Cost-effectiveness of First-line Antiepileptic Drug Treatments in the Developing World: A Population-level Analysis

Authors

  • Dan Chisholm,

    1. Department of Health System Financing, and Department of Mental Health and Substance Abuse, World Health Organization, Geneva, Switzerland
    Search for more papers by this author
  • WHO-CHOICE

    1. Department of Health System Financing, and Department of Mental Health and Substance Abuse, World Health Organization, Geneva, Switzerland
    Search for more papers by this author

Address correspondence and reprint requests to Dr. D. Chisholm at Costs, Effectiveness and Priority-setting (CEP), Department of Health System Financing, Expenditure and Resource Allocation (FER), Evidence and Information for Policy (EIP), World Health Organization, 1211 Geneva, Switzerland. E-mail: ChisholmD@who.int

Abstract

Summary: Purpose: To establish the population-level costs and cost-effectiveness of first-line antiepileptic drug (AED) treatments for reducing the treatment gap in developing countries.

Methods: A population model was applied to nine World Health Organization (WHO) developing subregions to estimate the impact of four first-line AEDs in the primary care management of (ICD-10 defined) idiopathic epilepsy and epileptic syndromes: phenobarbitone (PB), phenytoin (PHT), carbamazepine (CBZ), and valproic acid (VPA). The efficacy of treatment was gauged in terms of improvements to both disability and recovery, subsequently adjusted for treatment coverage, response, and adherence. Total population-level treatment effects (measured in disability-adjusted life years or DALYs averted) and treatment costs (measured in international dollars; I$) were combined to form ratios of cost-effectiveness.

Results: Across nine developing WHO subregions, extending AED treatment coverage to 50% of primary epilepsy cases would avert between 150 and 650 DALYs per one million population (equivalent to 13–40% of the current burden), at an annual cost per capita of I$ 0.20–1.33. Older first-line AEDs (PB, PHT) were most cost-effective on account of their similar efficacy but lower acquisition cost (I$ 800–2,000 for each DALY averted).

Conclusions: A significant proportion of the current burden of epilepsy in developing countries is avertable by scaling-up the routine availability of low-cost AEDs. Critical factors in the successful implementation of such a scaled-up level of service delivery, apart from renewed political support and investment, relate to appropriate training and continuity of drug supply.

Epilepsy is a significant cause of disability and disease burden in the world, contributing 7 million disability-adjusted life years or DALYs (0.5%) to the global burden of disease in 2000 (1). One DALY can be thought of as 1 lost year of healthy life, and the burden of disease as a measure of the gap between current health status and an ideal situation in which everyone lives into old age free of disease and disability. This burden is particularly manifest in developing countries on account of the lower rate of treatment provided to those in need, referred to as the seizure “treatment gap” in epilepsy (2). Economic assessments of the national burden of epilepsy have been conducted in a number of high-income countries (3,4) and more recently in India (5), each of which have clearly shown the significant economic implications the disorder has in terms of health care service needs, premature mortality, and lost work productivity. Inclusion of other, more elusive costs such as those associated with stigma—exclusion from school activities, careers, even marriage (6)—would only increase such estimates.

More than the attributable burden of epilepsy, two critical questions for decision making and priority setting relate to the avertable burden of a particular disease arising from the use of an evidence-based set of interventions, and the relative cost of their implementation. Such an analysis can reveal the technically most efficient response to the attributable burden of a particular disease. A small number of cost-effectiveness studies have been carried out in high-income countries with such a purpose in mind (7), but, in common with other areas of economic evaluation, their potential influence on policy and practice has been diminished by methodologic inconsistency or lack of generalizability beyond the immediate study setting (8). No published economic evaluations exist of epilepsy treatment in developing countries, despite the likelihood that, because of their low price and established efficacy (9), use of older AEDs in primary health care settings can be expected to represent a very cost-effective use of scarce health care resources.

