Lamotrigine versus levetiracetam or zonisamide for focal epilepsy and valproate versus levetiracetam for generalised and unclassified epilepsy: two SANAD II non-inferiority RCTs

Article history

The research reported in this issue of the journal was funded by the HTA programme as project number 09/144/09. The contractual start date was in June 2012. The draft report began editorial review in October 2020 and was accepted for publication in August 2021. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HTA editors and publisher have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this report.

Copyright statement

Copyright © 2021 Marson et al. This work was produced by Marson et al. under the terms of a commissioning contract issued by the Secretary of State for Health and Social Care. This is an Open Access publication distributed under the terms of the Creative Commons Attribution CC BY 4.0 licence, which permits unrestricted use, distribution, reproduction and adaption in any medium and for any purpose provided that it is properly attributed. See: https://creativecommons.org/licenses/by/4.0/. For attribution the title, original author(s), the publication source – NIHR Journals Library, and the DOI of the publication must be cited.

2021 Marson et al.

Study design

The SANAD II trial was a pragmatic Phase IV, multicentre, unblinded randomised controlled trial that was conducted in NHS adult neurology and paediatric services. The study was essentially two separate RCTs: the first trial recruited participants with newly diagnosed focal epilepsy who were randomised to start treatment with the ‘standard’ drug lamotrigine or with the ‘new’ drugs levetiracetam or zonisamide, and the second trial recruited participants with newly diagnosed generalised epilepsy or epilepsy that was unclassified at the time of randomisation, who were randomised to start treatment with the ‘standard’ drug valproate or with the ‘new’ drug levetiracetam. Both trials followed the previously published protocol. 30 An economic evaluation was performed to consider the cost-effectiveness of newer drugs compared with the standard drugs. A schematic of the study design is provided in Figure 1.

FIGURE 1.

Schematic of study design.

Study sites

Participants were recruited from NHS outpatient epilepsy, general neurology and paediatric (epilepsy and general) clinics in the UK. The study was co-ordinated through the UK Epilepsy Research Network, the Medicines for Children Research Network, the Wales Epilepsy Research Network and the Comprehensive Clinical Research Network. To be eligible to participate in the study, staff at the sites had to be experienced in treating epilepsy.

Participants

We aimed to recruit 1510 patients (990 with focal onset seizures and 520 with generalised onset seizures or difficult to classify seizures) with the following characteristics.

Recruitment procedure

Patients aged ≥ 5 years who had had two or more spontaneous seizures that required anti-seizure medication and had not previously been treated with anti-seizure medication were screened at the study centre sites to identify participants potentially eligible for the study. Potentially eligible patients (i.e. those meeting the eligibility criteria listed), or their parent/legally acceptable representative, where appropriate, were invited to participate in the study and were provided with a patient information sheet and consent form. The patient (or their parent/legally acceptable representative) was allowed sufficient time to discuss the trial and to decide whether or not to consent to trial entry.

Informed consent

Informed, written consent to enter the SANAD II study was obtained at the baseline visit. The original copy of the signed and dated consent form was filed in the participant’s notes. One copy of the signed consent form was given to the patient (or their parent or legal representative in the case of minors and adults with incapacity) for their records, one copy was retained in the investigator site file, and a final copy was sent to the co-ordinating centre.

If capable, and under appropriate circumstances, minors were approached to provide assent by a member of the research team with experience working with minors. The absence of assent did not exclude the patient, provided that consent had been obtained from the parent/legal representative.

For adults lacking capacity, trial participation was discussed with a personal (or professional) legal representative by a suitably experienced member of the research team. For England, Wales and Northern Ireland, a personal legal representative is someone suitable by virtue of their relationship with the adult and who is available and willing to be the personal legal representative. For Scottish sites, a personal legal representative is a welfare guardian, welfare attorney or nearest relative. They were provided with written information and asked to sign the patient representative consent form.

Randomisation, concealment and blinding

Once eligibility criteria had been confirmed and informed consent and assent, when appropriate, had been obtained, the recruiting clinician selected the appropriate trial based on the patient’s epilepsy classification (focal vs. generalised or unclassified). Patients with focal epilepsy were then randomised in a 1 : 1 : 1 ratio to lamotrigine, levetiracetam or zonisamide; patients with generalised and unclassifiable epilepsy were randomised in a 1 : 1  ratio to levetiracetam or valproate.

Randomisation was performed using a secure (24-hour) web-based randomisation program that was controlled centrally by the Liverpool Clinical Trials Centre (LCTC). A personal login (username and password), provided by the LCTC, was required to access the randomisation system.

Patients were allocated a unique study number (randomisation number) and treatment allocation, displayed to the authorised randomiser on a secure web page, and an automated e-mail confirmation was sent to the authorised randomiser, principal investigator and the trial co-ordinator.

Randomisation used a minimisation program with a built-in random element utilising factors for centre, sex (male or female) and number of previous seizures (two, three to five, or six or more). The factors used for minimisation were not made known to the recruiting sites to avoid any risk of them predicting allocation. The recruiting clinicians were required to initiate trial treatment within 7 days of randomisation.

The SANAD II trial was unblinded, trial treatments were prescribed as per routine NHS practice and dispensed by hospital and community pharmacies, and clinicians prescribed the formulation they considered most appropriate.

Treatment group allocation

The aim of treatment was to control seizures with the minimum effective dose of drug. The trial protocol30 provided guidance on initial drug titration and maintenance doses based on the routine practice at the time that the trial was initiated, although clinicians were able to tailor this as they considered appropriate.

Data collection and management

The majority of clinical data were collected using paper CRFs that were completed by personnel (usually the research nurse) during clinic visits. The paper CRFs were photocopied for local records and originals were returned to the co-ordinating centre for data entry onto a MACRO 4 database (Macro 4 Ltd, Crawley, UK). Where patients defaulted from clinic follow-up, additional information was sought from general practitioners (GPs). Patients were asked to record data on seizures in patient seizure diaries, which were used as an aide-memoire at clinic follow-up visits.

Baseline assessment

Following consent from the patient (or parent/legal representative), the delegated member of the research team completed the baseline CRF to collect data, including seizure history, history of neurological insult or febrile seizures, family history of epilepsy, and the results of electroencephalography (EEG) or imaging [computerised tomography (CT) or magnetic resonance imaging (MRI)]. If further investigations (EEG or imaging) were requested at this visit, data on the results were collected when available, but randomisation was not delayed. If a DNA sample was provided, then the DNA sample CRF was completed. Once all eligibility criteria had been assessed, full eligibility was confirmed by a doctor who had been authorised to do so on the site delegation log; a record of this confirmation was made in the patient’s medical notes. Following the eligibility confirmation, the patient was then randomised.

Follow-up

The expected duration of follow-up for each participant was between 2 and 6.5 years, with visits planned as per routine practice: typically at 3, 6, and 12 months and annually thereafter. Patients could be seen at other times as clinically indicated. All patients were to be followed up even if allocated treatment had been withdrawn. We aimed to complete recruitment over a 4.5-year period, but a 12-month extension was required to meet the sample size target for the focal epilepsy trial, after which the trial cohort was followed up for a further 2 years, allowing a minimum follow-up of 2 years and maximum of 7.5 years for patients in the focal epilepsy trial.

Outcome measures

Sample size

The SANAD II trial was powered to detect non-inferiority of the new anti-seizure medications (levetiracetam and zonisamide) compared with standard treatments [lamotrigine (for focal epilepsy) or valproate (for generalised or unclassified epilepsy)] for the primary outcome time to 12-month remission. A new drug might become a standard first-line treatment if it is proven to be non-inferior for efficacy but superior for tolerability when compared with a standard treatment; tolerability is examined in secondary outcomes, including time to treatment failure for adverse effects. Powering the study for non-inferiority would also provide sufficient power to detect important differences between treatment policies.

