Cannabidiol for treating seizures caused by tuberous sclerosis complex

3.1 Tuberous sclerosis complex is a rare genetic disorder characterised by growth of noncancerous tumours (known as tubers). Tubers can form in many parts of the body, but most commonly occur in the brain, eyes, kidneys, heart, lungs and skin. Seizures resulting from tubers that have formed in the brain affect up to 85% of people with the condition. Seizures usually start by the age of 2 years as focal onset seizures (which begin in 1 side of the brain) but may progress to generalised seizures involving both sides of the brain. Clinical experts explained that there is high variability in seizure type, severity and frequency depending on where the tubers are located. Any seizure type is possible (with or without the person retaining consciousness). The patient experts explained that seizures can be traumatic for people with tuberous sclerosis complex and their families. Drop seizures, in which people lose muscle tone or muscles stiffen, are particularly dangerous and are associated with a risk of injury as people crash to the ground. The clinical experts also highlighted that tubers that form in the brain can cause a range of cognitive, behavioural, and psychiatric manifestations (known as tuberous sclerosis complex-associated neuropsychiatric disorders or TAND) in around half of people with seizures caused by tuberous sclerosis complex. Patient experts highlighted the significant quality-of-life impact from TAND for patients and families. Severe learning difficulties occur in around 30% of people with tuberous sclerosis complex and may impede speech. Impaired movement may also limit all aspects of daily living. Poor behaviour including anger, mood swings, aggression and a lack of perception of risk often makes daily activities impossible and people may need round-the-clock care. People with the condition also have disrupted sleep, which can impact on the mental health of the entire family. The committee concluded that seizures caused by tuberous sclerosis complex severely affect the quality of life of patients, families and carers.

3.2 The clinical experts explained that it is difficult to control seizures associated with tuberous sclerosis complex. They highlighted that the aim of treatment is seizure freedom, but this is rarely achieved in this population. This is because of the complexity and heterogeneity of the underlying disease and variability of tuber locations in the brain. Also, it is harder to control seizures in people with moderate to severe learning difficulties. At the second committee meeting, the patient and clinical experts explained that reducing the frequency of seizures is important in improving quality of life for people with tuberous sclerosis complex, their families and carers. Fewer, less severe and more predictable seizures allow people with severe forms of the disease to leave the house, carry out daily activities and attend mainstream school. Indirectly, fewer seizures can also improve behaviour, reduce anxiety and decrease the risk of sudden unexpected death in epilepsy (SUDEP). Clinical experts explained that the treatment pathway for treating seizures associated with tuberous sclerosis complex is complicated. However, resective surgery should always be considered in people with structural brain lesions. Because there is a long waiting list for surgery, and a limited number of centres providing it, combinations of other treatments are often used. First-line treatment is monotherapy with an antiseizure medication. If seizures remain uncontrolled, a range of antiseizure medications are added. A clinical expert explained that a person's condition would be classed as refractory to treatment if their seizures are inadequately controlled by 2 or more antiseizure medications at a therapeutic dose. For these people, further adjunctive antiseizure medications and non-pharmacological treatments (ketogenic diet and vagus nerve stimulation) are used. Everolimus is also commissioned under an NHS clinical commissioning policy for treating refractory focal onset seizures associated with tuberous sclerosis complex in people aged 2 years and above, when surgery and vagus nerve stimulation has failed or is not considered appropriate. A clinical expert stressed that there is an unmet need for treatments to control seizures and behavioural problems associated with the condition. This is because the range of current antiseizure medications do not control seizures for long and are often associated with side effects. The committee concluded that people with tuberous sclerosis complex and their carers would value a new treatment option.