The purpose of this economic analysis is to estimate the proportion of the attributed burden of epilepsy that is averted currently or could be averted (via scaling up) by proven efficacious treatments for this common neurologic condition and to calculate the expected costs and cost-effectiveness of such first-line treatments. In so doing, it seeks to respond directly (and contribute) to the methodologic agenda laid down by the International League Against Epilepsy (ILAE) Subcommission on the Economic Burden of Epilepsy (10).

METHODS

Cost-effectiveness framework

The World Health Organization is currently engaged in a work program entitled CHOosing Interventions that are Cost-Effective (WHO-CHOICE). By using uniform methods, the project has generated cost-effectiveness data in 14 epidemiologic subregions of the world for key health interventions capable of reducing leading contributors to (and risk factors for) disease burden (http://www.who.int/evidence/cea). A standardized approach for cost-effectiveness analysis has been developed for all interventions in different settings (11), a core feature of which is that the costs and effects of both current and new strategies are compared with a starting point of providing no intervention. Use of such a starting point enhances the generalizability of findings (because “usual care” can vary substantially between geographic and socioeconomic settings) and also enables assessment of the relative efficiency of existing policies or practices. In addition to this primary comparison of treatment versus no treatment, a secondary level of analysis concerns the relative costs and effects of competing treatments (i.e., which AED is most efficient from the economic perspective?).

Setting

The 192 member states of the WHO were divided into five mortality strata on the basis of their levels of child and adult mortality (1). When these mortality strata were applied to the six regions of WHO, they gave rise to 14 epidemiologically defined subregions. Given the focus on developing regions of the world, five high-income subregions with low rates of child and adult mortality (capturing North America, Europe, and industrialized countries of Oceania) were not included in the analysis. Costs and effects of interventions for treating epilepsy were therefore modeled at the level of the total population in nine developing subregions of the world but have been derived in a way that allows contextualized analyses by country-level analysts interested in generating nationally applicable results and recommendations (8).

Disease model

To model the impact of health interventions relative to the situation of doing nothing, a state-transition model was used (12), which traces the development of a subregional population, taking into account births, deaths, and the disease in question (Fig. 1). The population is divided into three possible states (healthy or disease-susceptible, diseased, and dead), with movement between these states determined by a set of epidemiologic transition rates (incidence, remission, and cause-specific plus residual mortality); a health state valuation (akin to a disability weight) is applied to both the disease in question and to the nondiseased population. The output of the model is an estimate of the total healthy life years experienced by the population over a lifetime period (100 years). The model was run for a number of possible scenarios, including no treatment at all (natural history), current treatment coverage, as well as scaled-up coverage of current and potentially new interventions. For the treatment scenarios, a treatment-program implementation period of 10 years was used (thereafter epidemiologic rates and health-state valuations return to natural history levels), from which the number of additional DALYs averted each year in the population compared with no treatment was derived (DALYs were age-weighted and discounted at 3% per year).

Figure 1.

Population model for epilepsy.

Definition and classification of epilepsy

Epilepsy was modeled as a disabling, chronic condition with an increased risk of premature mortality. In line with the WHO's Global Burden of Disease study (13), analysis was restricted to active cases of idiopathic (as opposed to symptomatic) epilepsy and epileptic syndromes, defined as two or more epileptic seizures in the last 5 years that are unprovoked by any immediate identified cause (ICD-10 codes G40.0, G40.3-4, G40.6-8; http://www.who.int/classifications (14)). Unprovoked cases represent ∼70% of incident seizures, with the remaining 30% having a clear attributable cause (“provoked” by known cerebral dysfunction, CNS infections etc.). ICD-10–defined idiopathic epilepsy and epileptic syndromes can be manifested by seizures that have a localized onset (partial/focal, which may be simple or complex, depending on levels of impaired consciousness) or start in both hemispheres of the brain (generalized, including tonic–clonic seizures and absence seizures in children); both partial and generalized seizures were included in the WHO epidemiologic estimates.