The International League Against Epilepsy (ILAE) Commission on Antiepileptic Drugs defined limits of equivalence of ± 10% for the primary outcome in anti-seizure medication monotherapy studies. 41 However, the Commission was not explicit as to whether this should be on the hazard ratio (HR) or absolute scale. No empirical work had been undertaken to underpin the choice of equivalence or non-inferiority margins in epilepsy trials. The chief investigator had given numerous seminars and lectures in the UK and elsewhere about epilepsy trial methodology, and the audience had typically voted for a margin of 10% around absolute differences between anti-seizure medications for monotherapy studies when given examples of margins ranging from 20% to 5%. Communicating treatment differences to patients on a HR scale is also difficult compared with a discussion of absolute differences at specific time points. Given that the ultimate purpose of the SANAD II trial is to provide information that patients and clinicians can use to help them to make treatment decisions, the non-inferiority margin for the SANAD II trial was been chosen according to absolute differences.

Calculations were informed by the SANAD I study, which estimated the 12-month remission-free probability (at 24 months) as 0.43 (exponential hazard rate of 0.0352) for lamotrigine (focal standard), and 0.31 (exponential hazard rate of 0.0488) for valproate (generalised and unclassified epilepsy standard). The calculations assumed a HR of 1.0, 80% power, and allowance for approximately 5% losses to follow-up throughout, as observed in the SANAD I trial. For the focal epilepsy trial, two primary comparisons were of interest (i.e. levetiracetam vs. lamotrigine, and zonisamide vs. lamotrigine); therefore, the one-sided significance level was divided by 2 (one-sided alpha 0.0125). Assuming a 10% absolute difference in survival probability, the non-inferiority margin on the HR scale was:

After adjusting for 5% losses to follow-up, 330 patients were required in each of the three treatment groups (i.e. a total of 990 patients). For the generalised or unclassified epilepsy trial, there was only one comparison of interest (levetiracetam vs. valproate). Assuming a 10% absolute difference in survival probability, the non-inferiority margin on the HR scale was as follows for the trial in generalised or unclassified epilepsy:

Therefore, with a one-sided alpha of 0.025, 260 patients were required in each of the two treatment groups, allowing for 5% losses to follow-up (i.e. a total of 520 patients). The total number of patients required for both trials was 1510. The sample size was calculated using nQuery software (Statistical Solutions Ltd, Cork, Ireland).

Statistical analysis

The statistical analysis and reporting of the SANAD II trial were undertaken in accordance with the Consolidated Standards of Reporting Trials (CONSORT) guidelines42 and the International Conference on Harmonisation E9 guidelines. 43 Primary analyses were undertaken on an intention-to-treat (ITT) basis, including all randomised patients retained in their randomised treatment groups. The statistical and health economic analysis plans were developed before conducting final analyses. Analyses were conducted using SAS^® software (version 9.4; SAS Institute, Cary, NC, USA).

All analyses were conducted separately for the trial in focal epilepsy and the trial in generalised and unclassified epilepsy. A 97.5% two-sided confidence interval (CI) was used for the primary outcome analysis for the focal epilepsy trial (see Sample size for justification). All other CIs (focal epilepsy trial and generalised or unclassified epilepsy trial) were calculated at the 95% level (two-sided), with a two-sided p-value of ≤ 0.05 used to declare statistical significance for all analyses. No formal adjustment was made for multiple testing of secondary outcomes, but conclusions drawn from the analysis of all secondary outcomes would be cautionary unless the p-value was < 0.001.

The time-to-event outcomes were summarised using Kaplan–Meier curves for each treatment group and explored using two different Cox proportional hazards regression models: (1) including the treatment effect only using an indicator variable and (2) including the treatment effect together with minimisation factors included as indicator variables for gender (male or female), number of seizures prior to randomisation (two, three to five, or six or more) and random effects for centre. The assumption of proportional hazards was investigated by examining Schoenfeld residual plots, and incorporating time-dependent covariates in all models. If the residuals were not time dependent and the parameter estimate for the time-dependent covariate was not significant at the 5% level, then the assumption of proportional hazards was assumed to hold; otherwise, an additional extended Cox model with the addition of time-dependent covariates was used. The HR and relevant CI (95% CI unless indicated otherwise) are presented for the comparison of lamotrigine with levetiracetam (focal epilepsy trial), lamotrigine with zonisamide (focal epilepsy trial) and valproate with levetiracetam (generalised or unclassified epilepsy trial). For the primary outcome (12-month remission) non-inferiority hypothesis, the upper limit of the 97.5% CI should be < 1.329 to conclude non-inferiority for the focal epilepsy trial, whereas the upper limit of the 95% CI should be < 1.314 to conclude non-inferiority for the generalised and unclassified epilepsy trial.

A per-protocol (PP) analysis of the primary outcome of time to 12-month remission was also undertaken using a Fine and Gray model,44 with treatment failure included as a competing risk, and censoring participants with drug failure (withdrawn from study or drug or other anti-seizure medication added) before achieving a period of remission. This analysis excluded participants with major protocol deviations, those subsequently given an alternative diagnosis to epilepsy and those who did not receive the drug at all.

For time to treatment failure, a competing risks analysis, using the Fine and Gray model,44 was undertaken to assess the two main reasons for treatment failure (i.e. ISC and UAR). 45 Cumulative incidence curves are presented for each treatment group.

The difference in QoL measures between treatment groups was estimated for each population (child/adult/parent–carer), and for each outcome applicable within that population. This was carried out by fitting a repeated measures random-effects model, with baseline QoL variable as a covariate, along with treatment group and time in days, using spatial power covariance structure for repeated measures (appropriate for repeated measures that can be unevenly spaced), and unstructured covariance for the random effect. 46

Analysis sets for the summary of ARs include all patients who received any dose of a study drug. All ARs and serious adverse reactions (SARs) were coded using the MedDRA dictionary to the most appropriate lower-level term, preferred term and higher-level System Organ Classification. The number (and percentage) of patients experiencing each reaction, and the number (and percentage) of occurrences of each reaction are presented with no formal statistical testing undertaken.

Interim monitoring was carried out by an Independent Data and Safety Monitoring Committee (IDSMC), meeting approximately annually. This included analyses of the primary outcome and five of the secondary outcomes (all using the Haybittle–Peto approach). 47

Health economics

The economic analysis was conducted from the perspective of the NHS and Personal Social Services in the UK. The primary economic analysis compared the costs and consequences of each anti-seizure medication over the first 24 months post randomisation. An analysis at an extended 48-month time horizon was planned for those participants followed up for ≥ 4 years.

The within-trial economic analysis was performed using individual, patient-level data from the SANAD II trial. Cost-utility analyses were conducted to estimate incremental cost-effectiveness ratios, expressed as costs per quality-adjusted life-years (QALYs) gained.

The health economic analysis was carried out in Stata^® IC version 13 (StataCorp LP, College Station, TX, USA), and reported according to the Consolidated Health Economic Evaluation Reporting Standards (CHEERS) statement. 48

Patient and public involvement

The SANAD II trial was designed in collaboration with Epilepsy Action (Leeds, UK), which consulted its members. A patient and public involvement representative sat on the Trial Steering Committee (TSC) and attended regular meetings during the trial. The trial team will collaborate with Epilepsy Action on dissemination of results to the public.

