NATIONAL INSTITUTE FOR HEALTH AND CARE EXCELLENCE

SINGLE TECHNOLOGY APPRAISAL (STA)

Cannabidiol for treating seizures caused by tuberous sclerosis complex [ID1416]

Appraisal Committee Meeting – 15 September 2022 1[st] Committee meeting

The final scope and final stakeholder list are available on the NICE website.

Pre-technical engagement documents

1. Company submission summary from GW Pharmaceuticals

2. Clarification questions and company responses

3. Patient group, professional group and NHS organisation submissions from:

• a. Tuberous Sclerosis Association

• b. Association of British Neurologists

4. Evidence Review Group report – factual accuracy check

Post-technical engagement documents

5. Technical engagement response from company

6. Technical engagement responses and statements from experts:

• a. Dr Pooja Takhar – clinical expert, nominated by Tuberous Sclerosis Association

• b. Lisa Suchet – patient expert, Tuberous Sclerosis Association

7. Appraisal Committee Meeting presentation slides

Please note that the full submission, appendices to the company’s submission and company model will be available as a separate file on NICE Docs for information only.

NATIONAL INSTITUTE FOR HEALTH AND CARE EXCELLENCE

Single technology appraisal

Cannabidiol for treating seizures caused by tuberous sclerosis complex [ID1416]

Document A

Company evidence submission summary for committee

GW Research Ltd confirm that all information in the submission summary is an accurate summary or replication of evidence in the main submission and accompanying appendices and that wherever possible a cross reference to the original source is provided.

June 2022

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416]

© GW Research Ltd. (2022). All rights reserved

1 of 44

Instructions for companies

This is the template you should use to summarise your evidence submission to the National Institute for Health and Care Excellence (NICE) as part of the single technology appraisal (STA) process. This document will provide the appraisal committee with an overview of the important aspects of your submission for decisionmaking.

This submission summary must not be longer than 25 pages, excluding the pages covered by this template. If it is too long it will not be accepted. Please submit a draft summary with your main evidence submission. The NICE technical team may request changes later.

When cross referring to evidence in the main submission or appendices, please use the following format: Document, heading, subheading (page X).

For all figures and tables in this summary that have been replicated, cross refer to the evidence from the main submission or appendices in the caption in the following format: Table/figure name – document, heading, subheading (page X).

Companies making evidence submissions to NICE should also refer to the NICE guide to the methods of technology appraisal and the NICE guide to the processes of technology appraisal.

Highlighting in the template (excluding the contents list)

Square brackets and grey highlighting are used in this template to indicate text that should be replaced with your own text or deleted. These are set up as form fields, so to replace the prompt text in [grey highlighting] with your own text, click anywhere within the highlighted text and type. Your text will overwrite the highlighted section.

To delete grey highlighted text, click anywhere within the text and press DELETE.

Grey highlighted text in the footer does not work as an automatic form field but serves the same purpose – as prompt text to show where you need to fill in relevant details. Replace the text highlighted in [grey] in the header and footer with appropriate text. (To change the header and footer, double click over the header or

footer text. Double click back in the main body text when you have finished.) Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416] © GW Research Ltd. (2022). All rights reserved

2 of 44

Contents

Instructions for companies ......................................................................................... 2 Tables and figures ...................................................................................................... 4 Abbreviations ............................................................................................................. 5 Submission summary ................................................................................................. 6 Health condition ............................................................................................... 6 Clinical pathway of care ................................................................................... 7 Equality considerations .................................................................................... 7 The technology ................................................................................................ 8 Decision problem and NICE reference case .................................................. 10 Clinical effectiveness evidence ...................................................................... 14 Key results of the clinical effectiveness evidence .......................................... 16 Evidence synthesis ........................................................................................ 19 Key clinical issues ......................................................................................... 19 Overview of the economic analysis ............................................................... 20 Incorporating clinical evidence into the model ............................................... 24 Key model assumptions and inputs ............................................................... 28 Base case ICER (deterministic) ..................................................................... 32 Probabilistic sensitivity analysis ..................................................................... 34 Key sensitivity and scenario analyses ........................................................... 35 Innovation ...................................................................................................... 37 Disease severity modifier criteria ................................................................... 38 Budget impact ................................................................................................ 38 Interpretation and conclusions of the evidence.............................................. 41 References .................................................................................................... 43

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416] © GW Research Ltd. (2022). All rights reserved

3 of 44

Tables and figures

Table 1: Technology being appraised – B.1.2 (page 11) ............................................ 8 Table 2: The decision problem – B.1.1 (page 9) ...................................................... 10 Table 3: Clinical effectiveness evidence ................................................................... 14 Table 4: Overview of the model approach ................................................................ 20 Table 5: Key model assumptions and inputs ............................................................ 29 Table 6: Base case results (deterministic) – B.3.8 (Table 31, page 152) ................. 33 Table 7: Base case results (deterministic), including disease severity modifier – B.3.8 (Table 32, page 152) ................................................................................................ 33 Table 8: Base case results (probabilistic) – B.3.9 (Table 36, page 156) .................. 34 Table 9: Key scenario analyses – B.3.9 (Table 37, page 160) ................................. 36 Table 10: Decision modifier – severity criteria .......................................................... 38 Table 11: Budget impact – company BIA document ................................................ 39

Figure 1: Clinical pathway for TSC-associated seizures including cannabidiol .......... 7 Figure 2: Model process diagram – B.3.2.2 (page 73) ............................................. 22 Figure 3: Seizure-free days proportions for all observed cycles – B.3.3.2 (page 91) 25 Figure 4: Seizure frequency on seizure days for all observed cycles – B.3.3.2 (page 91) ............................................................................................................................ 25 Figure 5: Observed and estimated seizure frequency on observed and estimated seizure days during the OLE – B.3.3.2 (page 93) .................................................... 26 Figure 6: Scatterplot of probabilistic results – B.3.9 (Figure 20, page 156) .............. 34 Figure 7: Tornado diagram – B.3.9 (Figure 22, page 158) ....................................... 35

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416] © GW Research Ltd. (2022). All rights reserved

4 of 44

Abbreviations

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416] © GW Research Ltd. (2022). All rights reserved

5 of 44

Submission summary

==> picture [30 x 13] intentionally omitted <==

Health condition

Tuberous sclerosis complex (TSC) is a rare (orphan; prevalence 1 in 18,861), autosomal dominant, multisystemic disorder characterized by the formation of benign tumours (known as hamartomas) in multiple organ systems, most notably the brain, skin, kidneys, lungs and heart. TSC leads to severe, often debilitating neurological disorders, including epilepsy, which is experienced by around 80% of patients.

TSC-associated epilepsy is a devastating and life-threatening form of epilepsy that presents early in childhood and is associated with refractory seizures and poor outcomes. In addition to the high seizure burden, there are cognitive and behavioural difficulties collectively known as TAND (TSC-associated neuropsychiatric disorders) that prevent children from achieving independence in adult life. This has a profound impact on the quality of life experienced not only by the patients but also by their families and carers. Mortality rates are higher than in the general population. Uncontrolled epilepsy is among the most common causes of death in TSC as a result of status epilepticus or sudden unexpected death in epilepsy (SUDEP).

Despite the availability of a broad range of anti-epileptic drugs (AEDs), nonpharmacological interventions and invasive resective surgery (see Figure 1), up to two-thirds of patients with TSC-associated epilepsy currently do not achieve seizure control and continue to be at risk of hospitalization and death. Patients entering the Epidyolex[®] GWPCARE6 trial had already tried and discontinued up to 15 AEDs (median: four), with some taking up to five AEDs concurrently (median: three).

There remains a substantial unmet need for a well-tolerated treatment to provide early and effective seizure control and improve the overall condition of patients with TSC-associated epilepsy without markedly increasing adverse events.

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416] © GW Research Ltd. (2022). All rights reserved

6 of 44

==> picture [30 x 13] intentionally omitted <==

Clinical pathway of care

For patients with TSC-associated seizures considered for treatment with cannabidiol, it will be an add-on treatment for refractory seizures in people aged 2 years and older once two other appropriate AEDs, trialled to a maximally tolerated dose, have failed to achieve seizure freedom (Figure 1). This is in line with the International League Against Epilepsy’s (ILAE) definition of a refractory patient.

Figure 1: Clinical pathway for TSC-associated seizures including cannabidiol

==> picture [487 x 237] intentionally omitted <==

----- Start of picture text -----
Pharmacological therapy
First-line therapy
Vigabatrin
Non-pharmacological therapy
Adjunctive therapy
(e.g. topiramate, carbamazepine, Ketogenic diet
oxcarbazepine)
Subsequent adjunctive
therapies Vagus nerve stimulation
including cannabidiol Everolimus
≥2 years; only if surgery or VNS has
failed/is not suitable; refractory focal
onset seizures only (not licensed for
Resective surgery generalized onset seizures)
(In carefully selected cases; preceded by
a comprehensive presurgical evaluation;
only after failure of ≥2 AEDs)
----- End of picture text -----

Key: AED, anti-epileptic drug; TSC, tuberous sclerosis complex; VNS, vagus nerve stimulation.

==> picture [30 x 13] intentionally omitted <==

Equality considerations

The use of cannabidiol is unlikely to raise any equality issues. Patient age is defined in the indication: Epidyolex is indicated for use as adjunctive therapy for seizures associated with TSC in patients 2 years of age and older.

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416] © GW Research Ltd. (2022). All rights reserved

7 of 44

==> picture [30 x 13] intentionally omitted <==

The technology

Table 1: Technology being appraised – B.1.2 (page 11)

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416]

© GW Research Ltd. (2022). All rights reserved

8 of 44

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416] © GW Research Ltd. (2022). All rights reserved

9 of 44

==> picture [30 x 13] intentionally omitted <==

Decision problem and NICE reference case

The submission covers the technology’s full marketing authorization for this indication.

The company submission differs from the final NICE scope and the NICE reference case (see Table 2).

Table 2: The decision problem – B.1.1 (page 9)

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416]

© GW Research Ltd. (2022). All rights reserved

10 of 44

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416]

© GW Research Ltd. (2022). All rights reserved

11 of 44

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416]

© GW Research Ltd. (2022). All rights reserved

12 of 44

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416]

© GW Research Ltd. (2022). All rights reserved

13 of 44

==> picture [30 x 13] intentionally omitted <==

Clinical effectiveness evidence

Table 3: Clinical effectiveness evidence

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416] © GW Research Ltd. (2022). All rights reserved 14 of 44

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416] © GW Research Ltd. (2022). All rights reserved 15 of 44

==> picture [30 x 13] intentionally omitted <==

Key results of the clinical effectiveness evidence

A.7.1. Primary endpoint: change in number of TSC-associated seizures

In the GWPCARE6 randomised controlled trial, the ‘treatment period’ lasted 16 weeks: a 4-week ‘titration period’ followed by a 12-week ‘maintenance period’. The primary endpoint was the change in number of TSC-associated seizures during the treatment period compared to the baseline period (intention-to-treat analysis set).

As noted in Document B, Section B.2.6.1, the GWPCARE6 trial met its primary endpoint, demonstrating that cannabidiol had a statistically significant and clinically meaningful effect compared with placebo (in addition to usual-care) in the median percentage reduction from baseline in TSC-associated seizure frequency. The reduction in seizure frequency compared with baseline for the cannabidiol 25 mg/kg/day plus usual-care group was 49% vs 27% for placebo plus usual-care (p = 0.0009).

During the maintenance period only of GWPCARE6 (i.e. the 12-week stable dosing period after titration), cannabidiol 25 mg/kg/day plus usual-care demonstrated a 56% reduction in TSC-associated seizures vs 30% with placebo plus usual-care (p = 0.0004).

A.7.2. Key secondary endpoints

All key secondary endpoints in GWPCARE6 were supportive of the primary endpoint:

 Treatment responders (≥ 50% TSC-associated seizure reduction)

• 36% of patients taking cannabidiol (25 mg/kg/day) plus usual-care had ≥ 50% seizure reduction vs 22% with placebo plus usual-care (p = 0.0692)

 Treatment responders (≥ 75% TSC-associated seizure reduction)

• 16% of patients taking cannabidiol (25 mg/kg/day) plus usual-care had ≥ 75% seizure reduction vs 0% with placebo plus usual-care (p = 0.0003)

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416] © GW Research Ltd. (2022). All rights reserved

16 of 44

 TSC-associated seizure-free days

• Patients taking cannabidiol (25 mg/kg/day) plus usual-care demonstrated a nominally statistically significant percentage of improvement in TSC-associated seizure-free days. Patients taking cannabidiol experienced an additional 2.8 seizure-free days per month vs the placebo group (p = 0.0047)

 TSC-associated seizure-freedom and total seizure-freedom

• The patients in the GWPCARE6 trial were particularly treatment-resistant, having tried and failed a median of four AEDs prior to entering the trial and continuing to take a median of three AEDs throughout the trial. Given the highly refractory nature of the trial population, achieving seizure-freedom is a key result

• Despite the refractory nature of their epilepsy, in the maintenance period of the trial (the 12-week period when patients had completed titration and were on a stable dose), TSC-associated seizure-freedom was achieved in four of the 75 patients (5.4%) taking cannabidiol (25 mg/kg/day) plus usual-care compared with none of the 76 patients in the placebo plus usual-care group (p = 0.0354)

• In the treatment period (the 16-week period including a 4-week titration followed by 12 weeks on a stable dose), one patient in the cannabidiol (25 mg/kg/day) plus usual-care arm achieved TSC-associated seizure-freedom vs none in the placebo plus usual-care arm (p = 0.3173)

• Overall, one patient in the cannabidiol (25 mg/kg/day) plus usual-care group, experienced total seizure-freedom (i.e. no seizures of any type) during the treatment period, an important result in this highly refractory population

A.7.3. Quality of life - Subject/Caregiver Global Impression of Change score

The Subject/Caregiver Global Impression of Change (S/CGIC) is a patient-reported outcome (PRO) assessment completed by patients and caregivers. S/CGIC has been included in other Phase III studies for severe epilepsies and is an accepted measure of the patient’s overall condition as a proxy for quality of life.

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416] © GW Research Ltd. (2022). All rights reserved

17 of 44

Scores are not reported at baseline as S/CGIC is a measure of change. Prior to randomization, the patient or caregiver is asked to write a brief description of the patient’s overall condition as a memory aid. At the end of the treatment period, the patient/caregiver assesses the status of the patient’s overall condition (compared with before treatment) on a seven-point scale: ‘Very Much Improved’; ‘Much Improved’; ‘Slightly Improved’; ‘No Change’; ‘Slightly Worse’; ‘Much Worse’; and ‘Very Much Worse’.

S/CGIC represents a meaningful measure of health-related quality of life (HRQL) as it captures an estimate of the effect of treatment on the patient’s overall condition based on his/her entire seizure and co-morbidity burden, thereby providing valuable information on the clinical meaningfulness of the therapy.

Using this measure, patients and caregivers reported an improvement in patients’ overall condition in 69% of those receiving cannabidiol (25 mg/kg/day) plus usualcare vs 39% receiving placebo plus usual-care (Odds ratio [OR]: 2.25; nominal p = 0.0074).

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416] © GW Research Ltd. (2022). All rights reserved

18 of 44

==> picture [30 x 13] intentionally omitted <==

Evidence synthesis

No meta-analyses, indirect treatment comparisons or mixed treatment comparisons were conducted.

Refractory epilepsy (also known as drug-resistant epilepsy) has been defined by the ILAE as failure of adequate trials of two tolerated and appropriately chosen and used AED regimens (as monotherapies or in combination) to achieve sustained freedom from seizures. A high proportion of patients with TSC-associated seizures are refractory despite taking a variety of AEDs, reflecting the complexity of the condition and the fact that patients are resistant to or are unable to tolerate current AEDs.

In the Phase III clinical trial of cannabidiol in TSC-associated epilepsy, the intervention was cannabidiol in addition to usual-care and the comparator was usualcare without cannabidiol (i.e. usual-care plus placebo). For patients considered for treatment with Epidyolex, it will be an add-on treatment for refractory seizures in people aged 2 years and older once two other appropriate AEDs, trialled to a maximally tolerated dose, have failed to achieve seizure freedom.

Therefore, the only viable comparator is established clinical management (usualcare).

==> picture [30 x 14] intentionally omitted <==

Key clinical issues

Not applicable.

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416] © GW Research Ltd. (2022). All rights reserved

19 of 44

==> picture [39 x 13] intentionally omitted <==

Overview of the economic analysis

In line with the Epidyolex licence for TSC-associated epilepsy, the economic analysis estimates the cost-effectiveness of cannabidiol (Epidyolex[®] ) plus usual-care as an adjunctive therapy for seizures associated with TSC in patients aged ≥ 2 years versus usual-care.

An overview of the modelling approach is provided in Table 4, with full details provided in Document B, Section B.3.2.2 (pages 70–80). A model process diagram is presented in Figure 2.

Table 4: Overview of the model approach

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416]

© GW Research Ltd. (2022). All rights reserved

20 of 44

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416]

© GW Research Ltd. (2022). All rights reserved

21 of 44

Figure 2: Model process diagram – B.3.2.2 (page 73)

==> picture [446 x 117] intentionally omitted <==

==> picture [446 x 118] intentionally omitted <==

Key : HCRU, healthcare resource use; HRQL, health-related quality of life; IPD, individual patient-level data.

Note : *, Seizure frequency per day categories are aligned to HRQL data collected by seizure type.

Model seizure frequency categories for HCRU and HRQL

Previous health technology assessment (HTA) submissions evaluating cannabidiol in Dravet syndrome (DS) and Lennox–Gastaut syndrome (LGS) defined model health states based on low, medium and high seizure frequency and associated seizurefree days.[5, 6] The categorization of seizure frequency and seizure-free days allows for differential healthcare resource use (HCRU) and HRQL values to be calculated based on levels of seizure frequency and seizure-free days.

To model the HRQL and HCRU associated with differential seizure frequency and seizure-free days, the ‘alive health state’ is divided into sub-health states

representing different seizure frequency categories. Different seizure frequency categories were defined for HRQL and HCRU to appropriately capture the patient, caregiver and clinician experience and impact on costs to the NHS and Personal Social Services.

In line with feedback from NICE during the DS/LGS appraisals, HRQL data were collected using vignettes in a general population valued using time trade-off (TTO)

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416] © GW Research Ltd. (2022). All rights reserved

22 of 44

methodology. Taking on board the previous ERG and NICE feedback, the vignette descriptions were designed to accurately capture the HRQL profile of people with TSC-associated epilepsy and their caregivers. The vignette health states varied in terms of seizure frequency and severity on seizure days.

HCRU in the model was informed by the results of a two-round Delphi panel study involving 10 clinical experts. The resource utilization questions in the Delphi panel study were framed using weekly seizure frequency rather than daily seizure frequency. The weekly time period and categories used were defined based on clinician feedback on the most easily understood units of measurement when thinking about average patients and corresponding annual HCRU.

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416] © GW Research Ltd. (2022). All rights reserved

23 of 44

==> picture [39 x 13] intentionally omitted <==

Incorporating clinical evidence into the model

Data for cannabidiol plus usual-care and placebo plus usual-care from the treatment period of the blinded GWPCARE6 clinical trial (Weeks 1–16) and data for cannabidiol from the 72 weeks of the open-label extension (OLE) Period (Weeks 17– 88) were used to inform model inputs and validate model assumptions.

Efficacy data for cannabidiol plus usual-care and placebo plus usual-care were sourced from the treatment period of the GWPCARE6 trial.

First, a binomial regression model was used to predict the proportion of seizure-free days per 7-day cycle; then, a negative binomial model was used to predict the total seizure frequency on the non-seizure-free days per 7-day cycle. The regression coefficients were applied to each patient’s baseline seizure frequency from the GWPCARE6 trial at each 7-day model cycle to predict seizure-free-day and seizure frequency distributions. These individual calculations were used to distribute the cohort amongst sub-health-states representing different seizure frequency categories. Figure 3 and Figure 4 present the fitted values from the regression compared with the observed values from the GWPCARE6 (treatment period) and show that the model provides a good fit for the data.

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416] © GW Research Ltd. (2022). All rights reserved

24 of 44

Figure 3: Seizure-free days proportions for all observed cycles – B.3.3.2 (page 91)

==> picture [351 x 149] intentionally omitted <==

==> picture [351 x 149] intentionally omitted <==

Key : CBD, cannabidiol.

Figure 4: Seizure frequency on seizure days for all observed cycles – B.3.3.2 (page 91)

==> picture [345 x 151] intentionally omitted <==

==> picture [345 x 151] intentionally omitted <==

Key : CBD, cannabidiol.

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416] © GW Research Ltd. (2022). All rights reserved 25 of 44

The GWPCARE6 OLE data were used as validation of the predicted seizure frequency data. The OLE data were not used in the regression model as data were collected on a weekly basis rather than a daily basis, the number of seizure-free days was not recorded during the OLE and the number of missing days was not recorded. Figure 5 presents the extrapolated seizure frequency data from the fitted seizure frequency model compared with the observed OLE data. Overall, the regression model provides reasonable predictions for seizure frequency throughout the blinded trial and the OLE period.

Figure 5: Observed and estimated seizure frequency on observed and

estimated seizure days during the OLE – B.3.3.2 (page 93)

==> picture [450 x 210] intentionally omitted <==

Notes : Dashed line indicates the start of the OLE. Data are presented for the cannabidiol 25 mg/kg/day arm from the core trial.

The OLE data were also used to inform other model parameters such as stopping rule rates, discontinuation rates and TAND response rates.

Health-related quality of life data

The utility data collected in GWPCARE6 using the Quality of Life in Childhood Epilepsy (QOLCE) and Quality of Life in Epilepsy (QOLIE-31-P) questionnaires was deemed to be unsuitable for use in the economic model due to the following: the low response rates to the QOLIE-31-P (n = 11 in the cannabidiol arm and n = 10 in the

placebo arm); lack of a published mapping algorithm for the QOLCE for the

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416]

© GW Research Ltd. (2022). All rights reserved

26 of 44

paediatric population (described in Document B, Section B.3.4.2 and Section B.3.4.3); and the authors of the only mapping algorithm for QOLIE-31-P considering it to have only weak-to-moderate correlation with the EQ-5D.[7, 8]

A systematic literature review and two targeted literature reviews identified no suitable published utility data to inform patient or caregiver utilities for the analysis (Document B, Section B.3.4.4). Alternative options to collect HRQL data were therefore explored in line with NICE guidance[9] , including patient and caregiver surveys and vignettes.

