The SANAD II study of the effectiveness and cost-effectiveness of levetiracetam, zonisamide, or lamotrigine for newly diagnosed focal epilepsy: an open-label, non-inferiority, multicentre, phase 4, randomised controlled trial

Summary Background Levetiracetam and zonisamide are licensed as monotherapy for patients with focal epilepsy, but there is uncertainty as to whether they should be recommended as first-line treatments because of insufficient evidence of clinical effectiveness and cost-effectiveness. We aimed to assess the long-term clinical effectiveness and cost-effectiveness of levetiracetam and zonisamide compared with lamotrigine in people with newly diagnosed focal epilepsy. Methods This randomised, open-label, controlled trial compared levetiracetam and zonisamide with lamotrigine as first-line treatment for patients with newly diagnosed focal epilepsy. Adult and paediatric neurology services across the UK recruited participants aged 5 years or older (with no upper age limit) with two or more unprovoked focal seizures. Participants were randomly allocated (1:1:1) using a minimisation programme with a random element utilising factor to receive lamotrigine, levetiracetam, or zonisamide. Participants and investigators were not masked and were aware of treatment allocation. SANAD II was designed to assess non-inferiority of both levetiracetam and zonisamide to lamotrigine for the primary outcome of time to 12-month remission. Anti-seizure medications were taken orally and for participants aged 12 years or older the initial advised maintenance doses were lamotrigine 50 mg (morning) and 100 mg (evening), levetiracetam 500 mg twice per day, and zonisamide 100 mg twice per day. For children aged between 5 and 12 years the initial daily maintenance doses advised were lamotrigine 1·5 mg/kg twice per day, levetiracetam 20 mg/kg twice per day, and zonisamide 2·5 mg/kg twice per day. All participants were included in the intention-to-treat (ITT) analysis. The per-protocol (PP) analysis excluded participants with major protocol deviations and those who were subsequently diagnosed as not having epilepsy. Safety analysis included all participants who received one dose of any study drug. The non-inferiority limit was a hazard ratio (HR) of 1·329, which equates to an absolute difference of 10%. A HR greater than 1 indicated that an event was more likely on lamotrigine. The trial is registered with the ISRCTN registry, 30294119 (EudraCt number: 2012-001884-64). Findings 990 participants were recruited between May 2, 2013, and June 20, 2017, and followed up for a further 2 years. Patients were randomly assigned to receive lamotrigine (n=330), levetiracetam (n=332), or zonisamide (n=328). The ITT analysis included all participants and the PP analysis included 324 participants randomly assigned to lamotrigine, 320 participants randomly assigned to levetiracetam, and 315 participants randomly assigned to zonisamide. Levetiracetam did not meet the criteria for non-inferiority in the ITT analysis of time to 12-month remission versus lamotrigine (HR 1·18; 97·5% CI 0·95–1·47) but zonisamide did meet the criteria for non-inferiority in the ITT analysis versus lamotrigine (1·03; 0·83–1·28). The PP analysis showed that 12-month remission was superior with lamotrigine than both levetiracetam (HR 1·32 [97·5% CI 1·05 to 1·66]) and zonisamide (HR 1·37 [1·08–1·73]). There were 37 deaths during the trial. Adverse reactions were reported by 108 (33%) participants who started lamotrigine, 144 (44%) participants who started levetiracetam, and 146 (45%) participants who started zonisamide. Lamotrigine was superior in the cost-utility analysis, with a higher net health benefit of 1·403 QALYs (97·5% central range 1·319–1·458) compared with 1·222 (1·110–1·283) for levetiracetam and 1·232 (1·112, 1·307) for zonisamide at a cost-effectiveness threshold of £20 000 per QALY. Cost-effectiveness was based on differences between treatment groups in costs and QALYs. Interpretation These findings do not support the use of levetiracetam or zonisamide as first-line treatments for patients with focal epilepsy. Lamotrigine should remain a first-line treatment for patients with focal epilepsy and should be the standard treatment in future trials. Funding National Institute for Health Research Health Technology Assessment programme.

