3 The company's submission

The appraisal committee (section 6) considered evidence submitted by Daiichi Sankyo and a review of this submission by the evidence review group (ERG; section 7).

Clinical effectiveness

Overview of clinical trials

3.1

The company did a systematic review of the literature to identify studies evaluating the clinical effectiveness and safety of edoxaban for the treatment and secondary prevention of venous thromboembolism (VTE). It identified 1 relevant randomised clinical trial (Hokusai‑VTE). The company did not find any head‑to‑head studies, so it conducted a network meta‑analysis that compared edoxaban against treatment with rivaroxaban, dabigatran etexilate, and warfarin. The company did not find any relevant non‑randomised studies.

3.2

Hokusai‑VTE was an international (37 countries including the UK) randomised, double‑blind, non‑inferiority trial. It compared initial treatment with heparin followed by edoxaban or warfarin for treating acute symptomatic VTE or preventing symptomatic recurrent VTE. Eligible adults were randomised in a 1:1 ratio with stratification by presenting diagnosis, temporary baseline risk factors (such as trauma, surgery or immobilisation) and the dose of edoxaban (which was reduced for patients with moderate renal impairment, those who were having concomitant treatment with potent permeability glycoprotein [P‑glycoprotein] inhibitors, or those who weighed 60 kg or less). A total of 8,292 patients were randomly assigned to either the edoxaban group (n=4,143) or the warfarin group (n=4,149).

3.3

All patients had initial therapy with open‑label heparin for at least 5 days. Edoxaban or warfarin was administered in a double‑blind, double‑dummy fashion:

  • Patients in the edoxaban group had placebo warfarin during initial heparin therapy. After stopping heparin, they continued placebo warfarin (adjusted to maintain a sham international normalised ratio [INR] of 2.0 to 3.0) and started 60 mg of edoxaban once daily (or 30 mg once daily in patients who needed dose reduction at randomisation or during the study).

  • Patients in the warfarin group started warfarin during initial heparin therapy. After stopping heparin, they continued warfarin (adjusted to maintain an INR of 2.0 to 3.0) and started placebo edoxaban.

    Treatment with edoxaban or warfarin continued for at least 3 months and up to a maximum of 12 months, with treatment duration based on risk of recurrent VTE, risk of bleeding, and patient preference. The company noted that people who stayed in the trial for longer than 3 months were mostly those identified as being at higher risk of recurrence. In both groups, the median duration of treatment was about 260 days (8.5 months). Around 60% of patients had treatment for more than 6 months; 40% of patients continued treatment for 12 months.

3.4

Patient characteristics were similar between the treatment groups: mean age was 56 years, the majority of patients were male (57%) and patient ethnicity was reported as white (70%), Asian (21%), black (4%) or 'other' (5%). A total of 4,921 patients presented with deep vein thrombosis (DVT) only, and 3,319 with pulmonary embolism (PE; with or without DVT).

3.5

The primary efficacy outcome measure in the trial was the incidence of symptomatic recurrent VTE (a composite measure of recurrent DVT, new non‑fatal symptomatic PE, and fatal PE) during the 12‑month study period. Secondary outcomes included clinically relevant bleeding during treatment or within 3 days of interrupting or stopping the study drug (this was referred to by the company as the primary safety outcome; a composite of major bleeding and clinically relevant non‑major bleeding), and a composite clinical efficacy outcome of recurrent VTE and all‑cause mortality during the 12‑month study period.

3.6

The trial included 3 analysis sets (modified intention to treat, per protocol analysis, and safety analysis) and 2 study periods (overall study period and on‑treatment period). The primary efficacy analyses were done in the modified intention to treat population (analyses based on randomised treatment, even if the patient did not receive this) for the overall study period, which included 4,118 patients randomised to edoxaban and 4,122 randomised to warfarin. Summary statistics were provided for the on‑treatment period. The safety population (analyses based on treatment received) was used for outcomes related to safety; this population was identical to the modified intention to treat population, because all patients had the treatment to which they were randomised. The company did pre‑specified subgroup analyses for the primary efficacy outcome for various patient and disease characteristics, including whether the presenting diagnosis was PE with or without DVT (n=1,650 in the edoxaban group; n=1,669 in the warfarin group), or DVT only (n=2,468 in the edoxaban group; n=2,453 in the warfarin group).