Patient and caregiver surveys were ruled out as an option due to the lack of a valid preference-based measure to use in a population with TSC-associated epilepsy. The validity of the EQ-5D has been questioned in this patient population.[7, 8, 10] A study which investigated mapping the QOLIE-31-P to the EQ-5D-3L found that it was unable to capture small changes in seizure frequency over time and could not measure epilepsy-specific concepts such as seizure anxiety.[8] In addition, a comparison of the instruments used in the GWPCARE6 trial to collect HRQL data with the EQ-5D found there was a limited overlap between the instruments and the EQ-5D, indicating that the EQ-5D is not suitable in this population to capture HRQL (see Document B, Section B.3.4.4 for more detail).

Vignettes were explored as an option as they can be constructed to suit the model health states and designed to reflect the patient condition and caregiver burden. Vignettes have been used in previous submissions for cannabidiol in DS and LGS.[5, ] 6, 11 Given the unsuitability of the trial HRQL data, lack of published utility data and a suitable preference-based measure, vignettes were selected as the best approach to collect HRQL for patients and caregivers.

To address issues associated with the methodology used for the collection of HRQL noted by NICE in the submissions for DS and LGS[12, 13] , the health state vignettes were valued by a representative sample of the UK general population using TTO methods. The vignette descriptions included other aspects of TSC and TSCassociated epilepsy and were developed with input from healthcare professionals and caregivers (Document B, Appendix R).

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416] © GW Research Ltd. (2022). All rights reserved

27 of 44

==> picture [39 x 13] intentionally omitted <==

Key model assumptions and inputs

The key assumptions of the economic analysis are described in Table 5. TSCassociated epilepsy is a rare (orphan) condition predominantly impacting children, limiting the data available for this small population. While the modelling approach made the best use of the clinical data available, in the absence of data, use of published literature and elicitation of data from experts was necessary. As acknowledged by NICE in their updated manual, there are certain conditions, such as rare diseases, for which evidence generation is particularly difficult. Therefore, in this circumstance, acceptance of a higher degree of uncertainty may be considered by the Committee.[9]

Table 5 presents the range of assumptions embedded in the economic analysis and how they have been made to minimize potential bias.

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416] © GW Research Ltd. (2022). All rights reserved

28 of 44

Table 5: Key model assumptions and inputs

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416] © GW Research Ltd. (2022). All rights reserved 29 of 44

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416] © GW Research Ltd. (2022). All rights reserved 30 of 44

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416] © GW Research Ltd. (2022). All rights reserved 31 of 44

==> picture [39 x 13] intentionally omitted <==

Base case ICER (deterministic)

Table 6 displays the base case cost-effectiveness results for cannabidiol plus usualcare compared with placebo plus usual-care. All results are presented including the current patient access scheme (PAS) in place for cannabidiol in DS/LGS, with a list price of £850.29 and a PAS price of '''''''''''''''''' per 100 ml bottle (100 mg/ml).

Cannabidiol is estimated to provide additional discounted quality-adjusted life years (QALYs) of ''''''''''' (when including caregiver QALYs) at an additional discounted cost of ''''''''''''''''''. The estimated incremental cost-effectiveness ratio (ICER) for cannabidiol is £12,876. As discussed in Section A.17 below, under the latest NICE methods, cannabidiol would meet the criteria for the application of a disease severity QALY modifier of 1.2. The results with consideration of the disease severity modifier, as detailed in Document B, Section B.3.3.9 and Section A.17 below, are presented in Table 7 with the ICER reducing to £10,730. The results demonstrate that cannabidiol is a cost-effective use of NHS resources when considering the severity of TSCassociated epilepsy.

As outlined in Section A.1, TSC-associated epilepsy is a rare condition (with n = 1,215 patients expected in England) impacting predominantly children. As acknowledged by NICE in their updated manual, cost-effectiveness estimates are inherently uncertain for small populations such as these; therefore, these analyses should be considered within that context, with a higher degree of uncertainty acceptable.[9] Nevertheless, the conservative nature of the cost-effectiveness approaches outlined in Table 5 means that even the results presented in Table 6 in isolation strongly suggest that cannabidiol is a valuable option in England for NHS patients with TSC-associated epilepsy and their families.

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416] © GW Research Ltd. (2022). All rights reserved

32 of 44

Table 6: Base case results (deterministic) – B.3.8 (Table 31, page 152)

Table 7: Results (deterministic), including disease severity modifier – B.3.8 (Table 32, page 152)

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416] © GW Research Ltd. (2022). All rights reserved 33 of 44

==> picture [39 x 13] intentionally omitted <==

Probabilistic sensitivity analysis

Probabilistic sensitivity analysis was conducted where all inputs were varied simultaneously over 1,000 iterations. Distributional assumptions were driven by data wherever possible, and, in the absence of data, loose distributional assumptions were employed to ensure robust parameter uncertainty analysis (Document B, Section B.3.6). There is a '''''''''''''''' and ''''''''''''''' likelihood that cannabidiol is costeffective at willingness-to-pay (WTP) thresholds of £20,000 and £30,000 per QALY gained, respectively.

Table 8: Base case results (probabilistic) – B.3.9 (Table 36, page 156)

Figure 6: Scatterplot of probabilistic results – B.3.9 (Figure 20, page 156)

==> picture [452 x 125] intentionally omitted <==

==> picture [452 x 126] intentionally omitted <==

Key: ICER, incremental cost-effectiveness ratio; QALY, quality-adjusted life year.

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416]

© GW Research Ltd. (2022). All rights reserved

34 of 44

==> picture [39 x 13] intentionally omitted <==

Key sensitivity and scenario analyses

Figure 7 shows a tornado diagram depicting the 10 parameters that have the greatest influence on the ICER versus placebo plus usual-care in one-way sensitivity analyses.

These results demonstrate that the ICER is most sensitive to the stopping rule assessment rate applied at 6 months for patients with a seizure frequency greater than seven seizures per week (the highest seizure frequency category), the patient utility value applied to seizure-free patients, and response rates used to estimate the proportion of patients who benefit from a reduction in TAND symptoms.

Figure 7: Tornado diagram – B.3.9 (Figure 22, page 158)

==> picture [452 x 125] intentionally omitted <==

==> picture [452 x 125] intentionally omitted <==

Key : CBD, cannabidiol; HS, health state; ICER, incremental cost-effectiveness ratio; pw, per week; TAND, TSC-related neuropsychiatric disorders; UC, usual-care.

Results of the top seven most influential scenario analyses are shown in Table 9. Further scenario analyses exploring parameter, methodological and structural uncertainties around the base case results are reported in Document B, Section B.3.9.3.

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416] © GW Research Ltd. (2022). All rights reserved

35 of 44

The results were relatively insensitive for most analyses, with cannabidiol remaining cost-effective in all scenarios at a WTP threshold of £20,000 per QALY gained.

The most influential scenario, which resulted in a dominant ICER, and has the largest impact resulting in cost-savings for the cannabidiol arm, is associated with the inclusion of wider social costs: social and educational care resource use, which was elicited via the two-round Delphi panel study.

Table 9: Key scenario analyses – B.3.9 (Table 37, page 160)

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416] © GW Research Ltd. (2022). All rights reserved

36 of 44

==> picture [39 x 13] intentionally omitted <==

Innovation

Up to two-thirds of patients with TSC-associated epilepsy are refractory to currently available treatments, many of which were developed more than 20 years ago, and some more than 50 years ago.

Epidyolex is an innovative therapy for patients with TSC-associated seizures. It is the first cannabinoid in its class, with a novel, multi-modal mechanism of action, different to that of other AEDs. In clinical trials involving patients with TSC-associated epilepsy (who had tried numerous other treatments without achieving seizure control), cannabidiol has demonstrated clinically and statistically significant

reductions in seizure frequency and has a consistent, well-defined and manageable safety and tolerability profile.

Additional benefits of cannabidiol that are not captured in the QALY calculation include the value of:

• A beneficial impact on the mortality risk related to sudden unexpected death in epilepsy (SUDEP)

• Improving the quality of life of the wider family, including siblings

• Increasing caregiver productivity and the associated societal benefits of the parent(s)/primary caregiver(s) not needing to give up work to care for a patient with TSC-associated epilepsy

• Reducing the duration/severity of seizures (the model only captures seizure frequency)

• The long-term impact of improved seizure control on co-morbidities and injuries

Cannabidiol represents a step-change in the treatment of TSC-associated epilepsy, offering patients who live with the constant threat of seizures and who otherwise have extremely limited treatment options the opportunity of a well-tolerated, longterm treatment with sustained efficacy. It reduces seizure frequency and gives patients the potential for seizure-freedom or additional seizure-free time, as well as the possibility of reducing TAND, limiting neurological decline and improving developmental outcomes.

For further information see the section on Innovation in the main submission: Document B, Section B.2.12 (page 58).

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416]

© GW Research Ltd. (2022). All rights reserved

37 of 44

==> picture [39 x 13] intentionally omitted <==

Disease severity modifier criteria

The updated 2022 NICE methods guidance introduced new criteria to reflect in exceptional circumstances the severity of disease within decision making.[9] The company is aware that this submission is being assessed under the old methods guidance. However, given that the guidance was recently updated in order to support patients with severe diseases, we considered it appropriate to provide an indication of the severity of TSC-associated epilepsy and the impact of this on the economic analysis.

As described in Section A.1, patients with TSC-associated epilepsy have reduced life expectancy and significant lifetime morbidity. Patients with TSC-associated epilepsy are expected to experience an absolute shortfall of between 12 and 18 QALYs over a lifetime horizon. As summarized in Table 10, cannabidiol as a treatment within the population defined in this decision problem satisfies the criteria for the application of a QALY weight of 1.2.

Table 10: Decision modifier – severity criteria

==> picture [39 x 14] intentionally omitted <==

Budget impact

Aligned with the outcomes of the cost-effectiveness analyses, when the PAS is applied, the net budget impact of cannabidiol remains under the £20 million threshold throughout the 5-year time horizon, as summarized in Table 11.

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416]

© GW Research Ltd. (2022). All rights reserved

38 of 44

Table 11: Budget impact – company BIA document

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416]

© GW Research Ltd. (2022). All rights reserved

39 of 44

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416] © GW Research Ltd. (2022). All rights reserved

40 of 44

==> picture [39 x 13] intentionally omitted <==

Interpretation and conclusions of the evidence

TSC-associated epilepsy is a debilitating, lifelong and treatment-resistant form of epilepsy, with an increased risk of death as a result of SUDEP or status epilepticus. Up to two-thirds of patients are refractory to current treatments.

Cannabidiol has been robustly evaluated in patients with refractory TSC-associated epilepsy in a global randomized controlled trial (GWPCARE6) and its associated ongoing OLE study.

• The blinded study met its primary endpoint: patients taking cannabidiol experienced a median 49% reduction in TSC-associated seizures versus 27% on usual-care (p = 0.0009), a clinically meaningful result. Cannabidiol also increased the chance of achieving TSC-associated seizure-freedom and/or additional seizure-free days

• Cannabidiol has a well-defined safety and tolerability profile. Most adverse events were mild to moderate, transient and resolved by the end of the trial

• The ongoing OLE study demonstrates that the efficacy of cannabidiol in reducing seizures is sustained in the longer term. Interim data up to 72 weeks demonstrating durability of outcomes have been reported

As outlined in section A.16 (Innovation) above, additional benefits of cannabidiol that are not captured in the economic analysis include: a beneficial impact on the SUDEP mortality risk; improving the quality of life of the wider family, including siblings; increasing caregiver productivity; reducing the duration/severity of seizures; the longterm impact of improved seizure control on co-morbidities and injuries.

Cannabidiol offers patients with refractory TSC-associated seizures the opportunity of a well-tolerated, long-term treatment with sustained efficacy. It reduces seizures and gives patients the potential for seizure-freedom or additional seizure-free time, as well as the possibility of reducing TAND, limiting neurological decline and improving developmental outcomes.

Over a lifetime time horizon, the ICER for cannabidiol in addition to usual-care versus usual-care alone is £12,876 per QALY gained. Thus, cannabidiol is costeffective in patients who have extremely limited other treatment options.

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416] © GW Research Ltd. (2022). All rights reserved

41 of 44

The number of refractory patients with TSC-associated epilepsy eligible for treatment with Epidyolex is estimated to be 1,215. Cannabidiol will have a limited budget impact due to the orphan nature of TSC-associated epilepsy, as well as cost offsets associated with disease management.

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416] © GW Research Ltd. (2022). All rights reserved

42 of 44

==> picture [39 x 13] intentionally omitted <==

References

1. Thiele EA, Bebin EM, Bhathal H, et al. Add-On Cannabidiol Treatment for Drug-Resistant Seizures in Tuberous Sclerosis Complex: A Placebo-Controlled Randomized Clinical Trial. JAMA Neurol . 2020.

2. Wheless J, Bebin EM, Filloux F, et al. Long-term Safety and Efficacy of Addon Cannabidiol (CBD) for Treatment of Seizures Associated with Tuberous Sclerosis Complex (TSC) in an Open-Label Extension (OLE) Trial (GWPCARE6) (1127). Neurology . 2021; 96(15 Supplement):1127.

3. Office for National Statistics. National Life Tables, England, 1980-1982 to 2016-2018. 2018. (Updated: 25 September 2019) Available at: https://www.ons.gov.uk/peoplepopulationandcommunity/birthsdeathsandmarriages/lif eexpectancies/datasets/nationallifetablesenglandreferencetables . Accessed: 25 January 2022.

4. Nabbout R, Belousova E, Benedik MP, et al. Epilepsy in tuberous sclerosis complex: Findings from the TOSCA Study. Epilepsia Open . 2019; 4(1):73-84. 5. National Institute for Health and Care Excellence (NICE). TA614: Cannabidiol with clobazam for treating seizures associated with Dravet syndrome. 2019. (Updated: 18 December 2019) Available at: https://www.nice.org.uk/guidance/TA614 . Accessed: 25 January 2022. 6. National Institute for Health and Care Excellence (NICE). TA615: Cannabidiol with clobazam for treating seizures associated with Lennox–Gastaut syndrome. 2019. (Updated: 18 December 2019) Available at: https://www.nice.org.uk/guidance/ta615/ . Accessed: 25 January 2022.

5. Wijnen BFM, Mosweu I, Majoie M, et al. A comparison of the responsiveness of EQ-5D-5L and the QOLIE-31P and mapping of QOLIE-31P to EQ-5D-5L in epilepsy. Eur J Health Econ . 2018; 19(6):861-70.

6. Mukuria C, Young T, Keetharuth A, et al. Sensitivity and responsiveness of the EQ-5D-3L in patients with uncontrolled focal seizures: an analysis of Phase III trials of adjunctive brivaracetam. Qual Life Res . 2017; 26(3):749-59.

7. National Institute of Clinical Excellence (NICE). NICE health technology evaluations: the manual: Process and methods [PMG36]. 2022. (Updated: 31 - January) Available at: https://www.nice.org.uk/process/pmg36/chapter/introduction to-health-technology-evaluation . Accessed: 31 January 2022.

8. Mulhern B, Pink J, Rowen D, et al. Comparing Generic and ConditionSpecific Preference-Based Measures in Epilepsy: EQ-5D-3L and NEWQOL-6D. Value Health . 2017; 20(4):687-93.

9. Lo SH, Lloyd A, Marshall J and Vyas K. Patient and Caregiver Health State Utilities in Lennox-Gastaut Syndrome and Dravet Syndrome. Clin Ther . 2021; 43(11):1861-76 e16.

10. National Institute for Health and Care Excellence (NICE). TA614: Cannabidiol with clobazam for treating seizures associated with Dravet syndrome. Final appraisal document. 2019. (Updated: 12 May 2021) Available at: https://www.nice.org.uk/guidance/ta614/documents/final-appraisal-determinationdocument . Accessed: 25 January 2022.

11. National Institute for Health and Care Excellence (NICE). TA615: Cannabidiol with clobazam for treating seizures associated with Lennox–Gastaut syndrome. Final appraisal document. 2019. (Updated: 18 December 2019) Available at: https://www.nice.org.uk/guidance/ta615/documents/final-appraisal-determinationdocument . Accessed: 25 January 2022. Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416]

© GW Research Ltd. (2022). All rights reserved

43 of 44

14. GW Pharma (International) B.V. Summary of product characteristics (SPC). Epidyolex (cannabidiol) 100 mg/ml oral solution. Summary of product characteristics 2021.

15. Thiele E, Bebin EM, Filloux F, et al. Long-term Safety and Efficacy of Cannabidiol (CBD) for the Treatment of Seizures in Patients with Tuberous Sclerosis Complex (TSC) in an Open-label Extension (OLE) Trial (GWPCARE6)(677). Neurology . 2020; 94.

16. Patel A, Chin R, Mitchell W, et al. Long-Term Safety and Efficacy of Cannabidiol (CBD) Treatment in Patients with Lennox Gastaut Syndrome (LGS): 3- Year Results of an Open-Label Extension (OLE) Trial (GWPCARE5) (668). Neurology . 2020; 94(15 Supplement):668.

17. Halford JJ, Scheffer I, Nabbout R, et al. Long-Term Safety and Efficacy of Cannabidiol (CBD) Treatment in Patients with Dravet Syndrome (DS): 3-Year Interim Results of an Open-Label Extension (OLE) Trial (GWPCARE5) (439). Neurology . 2020; 94(15 Supplement):439.

18. Weinstock A, Bebin M, Checketts D, et al. Long-term Efficacy and Safety of Cannabidiol (CBD) in Patients with Tuberous Sclerosis Complex (TSC): 4-year Results from the Expanded Access Program (EAP) (2405). The American Epilepsy Society Annual Meeting . Virtual event2020.

19. National Institute for Health and Care Excellence (NICE). TA615: Cannabidiol with clobazam for treating seizures associated with Lennox–Gastaut syndrome. Appraisal consultation committee papers. 2019. (Updated: 18 December - 2019) Available at: https://www.nice.org.uk/guidance/ta615/documents/committee papers . Accessed: 25 January 2022.

20. GW Pharmaceuticals. GW Pharma Health Technology Assessment (HTA) Advisory Board for Epidyolex to Treat Tuberous Sclerosis Complex – Summary Report. 2020. (Updated: 25 January 2022) Data on File.

21. Ryvlin P, Cucherat M and Rheims S. Risk of sudden unexpected death in epilepsy in patients given adjunctive antiepileptic treatment for refractory seizures: a meta-analysis of placebo-controlled randomised trials. Lancet Neurol . 2011; 10(11):961-8.

22. Lagae L, Irwin J, Gibson E and Battersby A. Caregiver impact and health service use in high and low severity Dravet syndrome: a multinational cohort study. Seizure . 2019; 65:72-9.

23. Shepherd C, Koepp M, Myland M, et al. Understanding the health economic burden of patients with tuberous sclerosis complex (TSC) with epilepsy: a retrospective cohort study in the UK Clinical Practice Research Datalink (CPRD). BMJ Open . 2017; 7(10):e015236.

24. Tritton T, Bennett B, Brohan E, et al. Health utilities and quality of life in individuals with tuberous sclerosis complex (TSC) who experience epileptic seizures: A web-based survey. Epilepsy Behav . 2019; 92:213-20.

25. Vergeer M, de Ranitz-Greven WL, Neary MP, et al. Epilepsy, impaired functioning, and quality of life in patients with tuberous sclerosis complex. Epilepsia Open . 2019; 4(4):581-92.

Summary of company evidence submission template for cannabidiol for treating seizures caused by TSC [ID1416] © GW Research Ltd. (2022). All rights reserved

44 of 44

NATIONAL INSTITUTE FOR HEALTH AND CARE EXCELLENCE

Single technology appraisal

Cannabidiol for treating seizures caused by tuberous sclerosis complex [ID1416] Clarification questions

May 2022

Section A: Clarification on effectiveness data

Searches

A1. The ERG noted that the Medline search appears to contain Emtree subject headings rather than MeSH terms. It also states that the search was conducted via Embase. Please can you confirm that by this you mean a search of the Embase database conducted on the understanding that it now contains all records from Medline or was this a separate search of both resources using the same strategy?

We confirm that we searched Medline via the Embase database, on the understanding that it now contains all records from Medline.

A2. Whilst reference checking is mentioned in Appendix H for the targeted literature review, there is no mention of it in Appendix D for the main searches. Please confirm whether this took place for the main systematic literature review (SLR).

Yes, reference checking of existing systematic literature reviews was conducted for the main SLR, to identify additional relevant publications.

Decision problem

A3. Priority: Whereas the final NICE scope does not include people with Tuberous sclerosis complex (TSC) where usual care is unsuitable or not tolerated by current or prior established clinical management, the company does. The company claims that this modified population is justified because it is “ in line with recommendations in NICE Clinical guideline 137 (CG137). ”

a. Please specify where in NICE Clinical Guideline 137 it explicitly makes recommendations for people with TSC where usual care is unsuitable or not tolerated.

Nice Clinical Guideline 137 does not explicitly make recommendations for people with TSC-associated epilepsy.

However, more generally for all patients with epilepsy, NICE CG137 makes the following recommendation: “Treatment should be reviewed at regular intervals to ensure that children, young people and adults with epilepsy are not maintained for

long periods on treatment that is ineffective or poorly tolerated”. This applies to patients with TSC-associated epilepsy.

Furthermore, it is well known that certain AEDs are not suitable for all patients. For example, sodium valproate is associated with a risk of teratogenicity, therefore it is often deemed to be unsuitable for women of childbearing age. NICE CG137 states “Do not offer sodium valproate to women and girls of childbearing potential (including young girls who are likely to need treatment into their childbearing years) unless other options are unsuitable, ineffective or not tolerated and the pregnancy prevention programme is in place”.

b. Please specify what constitutes usual care.

