3 Evidence

The appraisal committee (section 7) considered evidence from a number of sources. See the committee papers for full details of the evidence.

Clinical effectiveness

3.1 The company did a systematic literature review, and identified 4 randomised controlled trials (RCTs) evaluating the efficacy and safety of evolocumab for primary hypercholesterolaemia or mixed dyslipidaemia: LAPLACE‑2; RUTHERFORD‑2; GAUSS‑2; and DESCARTES. Of these, LAPLACE‑2 and GAUSS‑2 gave head-to-head evidence for evolocumab compared with ezetimibe, whereas RUTHERFORD‑2 and DESCARTES compared evolocumab with placebo only. GAUSS‑2 and RUTHERFORD‑2 only studied evolocumab in subgroups specified in the scope; people who cannot tolerate statins (defined as people who had tried at least 2 statins, but could not tolerate any dose or increase the dose above the smallest tablet strength because of intolerable muscle-related side effects), and those with heterozygous-familial hypercholesterolaemia respectively.

3.2 All the trials were phase III, double-blind RCTs, including a combined total of 3500 patients who were included only if they had an low-density lipoprotein cholesterol (LDL‑C) concentration equal to or greater than a certain concentration; this was 2.1 mmol/litre in LAPLACE‑2, 2.6 mmol/litre in RUTHERFORD‑2 and GAUSS‑2, and 1.9 mmol/litre in DESCARTES. All patients had background therapy during the trials: moderate- to high-intensity statin therapy (LAPLACE‑2), a statin with or without other lipid-lowering therapies (RUTHERFORD‑2), non-ezetimibe lipid-lowering therapy (GAUSS‑2), or diet alone or in combination with atorvastatin, ezetimibe, or both (DESCARTES). All trials except DESCARTES lasted for 12 weeks; DESCARTES was a long-term study that lasted for 52 weeks.

3.3 All the trials used a 2:1 randomisation to the evolocumab or control treatment arms. They gave evidence on the following treatment comparisons:

LAPLACE‑2 (n=1,899): eligible patients were randomised to 1 of 5 open-label statin cohorts; atorvastatin 10 mg or 80 mg, rosuvastatin 5 mg or 40 mg, or simvastatin 40 mg.
- Within the atorvastatin cohorts: evolocumab 140 mg every 2 weeks or 420 mg monthly in combination with placebo was compared with placebo every 2 weeks or monthly in combination with ezetimibe or placebo respectively.
- Within the rosuvastatin and simvastatin cohorts: evolocumab 140 mg 2 weekly or 420 mg monthly alone was compared with placebo every 2 weeks or monthly alone respectively.
RUTHERFORD‑2 (n=331): evolocumab 140 mg every 2 weeks or 420 mg monthly was compared with placebo every 2 weeks or monthly respectively.
GAUSS‑2 (n=307): evolocumab 140 mg every 2 weeks or 420 mg monthly in combination with placebo was compared with placebo every 2 weeks or monthly in combination with ezetimibe respectively.
DESCARTES (n=905): evolocumab 420 mg monthly was compared with placebo monthly.

3.4 The co-primary end points in LAPLACE‑2, RUTHERFORD‑2 and GAUSS‑2 were the percent change from baseline in LDL‑C level at week 12, and the mean percent change from baseline in LDL‑C level at weeks 10 and 12. In DESCARTES, the primary end point was the percent change from baseline in LDL‑C level at week 52. Other lipid parameters (including triglycerides and high-density lipoprotein cholesterol implicated in mixed dyslipidaemia) were measured in the trials as secondary end points, but the company's submission did not focus on them. None of the trials collected data on health-related quality of life.

Evidence review group's comments

3.5 The ERG considered the trials identified for evolocumab to be relevant, good-quality RCTs. It noted that the patient and disease characteristics at baseline were generally well-balanced across treatment arms. However, all 4 trials excluded patients with type 1 diabetes, or newly diagnosed or poorly controlled type 2 diabetes. The ERG questioned whether this could affect the generalisability of the trials because, in clinical practice, these patients are likely to present with co-morbid hypercholesterolaemia or mixed dyslipidaemia.

3.6 The ERG pointed out that the change in LDL‑C concentration is clinically important if it can be used as a surrogate for cardiovascular disease (CVD). Although the effect of statins on cardiovascular (CV) events is established, that of evolocumab has not been robustly shown in purposely designed clinical trials. The ERG noted that the ongoing FOURIER RCT will test whether LDL‑C is a valid surrogate for CV outcomes for evolocumab, which it considered to be a key area of uncertainty in the current evidence.

Clinical trial results

3.7 All efficacy and safety analyses were based on the modified intention-to-treat populations, that is, all patients who had at least 1 dose of study treatment. The results for the primary end points are shown in table 2 and table 3.

