For the economic evaluation the raw data from a detailed, open-label, single-arm clinical trial on the efficacy and safety of LEV in Korea was used.¹² The study population consisted of 100 adult patients aged 18 years or older who received LEV as add-on therapy in addition to conventional AEDs. All the patients came from nine research centres in Korea and had uncontrolled epilepsy and/or intolerable adverse events on conventional AEDs.

Data from the 16-week clinical trial were recorded prospectively. It consisted of two periods: 1) an up-titration period (first 4 weeks) and 2) a maintenance period (for the following 12 weeks). For purposes of our analysis, data were extracted from the collected dataset for the period of 12 weeks before and 16 weeks after the first day of add-on therapy of LEV. Recorded data covered the following domains:

When estimates of input parameters were not available, endorsed comments from the Scientific Advisory Board (SAB), convened by the Korean Epilepsy Society, were considered.

Epilepsy-related QoL was assessed twice using the Korean translation of the QOLIE-31 at baseline before the beginning of add-on therapy of LEV (Week 0) and at the end of the clinical trial (Week 16), or earlier if withdrawn. The QOLIE-31 is a self-administered questionnaire designed to be completed by patients.¹³ Validation studies of the QOLIE-31 in different language versions also demonstrated good reliabilities and validities.^14-16 QOLIE-31 was derived from the longer QOLIE-89, an instrument with 17 subscales, including generic and epilepsy-specific issues, and a separate item on health status (not scored).¹⁷ The QOLIE-31 includes 7 subscales: seizure worry, overall QoL, emotional wellbeing, energy-fatigue, cognitive functioning, medication effects, social function, and health status (not scored). The subscale and the total scores are calculated according to the algorithm defined by the author with scores ranging from 0 to 100 and higher scores indicating better function.¹³

A 1-year dose-escalation decision-tree model was designed for economic evaluation of LEV as an add-on therapy and further, was adapted to Korean data.^18,19 Two hypothetical cohorts of patients were started either on combination therapy (LEV 1,000 mg/day plus ST) or on ST alone. For patients in the LEV arm the dose was increased up to 3,000 mg/day if freedom from seizures was not attained at lower doses (Figure 1). Patients who started on ST alone were kept in that arm for the whole 1-year duration of the economic model. In this model, success was defined as freedom from seizure, while failure was defined as withdrawal leading to being returned to ST in the combination therapy arm or not-seizure-free, continued refractory partial epilepsy state in the ST alone arm. All these assumptions were reviewed and endorsed by the SAB.

More specifically, patients in the combination therapy arm started with LEV 1,000 mg/day plus ST for a 0.5-month observation period, after which they entered into one of the following pathways:

Patients increased to LEV 2,000 mg/day and maintained that dosage for a 0.5-month observation period after which they entered one of the following pathways:

Patients increased to LEV 3,000 mg/day and maintain for a 0.5-month observation period after which they entered one of the following pathways:

The probabilities of each state for the combination therapy (LEV+ST) were derived from the outcomes of the 16-week duration of the aforementioned, open-label clinical trial.¹² The transition probabilities for ST were adopted from a previous report.¹⁸