Natural history of epilepsy

Estimation of the epidemiologic situation that would prevail without epilepsy treatment used prevalence estimates from WHO's GBD 2000 study [age-standardized rates ranged from 4 to 14 cases per 1,000 population (13)], whereas case fatality rates were calculated to be in the range of 3 to 26 cases per 1,000 population, reflecting an all-age standardized mortality ratio of 1.6 for idiopathic epilepsy (15, 16). Expected rates of spontaneous recovery—defined as being seizure free for 5 years (17, 18) —were calculated to be close to 5.0% (equivalent to an average case duration of 11.8–16.5 years), based on a review of the available literature (18–21). Finally, a baseline disability weight of 0.15 (where 0 equals no disability) was assigned to untreated epilepsy, which is the same weight used in the GBD study (22) (thereby enabling calculation of the proportion of current disease burden that can be avoided by effective interventions). Sensitivity analysis was performed on the impact of a higher disability weight for untreated epilepsy.

Intervention effectiveness

In the context of disease-control priorities for developing countries, the focus is on widely available, low-cost AEDs as the first-line intervention strategy for reducing the global burden of epilepsy. Four AEDs indicated for the treatment of both generalized and partial seizures were compared: phenobarbitone (PB), phenytoin (PHT), carbamazepine (CBZ), and valproic acid (VPA). It should be noted, however, that some of these AEDs are more often used for certain seizure types (for example, CBZ for complex partial seizures), and not at all for others (for example, PB, PHT, and CBZ are not indicated for absence seizures).

The impact of AED treatment was gauged in terms of improvements both to disability and recovery. In terms of efficacy, the average level of disability was estimated to improve by 56%, based on the relative reduction in the GBD disability weight for treated versus untreated epilepsy [(0.150–0.065)/0.150] (22), whereas studies of remission under treatment indicate an improvement of 80% over untreated natural history [an instantaneous remission rate of 9% compared to 5% (17, 18, 23)]. No differences in efficacy were modeled for older versus newer first-line AEDs at the population level (9) (thereby rendering the secondary analysis of competing treatments as a cost-minimization analysis; i.e., which AED can produce this equivalent effect at lowest cost?). Adjustments for treatment response (baseline estimate, 75%) and adherence (70%) were subsequently made to reflect expected real-world effectiveness (as opposed to efficacy) of these interventions (9, 24, 25), which was then applied at varying rates of population-level intervention coverage (see Table 1 for the derivation of population-level effectiveness of AEDs). Three population-level treatment-coverage rates were considered: 25%, a proxy estimate for the currently large treatment gap (low); 50%, a substantially scaled-up level of treatment coverage (medium); and 80% (high).

Table 1. Estimated population-level effect of anti-epileptic drug treatment on disability and remission
InterventionCoverageDisability improvement (%)Remission improvement (%)
Efficacy 1EffectivenessEfficacy 1Effectiveness
  1. Example: At a treatment coverage level of 50%, and assuming 75% treatment response and 70% adherence, AEDs are estimated to improve remission by 21% (80% improvement in efficacy * 75% treatment response * 70% adherence * 50% treatment coverage).

  2. 1 Efficacy improvements reported here already adjusted for treatment response of 75%.

Adherence rate 60%70%80% 60%70%80%
Anti-epileptic drug25%42%6%7%8%60%9%11%12%
 in primary care50%42%13%15%17%60%18%21%24%
 80%42%20%24%27%60%29%34%38%

Intervention costs

Analysis of the relative costs and effects of key epilepsy treatments was performed at the level of the primary health center, although costs incurred at higher levels of the health care system (referral to hospital outpatient care for a proportion of patients) were also included. Not only patient-level costs but also program-level costs were identified and estimated. Program costs included training of primary health care providers (2 initial days of training, 1 supervision day in each successive year) as well as the administrative support for implementation of an epilepsy treatment program in primary care (26). Patient-level resource inputs included daily medication (e.g., 60–100 mg PB), laboratory tests [EEG, plus a small proportion requiring a computed tomography (CT) scan], quarterly primary care visits, an average of two outpatient visits depending on severity, and a hospital admission of 1 week's duration for 5% of the most severe cases. Information sources included data from costing studies and also a multinational Delphi consensus study on resource utilisation for neuropsychiatric disorders in developing countries undertaken for WHO-CHOICE (5, 27). Nondrug patient-level resource quantities for each subregion are documented in Table 4.