Protocol amendments

During the course of the SANAD II trial, a number of amendments were made to the trial protocol. These are further detailed in Appendix 2. Each amendment was assessed by the Trial Management Group (TMG), TSC, co-sponsors and funder prior to being submitted for approval. Approval for amendments was sought from the Research Ethics Committee, MHRA (if appropriate) and (post 2015) from the Health Research Authority.

Trial funder

The SANAD II trial was funded by the NIHR Health Technology Assessment programme (09/144/09).

Trial co-sponsors

The SANAD II trial was co-sponsored by the University of Liverpool and the Walton Centre NHS Foundation Trust.

Trial management and quality assurance

The SANAD II trial was managed by the LCTC. A risk assessment was performed by the LCTC in conjunction with co-sponsors and the chief investigator. The risk assessment indicated that the SANAD II trial was low risk. As such, monitoring/quality assurance was carried out centrally. This included confirming informed consent; the MACRO database containing predefined ranges that flagged data queries; and the trial statistician producing 6-monthly reports to look for errors, inconsistencies in data, assess safety and to highlight any protocol deviations.

Trial oversight

The SANAD II trial was overseen by the TMG, TSC and IDSMC.

Ethics considerations, regulatory requirements and research governance framework

The SANAD II trial was conducted in accordance with the European Clinical Trials Directive,69 ICH GCP Guidelines,70 the Declaration of Helsinki,71 UK Policy Framework for Health and Social Care Research,72 and the Medicines for Human Use (clinical trials) regulations (2004). 73 The SANAD II trial was issued a EudraCT number (2012-001884-64) and approved by the MHRA on 22 May 2012 (‘effective date’). We also sought and received approval from the North West – Liverpool East Research Ethics Committee for the SANAD II trial to proceed. This was granted on 7 June 2012.

Research Ethics Committee approval was sought for all amendments made to the protocol. MHRA approval was sought for all amendments that related to the trial investigational medicinal products. The SANAD II trial was brought under the HRA umbrella in 2016.

Recruitment and baseline characteristics

The first participant was randomised on 2 May 2013 and the last participant on 20 June 2017 (see Appendix 3, Figure 23), after which every effort was made to follow the trial cohort for a further 2 years; the last participant follow-up visit was on 17 October 2019. Sixty-five UK centres recruited between 1 and 130 patients each, and randomised a total of 990 participants: 330 to start treatment with lamotrigine, 332 to start treatment with levetiracetam and 328 to start treatment with zonisamide (Figure 2). Baseline characteristics were well balanced across treatment groups (Table 2 and see Appendix 3, Table 31). The mean age of participants was 39.3 years [standard deviation (SD) 21.2 years] and 177 (17.9%) participants were aged < 18 years. There was a predominance of males (56.7%), 4.5% of participants had a learning disability, 16.5% had a previous or current neurological disorder, 10.8% a first-degree relative with epilepsy and 4.4% had a history of febrile convulsions. A total of 35.9% of participants were classified with temporal lobe epilepsy, 6.3% with frontal lobe epilepsy, 2.1% with occipital lobe epilepsy, 2.0% with parietal lobe epilepsy and 50.4% with focal epilepsy where localisation was not specified. The median number of seizures before randomisation was 6 [interquartile range (IQR) 3–24] and participants were randomised a median of 13 days (IQR 3–36 days) after their most recent seizure.

FIGURE 2.

The CONSORT participant flow diagram: focal epilepsy trial.

The median number of days of follow-up was 462.5 (IQR 365–777 days) in the lamotrigine group, 449.5 (IQR 365–824) days in the levetiracetam group and 447 (IQR 365–730) days in the zonisamide group, with completeness of follow-up statistics for the primary outcome of 77.2% in the lamotrigine group, 78.3% in the levetiracetam group and 75.6% in the zonisamide group (see Appendix 3, Table 30 and Figure 24).

Time to 12-month remission

Estimates from the primary and secondary analyses are provided in Table 3. For the ITT analysis of time to 12-month remission, there is insufficient evidence to conclude non-inferiority of levetiracetam compared with lamotrigine, as the 97.5% confidence interval for the HR (1.18, 97.5% CI 0.95 to 1.47, unadjusted; 1.13, 97.5% CI 0.91 to 1.41, adjusted) includes the predefined non-inferiority margin of 1.329, but there was sufficient evidence to conclude non-inferiority of zonisamide compared with lamotrigine (HR 1.03, 97.5% 0.83 to 1.28, unadjusted; 1.01, 97.5% CI 0.81 to 1.25, adjusted). There was no evidence of violation of the assumption of proportional hazards (p = 0.90). We also present the annual 12-month remission probabilities (Table 4); for example, we estimate that, at 2 years’ follow-up, compared with the lamotrigine group, the proportion of participants who had achieved remission was 5% lower (97.5% CI –13% to 3%) in the levetiracetam group and 1% lower (97.5% CI –9% to 7%) in the zonisamide group. The Kaplan–Meier estimates of the median number of days to achieve 12-month remission were 516 (97.5% CI 457 to 577) days in the lamotrigine group, 588 (97.5% CI 472 to 706) days in the levetiracetam group and 530 (97.5% CI 453 to 601) days in the zonisamide group (Figure 3).

FIGURE 3.

Kaplan–Meier plot of time to 12-month remission: lamotrigine vs. levetiracetam vs. zonisamide for focal epilepsy.

The PP analyses for time to 12-month remission excluded patients with major protocol deviations (1.5%) and patients who were later diagnosed as ‘not epilepsy’ (1.6%) and accounted for treatment failures prior to achieving 12-month remission (lamotrigine group, 24%; levetiracetam group, 35%; zonisamide group, 39%) in a competing risks analysis (Figure 4). The results indicate that lamotrigine is superior to both levetiracetam (HR 1.32, 97.5% CI 1.05 to 1.66) and zonisamide (HR 1.37, 97.5% CI 1.08 to 1.73).

FIGURE 4.

Cumulative incidence of time to 12-month remission, PP competing risks analysis: lamotrigine vs. levetiracetam vs. zonisamide for focal epilepsy.

Additional prespecified sensitivity analyses, the results of which are shown in Appendix 3, Table 32, did not change the conclusions of the primary analyses.

Time to 24-month remission

The ITT analysis of time to 24-month remission (Figure 5) indicates no significant difference between initiating treatment with lamotrigine or levetiracetam (HR 1.04, 95% CI 0.81 to 1.33) or between initiating treatment with lamotrigine of zonisamide (HR 0.96, 95% CI 0.75 to 1.23).

FIGURE 5.

Kaplan–Meier plot of time to 24 month remission: lamotrigine vs. levetiracetam vs. zonisamide for focal epilepsy.

Time to first seizure

The ITT analysis of time to first seizure (Figure 6) indicates no significant difference between initiating treatment with lamotrigine or levetiracetam (HR 1.07, 95% CI 0.89 to 1.29) or between initiating treatment with lamotrigine of zonisamide (HR 1.04, 95% CI 0.86 to 1.25).

FIGURE 6.

Kaplan–Meier plot of time to first seizure: lamotrigine vs. levetiracetam vs. zonisamide for focal epilepsy.

Time to treatment failure

The analysis of overall time to treatment failure for any reason (Figure 7) indicates a significant advantage of lamotrigine when compared with both levetiracetam (HR 0.60, 95% CI 0.46 to 0.77) and zonisamide (HR 0.46, 95% CI 0.36 to 0.60), with no evidence to suggest violation of the assumption of proportional hazards (p = 0.77). Table 5 provides annual treatment failure rates and differences in failure rates between lamotrigine and both levetiracetam and zonisamide. At 2 years, there was a 16% (95% CI 8% to 23%) difference in the treatment failure rate on levetiracetam and lamotrigine and a 23% (95% CI 15% to 30%) difference between zonisamide and lamotrigine.

