory epilepsy (N = 125) who entered the study at the point when they were starting a new adjunctive AED. The choice of new AED was left to the treating physician. Utilities were analysed according clinical response. In the model, utilities for the initial period are represented by the utility scores measured in the UK study 6 months after the start of a newer adjunctive AED. Five-year utility scores from the study represent utilities during the maintenance phase of the model. Utility values are assigned in the model according to whether or not the patient has a response (50% reduction in seizure frequency). AEs impact utilities in the initial period of the model if they are treatment-limiting, in which case the ‘no response’ utility value from the UK study is assigned.

Utilities were also evaluated in the UK study for patients who discontinued their study drug for any reason (e.g. received surgery, switched to a new drug, etc.). The mean utility score of discontinuing patients is used to represent the utility associated with the third line in the model after failure of the second adjunctive AED (Table 4).

Resource use

Resource use and unit costs were expressed separately; routine resource utilisation and resource utilisation related to AEs were estimated separately. Routine resource use includes the estimated number of annual general practitioner (GP) and specialist visits and the number of routine electroencephalogram (EEG) tests performed in clinical practice in Scotland (Table 5). As published data are not available, these estimates come from four Scottish clinical opinion leaders. Each physician was asked to estimate the number of neurologist visits, GP visits and EEGs they would expect for each patient, based on seizure frequency. According to clinical opinion leaders surveyed, it was estimated that patients who experience seizures had 3 specialist visits per annum and 2.5–5 GP visits depending on seizure frequency. All patients were assumed to have one additional GP visit during the initial period to monitor dose titration. All four clinical opinion leaders indicated that EEGs are not performed routinely, so it was assumed that patients experiencing ≥1 seizure/month have an EEG every 4 years (or 0.25 EEGs/year); patients with <1 seizure/month were assumed to have no EEGs.

Table 5. Estimated annual direct medical resource use * for model health states

Seizure frequency	Neurologist visits per year	GP visits per year	EEGs per year
<1/month	3	2.5	0
≥1/month	3	5	0.25
Third-line treatment	3	5	0.25
GP, general practitioner; EEG, electroencephalogram.
* Source: Information on clinical practice from clinical opinion leaders in Scotland.

Additionally, resource use was estimated separately for those patients who switch from the second adjunctive AED treatment to third-line treatment. Patients receiving third-line treatment were assumed to have the same resource use and mortality risk as those experiencing ≥1 seizure/month. This assumption is based on the idea that patients who have not achieved a ≥50% reduction in seizures with two different adjunctive AED treatments are likely to have more severe refractory epilepsy.

Most side effects during the initial period are related to dose titration and are dealt with by dose adjustments. Therefore, it is assumed that they are managed during the initial period GP visit, and are not modelled explicitly.

Four side effects are modelled in the maintenance period: weight loss; problems in concentration; nausea/vomiting; and rashes. These side effects were chosen because an overview of the literature demonstrated that either they were important to patients or had the potential to become severe, even life-threatening.

If a patient switched an AED owing to lack of response, they were assumed to continue taking both the first-choice and the second-choice AED for three months while the ineffective drug was tapered down slowly and the second choice drug was slowly uptitrated to reach its effect.

Costs

Costs were considered from the perspective of the Scottish national healthcare system and were assembled from various sources. All costs were expressed in 2004 values. The analysis includes only healthcare costs; indirect costs due to potential time lost from work are excluded. Unit costs of healthcare services and AEDs used in the base-case analysis of the model are summarised in Table 6.