Table 4. Costs and cost-effectiveness of first-line anti-epileptic drugs in primary care (50% treatment coverage)
First-line AEDs (50% treatment coverage)AfricaThe AmericasEastern MediterraneanSouth East AsiaWestern Pacific
WprB
Non-drug costsMean QuantityAfrDAfrEAmrBAmrDEmrBEmrDSearBSearD
  1. Notes:

  2. 1 Total patient-level costs per treated case per year (drugs, lab tests and primary/secondary health care).

  3. 2 Total patient-level costs per treated case per year (drugs, lab tests and primary/secondary health care).

  4. 3 Total costs divided by total health gain (average cost-effectiveness ratio, compared to the situation of no treatment).

  5. 4 Incremental cost of moving from 50% to 80% coverage, divided by incremental health gain (see Table 3).

Primary care visits4.010.911.313.411.013.49.711.111.011.3
Outpatient care visits2.0 8.910.041.922.124.314.522.5 9.323.9
Inpatient care (days) 0.25 4.7 5.217.610.210.2 7.110.4 4.910.9
EEG1.017.717.617.717.318.617.717.517.118.0
CT scan 0.2023.523.423.623.024.823.623.322.824.0
Total non-drug cost 65.767.5114.283.691.372.784.965.088.2
 Notes 
Patient cost per year (I$)1 
 Phenobarbitone 7172119899678907093
 Phenytoin 74761239210081947497
 Carbamazepine 105107153123130112124104127
 Valproic acid 159161207177184166178158181
Cost per capita (I$ per year)2 
 Phenobarbitone 0.460.550.820.520.310.180.280.230.29
 Phenytoin 0.480.570.840.540.320.190.290.240.30
 Carbamazepine 0.640.761.020.680.370.250.360.320.37
 Valproic acid 0.911.091.330.940.470.340.490.470.50
Cost per DALY averted (I$)3 
 Phenobarbitone 8128441,7051,2072,0631,2121,3219721,534
 Phenytoin 8458801,7501,2482,1071,2561,3641,0131,581
 Carbamazepine 1,1171,1732,1181,5812,4711,6171,7201,3471,971
 Valproic acid 1,5981,6942,7722,1723,1182,2572,3511,9412,662
Incremental CER (to 80% coverage)4 
 Phenobarbitone 7476608909932,8222,7792,0171,7612,233
 Phenytoin 7826909311,0402,9532,9082,1111,8432,336
 Carbamazepine 1,0659401,2671,4164,0223,9602,8742,5093,181
 Valproic acid 1,5671,3831,8652,0835,9185,8284,2293,6934,682

Unit costs of primary and secondary care services were derived from an econometric analysis of a multinational dataset of hospital costs, by using gross national income per capita (plus other explanatory variables) to predict unit costs in different WHO subregions (28). Supplier prices for generically produced AEDs were obtained from the International Drug Price Indicator Guide, subsequently adjusted upward to take into account the cost of shipping and distribution (http://erc.msh.org/dmpguide; see Table 2). All baseline analysis costs for the 10-year implementation period were discounted at 3% and expressed in international dollars (I$), which adjust for differences in the relative price and purchasing power of countries and thereby facilitate interregional analysis. That is, one international dollar buys the same quantity of health care resources in China or India as it does in the United States of America (see Technical Appendix for more details).

Table 2. Anti-epileptic drug costs (US $, 2000) (Source: International Drug Price Indicator Guide, 2000; http://erc.msh.org/dmpguide/)
Drug nameStrength
(mg)
Package
(tablets)
Pack price 1
($US; FOB)
CIF price2
(FOB * 1.25)
Unit cost
(per tablet)
Annual total per patient

Tablets 3
Cost
  1. 1‘Free on board' price based on median value from 6 suppliers: TRI-MED; IDA; ORBI; ECHO; JMS; MISSION.