In TSC-associated epilepsy, because there is no standard of care once a patient is refractory, various terms have been used (interchangeably) to describe the variety of treatments used in an attempt to gain seizure control. These terms include, for example, ‘current clinical management’, ‘established clinical management’, ‘usualcare’.

In the GWPCARE6 trial of cannabidiol, patients entering the trial were permitted to be taking any AEDs/other treatments (except those listed as exclusion criteria) as long as they were stable during baseline and during the trial. These treatments/ combinations of treatments are what are referred to as 'usual-care' in our submission. Cannabidiol was an add-on to usual-care.

Please see question A9 below for further detail.

c. Is there a difference between the treatments termed as usual care and

established clinical management? If yes, please specify.

No, the terms are used interchangeably.

d. Please specify the main reasons that account for the unsuitability or

intolerability of usual care.

Reasons for unsuitability include: teratogenicity risk (e.g. sodium valproate for females of childbearing age); contraindications/cautions (e.g. carbamazepine is contra-indicated for patients with a history of bone marrow depression; topiramate

should be used with caution in patients with acute porphyrias, metabolic acidosis and/or glaucoma); patient/caregiver preference (for example, some parents choose not to allow their child to have certain drugs due to their side-effect profile/perceived risk).

Reasons for intolerability: this is usually related to side-effects. Examples include: vigabatrin is associated with visual defects; benzodiazepines are associated with drowsiness. Patients/caregivers may choose to discontinue a treatment as a result of these side-effects.

e. Please specify the criteria used to judge whether usual care is classified as unsuitable.

This is a decision taken for each individual patient by the clinician, in conjunction with the patient, the patient’s caregivers and/or the multidisciplinary team responsible for the patient’s care.

f. Please specify the criteria used to judge whether usual care is classified as not tolerated.

This is a decision taken for each individual patient by the clinician, in conjunction with the patient, the patient’s caregivers and/or the multidisciplinary team responsible for the patient’s care.

g. Please provide a comparison of effectiveness, safety and cost-effectiveness for cannabidiol versus all relevant comparators for patients who do and do not receive usual care.

As outlined above, all patients receive usual-care. Cannabidiol is an add-on to usualcare.

A4. Priority: Whereas the final NICE scope lists everolimus as a comparator, in their submission (Document B), the company lists it as a later line of therapy. The company justifies this based on:

• a. NHS England Clinical Commissioning Policy (Everolimus for refractory focal onset seizures associated with tuberous sclerosis complex (ages 2 years and above) 2018)[1] . Please elaborate on how this document supports using

everolimus as a later line therapeutic option as opposed to its indication specified in the final NICE scope?

The comparator in the final NICE scope is “Established clinical management without cannabidiol”.

In the NHS England Clinical Commissioning Policy, everolimus is positioned as a later line treatment. For a patient to be eligible for treatment with everolimus, there is a need not only for inadequate response to two or more AEDs, but also for failure/ruling out of both epilepsy surgery and vagus nerve stimulation (VNS) by a multidisciplinary team including, but not limited to, a radiologist, and a neurology specialist with experience in TSC management. There is also a stated requirement to have local availability of services to support therapeutic drug monitoring.

It should be noted that everolimus is only licensed for a subset of TSC-associated seizures. It is indicated as adjunctive treatment of patients aged 2 years and older whose refractory partial onset seizures , with or without secondary generalisation, are associated with TSC. Unlike cannabidiol, everolimus is not licensed for generalised onset TSC-associated seizures.

In the NHS England Clinical Commissioning Policy, the ‘Criteria for Starting treatment’ are as follows:

Patients aged 2 years and older with a confirmed diagnosis of TSC related seizures whose refractory partial-onset seizures (focal onset seizures) are associated with TSC, AND

• whose TSC-related seizures have not adequately responded (meaning 2 or more partial onset seizures per month OR recurrent status epilepticus to treatment with at least 2 different AEDs titrated to a therapeutic dose; AND

• who have previously been considered for surgical resection as assessed by a designated Children’s Epilepsy Surgery Service (CESS) or adult specialised epilepsy surgery service. Specifically, the CESS / specialised adult service will have previously decided that:

  • there is no brain abnormality which can be identified as causing seizures that can be removed surgically without unacceptable risks; OR

• there are multiple or infiltrative brain abnormalities which may be causing seizures which cannot be removed surgically; OR

• surgery has been performed and the seizures have not adequately reduced in frequency or severity; AND

• who have been considered for VNS and:

  • VNS was not considered appropriate as the next treatment option by the patient, or their carer in discussion with the treating clinician; OR

  • VNS has been performed and seizures have not adequately reduced in frequency or severity; AND

• for whom, in the opinion of a properly constituted multi-disciplinary team (MDT) (as defined in the governance arrangements), everolimus is considered more appropriate than a trial of an alternative AED (in line with NICE CG137 which states that treatment strategies should be individualised).

b. “ The drug’s [everolimus’] safety/tolerability profile ” (stated in company submission, Document B). Please provide evidence contrasting the safety/tolerability of cannabidiol with that of everolimus?

Everolimus is an mTOR inhibitor. The mTOR inhibitors are anti-tumour agents with immunosuppressive properties. In addition to TSC-associated seizures, everolimus is indicated for treatment of advanced or progressive malignancies (breast cancer, neuroendocrine tumours, and renal cell carcinoma), as well as non-cancerous tumours in patients with TSC. Adverse reactions to everolimus include non-infectious pneumonitis, infections, severe hypersensitivity reactions, angioedema, stomatitis, renal failure, impaired wound healing, metabolic disorders, and myelosuppression.

The safety and tolerability profile of cannabidiol is consistent, well-defined and manageable, as demonstrated across five randomised controlled Phase III trials in severe refractory epilepsies, including TSC-associated epilepsy, DS and LGS. In GWPCARE6, most adverse events were mild to moderate, transient and resolved by the end of the trial. The most common adverse events were diarrhoea (31% of patients treated with 25 mg/kg/day cannabidiol vs 25% of patients in the placebo group), decreased appetite (20% and 12%, respectively), vomiting (17% and 9%, respectively) and somnolence (13% and 9%, respectively).

c. That everolimus is not specifically an anti-seizure medication (ASM). Can the company please state why not being specifically an ASM is relevant to whether it is included in the final NICE scope and should be used as a comparator by the company?

In the draft scope prior to the decision problem meeting, everolimus was included as follows: “Anti-epileptic drugs, which may include everolimus”.

The company was simply pointing out that everolimus is not an AED/ASM. It is an mTOR inhibitor with prior indications in immunosuppression and cancer.

d. Everolimus’ “ safety and tolerability burden ” (stated in company submission Document B). Given that all the comparators (as well as the intervention) also have potential adverse effects, please justify excluding everolimus as a direct comparator on this basis.

The comparator in the NICE final scope is “Established clinical management without cannabidiol”. Therefore everolimus alone is not a direct comparator, regardless of its safety and tolerability burden.

e. In the company submission (Document B), the company states that

everolimus “ may be considered in TSC earlier in the pathway, but not specifically for the treatment of seizures. ” Can the company please provide additional rationale for this assertion?

Everolimus has a separate indication/dosing schedule in TSC that is not specifically related to seizures: for the treatment of subependymal giant cell astrocytoma (SEGA), a benign tumour of the brain, where it is used in adults and children whose brain tumour cannot be surgically removed. For this indication and dosage, it may be considered in TSC earlier in the pathway, but it is licensed to treat SEGA and not specifically for the treatment of TSC-associated seizures.

A5. Regarding the addition of seizure-free days as an outcome, please provide evidence regarding the extent to which seizure-free days are correlated with seizures.

Seizure-free days is not an addition to the outcomes. Change in seizure-free days was a pre-planned secondary endpoint in the GWPCARE6 clinical trial and has been

included as an outcome in prior submissions for refractory epilepsies: TA614 and TA615 and ID1109.[2-4] It is a clinically important outcome for patients and caregivers and a crucial element when considering the impact and efficacy of cannabidiol in patients with TSC-associated epilepsy. In the GWPCARE6 trial, patients taking cannabidiol (25 mg/kg/day) experienced an additional 2.8 seizure-free days per month vs the placebo group (p=0.0047).[5]

The previous models accepted by NICE for cannabidiol in DS and LGS included seizure-free days as an important aspect of quality of life. Auvin et al. reinforce the importance of seizure-free days as a specific outcome, concluding that, whilst fewer seizures and additional seizure-free days both improved quality of life in caregivers and patients, seizure-free days had the greatest impact on patient quality of life.[6]

As detailed in Document B, improvements in quality of life and patient wellbeing are linked to both the number of seizures experienced, as well as how these seizures are distributed over time. A period of seizure-free time (whether several hours in a day, or seizure-free days) has the potential to improve quality of life for patients and their families.

Additionally, feedback from clinicians and patient organisations also highlights the importance of seizure-free days:

• Clinical experts at an advisory board meeting highlighted that seizure-free days matter more in terms of quality of life than a change in seizure frequency.[7]

• “Whilst reductions in convulsive seizures and drop seizures are of most medical benefit, other changes in seizure activity, including altering patterns of seizures leading to increased seizure-free days, should be viewed as clinically/statistically significant.” - Epilepsy Action, comment on HTA for cannabidiol in DS and LGS.[8, 9]

Therefore, it is clinically important to include seizure-free days as an outcome in the analysis.

We would expect to see a moderate negative correlation between seizure frequency and seizure-free days, i.e. a reduction in seizure frequency might lead to an increase in seizure-free days. However, although seizure frequency and seizure-free days

may be moderately correlated, it is possible to experience a reduction in seizure frequency without a corresponding increase in seizure-free days and vice versa.

The NICE committee conclusion from the ACD for TA614 and TA615 considered it appropriate to capture the benefits of having more seizure-free days. However, the committee also considered that the approach used (categorisation into number of seizures, and then subdivision of these into number of seizure-free days) may have resulted in ‘double-counting’ the benefits of reducing the frequency of seizures.

To address this, the modelling approach used in the current submission allows for the separate modelling of seizure-free days and seizure frequency, whilst also accounting for the correlation between both. Firstly, a binomial regression model was used to predict the proportion of seizure-free days per cycle. Secondly, a fitted negative binomial model was used to predict the total seizure frequency on the nonseizure-free days per cycle. The correlation between both outcomes is therefore captured, as seizure frequency is only estimated for the days in each cycle when patients are expected to have seizures.

GWPCARE6 trial and data analysis

Population

A6. Priority: Table 4 in the company submission lists the UK as one of the GWPCARE6 study locations. Please provide the number of UK patients randomised and provide the baseline characteristics of these patients by study arm.

Eleven patients from the UK were screened, and 7 of the 11 were randomised. Two of the seven were randomised to the 50 mg/kg/day dose and are not included here. The baseline characteristics of the other 5 patients from the UK are shown in Table 1 below.

Table 1: UK patients - baseline characteristics

Key: F, female; M, male; W/C, White/Caucasian

A7. Priority: Please discuss the generalisability of the study baseline characteristics to the general UK population (with supporting documents).

The company considers that the GWPCARE6 study baseline characteristics are generalisable to the general UK population:

• UK specialist clinicians agree that the participants with TSC-associated epilepsy in the GWPCARE6 trial broadly reflect the characteristics of people seen in their clinical practice in the UK National Health Service (NHS). This was noted in an HTA advisory board meeting[7] and also confirmed in recent discussions (conducted to inform our responses to the ERG) with two UK clinical experts - consultant neurologists Professor Finbar O’Callaghan and Dr Sam Amin.

• The GWPCARE6 trial included UK patients (see Table 1 in question A6 above)

• The diagnostic criteria for TSC-associated epilepsy in the trial were based on international guidelines, which are applicable to UK patients.

Comparator

A8. Priority: Please provide evidence that the comparator treatments in the GWPCARE6 trial are representative of UK clinical management. In particular, please compare and contrast the comparator in the trial with established clinical management or usual care as would be the case in NHS clinical practice.

A9. Cannabidiol appears to have been given to patients who remain on the clinical management (usual care) that they were on before the start of the trial.

• a. Please clarify if the usual care interventions permitted during the trial were protocol-specified.

• b. Please list all permitted usual care interventions, and provide data about the patients who received these, divided according to treatment group.

• c. Please list any other permitted concomitant medications (non-usual care) and provide data about how many patients in each arm received these.

A9. a, b and c. We confirm that cannabidiol was an add-on to usual-care. The usualcare interventions were not protocol-specified. Patients entering the GWPCARE6 trial were permitted to be taking any AEDs/other treatments (except those in the exclusion criteria) as long as they were stable during baseline and during the trial. These treatments/combinations of treatments are what are referred to as 'usual-care' throughout our submission.

Refractory patients with TSC-associated epilepsy may cycle through numerous AEDs in an attempt to achieve seizure control. As a result, ‘usual-care’ comprises many different AEDs/combinations of AEDs - demonstrating that there is no standard of care once a patient is refractory.

Table 2 and Table 3 below show the range of different drugs available, and thus the huge number of potential combinations.

Table 2: GWPCARE6 – prior and concomitant AEDs

Table 3: Concomitant AEDs

==> picture [359 x 151] intentionally omitted <==

==> picture [359 x 150] intentionally omitted <==

==> picture [359 x 151] intentionally omitted <==

A10. Please provide more detail on the decision-making process underlying choice of the ‘usual care’ treatments. Were these decisions made freely on the basis of clinical need at the discretion of the attending clinician, or were they taken from a list of pre-hoc agreed ‘usual care’ treatments (if so please specify them)?

The patient's 'usual-care' was at the discretion of the clinician (for the reasons outlined in question A9 above).

A11. There was a large discrepancy between groups at baseline for use of vigabatrin, with 37% of the cannabidiol group using it compared to 22% of the placebo group. This was offset by slightly larger uses of valproic acid, levetiracetam and clobazam in the placebo group. Given that vigabatrin is considered the first line drug (company submission, Figure 4, Document B), this discrepancy may have important influences on outcome. Please explain how this potential threat to internal validity has been accounted for in the analysis.

The company does not consider that there is a threat to internal validity.

As stated in our submission, vigabatrin is recommended as first-line monotherapy for TSC-associated infantile spasms and/or focal seizures in children <1 year old. Vigabatrin is associated with irreversible visual field defects, including blindness in severe cases, and is therefore not suitable for all patients.

As explained in question A9, refractory patients with TSC-associated epilepsy may cycle through many AEDs in an attempt to achieve seizure control.

The majority of patients with refractory TSC-associated epilepsy in the GWPCARE6 trial had already tried and failed vigabatrin i.e., the drug did not lead to seizure control. The GWPCARE6 trial population had failed to achieve seizure control with a median of 4 AEDs prior to entering the study. Vigabatrin was among the most common of these AEDs, having already been tried and stopped by 43% of patients prior to entering the study. In addition, a further 33% of patients were taking vigabatrin on entering the study, meaning that, by definition, it was not working as they were not achieving adequate seizure control. Therefore, in total, >75% of the GWPCARE6 trial population had already failed to achieve seizure control with vigabatrin.

The results of a pre-specified subgroup analysis show no effect on the primary endpoint whether the patient was taking or not taking vigabatrin (see Figure 1 below).

Figure 1: Subgroup analysis of the primary endpoint

==> picture [353 x 134] intentionally omitted <==

==> picture [353 x 135] intentionally omitted <==

Outcomes

A12. Priority: Section 5.5.2.5 of the clinical study report (CSR) states, “ Effects on quality of life were therefore measured using the QOLCE/QOLIE-31-P questionnaires which have good construct validity, internal consistency, testretest reliability, and sensitivity to epilepsy severity .”[10] Section 10.1 of the CSR states: “ these patients had a poor quality of life based on the low mean overall QOLCE and QOLIE-31-P scores of approximately 44–50 at baseline .”[10] Yet, the company did not present detailed data for this outcome in their submission.

• a. Please provide all outcome data that reports results of the QOLCE/QOLIE-31-P questionnaires.

Please see Figure 2 below for QOLCE and Table 4 below for QOLIE-31-P.

Figure 2: Change from Baseline to End of Treatment in QOLCE

==> picture [357 x 159] intentionally omitted <==

==> picture [357 x 158] intentionally omitted <==

Table 4: Change from Baseline to End of Treatment in QOLIE-31-P

==> picture [345 x 170] intentionally omitted <==

==> picture [345 x 169] intentionally omitted <==

b. Please state how the QOLCE/QOLIE-31-P questionnaire results were accounted for in the efficacy conclusions.

GWPCARE6 attempted to capture HRQL using the general epilepsy quality of life instruments QOLCE and QOLIE-31-P. However, as outlined in the company’s submission, there are significant challenges in collecting HRQL data in clinical trials involving patients with severe and refractory epilepsies such as TSC-associated epilepsy:

• There are no validated disease-specific instruments

• These types of general epilepsy QoL instruments do not work well for severe epilepsy (for example, asking questions about social interactions for a patient with TSC-associated epilepsy who has physical/learning disabilities that mean they do not attend school, go to work, or have any social interactions)

• Since many of the questions in general QoL instruments are unsuitable, there are missing data, making it difficult to draw meaningful conclusions

Over the GWPCARE6 trial period, most differences in the QOLIE-31-P and QOLCE were in favour of cannabidiol, but none were statistically significant. However, missing or non-applicable items were an issue for both instruments.

Neither the QOLIE-31-P or QOLCE are validated measures of HRQL for TSCassociated epilepsy. Although they are used in clinical trials of patients with general epilepsy, they are not appropriate for the very severe end of the epilepsy spectrum, where patients have an exceptionally high frequency/burden of seizures and substantial levels of learning difficulties, making many of the questions unsuitable. This is a likely reason for the high levels of missing data evidenced for both measures used in the GWPCARE6 study. The extent of the missing data for both instruments used in the clinical trials limited their ability to draw conclusions from the data.

A more meaningful measure of HRQL from the GWPCARE6 clinical trial is the Subject/Caregiver Global Impression of Change (S/CGIC), which captures an estimate of the effect of treatment on the patient’s overall condition based on his/her entire seizure and comorbidity burden, thereby providing valuable information on the clinical meaningfulness of the therapy. Using S/CGIC, patients and caregivers reported an improvement in patients’ overall condition in 69% of those receiving cannabidiol (25 mg/kg/day) plus usual-care versus 39% receiving placebo plus usual-care. However, the S/CGIC measure is not preference-based and therefore could not be used to derive utilities.

A13. Many of the outcomes were measured by changes from baseline. The ERG would like to know more about how baseline measurements compared with measurements taken during the trial period. Specifically, in the trial period, seizure frequency and seizure free days were assessed on each day from the baseline to the completion of dosing using an IVRS diary

• a. Please state how baseline seizure free days were measured.

• b. Please state how baseline seizure frequency was measured.

• c. Please highlight all differences between how seizure frequency and seizure free days were measured at baseline and for the trial period.

The measurements were the same during the baseline period baseline and the trial period.

Eligible patients entered the trial at the screening visit (Day −35) and began a 7-day screening period. Patients who successfully completed this then began a 28-day baseline period on Day −28. Patients who satisfied all eligibility criteria were then randomized on Day 1.

An interactive voice response (IVRS) system was used daily to record information on seizures. During the baseline period and the double-blind participation in the trial, the caregiver made daily calls into an interactive voice response system (IVRS) to log the seizures experienced by the subject within the previous 24 hours.

Adverse events and drug-drug interactions

A14. Appendix F and Section B.2.10 of the company submission refer to adverse reactions from the GWPCARE6 trial.

a. Please provide the follow-up period and tool used for adverse events reporting.

A paper diary was used daily to record information on adverse events (AEs).

All AEs (including serious AEs) observed by the investigator or reported by the patient/caregiver during the trial were recorded on the patient’s Case Report Form at all trial visits, questioning the patient/caregiver further if necessary.

An AE was defined as any new unfavourable/unintended signs/symptoms (including abnormal laboratory findings), or diagnosis or worsening of a pre-existing condition, which occurred following screening (Visit 1) and at any point up to the post-treatment safety follow-up visit (Visit 12, for patients who did not enter the OLE), which may or may not be considered related to the IMP. Any event that was the result of a trial procedure was to be recorded as an AE.

Unless entering the OLE trial (in which case patients would have been monitored for AEs for the duration of the OLE) the trial required that patients be actively monitored for AEs up to 28 (+3) days after the last dose of IMP, until Visit 12.

• b. Please provide the metric used to classify the severity of adverse events in Table 3 of Appendix F and Table 8 in the company submission.

For all AEs and serious AEs, the clinical trial investigators were required to assign severity and document this on the Case Report Form.

The method is described in the trial protocol as follows:

“When describing the severity of an AE, the terms mild, moderate, or severe should be used. Clinical judgment should be used when determining which severity applies to any AE.

If the severity of an AE fluctuates day-to-day, e.g., a headache or constipation, the change in severity should not be recorded each time; instead, only the worst observed severity should be recorded with AE start and stop dates relating to the overall event duration, regardless of severity.

A severe AE is not the same as a SAE. For example, a patient may have severe vomiting but the event does not result in any of the SAE criteria above. Therefore, it should not be classed as serious.”

An AE was considered serious if it: (1) was fatal; (2) was life-threatening; (3) required inpatient hospitalization or prolonged existing hospitalization; (4) was persistently or significantly disabling or incapacitating; (5) was a congenital anomaly/birth defect; or (6) was a medically significant event that, based upon appropriate medical judgment, may have jeopardized the patient and may have required medical or surgical intervention to prevent one of the outcomes listed above.

• c. Please clarify if the published SAEs in Table 3 of Appendix F were reported in all participants or in ≥10% of participants.

All participants.

A15. It is well known that cannabis-based medications through interference with CYP3A4 enzymes, have the potential to initiate drug-drug interactions that may lead to serious drug toxicities and side effects in real world practice.

The company is somewhat concerned that no supporting evidence/references have been provided to support the speculation (about ‘cannabis-based medications’ and not specifically Epidyolex) that “ It is well known that cannabis-based medications through interference with CYP3A4 enzymes, have the potential to initiate drug-drug interactions that may lead to serious drug toxicities and side effects in real world practice”.