3.3 The scoped positioning for cannabidiol was as an adjunctive therapy for people whose seizures are inadequately controlled by established clinical management. In its submission, the company updated the population to include people in whom usual care is not tolerated. This was based on the International League Against Epilepsy's definition of refractory epilepsy as ʻfailure of adequate trials of 2 tolerated and appropriately chosen' antiseizure medications to achieve sustained seizure freedom. The committee noted that the company's population was narrower than the marketing authorisation which specified only that cannabidiol should be used adjunctively. The ERG was concerned that usual care may differ for people when antiseizure medications are not tolerated rather than when they are ineffective. This is because antiseizure medications can be continued adjunctively if seizures remain inadequately controlled, but are stopped if not tolerated. The ERG also highlighted the possibility that everolimus may form usual care in the company's updated population. The clinical experts explained that everolimus has a different indication to cannabidiol because of its mode of action. Everolimus treats the underlying cause of tuberous sclerosis complex. It has also been shown to reduce the size of existing tumours and slow the formation of new tuberous sclerosis complex-related tumours including astrocytomas (tumours arising from the glial cells in the brain), kidney and lung lesions, and skin conditions. The committee noted that everolimus requires close monitoring and can only be used in people with focal seizures after consideration of surgery and vagus nerve stimulation. So, the population who would have everolimus at the same point in the pathway as cannabidiol was likely to be very small. The committee agreed that usual care including antiseizure medications with or without surgery or vagus nerve stimulation was the appropriate comparator.

3.4 Cannabidiol (plus usual care) has been compared with placebo (plus usual care) in the randomised controlled trial, GWPCARE6. In this, 2 maintenance doses of cannabidiol (25 mg/kg/day [n=75]; and 50 mg/kg/day [n=73]) were compared with placebo (n=76). However, because the marketing authorisation for cannabidiol is granted for a maximum 25 mg/kg/day, the company did not present results from the 50 mg/kg/day dose in its submission. GWPCARE6 had a dose titration period of 4 weeks, followed by 12 weeks at the maintenance dose. It included people aged 1 year to 65 years, 7 of whom were from the UK. A clinical expert stated that the baseline characteristics and usual care treatments in GWPCARE6 were similar to those expected in UK clinical practice. However, the committee noted differences between arms in the amount of people who had vigabatrin and clobazam (2 of the most common antiseizure medications). A clinical expert stated that vigabatrin is unlikely to impact the effectiveness of cannabidiol because it treats infantile spasms as opposed to seizures. However, both patient and clinical experts reported that clobazam increases the effectiveness of cannabidiol when taken adjunctively. So, the GWPCARE6 results may not be generalisable to UK clinical practice if clobazam use differs between the trial and the NHS. The committee noted that cannabidiol's marketing authorisation mandates use with clobazam for Lennox–Gastaut and Dravet syndromes, but not for tuberous sclerosis complex. A clinical expert explained that clobazam is currently used in NHS practice in a similar way to GWPCARE6. Also, its use would not be expected to increase if cannabidiol was recommended. This is because clobazam is associated with side effects so its risks and benefits would be evaluated on an individual basis and dose adjustments of both drugs may be needed. The company also submitted data from the following ongoing studies to support cannabidiol's long-term effect:

3.5 The primary endpoint for GWPCARE6 was the percentage change in tuberous sclerosis complex-associated seizures during the treatment period (4‑week titration plus 12‑week maintenance periods). For people taking 25 mg/kg/day cannabidiol with usual care, there was a statistically significant reduction in tuberous sclerosis complex-associated seizures compared with placebo with usual care (30% reduction, 95% confidence interval 14% to 43%, p<0.001). There were 5% of people having cannabidiol 25 mg/kg/day who were seizure-free throughout the maintenance period compared with 0% taking placebo. People in the cannabidiol arm also experienced an additional 2.8 seizure-free days compared with placebo. At consultation, the company provided 3‑year follow-up data from the open-label extension study. For people still having treatment at 3 years, seizure reductions of 50% and over, 75% and over and 100% were observed in 78%, 69% and 31% of people, respectively. At least 6% of people taking cannabidiol were seizure-free at any given time during the follow-up period. The committee noted that only people whose disease showed some level of response would continue cannabidiol in clinical practice. Also, the primary outcome of the open-label study was a safety endpoint and data on seizure-free days had not been collected. However, it acknowledged that the 3‑year data from the extension study suggested ongoing seizure control in people who continued treatment. Clinical experts confirmed that, based on their experience of people with refractory epilepsy taking cannabidiol, the treatment effect did not wane over time. The committee recalled that both seizure freedom and reducing the number of seizures are important outcomes to patients and their families (see section 3.2). However, it noted that GWPCARE6 did not show an association between reducing seizure frequency and improving quality of life (as assessed by the Quality of Life in Childhood Epilepsy (QOLCE) and Quality of Life in Epilepsy [QOLIE]-31-P questionnaires). However, the clinical experts suggested this was because the 16‑week follow-up period in GWPCARE6 was insufficient to fully capture the benefits of reducing or stopping seizures on daily activities and mental health. They explained that, in their experience, seizure reduction did align with improved quality of life. The committee remained concerned that, by pooling usual care treatments, cannabidiol had not been compared with individual antiseizure medications. However, it concluded cannabidiol plus usual care likely reduces seizure frequency and, at least in the short term, increases the number of seizure-free days compared with usual care alone.

3.6 GWPCARE6 showed that a large proportion of patients having cannabidiol had adverse events. The most common adverse events in this group were diarrhoea, decreased appetite and somnolence (drowsiness). A clinical expert noted that people with tuberous sclerosis complex often experience adverse effects from their medications. They also noted that cannabidiol's adverse effects are mostly, but not always, mild and tolerated. They stated that the choice of treatment depends on the balance of its safety and tolerability, with adverse events representing an important consideration. The committee also recalled that cannabis-based medicines can increase the exposure of other antiseizure medications, so dose modifications for concomitant medications may be needed in clinical practice. It agreed that, while cannabidiol's adverse effects are mostly manageable, they are an important consideration when making decisions about whether to start or continue treatment. The committee concluded that cannabidiol is associated with adverse events that are manageable but monitoring and adjustment of concurrent antiseizure medications may be needed.

3.7 Cannabidiol's marketing authorisation does not specify a stopping rule, that is, stopping treatment if or when it does not work. However, the company proposed that cannabidiol should be stopped if the frequency of seizures caused by tuberous sclerosis complex does not reduce by 30% from baseline. Baseline seizure frequency in the model was taken from the 28‑day baseline period in GWPCARE6, assumed to represent the 6 months before starting treatment which would be used in clinical practice. The company implemented this stopping rule every 6 months in its model. The committee noted that this aligned with the stopping rule in NICE's technology appraisal guidance on cannabidiol with clobazam for treating seizures associated with Dravet syndrome and cannabidiol with clobazam for treating seizures associated with Lennox–Gastaut syndrome. However, the company had implemented an additional stopping rate for cannabidiol in tuberous sclerosis complex. This was based on stopping rates in GWPCARE6 in the short term, the open-label extension study in the medium term and NICE's guidance on cannabidiol for Lennox–Gastaut syndrome in the long term. The committee concluded that this was acceptable, and it was appropriate to assess response to treatment every 6 months and stop cannabidiol if it is not effective.

3.8 The company presented a cohort-level model to estimate the cost effectiveness of cannabidiol. It had 3 health states: alive having cannabidiol with usual care, alive having usual care alone, and dead. The alive health states were further divided by the number of seizures per week (seizure-free, 2 and under, over 2 to 7, and over 7) and number of seizures per day (seizure-free, 1 and under, over 1 to 2, over 2 to 4, and over 4). The model used a lifetime horizon and a cycle length of 1 week. The company did not include vagus nerve stimulation, ketogenic diet or surgery as comparators in its model because it stated these would be used equally between arms. So, the comparator was usual care with antiseizure medications. The committee concluded that the company's general model structure was appropriate.