  2. 2Total price after allowance for ‘carriage, insurance and freight' (an average 25% on-cost of FOB price).

  3. 3Mean dosage is a weighted average of low (30%) and high (70%) daily maintenance doses.

Phenobarbitone301000 $3.30 $4.13$0.0041,241 $5.12
Phenytoin1001000 $5.67 $7.09$0.0071,241 $8.80
Carbamazepine2001000$16.50$20.63$0.0211,862$38.39
Valproic acid2001500$36.20$45.25$0.0303,103$93.59

Uncertainty analysis

A series of one-way and multi-way sensitivity analyses were carried out on baseline results, including assessment of the impact of discounting and age-weighting as well as plausible changes to cost estimates (e.g., the price of AEDs and the intensity of secondary-level health care use) and effectiveness calculations (untreated disability weight, efficacy, and adherence). Baseline results (with pessimistic and optimistic scenarios as lower and upper ranges) were also entered into the MCLeague analytical software package (11) , which performs a probabilistic uncertainty analysis by using Monte Carlo simulation (2,000 runs were made, by using a truncated normal distribution).

RESULTS

Population-level effect of first-line AEDs in primary care

Table 3 provides estimates of the population-level effects of first-line AEDs at differing coverage levels in nine WHO developing regions. At the population level, no discernible difference was found between different AEDs with respect to efficacy or effectiveness. At an intervention coverage rate of 25%—taken to be a proxy for the current situation—DALYs averted per 1 million population ranged from 90 in Eastern Mediterranean to ∼350 in the Africa subregions (equivalent to 8–25% of current burden). Doubling the coverage level to 50% would avert 67% more DALYs (150–650 per 1 million population; 13–40% of current burden), whereas an 80% coverage rate would avert 21–62% of the current burden (240–1,000 DALYs per 1 million population). Variations across regions in terms of population-level effectiveness reflect differences in underlying incidence and prevalence estimates (for example, an age-standardized prevalence of >10 per thousand population in Africa and the Americas, compared with 4–7 in Asian and Eastern Mediterranean regions) (13).

Table 3. Attributable and avertable burden of epilepsy in WHO developing regions
WHO Region
WHO sub-region1
Example countries
AfricaThe AmericasEastern MediterraneanSouth East AsiaWestern Pacific
WprB
China
Vietnam
AfrD
Nigeria
Senegal
AfrE
Botswana
Kenya
AmrB
Brazil
Mexico
AmrD
Ecuador
Peru
EmrB
Iran
Jordan
EmrD
Egypt
Pakistan
SearB
Indonesia
Thailand
SearD
India
Nepal
  1. 1Sub-regional suffix letter refers to levels of child and adult mortality as follows: B (low, low); D (high, high); E (high, very high).

Total population (million)294346431711393432941,2421,533
Burden (DALYs per year, thousands)4236908481901314003711,5281,291
Intervention: first-line anti-epileptic drugs in primary care
  AEDs at 25% coverage100,612133,936124,50318,40212,49431,25736,940178,818174,108
  AEDs at 50% coverage167,687223,226207,50530,66920,82352,09561,567298,029290,180
  AEDs at 80% coverage264,156351,635326,26548,24532,90182,31097,276470,886458,485
Standardised effect (DALYs averted per 1 million popn)
  AEDs at 25% coverage3423882892589091126144114
  AEDs at 50% coverage570646482431150152210240189
  AEDs at 80% coverage8981,018757677237240331379299
Effect as% of current burden
  AEDs at 25% coverage24%19%15%10%10%8%10%12%13%
  AEDs at 50% coverage40%32%24%16%16%13%17%20%22%
  AEDs at 80% coverage62%51%38%25%25%21%26%31%36%