We respectfully request that this is removed before the document is in the public domain so that questions a, b and c are standalone without this introduction.

• a. As the GWPCARE6 CSR states that, “ Care was to be taken with drugs, or their metabolites, that are cytochrome P450 2C19 substrates… ”[10] please clarify if certain medications were disallowed during the trial to prevent these potentially toxic drug-drug interactions from occurring.

As above, we respectfully request that the words ‘potentially toxic’ are removed from this question.

The GWPCARE6 study protocol stated that “Care should be taken with drugs, or their metabolites, that are cytochrome P450 2C19 substrates, such as N- desmethylclobazam. Care should also be taken with drugs, or their metabolites, that are solely or primarily metabolized by UDP-glucuronosyltransferase 1A9 and 2B7.” However, these medications were not disallowed during the study.

As per the GWPCARE6 trial exclusion criteria, the following medications were prohibited for the duration of the trial:

• Any new medications or interventions for epilepsy (including ketogenic diet and VNS) or changes in dosage

• Recreational or medicinal cannabis or synthetic cannabinoid-based medications (including Sativex) within 3 months prior to or during the trial

• Any other IMP taken as part of a clinical trial

• Felbamate if taken for less than 1 year prior to screening

• Oral mTOR inhibitor

• b. Please clarify if the dosage of concomitant ASMs that patients were stable on for 1 month prior to screening, would have been modified by the investigator during the trial considering potential drug-drug interactions.

Throughout the duration of the trial, doses of concomitant AEDs and any nonpharmacological regimens for epilepsy were to remain stable. However, due to potential pharmacological interactions between cannabidiol and other concurrently administered drugs, the doses of concomitant AEDs could be adjusted following discussion with the GW medical monitor(s) if there were any clinical symptoms indicative of a safety concern. If, during the blinded phase, plasma concentrations of concomitant AEDs were found to be altered following administration of IMP, or if there were side effects suspected of being related to an elevation in the concomitant AED concentration, the investigator was to contact the GW medical monitor to discuss best management. Decisions were to be based on clinical symptoms and not plasma levels of AEDs.

• c. With an emphasis on adverse events, please discuss the external validity of the trial to real world practice considering this issue.

In this orphan population of refractory patients with severe TSC-associated epilepsy, polypharmacy is common in an attempt to achieve seizure control. Because of this, patients in the GWPCARE6 trial were taking up to 5 concomitant AEDs/numerous different combinations of AEDs, so any interpretation of potential drug-drug interactions is limited.

The Epidyolex[®] Summary of Product Characteristics includes comprehensive information for prescribing clinicians on “Interaction with other medicinal products and other forms of interaction” . As is mandatory for licensed drugs, the SmPC describes potential interactions with a number of drug classes, not just CYP3A4.

As stated clearly in the SmPC, it will be the decision of the prescribing clinician as to whether dose adjustments to other medicinal products used in combination with cannabidiol should be made in real world practice:

“A physician experienced in treating patients who are on concomitant antiepileptic drugs (AEDs) should evaluate the need for dose adjustments of cannabidiol or of the concomitant medicinal product(s) to manage potential drug interactions”.

A16. The GWPCARE6 CSR states that, “The use of rescue medication was allowed when necessary… Overall, 78 patients (34.8%) were recorded as taking rescue medications. ”[10]

• a. Please discuss which rescue medications were used during the trial.

The most common class of rescue medication was benzodiazepine derivatives. These included diazepam, clonazepam, midazolam, midazolam hydrochloride, lorazepam, clobazam and clorazepate dipotassium (see Table 5).

The most common rescue medication was diazepam.

• b. Please provide the frequency of use of rescue medications (by common class) per arm.

Similar proportions of patients across the treatment groups were recorded as taking rescue medications: 25 patients [33.3%] in the 25 mg/kg/day cannabidiol group, and 26 patients [34.2%] in the placebo group.

Table 5 below provides a summary of the use of rescue medications in the GWPCARE6 trial.

Table 5: Summary of rescue medications

==> picture [472 x 192] intentionally omitted <==

Data analyses

A17. Priority: Please provide subgroup analyses for:

• a. the efficacy and safety of cannabidiol as an add-on to usual care based on GWPCARE6 patients’ prior and concomitant seizure interventions; and

As outlined in questions A3.b and A9 above, in the GWPCARE6 trial of cannabidiol, patients entering the trial were permitted to be taking any AEDs/other treatments (except those listed as exclusion criteria) as long as they were stable during baseline and during the trial. These treatments/combinations of treatments are what are referred to as 'usual-care' in our submission. Cannabidiol was an add-on to usualcare.

Refractory patients with TSC-associated epilepsy may cycle through numerous AEDs in an attempt to achieve seizure control. As a result, ‘usual-care’ comprises many different AEDs/combinations of AEDs - there is no standard of care once a patient is refractory.

This was seen in the GWPCARE6 trial population. Patients entering the GWPCARE6 trial had already failed to achieve seizure control with a median of 4 (and up to 15) AEDs prior to entering the study, and were currently taking a median of 3 (and up to 5) AEDs.

As discussed during the ERG clarification meeting, and as shown in Table 2 and Table 3 above, the range of different drugs being taken, and thus the number of potential combinations, is huge.

The pre-specified subgroup analysis (see Figure 3 below) provided here demonstrates that the main concomitant AEDs in the GWPCARE6 study (clobazam, valproic acid, levetiracetam and vigabatrin) have no impact on the efficacy of cannabidiol.

This was similar for LGS and DS, and is why NICE decided for LGS and DS that the only relevant comparator was 'current clinical management' or 'usual-care'.

Figure 3: Subgroup analysis of the primary endpoint

==> picture [423 x 101] intentionally omitted <==

==> picture [423 x 100] intentionally omitted <==

==> picture [423 x 101] intentionally omitted <==

The safety and tolerability profile of cannabidiol is consistent, well-defined and manageable, as demonstrated across five randomised controlled Phase III trials in severe refractory epilepsies, (including TSC-associated epilepsy, Dravet syndrome and Lennox-Gastaut syndrome), where patients had tried and failed various AEDs, or were taking numerous combinations of concomitant AEDs.

In GWPCARE6, most adverse events were mild to moderate, transient and resolved by the end of the trial. The safety profile of cannabidiol observed in the GWPCARE6 study was consistent with findings from previous studies, with no new safety risks identified.

b. by the presence and absence of drug-resistant TSC-associated epilepsy.

All patients entering the trial had TSC-associated epilepsy that was not responding to their prior or current AEDs. One of the GWPCARE6 trial inclusion criteria was as follows: “Experienced at least eight seizures during the first 28 days of the baseline period with at least one seizure occurring in at least three of the four weeks”.

A18. Please specify whether the trial was powered to show the superiority of cannabidiol to usual care as an add-on.

We confirm that the trial was powered to show the superiority of cannabidiol to usual care as an add-on.

A19. The CSR states that a patient’s treatment assignment could be unblinded if unblinding was “ essential to make a decision on the medical management of the patient .”[10] Please specify how many patients were unblinded in each group?

One site unblinded a patient after they experienced an adverse event of rash with eosinophilia. Sponsor medical approval was not in place prior to unblinding. The patient was withdrawn.

A20. It is stated in the statistical analysis section that secondary endpoints are analysed using a per protocol analysis.

• a. Please clarify if this means that an intention to treat (ITT) analysis has not been used for these outcomes.

• b. If so, please provide an ITT analysis for these outcomes.

We apologise if this caused confusion. In the Statistical Analysis section of Document B, the sentence reads “Primary analyses used the ITT analysis set. Only the primary and key secondary endpoints were analysed using the per protocol analysis set.” The ITT analysis set was the primary analysis set for all efficacy endpoints.

Systematic review

A21. Priority: The ERG notes that despite the broadness of the eligibility criteria, no efficacy/ safety studies on common ASMs such as valproate, lamotrigine, oxcarbazepine, vigabatrin, levetiracetam, etc, appear to have been identified in the SLR. The ERG reran the conditions facets (lines #1 to #3) from the Embase strategy reported in Appendix D Table 1 of the company submission. The strategy was run as closely as possible to the company’s strategy using Ovid syntax and the same limits. A facet for anti-epileptics including the named drugs Valproate (or Valproic acid), Vigabatrin, and Lamotrigine was added and finally a randomised controlled trial (RCT) study design limit was applied. From the resulting 167 hits, 51 were deemed to be includes at title and abstract screening (please see Appendix 1). Of those 51, 2 were duplicate references and 3 were published in 2022 after the company submission searches were undertaken. Of the remaining studies, 2 appeared in the company submission list of included studies and 3 were listed as excludes. For the remaining 41 studies, please confirm:

a. If the papers in Appendix 1 were retrieved by your searches?

A general note about the SLR: we conducted a full SLR to identify all relevant studies on the efficacy and safety of pharmacological interventions for TSCassociated seizures. However, our submission to NICE focused on those studies that were most relevant to the decision problem of the use of cannabidiol as add-on therapy to usual-care. The studies of standard anti-epileptic drugs were considered to be describing the efficacy and safety of usual-care interventions and so were not reported in the submission. We have included the details of these additional studies in a supplementary document for completeness (see separate document entitled “Supplement_SLR_Additional_Studies_Apr2022”).

The table has been updated with the detailed answers to this question. In brief, only 2 of the citations identified by the ERG were not identified by our search for the full SLR, and these were both earlier conference abstracts of papers that had been included ([13] Franz et al 2017, [22] French et al. 2016).

b. If so, explain why they were excluded. Please introduce a column to the table in Appendix 1 with the reason for exclusion.

The table has been updated with this information (see separate document entitled “NICE_ERG_QueryReferences_CompleteTable_Apr2022”). Most of these studies were included in the full SLR but not in the submission as they were secondary publications of included studies, were listed in the table of secondary publications but not cited in the reference list, or were not considered relevant to the decision problem. Data from these studies has been summarised in the supplementary report that accompanies this response (see separate document entitled “Supplement_SLR_Additional_Studies_Apr2022”).

A22. Pertaining to question A21:.

• a. Please confirm if restrictions were placed on outcomes of interest during title & abstract, and full paper screening stages.

Abstracts and full texts were excluded if they did not report outcomes relating to seizure rates, severity or frequency or other epilepsy-related outcomes or did not relate to a general population with TSC who may or may not have seizures. Studies were excluded if they only assessed non-seizure-related manifestations of TSC. No other restrictions were placed during abstract screening.

Old conference abstracts with no poster or additional data available were generally excluded for not reporting enough data to be useful, but those with useful data were included unless a corresponding full-text publication was available that reported the same outcome data as the abstract.

• b. Please provide a table clearly outlining the PICOS (inclusion/ exclusion criteria) used during title and abstract, and full text screening in the SLR.

Inclusion/exclusion criteria used in the full SLR are shown in the table below:

A23. Please specify which risk of bias (RoB) tool was used for conducting the quality assessment of the GWPCARE6 trial. In addition, provide brief justifications for domain decisions.

The quality assessment of the GWPCARE6 trial and other RCTs was completed using the Cochrane Risk of Bias (ROB2) tool.[11] This was summarised in the submission and the full evaluation reported below, with further justifications for domain decisions for GWPCARE6 added.

A24. Please provide more information on the study selection, data extraction, and quality assessment process. Specifically, please state how many reviewers were involved at each stage, if these processes were conducted independently, how consensus was carried out, and if a third reviewer was involved in resolving disagreements.

Studies were selected for inclusion in the SLR if they met the inclusion criteria detailed in our previous response.

Two researchers independently screened each abstract and any discrepancies were agreed in discussion with the project leader.

One researcher and the project leader independently screened each full text publication to confirm that it met the inclusion criteria, with any disagreements resolved by discussion.

One researcher extracted data from included papers into an Excel template and a second researcher validated the data extraction. Any areas of uncertainty were checked by the project leader during the report synthesis and sign-off process.

Two researchers independently evaluated the risk of bias of each included study and the assessment was signed off by the project leader.

Section B: Clarification on cost-effectiveness data

Model structure

B1. Priority: The proportions for seizure type (generalized, focal with impairment, or combined) were based on week 16 data from the GWPCARE6 trial.

a. Please justify why the proportions were based on week 16 data and not on the average proportions from the trial (company submission, Appendix L).

As detailed in Document B, Section B.3.2.2, and as shown in Figure 4 below, the proportions change minimally over the 16-week trial period. Therefore, the base case analysis uses the Week 16 data from the GWPCARE6 trial as this was the point of completion of the Core Trial Period and is expected to reflect the distribution of seizure types following treatment over a longer time horizon more accurately.

b. Please explore the effect of using the average proportions in a scenario analysis.

The results for the scenario analysis based on the average proportions data reported are provided in Table 6. Note that the scenario is inclusive of the change in the base case requested as per question B13.a.

The scenario shows minor sensitivity to the change, with cannabidiol demonstrating a marginally higher incremental QALY and a slightly lower ICER.

Table 6: Scenario analysis results (PAS price)

• c. Please comment on comparability of seizure type proportions as observed in the blinded trial period of GWPCARE6 versus proportions observed in the open label extension (OLE) study. Are these values assumed to remain stable, or would non-response and discontinuation affect the relative proportions? Please provide clinical expert opinion or other evidence of clinical plausibility to support your answer.

The observed data in Figure 4 demonstrate that seizure type proportions are consistent over time. The data show that there is minimal change over the openlabel extension. Therefore, it is reasonable to assume that the Week 16 data from the GWPCARE6 trial is reflective of the OLE period and of the extrapolated Week 16 data over time. The Week 16 proportions are modelled to remain constant over time in both treatment arms; this assumes that discontinuation and stopping due to nonresponse does not affect the relative proportions.

This is supported by UK clinical expert opinion. In recent discussions conducted to inform our responses to the ERG, Professor Finbar O’Callaghan and Dr Sam Amin confirmed that the seizure type proportions are generally stable over time.

Figure 4: Observed proportion of patients by seizure type: Core Trial Period and OLE

==> picture [470 x 220] intentionally omitted <==

Key: Focal Imp - Focal with impairment of awareness seizures Note: Patients in the cannabidiol 25 mg/kg and placebo arms are pooled to calculate the average observed seizure frequency by seizure type

Population

B2. Although the modelled population is said to be patients ≥2 years old, the GWPCARE6 ITT population that was used to inform the model included also children aged 1 year-old.

• a. Please comment on the impact of including the 1-year-olds for estimating efficacy inputs, discontinuation rates, TSC-associated neuropsychiatric disorders (TAND) responses and patient baseline characteristics.

The population enrolled in the GWPCARE6 trial included nine children aged <2 years old at the date of screening (three in the cannabidiol 25 mg/kg/day arm and six in the placebo arm) who were included in the GWPCARE6 ITT analysis.

The GWPCARE6 ITT analysis, including data for the children aged <2 years old at screening, was used to inform model inputs including efficacy inputs, discontinuation rates and TAND response rates and patient baseline characteristics in the costeffectiveness analysis as detailed below:

• Patient level data from the GWPCARE6 blinded trial period are utilized within the regression analysis to predict the expected probability of seizure-free days and associated seizure frequency

• Data on patients discontinuing treatment from the GWPCARE6 blinded trial period and open-label extension are used to calculate the proportion of patients expected to discontinue treatment during the trial period and OLE period

• Patient level data on treatment response (≥ 30% reduction in seizure frequency) from the GWPCARE6 blinded trial period and the open-label extension are used to calculate stopping rates

• Data on patients who experienced a ≥50% response (reduction in seizure frequency) from the GWPCARE6 blinded trial period and open-label extension are used to calculate TAND response rates, which are then used to calculate the proportion of patients who may benefit from TAND mitigation.

• Patient baseline characteristics are used to calculate drug costs and mortality. Note: patient characteristics used to calculate drug costs exclude patients aged 1 year-old.

As shown in Appendix L, Figure 7, the subgroup analysis demonstrates that age is not a treatment effect modifier.

In addition, as shown in Table 7 below, all but two of the patients who were <2 years old at the screening visit had reached age 2 by the end of the trial. Therefore, it is reasonable to assume that the outcomes in the <2 year olds would be similar to the overall trial outcomes. Excluding these patients would unnecessarily reduce patient numbers in an already small trial population (due to the nature of the orphan disease), break trial randomisation and increase model uncertainty by reducing sample size.

• b. Please provide these data separately for the 1-years olds in the GWPCARE6 ITT population or perform a scenario excluding the 1-year-olds from the ITT population in estimating model inputs.

Table 7 shows the baseline characteristics of the patients from the cannabidiol 25 mg/kg/day and placebo arms who were <2 years old at the date of screening for the GWPCARE6 study.

These baseline characteristics are similar to the baseline characteristics of the overall population.

Due to the very small number of patients aged <2 years and considering that all except two patients had reached age 2 by the end of the trial, it is not expected that stopping rule rates, discontinuation rates or TAND response rates would be significantly different for these patients compared to the overall outcomes.

Therefore, separate outcome data for these patients are not provided.

Table 7: Patients <2 year of age at screening - baseline characteristics

*Note : patient’s actual age was ***** years

Key: F, female; M, male; W/C, White/Caucasian

B3. In Table 12 of the company submission which displays the baseline characteristics of the model population, the % female varies quite drastically between age categories. Please check this is correct and highlight whether and where in the model the % female has an effective impact.

We confirm that the data are correct. Please note the % female input by age group is not used in the cost effectiveness analysis. It is provided to allow for a measure of comparability across age groups, as mean age, mean weight and mean BSA all change with age (as shown in Table 8). The observed range reflects the demographics of a rare orphan disease population. It is reasonable to observe variation across patient groups where there are small patient numbers.

Please note that we do not consider the variation to be ‘drastic’ (p-value: 0.453); the largest variation relative to the other age categories is observed in the patient group with the smallest patient numbers (aged 12-17 years [Table 8]).

Table 8: GWPCARE6 population baseline characteristics by age group

Intervention, technology and comparators

B4. Priority: The final scope mentions the following treatments as comparators: anti-seizure medications (ASMs), everolimus, vagus nerve stimulation, ketogenic diet, and surgical resection. In the company submission the comparator reflects the control arm of the GWPCARE6 trial, in which patients were treated with ASMs, and on ketogenic diet, vagus nerve

stimulation, and who underwent surgery more than 6 months before screening.

• a. In Section 3.2.3.2. it is stated that the comparator in the costeffectiveness analysis is usual care, consisting of a combination of ASMs. Please provide a justification for not including the ketogenic diet, vagus nerve stimulation, and surgery, as comparators in the economic model.

The comparator in the final NICE scope is “Established clinical management without cannabidiol”.

Ketogenic diet, vagus nerve stimulation (VNS) and resective surgery are part of the treatment pathway for TSC-associated epilepsy (see Figures 3 and 4 in Document B of the company’s submission), and therefore part of the ‘established clinical management’ (or ‘usual-care’) mix that cannabidiol would be added to.

Use of a ketogenic diet, VNS and/or prior surgery was not an exclusion criterion in the GWPCARE6 clinical trial. Any non-pharmacological interventions for epilepsy (e.g. ketogenic diet and VNS) had to have been stable for 1 month prior to screening and throughout the duration of the trial. Patients who had undergone prior surgery for epilepsy were also not excluded, provided that the surgery was not within the 6 months prior to screening. Approximately 1.3% and 11.2% of patients in GWPCARE6 were on a ketogenic diet and VNS, respectively, at baseline, spread fairly evenly across the treatment arms.

There is no evidence to suggest that levels of use of ketogenic diet, VNS and/or resective surgery would differ greatly between patients receiving usual-care only or receiving usual-care plus cannabidiol. Therefore, any effects of these treatments would apply equally to both arms of the model. Similarly, costs of the ketogenic diet, VNS and/or surgery, and disutilities associated with adverse events/complications, would apply equally to both arms.

Based on the above, ketogenic diet, VNS and resective surgery are already included in the comparator by virtue of their contribution to both arms of the model as part of the ‘usual-care’ mix.

b. Please provide scenario analyses with usual care including a combination of ASMs, ketogenic diet, vagus nerve stimulation, and surgery, as comparator in the economic model.

Please see the response to question B4.a above.

Given that ketogenic diet, VNS and/or resective surgery are part of ‘usual-care’ and can reasonably be considered to contribute equally to the treatment effect with and without cannabidiol, none of these interventions in isolation is considered an appropriate comparator to cannabidiol.

c. Everolimus was not included in the company submission as a

comparator but as later line treatment (see question A4). Please provide

a scenario with everolimus as a comparator.

The comparator in the final NICE scope is “Established clinical management without cannabidiol”.

For the reasons outlined in Document B of the company’s submission and in the answers to question A4. above, the company does not consider everolimus in isolation to be a relevant comparator. However, as noted here, we have included it in the model as a later line treatment.

Please also refer to the answers to question B6. below.

B5. The efficacy (and safety) data of cannabidiol 25 mg/kg/day from the GWPCARE6 trial was used to populate the cannabidiol arm in the cost-effectiveness model. The average dose of cannabidiol used in the cost-effectiveness model to calculate the drug costs is 12 mg/kg/day based on the early onset of effect observed in the GWPCARE6 trial and similar results found in a real-world study.

• a. Please explain the choice for an average dose of 12 mg/kg/day in the basecase analysis and how this would reflect ‘a spectrum of doses ranging from ≤10 mg/kg/day to the maximum of 25 mg/kg/day’.

Since the objective of the cost-effectiveness modelling is to represent a cohort in UK clinical practice, a dose that represents this cohort has been used in the model,

rather than considering the dosing of individuals. For this reason, the dose used in the model to calculate ICERs is an average dose.

According to the Epidyolex Summary of Product Characteristics, for TSC-associated seizures, the dose should be increased to 10 mg/kg/day and then the clinical response and tolerability should be assessed. Based on individual clinical response and tolerability, each dose can then be further increased only if needed .

Thus, an average dose of 25 mg/kg/day would imply that some patients are above the maximum licensed Epidyolex dose in TSC of 25 mg/kg/day (i.e. off-label use) or that all patients are treated with the maximum daily dose of 25 mg/kg/day, which is not plausible.

