TA615 · STA

Cannabidiol with clobazam for treating seizures associated with Lennox–Gastaut syndrome

Recommended with restrictionsNovember 2019

Frequency of drop seizures checked every 6 months and cannabidiol stopped if the frequency has not fallen by at least 30% compared with the 6 months before starting treatment. Company to provide cannabidiol according to commercial arrangement.

Source documents

Intervention

cannabidiol (Epidyolex)
· cannabinoid receptor agonist · oral

Condition

lennox–gastaut syndromeneurology_psychiatry · chronic

Comparators

NameType Established Committee preferred
placeboplacebo
usual carestandard of careYes

Clinical trials

TrialDesignPhasePivotal
GWPCARE3RCTPhase 3Yes
GWPCARE4RCTPhase 3Yes
GWPCARE5open_label

Economic model

markov (company)
Time horizon: 36 months minimum (based on open-label extension follow-up)
Cycle length: 3 months (based on references to cycles and 27 months = 9 cycles)

Methodological decisions (16)

cost assumption

Model validity concern with QALYs for patients stopping cannabidiol

ERG: Identified that model generates more favourable results for patients stopping cannabidiol than expected; when all clinical inputs equal, model estimated higher QALYs for cannabidiol than expected

ICER impact: increases

cost assumption

Company used median body weight from trials to calculate weight-based dose. Committee preferred mean weight, noting median was lower than mean and good practice recommends mean values.

Company: Median weight appropriate due to significant outliers in trial data

Committee: Mean body weight should be used; committee agreed outlier patients would receive treatment in NHS

ICER impact: decreases

cost assumption

Company assumed maintenance dose of 10 mg/kg/day for all patients despite 10-20 mg/kg/day range in summary of product characteristics. Company provided scenario with 12 mg/kg/day average (20% at 20 mg/kg/day).

Company: Base case 10 mg/kg/day appropriate; scenario with 12 mg/kg/day average and another with 9 mg/kg/day provided

Committee: Scenario with 12 mg/kg/day average dose preferred as it captures dose increases

ICER impact: increases

cost assumption

Company assumed 2 carers per patient based on clinical expert opinion. Provided scenarios with 1.8 carers (literature-based, Lagae et al. 2017) and 3 carers. Linear multiplication of disutility by number of carers.

Company: 2 carers base case; scenarios with 1.8 and 3 provided; vignette study accounted for care-sharing

Committee: 1.8 carers scenario preferable; linear multiplication approach inappropriate but limited impact with 1.8 value

ICER impact: increases

model structure

Definition of health states by seizure frequency ranges

Company: Initially used health states with very wide seizure ranges (e.g. 45-110 seizures) divided by overall trial population

Committee: Health states defined by narrower ranges of seizures based on clinical rationale, with most patients experiencing 50% seizure change moving to different health state

ICER impact: decreases

model structure

Health state definition based on seizure frequency ranges. Company initially used wide ranges (e.g. 45-110 seizures), committee was concerned these lacked clinical rationale. Company revised to narrower ranges based on 50% seizure change being clinically meaningful.

Company: Wide seizure ranges initially, then revised to narrower ranges ensuring 50% change moves patients to different health state

Committee: Narrower ranges of seizures appropriate for decision making

ICER impact: negligible

model structure

Company divided each of 3 drop seizure health states into substates based on seizure-free days (exploratory endpoint). Committee noted other approaches like discrete event simulation may be more appropriate but accepted this method.

Company: Substates based on seizure-free days to capture both seizure frequency and days without seizures benefits

Committee: Approach acceptable, though other methods like discrete event simulation may be more appropriate

ICER impact: negligible

model structure

Model reassignment of patients who stopped cannabidiol. Company assumed those stopping return to health states in same proportions as usual care patients, but most stopping were in highest seizure frequency state initially. This could result in some being reassigned to lower seizure states than baseline.