Treatment costs of first-line AEDs

As shown in Table 4, the average cost associated with drug supply, laboratory tests, and health care ranged from as little as I$ 70 per year in very low income subregions for PB and PHT, to I$ 160–210 per year for VPA. For estimating total costs of providing AEDs at a specified level of treatment coverage in the population (a coverage rate of 50% is used in Table 4), program-level costs of central administration and training of primary care providers also must be included (accounting for 10–15% of total costs for more expensive first-line AEDs; 20–25% of total costs for primary care management of epilepsy with PB or PHT). In all subregions (except American subregion AmrB), the total annual cost per capita for PB and PHT ranged between I$ 0.20–0.50 (equivalent to an investment of I$ 200,000–500,000 per 1 million population).

Cost-effectiveness of AEDs

Compared with the epidemiologic situation of no treatment (natural history), the most efficient strategy for reducing the burden of epilepsy in developing regions is the use of PB or PHT (I$ 800–2,000 per DALY averted; Table 4). More expensive AEDs such as CBZ and VPA produce (average) cost-effectiveness ratios in the range of I$ 1,100–3,000. The incremental cost-effectiveness ratio (ICER) of moving from a treatment coverage rate of 50% to 80%—which is calculated as the additional cost divided by the additional health gain associated with this scaled-up activity—is favorable in Africa and the Americas (<I$ 2,000 per extra DALY), and I$ 2,000–6,000 in other subregions. The regional differences in the additional cost per extra unit of health outcome are influenced by the varying cost of scaling up health program activities in the nine different subregions (higher in Eastern Mediterranean, South Asia, and Western Pacific regions).

Uncertainty analysis

Substitution of the baseline discount rate of 3% with values of 0 and 6% per year altered total costs and average CERs for all interventions by +10% and –9%, respectively. The removal of age-weighting had a more significant impact on results, reducing total health gain estimates by 13–19% across subregions (resulting in a corresponding increase of 15–23% in average CERs). A one-way sensitivity analysis that used a higher untreated disability weight of 0.3 rather than 0.15 (but the same proportionate improvement of 56% after treatment, i.e., a treated weight of 0.13 rather than 0.065) increased total population health gain by as much as 40%, indicating that results are sensitive to this parameter. Best- and worst-case scenarios were derived by according lower and upper values to the price of AEDs, the proportion of cases using primary and outpatient services, in addition to the level of treatment efficacy (remission and disability weight; ±25%) and adherence (±10%). Under the best-case scenario, total costs were 15–30% lower and total effects 43% higher than base case results, lowering the overall cost per DALY averted by approximately half. Results for the worst-case scenario were in the same range, this time with increases of close to 20–35% in costs and 36% less health gain, leading to average CERs being double their baseline value. To illustrate, the expected cost per DALY for PB treatment in Western Pacific subregion WprB (baseline value, I$ 1,534) ranged from I$ 739 to I$ 2,643. Under best- and worst-case scenarios, the rank order of cost-effectiveness was unchanged.

By entering cost and effect data ranges into a stochastic uncertainty framework, a sense of the probable attractiveness of interventions can be elicited. Figure 2 charts for South East Asia subregion SearD the expected probability of each intervention being considered cost-effective at different levels of resource availability, after taking into account the uncertainty around baseline estimates. It shows that at very restricted levels of resource availability, only the low-cost AEDs have a high probability of being selected as a cost-effective choice, but as this resource constraint is lifted, the probability of choosing one AED over another becomes indistinguishable (i.e., any of the AEDs could be included in a cost-effective package of care).

Figure 2.

Probability of AEDs being cost-effective at different budget levels (WHO South East Asian sub-region SearD).