By using an average dose of 12 mg/kg/day case in the model, we can account for the range of doses seen in clinical practice across a cohort of UK patients with TSCassociated epilepsy, as clinicians aim to optimize the dose for individual patients. As stated in our original submission, in real-world clinical practice, there will be a spectrum of doses ranging from ≤10 mg/kg/day to the maximum of 25 mg/kg/day.

Feedback from clinical experts and real-world data support this assumption of using an average dose of 12 mg/kg/day in the model:

• Since the NICE submission, the company has obtained data from a German dispensing database (INSIGHTS) on real-life dosing. The daily dose was estimated from a group of patients with Epidyolex prescriptions in 2021. The indication for the prescription was not available in the database, therefore a TSC diagnosis was inferred from a record of any use of vigabatrin or everolimus in the preceding 3 years. Body weight of patients was estimated from age and gender average weight in the German general population. From a total of 118 patients, the observed median dose was 12.21 mg/kg/day in children (inter-quartile range 6.67) and 7.77 mg/kg/day (IQR: 5.68) in adults

• Discussions across Europe with expert clinicians who have experience of using Epidyolex in clinical practice suggests that the average dose in real-world clinical practice will be around or below 12 mg/kg/day

• b. Please add options for 15, 20 and 25 mg/kg/day to the cannabidiol dose scenario in the model (instead of now only 10 and 12 mg/kg/day) to explore effects on costs and ICER.

This option has now been included in the model.

However, for the reasons outlined in question B5.a above and B5.c below, we strongly disagree that average doses of 15, 20 and 25 mg/kg/day would be representative of the likely dosing of cannabidiol in real-word practice in TSCassociated epilepsy.

• c. Please justify the assumption that a dose of 12 mg/kg/day has the same efficacy and safety as 25 mg/kg/day. The lack of a dose response relation in Lennox-Gastaut syndrome and Dravet syndrome may not be generalizable to TSC; please also provide evidence on lack of a dose response relation (between the 25 and 50 mg/kg/day doses) from GWPCARE6.

The company considers that the efficacy outcomes from the GWPCARE6 study in TSC-associated epilepsy will be reflective of patients receiving lower average doses in clinical practice.

The totality of the evidence in the cannabidiol clinical trial programme in refractory epilepsies to date does not support a clear dose response above 10 mg/kg/day. This was also concluded by the EMA in setting the maintenance dose of 10 mg/kg/day for DS and LGS.

According to the Epidyolex Summary of Product Characteristics, for TSC-associated seizures, the dose should be increased to 10 mg/kg/day and then the clinical response and tolerability should be assessed. Based on individual clinical response and tolerability, each dose can then be further increased only if needed .

As outlined in Document B of the company’s submission, the early onset of efficacy at lower doses observed during titration in GWPCARE6 (TSC-associated epilepsy) is consistent with the demonstrated efficacy of cannabidiol at 10 mg/kg/day from the Phase 3 studies GWPCARE3 (in LGS) and GWPCARE2 (in DS).

Together with the lack of dose response observed between 10 mg/kg/day and 20 mg/kg/day in the DS and LGS studies, the available clinical data across five pivotal

Phase 3 trials support the potential for cannabidiol efficacy to become clinically apparent at doses much lower than 25 mg/kg/day.

Furthermore, the company considers that the evidence from the other severe epilepsy syndromes, LGS and DS, is applicable to TSC-associated epilepsy for the following reason:

• Seizures are not a disease in themselves: they are a manifestation of different disorders that can lead to abnormal neuronal activity in the brain. The term epilepsy refers to the disease, disorder or syndrome involving recurrent seizures. Terms to describe and classify types of seizure have been developed by the International League Against Epilepsy (ILAE) broadly based on how and where the seizure begins in the brain (focal or generalised) and the person’s level of awareness during a seizure (aware or impaired awareness). This ILAE operational classification of seizure types can be used to classify seizures across different aetiologies, that is, it is specifically designed to be applied to seizures where the underlying disease/syndrome is different. Thus, although a patient may be diagnosed with a particular epilepsy/epilepsy syndrome based on specific clinical/diagnostic features, each individual seizure type experienced within those epilepsy syndromes/disorders will be the same, according to the ILAE classification.

• Therefore, although the type(s) of seizures that predominate may vary between severe epilepsies/syndromes such as those caused by TSC, DS and/or LGS, the individual seizures are the same. For example, a generalised tonic-clonic seizure resulting from TSC-associated epilepsy would be similar to a generalised tonicclonic seizure as a result of Dravet syndrome, or an atonic seizure resulting from TSC-associated epilepsy would be similar to an atonic seizure as a result of Lennox-Gastaut syndrome.

• Since cannabidiol is specifically treating the seizures caused by TSC (and DS/LGS) and not the underlying disease or syndrome itself, it will work in the same way to reduce seizures whether they are caused by TSC, DS or LGS.

• Based on the above, we consider that the evidence for cannabidiol in DS and LGS is generalizable to TSC-associated epilepsy.

B6 . Everolimus was considered as a subsequent treatment in the model for 7.7% of the cohort. Patients in the cannabidiol arm receive everolimus after discontinuation of cannabidiol, patients in the placebo arm are assumed to receive everolimus at 2 years after the trial period.

• a. Please justify the 2yr + 16-week year time point at which patients in the placebo arm would receive everolimus and provide supporting evidence.

There is limited information regarding the positioning or timing of everolimus as a later line treatment in TSC-associated epilepsy.

In the NHS England Clinical Commissioning Policy (see question A4.a above), everolimus is positioned as a later line treatment due to the need not only for inadequate response to two or more AEDs, but also for failure/ruling out of both epilepsy surgery and vagus nerve stimulation (VNS) by a multidisciplinary team.

The TOSCA registry indicates that everolimus is used in only a small proportion

• (7.7%) of patients.[12] There may be various reasons for this, including, for example:

• Clinicians may have concerns regarding the use of mTOR inhibitors in children, specifically regarding their effect on growth and gonadal function over the longterm[13]

• Standard recommendations advise against concomitant use of everolimus with ketogenic diets in children with epilepsy, due to additive toxicity with hyperlipidaemia[13]

• Patients/caregivers may prefer to try other AEDs with titration/monitoring/sideeffect profiles perceived as being more manageable before moving to everolimus, due to concerns about using a drug that has a toxicity profile associated with its original indications in oncology/immunosuppression.

For the purposes of the model, it was assumed that, at 2 years post the trial, placebo plus usual-care patients will have tried other AEDs, and a proportion of these patients would then receive treatment with everolimus as an option later in the treatment pathway.

Two scenarios are provided to address the uncertainty in this assumption, assuming a 1-year delay and 3-year delay post the trial for patients to cycle through other AEDs before considering everolimus as a treatment option. Note that the scenario is inclusive of the change in the base case as per question B13.a.

The results for the scenario analysis based on varying the start date for a proportion of usual-care patients receiving everolimus, are provided in Table 9 (1-year delay) and Table 10 (3-year delay).

The scenarios shows low sensitivity to the change, with cannabidiol demonstrating a marginal change in incremental costs (lower for a 1-year delay, higher for a 3-year delay) and ICER (lower for a 1-year delay, higher for a 3-year delay).

Table 9: Scenario analysis results (PAS price) – usual-care proportion of

patients start everolimus delay – 1 year

Table 10: Scenario analysis results (PAS price) – usual-care proportion of

patients start everolimus delay – 3 years

• b. The proportion of 7.7% was based on the TOSCA registry.[12] Please comment on how representative the population in the TOSCA registry is when considering UK clinical practice for the population in this appraisal.

TOSCA is a multicentre, international disease study designed to collect data, retrospectively and prospectively, on patients with TSC from countries worldwide.

Recognising that TSC-associated epilepsy is an orphan disease, we consider that the TOSCA registry provides the most comprehensive and representative data source currently available. Almost 60% of the patients in the TOSCA registry were from European countries, with 32 patients from the UK.[12]

• c. Please justify that the proportion of 7.7% may be considered to apply to both the cannabidiol and usual care (placebo) arm and provide supporting evidence.

In the model, we have assumed that 7.7% of patients discontinuing cannabidiol or continuing on usual-care would receive treatment with everolimus later in the TSC-associated epilepsy treatment pathway. This proportion was based on data from the TOSCA registry, which represents the best currently available data for patients taking everolimus for TSC-associated epilepsy.

Given that refractory patients with TSC-associated epilepsy may cycle through numerous AEDs in an attempt to achieve seizure control and that there is no standard of care once a patient is refractory, we do not consider it unreasonable to assume that a similar proportion of patients in both arms will eventually receive everolimus.

• d. Only drug costs for everolimus were considered in the model. Please justify why effects were not considered.

The effect of everolimus is not incorporated into the model for several reasons:

• A similar proportion (7.7%) of cannabidiol patients who discontinue treatment and patients in the placebo plus usual-care arm receive everolimus as a subsequent treatment, therefore the impact on effects would be expected to be similar across arms

• There are no data to support the effect of everolimus post treatment with cannabidiol

• The licence for everolimus is restricted to partial-onset seizures only: this does not fully reflect the modelled population

Therefore, in the absence of any relevant clinical data, the efficacy impact could not be included in the analysis. We have adopted a pragmatic approach and examined everolimus as a later line treatment by applying the cost impact (applied as a one-off cost) to a proportion of patients in both model arms.

• e. Please provide a scenario taking into account an effect of everolimus on seizure frequency.

As per our response to question B6.d above, it is not possible to provide a scenario taking into account the effect of everolimus on seizure frequency when given as a later line treatment.

• f. Please provide a scenario where everolimus is excluded as a subsequent treatment.

A scenario excluding everolimus as a subsequent treatment is provided below in Table 11. Additionally, it is provided in Appendix 2, alongside the updated sensitivity analysis that was included in the original submission. Note that the scenario is inclusive of the change in the base case as per question B13.a. The scenario shows low sensitivity to the exclusion of everolimus, with a marginal change in incremental costs.

Table 11: Scenario analysis results (PAS price)

Clinical parameters and variables

B7. Priority: Regression models were used to estimate the effectiveness of cannabidiol compared with usual care.

• a. Please provide any predefined statistical plan that was made for the regression modelling and justify the appropriateness of the analyses if no predefined statistical plan was used.

Text from the analysis plan designed to support the economic analysis is provided below.

The previous NICE model for cannabidiol developed for TA614 and TA615 used transition matrices to capture the efficacy for each treatment in terms of seizure frequency and seizure-free days, as shown in Figure 5 (for DS; the LGS model structure is similar but with different seizure numbers/seizure-free days). Given the incorporation of both seizure frequency and seizure-free days in this structure, a total of nine separate transitions were required to be populated.

The transition matrices provided a simple model structure but led to criticism due to the perceived arbitrary choice of bands and wide range of seizure frequency.[14, 15] In addition, given the low sample size in the DS and LGS on-clobazam data (86 patients and 74 patients, respectively), using the health states to further divide the patient population led to very small sample sizes in some transitions. As both of these structural issues are still relevant for the analysis of the GWPCARE6 trial data, alternative methods were considered.

Figure 5: DS model structure (28-day cycle)

==> picture [373 x 108] intentionally omitted <==

==> picture [373 x 107] intentionally omitted <==

Key: DS, Dravet syndrome.

Regression methods

Regression approaches were considered beneficial in terms of their ability to estimate the proportion of seizure-free days and not just seizure frequency without further subdividing the data into additional heath states to accurately capture the impact on quality of life in patients and caregivers. Based on clinical feedback, the need to capture the benefit of seizure-free days specifically is important given the substantial quality of life benefit for patients and caregivers associated with seizurefree days.

The daily data from the core randomised controlled trial aspect of GWPCARE6 were used for the regression analyses. It was not feasible to incorporate the OLE data into the regression analysis to predict seizure frequency and seizure-free days as the OLE data were collected on an approximately weekly basis and the number of seizure-free days per week were not recorded. The OLE data were instead used for validation (see Document B, Figure 16 in the company submission).

Three model structures were considered to model the seizure frequency and seizure-free days of patients in GWPCARE6:

• Generalized estimating equations (GEEs) for generalized linear models – this approach estimates population-averaged model parameters and their standard

errors. In the presence of missing data, GEE requires the strong assumption of missing completely at random; however, it is unclear whether this assumption holds for the GWPCARE6 data.[16] In addition, GEE models may yield biased results when the sample size is small, as such this method was not considered further.[17]

• Hurdle models – a two-part model that specifies one process for zero counts (seizure-free days) and another process for positive counts (seizure days)

• Generalized linear mixed models (GLMMs) – independent mixed effects regression modelling of the proportion of seizure-free days and seizure frequency was considered over more simplistic regression models to estimate count data (such as standard Poisson and single negative binomial models) as it allowed for more explicit estimation of the proportion of seizure-free days (which may be underestimated using more simplistic methods)

The independent modelling of seizure-free days and seizure frequency using GLMMs was favoured over hurdle models due to the ease of interpretation and increased flexibility in the modelling. As such, results for the independent modelling of seizure-free days and seizure frequency were preferred and were included within the company’s submission.

Independent modelling of the proportion of seizure-free days and seizure frequency

This approach analysed weekly data utilizing a two-step regression approach (described below) to predict seizure-free days and seizure frequency on seizure days; regression models were fitted using the lme4 R package.[18, 19]

1. The proportion of seizure-free days per cycle (7 days) was modelled using binomial regression. Binomial regression was used as the data on seizure-free days are dichotomous with two possible outcomes: a seizure-free day or a nonseizure-free day. Therefore, the proportion of seizure-free days per week can be estimated

   • a. It is acknowledged that, by analysing these data as a proportion, the probability of each seizure-free days in the cycle per week should be

independent and constant; however, independence is not satisfied as multiple records per week are used to calculate this proportion per patient (as the data are recorded daily). Given the variability in seizure frequency day to day and the expectation that, within a cycle of one week, the chance of a seizure-free day is anticipated to be similar, it was assumed that seizure-free days are independent of each other within the analysis

• b. There are a limited number of cycles where data are not collected each day in the cycle. The estimated proportions become more variable for cycles where the data collected are very limited. As such, the proportion of seizurefree days is only estimated for cycles where 3 or more days of data are available

2. Subsequently, a negative binomial regression model was fitted to the subset of seizure frequency data for days on which patients experienced at least one seizure. Negative binomial regression was used as the seizure frequency data are count data. A negative binomial model was chosen over Poisson regression as the assumption that the variance is equal to the mean is relaxed. This regression type been described as a good choice to estimate transition probabilities in the presence of small sample sizes and where data is overdispersed[20]

   • a. Specifically, the subset of seizure frequency data for days on which patients experienced at least one seizure is generated, and the total seizures in the cycle are calculated along with the number of records for which data are available. For example, for a patient who has three seizure-free days in a cycle, the total number of seizures in the remaining 4 days would be calculated (and the number of records would be four)

   • b. An offset to account for the differing ‘exposure’ is included in the model. This accounts for any missing days. For the example above, it is important to recognize that the number of seizures that the patient experienced was over 4 days rather than the possible seven.

   • c. The fitted negative binomial model estimates the seizure frequency per day, conditional on patients experiencing at least one seizure (i.e. seizure

frequency on non-seizure-free days), as the number of records with seizures per week is included as an offset within the analysis. Seizure frequency for the appropriate number of seizure days can be estimated by including the offset when using the model to predict seizure frequency.

• b. The regression models have both a random intercept and random slopes. Please justify adopting random slopes (instead of only using a random intercept).

Random effects were incorporated into the regression analyses as the analysis datasets include repeated measures for each patient for which there is an inherent correlation between observations of the same patient. Two levels of random effect were applied in the model:

• Random intercept – the intercept value is assumed to follow a distribution and each patient may have a different intercept value

• Random slope – the change in outcomes over time will follow a distribution and the rate of change will vary by each patient (i.e. some patients may improve faster than others, while others may decline over time). Without the inclusion of the random slope, the analysis assumes that the change in outcomes is the same for all patients, which is a strong assumption to make

To assess further the inclusion of the random slope, the regression model was fit with and without the random slope and the goodness of fit statistics were compared. Table 12 displays the goodness of fit statistics for each of the regression models with and without the inclusion of the random slope.

Typically, if the difference in the AIC and BIC score between two models is <5, the models are deemed to fit equally as well. When random slopes are included within the regression models, both the AIC and BIC scores (presented in Table 12) improve in both models. The goodness of fit statistics improve by over 280 for the seizurefree days regression model, and by over 160 for the seizure frequency (on seizure days) regression model.

This improvement therefore suggests that the inclusion of the random slope drastically improves the goodness of fit for both regression models. In turn, this supports the assumption that all patients do not have the same response over time.

Table 12: Goodness of fit statistics – regression models with and without a

random slope

c. Please provide an option in the model to allow the regression models to have only random effects in the intercept (and not in the slopes).

An option has been included in the cost-effectiveness model to exclude the random slope from the regression. The results of this analysis is presented in Table 15 and demonstrates that removing the random slope coefficient has a limited impact on the ICER. Table 13 and Table 14 present the coefficients for the seizure-free days and seizure frequency (on seizure days) regression models respectively.

As per the response to question B7.b above, the inclusion of the random slope improves the goodness of fit for both models and it is clinically implausible for all patients to have the same response over time. As such, we maintain that the base case analyses including this random effect are the most appropriate.

Therefore, the scenario result excluding the random slope (Table 15) should be interpreted with caution.

Table 13: Binomial seizure-free day model without random slope – coefficients

Table 14: Negative binomial seizure frequency (on seizure days) model without

random slope – coefficients

Table 15: Scenario analysis results (PAS price) – using regression model

without a random slope

d. Please elaborate on the procedures to select covariates for the

regression models. What candidate covariates and interaction terms were considered and why? What were the criteria to in- or exclude covariates and interaction terms?

Covariate adjustment was generally considered within the regression analyses for both prognostic factors and treatment effect modifiers to estimate outcomes at a cohort level. However, adjustment for prognostic factors was not considered necessary within this analysis as patients in the GWPCARE6 trial were randomized and patient characteristics are observed to be well balanced.

To determine whether adjustment for treatment effect modifiers was required, the subgroup analysis results for the primary endpoint from GWPCARE6 were explored. This showed that age group, sex, region, clobazam use, valproic acid use, levetiracetam use, vigabatrin use, baseline TSC-associated seizure frequency, number of concurrent AEDs, number of prior AEDs and composite of number of prior AEDs and concurrent AEDs were not observed as statistically significant treatment effect modifiers (see Figure 7, in Appendix L, Document B in the company submission). Therefore, covariate adjustment for treatment effect modifiers was also not deemed necessary. If a characteristic had been observed to be a treatment effect modifier, then an interaction term with treatment would have been considered. Although additional adjustment is not required, to accurately capture changes over time, the following covariates were included within the regression analyses:

• Treatment (cannabidiol 25 mg/kg/day or placebo)

• Cycle (log transformed; continuous covariate) – by including treatment cycle as a continuous covariate, extrapolating the covariate would assume that a patient’s outcomes would continue to improve indefinitely with time. A log transformation was therefore also considered, which slows the improvement of outcomes over time. Note, this is discussed further in the response to question B7.e.

• Treatment by cycle interaction – it is anticipated that the change in seizure frequency and seizure-free days over time will differ by treatment. As such, an interaction term was considered

• Baseline seizure frequency per week (continuous covariate) – the impact of treatment is anticipated to differ depending on the level of seizure frequency the patient is experiencing prior to treatment. As such, a covariate for baseline seizure frequency was estimated to accurately capture the variation in the patient population. To provide one overall estimate of baseline seizure rate to include in

both the regression models, the baseline seizure frequency covariate includes days where patients had no seizures in the estimate. Note that the baseline seizure frequency covariate was centred to prevent issues with convergence

• Offset for number of records. The offset term is used to account for the different ‘exposure’ for patients, as some patients may have less than 7 days of data recorded in a week. The offset term assumes there is a linear relationship between the number of seizures and the number of records. To incorporate an offset term into a negative binomial analysis, an additional covariate is added to the regression equation; however, the coefficient for the offset covariate is fixed to a value of one and the log of this value is taken to match the link function

• Random intercept – the intercept value is assumed to follow a distribution and each patient may have a different intercept value

• Random slope – the change in outcomes over time will follow a distribution and the rate of change will vary by each patient (i.e., some patients may improve faster than others while others may decline over time). Note: this is discussed further in the response to question B7.b.

e. A log-transformation was performed to reduce the improvement that was associated with increasing cycles (over time). Please justify why a log-transformation is preferred for this purpose.

Cycle was considered as a proxy for time within the regression analysis. However, it was determined that the inclusion of cycle as a continuous covariate without transformation would not be appropriate for inclusion within the regression model, as this would assume that patients would continuously improve over time, which is unlikely to hold for the full-time horizon.

The log transformation is a common transformation for positive, right skewed data which in this case allows for the improvement in patients to reduce over time. As such, this was utilized within the regression analysis.[21] Use of a categorical covariate for cycle was also considered. However, this would rely on arbitrarily selecting a number of categories and the cut-off times for each of the categories.

f. Please provide details/results of any other transformations considered and/or tested.

Table 16 presents the goodness of fit statistics for the regression models using no transformation, square root transformation and the log transformation on cycle. The square root transformation was not included in the response to question B7.e but is included here as it is considered as an alternative to the log-transformation for right skewed data.[21] Similar to the log-transformation, the square root transformation also reduces the improvement of patients over time, but the rate of change may be slower.

For each of the regression models, the results in Table 16 indicate that the logtransformation on cycle provides the best fitting model, whereas the models without any transformation provide the worst fitting models based on both AIC and BIC. For the seizure frequency on seizure days regression model, both the AIC and BIC for the model using the square root transformation are within 5 points of the scores for the log-transformation model, suggesting the goodness of fit for these models are comparable.

Using the square root transformation on cycle has very little impact on the ICER (see Table 17). This option has been included in the model.

Table 16: Goodness of fit statistics – regression models using different transformations on cycle

Table 17: Scenario analysis results (PAS price) – using square root

transformation on cycle.

g. Please provide diagnostics of the regression models, including multicollinearity.