3.9 The company derived efficacy data for the sub-health states for number of seizures per week and number of seizures per day from GWPCARE6 at week 16. For the long-term effect on seizures, the company applied 2 linear regression models sequentially. The first predicted the number of seizure-free days per 7‑day cycle in the model using a binomial regression model based on individual patient data from GWPCARE6 at week 16. The second, a negative binomial model, predicted the total seizure frequency on non-seizure-free days per cycle. The company chose this approach to capture correlation between seizure frequency and seizure-free days. It applied the regression coefficients to each patient's baseline seizure frequency to predict outcomes for each 7‑day model cycle. Using this approach, the relative treatment effect was maintained over the modelled time horizon. Results showed a trend for reduced seizure frequency and increased odds of seizure-free days for cannabidiol plus usual care compared with placebo plus usual care. The committee noted that the relative treatment effect was not statistically significant. The company stated this was because the powers of the analyses were low because of use of weekly model cycles and 2 different regression models. The committee noted that the seizure frequency per day predicted by the company's regression model broadly aligned with data from the open-label extension study, but the study did not collect data on seizure-free days. It acknowledged the uncertainty in the company's approach but noted that the ERG had not suggested an alternative method for modelling seizures. Given this, it considered analyses using the company's linear regression models for seizure-free days and seizure frequency were acceptable for decision making.

3.10 At the first committee meeting, the ERG was concerned that the company had modelled seizure freedom over a 7‑day cycle by using the proportion of people who were seizure-free over 6.5 days. This was because a negative binomial regression model cannot predict discrete outcomes (that is, seizure-free over 0 days or 7 days). Using this approach, 0% of the placebo and 11% of the cannabidiol arm were seizure-free at 10 weeks (the latter reduced as people stopped treatment over time). The ERG highlighted that using a cut-off of 6.5 days would overestimate the number of people having cannabidiol who were seizure-free per cycle. This is because the predicted number of seizure-free days using the linear regression model results was 6.62 for cannabidiol compared with 5.89 for placebo. At technical engagement, the company provided a scenario using a cut-off of 6.61 days to model the number of people per cycle who were seizure-free to better align with the regression model results. The committee also considered a scenario provided by the ERG using a cut-off of 7 days, in which no one was seizure‑free over the course of a week in either arm but still gained a benefit for cannabidiol in other seizure frequency categories. The committee recalled that seizure-free days per week had not been collected as an outcome in GWPCARE6 so could not directly inform the modelling. However, it compared the number of people who were seizure-free over 7 days using different cut-offs in the model with that expected in clinical practice to determine the most clinically plausible approach. It recalled that achieving seizure freedom is challenging in this population (see section 3.2). A clinical expert agreed it was plausible that a small number of people taking cannabidiol may achieve seizure freedom but that this would be unlikely in people only having usual care. This was supported by data from GWPCARE6 in which 5% of people taking cannabidiol and 0% of people taking placebo were seizure-free during the maintenance period. So, the ERG's scenario using a 7‑day cut-off was likely conservative. This also aligned with testimonies from patients who had taken cannabidiol. The committee recognised that the number of seizure-free days is a useful outcome for determining response to treatment. However, it was concerned that measuring this outcome over the course of a week introduced artificial effects into the model that favoured cannabidiol or usual care depending on which cut-off was chosen. It would have preferred the company to use a longer time-period, as in NICE's technology appraisal guidance on fenfluramine for treating seizures associated with Dravet syndrome which modelled the number of seizure-free days in a month. The committee agreed that the proportion of people who are seizure-free over 7 days in the model was uncertain because the company's methodology was problematic. It recalled that a 6.5‑day cut-off likely overestimated the proportion who were seizure-free per week in clinical practice while a 7‑day cut-off underestimated this. So, it concluded that a 6.61‑day cut-off was most appropriate for decision making.