The economic evaluation focused on the direct medical costs incurred by the patients from the third-party payer perspective. LEV has been available since January 2007. Costs for ST should not include cost for LEV. Therefore, unit costs were obtained from the "2006 Korean National Health Insurance Statistical Year-book" and the "2006 Major Health Insurance Statistics".^20,21 The average currency exchange rate in the year of 2006 was 955.51 Korea Won per US$ 1 (http://www.keb.co.kr/IBS/welcome.jsp). Resource use related to the routine care provided to patients with refractory partial epilepsy could not be derived from the aforementioned prospective clinical trial because information on healthcare resource utilization had not been collected.¹² Furthermore, specific data on health resource utilization of patients with refractory epilepsy were not available in Korea. Therefore, routine follow-up of patients with epilepsy in typical practices was based on national data sources (gross costing method).²⁰ It was assumed that patients with refractory epilepsy would incur higher costs as a consequence of frequent outpatient visits and hospital admissions. Success in the treatment of patients with refractory epilepsy would save greater amounts of money than success in all patients with epilepsy. It was also expected that use of the costs of all patients with epilepsy could underestimate potential economic benefits in this study. It is also a more conservative estimate than the usual cost measure. To cover medical costs for patients with refractory partial epilepsy, costs per patient for inpatient, outpatient, and pharmacy for a 1-year period were considered in this study. Costs consist of: inpatient (US$ 1,841 per person per year); outpatient (US$ 183 per person per year); and pharmacy (US$ 368 per person per year) for ST (Table 1); while for combination therapy, costs for LEV (US$ 0.97, US$ 1.45, and US$ 2.18 per tablet with LEV 250 mg, 500 mg, and 1,000 mg respectively)(http://www.kpis.or.kr/index.jsp) were added to the costs for s ST alone.

Generally speaking, reduced costs are assumed when seizures occur less frequently. The SAB endorsed an assumption of a linear relation between costs for inpatient and seizure frequency among patients with refractory partial epilepsy. For the base case scenario, in patients with seizure freedom (seizure frequency=0), the inpatient cost was considered to be zero. The better response was assumed for those patients who stayed at a certain dose of LEV per the clinician's judgement that the dose was sufficiently effective and tolerable for the patient. P^{better response}=0.5 means that those with a better response may only need half the given inpatient cost. On the other hand, worse response was assumed for those patients who returned to a lower dose of LEV per the clinician's judgement that the patient was unable to tolerate a higher dose, but experienced a significant reduction in the frequency of seizures while treated with the lower dose of LEV. P^{worse response}=0.8 means that those patients with worse response may need 80% of the inpatient cost.

Measurement of the incremental cost-effectiveness ratios (ICER), i.e. incremental cost per SFD and the QALY gained, was derived from the trial in Korea. ICER was calculated as follows^*:

ICER=(C_LEV+ST-C_ST)/(E_LEV+ST-E_ST)=ΔC/ΔE

^*C: cost, E: effectiveness, Δ: difference

The net cost is equal to the total cost of combination therapy less the total cost of ST. The cost per SFD gained is equal to the net costs for LEV treatment per person per year, divided by the net SFD gained. The cost per QALY gained in this study is equal to the net costs for LEV treatment per person per year, divided by the net QALYs gained.

In order to test the robustness of the model and to evaluate the effect of uncertainty around model parameters on the ICER, we conducted extensive sensitivity analyses on all the variables in this study. To test the cost parameter estimates, costs for inpatient, outpatient and pharmacy and the acquisition price for LEV were both increased and decreased by 50%, the range that endorsed by the SAB. For parameters associated with transitions from or to certain doses of LEV, a full range (0-1) of two probabilities of better response (originally 0.5) and worse response (originally 0.8) were considered.

Of a total of 100 Korean patients, 52 (52%) were male and 48 (48%) were female. Patients' mean age (±SD) was 35.3 (±11.7) years, while the mean age (±SD) at onset of epilepsy was 17.8 (±9.5) years. Eighty patients presented with at least one complex partial seizure, 28 patients with at least 1 secondarily generalized tonicclonic seizure, and 13 patients with at least 1 simple partial seizure during the baseline 3-month period.

In the history of previous AED treatment, 76 patients had taken two or more AEDs prior to entry to the study. The majority (80%) of patients entered the trial on two concomitant AEDs. The most frequently used concomitant AEDs were carbamazepine, valproic acid, lamotrigine, and topiramate.