Diagnostic plots to assess the suitability of the fitted binomial seizure-free days model are presented in Figure 6 and Figure 7. There are no clear issues with the model fit (with the exception of one outlying value for the Pearson residuals; see Figure 7). The assumption of normality of the random effects is also demonstrated to be an approximately appropriate assumption (see Figure 6). The slight bow in the slope random effect quantile-quantile plot should be considered as a potential limitation of the model. However, the need to include a random effect for slope is clear based on the goodness of fit statistics (see response to B7.b).

Diagnostic plots to assess the suitability of the fitted negative binomial model for seizure frequency are presented in Figure 8 and Figure 9. The assumption of normality of the random effects is demonstrated to be a reasonable assumption (see Figure 8). The slight bow observed in both the intercept and slope random effect quantile-quantile plot should be considered as a potential limitation of the model. However, the need to include both random effect terms is clear based on the goodness of fit statistics (see response to B7b). The residual plots in Figure 9 show some outlying observations for the lower fitted values (i.e. fewer seizures); however,

this may be expected given the data are highly concentrated. In addition, the associated histograms suggest the residuals are approximately normally distributed suggesting there are no major issues with the model fit.

The Spearman’s rank correlation coefficient has been used to assess the correlation between the predictors included in the model (treatment, baseline seizure frequency (scaled) and log[cycle]). Results for the seizure-free days and seizure frequency model are presented in Table 18 and Table 19, respectively. Results suggest that there is minimal correlation between variables (values < 0.1).

Figure 6: Diagnostic plots for the random effects terms – binomial seizure-free

days model

==> picture [452 x 170] intentionally omitted <==

==> picture [452 x 170] intentionally omitted <==

Key: QQ, quantile-quantile; REs, random effects.

Figure 7: Diagnostic plots for the residuals – binomial seizure-free days model

==> picture [452 x 170] intentionally omitted <==

==> picture [452 x 171] intentionally omitted <==

Figure 8: Diagnostic plots for the random effects terms – negative binomial seizure frequency model

==> picture [452 x 170] intentionally omitted <==

==> picture [452 x 170] intentionally omitted <==

Key: QQ, quantile-quantile; REs, random effects.

Figure 9: Diagnostic plots for the residuals – negative binomial seizure

frequency model

==> picture [452 x 170] intentionally omitted <==

==> picture [452 x 170] intentionally omitted <==

Table 18: Spearman's rank correlation coefficient – binomial seizure-free days model

Table 19: Spearman's rank correlation coefficient – negative binomial seizure frequency model

h. Please clarify and justify how missing data were handled for the estimation of the regression models.

Within the core trial period, data were reported on 15,339 out of 16,180 possible days (~95%). Therefore, the amount of missing data during the core trial period is low, suggesting that the impact of the missing data may be negligible and that imputation approaches are not required.[22]

However, as detailed in response to B.7a, additional considerations were made within the regression analysis:

• For the seizure-free days regression analysis, the proportion of seizure-free days per week is estimated rather than the absolute number of seizure-free days per week. Given that the estimated proportions become more variable where data are limited, the proportions are estimated for cycles where at least 3 days of data were available; note, the seizure days in these cycles were still included within the seizure frequency analysis

• For the seizure frequency on seizure days regression analysis, an offset term was included to account for the different ‘exposure’ for patients, as some patients may have less than 7 days of data recorded in a week. For example, if data are only available for 5 days in week, then the proportion of seizure days and the seizure frequency on seizure days may be applied to a 7-day period.

B8. Priority: Long-term extrapolation is based on 16 weeks data. The company submission states that ‘ The base case analysis assumes that the relative treatment effect is consistent and maintained over the model time horizon, whilst patients are on cannabidiol plus usual-care treatment .’

• a. Please confirm that this is implemented by keeping all proportions fixed after 16 weeks, which implies that all patients maintain week 16 seizure frequency and seizure free days and can only move (worsen) when they discontinue cannabidiol or die.

We confirm that all patients on treatment with cannabidiol plus usual-care maintain week 16 seizure frequency and seizure-free days, and move when they discontinue cannabidiol or die.

• b. Please comment on the clinical plausibility of the model result that none of the patients in the usual care arm are seizure-free during the entire time horizon, given that natural variation is seen in TSC-related epilepsy. Please provide clinical expert opinion as well as supporting empirical evidence to accompany the response.

As described in Document B of the company’s submission, TSC-associated epilepsy presents a significant clinical challenge as up to two-thirds of patients are refractory to currently available treatments and do not achieve seizure control.

In the GWPCARE6 Phase 3 clinical trial, patients had tried and discontinued up to 15 AEDs (median 4) in order to try to achieve seizure control, with some taking up to 5 AEDs concurrently (median 3) and still not gaining control of their seizures.

Therefore, it is highly unlikely that a patient who is already refractory would become seizure-free just by continuing on their existing usual-care treatment.

The clinical plausibility of this was supported by feedback from recent discussions with two UK clinical experts, Professor Finbar O’Callaghan and Dr Sam Amin.

c. Please comment on the clinical plausibility of the assumption that the relative treatment effect of cannabidiol is maintained over a patient’s lifetime, whilst on treatment.

There is currently no clinical evidence to suggest that the Epidyolex treatment effect falls over time. Epidyolex has shown sustained durability of effect over the OLE period of GWPCARE6 (TSC-associated epilepsy) and in the GWPCARE5 OLE study (DS/LGS). In addition, a stable and durable long-term effect with Epidyolex has also been evidenced in a US Expanded Access Programme.

A reduction in treatment effect is already ‘built in’ to the economic model via conservative assumptions on long-term discontinuation rates and the use of stopping rules. This reflects clinical practice, and is evidence-based.

The use of these conservative and evidence-based discontinuation assumptions and stopping rules throughout the model means that patients stay in the model in the longer term only as long as they remain on treatment and are receiving a sufficient treatment effect . For many patients, this is a relatively short time in comparison to the time horizon of the model.

Table 20 below shows that, at 10 years, around 9% of patients remain on treatment. By 20 years, this has fallen to less than 6% of patients.

Table 20: Proportion of patients remaining on treatment (cannabidiol) over

time

Note: the proportion of patients on treatment is inclusive of the change in the base case as per question B13.a.

d. Please provide a scenario where the effect of cannabidiol whilst on treatment decreases over time, to explore the impact of the above assumption on the costs, QALYs, and the ICER.

Feedback from recent discussions (conducted to inform our responses to the ERG) with two UK clinical experts (Professor Finbar O’Callaghan and Dr Sam Amin) indicates that, in real-world clinical practice, a patient with TSC-associated epilepsy who was not receiving benefit from an AED would be taken off that treatment quickly. This is a particularly common practice in severe epilepsy, where patients cycle through a large number of AEDs over their lifetime. In line with responsible prescribing practices, clinicians, patients and caregivers would not want a patient to remain on a treatment that was no longer adding value.

This means that a situation in which a patient was on a treatment for a period of time and then remained on that treatment long term if it reduced in effectiveness is not a plausible scenario.

However, in response to this request from the ERG, efforts to model the consequences of a reduction in treatment effect have been made within the confines of the economic model structure.

We have provided a scenario where patients who would otherwise discontinue cannabidiol treatment continue for an extra 16 cycles and incur the extra cost of cannabidiol but with no treatment benefits. (Note that the scenario is inclusive of the change in the base case as per question B13.a).

This is applied assuming a lag between a loss of treatment effect and discontinuing cannabidiol by modelling that a proportion of patients (range: 10-100%) who discontinue incur all the cannabidiol treatment cost but receive no treatment benefit for an extra 16 weeks.

This delay of 16 weeks is likely to be a conservative assumption. The two UK clinical experts confirmed that, if a patient with TSC-associated epilepsy under their care were to stop having a satisfactory response to their anti-seizure medication (ASM), the ASM would be stopped almost immediately and certainly within ‘a couple of months’.

The results in Table 21 demonstrate that there is a minor impact on the ICER when assuming that the maximum number (100%) of discontinuing cannabidiol patients incur the extra cost of cannabidiol but receive no treatment benefit.

Table 21: Scenario analysis results (PAS price)

B9. Priority: The effect on TAND was included in the model (as a reduction in resource use and an increase in health-related quality of life (HRQoL)) when the seizure frequency was reduced by at least 50%.

• a. Please justify the use of seizure frequency based on a relative scale (50% reduction). Please consider in your answer company submission section B.3.1.2 which discusses how a method based on a percentage reduction in seizure frequency may not accurately capture costs and quality of life outcomes.

As outlined in Document B of the company’s submission, the cognitive and behavioural difficulties known as TSC-associated neuropsychiatric disorders (TAND) that prevent children from achieving independence in adult life can have a profound impact on the quality of life experienced not only by the patients but also by their families and carers.

We acknowledge that the approach taken to modelling the devastating impact of TAND is not ideal. However, we have made our best efforts with the limited available evidence. In this case, the evidence available was seizure frequency based on a relative scale (% reduction).

This pragmatic approach was an attempt to broadly quantify the impact on costs and quality of life outcomes of this important aspect of TSC-associated epilepsy.

b. Please clarify whether the 50% reduction threshold was defined as a reduction in seizure frequency, in seizure free days, or a combination of both, and how this definition aligns with the definition of ‘seizure frequency’ the experts in the Delphi panel used.

The 50% reduction threshold was defined as a reduction in seizure frequency.

The Delphi panel returned a near consensus that the reduction in seizure frequency required to reduce the progression of TAND aspects would be 47.5%. To be conservative, we used a 50% reduction in the model.

This was implemented in the model by applying the benefit of reducing the progression of TAND to the proportion of patients who experienced a reduction of seizure frequency of at least 50% following treatment initiation with cannabidiol plus usual-care or placebo plus usual-care.

c. Please provide a scenario where the threshold for reduction of seizure frequency to include an effect on TAND is 75%.

The results for the scenario analysis based on the threshold for reduction of seizure frequency to include an effect on TAND of 75%, are provided in Table 22. Note that the scenario is inclusive of the change in the base case as per question B13.a.

The scenario shows minor sensitivity to the change, with cannabidiol demonstrating a marginally lower incremental cost, higher incremental QALY and a slightly lower ICER.

Table 22: Scenario analysis results (PAS price)

d. Please provide a scenario analysis excluding TAND.

The results for the scenario analysis based on the exclusion of any TAND mitigation, are provided in Table 23. Note that the scenario is inclusive of the change in the base case as per question B13.a.

The scenario shows minor sensitivity to the change, with cannabidiol demonstrating a marginally higher incremental cost, lower incremental QALY and a slightly higher ICER.

Table 23: Scenario analysis results (PAS price)

e. Seizure frequency has an impact on resource use and HRQoL, but also

on TAND which in turn has again impact on resource use and HRQoL. Although the TAND aspects have a distinct impact on resource use and HRQoL, there may also be overlap with aspects already captured in the vignettes for instance. Please comment on the possible impact of this duplicating effect (i.e., double counting) and explain how it could be corrected.

We acknowledge that there may be some overlap. However, as explained in question B9.a above, we took a pragmatic approach in an attempt to model the devastating impact of TAND on patients and their carers/families.

Whilst not ideal, we made our best efforts with the limited data available. As can be seen in question B9.d above, there is minor sensitivity in the model to TAND.

• f. The company submission states on p110 that the proportion of patients who experienced a reduction of seizure frequency of at least 50% following treatment initiation was informed by an analysis of GWPCARE6 data, and that the reduction in progression of TAND was applied for that proportion. From the model it would seem however that the proportion is informed by the modelled percentages per seizure frequency category. Please provide from the GWPCARE6 data the proportions of patients experiencing at least 50% reduction in seizure frequency at 6 months, by age category (and separately for the 1-yearolds).

To clarify, the proportion of patients who achieved at least a 50% response (reduction) in seizure frequency compared to baseline was calculated based on the same method used to calculate the stopping rule rates, as detailed in Appendix O. The same method was used to maintain consistency across model inputs.

The proportion of patients who achieved at least a 50% response (reduction) in seizure frequency compared to baseline for patients aged 2-6 years is provided in Table 24 . The data is provided for patients aged 2-6 years, which aligns with the population for whom TAND mitigation benefits are applied in the base case analysis, and for all patients. Other age groups (including patients aged 1 year old) are not provided as these are not used in the model analysis.

A total of 11 patients (cannabidiol arm) and 14 patients (usual-care arm) were available to inform the TAND response rates for patients aged 2-6 years. Given this limited sample size, using only data for the 2-6 year old age category produces unrealistic outcomes, with responses for all but one health state based on 4 patients or less. Therefore, in both the model base case, where benefits are only applied for patients aged 2-6 years, and in the scenario which applies benefits to all patients, the data on TAND response rates were taken from the full population.

Table 24: Proportion of patients who are TAND responders per seizure frequency category (≥ 50% rates) – aged 2-6 and all patients

B10. The following questions relate to validation regression estimates versus OLE data:

• a. Please confirm that the data presented for the OLE in Figure 16 of the company submission do not contain patients who were on placebo during the blinded phase.

We confirm that the data presented for the OLE in Figure 16 in Document B of the company submission is based on patients who received cannabidiol 25 mg/kg in the blinded phase.

• b. In the OLE, there was no data on seizure free days, only on seizure frequency. Please clarify what is meant in Figure 16 on the y-axis with:

‘seizure frequency on seizure days’. How is this defined and how does this compare to what was estimated in the regression model?

During the blinded phase, data for seizure frequency were collected daily for each patient. In the OLE, patients/caregivers were instructed to complete a weekly seizure reporting diary meaning that seizure frequency data were collected approximately every 7 days. Due to the different nature of seizure collection in the OLE, the following assumptions were necessary when summarising the data per week:

• The number of days between collections was the number of days seizures occurred over. However, a cut-off of 7 days was used where there were more than 7 days between records as in some cases this value was high

• Seizures collected over periods of 12 weeks occurred evenly over each week – seizures were grouped into 12-week blocks (Week 1 to 12, Week 13 to 24, etc.) which were then used to derive the average number of seizures per 7 days. This was done to adjust for multiple collections in some weeks

• Seizures occurred on every day of the collection period – this assumption was required as it was not possible to extract the number of seizure-free days from the data

Therefore, the observed data in Figure 16 of the company submission represents the average 7-day seizure frequency based on all assumed days of the collection period.

• c. Please comment on the fact that in Figure 16, on average there seems to be underprediction of the seizures. Please provide in a table the numerical data behind the figure with percentual deviation from observed seizure frequency per cycle.

Figure 10 presents the observed and estimated (using both the fixed and random effect) seizure frequencies with 95% confidence intervals for the observed data. Note that all point estimates are identical to those shown in Figure 16 of the company submission, except at Week 17, where data were re-estimated to include only patients with observed values. The predicted values (shown in red) fall well within the 95% confidence interval of the observed data (shown in blue). In addition, the area under the curve (AUC) for the observed data is less than 5% different from the

predicted curve [755 (95% confidence interval: 430, 1080), compared with 715]. Again, the AUC for the predicted data falls well within the 95% confidence interval for the observed curve. Although we note that the AUC for the predicted curve is marginally lower than the AUC for the observed curve, there is variability in the observed OLE data, which is likely due to the nature of the data collection and derivation (described in response to B10.b).

Although it is possible that the number of seizure days per week of the OLE are underestimated in the observed data due to the assumption that seizures occur on all collection days (bullet 3 above), this may be counteracted with the number of collection days assumption (bullet 1 above). Although we acknowledge there are limitations with the data collection in the OLE, the estimates suggest that the model provides reasonable predictions for seizure frequency throughout the OLE period.

Figure 10: Observed and estimated seizure frequency

==> picture [452 x 211] intentionally omitted <==

Notes : Dashed line indicates the start of the open-label extension. Data are presented for the cannabidiol 25 mg/kg/day arm from the core trial. 95% confidence intervals are presented for the observed data.

Table 25: Observed and estimated seizure frequency data

B11 . At baseline (cycle 0) the proportion of seizure-free patients is 7% in both arms.

• a. Please explain why the proportion of seizure-free patients in both arms drops to 0% in the first week (cycle 1).

• b. Please justify, providing clinical expert opinion, the clinical plausibility of this sudden change.

• c. Please provide a scenario where proportion of seizure-free patients is 0% at baseline.

Please note that, while relooking at these data in response to this question, we noticed that erroneous baseline (cycle 0) data in the model was included. Data was left over from the exploratory phase of the model development and has therefore been removed. As the data was not used, there is no impact on the modelled ICERs. We apologise for the error.

B12 . Table 16 of the company submission presents the discontinuation rate per cycle per seizure frequency category. Please explain:

• a. How were the discontinuation rates in Table 16 calculated, were the 3-month rates divided by 13 to get from a 3-month to a 1-week rate? Also please clarify

whether the original rates from technology appraisal (TA) 615[23] were actually rates or probabilities.

The discontinuation rates in Document B, Table 16, for long term discontinuation, were 3-month rates sourced from NICE TA615. To calculate a weekly rate suitable for use in the economic model, these were divided by 91 days (equivalent to 13 weeks) and multiplied by 7 days (1 week) to get from a 3-month to a 1-week rate .

To clarify, the discontinuation rates used in technology appraisal TA615 were presented as rates per cycle (3-month cycle).

b. How the rates compare to other cannabidiol trials or TAs, such as TA614[24]

Table 26 presents the rates used in TA614, TA615 and ID1416. For comparative purposes, the rates used in TA614 and TA615 have been converted from a 3-month to a 1-week rate, using the calculation outlined in response to B12.a. The 3-month rates from TA614 and TA615 are presented in Appendix 3. The rates calculated from the GWPCARE6 trial and open-label extension data for ID1416 are broadly comparable to the rates used in TA614 and TA615.[8, 9]

Table 26: Discontinuation rates from TA614, TA615 and ID1416 (per week)

B13 . Table 17 of the company submission shows the proportion of patients

discontinuing treatment based on the stopping rule.

a. The proportion of patients with > 2 and ≤ 7 seizure per week stopping at 12, 18, and 24 months was 0% because of small patient numbers in the OLE study. Please correct this inconsistency by using the average of the proportion with ≤ 2 seizures per week and ≥7 seizures per week in the base case.

As discussed on the ERG clarification call (28[th] April 2022), the base case analysis has been updated, to include the scenario where an average of the ≤ 2 seizures per week and ≥7 seizures per week is applied to the >2 –≤7 seizures per week health state.

The updated base case results of the comparison between cannabidiol plus usualcare and placebo plus usual-care are presented in Table 27. The results demonstrate that cannabidiol is a cost-effective use of NHS resources to treat seizures in patients aged ≥ 2 years with TSC-associated epilepsy at a WTP threshold of £20,000 per QALY gained.

The base case results with consideration of the disease severity modifier, as detailed in Document B, Section B.3.3.9, are presented in Table 28. These results also demonstrate that cannabidiol is a cost-effective use of NHS resources when considering the severity of TSC-associated epilepsy.

Updated sensitivity analysis is provided in Appendix 2.

Table 27: Updated base case results

Key: ICER, incremental cost-effectiveness ratio; QALYs, quality-adjusted life years. Note: *Discounting is applied to QALYs and cost.

Table 28: Updated base case results (including disease severity multiplier)

• b. Please justify why the stopping rule was not applied after 24 months.

Patient level data on treatment response (≥ 30% reduction in seizure frequency) from the GWPCARE6 blinded trial period and the open-label extension are used to calculate stopping rates.

There are currently no data beyond this to enable us to apply an evidence-based stopping rule. For this reason, a long-term discontinuation rate is also applied post the OLE period to reflect patients discontinuing therapy due to non-response and other factors over the long-term.

To note: applying the stopping rule up to 24 months is in line with the stopping rule frequency requested by NICE/NHS England in the technology appraisals for cannabidiol in LGS and DS.

• c. Provide scenarios with the stopping rule also applied after 24 months.

There are currently no data available to do this, so any scenario provided would be arbitrary.

However, it would be expected that if an arbitrary extrapolated stopping rule were applied after 24 months, this would result in a lower ICER.

B14 . Please provide the formula for calculating the TSC-related mortality and sudden unexpected death in epilepsy (SUDEP) per 11.08 and 8 years, respectively, to annual mortality risks (page 105 of company submission).

The TSC-related mortality rate calculation is detailed in Appendix P. Briefly the calculation is as follows:

��� ������

==> picture [381 x 27] intentionally omitted <==

The SUDEP mortality rate calculation is detailed in Appendix P. Briefly the calculation is as follows:

� ������

==> picture [349 x 27] intentionally omitted <==

Adverse events (AEs)

B15 . Section 3.3.4 of the company submission describes the inclusion of adverse events in the cost-effectiveness model. Please explain:

• a. The difference in definition of serious and severe events.

Serious adverse event: In the GWPCARE6 trial, an AE was considered serious if it: (1) was fatal; (2) was life-threatening; (3) required inpatient hospitalization or prolonged existing hospitalization; (4) was persistently or significantly disabling or incapacitating; (5) was a congenital anomaly/birth defect; or (6) was a medically significant event that, based upon appropriate medical judgment, may have jeopardized the patient and may have required medical or surgical intervention to prevent one of the outcomes listed above.

Severe adverse event: For all AEs and serious AEs, the clinical trial investigators were required to assign severity and document this on the Case Report Form. The method is described in the trial protocol as follows:

“When describing the severity of an AE, the terms mild, moderate, or severe should be used. Clinical judgment should be used when determining which severity applies to any AE.

If the severity of an AE fluctuates day-to-day, e.g., a headache or constipation, the change in severity should not be recorded each time; instead, only the worst observed severity should be recorded with AE start and stop dates relating to the overall event duration, regardless of severity.

A severe AE is not the same as a SAE. For example, a patient may have severe vomiting but the event does not result in any of the SAE criteria above. Therefore, it should not be classed as serious.”

• b. How exactly the event rates in Table 15 of the company submission can be derived from, or are related to, Table 3 of Appendix F.

The event rates in Table 15 of the company submission are Serious treatmentemergent adverse events (TEAEs), that were classified as severe by the investigator and were considered to be treatment-related .

These are the adverse events that would drive the most costs associated with the drug and are therefore the most appropriate to use in the model.