Company: Patients who stop cannabidiol should be reassigned to health states in same proportions as usual care; justified by lack of clinical data on stopping outcomes

Committee: Patients should be split into quantiles and redistributed to corresponding quantiles in usual care arm to limit inappropriate transitions

ICER impact: increases

mortality assumption

Company assumed people without drop seizures have lower epilepsy-related mortality risk (58% reduction per Trinka et al. 2013), later halved to 29%. Committee concerned observational evidence was confounded and not supported by trial evidence.

Company: Reasonable to assume 29% mortality reduction for seizure-free patients based on observational data

Committee: Insufficient evidence to prove cannabidiol prolongs life; prefer scenario without mortality benefit assumption

ICER impact: decreases

stopping rule

Whether to apply stopping rule at 3 months or 6 months

Company: Stopping at 3 months inappropriate because titration to therapeutic dose takes longer than 3 months

Committee: Stopping rule should be assessed every 6 months, based on reduction in drop seizures compared with 6 months before starting cannabidiol

ICER impact: uncertain_direction

treatment effect waning

Company assumed cannabidiol effect maintained in same health state beyond 27 months based on 36-month open-label extension data. Clinical experts expected efficacy to diminish over time as with other antiepileptic drugs.

Company: Effect maintained as long as patient takes cannabidiol; captured diminishment only through increased stopping rates in scenario

Committee: Effect likely to diminish over time; scenario analysis capturing stopping captures some but not all effects of efficacy waning

ICER impact: increases

treatment sequencing

Company model originally assumed usual care patients returned to baseline health states. Revised to assume patients remained in same health states from cycle 2 onward to address lack of comparator data in open-label extension.

Company: Patients remain in same health states from 6 months onward; assumption disadvantages cannabidiol by overestimating usual care effectiveness

Committee: New base-case assumption in line with preferences and appropriate for lack of comparator data

ICER impact: negligible

utility source

Capturing benefits of reducing non-drop seizures

Company: Applied additional disutility value from de Kinderen et al. 2016 epilepsy preference study, assuming patients with fewer drop seizures benefit from fewer non-drop seizures

ERG: Concerned about double counting of benefits and unable to reproduce company's utility estimates; clinical trial data on non-drop seizure reduction not used directly in model

Committee: Company's approach may not be valid and may double-count benefits; however, benefit of reducing non-drop seizures should be considered in decision-making

ICER impact: uncertain_direction

utility source

Company included non-drop seizure benefits via disutility adjustment from public preference study (de Kinderen et al. 2016). ERG concerned about double counting and could not reproduce utility estimates. Clinical trial data on non-drop seizure reduction not directly used.

Company: Applied additional disutility value from literature for non-drop seizure reduction

ERG: Approach may have led to double counting; unable to reproduce estimates

Committee: Approach may not be valid, but acknowledged importance of reducing non-drop seizures and would consider in decision-making

ICER impact: uncertain_direction

utility source

Company used vignette study values from patients and carers instead of literature data. Vignette study asked patients/carers to rate health states; lowest value was 0 (death) vs EQ-5D allowing negative values. Committee noted limitations but found values appropriate.

Company: Vignette study necessary due to lack of literature data matching model structure; company considered values confidential

Committee: Vignette approach justified but has limitations; values from Verdian et al. 2018 general population study better aligned with NICE preference, but vignette values acceptable given model structure constraints

ICER impact: uncertain_direction

utility source

Company included carer quality-of-life decrements based on vignette study, applied to only 2 highest seizure frequency health states. Committee concerned about limited scope and preference for public preference study values.

Company: Vignette study values appropriate for captured health states; other family members may benefit from improved seizure control

Committee: Appropriate to include carers' quality of life; vignette study best available source despite limitations

ICER impact: negligible

Evidence gaps

short follow upClinical trials had follow-up of 14 weeks; long-term efficacy after 3 years uncertain despite interim analysis from GWPCARE5
immature overall survivalLong-term safety data limited; GWPCARE5 ongoing extension study will provide more information on safety over up to 5 years

Commercial arrangement

simple discount pas · confidential · critical for recommendation

Special considerations

Innovation acknowledged