FIGURE 7.

Kaplan–Meier plot of time to treatment failure: lamotrigine vs. levetiracetam vs. zonisamide for focal epilepsy.

Table 6 summarises the doses taken at treatment failure or last follow-up and indicates that reasonable dose ranges were tried before deciding that failure had occurred. The competing risks analysis shows that levetiracetam treatment was significantly more likely than lamotrigine treatment to fail due to ARs (HR 0.53, 95% CI 0.35 to 0.79) (see Figure 7), but not ISC (HR 0.67, 95% CI 0.45 to 1.01) (Figure 8). Similarly, zonisamide was significantly more likely to fail than lamotrigine due to ARs (HR 0.37, 95% CI 0.25 to 0.55), but not ISC (HR 0.76, 95% CI 0.50 to 1.15) (Figure 9).

FIGURE 8.

Cumulative incidence of treatment failure because of UARs from competing risks analysis: lamotrigine vs. levetiracetam vs. zonisamide for focal epilepsy.

FIGURE 9.

Cumulative incidence of treatment failure because of ISC from competing risks analysis: lamotrigine vs. levetiracetam vs. zonisamide for focal epilepsy.

Safety

Data were recorded on ARs for the SANAD II trial, which were defined as AEs judged by the treating clinicians to be possibly, probably or definitely caused by anti-seizure medication. Table 7 provides an ITT (by treatment policy) summary of ARs according to the MedDRA System Organ Classification. Summaries by MedDRA-preferred term are presented in Appendix 3, Table 33.

There were 251 ARs experienced by 108 (33%) participants starting treatment with lamotrigine, 328 ARs experienced by 144 (44%) participants starting treatment with levetiracetam and 351 ARs in 146 (45%) participants starting treatment with zonisamide. The main difference in adverse effect profiles was in the prevalence of psychiatric symptoms, which were reported in 13.1% of those starting on lamotrigine, 29.7% of those starting on levetiracetam and 22.5% of those starting on zonisamide.

Seven events in two participants starting on lamotrigine were classified as a SAR, compared with one event in those starting on levetiracetam and four in those starting on zonisamide; there were no suspected unexpected serious adverse reactions (SUSARs) (see Appendix 3, Table 34). There were 37 deaths during the trial: 15 (four likely to be seizure related) in participants starting on lamotrigine, 12 (two likely to be seizure related) in those starting on levetiracetam and 10 (two likely to be seizure related) in those starting on zonisamide (see Appendix 3, Table 35).

There were 11 pregnancies in 11 women starting treatment with lamotrigine (10 with normal postnatal examination and one with minor malformations), six pregnancies in five women starting on levetiracetam (five with normal postnatal examination and one termination), and 17 pregnancies in 14 women starting treatment with zonisamide [eight with normal postnatal examination, eight miscarriages (in five women) and one termination] (see Appendix 3, Table 36).

Quality of life

A total of 493 (49.8%) participants returned QoL questionnaires at baseline and at least one other time point during follow-up. A comparison of those who did and did not return questionnaires showed a similar proportion of male and females, and a similar proportion of those with learning disabilities and neurological deficits to those without, but non-responders were slightly younger (Table 8).

Overall, lamotrigine was associated with a better profile on self-reported measures than levetiracetam or zonisamide. A comparison of the treatment effects in adults (Table 9) revealed negative treatment effects for levetiracetam when compared with lamotrigine for patient-reported anxiety, depression stigma, epilepsy impact and overall QoL. Compared with lamotrigine, zonisamide had a negative treatment effect for depression, epilepsy impact and overall QoL. A comparison of the treatment effects in children is summarised in Table 10. Owing to the small sample size, it is not possible to make any reliable inference about QoL effects.

Data completeness

The HES data were available for a total of 772 participants, relating to 266 participants randomised to start treatment with lamotrigine, 261 participants randomised to start treatment with levetiracetam and 245 participants randomised to start treatment with zonisamide. A breakdown of missing data by treatment group and outcome is provided in Appendix 4, Table 48.

A total of 789 participants completed at least one self-report questionnaire (completing the resource use, EQ-5D or both sections); 621 participants completed two or more questionnaires. In total, questionnaires were available for 3039 participant time points (once child and proxy questionnaires had been resolved).

Questionnaires returned after the change in protocol were assigned to their nearest time point for presentation purposes. Self-report resource use data were available for 550 participants at 3 months, 527 participants at 6 months, 465 participants at 12 months and 398 participants at 24 months. Resource use data were also available from 496 questionnaires returned at the later time points (36, 48 and 60 months).

Utility data (EQ-5D) were available for 616 participants at baseline; data were interpolated to 12 months for 422 participants and to 24 months for 319 participants. These are lower than the figures reported in Appendix 4, Table 48, because the 12- and 24-month questionnaires were dated less than 365 and 730 days post randomisation, respectively. For the NEWQOL-6D, fewer utility data were available because of a large number of partially completed questionnaires.

A total of 50 data sets were imputed, based on the largest fraction of missing information (0.7) and accepting < 1% reduction in power compared with 100 imputations. For the bootstrapped results, this was reduced to 10 for efficiency purposes, accepting a higher reduction in power to achieve an acceptable computation time. 64 Owing to the level of missingness, models containing the NEWQOL-6D were non-convergent; hence only complete-case results are presented for the NEWQOL-6D.

Resource use and costs

Table 11 presents the observed mean disaggregated resource use based on the self-report questionnaires. Table 12 presents the most common admitted patient care episodes, outpatient and A&E-related HRGs, and costs observed during the trial period. During the 24-month follow-up period, 339 unique HRGs were recorded in admitted patient care, including 262 in outpatients and 35 in A&E.

Based on the imputed data, the majority of costs relate to secondary care, in particular admitted patient care and outpatient clinic attendance (Table 13). Comparing across treatment groups, zonisamide has higher secondary care costs and medicines costs than lamotrigine or levetiracetam. The total (unadjusted) costs were £5409 (97.5% CR £4584 to £6658) for participants randomised to start treatment with zonisamide, compared with £5074 (97.5% CR £4433 to £6049) for those randomised to levetiracetam and £4063 (97.5% CR £4842 to £6317) for those randomised to lamotrigine. The differences between the zonisamide and levetiracetam groups (£336, 97.5% CR –£926 to £1634) and between the levetiracetam and lamotrigine groups (£1011, 97.5% CR –£36 to £2066) were not statistically significant. However, the difference in cost between zonisamide and lamotrigine was significant (£1347, 97.5% CR £226 to £2550).

Based on imputed data, the mean baseline costs were £1215 (97.5% CR £1061 to £1375) in the zonisamide group, £1191 (97.5% CR £1035 to £1398) in the levetiracetam group and £1239 (97.5% CR £1036 to £1464) in the lamotrigine group. The base-case analysis that adjusted for baseline costs, age, gender and epilepsy type with centre as random effects yielded a mean 2-year total cost of £5400 (97.5% CR £4659 to £6770) in the zonisamide group, compared with £5104 (97.5% CR £4450 to £6141) in the levetiracetam group and £4042 (97.5% CR £3626 to £4983) in the lamotrigine group. The differences between the zonisamide and levetiracetam groups (£297, 97.5% CR –£900 to £1624) and between the levetiracetam and lamotrigine groups (£1062, 97.5% CR –£1174 to £2133) were not statistically significant. There was a significant difference of £1358 (97.5% CR £376 to £2563) between the zonisamide and lamotrigine groups.