The main health outcome of the analysis, i.e., SFDs, was improved by LEV with an incremental gain of 18.3 days per patient per year, thus resulting in a cost-effectiveness ratio of US$ 44 per SFD gained per patient per year. The ICER in this study looks grossly better than the Canadian report of CAN$ 80.7 per SFD gained.¹⁹ The superior ICER obtained here probably reflects different direct medical costs between Korea and Canada. Both models (Blais et al.¹⁹ and this study) share a common methodology using a decision-tree model for LEV add-on to ST for refractory epileptic patients and considering direct medical costs and SFDs gained as effectiveness. LeLorier et al.²³ reported the potential benefit of seizure freedom (one additional seizure free patient per year) could be gained with a cost of £5,301 by adding LEV to standard treatment when considering the potential benefit of epilepsy surgery in the model. For other AEDs, Markowitz et al.²² found that lamotrigine (LTG) therapy cost an additional US$ 728 compared with the patients' current monotherapy only, discounted at 3% over 10 years. LTG therapy resulted in a gain 106 additional SFDs (undiscounted) and an ICER of US$ 6.9 per SFD gained. However, Markowitz et al.²² used the 10-year long-term model, but they did not discount outcome (i.e., SFD). Therefore, direct comparison of cost per SFD with a 1-year time horizon may not be justified. Schachter et al.²⁴ estimated that adjunctive tiagabine (TGB), added to existing treatment of phenytoin, cost US$ 719 over 16 weeks including the cost of managing adverse events, compared with US$ 784 for adjunctive carbamazepine (CBZ). CBZ was more clinically efficacious (50% reduction in seizure frequency) but more detailed results were not provided. Within the baseline CBZ arm, add-on phenytoin (PHT) costs US$ 810 compared with US$ 958 for add-on TGB. Add-on PHT and add-on TGB had similar efficacy. Compared with current medication only, Messori et al.²⁵ found that adjunctive LTG cost an additional US$ 1,612,370 for a cohort of 100 patients over the patients' lifetimes, gained 39 QALYs, and calculated an ICER of US$ 41,343 per QALY. When we consider both results of this study, previous economic evaluations of other AEDs, and the Gross Domestic Product of Korea (US$ 19,722/year/person, 2006), LEV is comparatively cost-effective against older AEDs.

QALYs require a uni-dimensional measurement of QoL using both direct (i.e., standard gamble, time trade-off, person trade-off, visual analogue scale) and indirect utility assessments (i.e., EQ-5D, SF-6D, HUI). But, health-related QoL should be considered multidimensional, so it is often measured by disease-specific QoL scales. Therefore, analysis would be facilitated by converting disease-specific QoL measurement into utilities. Conversion algorithms were derived using 1 of 4 techniques: 1) transfer to utility regression, 2) response mapping, 3) effect size translation, and 4) revaluing outcome measures using preference-based scaling techniques.²⁶ The resultant health state classification system consists of several dimensions (i.e., mobility, usual activity, visual acuity, pain, cognition, depression) with several levels (i.e., level 1=no problem, level 2=some problem, level 3=extreme problems), thus a generating huge number of health states. The valuing processes used in this study may be considered relatively new. Contrary to usual methods, levels were not categorized in this study, even though we used quality weights for 7 dimensions which were provided by original authors of QOLIE.¹⁷ Instead, subscale scores ranging from 0 to 100 were used to represent multidimensional levels of the health state of a patient. In this study, it was assumed that level is a continuous (linear), not categorical (stepwise) variable. Whether linear or stepwise, a higher score or higher level reflects more severe problems. A utility score was computed by summing up all utility portions of different dimensions in this study, which was adopted from the methodology for the EQ-5D Korean version.²⁷

This study has some limitations. First, as with other cost-effectiveness studies, with regard to external validity, there is the problem of extrapolating clinical trial data (efficacy) to real world clinical practice (effectiveness). This problem was addressed through the sensitivity analyses. Second, because of the third-party payer perspective employed in this analysis, indirect costs were not considered in the model, even if they were estimated to represent a large proportion of the total costs associated with epilepsy. In order to gather such information, data derived from a long-term observation period must be used for this refractory population. Third, the Korean translation of the QOLIE-31 has not yet been validated, so all results of the cost-utility analysis should be considered in the light of data from relevant preference-based health economic studies. Fourth, we should use quality weights from other countries given that there are no prior studies that have been carried out in Korea. Special caution is needed when transferring these data to other contexts.