Discussion

In the context of the low priority and consequently high treatment gap for epilepsy care in developing regions of the world, this economic analysis set out to establish the expected costs and cost-effectiveness of first-line AEDs in primary care. The principal conclusion of the study—and one that reinforces previous international policy analyses (1, 29)—is that a significant proportion of the current burden of epilepsy in developing countries is avertable by scaling-up the routine availability of low-cost AEDs. Purely from an efficiency point of view, the most cost-effective technology for reducing this burden is first-line treatment in primary care with older AEDs such as PB and PHT, because their expected population-level impact is similar to those of other first-line AEDs (such as CBZ, VPA), yet their acquisition costs are appreciably lower. The estimated cost of each healthy year of life gained by the use of VPA may be double that of PB; however, it should be emphasized that the simple ratio of total cost to total health gain for all four AEDs considered in this analysis falls below average annual income per capita, which is a recently proposed threshold by the WHO Commission on Macroeconomics and Health (30) for considering a health intervention to be very cost-effective. Such a conclusion is further borne out by the favorable comparison of the cost per DALY averted by first-line AEDs against intervention strategies for other chronic, disabling, noncommunicable conditions such as depression, hypertension, and high cholesterol (31, 32). At the sector-wide level to which this analysis is principally directed, therefore, strong arguments exist for investing in epilepsy treatment as an efficient use of public health resources, quite apart from the additional economic benefits that such treatment would give in terms of reduced days of work lost to seizures (5).

Although the methodologic framework adopted for this analysis offers a new and potentially useful approach to the comparative assessment of intervention strategies across the health sector, it necessarily suffers from a number of constraints common to all disease modeling exercises. At the clinical level, analysis has been applied to the broad set of ICD-10 idiopathic epilepsy and epileptic syndromes without consideration of the preferred drug of choice for particular epilepsy types or syndromes. Because certain first-line AEDs are not indicated for specific seizure types (most obviously, PB, PHT, and CBZ for absence seizures), this points to an overestimate of expected population effect for broad-spectrum use of these individual drugs. The size of this overestimate, however, is likely to be modest (and within the uncertainty limits described earlier) because of the quite small proportion of all active ICD-defined idiopathic epilepsy and epileptic syndromes that fall into these specific seizure types (e.g., absence and myoclonic seizures each typically account for ≤5% of seizures) (33). Similarly, whereas monotherapy currently stands as the preferred pharmacologic strategy (9), the analysis does not take into account the significant proportion of cases managed with combination therapy, which may push up costs and/or have a differential impact on health outcomes. Comorbidity is a further (unmeasured) factor that can drive up the costs of treatment without corresponding health improvements. Also, the influence of different adverse effects on the choice of AED has not been incorporated into the analysis other than by the adoption of a constant nonadherence rate across all four drugs. Finally, cost-effectiveness estimates were restricted to drug treatments; epilepsy surgery was not included in the analysis on the basis of its limited indication (for intractable cases only), the high-technology requirements of its use, and compared with AEDs, the relatively high cost per healthy year of life reported to date in the literature [US$ 16,000–27,000 (34, 35)].

In terms of the population-level measurement of costs and effects, as well as the epidemiology of epilepsy that underlies it, a number of key assumptions have been made. In the absence of clear evidence to the contrary, for example, the efficacy of different first-line AEDs was assumed to be constant within and across regions. More fundamentally, the actual efficacy data on which population-level treatment effects were based—in particular, improvements in the rate of remission, relative to the untreated natural history of epilepsy—proved to be extremely scarce for developing regions. Accordingly, effect sizes are based on best available international evidence (mostly research-rich countries such as the United States or the United Kingdom), which may or may not accurately predict outcome in a specific developing region context. Indeed, the epidemiologically based modeling approach used in this study has highlighted the already expressed need for more robust estimates of outcome for AEDs in developing countries (29), particularly with respect to daily functioning, terminal remission, and even mortality (over and above natural history or spontaneous improvement). Concerning disability weight, GBD weights were used to allow calculation of averted burden, but evidence suggests that the untreated weight for epilepsy may be too low (36); use of a higher weight was shown to increase significantly the anticipated health benefits of treatment. Appreciable uncertainty remains concerning the expected frequency, intensity, and cost of health service use for an “average” case of idiopathic epilepsy in the different subregional populations. Sensitivity analysis of key drivers of cost and effectiveness can shed light on the expected extent of this uncertainty—showing that average cost-effectiveness ratios may plausibly be half or double their baseline value—but is no substitute for better empiric estimates from a variety of treatment settings in the first place. Such prospective evaluations are not difficult to carry out, can be conducted alongside ongoing projects (37), and would fill an important gap in health services research.