3.11 The summary of product characteristics for cannabidiol states that the dose can be increased from a maintenance dose of 10 mg/kg/day to a maximum of 25 mg/kg/day based on response. Yet, the company assumed an average dose of 12 mg/kg/day in its base case. This was based on clinical expert advice to the company and German dispensing data from 118 people with tuberous sclerosis complex, in which the median cannabidiol dose was 12.2 mg/kg/day in children and 7.8 mg/kg/day in adults. The committee noted that the European regulator mandates a maintenance dose of cannabidiol of 10 mg/kg/day for Lennox–Gastaut and Dravet syndromes and there is no evidence that increasing the doses above this results in a different treatment effect. It also noted data from the open-label extension study that suggested a maintenance dose above 15 mg/kg/day did not improve response. The ERG was concerned that a maintenance dose of 12 mg/kg/day was not verified by clinical trial data (because treatment was titrated up to 25 mg/kg/day or 50 mg/kg/day in GWPCARE6 regardless of response) and that the proportion of people needing a higher dose in the open-label extension study was unknown. It presented scenarios varying the average dose of cannabidiol. At the second meeting, clinical experts provided dosing information based on clinical experience of treating people with tuberous sclerosis complex in clinical practice (including people who had concurrent Lennox–Gastaut syndrome) in Scotland and Wales. They stated that most people would have a dose about 12 mg/kg/day in clinical practice. This was normally lower in adults, for whom 1 clinical expert estimated an average dose of 8 mg/kg/day. This was because, although there is potential that higher doses may be more effective, they cause more side effects than lower doses. So, 12 mg/kg/day is likely to be the average dose in clinical practice as a trade-off between efficacy and side effects. The clinical experts acknowledged a dose–response effect for cannabidiol but noted that, in their experience, they had observed similar effectiveness using lower doses to those reported for the trial dose of 25 mg/kg/day. This was because people with refractory epilepsy are taking other antiseizure medications that may increase the effectiveness of cannabidiol when used in combination (see section 3.6). Dose titration would also be much slower in clinical practice, allowing clinicians to establish response at a lower dose than used in the trial. However, the committee remained concerned that the efficacy data in the model came from people having 25 mg/kg/day in GWPCARE6 and it had no credible evidence that efficacy at half that dose would be identical. It also noted that a 15 mg/kg/day average dose for cannabidiol was accepted in NICE's technology appraisal guidance on fenfluramine for treating seizures associated with Dravet syndrome. It therefore considered that, although 12 mg/kg/day was likely the average dose used in clinical practice, some adjustment for loss of benefit compared to GWPCARE6 results should be included in the modelling. So, it preferred an average dose of 15 mg/kg/day for decision making.