Utilities and quality-adjusted life-years

The distributions of participants’ responses to the EQ-5D-3L-Y and the NEWQOL-6D questionnaires by randomised treatment group are presented in Appendix 4, Figures 27 and 28. Based on imputed data, mean baseline utilities were 0.766 (97.5% CR 0.733 to 0.804) in the levetiracetam group, 0.800 (97.5% CR 0.760 to 0.830) in the zonisamide group and 0.779 (97.5% CR 0.751 to 0.818) in the lamotrigine group. In the base-case adjusted analysis, levetiracetam was associated with a QALY gain of 1.474 years (97.5% CR 1.393 to 1.523 years) over the 2-year time horizon, whereas zonisamide was associated with a QALY gain of 1.502 years (97.5% CR 1.418 to 1.566 years) and lamotrigine was associated with a QALY gain of 1.605 years (97.5% CR 1.547 to 1.651 years). This corresponds to a negative incremental QALY gain of –0.025 years (97.5% CR –0.058 to 0.129 years) between levetiracetam and zonisamide. The incremental QALY gains of –0.103 years (97.5% CR –0.201 to –0.015 years) between zonisamide and lamotrigine and –0.128 years (97.5% CR –0.219 to –0.065 years) between levetiracetam and lamotrigine were significant.

The QALYs based on the NEWQOL-6D were calculated for complete-case data only, over the 2-year time horizon. Levetiracetam was associated with adjusted QALY gains of 1.703 years (97.5% CR 1.678 to 1.727 years), compared with 1.712 years (97.5% CR 1.690 to 1.735 years) for zonisamide and 1.710 years (97.5% CR 1.687 to 1.733 years) for lamotrigine. Levetiracetam was, therefore, associated with a negative incremental QALY gain of –0.007 years (97.5% CR –0.035 to 0.019 years) when compared with zonisamide, and with a negative incremental QALY gain of –0.007 years (97.5% CR –0.035 to 0.019 years) when compared with lamotrigine. The incremental QALY gain between zonisamide and lamotrigine was 0.002 years (97.5% CR –0.021 to 0.025 years).

The distribution of responses to the EQ-VAS is shown in Table 14. The adjusted analysis based on the EQ-VAS resulted in a QALY gain of 1.398 years (97.5% CR 1.324 to 1.479 years) in the levetiracetam group, 1.418 years (97.5% CR 1.351 to 1.456 years) in the zonisamide group and 1.431 years (97.5% CR 1.360 to 1.476 years) in the lamotrigine gorup. The negative incremental QALY gains of –0.020 years (97.5% CR –0.094 to 0.085 years) for levetiracetam compared with zonisamide, –0.013 years (97.5% CR –0.085 to 0.060 years) for zonisamide compared with lamotrigine and –0.033 years (97.5% CR –0.112 to 0.075 years) for levetiracetam compared with lamotrigine are consistent with the base-case EQ-5D.

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 INHB (–0.171, 97.5% CR –0.295 to –0.055) compared with lamotrigine, and levetiracetam is associated with a negative INHB compared with zonisamide (–0.010, 97.5% CR –0.142 to 0.112) at a cost-effectiveness threshold of £20,000 per QALY.

Sensitivity analyses

Table 15 presents the results of the sensitivity analyses, which are consistent with the base case for all analyses other than the NEWQOL-6D, where the NHB for levetiracetam is seen to be higher than for zonisamide at the £20,000 per QALY cost-effectiveness threshold. Table 15 also presents the complete-case analysis in which levetiracetam is associated with lower costs than lamotrigine, but lamotrigine is still associated with the higher NHB.

The cost-effectiveness acceptability curve (Figure 10) indicates that the probability of levetiracetam being the most cost-effective treatment at a cost-effectiveness threshold of £20,000 per QALY is 0, and the probability of zonisamide being the most effective treatment is 0.001.

FIGURE 10.

Cost-effectiveness acceptability curve. Dashed lines represent cost-effectiveness thresholds of £20,000 per QALY and £30,000 per QALY.

Subgroup analyses

The results of the subgroup analysis for adults are consistent with the base-case analysis for the whole population (Table 16). For children, however, lamotrigine is associated with the highest costs (£5076, 97.5% CR £3815 to £7219) compared with levetiracetam (£4972, 97.5% CR £3739 to £6840) and zonisamide (£4638, 97.5% CR £3826 to £6974). Levetiracetam is associated with a higher QALY gain than lamotrigine and, therefore, lamotrigine is dominated. Zonisamide has a lower cost and lower QALY gain than levetiracetam, but also a lower NHB at a cost-effectiveness threshold of £20,000 per QALY, and is therefore not cost-effective at that threshold.

Recommendations for research

1. A network meta-analysis making both direct and indirect comparisons is required of all available anti-seizure monotherapy trials to provide an overview of the entirety of current evidence regarding clinical effectiveness for focal epilepsy. This work is under way and funded by the National Institute for Health Research (NIHR) to inform the current NICE epilepsy guidelines update. 79

2. An economic model is required, utilising the results of direct and indirect comparisons to estimate the comparative cost-effectiveness of currently available treatments for focal epilepsy.

3. Prognostic modelling of data from the SANAD I and SANAD II trials is required to explore subgroup effects and to better stratify patients for likely outcome at the time of initiating treatment for focal epilepsy.

4. Methodological work, utilising data from the SANAD I and SANAD II trials is required to inform the design of future trials assessing the clinical effectiveness and cost-effectiveness of anti-seizure medications in people with newly diagnosed focal epilepsy. This includes the possibility of designs using data from the SANAD I and SANAD II trials as historical controls.

5. Future trials are required to assess the clinical effectiveness and cost-effectiveness of other focal epilepsy treatments including lacosamide, brivaracetam, perampanel and clobazam.

In conclusion, the SANAD II results indicate that lamotrigine should remain a first-line standard treatment for focal epilepsy and that neither levetiracetam nor zonisamide should be used as routine first-line anti-seizure medications.

Recruitment and baseline characteristics

The first participant was randomised on 30 April 2013 and the last participant was randomised on 2 August 2016 (see Appendix 3, Figure 25), after which every effort was made to follow the trial cohort for a further 2 years, and the last participant visit was on 13 January 2019. Sixty-nine UK centres recruited between 1 and 40 patients each, and randomised a total of 520 participants: 260 to start treatment with levetiracetam and 260 to start treatment with valproate (Figure 11). Baseline characteristics were well balanced across treatment groups (Table 17 and see Appendix 3, Table 37). The median age of participants was 13.9 years (IQR 8.9–19.7 years) with a predominance of males (64.8%), reflecting concern about randomising females to valproate. Approximately 10% of patients had a learning disability, 3.1% a neurological deficit and 19% had a first-degree relative with epilepsy. Approximately three-quarters of the participants had generalised epilepsy and in the remainder the type of epilepsy was unclassified. Of those with generalised epilepsy (397), 26.2% had childhood absence epilepsy, 9.1% juvenile absence epilepsy, 12.8% juvenile myoclonic epilepsy, 5.8% generalised epilepsy with tonic–clonic seizures on waking and 45.3% were classified as having ‘idiopathic generalised epilepsy not specified’. Participants were randomised a median of 4 days (IQR 0–26 days) after their most recent seizure.

FIGURE 11.

The CONSORT participant flow diagram: generalised and unclassified epilepsy trial.

The completeness of follow-up statistic (see Appendix 3, Table 38 and Figure 26) for the primary outcome was 87% for valproate and 83% for levetiracetam. In this analysis, the median number of days of follow-up was 427 (IQR 365–731) days for the valproate group and 550 (IQR 366–781) days for the levetiracetam group; follow-up was shorter in the valproate group, as participants allocated valproate achieved the primary outcome sooner.