The economic analysis was conducted from the perspective of the National Health Service (NHS) and 3 Personal Social Services (PSS) in the UK. The primary economic analysis compared the costs and 4 consequences of each antiepileptic drug over the first 24 months post randomisation. An analysis at 5 an extended 48-month time horizon was planned for those participants followed up for 4 years or 6 more. 7 The within-trial economic analysis was performed using individual, patient-level data from the 8 SANAD-II trial. A cost-utility analysis was conducted to estimate incremental cost-effectiveness 9 ratios, expressed as costs per quality-adjusted life years (QALY) gained. 10 The health economic analysis was carried out in Stata IC version 13 (StataCorp LLc, College Station, 11 TX), and reported according to the CHEERS statement. 1 12 Data sources 13 Resource-use 14 Participants' use of resources was considered in four broad categories: i) resource-use associated 15 with secondary care (inpatient, outpatient, accident and emergency), ii) other healthcare and social 16 services resource-use (primary care, community services), iii) use of anti-seizure medication, and iv) 17 use of other medications. 18 The measurement of resource-use was based on complementary approaches, using data collected as 19 part of the trial, and as part of routine care. Resource-use postal questionnaires, completed by the 20 parent or carer for participants under the age of 16, included a modified Client Service Receipt 21 Inventory (CSRI) based on that from the SANAD trial. [2][3][4] This was used to collect information on 22 participants' use of health service resources, personal social services and medicines. Questions 23 pertained to contacts with health professionals at the GP surgery, in the hospital and in the 24 community, the use of emergency services, and any tests or investigations which participants may 25 have had. The questionnaires were initially administered at 3, 6, 12 months and annually thereafter 26 (up to 60 months); however, from Protocol Version 7 onwards, this questionnaire was also provided 27 during outpatient visits to aid completeness. Questionnaires completed following visits were 28 matched to respective time points for analysis . 29 In all cases, participants were asked to report their primary and secondary care and social services 30 resource-use for the 3-month period prior to completing the questionnaire, and to report their 31 medicines use over a 4-week period prior to completing the questionnaire due to the additional 32 complexity in the recall. The self-report questionnaires additionally contained free-text sections 33 which allowed participants to record any resource-use which would not otherwise be captured by 34 the questionnaire. During analysis these were assessed for duplication against those resources 35 captured by the questionnaire, and any relevant, non-duplicated resources were extracted. Prior to 36 Protocol Version 7 the questionnaire included additional questions relating to a broader 37 perspective, 5 however these were removed in order to shorten the questionnaire, improve 38 completion rates and to prioritise the NHS and PSS perspective, consistent with the NICE guidance 39 for technology appraisal. 6 40 Self-report data were therefore available for months 0-3, 3-6, 9-12, and 21-24. Self-reported 41 resource-use for year 1 was estimated by multiplying the resource-use from months 9-12 by two, 42 and adding the resource-use reported for months 0-3 and 3-6. Self-reported resource-use for year 2 43 was estimated by multiplying resource-use for months 21-24 by four, and similarly for years 3, 4 and 44 5. Participants' use of concomitant medicines was multiplied by three (due to the shorter, 4-week 45 recall period), before estimation following the same method. 46 Anti-seizure medications and their respective doses were recorded directly within case report forms. 47 Routine Hospital Episode Statistics (HES) were the primary source of data on participants' use of 48 secondary care resources over the trial period. HES data were obtained from NHS Digital (for 49 patients in England) 7 and, from the Secure Anonymised Information Linkage (SAIL) databank (for 50 patients in Wales). 