Table 2 Difference in percent change from baseline in LDL‑C concentration at week 12 (week 52 in DESCARTES)

Evolocumab dosage	Cohort	Versus placebo (%, 95% CI)	Versus ezetimibe (%, 95% CI)
LAPLACE‑2
140 mg every 2 weeks	Atorvastatin 10 mg	−74* (−81 to −68)	−44* (−50 to −37)
	Atorvastatin 80 mg	−80* (−91 to −68)	−50* (−61 to −39)
	Any atorvastatin	N/A	−47* (−53 to −41)¹
	Rosuvastatin 5 mg	−71* (−78 to −64)	N/A
	Rosuvastatin 40 mg	−71* (−80 to −63)	N/A
	Simvastatin 40 mg	−74* (−80 to −67)	N/A
	Any statin	−74* (−77 to −70)¹	N/A
420 mg monthly	Atorvastatin 10 mg	−61* (−68 to −54)	−43* (−50 to −36)
	Atorvastatin 80 mg	−74* (−84 to −65)	−41* (−51 to −32)
	Any atorvastatin	N/A	−43* (−48 to −37)¹
	Rosuvastatin 5 mg	−66* (−73 to −59)	N/A
	Rosuvastatin 40 mg	−59* (−70 to −48)	N/A
	Simvastatin 40 mg	−62* (−71 to −52)	N/A
	Any statin	−65* (−69 to −61)¹	N/A
GAUSS‑2
140 mg every 2 weeks	N/A	N/A	−39* (−45 to −34)
420 mg monthly	N/A	N/A	−38* (−43 to −33)
RUTHERFORD‑2
140 mg every 2 weeks	N/A	−61* (−67 to −55)	N/A
420 mg monthly	N/A	−60* (−68 to −53)	N/A
DESCARTES
420 mg monthly	N/A	−59* (−64 to −55)	N/A
*p<0.001
¹Using a fixed-effects model
Abbreviations: CI, confidence interval; LDL‑C, low-density lipoprotein cholesterol; N/A, not applicable.

Table 3 Difference in mean percent change from baseline in LDL‑C concentration at weeks 10 and 12

Evolocumab dosage	Cohort	Versus placebo (%, 95% CI)	Versus ezetimibe (%, 95% CI)
LAPLACE‑2
140 mg every 2 weeks	Atorvastatin 10 mg	−73* (−78 to −67)	−41* (−47 to −35)
	Atorvastatin 80 mg	−78* (−88 to −68)	−48* (−58 to −38)
	Any atorvastatin	N/A	−45* (−50 to −39)¹
	Rosuvastatin 5 mg	−70* (−76 to −63)	N/A
	Rosuvastatin 40 mg	−69* (−77 to −62)	N/A
	Simvastatin 40 mg	−73* (−78 to −67)	N/A
	Any statin	−72* (−75 to −69)¹	N/A
420 mg monthly	Atorvastatin 10 mg	−65* (−71 to −58)	−45* (−52 to −39)
	Atorvastatin 80 mg	−78* (−86 to −70)	−46* (−54 to −38)
	Any atorvastatin	N/A	−46* (−51 to −41)¹
	Rosuvastatin 5 mg	−69* (−75 to −62)	N/A
	Rosuvastatin 40 mg	−67* (−76 to −58)	N/A
	Simvastatin 40 mg	−70* (−79 to −61)	N/A
	Any statin	−70* (−73 to −66)¹	N/A
GAUSS‑2
140 mg every 2 weeks	N/A	N/A	−38* (−44 to −33)
420 mg monthly	N/A	N/A	−39* (−44 to −35)
RUTHERFORD‑2
140 mg every 2 weeks	N/A	−61* (−67 to −55)	N/A
420 mg monthly	N/A	−66* (−72 to −61)	N/A
*p<0.001
¹Using a fixed-effects model
Abbreviations: CI, confidence interval; LDL‑C, low-density lipoprotein cholesterol; N/A, not applicable.

3.8 The company presented subgroup analyses. It stated that in all of these evolocumab, compared with placebo or ezetimibe, was consistently effective in lowering LDL‑C, with no notable differences between subgroups.

3.9 The company presented interim results from 2 ongoing, long-term, extension studies, OSLER and OSLER‑2, which compared evolocumab plus standard of care (defined according to local guidelines) with standard of care alone. Eligible patients were those who completed treatment in a 'parent' study, including the RCTs identified for evolocumab. The company stated that OSLER and OSLER‑2 showed that the effect of evolocumab continued for over 2 years. It also presented a pre-specified exploratory analysis, which combined data from OSLER and OSLER‑2 (n=4465) on adjudicated CV events including death, myocardial infarction, unstable angina, coronary revascularisation, stroke, transient ischaemic attack, and heart failure. The rate of CV events at 1 year was 0.95% and 2.18% among patients randomised to evolocumab or standard of care respectively (hazard ratio 0.47, 95% CI 0.28 to 0.78; p=0.003).

3.10 TAUSSIG is an ongoing non-randomised, non-controlled, 5‑year extension study of evolocumab for severe familial hypercholesterolaemia. Among 142 patients with severe heterozygous-familial hypercholesterolaemia, the percent reduction from baseline in LDL‑C level at week 36 was 50.5%, with reductions ranging from 42.0% to 54.3% at earlier time points.

ERG's comments

3.11 The ERG noted that evolocumab, at both licensed doses, effectively reduced LDL‑C concentration from baseline compared with ezetimibe or placebo (p<0.001), with consistent results seen across all subgroups, including patients who can or cannot tolerate statins.