Time to 12-month remission

For the ITT analysis of time to 12-month remission, there is insufficient evidence to conclude non-inferiority of levetiracetam compared with valproate, as the 95% CI for the HR (unadjusted: 1.19, 95% CI 0.96 to 1.47; adjusted: 1.23, 95% CI 0.99 to 1.52) includes the predefined non-inferiority margin of 1.314. There is crossing of the Kaplan–Meier survival curves (Figure 12) and evidence of non-equality of HRs across time (p = 0.001), violating the assumption of proportional hazards. To supplement presentation of the average treatment effect over time, we also present interval-specific HR estimates (Table 18), which indicate a significant beneficial effect of initiating treatment with valproate within the first year. Thereafter there are no statistically significant effects but the direction of benefit is in favour of valproate during the interval from 1 to 2 years and in favour of levetiracetam during the interval from 2 to 3 years. These results should be viewed cautiously, as they are sensitive to interval choice and subject to selection bias. We also present the annual difference in 12-month remission probabilities (Table 19) showing, for example, that at 1 year the proportion of patients entering 12-month remission was 9% lower in the levetiracetam group than in the valproate group. The Kaplan–Meier estimate of the median number of days to achieve 12-month remission was also shorter for valproate (445 days, 95% CI 406 to 531 days) than for levetiracetam (636 days, 95% CI 553 to 728 days).

FIGURE 12.

Kaplan–Meier plot of time to 12-month remission ITT analysis: levetiracetam vs. valproate for generalised or unclassified epilepsy.

The PP analyses for time to 12-month remission (Figure 13 and see Table 18) excluded, patients with major protocol deviations (1%) and patients later diagnosed as ‘not epilepsy’ (1%) and accounted for treatment failures prior to achieving 12-month remission (32% valproate, 47% levetiracetam). This analysis indicates superiority of valproate over levetiracetam (HR 1.68, 95% CI 1.30 to 2.15). Furthermore, the assumption of constant HR across time appears reasonable in the PP analysis, suggesting that treatment failures prior to remission largely explain the non-constant effect seen in the ITT analysis.

FIGURE 13.

Cumulative incidence of time to 12-month remission from competing risks PP analysis: levetiracetam vs. valproate for generalised or unclassified epilepsy.

Additional prespecified sensitivity analyses, the results of which are detailed in Appendix 3, Table 39, did not change the conclusions of the primary analysis.

Subgroup effects were explored in a post hoc analysis (Figure 14) and indicate an important advantage for initiating valproate in those with other idiopathic generalised epilepsies (HR 1.55, 95% CI 1.14 to 2.11), as the difference in immediate 12-month remission rates was 19.1% (95% CI 6.6% to 31.7%), but not for absence epilepsies (HR 0.90, 95% CI 0.60 to 1.35) or unclassified epilepsy (HR 1.07, 95% CI 0.69 to 1.67).

FIGURE 14.

Kaplan–Meier plots of time to 12-month remission epilepsy type subgroup analysis: levetiracetam vs. valproate. (a) Absence epilepsy; (b) other generalised epilepsy; and (c) unclassified epilepsy.

Time to 24-month remission

In the ITT analysis of time to 24-month remission, initiating treatment with valproate was superior to initiating treatment with levetiracetam (HR 1.43, 95% CI 1.06 to 1.92). As with time to 12-month remission, there is crossing of the Kaplan–Meier survival curves (Figure 15) and evidence of non-equality of HRs across time (p = 0.002), suggesting a violation of the assumption of proportional hazards. The difference in 24-month remission rates was –12% (95% CI –20% to –4%) at 24 months’ follow-up, diminishing to –4% (95% CI –17% to 8%) at 4 years.

FIGURE 15.

Kaplan–Meier plot of time to 24-month remission: levetiracetam vs. valproate for generalised or unclassified epilepsy.

Time to first seizure

For time to first seizure (Figure 16), valproate was superior to levetiracetam (HR 0.82, 95% CI 0.67 to 1.00), and there was insufficient evidence to suggest a violation of the assumption of proportional hazards (p = 0.39) – most likely because this analysis would not be affected by treatment failures for inadequate seizure control.

FIGURE 16.

Kaplan–Meier plot of time to first seizure: levetiracetam vs. valproate for generalised or unclassified epilepsy.

Time to treatment failure

The analysis of overall time to treatment failure for any reason (Figure 17) shows a significant benefit for valproate (HR 0.65, 95% CI 0.50 to 0.83), with no evidence to suggest a violation of the assumption of proportional hazards (p = 0.22). Table 20 provides annual treatment failure rates and differences in failure rates between valproate and levetiracetam. At 2 years there was a difference of –15% (95% CI –23% to –6%) in the treatment failure rate on levetiracetam compared with the treatment failure rate on valproate.

FIGURE 17.

Kaplan–Meier plot of time to treatment failure: levetiracetam vs. valproate for generalised or unclassified epilepsy.

Table 21 summarises the doses taken at treatment failure or last follow-up and indicate that reasonable dose ranges were tried before deciding failure had occurred. The competing risks analysis shows that this difference is predominantly driven by failures due to ISC with a subdistribution (HR 0.43, 95% CI 0.30 to 0.63) (Figure 18), whereas there is little difference between treatments for treatment failure due to unacceptable ARs (HR 0.93, 95% CI 0.61 to 1.40) (Figure 19).

FIGURE 18.

Cumulative incidence of treatment failure for ISC from competing risks analysis: levetiracetam vs. valproate for generalised or unclassified epilepsy.

FIGURE 19.

Cumulative incidence of treatment failure due to UARs from competing risks analysis: levetiracetam vs. valproate for generalised or unclassified epilepsy.

The HR for overall treatment failure was not consistent across epilepsy types (Figure 20), with a significant benefit for valproate for absence (HR 0.58, 95% CI 0.37 to 0.89) and other generalised types of epilepsy (HR 0.44, 95% CI 0.30 to 0.66) but not for unclassified epilepsy (HR 1.44, 95% CI 0.85 to 2.45; test for interaction p = 0.002). The competing risks analysis shows a similar pattern of effects for failure due to ISC [absence epilepsies, HR 0.35 (95% CI 0.19 to 0.63); other generalised eplipsy, HR 0.27 (95% CI 0.14 to 0.49); and unclassified epilepsy, HR 2.15 (95% CI 0.79 to 5.86)]. There was no difference between treatments in failure due to UARs by epilepsy type.

FIGURE 20.

Hazard ratios for treatment failure by subgroups: absence epilepsies, other idiopathic generalised epilepsy and unclassified epilepsy. (a) Time to treatment failure any reason; (b) time to treatment failure for UARs; and (c) time to treatment failure for ISC. IV, inverse variance.

Safety

The SANAD II trial recorded data on ARs, which were AEs judged by the treating clinicians to be possibly, probably or definitely caused by anti-seizure medicines. ARs according to the MedDRA System Organ Classification are shown in Table 22, and Appendix 3, Table 40, shows ARs by MedDRA-preferred term. There were 220 ARs experienced by 96 (37.4%) patients who were randomised to start treatment with valproate and 223 ARs experienced by 107 patients (41.5%) randomised to start treatment with levetiracetam. The profile of ARs is different from most notably psychiatric symptoms reported in those allocated to levetiracetam (109 events, 66 participants) compared with those allocated to valproate (54 events, 36 participants). There were more reports of weight gain in those starting on valproate (26 participants; 10.1%) than in those starting on levetiracetam (eight participants; 3.1%). Of those randomised to start treatment with valproate, 10 (3.9%) had a total of 15 severe ARs and, of those randomised to start treatment with levetiracetam, 10 (3.5%) had a total of 16 severe ARs. For two (0.8%) patients on valproate and four (1.6%) patients on levetiracetam, the ARs were classified as serious (see Appendix 3, Table 41). None of the ARs were classified as SUSARs. There were two deaths, one in each group, that were unrelated to trial treatments (see Appendix 3, Table 42). One participant randomised to start treatment with valproate became pregnant, and the pregnancy was conceived while taking levetiracetam monotherapy and resulted in a normal healthy baby at postnatal examination. Nine participants randomised to start treatment with levetiracetam reported a pregnancy, which resulted in four normal healthy babies at postnatal examination (levetiracetam, n = 3; levetiracetam plus pregabalin being taken at the time of reporting pregnancy, n = 1), three miscarriages (all levetiracetam taken at time of reporting pregnancy), one low-birthweight baby (levetiracetam being taken at the time of reporting pregnancy) and one baby with major malformations (carbamazepine being taken at the time of reporting pregnancy) (see Appendix 3, Table 43).