8 HES data were not obtained for patients in Scotland or Northern Ireland. HES 51 provided Health Resource Group (HRG) data on the type of care patients receive at a ward level, 52 outpatient visits and Accident and emergency admissions. HES data were used as the source for 53 baseline resource-use and costs, based on the 6 months prior to randomisation. Adjustments were 54 made where hospital episodes overlapped with randomisation date, in order to apportion the 55 resource-use to the periods prior to, and subsequent to, randomisation. 56 All resource-use was measured irrespective of whether they were epilepsy related or otherwise. 9 57 Unit costs 58 Resource-use was valued in monetary terms (£ sterling) using sources of national unit costs. 10-13 59 For data pertaining to participants from Wales an initial mapping step was performed using the 60 Welsh NHS Data dictionary. 14 Subsequently, HRG codes were obtained from the HES data using the 61 NHS Digital costing grouper. 15 Unit costs were allocated based on the latest available National 62 Schedule. 10 63 Unit costs for primary care and community care were taken from the compendium of Unit Costs of 64 Health and Social Care. 11 Unit costs and their sources relating to items within the self-report 65 questionnaire, are presented in Table 1. Unit costs relating to the most commonly reported HRGs 66 are presented in Table 2. 67 Total costs for resource-use were calculated by multiplying the unit cost per item by the recorded 68 number of times that each resource was used. 69  England. 13 Unit costs for trial anti-seizure medications are presented in Table 3. Unless otherwise 5 specified in the data, children aged 9 and over were assumed to be prescribed tablets or capsules, 6 whilst children aged 8 and under were assumed to be prescribed an alternative form (e.g. solution, 7 dispersible) where available. 8 The cost of each medicine was calculated by calculating the price per dose and multiplying by the 9 quantity prescribed (e.g. number of tablets, capsules, inhalers or prefilled syringes), and the number 10 of days of treatment. 11 All costs are at 2019/2020 prices and were discounted in the base-case analysis at the NICE 12 recommended rate of 3.5% per annum. 6 13 14 Health Utilities 15 The primary health outcome measure for the economic analysis was the quality-adjusted life year 16 (QALY), generated from utility data measured using the EuroQol 5-dimension 3-level (EQ-5D-3L) 17 questionnaire. 22 Secondary economic outcome measures were the EQ-VAS, and an epilepsy-specific 18 utility measure, the NEWQOL-6D. 23

19
The EQ-5D descriptive system includes five dimensions, relating to mobility, self-care, usual 20 activities, pain and discomfort, and anxiety. For the EQ-5D-3L and EQ-5D-3L-Y, each dimension is 21 measured against 3 statements (no problems, some problems and extreme problems), scored 1, 2 22 and 3, respectively. The NEWQOL-6D is an epilepsy-specific measure that includes domains of worry, 23 depression, memory, concentration, control and stigma. Responses are measured according to 4 24 categories. Utility scores are obtained from the EQ-5D-3L-Y, EQ-5D-3L, EQ-5D-3L proxy and 25 NEWQOL-6D using UK tariff values. 23,24 26 For participants aged 8 to 15, self-reported responses to the EQ-5D-3L-Y were used, or if not 27 available, proxy questionnaire responses (EQ-5D-3L and NEWQOL-6D), completed by a parent or 28 carer. For participants aged 5-7 years, only proxy questionnaires were administered. All participants 29 aged 8 years or over were administered the EQ-VAS. 30 All economic outcome measures were completed during the baseline visit, and annually thereafter 31 (up to 60 months), and from Protocol version 7 onwards, were also provided during outpatient visits 32 to aid completeness. Utility scores at 365 days (12 months) and at 730 days (24 months) were 33 interpolated, based on recorded utility scores and actual dates of questionnaire completion. QALY 34 profiles were derived from these utilities, estimated based on the area under the curve (AUC) 35 assuming the trapezoidal rule using all available data. QALYs derived from the secondary health 36 economic outcomes (EQ-VAS and NEWQOL-6D) were estimated in the same way, based on AUC. 37 All QALYs were discounted at the NICE recommended rate of 3.5% per annum. 6 38 39 Data analysis 40 Analysis consisted of all randomised participants, which is consistent with the intention to treat 41 approach. All statistical tests were two-sided, with confidence intervals (CIs) and central ranges (CRs) 42 reported at 97.5%. 