3.12 The ERG noted that, although none of the RCTs studied evolocumab in combination with ezetimibe, RUTHERFORD‑2 and DESCARTES included a subgroup in which patients had ezetimibe as background therapy with (DESCARTES) or without (RUTHERFORD‑2) high-dose atorvastatin. The ERG reported the results for these subgroups:

DESCARTES: The difference in percent change from baseline in LDL‑C level at week 52 between evolocumab 420 mg monthly plus ezetimibe plus statin and placebo plus ezetimibe plus statin was −49.3% (95% CI −59.5 to −39.1; p<0.001) in favour of evolocumab.
RUTHERFORD‑2: At week 12, the percent change from baseline in LDL‑C level favoured evolocumab 140 mg every 2 weeks plus ezetimibe compared with placebo plus ezetimibe, with a difference of −58.4% (95% CI −67.1 to −49.7; p<0.001). Evolocumab monthly plus ezetimibe was also more effective than placebo plus ezetimibe, with a difference of −60.9% (95% CI −71.0 to −50.8; p<0.001).

3.13 The ERG considered that the evidence from OSLER and OSLER‑2 was arguably not relevant to this appraisal. This was because the studies included populations from trials that were themselves excluded from the systematic review of clinical evidence.

Adverse effects of treatment

3.14 In addition to the data on adverse effects from the individual studies for evolocumab, the company presented integrated analyses of safety data from 6,026 patients with primary hypercholesterolaemia or mixed dyslipidaemia who had any dose of evolocumab. The key results of these analyses are summarised below:

Overall, evolocumab had a safety profile similar to the control treatment (placebo or ezetimibe) arms, with the incidence of adverse events being 51.1% compared with 49.6%. Most adverse events were mildly to moderately severe.
Serious adverse events occurred in 2.8% and 2.1% of patients who had evolocumab or any control treatment (placebo or ezetimibe) respectively.
Of patients who had evolocumab, 1.9% stopped treatment because of an adverse event compared with 2.3% of those who had placebo or ezetimibe.
The most common adverse events for evolocumab compared with placebo or ezetimibe were: nasopharyngitis (5.9% compared with 4.8%), upper respiratory tract infection (3.2% compared with 2.7%), headache (3.0% compared with 3.2%) and back pain (3.0% compared with 2.7%).
The company stated that anti-evolocumab antibodies were infrequent, non-neutralising, and not associated with clinically relevant adverse events.

ERG's comments

3.15 The ERG stated that evolocumab seemed to have an acceptable safety profile.

Cost effectiveness

3.16 The company submitted a de novo Markov economic model to assess the cost effectiveness of evolocumab in reducing CVD for primary hypercholesterolaemia (heterozygous-familial and non-familial) or mixed dyslipidaemia. The perspective of the analysis was that of the NHS and personal social services. Costs and health effects were modelled over a lifetime time horizon, and discounted at an annual rate of 3.5%. The cycle length in the model was 1 year.

Population, intervention and comparators

3.17 The company modelled 3 separate subpopulations:

non-familial hypercholesterolaemia without CVD
non-familial hypercholesterolaemia with CVD
heterozygous-familial hypercholesterolaemia (with or without CVD).

The company modelled the 2 non-familial hypercholesterolaemia populations based on the characteristics of the respective populations in LAPLACE‑2 with or without CVD. However, it only used data from the subset of patients who had an LDL‑C concentration over 2.5 mmol/litre to represent a population at high risk of CVD. For patients with heterozygous-familial hypercholesterolaemia, the company used the modified intention-to-treat population in RUTHERFORD‑2.

3.18 The intervention modelled in the base case was evolocumab 140 mg every 2 weeks; the company explored using the monthly dosage of evolocumab in scenario analyses (see section 3.48). For each population modelled, the company presented separate results for 4 treatment comparisons; 2 relevant to patients who can tolerate statins (who had atorvastatin as background therapy), and 2 relevant to those who cannot (who did not have any background lipid-lowering therapy):

For patients who can tolerate statins:
- evolocumab plus atorvastatin compared with ezetimibe plus atorvastatin
- evolocumab plus ezetimibe plus atorvastatin compared with ezetimibe plus atorvastatin.
For patients who cannot tolerate statins:
- evolocumab compared with ezetimibe
- evolocumab plus ezetimibe compared with ezetimibe.
  
  The company represented statins with atorvastatin because this is the statin recommended in NICE's guideline on lipid modification for people with or without CVD.

ERG's comments

3.19 The ERG's clinical advisers suggested that modelling a non-familial hypercholesterolaemia population with an LDL‑C concentration over 2.5 mmol/litre only was likely to have excluded many patients within this population. This was because many UK patients will have an LDL‑C concentration lower than 2.5 mmol/litre after taking statins.