Table 3 of Appendix F is simply a list of treatment-emergent adverse events (serious and non-serious).

• c. Why AE rates were based on the first 16 weeks, and not the adverse events from the extended trial period as the serious AE incidence in the OLE study was still 15%. Please validate 16-week AE data with the OLE study data.

The company considered this to be a conservative approach. The inclusion of serious AEs from the randomized clinical trial period is likely to be an overestimate of costs.

For example, the rate of serious adverse events in the cannabidiol 25 mg/kg/day arm in the randomized trial period was 21.3%. As noted by the ERG above, this decreased to 15% in the OLE.

Based on real-world evidence, it is likely that this rate might be even lower in realworld clinical practice. For example, a real-world study of cannabidiol use in patients demonstrated that slower titration of the cannabidiol dose delivered better tolerability with comparable efficacy to previous trials.[25]

Health related quality of life

B16. Priority: A vignette study was used to estimate the HRQoL for the different health states.

• a. According to the health-related quality of life task and finish group report (2020),[26] the DSU recommends that EQ-5D data collected for each vignette should be valued using a relevant value set. Please explain which EQ-5D value set was used.

The company did not use the EQ-5D value set, as the accepted preference elicitation TTO method was used to value the vignettes directly. This method was used as it is a standardised interview method for valuing health states, in line with methods used to value the EQ-5D-3L. According to the DSU report, vignettes can be

valued using a range of different techniques: clinical experts (or general population or patients) proxy-completing EQ-5D for the vignette health state; valuation with members of the general public or patients using preference elicitation techniques such as time trade-off (TTO); or eliciting values from clinical experts (for example using a Delphi method).

Standard measures such as the EQ-5D questionnaire fail to capture many aspects of the severity and complexity of severe and refractory epilepsies such as TSCassociated epilepsy. A feasibility assessment was conducted to determine the strength of evidence for the suitability of the EQ-5D and other generic or condition specific preference-based measures in refractory epilepsies (see Appendix R). Two studies evaluating the use of the EQ-5D found only weak correlations between the EQ-5D and condition specific measures. Mukuria et al concluded that, given the lack of correlation and joint responsiveness between the measures, using the EQ-5D for a cost-effectiveness analysis, including for mapping, is not recommended.[27] No other preference-based measures with confirmed psychometric validity in the population of interest were identified. On this basis, given that we were unable to find a suitable measure that could be used directly in patients, the only alternative option was to generate vignettes that accurately reflected the TSC-associated epilepsy population and value these in the general population to generate utilities.

b. Please clarify the exact methods used to calculate time trade-off (TTO) weights for the combined seizure health states (Table 21 of the company submission).

As detailed in Lo et al, 2021, in order to keep the length of the TTO valuation interviews manageable, only one vignette with a combination of both seizure types was developed.[28] For the other possible health states with a combination of seizure types, utility weights were estimated through linear interpolation of the differences between the TTO-values states. The methods used to calculate the TTO weights are detailed in Document B, Section B.3.4.4.1, and are consistent with the method outlined in Lo et al, 2021 (see Table 29 below).

The estimates were made assuming that disutility increases in a linear fashion with increasing seizure frequency. Values for the states with the lowest and highest number of seizures (valued using the TTO vignette method) were used as anchoring

points to allow values of other states with seizure frequencies falling within this range to be estimated.

Table 29: Estimation of utility values using TTO-valued heath states

c. Please provide a comparison of the utility values by seizure frequency and seizure type in Tables 20-24 of the company submission with those from TA614[24] and TA615.[23]

The VAS was used to elicit QoL data in TA614 and TA615. Since the VAS scores were not considered to be the most appropriate estimates to elicit utility values, we followed feedback from NICE and adopted the more rigorous preference-based TTO method for this submission.

For comparison, utility values from TA614 and TA615 are provided below.

TA614

- Summary of mean VAS scores for cost effectiveness analysis

Summary of mean caregiver VAS score utility decrements

TA615

- Summary of mean VAS scores for cost effectiveness analysis

Summary of mean caregiver VAS score utility decrements

d. Justify the relatively large utility decrements associated with the seizure

frequencies health states (for patients and caregivers), using expert opinion, empirical evidence and reference values from other diseases as benchmarks.

In recent discussions (conducted as part of providing answers to the ERG clarification questions), two UK clinical experts (Professor Finbar O’Callaghan and Dr

Sam Amin) confirmed that TSC-associated epilepsy and its associated consequences would have a ‘high negative impact’ on the quality of life of both patients and caregivers.

The relatively low utility values for the seizure health states are not unexpected given the number of seizures, TAND aspects and comorbidities, and the consequent impacts on the patient and his/her caregivers, as described by the vignettes in the TTO study. The detailed vignette descriptions outline the devastating situation in which patients or caregivers may find themselves. The vignettes were drafted with input from caregivers who highlighted the considerable impact that caring for a patient with TSC-associated epilepsy has on their wellbeing and quality of life.

We would like to reiterate the severe impact of TSC-associated epilepsy on the quality of life not only of patients but also on that of their families and caregivers. The burden of care and the effects of TSC-associated seizures on the patient can necessitate changes in virtually all aspects of the lives of caregivers and family members. The various aspects of TAND are an additional burden for caregivers already trying to cope with seizures.

Caring for a child or loved one with TSC-associated epilepsy is a 24-hour responsibility. Patients often have complex needs, severe impairment of daily functioning and a history of epilepsy-related events, such as injuries and hospitalisations related to seizures. Caring for them dictates work schedules, family time and leisure activities. Caregivers suffer significant anxiety and depression, social isolation, poor sleep quality, emotional distress, and a considerable impact on their ability to work, often resulting in financial difficulties.

Whilst ‘general epilepsy’ undoubtably has an impact on the quality of life of patients and caregivers, TSC-associated epilepsy (in common with other forms of epilepsy that are at the extremely severe, refractory and life-threatening end of the disease spectrum, such as DS and LGS) takes this impact to another level.

Thus, in the context of the severe impact on both patients and caregivers outlined above, the relatively large utility decrements associated with the seizure frequencies health states (for patients and caregivers) are not unexpected.

e. Please provide external validation (with data if available or else with expert opinion) for the health state utilities from the vignette study.

Scenario analyses have been run applying alternative utilities from the studies by Tritton 2019[29] and Vergeer 2019[30] previously described in Document B of the company’s submission.

Using these alternative utility values has a minor impact on the ICER (see Table 35 below).

• f. Please explain the calculations used to transform the vignette health states into the health state utilities for the model.

An explanation of the calculation, including a worked example to calculate the utility for the health state ‘one seizure’ per day is provided in Appendix R, Section R4. The utility calculations for two seizures per day, three or four seizures per day and five or more seizures per day are calculated using the same method as outlined in the worked example. Utility values for caregivers are calculated using the same method.

B17. Priority: The utilities in the model were not corrected for age. Particularly in the seizure-free state this results in unrealistically high QALYs at older age. Please provide an updated cost-effectiveness model including an age-related decrement for the health state utilities or apply a cap to ensure that utilities do not go above those of the general population for the corresponding age.

As requested, a general population utility value cap on the patient utility values, ensuring that utilities do not exceed the age- and gender-matched population norms, has been added into the model. The seizure-fee health state is the only health state that exceeds population norms over a lifetime horizon. A cap on this utility has been added to the model and is applied when the seizure free utility of 0.725 exceeds population norms which first occurs at age 86 (0.723). The gender-matched population norms reported by Alava et al, 2017 were used to inform the cap, in line with NICE DSU guidance.[31, 32]

The results for the scenario analysis based on the addition of a patient utility cap, are provided in Table 30. The scenario shows very little sensitivity to the addition of a patient utility cap, with a minimal change to the ICER.

Table 30: Scenario analysis results (PAS price)

B18 . Caregiver disutilities were assumed to apply additively to two caregivers. However, it is unlikely that both caregivers provide equal care to the patient. For example, Vyas et al.[33] showed that all household members together spent a total of 11 seizure-specific hours per week, of which 7 hours were spent by the primary caregiver. This indicates that disutilities might also differ between caregivers. Also, in TA614[24] and TA615,[23] the committee concluded linearly extrapolating to 2 caregivers was not preferable.

• a. Please justify that caregiver utility decrements included in the cost-

  • effectiveness model are on an additive scale, assuming both caregivers would have the same disutility.

The vignette study specifically assessed caregiver quality of life in the context of the respondent being one of two primary caregivers. Therefore, it is reasonable to assume that the caregiver utility estimates are representative of an individual’s quality of life decrement where another carer was present.

In previous NICE submissions with a carer disutility included for more than one carer (e.g. ataluren for DMD (HST3); disutility applied for 2 and 3 carers), there is a precedent where each carer was allocated the same disutility.

In a broader context, including utilities for two caregivers also reflects the impact of TSC-associated epilepsy on the quality of life of the wider family. Whilst we acknowledge that not every patient will have two primary caregivers (although many will need more than two), the company is aware that there are often many other people contributing to the care of the patient (for example, siblings, grandparents, aunts, uncles, family friends). The cumulative impact on the quality of life of all these

other caregivers more than compensates for any additive effect in the two main caregivers.

In TA614 and T615, NICE particularly recognised the negative impact of severe epilepsies on a patient’s siblings.

• b. Please provide a scenario where disutilities for only 1 caregiver are included.

The improvement in quality of life for both patients and caregivers is an extremely important part of the value of cannabidiol in TSC-associated epilepsy. This was recognised by NICE in the appraisals of cannabidiol in other severe epilepsies (DS and LGS).

For the following reasons, we do not consider that a scenario where disutilities for only 1 caregiver are included is reasonable:

• In a severe and life-threatening disease such as TSC-associated epilepsy, patients are at a significant ongoing risk of injury and death from their seizures, have multiple co-morbidities and often require lifelong round-the-clock care from multiple caregivers

• The DISCUSS study reported by Lagae et al[41] (discussed in Document B of the company’s submission) indicated that the minimum number of caregivers required for a patient with refractory severe epilepsy is at least two

• Two UK clinical experts consulted as part of this response to the ERG clarification questions indicated that having at least two caregivers was usual for patients with TSC-associated epilepsy

• Including disutilities for 2 caregivers reflects the quality of life impact of round-the clock care of a patient with TSC-associated epilepsy, but does not account for the impacts on the wider family, such as siblings. Each family carer has the burden, worry and psychological distress of caring for a patient at risk of injury and even death from their seizures. It would be expected that incorporating the full effect of TSC-associated epilepsy on the lives of all family members would result in a further reduction of the ICER.

For these reasons, the company considers that only including the impact on 2 caregivers in the base case is conservative.

B19 . The following questions relate to caregiver utility values:

• a. Please comment on the fact that the seizure-free health state value estimated in the vignette study is 0.905 for caregivers, which is used as a base for calculating the disutilities in the non-seizure free health states. The study by Vyas et al.[33] shows that overall time spent caring for patients with TSC is 128 hours per week, of which 11 hours per week seizure specific. This implies that when a patient is seizure-free, the caregiver may still have many worries, so a utility score of 0.905 may overestimate carer HRQoL for this health state.

The vignette study estimated a caregiver utility of 0.905 for the seizure-free health state value. The vignette health state description is shown below (see Figure 11). Participants (as caregivers) were asked to consider some worry for the care of their child. Additionally, participants were asked to adopt the perspective of one of two primary caregivers/parents of a 13-year-old child.

Assuming the caregiver is a parent of the child (mother), based on ONS data, the average age of mothers at childbirth in 2019 was 30.7 years.[34] For a 13-year old child, this implies an average parent age of 43 years. The utility for an adult (female) aged 43 is 0.897, which is very similar to the utility estimated for the seizure-free health state value (potentially over-estimated by 0.008). Therefore, the utility value for the seizure-free health state elicited by the vignettes is reflective of a carer population in real-world practice.

It should also be noted that the vignette study included all background information relating to ‘worries’ equally in the vignettes for the seizure-free state and all the seizure states. As a result, even if the seizure-free utility value was a slight overestimate, it would be expected that the utility values for the seizure states would also be slightly over-estimated. Therefore, the calculated decrements would be unchanged and the same as those used in the model.

Figure 11: Caregiver vignette – seizure-free health state

==> picture [390 x 117] intentionally omitted <==

==> picture [390 x 118] intentionally omitted <==

==> picture [390 x 117] intentionally omitted <==

b. Please also comment on the possibility of applying an age-related decrement or general population cap on the utility value for the caregivers and the impact on the ICER.

We would not expect that applying an age-related decrement or general population cap on the utility value for the caregivers would impact the ICER.

The impact of caring for a patient with TSC-associated epilepsy would reasonably be expected to be higher as the caregiver gets older. Therefore, as the age-related utility goes down, it would do so for all health states such that the absolute difference in the utility value between the seizure-free state and the seizure states would stay the same.

Cost and healthcare resource use

B20 . Please provide a scenario analysis where resource use was informed by the health care resource use reported by Shepherd et al.[35] instead of the Delphi panel values.

Shepherd et al, 2017 was reviewed as a potential source of data to inform the economic model during the development phase. However, the study was not considered suitable to inform healthcare resource use for several reasons. These include:

• The Shepherd study did not collect data based on any TSC-associated epilepsy outcomes such as seizure frequency. As healthcare resource use is anticipated to be higher with increasing seizure frequency, using the data collected by Shepherd in the model will overestimate the healthcare resource use burden associated with the lowest seizure frequency health states and underestimate resource use associated with the higher seizure frequency health states and fail to capture the benefit of cannabidiol in reducing seizure frequency and therefore healthcare resource use burden, which was included in previous submissions in rare epilepsies.[4, 8, 9]

• The study population in Shepherd et al. does not fully align with the modelled population. Only 60% of the study population were defined as having refractory epilepsy, based on a definition of being prescribed ≥ 2 concomitant antiepileptic drugs during their entire recorded history (data collected based on CPRD and HES records between 1997-2012); with the remainder receiving no AEDs (12%) or only one AED (28.2%). As patients with refractory epilepsies are associated with higher healthcare burden[36, 37] , the data presented by Shepherd is likely to underestimate the healthcare resource use burden associated with TSCassociated epilepsy.

• Using the collected data recorded between 1997 to 2012, Shepherd et al. calculated mean healthcare resource use based on the most recent 3-year period of continuous data for each person. As part of the validation of the Delphi panel, the elicited resource use estimates were compared to the Shepherd data. The clinician who assisted in this, Dr Kingswood, noted that treatment practice has

changed since the Shepherd study was conducted, with new guidance issued in 2018, changing the treatment paradigm.[38] Treatment practice has moved more care from general practice to specialist centres, leading to a possible overreporting of GP visits and under-reporting of outpatient visits in the Shepherd et al. study.

Given these limitations, the Shepherd data is not considered suitable for use as a scenario for healthcare resource use.

Company’s base-case results and sensitivity analyses

B21 . The following questions relate to the probabilistic sensitivity analysis (PSA)

• a. Please explain what could be causing the outliers to the northeast in the incremental cost-effectiveness plane (Figure 20 in the company submission)

The probabilistic analysis (PSA) results are presented in Document B, Section B.3.9.1. The PSA results are broadly comparable to the deterministic analysis results; however, the PSA contains several outlier results, as shown in in Document B, Section B.3.9.1, Figure 20. The following tests were undertaken during model development to test the PSA analysis.

Firstly, all probabilistic distributions were checked to ensure the correct distributions were being used. Next, the most important sets of parameters were tested, by individually switching these parameters to a deterministic setting for each PSA run including utilities, costs, stopping and discontinuation rates and the output of the regression analysis, the coefficients which are used to predict seizure frequency and seizure-free days. The skewed costs effectiveness plane and outlier results are driven by the regression analysis, specifically the non-linearity of the regression coefficients combined with the non-linear probabilistic analysis. In this circumstance, it is reasonable to expect a non-uniform cost-effective plane. Additionally, as acknowledged in the updated NICE methods guidance, cost-effectiveness estimates are inherently uncertain for small populations such as those associated with orphan diseases. Therefore, these analyses should be considered within that context, with a higher degree of uncertainty considered acceptable.[39]

b. Please run a PSA with 5,000 simulations

The model has been updated to allow a maximum of 5,000 PSA iterations. As requested, a PSA was conducted in which all inputs were varied simultaneously over 5,000 iterations, based on their distributional information. Note that the updated PSA was run inclusive of the change to the base case requested for B13.a.

The results are summarized in Table 31 and are also presented on a costeffectiveness plane in Figure 12 and cost-effectiveness acceptability curve in

Figure 13 . The PSA results are broadly comparable with the deterministic analysis, and the PSA results presented in Document B, Section 3.9.1.

Table 31: Mean probabilistic sensitivity analysis results

Figure 12: Probabilistic sensitivity analysis cost-effectiveness plane (5,000) simulations

==> picture [441 x 241] intentionally omitted <==

Figure 13: Probabilistic sensitivity analysis cost-effectiveness acceptability curve (5,000 iterations)

==> picture [441 x 241] intentionally omitted <==

The cost-effectiveness acceptability curve suggests that there is a ***** and ***** likelihood that cannabidiol is cost-effective at WTP thresholds of £20,000 and £30,000 per QALY gained, respectively.

• c. Please add a worksheet to the model recording the disaggregated outcomes

per simulation of the PSA – to enable diagnosing the cause of outliers

The model has been updated to include an additional sheet providing disaggregated outcomes per simulation of the PSA (see the “PSA_data” sheet).

Validation and transparency

B22 . The results of the validity assessments or detailed validation exercises (i.e., specific black-box tests) are not described in company submission section B.3.11.

• a. Please provide a detailed description of the validity assessment performed as well as the results.

• b. Please provide complete the TECH-VER checklist (Büyükkaramikli et al. 2019, https://pubmed.ncbi.nlm.nih.gov/31705406/) and provide the results.

A technical review of the cost effectiveness model was independently undertaken by Swansea Centre for Health Economics (SCHE).

The SCHE technical review followed internationally recognised ISPOR principles of model validation. The overall integrity of the data was checked. SCHE also

conducted internal validation to ensure that the model engine works in the manner which it was designed, and the coding of the model was quality checked to detect any errors.

The validity of the model base case analysis and sensitivity analyses was explored This included extreme values testing to ensure that varying parameter input values had a viable and expected impact on outputs.

The process followed the methods outlined in the TECH-VER checklist.

The SCHE report is provided in the reference pack.[40]

B23 . Please provide cross validations, i.e., comparisons with other relevant NICE TAs focussed on similar, potentially relevant, diseases (e.g., related NICE recommendations and NICE Pathways listed in the final scope) and elaborate on the identified differences regarding:

• a. Model structure and assumptions

• b. Input parameters related to:

• i. Clinical effectiveness

• ii. Health state utility values

• iii. Resource use and costs

• iv. Estimated (disaggregated) outcomes per comparator/ intervention v. Life years

• vi. QALYs

• vii. Costs

Please refer to Appendix G of the company’s original submission.

Section C: Textual clarification and additional points

C1. When describing the population in Table 3 of the company submission, the following sentence does not seem to make sense: “All medications or interventions for epilepsy (including a ketogenic diet and vagus nerve stimulation, which were not counted as ASMs) stable for 4 weeks before screening.” Please clarify.

As for the rest of the ‘Population’ row in Table 3, Document B, the wording is abbreviated as appropriate for a table. As a complete sentence, it would read “All medications or interventions for epilepsy (including a ketogenic diet and vagus nerve stimulation, which were not counted as AEDs) had to be stable for 4 weeks before screening”.