Table 6. Unit costs for adult epilepsy treatment

Resource item	Unit	Unit cost (2004)	Source
Full blood count	Test	£13.31	MEDTAP Unit Cost Database 2004. United BioSource Corporation
EEG	Test	£191.00	University College London Hospitals NHS Trust 2004–2005
Biochemistry	Test	£28.41	MEDTAP Unit Cost Database 2004
GP visit	Visit	£20.00	Personal Social Services Research Unit 2003
Outpatient visit	Visit	£150.00	The Chartered Institute of Public Finance and Accountancy 2003 (CIPFA/HFM)
ICU	Day	£1,091.10	CIPFA/HFM
Hospital stay	One night (neurology ward/day)	£328.00	CIPFA/HFM
AED	Daily dose	Daily costs	Source
Lamotrigine	300 mg	£3.08	British National Formulary 2004 (BNF)
Levetiracetam	2,000 mg	£3.37	BNF
Zonisamide	300mg	£3.36	Eisai Ltd. 2004
EEG, electroencephalogram; GP, general practitioner; ICU, Intensive Care Unit, AED, antiepileptic drug.

Annual direct medical costs (except the cost of AEDs) were obtained from various UK sources and inflated to 2004 prices using Health and Community Health Services inflation indices21. Annual direct medical costs were estimated for patients categorised by disease severity: patients experiencing <1 seizure/month; and those experiencing ≥1 seizure/month. Estimated annual direct routine medical costs for each health state are shown in Table 7. The health states incorporate weighted averages of medical resource use from Table 5.

Table 7. Estimated annual resource use and costs by model health states (2004)

Health state	Seizure frequency		Annual routine costs without AEDs
	<1/month	≥1/month
Response	55%	45%	£543.99
No response	16%	84%	£582.11
Treatment-limiting event	0%	100%	£597.75
AED, antiepileptic drug.

Costs related to side effects were estimated based on clinical practice patterns in Scotland. Each of the four modelled side effects was associated with blood biochemistry and full blood count tests. Rash is most often limited to a simple morbilliform rash with no evidence of systemic involvement, however with lamotrigine the incidence of Stevens–Johnson syndrome (SJS) in clinical trials is approximately 3 in 1,000 adults23. Lamotrigine-related SJS is assumed to be treated in an Intensive Care Unit for 7 days plus an additional 14-day stay in the hospital.

Analysis

The model enables comparison of zonisamide adjunctive treatment with levetiracetam adjunctive treatment, with both having a common second-choice adjunctive treatment. The expected costs and utilities associated with the treatment strategies were calculated and compared. Incremental cost-effectiveness ratios (ICERs) were then calculated. If one treatment strategy is more costly and generates fewer quality-adjusted life-years (QALYs) than another strategy, it is said to be dominated by the other strategy and cannot be economically preferred.

The values of model parameters such as the effectiveness of AEDs, resource use, health outcome and mortality consequences of AED use are subject to uncertainty. Sensitivity analyses report a range of possible values for these parameters and report the effect of altering a parameter value individually (univariate sensitivity analysis) or in tandem (two-way sensitivity analysis). Sensitivity analyses were conducted for the zonisamide–levetiracetam comparison using the following variables:

• • discounting both costs and benefits at 0% and 6%, and costs at 6% with benefits at 0%;

• • 5- and 10-year time horizons;

• • varying the proportion of responding patients who have <1 seizure/month from 55%, to 45% and 65%;

• • varying the proportion of non-responding patients who have <1 seizure/month from 16%, to 6% and 26%;

• • varying the cost of switching after second-line treatment;

• • varying the annual costs of zonisamide, levetiracetam and lamotrigine;

• • varying the proportion of zonisamide responders from 42%, to 32% and 52%;

• • varying the proportion of levetiracetam responders from 33%, to 23% and 43%; and

• • varying utilities gained for responders, non-responders and those starting a new treatment.

Results

The treatment scenario of zonisamide followed by lamotrigine is more costly but also more beneficial compared with the scenario of levetiracetam followed by lamotrigine. In terms of costs and outcomes per patient, zonisamide costs £20 more with 0.026 more QALYs. The ICER of zonisamide–lamotrigine compared with levetiracetam–lamotrigine is £761/QALY (Table 8). This ICER is lower than £20,000/QALY and would generally be regarded as cost effective in a UK context24. Therefore, zonisamide first-line adjunctive treatment followed by lamotrigine as second-line adjunctive treatment appears to be a cost-effective option compared with levetiracetam followed by lamotrigine for patients with partial seizures.