3.20 The ERG noted that the company assumed that patients who can tolerate statins have the same characteristics as those who cannot. However, the risk of CVD was likely to be related to whether LDL‑C concentration can be controlled on statins. The ERG advised that GAUSS‑2 would have better represented patients with non-familial hypercholesterolaemia who cannot tolerate statins than LAPLACE‑2, noting that the company's analyses reflected the overall populations in LAPLACE‑2 and RUTHERFORD‑2, which included both patients who can and cannot tolerate statins, rather than either of these individual groups.

3.21 The ERG noted that the modelled heterozygous-familial hypercholesterolaemia population included patients with, and those without, CVD. It advised that modelling these groups separately may be more clinically appropriate.

Model structure

3.22 The company's model consisted of 24 mutually exclusive states:

3 acute states (in which the patient could stay for a maximum of 1 year unless the same event occurred in the next cycle)
- acute coronary syndrome (including myocardial infarction and unstable angina)
- ischaemic stroke
- heart failure
5 chronic states
- no CVD
- established CVD (including patients who had a history of stable angina, transient ischaemic attack, carotid stenosis, revascularisation without a history of myocardial infarction, abdominal aortic aneurism, or peripheral vascular disease)
- 3 post-event states
  - post-acute coronary syndrome
  - post-ischaemic stroke
  - post-heart failure
13 composite CVD states (formed of a combination of 2 or 3 acute and post-event states; these were used to remember the history of CV events and model the corresponding outcomes of recurring CV events)
3 death states: death from coronary heart disease, death from stroke and death from other causes.

Patients who had CVD could have either 1 of the events modelled separately (acute coronary syndrome, ischaemic stroke or heart failure), or 1 of the events in the established CVD state. This was because the events in the established CVD state were less severe than those modelled separately, and so would be associated with lower costs and better health outcomes. The company assumed that patients who started treatment in the model had it continuously over their lifetime.

3.23 Patients entered the model in different states depending on the population to which they belonged:

All patients with non-familial hypercholesterolaemia who did not have CVD entered the model in the no CVD state.
Patients with non-familial hypercholesterolaemia who had CVD entered the model in one of the 3 post-event states, or the established CVD state.
Patients with heterozygous-familial hypercholesterolaemia (with or without CVD) entered the model in one of the 3 post-CVD event states, the established CVD state, or the no CVD state.

Patients who entered the model in the no CVD state stayed in this state until they had CVD (that is, acute coronary syndrome, ischaemic stroke, heart failure, or one of the events in the established CVD state), or died. After the first CV event, patients could have no further CV events and move to the corresponding post-event state, have the same event again and stay in the same acute event state, have a different acute event and move to a composite state representing the post-event state for previous events and the new event, or die. Patients in a post-event state could have the same acute event and move to the corresponding acute state or composite state (if the patient had had other CV events), a different acute event and move to the relevant composite state, or die.

ERG's comments

3.24 The ERG considered that the company did not describe how it selected the states in the model, nor did it explain why they were more relevant than those used in previous models for primary hypercholesterolaemia or mixed dyslipidaemia, including the model for the NICE clinical guideline on lipid modification. The ERG was particularly concerned about the composite states in the model. This was because there were no data to inform them, and the company made several arbitrary assumptions about the costs and health effects in these states, which the ERG considered to have increased the uncertainty in the model.

Estimation of CVD risks

3.25 To estimate the risk of CVD in the model, the company used a 3‑step approach. First, it predicted the risk of CVD before treatment in patients in LAPLACE‑2 with an LDL‑C concentration at baseline over 2.5 mmol/litre (non-familial hypercholesterolaemia), and the modified intention-to-treat population in RUTHERFORD‑2 (heterozygous-familial hypercholesterolaemia). To do so, the company used published risk equations from the Framingham Heart Study for patients without CVD, and the REACH registry for patients with CVD. Second, the company estimated calibration (adjustment) factors from an analysis of data from the Clinical Practice Research Datalink (CPRD) and Hospital Episode Statistics (HES). Third, it adjusted the predicted risks of CVD based on the Framingham and REACH registry equations using these calibration factors to reflect real‑world data (CPRD and HES data). Because there was no CV risk equation specifically for patients with heterozygous-familial hypercholesterolaemia, the company adjusted the predicted risks of CVD in this population using a rate ratio of 7.1 (relative to patients without heterozygous-familial hypercholesterolaemia) derived from a study by Benn et al. (2012).

ERG's comments

3.26 The ERG considered the process by which the company estimated the risks of CVD to be circular, and unnecessarily complicated, with several assumptions and adjustments needed to estimate these risks. The ERG noted that the company used published equations to predict risks and then adjusted these to reflect real‑world data, although it could have estimated the risks directly from the analysis of real‑world data (CPRD and HES) without using risk equations. In the ERG's opinion, the company's approach did not add information compared with the CPRD and HES data.

3.27 The ERG stated that the company did not sufficiently justify why it selected the US-based Framingham risk equations for patients without CVD, instead of alternative equations such as the QRISK2 assessment tool, which was used in the model for NICE's guideline on lipid modification.

3.28 The ERG noted that the company added several constraints to prevent the model from generating negative transition probabilities. It considered that some of these constraints seemed arbitrary, and it was difficult to follow the logic supporting them from the information given by the company.