3.12 Annual healthcare resource use based on seizure frequency for people with tuberous sclerosis complex-associated epilepsy was sourced from a 2‑round Delphi panel of clinical experts. Separate resource use estimates were produced for adults and children. The company validated these estimates using a study by Shepherd et al. (2017). This reported a cost of £44,259 over a period of 3 years for resource use (GP visits, hospitalisation, other drugs and outpatient visits) in UK patients with tuberous sclerosis complex. The committee noted that the costs in the company's model for the same 3‑year period for usual care (£55,578) were significantly higher than those in Shepherd et al. and NICE's technology appraisal guidance on cannabidiol with clobazam for Dravet syndrome and cannabidiol with clobazam for Lennox–Gastaut syndrome. The ERG modelled a scenario applying the resource use costs from these appraisals. The company explained that the discrepancy was because of inflation rates in NHS reference costs (which it updated at consultation to align with inflation). The ERG also highlighted that hospitalisation costs and number of hospital days (both in a general ward and intensive care) were higher in the company's model for tuberous sclerosis complex than in the appraisals for other cannabidiol indications and costs reported in Shepherd et al. In the second committee meeting the company explained that there is a spectrum of severity in tuberous sclerosis complex. Although many people do not need frequent hospitalisation, a small proportion are associated with high resource use from having up to 40 hospitalisations per year. It defended the use of the Delphi panel for sourcing expert opinion, noting the methodology was more robust than the interviews with clinicians in the appraisals for Dravet and Lennox–Gastaut syndromes. Clinical experts explained that although tuberous sclerosis complex is a multi-organ disease, non-epilepsy-related hospital admissions are rare. So, hospitalisation rates are likely comparable to other cannabidiol indications, and may be lower than for Dravet syndrome, which causes very frequent seizures. This was supported by more frequent use of rescue medication by people with Dravet syndrome compared to the other cannabidiol appraisals. The ERG's base case after consultation reduced hospitalisation rates by 50% to better align with the other cannabidiol appraisals (the values from the Dravet and Lennox–Gastaut syndrome appraisals could not be used because seizure frequency categories differed across indications). At the second committee meeting, the committee also considered additional scenarios reducing hospitalisation rates from those in the Delphi panel by 10%, 20%, 30% and 40% provided by the company and by 25% provided by the ERG. It noted that the hospitalisation rates in scenarios were not underpinned by evidence, so the true resource use for people with tuberous sclerosis complex was unknown. However, the committee agreed it was appropriate to use the scenario that best aligned with the hospitalisation rates used in the technology appraisals for Dravet and Lennox–Gastaut syndromes. It concluded that healthcare resource use is likely overestimated in the company's model and a 50% reduction from the Delphi panel hospitalisation rates should be used for decision making.

3.13 The committee recalled that the company had collected data from responses to the QOLCE and QOLIE‑31‑P questionnaires in GWPCARE6, but did not use the data in its model. This was because, during the trial, the company considered that the questions were inappropriate for assessing people with severe learning difficulties. The company also noted that data on quality of life in the literature does not consider health states and substates used in the company's model (that is, number of seizures and seizure-free days). So, the company instead asked members of the general public to estimate the quality of life associated with each health state and substate in the model. Respondents were asked to consider 'vignettes', that is, descriptions of each health state and, using a time trade-off method, give each a value. Values ranged between less than 0 (worse than death), 0 (death) and 1 (perfect health). The company considered the quality-of-life values it used in its model to be confidential. The committee understood why the company had chosen to use a vignette study given the lack of literature utility values. However, it noted a discrepancy in the health state utilities used for tuberous sclerosis complex compared with those for Lennox–Gastaut and Dravet syndromes. This was especially so for the values published in Lo et al. (2021) which used vignettes from the general public (collected after the publication of guidance on cannabidiol for Lennox–Gastaut and Dravet syndromes to align with the committee's preferred assumptions in these appraisals). The company explained that, in part, this might be because of the difference in seizure severity categories and types of seizures in the other appraisals. The vignettes were considered by the general public who may find it hard to quantify the difference between the worst health state for Lennox–Gastaut syndrome (more than 110 drop seizures per month) compared with that for tuberous sclerosis complex (more than 4 seizures per day). A clinical expert stated that, although the epilepsy aspect of the 3 conditions may be similar, it is difficult to compare the severity of the indications. This is because tuberous sclerosis complex is a multi-organ disease associated with complications such as those in the skin, eyes and lungs that affect quality of life and are not present in the other conditions. So, correcting seizures will only partly correct the quality-of-life decrement associated with severe tuberous sclerosis complex. The committee was concerned that this was not captured in the company's model because the utility values for the seizure-free health state for both people with the condition and their carers was relatively high. It also noted that the health state utility values used in NICE's technology appraisal guidance on fenfluramine for treating seizures associated with Dravet syndrome encompassed a much narrower range than those used in the company's modelling of cannabidiol for tuberous sclerosis complex. In the fenfluramine appraisal, values were derived using empirical measurements collected in clinical trials, which had been mapped to EQ‑5D values in a way that matches the NICE reference case. The committee noted that it had not been presented with scenarios that used alternative health state utilities for cannabidiol, so could not determine the impact on the cost-effectiveness estimates. Given the lack of utility values in the literature, it would have liked to see data supporting the plausibility of the vignette values, especially for the seizure-free and most severe health states. However, acknowledging the high level of uncertainty, the committee considered the analyses using the company's utility values in its decision making. It concluded that the company's health state utilities are uncertain and differ from utilities in other cannabidiol appraisals.