Table 8. Total costs and life-years per patient

Scenario	Total cost/patient *	QALYs/patient
ZNS–LTG	£15,630	7.923
LEV–LTG	£15,610	7.897
Difference	£20	0.026
ICER	£761 per QALY
QALY, quality-adjusted life-year; ZNS, zonisamide; LTG, lamotrigine; LEV, levetiracetam; ICER, incremental cost-effectiveness ratio.
* Over the 15-year model.

Results of the sensitivity analyses are compiled in Table 9. Recently, generic lamotrigine has become available, lowering the annual cost of this drug to £895.42. Sensitivity analysis showed that by lowering the cost of lamotrigine to the current generic price the ICER increased to £1,205/QALY. Changes in the discount rate, time horizon, utilities and the proportion of responder and non-responder patients with <1 seizure/month did not affect our conclusions: the ICER remained well below the £20,000 mark.

Table 9. Sensitivity analysis for ZNS–LTG vs. LEV–LTG comparison (numbers shown may not ‘calculate’ owing to rounding differences)

	ZNS (300 mg)–LTG (300 mg)		LEV (2,000 mg)–LTG (300 mg)		ICER (per QALY)
	Total costs	Total QALYs	Total costs	Total QALYs
Base case	£15,629,617	7,923	£15,610,289	7,897	£761
Sensitivity analysis on the discount rates
No discounting costs	£19,486,982	7,923	£19,459,369	7,897	£1,087
No discounting benefits	£15,629,617	9,879	£15,610,289	9,849	£633
6% discounting costs	£13,579,471	7,923	£13,564,750	7,897	£579
6% discounting benefits	£15,629,617	6,883	£15,610,289	6,860	£854
Sensitivity analysis on time horizon
At 5 years	£7,093,761	3,587	£7,091,667	3,573	£146
At 10 years	£12,078,029	6,118	£12,063,797	6,096	£667
Sensitivity analysis on % of responding patients with <1 seizure/month
45%	£15,646,416	7,920	£15,625,590	7,895	£826
65%	£15,612,817	7,925	£15,594,988	7,900	£696
Sensitivity analysis on % of non-responding patients with <1 seizure/month
6%	£15,629,325	7,922	£15,609,972	7,897	£761
26%	£15,629,909	7,923	£15,610,607	7,897	£760
Cost of switching after second-line treatment
£170.00	£19,779,845	7,923	£19,874,364	7,897	Dominates comparator
£1,195.50	£16,135,114	7,923	£16,070,532	7,897	£2,542
ZNS annual costs
£613.20	£14,428,871	7,923	£15,610,289	7,897	Dominates comparator
£1,839.60	£16,830,364	7,923	£15,610,289	7,897	£48,021
LEV annual cost
£615.03	£14,206,227	7,923	£13,054,934	7,897	£45,314
£1,845.08	£17,053,007	7,923	£18,165,644	7,897	Dominates comparator
LTG annual costs
£895.42	£15,153,974	7,923	£15,123,360	7,897	£1,205
£1,218.00	£16,184,345	7,923	£16,178,180	7,897	£243
ZNS responders
32%	£15,568,449	7,880	£15,610,289	7,897	Less effective at lower cost
52%	£15,689,055	7,966	£15,610,289	7,897	£1,151
LEV responders
23%	£15,629,617	7,923	£15,544,727	7,850	£1,175
43%	£15,629,617	7,923	£15,674,872	7,944	Less effective at lower cost
Utilities: Response-stay on treatment
0.8500	£15,629,617	7,840	£15,610,289	7,822	£1,075
0.9500	£15,629,617	8,027	£15,610,289	7,993	£555
Utilities: No response-stay on treatment
0.8000	£15,629,617	7,923	£15,610,289	7,897	£761
0.9000	£15,629,617	7,923	£15,610,289	7,897	£761
Utilities: Treatment-limiting event
0.7500	£15,629,617	7,602	£15,610,289	7,568	£565
0.8500	£15,629,617	8,297	£15,610,289	8,282	£1,277
ZNS, zonisamide; LTG, lamotrigine; LEV, levetiracetam; QALY, quality-adjusted life-year; ICER, incremental cost-effectiveness ratio.