Quality of life

Of the 520 randomised participants, 299 (58%) returned a baseline QoL questionnaire; the response rate was slightly higher for parents of children aged 5–15 (61% return) than for adults (50% return). At baseline, non-responders were more likely than the responders to be male (71% vs. 57%) and have unclassified epilepsy (27% vs. 19%), and were less likely to have generalised epilepsy (73% vs. 81%). Responders and non-responders were similar in terms of numbers of those with a learning disability or neurological disorder. The response rate diminished substantially after baseline for all subsequent time points, despite sending questionnaires to all participants from the trial office and intervention from the trial management team to reduce the length of the questionnaire and to encourage investigators to provide participants with questionnaires at clinical visits. Results from the repeated measures random-effects models (Table 23) suggest that there may be small differences in favour of levetiracetam for the QoL emotional (child), family (child and parent) and school (child) domains. However, because of the high level of missing data, these results cannot be considered as reliable. Imputation was also not considered reasonable because of the high level of missing data.

Data completeness

Hospital Episode Statistics data were available for a total of 412 participants, relating to 202 participants in the valproate treatment group and 210 participants randomised to start treatment with levetiracetam. A breakdown of missing data by treatment group and outcome is provided in Appendix 4, Table 49.

A total of 389 participants returned at least one self-report questionnaire (completing the resource use, EQ-5D or both sections); 280 participants returned two or more questionnaires. In total, questionnaires were available for 1238 participant-time points (once child and proxy questionnaires had been resolved).

Questionnaires returned after the change in protocol were assigned to their nearest time point for presentation purposes. Self-report resource use data were available for 243 participants at 3 months, 212 at 6 months, 185 at 12 months and 148 at 24 months. Resource use data were also available from 153 questionnaires returned at the later time points (36, 48 and 60 months). It was deemed that there were insufficient data available for any meaningful analysis beyond the primary time horizon of 24 months, and thus the planned 48-month analysis was not conducted.

Utility data (EQ-5D) were available for 274 participants at baseline, and data were interpolated to 12 months for 161 participants and to 24 months for 128 participants. These are lower than the figures reported in Appendix 4, Table 49, because some 12-month questionnaires were dated before 365 days post randomisation and some 24-month questionnaires were dated before 730 days post randomisation. For the NEWQOL-6D, fewer utility data were available because of a high level of partially completed questionnaires.

A total of 50 data sets were imputed, based on the fraction of missing information 0.7 and accepting < 1% power fall-off compared with 100 imputations. For the bootstrapped results, this was reduced to 10 for efficiency purposes, accepting a larger reduction in power to achieve an acceptable computation time. 64 Owing to the level of missingness, models containing the NEWQOL-6D were non-convergent; hence only complete-case results are presented for the NEWQOL-6D.

Resource use and costs

Table 24 presents observed mean disaggregated resource use based on the self-report questionnaires. Table 25 presents the most common outpatient and A&E-related HRGs and costs observed during the trial period. During the 24-month follow-up period, 108 unique HRGs were recorded in admitted patient care, 168 in outpatients and 30 in A&E. The most common inpatient attendances were WJ11Z (disorders of immunity) (n ≈ 35), PR02C [paediatric epilepsy with a complication or comorbidity (CC) score of 0] (n ≈ 15) and PR02B (paediatric epilepsy with a CC score of 1–5) (n ≈ 15).

Based on the imputed data, the majority of costs relate to secondary care, in particular outpatient clinic attendance (Table 26). Anti-seizure medications also account for a high proportion of the total cost. Participants randomised to start treatment with levetiracetam produced higher costs for community care and secondary care, but lower medication costs than those randomised to start treatment with valproate. Total (unadjusted) costs for participants randomised to start treatment with levetiracetam were £4267 (95% CR £3944 to £5462), compared with valproate £4205 (95% CR £3827 to £4956). The difference of £61 (95% CR –£651 to £1230) was not statistically significant.

Based on imputed data, the mean baseline costs were £1067 (95% CI £934 to £1234) in the levetiracetam group and £1088 (95% CI £978 to £1213) in the valproate group The base-case analysis, which adjusted for baseline costs, yielded a 2-year mean total cost of £4350 (95% CR £4136 to £5623) in the levetiracetam group, compared with £4246 (95% CR £3979 to £5090) in the valproate group. This corresponds to an incremental cost of £104 (95% CR –£587 to £1234).

Utilities and quality-adjusted life-years

The distributions of participants’ responses to the EQ-5D-3L-Y and the NEWQOL-6D questionnaires by randomised treatment group are presented in Appendix 4, Figures 29 and 30. Based on imputed data, baseline utilities were 0.831 (95% CR 0.779 to 0.850) in the levetiracetam group and 0.811 (95% CR 0.772 to 0.840) in the valproate group. In the base-case adjusted analysis, levetiracetam was associated with a QALY gain of 1.603 years (95% CR 1.500 to 1.631 years), whereas valproate was associated with a QALY gain of 1.637 years (95% CR 1.565 to 1.673 years) over the 2-year time horizon. This corresponds to an incremental QALY gain of –0.035 years (95% CR –0.137 to 0.032 years).

The QALYs based on the NEWQOL-6D were calculated for complete-case data only, over the 2-year time horizon. Levetiracetam was associated with an adjusted QALYs gain of 1.741 years (95% CR 1.695 to 1.784 years), compared with 1.727 years (95% CR 1.697 to 1.779 years) for valproate. This shows levetiracetam to be associated with an incremental QALY gain of 0.015 years (95% CR –0.051 to 0.056 years).

The distribution of responses to the EQ-VAS is presented in Table 27. The adjusted analysis based on the EQ-VAS resulted in a QALY gain of 1.453 years (95% CR 1.358 to 1.495 years) for levetiracetam and of 1.464 years (95% CR 1.369 to 1.502 years) for valproate; the incremental QALY of –0.011 years (95% CR –0.103 to 0.077 years) is consistent with the base-case EQ-5D-3L analysis.

Incremental analysis

Based on the point estimate mean costs and QALYs, levetiracetam was both more costly and less effective than valproate, and, therefore, dominated, meaning that it is not considered to be cost-effective. Levetiracetam is associated with a negative incremental NHB (–0.040, 95% CR –0.175 to 0.037) at a cost-effectiveness threshold of £20,000 per QALY.

Sensitivity analysis

Table 28 presents the results of the sensitivity analyses, which are consistent with the base case for all analyses other than the complete-case analysis and NEWQOL-6D analysis, both of which are limited by missing data.

The cost-effectiveness plane, which depicts the joint uncertainty in costs and QALYs, is presented in Figure 21. It shows that 62% of simulations were more costly and less effective, 23% were less costly but less effective, 8% were less costly and more effective, and 7% were more costly and more effective.

FIGURE 21.