Clinical trial results

3.7

The company presented results for the primary efficacy outcome (symptomatic recurrent VTE) using a pre‑specified non‑inferiority margin of 1.5 for the upper 95% confidence interval (CI) of the hazard ratio (HR) (that is, the non‑inferiority analyses demonstrated statistically significant non‑inferiority if the upper boundary of the 95% CI for the outcome was below 1.5). Edoxaban demonstrated statistically significant non‑inferiority for the primary outcome when compared with warfarin (p<0.0001; see tables 1 and 2). Similar results were obtained for the on‑treatment period (HR, 0.82; 95% CI 0.60 to 1.14; p<0.0001 for non‑inferiority).

Table 1 Symptomatic recurrent VTE (overall study period)

Primary outcome

Edoxaban

(n=4,118)

Warfarin

(n=4,122)

All patients with recurrent VTE, n (%)

HR 0.89 (95% CI 0.703 to 1.128; p<0.0001 for non-inferiority)

130 (3.2)

146 (3.5)

Table 2 Type of first recurrent VTE (overall study period)

Type of first recurrent VTE, n (%)

Edoxaban

(n=4,118)

Warfarin

(n=4,122)

PE with/without DVT

73 (1.8)

83 (2.0)

PE‑related deaths

24 (0.6)

24 (0.6)

Fatal PE

4 (<0.1)

3 (<0.1)

Non‑fatal PE

49 (1.2)

59 (1.4)

Unexplained death (with VTE not ruled out)

20 (0.5)

21 (0.5)

DVT alone

57 (1.4)

63 (1.5)

3.8

Some patients in Hokusai‑VTE completed EuroQoL‑5‑Dimensions (EQ‑5D) assessments and utility scores were determined for all patients using the UK time trade‑off (TTO) value set, at baseline and then at 3‑month intervals. The company reported that the results should be interpreted with caution because data were too limited to compare the effects of edoxaban and warfarin on health‑related quality of life.

Adverse effects of treatment

3.9

The company presented the results of safety analyses, all of which related to adverse events, using the safety analysis set for the on‑treatment population. For the safety outcomes, results included:

  • Bleeding (major or clinically relevant non‑major bleeding [CRNM], primary safety outcome): edoxaban was associated with fewer bleeding events (p=0.004 for superiority, see table 3). Major bleeding in critical sites included 5 intracranial haemorrhage events (none of which were fatal) in the edoxaban group, and 18 (6 fatal) in the warfarin group. Major bleeding in non‑critical sites included 27 gastrointestinal tract bleed events (1 fatal) in the edoxaban group and 18 (2 fatal) in the warfarin group.

  • Mortality: the rate for VTE‑related death, cardiovascular death and other known causes (cancer, bleeding, infectious disease or other) was 0.8% in both groups (no statistical analyses presented).

Table 3 Bleeding events (on‑treatment study period)

Events

Edoxaban

(n=4,118)

Warfarin

(n=4,122)

Major and CRNM bleeding, n (%)

HR 0.81 (95% CI 0.705 to 0.936; p=0.004)

349 (8.5)

423 (10.3)

Major bleeding, n (%)

HR 0.84 (95% CI 0.592 to 1.205; p=0.3521)

56 (1.4)

66 (1.6)

Fatal, n (%)

2 (<0.1)

10 (0.2)

CRNM bleeding, n (%)

HR 0.80 (95% CI 0.68 to 0.93; p=0.004)

298 (7.2)

368 (8.9)

All bleeding, n (%)

895 (21.7)

1,056 (25.6)

Network meta‑analysis

3.10

The company did a Bayesian network meta‑analysis that compared edoxaban with warfarin, apixaban, dabigatran etexilate and rivaroxaban for the treatment and secondary prevention of VTE. The company stated that, to reduce heterogeneity between the trials, it included only trials from which it could derive data for a fixed 6‑month study period (that is, either the trial was fixed for 6 months or, if the trial was longer than 6 months, only data up to 6 months were extracted). This led to the inclusion of 6 randomised controlled trials, all of which compared the treatment of interest with low‑molecular‑weight heparin plus warfarin.