3.14 The committee recalled that caring for someone with tuberous sclerosis complex affects carers' quality of life (see section 3.1), and that capturing this in the model is appropriate. The company included utility decrements in its model for seizure frequency health states (number of seizures per day) based on the vignette study. This was because carer quality of life had not been measured using a preference-based approach in GWPCARE6, so utilities could not be derived directly from the trial. In its base case at the first committee meeting, the company assumed that people with tuberous sclerosis complex would have a cumulative 2 carers (1 primary caregiver plus others such as partners, friends and family). This was based on a paper by Lagae et al. (2019), which reported that people with Dravet syndrome had an average 2.06 carers. The clinical experts explained that there are no significant differences in the burden of care for seizures between cannabidiol indications. However, the total burden of care for people with tuberous sclerosis complex may be higher because of the combined effect of the epilepsy, associated learning disabilities and other comorbidities that occur with the condition. The ERG was concerned that the carer disutilities were applied additively in the company's model. It used 1.8 carers in its base case, which aligned with the committee's preferred assumptions in NICE's technology appraisal guidance for Dravet syndrome and Lennox–Gastaut syndrome. The committee agreed that the company's approach implied that the caring burden increases linearly the more carers a patient has. However, for a patient with multiple carers, it expected there to be less effect on the quality of life of each carer because they would 'share' some of the burden. At consultation, the patient organisations highlighted that care given by 2 carers does not reduce burden of care. The company supported this, stating that each carer will have worry and psychological distress from caring for a patient at risk from injury and death because of seizures. However, it updated its modelling to align with the committee's preference at the first meeting of 1.8 carers. This was because, while the total burden for 2 carers may be greater than the burden for a sole carer, the committee agreed that it would likely not be 2 times greater. It recalled that people with tuberous sclerosis complex may need help with every aspect of daily life (see section 3.1). However, given the multifactorial nature of the disease and associated care needs of people living with it, reducing seizure frequency would only go some way towards relieving carer burden. The committee recalled that carer health state utility values could not be derived directly from the clinical trial and that the values from the company's vignette were uncertain (see section 3.13). Given this uncertainty, it agreed that the model should use 1.8 carers to align with assumptions used in previous appraisals of cannabidiol.

3.15 The vignette calculated the impact of worsening seizure frequencies on carers as an incremental decrease from the carer utility for the seizure-free health state. After technical engagement, the company updated its base case to use a utility value for the seizure-free health state for carers equal to that of an average 45‑year‑old. This was because in the vignettes, members of the general public were asked to adopt the perspective of the primary caregiver of a 13‑year‑old child and the average parent of a child this age was 45 years (Office for National Statistics data). However, the ERG was concerned that the company's value overestimated the impact of seizure freedom on carers. This was because it had not adjusted for carers ageing and that the burden of care for non-seizure-related aspects of tuberous sclerosis complex may not be captured in the company's estimate. At consultation, the ERG also highlighted that carer utility was unlikely to adjust the moment a patient becomes seizure-free. It provided scenarios with lower utility values for the seizure-free health state for carers. The committee considered these scenarios in its decision making. However, it noted that the absolute utility decrement was unchanged regardless of the value used for seizure freedom. Also, the ERG had maintained the value of 0.881 in its preferred assumption after consultation. It concluded that the seizure-free health state utility for carers is uncertain but the utility value in the company and ERG's base cases is appropriate for decision making.