Cost-effectiveness plane based on 1000 simulations. Simulations that were cost-effective only at the lower £20,000 per QALY threshold are coloured orange; those which were cost-effective only at the higher £30,000 per QALY threshold are in light purple; those cost-effective at both the £20,000 and £30,000 per QALY thresholds are in light blue; and those that were not cost-effective at either threshold are in dark blue.

The cost-effectiveness acceptability curve (Figure 22) indicates that the probability of levetiracetam being cost-effective at a cost-effectiveness threshold of £20,000 per QALY is 0.17.

FIGURE 22.

Cost-effectiveness acceptability curve.

Subgroup analyses

The results of the subgroup analysis for both children and adults are consistent with the base-case analysis (all participants) and are presented in Table 29.

Contributions of authors

Anthony G Marson (https://orcid.org/0000-0002-6861-8806) (Chief Investigator, Professor of Neurology and Honorary Consultant Neurologist) developed the trial protocol in collaboration with co-investigators. He oversaw the delivery of the trial, and oversaw clinical aspects of the statistical analysis plan and clinical interpretation of the trial data. He led the preparation of the final report (drafting, reviewing and editing). He was chairperson of the TMG.

Girvan Burnside (https://orcid.org/0000-0001-7398-1346) (Trial Statistician and Senior Lecturer) contributed to protocol development and data capture methods, undertook the final statistical analysis, prepared data for reports throughout the trial, prepared data tables and figures for the final report, contributed to the final report (drafting reviewing and editing) and was a member of the TMG.

Richard Appleton (https://orcid.org/0000-0002-0742-2113) (Consultant Paediatric Neurologist and Honorary Professor) contributed to the trial protocol development, interpretation of the trial data and the preparation (reviewing and editing) of the final report, and was a member of the TMG.

Dave Smith (https://orcid.org/0000-0003-3449-4777) (Consultant Neurologist) contributed to the clinical interpretation of the trial data and the preparation (reviewing and editing) of the final report.

John Paul Leach (https://orcid.org/0000-0003-2086-9937) (Professor and Consultant Neurologist) contributed to the trial protocol development, interpretation of the trial data and the preparation (reviewing and editing) of the final report, and was a member of the TMG.

Graeme Sills (https://orcid.org/0000-0002-3389-8713) (Senior Lecturer in Pharmacology) co-ordinated the collection and archiving of DNA samples, contributed to the trial protocol development, interpretation of the trial data and the preparation (reviewing and editing) of the final report, and was a member of the TMG.

Catrin Tudur-Smith (https://orcid.org/0000-0003-3051-1445) (Statistical Lead, Professor of Statistics) contributed to trial design, trial protocol and data capture methods in collaboration with co-investigators. She led the blind review of the data and supervised the final analysis. She also contributed to the preparation of the final report (drafting, reviewing and editing) and was a member of the TMG.

Catrin O Plumpton (https://orcid.org/0000-0003-2710-9199) (Study Health Economist) contributed to protocol development and data capture methods, undertook the economic analysis under the supervision of Dyfrig Hughes and contributed to the final report (drafting and editing).

Dyfrig A Hughes (https://orcid.org/0000-0001-8247-7459) (Senior Health Economist) led the economic evaluation, contributed to protocol development and data capture methods, contributed to the writing of the report (drafting, reviewing and editing), and was a member of the TMG.

Paula R Williamson (https://orcid.org/0000-0001-9802-6636) (Senior Statistician, Professor of Statistics) contributed to protocol development and data capture methods, contributed to the drafting (reviewing and editing) of the final report, and was a member of the TMG.

Gus Baker (https://orcid.org/0000-0002-5736-605X) (Professor of Neuropsychology) led on the QoL assessments, contributed to protocol development and QoL data capture methods, contributed to the drafting (reviewing and editing) of the final report, and was a member of the TMG.

Silviya Balabanova (https://orcid.org/0000-0001-6989-8099) (Trial Manager) contributed to protocol development, supported the preparation of progress reports, contributed to the final report (drafting and reviewing) and was a member of the TMG.

Claire Taylor (https://orcid.org/0000-0003-4746-9730) (Trial Manager) contributed to protocol development, supported the preparation of progress reports and contributed to the final report (drafting and reviewing).

Richard Brown (https://orcid.org/0000-0002-1591-3959) (Consultant Paediatrician) contributed to clinical interpretation of the trial data and the preparation of the final report.

Dan Hindley (https://orcid.org/0000-0002-4128-6995) (Consultant Community Paediatrician) contributed to clinical interpretation of the trial data and the preparation of the final report.

Stephen Howell (https://orcid.org/0000-0001-8504-9644) (Consultant Neurologist) contributed to clinical interpretation of the trial data and the preparation of the final report.

Melissa Maguire (https://orcid.org/0000-0002-5351-6743) (Consultant Neurologist) contributed to clinical interpretation of the trial data and the preparation of the final report.

Rajiv Mohanraj (https://orcid.org/0000-0002-2559-148X) (Consultant Neurologist) contributed to clinical interpretation of the trial data and the preparation of the final report.

Philip EM Smith (https://orcid.org/0000-0003-4250-2562) (Consultant Neurologist and Honorary Professor) contributed to the trial protocol development, interpretation of the trial data and preparation of the final report, and was a member of the TMG.

Publications

Balabanova S, Taylor C, Sills G, Burnside G, Plumpton C, Smith PEM, et al. Study protocol for a pragmatic randomised controlled trial comparing the effectiveness and cost-effectiveness of levetiracetam and zonisamide versus standard treatments for epilepsy: a comparison of standard and new antiepileptic drugs (SANAD-II). BMJ Open 2020;10:e040635.

Marson A, Burnside G, Appleton R, Smith D, Leach JP, Sills G, et al. 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. Lancet 2021;397:1363–74.

Marson A, Burnside G, Appleton R, Smith D, Leach JP, Sills G, et al. The SANAD II study of the effectiveness and cost-effectiveness of valproate versus levetiracetam for newly diagnosed generalised and unclassifiable epilepsy: an open-label, non-inferiority, multicentre, phase 4, randomised controlled trial. Lancet 2021;397:1375–86.

Data-sharing statement

All requests for data (pseudoanonymised and fully anonymous) should be sent to the corresponding author. For data which are not fully anonymous, the decision for data sharing also lies with the trial joint data controllers (University of Liverpool, Walton Centre NHS Foundation Trust and Bangor University). Access to available data may be granted following review.

Patient data

This work uses data provided by patients and collected by the NHS as part of their care and support. Using patient data is vital to improve health and care for everyone. There is huge potential to make better use of information from people’s patient records, to understand more about disease, develop new treatments, monitor safety, and plan NHS services. Patient data should be kept safe and secure, to protect everyone’s privacy, and it’s important that there are safeguards to make sure that it is stored and used responsibly. Everyone should be able to find out about how patient data are used. #datasaveslives You can find out more about the background to this citation here: https://understandingpatientdata.org.uk/data-citation.

Disclaimers

This report presents independent research funded by the National Institute for Health Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, NETSCC, the HTA programme or the Department of Health and Social Care. If there are verbatim quotations included in this publication the views and opinions expressed by the interviewees are those of the interviewees and do not necessarily reflect those of the authors, those of the NHS, the NIHR, NETSCC, the HTA programme or the Department of Health and Social Care.

Additional tables and figures for the focal epilepsy trial

FIGURE 23.

Recruitment graph for the focal epilepsy trial.

Additional baseline tables and figures for focal epilepsy trial

FIGURE 24.

Completeness of follow-up scatterplots of observed follow-up time by date of randomisation: focal epilepsy trial. (a) Lamotrigine; (b) levetiracetam; and (c) zonisamide.