Appendix 1

==> picture [452 x 644] intentionally omitted <==

----- Start of picture text -----
Reference In company
submission?
[1] Chiron C, Dulac O, Beaumont D, Palacios L, Pajot N, Mumford J. No
Therapeutic trial of vigabatrin in refractory infantile spasms. Journal of
Child Neurology 1991;6(2_suppl):2S52-2S59.
[2] Chiron C, Dulac O, Beaumont D, Palacios L, Pajot N, Mumford J. Duplicate
Therapeutic trial of vigabatrin in refractory infantile spasms. Journal of
child neurology 1991;Suppl 2:S52-59.
[3] Chiron C, Dulac O, Beaumont D, Palacios L, Pajot N, Mumford J. Duplicate
Therapeutic trial of vigabatrin in refractory infantile spasms. Journal of
Child Neurology 1991;6(SUPPL. 2):2S52-2S59.
[4] Chiron C, Dumas C, Jambaque I, Mumford J, Dulac O. Randomized No
trial comparing vigabatrin and hydrocortisone in infantile spasms due to
tuberous sclerosis. Epilepsy Research 1997;26(2):389-395.
[5] Cock H, Wu JY, Devinsky O, Joshi C, Miller I, Roberts C, et al. Time to No
onset of Cannabidiol (CBD) treatment effect and resolution of adverse
events in the tuberous sclerosis complex Phase 3 randomised controlled
trial (GWPCARE6). Developmental Medicine and Child Neurology
2021;Conference: Annual Meeting of the British Paediatric Neurology
Association. Virtual. 63(SUPPL 1):77.
[6] Collins JJ, Tudor C, Leonard JM, Chuck G, Franz DN. Levetiracetam No
as adjunctive antiepileptic therapy for patients with tuberous sclerosis
complex: A retrospective open-label trial. Journal of Child Neurology
2006;21(1):53-57.
[7] Curatolo P, Franz DN, Lawson JA, Yapici Z, Ikeda H, Polster T, et al. No
Adjunctive everolimus for children and adolescents with treatment-
refractory seizures associated with tuberous sclerosis complex: post-hoc
analysis of the phase 3 EXIST-3 trial. The Lancet Child and Adolescent
Health 2018;2(7):495-504.
[8] Cusmai R, Moavero R, Bombardieri R, Vigevano F, Curatolo P. Long- No
term neurological outcome in children with early-onset epilepsy
associated with tuberous sclerosis. Epilepsy and Behavior
2011;22(4):735-739.
[9] De Vries PJ, Franz DN, Lawson JA, Yapici Z, Polster T, Nabbout R, et No
al. Adjunctive everolimus therapy for the treatment of refractory seizures
in people with tuberous sclerosis complex. Journal of Intellectual Disability
Research 2016;Conference: 19th International Research Symposium of
the Society for the Study of Behavioural Phenotypes, SSBP. Siena Italy.
60(9):839.
[10] Ding Y, Wang J, Zhou Y, Yu L, Zhang L, Zhou S, et al. Quality of life Excluded: no
in children with tuberous sclerosis complex: A pediatric cohort study. CNS results
Neuroscience and Therapeutics 2021;27(3):280-288. reported
----- End of picture text -----

ERG: Agrees [11] Franz D, Lawson J, Nabbout R, Curatolo P, De Vries P, Berkowitz N, No et al. Adjunctive everolimus therapy for the treatment of refractory seizures associated with tuberous sclerosis complex: Results from a randomized, placebo-controlled, phase 3 trial. Annals of Neurology 2016;Conference: 45th Annual Meeting of the Child Neurology Society. Vancouver, BC Canada. 80(Supplement 20):S317. [12] Franz D, Lawson J, Yapici Z, Ikeda H, Polster T, Nabbout R, et al. No Efficacy and safety of everolimus based on prior and concomitant antiepileptic drugs in patients with tuberous sclerosis complex (TSC)associated treatment-refractory seizures: A subanalysis of the phase 3 EXIST-3 study. Neurology. Conference: 70th Annual Meeting of the American Academy of Neurology, AAN 2018;90(15 Supplement 1). [13] Franz D, Lawson J, Yapici Z, Ikeda H, Polster T, Nabbout R, et al. No Sustained seizure reduction with adjunctive everolimus for treatmentrefractory seizures associated with tuberous sclerosis complex (TSC): Long-term results from the phase 3 EXIST-3 study. Annals of Neurology 2017;Conference: 46th Annual Meeting of the Child Neurology Society. Kansas City, MO United States. 82(Supplement 21):S281-S282. [14] Franz D, Lawson J, Yapici Z, Ikeda H, Polster T, Nabbout R, et al. No Sustained seizure reduction with adjunctive everolimus for treatmentrefractory seizures associated with tuberous sclerosis complex (TSC): Long-term results from the phase 3 EXIST-3 study. Neurology 2017;Conference: 69th American Academy of Neurology Annual Meeting, AAN 2017. Boston, MA United States. 89(8):e100. [15] Franz D, Lawson J, Yapici Z, Ikeda H, Polster T, Nabbout R, et al. No Efficacy and safety of everolimus based on prior and concomitant antiepileptic drugs in patients with tuberous sclerosis complex (TSC)associated treatment refractory seizures: Subanalysis of the phase 3 exist-3 study. Annals of Neurology 2018;Conference: 47th National Meeting of the Child Neurology Society, CNS 2018. Chicago, IL United States. 84(Supplement 22):S337. [16] Franz DN, Lawson JA, Yapici Z, Ikeda H, Polster T, Nabbout R, et al. No Sustained seizure reduction with adjunctive everolimus for treatmentrefractory seizures associated with tuberous sclerosis complex (TSC): Long-term results from the phase 3 exist-3 study. Annals of Neurology 2017;Conference: 142nd Annual Meeting of the American Neurological Association, ANA 2017. San Diego, CA United States. 82(Supplement 21):S65-S66. [17] Franz DN, Tillema JM, Care MM, Holland-Bouley K, Agricola K, Tudor No C, et al. Everolimus reduces seizure activity in patients with tuberous sclerosis complex (TSC). European Journal of Neurology 2011;Conference: 15th Congress of the EFNS. Budapest Hungary. Conference Publication:(var.pagings). 18(SUPPL. 2):568. [18] French J, Franz D, De Vries P, Nabbout R, Vaury A, Berkowitz N, et No al. Trial in progress: Exist-3, a placebo controlled study of the efficacy and safety of two trough ranges of everolimus as adjunctive therapy in patients with tuberous sclerosis complex (TSC)who have refractory partial-onset

seizures (POS). Epilepsia 2013;Conference: 30th International Epilepsy Congress. Montreal, QC Canada. Conference Publication:(var.pagings). 54(SUPPL. 3):65.

[19] French J, Lawson J, Yapici Z, Polster T, Nabbout R, Curatolo P, et al. Exposure-response analysis of adjunctive everolimus therapy in patients with refractory partial-onset seizures associated with tuberous sclerosis complex (TSC). European Journal of Neurology 2016;Conference: 2nd Congress of the European Academy of Neurology. Copenhagen Denmark. Conference Publication:(var.pagings). 23(SUPPL. 2):54.

[20] French J, Lawson JA, Yapici Z, Polster T, Nabbout R, Curatolo P, et al. Adjunctive everolimus therapy for the treatment of refractory seizures associated with tuberous sclerosis complex: Results from a randomized, placebo-controlled, phase 3 trial. Neurology 2016;Conference: 68th American Academy of Neurology Annual Meeting, AAN 2016. Vancouver, BC Canada. 87(2):e23-e24.

[21] French JA, Lawson JA, Yapici Z, Ikeda H, Polster T, Nabbout R, et al. Adjunctive everolimus therapy for treatment-resistant focal-onset seizures associated with tuberous sclerosis (EXIST-3): a phase 3, randomised, double-blind, placebo-controlled study. The Lancet 2016;388(10056):2153-2163.

[22] French JA, Lawson JA, Yapici Z, Polster T, Nabbout R, Curatolo P, et al. Adjunctive everolimus therapy for the treatment of refractory seizures associated with tuberous sclerosis complex: Results from a randomized, placebo-controlled, phase 3 trial. Annals of Neurology 2016;Conference: 141st Annual Meeting of the American Neurological Association, ANA 2016. Baltimore, MD United States. 80(Supplement 20):S32.

[23] Hess EJ, Moody KA, Geffrey AL, Pollack SF, Skirvin LA, Bruno PL, et al. Cannabidiol as a new treatment for drug-resistant epilepsy in tuberous sclerosis complex. Epilepsia 2016;57(10):1617-1624.

[24] Ko TS, Yum M, Lee E, Jeong M. Vigabatrin in epilepsy caused by tuberous sclerosis complex: Comparison of infantile spasms and partial epilepsy. Epilepsy Currents 2011;Conference: 64th Annual Meeting of the American Epilepsy Society, AES and 3rd Biennial North American Regional Epilepsy Congress. San Antonio, TX United States. Conference Publication:(var.pagings). 11(1 SUPPL. 1):no pagination.

[25] Krueger D, Care MM, Holland-Bouley K, Agricola K, Tudor C, Mangeshkar P, et al. Effect of everolimus on seizure activity in patients with tuberous sclerosis (TS). Epilepsy Currents 2011;Conference: 64th Annual Meeting of the American Epilepsy Society, AES and 3rd Biennial North American Regional Epilepsy Congress. San Antonio, TX United States. Conference Publication:(var.pagings). 11(1 SUPPL. 1):no pagination.

[26] Krueger DA, Wilfong AA, Holland-Bouley K, Anderson A, Agricola K, Tudor C, et al. Everolimus improves seizure control in tuberous sclerosis complex. Epilepsy Currents 2013;Conference: 2012 Annual Meeting of the American Epilepsy Society, AES 2012. San Diego, CA United States. Conference Publication:(var.pagings). 13(SUPPL. 1):108-109.

No No

No No

No No

No

No

[27] Krueger DA, Wilfong AA, Holland-Bouley K, Anderson AE, Agricola K, No Tudor C, et al. Everolimus treatment of refractory epilepsy in tuberous sclerosis complex. Annals of Neurology 2013;74(5):679-687. [28] Kwan P, Thiele EA, Bebin EM, Filloux F, Jansen FE, Loftus R, et al. No Long-term safety and efficacy of add-on cannabidiol for treatment of seizures associated with tuberous sclerosis complex in an open-label extension. Epilepsia 2021;Conference: 34th international Epilepsy Congress. Virtual. 62(SUPPL 3):145-146. [29] Nickels K. Cannabidiol in patients with intractable epilepsy due to Excluded: No TSC: A possible medication but not a miracle. Epilepsy Currents relevant data 2017;17(2):91-92. ERG: Agrees [30] O'Callaghan FJ, Sparagana SP, Kwan P, Saneto RP, Wheless JW, Include Hyland K, et al. Efficacy of add-on cannabidiol (CBD) treatment in patients with Tuberous Sclerosis Complex (TSC) and a history of Infantile Spasms: Post hoc analysis of Phase 3 Trial GWPCARE6. Developmental Medicine and Child Neurology 2021;Conference: Annual Meeting of the British Paediatric Neurology Association. Virtual. 63(SUPPL 1):71. [31] O'Callaghan FJ, Thiele EA, Bebin EM, Sparagana SP, Jansen FE, Published Schreiber A, et al. Effect of add-on Cannabidiol (CBD) on seizure after frequency and seizure-free intervals in patients with seizures associated company with tuberous sclerosis complex: Phase 3 trial GWPCARE6 post-hoc submission analysis. Developmental Medicine and Child Neurology 2022;Conference: searches Annual Meeting of the British Paediatric Neurology Association. Virtual. 64(SUPPL 1):26. [32] Oommen B, Brand E, Volpe A, Krause A, Crino P, Pollard J. No Vigabatrin for treatment of complex partial-onset seizures in patients with tuberous sclerosis: Prospective trial and retrospective case series. Epilepsy Currents 2015;Conference: 68th Annual Meeting of the American Epilepsy Society, AES 2014. Seattle, WA United States. Conference Publication:(var.pagings). 15(SUPPL. 1):531-532. [33] Overwater IE, Rietman A, Bindels-De Heus GCB, Moll HA, Elgersma No Y, Wit MCY. RATE: Randomised clinical trial of rapamycin in children with tuberous sclerosis complex and intractable epilepsy. European Journal of Paediatric Neurology 2013;Conference: 10th European Paediatric Neurology Society Congress, EPNS 2013. Brussels Belgium. Conference Publication:(var.pagings). 17(SUPPL. 1):S14-S15. [34] Parain D, Penniello MJ, Berquen P, Delangre T, Billard C, Murphy JV. No Vagal nerve stimulation in tuberous sclerosis complex patients. Pediatric Neurology 2001;25(3):213-216. [35] Roberts C, Kuperman R, Koenig M, Wilfong A, Wu J, Kohrman M, et No al. Efficacy and safety of adjunctive everolimus for treatment-resistant focal-onset seizures in tuberous sclerosis according to age: Insights from EXIST-3. Annals of Neurology 2017;Conference: 46th Annual Meeting of the Child Neurology Society. Kansas City, MO United States. 82(Supplement 21):S295. [36] Roberts C, Kuperman R, Koenig MK, Wu JY, Kohrman M, Curatolo P, No et al. Efficacy and safety of adjunctive everolimus for treatment-resistant

focal onset seizures in tuberous sclerosis according to age: Insights from EXIST-3. Neurology. Conference: 69th American Academy of Neurology Annual Meeting, AAN 2017;88(16 Supplement 1).

[37] Sahebkar F, Thiele E, Bebin EM, Bhathal H, Jansen FE, Kotulska K, et al. Cannabidiol (CBD) treatment in patients with seizures associated with tuberous sclerosis complex (TSC): A randomised, double-blind, placebo-controlled phase 3 trial (GWPCARE6). Developmental Medicine and Child Neurology 2020;Conference: British Paediatric Neurology Association Annual Meeting. Belfast United Kingdom. 62(Supplement 1):13.

[38] Samueli S, Abraham K, Dressler A, Groppel G, MuhlebnerFahrngruber A, Scholl T, et al. Efficacy and safety of Everolimus in children with TSC - associated epilepsy - Pilot data from an open singlecenter prospective study. Orphanet Journal of Rare Diseases 2016;11(1):1-8.

[39] Samueli S, Dressler A, Groppel G, Scholl T, Feucht M. Everolimus in infants with tuberous sclerosis complex-related West syndrome: First results from a single-center prospective observational study. Epilepsia 2018;59(9):e142-e146.

[40] Sanchez-Carpintero R, Cock H, Wu JY, Devinsky O, Joshi C, Miller I, et al. Time to onset of cannabidiol treatment effect and resolution of adverse events in tuberous sclerosis complex randomised controlled trial (GWPCARE6). Epilepsia 2021;Conference: 34th international Epilepsy Congress. Virtual. 62(SUPPL 3):32.

[41] Saneto R, Sparagana S, Kwan P, O'Callaghan F, Wheless J, Hyland K, et al. Efficacy of Add-on Cannabidiol (CBD) Treatment in Patients with Tuberous Sclerosis Complex (TSC) and a History of Infantile Spasms (IS): Post Hoc Analysis of Phase 3 Trial GWPCARE6. Annals of Neurology 2021;Conference: 50th Annual Meeting of the Child Neurology Society. Boston, MA United States. 90(SUPPL 26):S126.

[42] Saneto R, Sparagana S, Kwan P, O'Callaghan F, Wheless J, Hyland K, et al. Efficacy of add-on cannabidiol (CBD) Treatment in patients with tuberous sclerosis complex (TSC) and a history of infantile spasms (IS): Post hoc analysis of phase 3 Trial GWPCARE6. Neurology. Conference: 73rd Annual Meeting of the American Academy of Neurology, AAN 2021;96(15 SUPPL 1).

No

Excluded: No UK data

ERG: Questions this eligibility criteria. Appendix D of company submission only stipulates necessity for UK data in cost and resource use studies.

No

No

No

No

[43] Thiele EA, Bebin EM, Bhathal H, Jansen FE, Kotulska K, Lawson JA, No et al. Add-on Cannabidiol Treatment for Drug-Resistant Seizures in Tuberous Sclerosis Complex: A Placebo-Controlled Randomized Clinical Trial. JAMA Neurology 2021;78(3):285-292. [44] Thiele EA, Bebin EM, Filloux F, Kwan P, Loftus R, Sahebkar F, et al. Include Long-term safety and efficacy of add-on Cannabidiol (CBD) for treatment of seizures associated with tuberous sclerosis complex (TSC) in an openlabel extension (OLE) trial (GWPCARE6). Developmental Medicine and Child Neurology 2021;Conference: Annual Meeting of the British Paediatric Neurology Association. Virtual. 63(SUPPL 1):69. [45] Thiele EA, Bebin EM, Filloux F, Kwan P, Loftus R, Sahebkar F, et al. Published Long-term cannabidiol treatment for seizures in patients with tuberous after sclerosis complex: An open-label extension trial. Epilepsia company 2022;63(2):426-439. submission searches [46] Weinstock A, Bebin EM, Checketts D, Clark G, Szaflarski J, Seltzer L, No et al. Long-term efficacy and safety of cannabidiol (CBD) in patients with tuberous sclerosis complex (TSC): 4-year results from the expanded access program (EAP). Neurology. Conference: 73rd Annual Meeting of the American Academy of Neurology, AAN 2021;96(15 SUPPL 1). [47] Wheless J, Bebin E, Filloux F, Kwan P, Jansen F, Loftus R, et al. No Long-term Safety and Efficacy of Add-on Cannabidiol (CBD) for Treatment of Seizures Associated with Tuberous Sclerosis Complex (TSC) in an Open-Label Extension (OLE) Trial (GWPCARE6). Annals of Neurology 2021;Conference: 50th Annual Meeting of the Child Neurology Society. Boston, MA United States. 90(SUPPL 26):S129-S130. [48] Wheless J, Bebin EM, Filloux F, Kwan P, Jansen FE, Loftus R, et al. No Long-term safety and efficacy of add-on cannabidiol (CBD) for treatment of seizures associated with tuberous sclerosis complex (TSC) in an openlabel extension (OLE) Trial (GWPCARE6). Neurology. Conference: 73rd Annual Meeting of the American Academy of Neurology, AAN 2021;96(15 SUPPL 1). [49] Wu JY, Cock HR, Devinsky O, Joshi C, Miller I, Roberts CM, et al. Published Time to onset of cannabidiol treatment effect and resolution of adverse after events in tuberous sclerosis complex: Post hoc analysis of randomized company controlled phase 3 trial GWPCARE6. Epilepsia 2022. submission searches [50] Yum MS, Lee EH, Ko TS. Vigabatrin and mental retardation in No tuberous sclerosis: Infantile spasms versus focal seizures. Journal of Child Neurology 2013;28(3):308-313. [51] Zou L, Liu Y, Pang L, Ju J, Shi Z, Zhang J, et al. Efficacy and safety No of rapamycin in treatment of children with epilepsy complicated with tuberous sclerosis. [Chinese]. Zhonghua er ke za zhi 2014;Chinese journal of pediatrics. 52(11):812-816. Both duplicate references and articles published after the company submission searches were undertaken are greyed out in the table above.

Appendix 2: Updated base case analyses (as per Question

B13.a.)

Updated base case incremental cost-effectiveness analysis results disaggregated outcomes

Table 32 summarizes the patient QALY gain by component. The QALY gain associated with seizure-free days is the biggest driver of the incremental QALY associated with cannabidiol plus usual-care, compared with placebo plus usual-care, equating to *** of the incremental QALYs. The placebo plus usual-care arm had greater QALYs associated with the higher seizure frequency health states, reflecting the greater distribution of placebo plus usual-care patients within these states compared with cannabidiol plus usual-care.

Table 32: Summary of patient QALY gain by component

Table 33 summarizes the caregiver QALY decrements by health state. The base case analysis uses the seizure-free health state as the reference health state to calculate caregiver decrements. Therefore, there are no QALY decrements in this health state. The incremental QALY difference is driven by the difference between the model arms in the higher seizure frequency health states, reflecting the higher proportion of placebo plus usual-care patients who are distributed within the higher seizure frequency health states.

Table 33: Summary of caregiver QALY gain by component

==> picture [452 x 329] intentionally omitted <==

----- Start of picture text -----
Component QALY QALY Increment Absolute %
Cannabidiol Placebo + increment absolute
+ usual-care usual- increment
care)
Health Seizure- **** **** **** **** ****
state free per
day
≤ 1 **** **** **** **** ****
seizure
per day
> 1 - ≤ 2 **** **** **** **** ****
seizures
per day
> 2 - ≤ 4 **** **** **** **** ****
seizures
per day
> 4 **** **** **** **** ****
seizures
per day
Total ***** ***** ***** Total 100%
absolute
increment
Key: QALY, quality-adjusted life year; TAND, TSC-associated neuropsychiatric disorders.
----- End of picture text -----

A breakdown of cost by resource use type is presented in Table 34. Costs are presented by seizure frequency health state per week. The total lifetime discounted cost per patient for cannabidiol plus usual-care is ********; for placebo plus usualcare patients the total lifetime discounted cost per patient is ********.

The treatment cost associated with cannabidiol is one of the main drivers of the incremental cost of treatment compared to usual-care; however, this is offset by the

increased health state costs in the higher seizure frequency heath state (> 7 seizures per week) in the usual-care arm. Cost drivers for both arms include TAND management and health state costs.

Use of cannabidiol plus usual-care results in lower costs in the higher seizure frequency health states (> 2–≤ 7 seizures per week, and > 7 seizures per week) compared to placebo plus usual-care.

Table 34: Summary of costs by component

==> picture [452 x 390] intentionally omitted <==

----- Start of picture text -----
Component Cost Cost Increment Absolute %
Cannabidiol Placebo + increment absolute
+ usual-care usual- increment
care
Treatment costs
Drug acquisition ******* *** ******* ******* ***
costs
Usual-care costs ******* ******* *** *** ***
Adverse events **** *** **** * ***
Monitoring costs *** *** *** *** ***
Subsequent ****** ****** ******* ****** ***
treatment
TAND related cost ******** ******** ******* ****** ***
Health Seizure- ******* *** ******* ******* ***
state free over
costs 7 days
≤ 2 ******* ******* ****** ****** ***
seizures
> 2 - ≤ 7 ******** ******** ******** ******* ***
seizures
> 7 ******** ******** ******** *******
seizures
Total ******** ******** ******* Total 100%
absolute
increment
Key: TAND, TSC-associated neuropsychiatric disorders.
----- End of picture text -----*

Probabilistic sensitivity analysis

A PSA was conducted in which all inputs were varied simultaneously over 5,000 iterations, based on their distributional information. These results are provided in response to question B21 above.

Deterministic sensitivity analysis

The top 10 influential parameters on the ICER are presented as a tornado diagram. Each value was varied based on its uncertainty parameters.

Figure 14 presents the updated OWSA for cannabidiol plus usual-care compared with placebo plus usual-care with parameters shown in descending order of ICER sensitivity.

These results demonstrate that the ICER is most sensitive to the stopping rule assessment rate applied at 6 months for patients with a seizure frequency greater than seven seizures per week (highest seizure frequency category), the patient utility values applied to seizure-free patients, and response rates used to estimate the proportion of patients who benefit from a reduction in TAND symptoms.

Figure 14: Results of one-way sensitivity analysis

==> picture [446 x 294] intentionally omitted <==

Key: CBD, cannabidiol; ICER, incremental cost-effectiveness ratio; HS, health state; pw, per week; TAND, TSCassociated neuropsychiatric disorders; UC, usual-care.

Scenario analysis

The scenarios (updated) explored in the model are presented in Table 35. The top 10 most influential scenarios on the ICER are presented below in Figure 15. The results were relatively insensitive for most analyses, with cannabidiol remaining costeffective in all scenarios at a WTP threshold of £20,000 to £30,000 per QALY gained.

The most influential scenario, which resulted in a dominant ICER, and has the largest impact resulting in cost savings for the cannabidiol arm, is associated with the inclusion of wider social costs: social and educational care resource use, which was elicited via the two-round Delphi panel study.

Figure 15: Top 10 most influential scenarios on the ICER (deterministic) (PAS

price)

==> picture [452 x 96] intentionally omitted <==

==> picture [452 x 97] intentionally omitted <==

==> picture [452 x 96] intentionally omitted <==

Key: AED, anti-epileptic drug; HRQL, health-related quality of life; ICER, incremental cost-effectiveness ratio; PAS, patient access scheme; SmPC, Summary of Product Characteristics; SUDEP, sudden unexpected death in epilepsy; TAND, TSC-associated neuropsychiatric disorders; UC, usual-care.

Table 35: Scenario analysis results