The most significant changes to the conclusions occurred when the cost of switching after second-line treatment, AED costs or response rates were varied. Zonisamide dominates levetiracetam in three scenarios: when the cost of zonisamide is decreased by 50% from base case; the cost of levetiracetam is increased by 50% from base case; or the switching cost after second-line treatment is doubled. The ICER increased beyond the £20,000 threshold when the cost of zonisamide increased by 50% or the cost of levetiracetam decreased by 50%, to £48,021/QALY and £45,314/QALY respectively. Variations in response rates made zonisamide less effective but less costly when the response rate of zonisamide was decreased or the response rate of levetiracetam was increased.

Two-way sensitivity analysis was also performed to determine the effect of varying two parameters simultaneously. The results presented in Table 10 confirm the results found using the one-way sensitivity analysis.

Table 10. Two-way sensitivity analysis measuring incremental cost-effectiveness ratio for ZNS–LTG vs. LEV–LTG

Discounting costs	Time horizon
	5 years	10 years	15 years (base case)
0%	£310	£932	£1,087
3.5% (base case)	£146	£667	£761
6.0%	£43	£511	£579
% having <1 seizure/month if response	Cost of switching
	£1,195.50	£597.75 (base case)	£170.00
45%	£3,021	£826	Dominates comparator
55% (base case)	£2,542	£761	Dominates comparator
65%	£2,070	£696	Dominates comparator
% having <1 seizure/month without response	Cost of switching
	£1,195.50	£597.75 (base case)	£170.00
6%	£2,542	£761	Dominates comparator
16% (base case)	£2,542	£761	Dominates comparator
26%	£2,542	£760	Dominates comparator
Discounting costs	Cost of switching
	£1,195.50	£597.75 (base case)	£170.00
0%	£3,148	£1,087	Dominates comparator
3.5% (base case)	£2,542	£761	Dominates comparator
6%	£2,203	£579	Dominates comparator
Time horizon	Responder rate
	32%	42% (base case)	52%
5 years	Less effective at lower cost	£146	£988
10 years	Less effective at lower cost	£667	£1,164
15 years (base case)	Less effective at lower cost	£761	£1,151
ZNS, zonisamide; LTG, lamotrigine; LEV, levetiracetam.

Discussion

The results of cost-effectiveness analyses aid policy-makers in determining the value of new treatments compared with their next best alternatives. A policy-maker's willingness to pay acts as a threshold to determine whether the additional benefit provided by a treatment is worth the additional cost. In the absence of head-to-head economic studies of newer AEDs, economic modelling combining data from different sources is an adequate alternative to estimate cost effectiveness so that costs and benefits may be studied and prioritised within a healthcare system. The model described in this study was adapted from a previous study25. It uses data from well designed RCTs of similar design and recruited population in order to evaluate efficacy. It uses current treatment patterns and costs from Scotland to establish cost effectiveness using the guidelines from the SMC. Results from the model demonstrate that zonisamide is a cost-effective add-on treatment.

Results of this analysis may not be generalisable to countries other than Scotland. This is particularly the case for policy-makers whose resource use and costs are outside the sensitivity analyses performed. Policy-makers may also provide for populations that differ significantly from those used in the clinical trials, in which case the results may not be applicable. However, other countries, particularly those in the UK, may find this analysis reflective of their own situations.

Assumptions must be made to combine data or to fill in gaps where no published information is available. The multiple assumptions used in this analysis are a limitation of the results. Each assumption allows for potential bias and increases the need for explicit reporting. To increase the robustness of the model, two clinicians validated the model structure and our assumptions surrounding the model. In addition, we performed extensive one- and two-way sensitivity analyses around the key assumptions identified. Because epilepsy is a chronic condition for patients, it is important to evaluate treatment strategies over the long-term rather than studying outcomes over a short 3–6-month timeframe. Modelling cost and benefit data over 15 years offers regulatory agencies, payers, clinicians and patients a more realistic view of the costs and benefits involved with different treatment strategies.