43 Costs relating to secondary care were primarily sourced from HES data, but where these data were 44 not available, costs were supplemented with resource-use recorded in the self-report 45 questionnaires. Primary and community care costs and concomitant medication costs were also 46 taken from the resource-use questionnaires. Where resource-use questionnaires were returned, but 47 no response was provided for a given resource, then use of that resource was assumed to be zero. 48 Where participants indicated that they had used a resource but had not given a number for how 49 many times the resource was used, then the number was assumed to be one. Data relating to anti-50 seizure medications were taken from the baseline and follow-up CRFs. Missing dose data were 51 assigned according to previous or subsequent prescriptions, based on questions relating to dose 52 changes, and where these were unavailable, from the BNF recommended doses. 53 Data were examined for missingness, and appropriate methods were applied dependent on the level 54 of missingness and likely mechanism of missingness. 25 Missing cost and QALY data were imputed 55 using multiple imputation with chained equations. 26 When the mechanism of data missingness is not 56 missing completely at random, complete case analysis can lead to serious bias which can reverse 57 decisions of cost-effectiveness. 25 Multiple imputation is a flexible approach which provides unbiased 58 results when data are missing at random. 25,26 59 In order to maximise data use, data were imputed at the level of utility scores (EQ-5D, EQ-VAS) at 60 baseline, 12 months and 24 months; primary care, community care and concomitant medications 61 costs at 3 months, 6 months, 12 months and 24 months; admitted patient care, outpatients, 62 accident and emergency and anti-seizure medication costs) at 12 months and 24 months. Baseline 63 costs (relating to admitted patient care, outpatients, accident and emergency) were also imputed for 64 those participants where HES data were not available. Imputation models were generated using 65 predictive mean matching, and data were imputed by randomised treatment group. Variables 66 pertaining to epilepsy classification, seizure type, age, gender, primary outcome and treatment 67 failure were included within the imputation models. Imputation models for baseline measures 68 omitted post-baseline outcomes in order to preserve randomisation. The number of imputations 69 required was based on the level of missingness, according to the fraction of missing information 70 (FMI). 27 71 Based on the imputed data, total costs and QALYs during the course of the trial were calculated, with 72 summary statistics generated by randomised treatment group. Differences between treatment 73 groups were compared with reference to bootstrapped central ranges, based on 10,000 replications. 74 Total costs and QALYs (at 24 months) were adjusted for any imbalances in baseline costs and utilities 75 respectively, and clinical or demographic variables (age, sex, epilepsy classification, with centre as 76 random effects), using ordinary least squares (OLS) regressions. 28, 29 OLS was considered to be 77 appropriate given the large sample size. Several sensitivity analyses were conducted to assess the robustness of the base-case results to key 93 assumptions. These were: 94 1) using discount rates of 0% and 6% per annum for costs and QALYs; 95 2) an unadjusted analysis (i.e. based on mean costs and QALYs, with no regression); 96 3) using results for complete case cost and QALY data (i.e. those without missing data) to 97 identify the impact of missing data and imputation; 98 4) based on the population as the per protocol cohort; and 99 5) using QALYs derived from the NEWQOL-6D and EQ-VAS 100 6) treating blank values in resource use questionnaires as missing, rather than zero. 101 A bootstrap analysis was conducted to consider the joint uncertainty in incremental costs and 102 QALYs. This was represented as a cost-effectiveness plane, and as a cost-effectiveness acceptability 103 curve (CEAC) illustrating the probability of each treatment being cost effective for a given cost-104 effectiveness threshold. 31  to zonisamide. A breakdown of missing data by treatment group and outcome is provided in Table 4. 