3.16 Based on the consensus in the Delphi panel, the company assumed that 31% of people with tuberous sclerosis complex-associated epilepsy were institutionalised at an average age of 27 years. However, in its base case it did not model any impact on carer utility during this period. The company stated that carers for people who were institutionalised would still have:

3.17 The company included a cost in the model associated with managing TAND aspects. It also assumed a benefit on TAND aspects for people whose condition responds to cannabidiol, which it modelled as a utility benefit and a reduced cost. The utility benefit for delaying TAND was estimated by weighting utility increments for common TAND aspects from various literature sources by their prevalence using data from the tuberous sclerosis registry to increase disease awareness (TOSCA; de Vries et al. 2015). 'Responders' to cannabidiol were defined as the proportion of people with a reduction in tuberous sclerosis complex-associated seizures of 50% or more from baseline over 6 months. Based on the Delphi panel consensus that treatment with cannabidiol is likely to be most beneficial at an early age, only people aged 2 years to 6 years accrued costs and benefits associated with TAND aspects. The ERG had the following concerns about the company's assumption for TAND:

3.18 NICE's guide to the methods of technology appraisal notes that above a most plausible incremental cost-effectiveness ratio (ICER) of £20,000 per quality-adjusted life-year (QALY) gained, judgements about the acceptability of a technology as an effective use of NHS resources will take into account the degree of certainty around the ICER. The committee will be more cautious about recommending a technology if it is less certain about the ICERs presented. Because of confidential commercial arrangements for cannabidiol and comparators, none of the cost-effectiveness results are reported here. The committee noted the differences between the ERG's and company's base case after consultation in that the ERG used a 15 mg/kg/day dose instead of 12 mg/kg/day and reduced hospitalisation rates by 50% compared with those from the Delphi panel. The company base case was under £20,000 per QALY gained compared with usual care. However, the ERG base case was over £30,000 per QALY gained compared with usual care. This was when considering confidential discounts. The committee agreed that its preferred assumptions to compare cannabidiol with usual care aligned with the ERG base case, that is, they included:

3.19 The committee noted NICE's principles for the development of NICE guidance. This emphasises the importance of considering the distribution of health resources fairly within society as a whole, and factors other than relative costs and benefits alone. It recalled that tuberous sclerosis complex is a severe and multisystemic disease that significantly affects the quality of life of patients, carers and families (see section 3.1). It agreed that the severity of disease should be considered in the decision making. At the second committee meeting, the committee also noted that the following benefits of cannabidiol were not captured in the company's model:

3.20 The clinical experts stated that they would welcome an additional treatment option for tuberous sclerosis complex. However, they considered that cannabidiol represents only a modest change when managing tuberous sclerosis complex because few people became seizure‑free (see section 3.10). The committee concluded that cannabidiol did not meet the criteria for an innovative treatment.

3.21 The committee noted that the population for which cannabidiol is indicated includes children and young people. It discussed the need to balance the importance of improving the lives of children and their families with fairness to people of all ages. The committee also noted that half of people with tuberous sclerosis complex have associated learning difficulties, which is a protected characteristic under the Equality Act 2010. At the second committee meeting, the clinical experts highlighted that access to care for people with learning difficulties may be harder. Also, tuberous sclerosis complex is a hereditary disease, so one or more parents might also have the condition, increasing the challenge in accessing appropriate healthcare. The committee acknowledged and considered the nature of the population as part of its decision making.

3.22 The committee recalled its conclusion to consider factors other than relative costs and benefits of cannabidiol alone (see section 3.19). It acknowledged that the cost-effectiveness estimates were uncertain for cannabidiol because of some of the assumptions in the company's model (see section 3.18). However, the committee concluded that, despite these uncertainties and considering the uncaptured benefits and severity of the disease, cannabidiol is an effective treatment and a good use of NHS resources. It therefore recommended cannabidiol for treating seizures caused by tuberous sclerosis complex. It agreed that seizure frequency should be checked every 6 months and that, if the frequency has not fallen by at least 30% compared with the 6 months before starting treatment, cannabidiol should be stopped (see section 3.7).