3.29 The ERG noted that the company predicted the risks of CVD in the heterozygous-familial hypercholesterolaemia population using the Framingham and REACH registry risk equations based on the entire RUTHERFORD‑2 population (which included patients with or without CVD). It did not consider this to be appropriate because these equations were only created for patients with or without a history of CVD. Also, the company used the study by Benn et al. (2012) to adjust the risk of CVD at baseline in patients with heterozygous-familial hypercholesterolaemia. The ERG noted that this study compared the risk of CV events between the general population and patients with heterozygous-familial hypercholesterolaemia. However, in the model, the rate ratio was not applied to the general population, but to the RUTHERFORD‑2 trial population, which was already at high risk of CVD. This was likely to overestimate the risk of CVD, and produce more favourable incremental cost-effectiveness ratios (ICERs) for evolocumab. The ERG also highlighted other studies, which suggested that the rate ratio derived from Benn et al. was likely to be an overestimate (see section 3.42).

Treatment effect

3.30 The objective of the model was to capture the lifetime progression of CVD among adults with primary hypercholesterolaemia (heterozygous-familial and non-familial) or mixed dyslipidaemia. Because none of the clinical trials for evolocumab had data on the direct effect of evolocumab on CVD, the company used estimates from the Cholesterol Treatment Trialists' (CTT) meta-analysis to convert the surrogate outcomes measured in the trials (LDL‑C concentration) to 'real‑world' outcomes (CV events).

3.31 The company used the estimates of treatment effect on LDL‑C concentrations from the head-to-head RCTs comparing evolocumab with ezetimibe. The data on other lipid parameters were not included in the model. For patients with non-familial hypercholesterolaemia, the company used LAPLACE‑2 for the treatment comparisons relevant to patients who can tolerate statins, and GAUSS‑2 for the comparisons relevant to those who cannot (see section 3.18). To source the clinical effectiveness in patients with heterozygous-familial hypercholesterolaemia who can tolerate statins, the company used RUTHERFORD‑2 for evolocumab and LAPLACE‑2 for ezetimibe because RUTHERFORD‑2 compared evolocumab with placebo only. For patients with heterozygous-familial hypercholesterolaemia who cannot tolerate statins, the company used GAUSS‑2. The company assumed that the treatment effect in the model lasted throughout the time horizon.

ERG's comments

3.32 The ERG noted that the company used LDL‑C concentration as a surrogate for CVD. It considered that, without robust data on the effect of evolocumab on CV outcomes, relying on a surrogate end point could be uncertain.

3.33 The ERG was concerned about the following assumptions in the model, which it considered uncertain:

For patients with non-familial hypercholesterolaemia who can tolerate statins, the treatment effect from LAPLACE‑2 could be generalised to the subset of patients with an LDL‑C concentration over 2.5 mmol/litre.
The treatment effects from LAPLACE‑2 and GAUSS‑2 would be the same in all patients whether or not they have diabetes or other risk factors for CVD.
The treatment effect would last indefinitely in the model.

3.34 The ERG considered the following assumptions made by the company to estimate the relationship between changes in LDL‑C concentration and CV events to be arbitrary, implausible or uncertain:

The relationship between LDL‑C concentration and CVD was the same for patients with or without a history of CVD.
The effect of reducing LDL‑C concentration on non-fatal myocardial infarction was the same as that on heart failure (first event). The ERG also noted that the company assumed that reducing LDL‑C concentration in patients with heart failure (either acute, post-event state or combined state) would reduce death from coronary heart disease even though it recognised the lack of benefit for lipid-lowering therapies once patients had heart failure.
The relationship between LDL‑C concentration and non-fatal myocardial infarction (secondary prevention) would apply to patients moving from the no CVD state to the established CVD state.
Reducing LDL‑C concentration had no effect on death from stroke.

Health-related quality of life and costs

3.35 To populate the base-case model with utility data, the company used the utility values informing the model developed for NICE's guideline on lipid modification, with some adjustments to match the states in the model:

Established CVD: in the company's model, this state included various CV events, 1 of which was stable angina. The original utility value for stable angina was 0.808 (for both acute and post event). This was unexpectedly lower than the value for post myocardial infarction (0.880) and post unstable angina (0.880), which are considered more severe than stable angina. Because of this, the company used the utility value for the post-acute coronary syndrome state (0.880) for the established CVD state.
Acute states: in the model, the acute coronary syndrome state included myocardial infarction and unstable angina. The original utility values for the acute events of these 2 diseases were 0.760 and 0.77 respectively. The company chose the higher utility value (0.77) for the acute coronary syndrome state. The utility values for the ischaemic stroke and heart failure were 0.63 and 0.68 respectively.
Post-event states: the utility value was 0.88 for acute coronary syndrome, 0.63 for ischaemic stroke, and 0.68 for heart failure.
Composite states: the company assumed the lowest utility value in the individual acute or post-event states included in that composite state.

In line with NICE's guideline on lipid modification, the company assumed that the utility depends on age, and so it multiplied the utility values (multipliers) by age-adjusted utility values for the general population based on a study by Dolan et al. (1996). The company also gave details of a company-sponsored study that used the time trade-off method to estimate utility values for patients with CVD. It explored using utility values from this study in scenario analyses (see section 3.48).