There are few other published studies of economic modelling of the impact of newer AEDs on the medical care costs for epilepsy. A systematic review26 by Wilby et al reviewed the cost effectiveness of gabapentin, levetiracetam, tiagabin, oxcarbazepine, topiramate and vigabatrin. At the time of the review, zonisamide was not marketed in the UK and so was not included. They found that the drugs considered were more effective and more costly than existing treatment alone, concluding that at a willingness to pay of £20,000/QALY the newer drugs should only be used in cases where patients had not benefited from treatment with the older AEDs, or where the older drugs were considered unsuitable. They further reasoned that adjunctive therapy should be used where trials of monotherapy had not resulted in seizure freedom. Our current cost-utility analysis was applied within the conditions recommended by Wilby et al. Zonisamide was compared with levetiracetam in an adjunctive setting, and the most frequent concomitant AED in the studies utilised was one of the older drugs, carbamazepine. Within this context, the costs and benefits observed between the two products were very similar.

Another study published recently examined the cost of levetiracetam as add-on therapy for patients with refractory epilepsy using a 1-year dose escalation model to compare levetiracetam plus standard therapy with standard therapy alone4. The authors developed the model based on Canadian guidelines for pharmacoeconomic evaluations. Results of their study showed that patients in the levetiracetam plus standard therapy group had an average of 279 seizure-free days/year at an average cost of CAN$3,925 compared with 260 seizure-free days/year and CAN$2,403, respectively, for standard therapy. The ICER was CAN$80.70 per seizure-free day gained per patient per year.

Our study has a number of substantial differences from the one by Blais et al4. They did not include utilities, stating that scales used to develop utilities are not suitable for disease areas such as epilepsy. We believe gathering utility data and developing QALYs is essential to help make decisions about paying for health care. By using generic measures, cost-effectiveness evaluations can be compared across diseases so that regulatory agencies and payers have the best available information at hand with which to make reimbursement decisions. Blais et al also developed their model for a 1-year time period4, whilst we evaluated treatment strategies over a 15-year time horizon. Chronic diseases, by their nature, need to have longer-term costs and benefits evaluated. Finally, Blais et al used 1.5 months as the length of their treatment cycles, including the initial treatment period4. We used a slightly longer treatment cycle, giving enough time to titrate the AED and to determine the balance between treatment effectiveness and side effects.

Despite the differences in study methodology, these two evaluations indicate that individuals with partial epilepsy who are refractory to treatment incur great costs. Our results indicate an average annual cost per patient of approximately £1,042, but with improvement in patient outcomes. In an environment of constrained healthcare budgets with an emphasis on the efficient use of resources, new interventions must demonstrate cost effectiveness. The estimated ICER of £761/QALY for zonisamide compared with levetiracetam is well within the willingness-to-pay threshold of £20,000/QALY assumed in the UK. The difference in costs and benefits that make up this ICER are small. However, this result was robust to sensitivity analysis; the only analysis to raise the ICER above £20,000 was a substantial increase in the average annual cost of zonisamide per patient. In every-day clinical practice, however, it is anticipated that most patients would receive a daily dose of 300 mg, the base case in the economic evaluation, at which zonisamide is very cost effective.

Recognising the limitations of these analyses, our findings are evidence of the promising role of zonisamide as a therapeutic option for refractory epilepsy. These results remain to be confirmed in a long-term comparative trial incorporating resource use and utility data.

Acknowledgements

The authors would like to thank Edit Remak and Ruth Brown for their review of earlier versions of the manuscript. We would also like to thank Jodi Shorr for her assistance in preparation of this manuscript.

Funding: Funding for this study was provided by Eisai Ltd.

Conflicts of interest: Dr Yeates is an employee of Eisai Ltd. Mr Spackman, Ms Rentz and Mr Hutton are employees of UBC and received research study support from Eisai Ltd.