A further important analytic choice relates to the use of whole WHO subregions as the demographic units of analysis (a compromise between a single global estimate and assessment for each of WHO's 192 member states). Ultimately, the purpose of sectoral cost-effectiveness analysis is to inform national-level priority setting and resource allocation; therefore analyses should be conducted at this level. Such a process, which involves the respecification of key input parameters such as effective treatment coverage, drug prices, and unit costs of health services (expressed in local currency units), is now under way in a number of developing countries as part of the continuing “roll-out” of the WHO-CHOICE work program (8). Even then, generation of context-specific cost-effectiveness data provides no guarantee that results or recommendations will find their way into policy or practice, either because of reluctance by physicians to override clinical judgments with economic arguments when prescribing AEDs, or because of more entrenched barriers at the level of the health system (e.g., lack of a continuous drug supply, inequitable health financing arrangements) or society itself (e.g., cultural beliefs about the causes of epilepsy) (2, 38, 39). For example, significantly scaled-up availability or coverage of epilepsy treatments in the community is much more likely to take place within a health system that has national or social health insurance (particularly if such treatments are part of any essential package of health interventions), as opposed to a health system dominated by out-of-pocket expenditure, which offers a far less equitable or responsive mechanism for safeguarding at-risk populations from the adverse financial consequences of epilepsy. Accordingly, a number of significant obstacles remain to scaling up treatment coverage and consequently reducing the current treatment gap.

In conclusion, cost-effectiveness analysis represents only one criterion in the broader task of priority setting and the allocation of public health resources. Behind the analysis of the projected costs of scaling up epilepsy treatment, for example, lie the critical issues of training and supervision of primary health workers, and the efficient acquisition and distribution of a continuous supply of essential AEDs. Beyond the wider availability of effective treatments, greater attention must be given to the psychosocial needs and social integration of individuals with epilepsy, particularly children (40). Critical factors in the successful implementation of such a scaled-up level of service delivery, apart from renewed political support and investment, therefore relate to appropriate training, continuity of drug supply, and enhanced consumer or community involvement (2, 38).

Acknowledgments

Acknowledgment:  The following colleagues are warmly thanked for their active contribution to the conceptual and methodologic development of WHO-CHOICE (CHOosing Interventions that are Cost-Effective): Taghreed Adam, Rob Baltussen, David Evans, Raymond Hutubessy, Benjamin Johns, Jeremy Lauer, Christopher Murray, and Tessa Tan Torres. In addition, the comments of Dr. Leonid Prilipko and Dr. Charles Begley on an earlier version of this manuscript are gratefully acknowledged. The views expressed are those of the author and not necessarily those of the World Health Organization.

Appendices

Technical Appendix: International Dollars

Results of WHO-CHOICE's analyses are presented in international dollars for the year 2000. An international dollar has the same purchasing power as the U.S. dollar has in the United States. Costs in local currency units are converted to international dollars by using purchasing power parity (PPP) exchange rates. A PPP exchange rate is the number of units of a country's currency required to buy the same amounts of goods and services in the domestic market as the U.S. dollar would buy in the United States. An international dollar is therefore a hypothetical currency that is used as a means of translating and comparing costs from one country to the other by using a common reference point, the U.S. dollar. The PPP exchange rates used in this analysis were developed by WHO and are listed on the WHO-CHOICE website under “Prices of goods and services” (http://www.who.int/evidence/cea).

To convert international dollars to local currency units, multiply the international dollar figure by the PPP exchange rate. For example, 2 international dollars are equal to 24.102 Thai Baht for the year 2000 (2 × 12.051). To convert local currency units to international dollars, divide the local currency unit by the PPP exchange rate.

Ancillary