3.36 The company's model included treatment and monitoring costs, and those associated with the model health states. The cost of evolocumab in the model included the patient access scheme discount. The company assumed that patients who started treatment with evolocumab had 1‑hour training by a nurse to self-administer the treatment at a cost of £84.00; no additional monitoring was assumed for patients having evolocumab compared with those having ezetimibe. The company equated the costs in the composite states to the highest cost in the individual states included in that state.

ERG's comments

3.37 The ERG stated that, of the 7 acute and post-event states in the model, only 3 states (acute coronary syndrome, heart failure and post heart failure) were based on the EQ‑5D questionnaire. The other utility multipliers were taken from studies that used the time trade-off method, and so did not meet the NICE reference case (the methods considered by NICE to be the most appropriate for technology appraisals). The ERG also noted that some of the utility multipliers did not match the states in the model for which they were used.

3.38 Overall, the ERG did not have major concerns about the costs used in the company's model.

Original base-case results (including the patient access scheme)

3.39 In its patient access scheme submission accompanying the original submission, the company presented incremental cost-effectiveness analyses for all 3 populations. The original base-case ICERs, including the patient access scheme, are shown in table 4. All of these ICERs are based on the every 2 weeks dosage of evolocumab 140 mg.

Table 4 Company's original base-case ICERs (including the patient access scheme)

Treatment comparison	ICER (£/QALY)
	Non-familial hypercholesterolaemia		Heterozygous-familial hypercholesterolaemia
	Without CVD	With CVD	With or without CVD
Ezetimibe plus statin	N/A	N/A	N/A
Evolocumab plus statin	74,331	46,005	22,902
Ezetimibe	N/A	N/A	N/A
Evolocumab	78,879	49,278	23,927
Ezetimibe	Not presented	N/A	N/A
Evolocumab plus ezetimibe	Not presented	52,811	25,609
Ezetimibe plus statin	Not presented	N/A	N/A
Evolocumab plus ezetimibe plus statin	Not presented	50,880	24,826
Abbreviations: CVD, cardiovascular disease; ICER, incremental cost-effectiveness ratio; N/A, not applicable; QALY, quality-adjusted life year.

The company also presented sensitivity analyses (deterministic and probabilistic), scenario analyses, and subgroup analyses for selected populations and treatment comparisons. All the analyses presented in the company's patient access scheme submission accompanying the original submission were superseded by the company's new evidence in response to consultation (see sections 3.44–3.50).

3.40 At the start of treatment evolocumab is prescribed in specialist secondary care clinics, but people may move from secondary to primary care after 2–3 years because routine lipid management is an area of standard GP practice. This has potential implications for the proposed simple discount patient access scheme because simple discounts do not apply when drugs are prescribed by GPs using FP10 prescriptions. In response to a request from NICE, the company presented sensitivity analyses varying the proportion of patients who may move from secondary care to primary care (after which point the simple discount does not apply), and the time patients spend in secondary care before this happens.

ERG's comments

3.41 In summary, the ERG advised some caution in the interpretation of the company's results because of:

the selected populations used in the model (see sections 3.19–3.21)
the use of multiple composite states, which were populated using many assumptions and little evidence (see section 3.24)
the circular approach used by the company to predict risks of CVD (see section 3.26)
the likely overestimation of the risk of CVD in the heterozygous-familial hypercholesterolaemia population (see section 3.29)
the uncertainty about the relationship between LDL‑C reduction and reductions in CV events (see section 3.32).

3.42 The ERG stated that calibration rate of 7.1, which was likely to be overestimated (see section 3.29), was a key driver of the cost effectiveness of evolocumab for heterozygous-familial hypercholesterolaemia. It noted that the company estimated that about 50% of the patients having statins would have a CV event or die from other causes 8–9 years after starting treatment. In comparison, a long-term cohort study identified by the ERG (Versmissen et al. 2008) indicated that, within the same time period, 10% of patients with heterozygous-familial hypercholesterolaemia having statins would have coronary heart disease. The ERG also highlighted other studies, which suggested that the rate of death from cardiovascular or coronary artery disease may increase in patients with heterozygous-familial hypercholesterolaemia, although not to the extent assumed by the company; these studies also reported no statistically significant difference for all-cause mortality. Specifically, a UK study by Neil et al. (2008) reported standardised mortality ratios in patients with heterozygous-familial hypercholesterolaemia treated with statins of 1.03 (primary prevention) and 3.88 (secondary prevention) for death from coronary artery disease, and 0.94 for all-cause mortality, which was not statistically significant (p=0.31). Similar results were also reported by a recent Norwegian study by Mundal et al. (2014).

3.43 The ERG did a threshold analysis to determine the minimum calibration factor that must be applied to the predicted CV risks in the heterozygous-familial hypercholesterolaemia population for the ICER comparing evolocumab with ezetimibe to be below £30,000 per quality-adjusted life year (QALY) gained. This suggested that the ICER increased considerably as the assumed calibration factor decreased, with calibration factors greater than 4.5–5.6 needed for evolocumab compared with ezetimibe to have an ICER below £30,000 per QALY gained.