114 Seven-hundred and eighty-nine participants completed at least one self-report questionnaire 115 (completing either resource use, EQ-5D, or both sections); 621 completed two questionnaires or 116 more. In total, questionnaires were available for 3039 participant-time points (once child and proxy 117 questionnaires had been resolved). 118 Questionnaires returned after the change in protocol were assigned to their nearest time-point for 119 presentation purposes. Self-report resource use data were available for 550 participants at 3 120 months, 527 at 6 months, 465 at 12 months and 398 at 24 months. Resource use data were also 121 available from 496 questionnaires returned at the later time points (36 months, 48 months, 60 122 months). 123 Utility data (EQ-5D) were available for 616 participants at baseline, data were interpolated to 12 124 months for 422 participants and for 319 participants at 24 months. These are lower than the figures 125 reported in Table 4 due to 12-and 24-month questionnaires being dated less than 365 and 730 days 126 post randomisation, respectively. For the NEWQOL-6D, less utility data were available due to a high 127 level of partially completed questionnaires. 128 A total of 50 data sets were imputed, based on the largest FMI (0.7) and accepting <1% reduction in 129 power compared with 100 imputations. For the bootstrapped results, this was reduced to 10 for 130 efficiency purposes, accepting a higher reduction in power in order to achieve an acceptable 131 computation time. 27 Due to the level of missingness, models containing the NEWQOL-6D were non-132 convergent, hence only complete case results are presented for the NEWQOL-6D. 133  -6D  60 months 25  55  80  16  63  79  17  60  77  EQ-VAS  60 months 31  49  80  30  49  79  25  52  77 Resource use and costs Table 5 presents observed mean disaggregated resource-use based on the self-report questionnaires. Table 6 presents the most common admitted patient care episodes, outpatient and accident and emergency related HRGs and costs observed during the trial period. During the 24month follow-up period, 339 unique HRGs were recorded in admitted patient care, 262 in outpatients, and 35 in accident & emergency.
Based on the imputed data, the majority of costs related to secondary care, in particular admitted patient care and outpatient clinic attendance (

Utilities and Quality adjusted life years
The distribution of participants' responses to the EQ-5D-3L-Y and the NEWQOL-6D questionnaires by randomised treatment group are presented in Figures 1 and 2

Incremental analysis
Based on the point estimate mean costs and QALYs, both levetiracetam and zonisamide were more costly and less effective than lamotrigine, and were therefore dominated, meaning that they are not considered to be cost-effective. Zonisamide is associated with a negative incremental net health benefit of -0.171 (97.5% CR -0.295, -0.055) compared with lamotrigine, whilst levetiracetam is associated with a negative health benefit compared with zonisamide -0.010 (97.5% CR -0.142, 0.112) at a cost-effectiveness threshold of £20,000 per QALY. Table 9 presents the results of the sensitivity analyses, which are consistent with the base-case for all analyses other than the NEWQOL-6D, where the net health benefit for levetiracetam is higher than for zonisamide at the £20,000 per QALY cost-effectiveness threshold, and the complete case analysis where levetiracetam is associated with lower costs than lamotrigine, though lamotrigine is still associated with the higher net health benefit.

Sensitivity analyses
The cost-effectiveness acceptability curve (Figure 3) indicates that the probability of levetiracetam being the most cost-effective at a cost-effectiveness threshold of £20,000 per QALY, is 0, whilst the probability for zonisamide is 0.001.

Sub-group analyses
The results of the subgroup analysis for adults, are consistent with the base-case analysis for the whole population (Table 10). For children, however, lamotrigine is associated with the highest costs £5076 (97.5% CR £3815, £7219), compared with levetiracetam £4972 (97.5% CR £3739, £6840), and zonisamide £4638 (97.5% CR £3826, £6974). Levetiracetam is associated with higher QALYs than lamotrigine, and therefore lamotrigine is dominated. Zonisamide has a lower cost, and lower QALYs than levetiracetam, but also a lower net health benefit at a cost-effectiveness threshold of £20,000 per QALY, and is therefore not cost effective at that threshold.