Company's new evidence in response to consultation

3.44 In response to consultation on the first appraisal consultation document, in which evolocumab was not recommended for primary hypercholesterolaemia (heterozygous-familial and non-familial) or mixed dyslipidaemia, the company was permitted to submit revised cost-effectiveness analyses. These incorporated the following changes, which reflected the committee's preferred analyses in the first appraisal consultation document:

Use of the baseline characteristics of the population in GAUSS‑2 to model patients with non-familial hypercholesterolaemia who cannot tolerate statins.
Modelling of the heterozygous-familial hypercholesterolaemia population with or without CVD separately.
Use of the QRISK2 assessment tool to estimate the level of CVD risk in people without CVD (non-familial or heterozygous-familial hypercholesterolaemia).
Adjustment of the risk of CVD in the heterozygous-familial hypercholesterolaemia population using a rate ratio of 6.1 derived from Benn et al. (2012).
Use of the equation from the Health Survey for England to inform the relationship between age and background health-related quality of life.
Modelling of subgroups reflecting all the characteristics of the actual subgroup in clinical trials.

3.45 The company's revised base-case ICERs, including the patient access scheme, are presented in table 5. All of these ICERs are based on the every 2 weeks' dosage of evolocumab 140 mg.

Table 5 Company's revised base-case ICERs (including the patient access scheme)

Treatment comparison	ICER (£/QALY)
	Non-familial hypercholesterolaemia		Heterozygous-familial hypercholesterolaemia
	Without CVD	With CVD	Without CVD	With CVD
Ezetimibe plus statin	N/A	N/A	N/A	N/A
Evolocumab plus statin	69,249	45,439	23,536	29,910
Ezetimibe	N/A	N/A	N/A	N/A
Evolocumab	38,458	30,985	21,921	25,293
Ezetimibe	N/A	N/A	N/A	N/A
Evolocumab plus ezetimibe	41,911	33,814	23,602	27,390
Ezetimibe plus statin	N/A	N/A	N/A	N/A
Evolocumab plus ezetimibe plus statin	78,459	50,257	25,583	32,698
Abbreviations: CVD, cardiovascular disease; ICER, incremental cost-effectiveness ratio; N/A, not applicable; QALY, quality-adjusted life year.

3.46 The company revised its deterministic sensitivity analyses, varying the input values in the model for 1 parameter at a time. Among the most influential parameters were the treatment duration assumed in the model, the effect of reducing LDL‑C concentration on death from coronary heart disease or ischaemic stroke, and the heterozygous-familial hypercholesterolaemia calibration rate ratio.

3.47 The company revised its probabilistic sensitivity analyses, varying parameters simultaneously with values from a probability distribution. The probabilistic ICERs were slightly higher than the deterministic ones. The company reported the following probabilities of evolocumab being cost effective:

Non-familial hypercholesterolaemia population without CVD: 0% compared with any comparator at a maximum acceptable ICER of £30,000 per QALY gained.
Non-familial hypercholesterolaemia population with CVD: 0% compared with ezetimibe plus statin at a maximum acceptable ICER of £30,000 per QALY gained. Compared with ezetimibe alone, the probability was 5.4% when used as an add-on to ezetimibe, and 30.1% when used as an alternative to ezetimibe.
Heterozygous-familial hypercholesterolaemia without CVD: there was a low probability of evolocumab being cost effective at a maximum acceptable ICER of £20,000 per QALY gained (less than 20%). At a maximum acceptable ICER of £30,000 per QALY gained, the probability exceeded 85% for all comparisons.
Heterozygous-familial hypercholesterolaemia with CVD: there was a low probability of evolocumab being cost effective at a maximum acceptable ICER of £20,000 per QALY gained (less than 5%). At a maximum acceptable ICER of £30,000 per QALY gained, the probability ranged from 15% to 42% compared with ezetimibe plus statin, and from 69% to 84% compared with ezetimibe alone.

3.48 The company revised its scenario analyses, varying the input values for certain parameters. The model was most sensitive to applying alternative discount rates (0% for costs, and 0% or 6% for health effects), having evolocumab monthly (as opposed to every 2 weeks), having treatment for a shorter duration of 5 or 10 years, and using utility values from the company-sponsored time trade-off study.

Subgroups

3.49 The company presented a range of subgroup analyses based on actual patient-level characteristics when possible. It used LAPLACE‑2 and GAUSS‑2 for the subgroups of the non-familial hypercholesterolaemia population with CVD who can or cannot tolerate statins respectively, and RUTHERFORD‑2 for the subgroups of the heterozygous-familial hypercholesterolaemia population. For the subgroups of the non-familial hypercholesterolaemia population with CVD who can tolerate statins, the company also used CPRD data to model additional high-risk subgroups. The company presented results for the following subgroups:

Non-familial hypercholesterolaemia population with CVD who can tolerate statins:
- People with 1, and separately those with 2, of the following risk factors:
  - Based on LAPLACE‑2: mean LDL‑C concentration of 3.0–6.0 mmol/litre (intervals of 0.5 mmol/litre), diabetes, and acute coronary syndrome.
  - Based on CPRD: mean LDL‑C concentration of 3.0–6.0 mmol/litre (intervals of 0.5 mmol/litre), diabetes, 2 vascular beds, 3 vascular beds, atrial fibrillation, and acute coronary syndrome.
- People with 3 of the following risk factors (based on CPRD): mean LDL‑C concentration of 3.0–6.0 mmol/litre (intervals of 0.5 mmol/litre), diabetes, 2 vascular beds, 3 vascular beds, atrial fibrillation, and acute coronary syndrome.
Non-familial hypercholesterolaemia population with CVD who cannot tolerate statins:
- People with 1, and separately those with 2, of the following risk factors (based on GAUSS‑2): mean LDL‑C concentration of 3.0–6.0 mmol/litre (intervals of 0.5 mmol/litre), diabetes, and acute coronary syndrome.
Heterozygous-familial hypercholesterolaemia population with or without CVD who can tolerate statins:
- People with 1 risk factor (based on RUTHERFORD‑2): mean LDL‑C concentration of 3.0–6.0 mmol/litre (intervals of 0.5 mmol/litre).

3.50 The ICER ranges from the company's analyses are presented below:

Non-familial hypercholesterolaemia population with CVD who can tolerate statins (evolocumab plus statin compared with ezetimibe plus statin; base-case ICER £45,439 per QALY gained):
- People with 1 risk factor:
  - Based on LAPLACE‑2: from £34,277 to £51,571 per QALY gained (mean LDL‑C concentrations of 6.0 mmol/litre and 3.0 mmol/litre respectively).
  - Based on CPRD: from £32,622 (mean LDL‑C concentration of 6.0 mmol/litre) to £49,404 (diabetes) per QALY gained.
- People with 2 risk factors:
  - Based on LAPLACE‑2: from £31,340 (mean LDL‑C concentration of 4.5 mmol/litre and diabetes) to £41,509 (mean LDL‑C concentration of 3.5 mmol/litre and acute coronary syndrome) per QALY gained.
  - Based on CPRD: from £21,203 (mean LDL‑C concentration of 4.5 mmol/litre and 3 vascular beds) to £33,631 (mean LDL‑C concentration of 3.5 mmol/litre and diabetes) per QALY gained.
- People with 3 risk factors (based on CPRD): from £18,343 (mean LDL‑C concentration of 4.0 mmol/litre, acute coronary syndrome and 3 vascular beds) to £30,524 (mean LDL‑C concentration of 3.0 mmol/litre, diabetes and 2 vascular beds) per QALY gained.
Non-familial hypercholesterolaemia population with CVD who cannot tolerate statins (evolocumab compared with ezetimibe; base-case ICER £30,985 per QALY gained):
- People with 1 risk factor (based on GAUSS‑2): from £24,007 (acute coronary syndrome) to £43,180 (mean LDL‑C concentration of 3.0 mmol/litre) per QALY gained.
- People with 2 risk factors (based on GAUSS-2): from £25,347 (mean LDL‑C concentration of 4.5 mmol/litre and diabetes) to £31,842 (mean LDL‑C concentration of 3.5 mmol/litre and acute coronary syndrome) per QALY gained.
Heterozygous-familial hypercholesterolaemia population without CVD who can tolerate statins (evolocumab plus statin compared with ezetimibe plus statin; base-case ICER £23,536 per QALY gained):
- People with 1 risk factor (based on RUTHERFORD‑2): from £18,436 to £29,304 per QALY gained (mean LDL‑C concentrations of 6.0 mmol/litre and 3.0 mmol/litre respectively).
Heterozygous-familial hypercholesterolaemia population with CVD who can tolerate statins (evolocumab plus statin compared with ezetimibe plus statin; base-case ICER £29,910 per QALY gained):
- People with 1 risk factor (based on RUTHERFORD‑2): from £23,244 to £38,133 per QALY gained (mean LDL‑C concentrations of 6.0 mmol/litre and 3.0 mmol/litre respectively).

ERG's critique of the company's new evidence

3.51 The ERG noted that although the company appeared to have used the QRISK2 assessment tool appropriately, several assumptions and adjustments were still needed to estimate and apply the calibration factors for the non-familial hypercholesterolaemia population.

3.52 The ERG was concerned about the rate ratio of 6.1 used to adjust the risk of CVD for heterozygous-familial hypercholesterolaemia, reiterating that this was inappropriately applied to the RUTHERFORD-2 trial population, which was already at high risk of CVD (see section 3.29). The ERG maintained that it would be more appropriate to estimate the risk of CVD directly from the CPRD and HES data, or other routine data.

3.53 The ERG considered that the company's revised subgroup analyses were broadly reasonable. However, it highlighted uncertainties relating to the following:

Applying the same calibration factors from the whole non-familial hypercholesterolaemia population with CVD to the subgroups of that population.
Assuming that the association between reduced LDL‑C concentrations and improved CV outcomes did not depend on risk factors.
Assuming that the treatment effect in subgroups was the same as in the full trial populations.