3.11

The company used a fixed‑effects Bayesian model to synthesise the results for each outcome, and used the posterior distribution to derive comparative treatment effects (shown in the company submission as odds ratios with 95% credible intervals [CrI] for ease of interpretation), using warfarin as the reference treatment. It did network meta‑analyses for the whole VTE population, and also secondary analyses for the PE population only, for the following outcomes:

  • VTE recurrence

  • Bleeding (major bleeding, non‑major bleeding and a composite of both outcomes)

  • VTE‑related death

  • Net clinical benefit (composite of VTE recurrence, major bleeding and all‑cause mortality).

    The base case network meta‑analysis was for the outcome of VTE recurrence at 6 months. The results demonstrated no large differences in treatment effects for the outcomes, and the odds ratios had wide credible intervals that approached or crossed '1' (that is, there was no difference in effect). Results for the PE subgroup were consistent with the main population.

3.12

The company did a qualitative assessment of the consistency of evidence generated by direct and indirect comparisons, and stated that the network meta‑analysis results for each oral anticoagulant compared with warfarin were generally consistent with the direct evidence from the original trials.

3.13

The company did not quantitatively measure heterogeneity between the trials, because a low number of the trials had similar designs or comparators. Instead, it did a qualitative analysis of heterogeneity and found that there was variance across the trials in blinding, heparin lead‑in time, duration of treatment and the proportion of patients with extensive PE. The company concluded that there were substantial differences between the trial designs and the populations used for the network meta‑analyses, which meant that any results should be interpreted with caution. For example, Hokusai‑VTE was designed differently to other studies because it:

  • included initial heparin therapy for at least 5 days, whereas trials for rivaroxaban did not include initial heparin therapy

  • allowed the dosage of edoxaban to be reduced at any point, which was not allowed in other trials

  • was the only trial to have a flexible treatment duration (so that later data included more high‑risk patients who needed to remain on treatment for longer)

  • did not allow for an extension, whereas patients in other trials were entered into extension studies

  • was double‑blind, whereas trials for rivaroxaban had open‑label designs.

Evidence review group's comments on the company's clinical effectiveness evidence

3.14

The evidence review group (ERG) stated that the systematic review done by the company was reasonable, and that the identified trial (Hokusai‑VTE) was well conducted and appropriately powered to demonstrate non‑inferiority. However, the ERG stated that while the trial population was similar to trials for other newer anticoagulants, the generalisability of the trial to the UK population was uncertain because it included more people who were reported as 'Asian' and a younger population than would be expected in UK clinical practice. The treatment duration also suggested a higher‑risk population (60% had treatment for longer than 6 months). The ERG noted that the design of Hokusai‑VTE was different from other trials evaluating newer anticoagulants for this condition because patients could have their drug dosage altered at the beginning of or during the trial. The ERG was unsure about whether the monitoring undertaken in the trial would be needed in clinical practice but it stated that, because the dose of edoxaban was related to variable patient factors, it should be expected that some periodic monitoring would be needed to check whether a patient's dose needed to be reduced.

3.15

The ERG noted that the company had not presented evidence for the subgroups of:

  • people with active cancer: the ERG noted that Hokusai‑VTE included people with a diagnosis of active cancer and an analysis of this subgroup had been planned but not conducted, which may have been because of limited data (n=208)

  • people for whom treatment with warfarin was not appropriate: the ERG agreed that this exclusion was legitimate because of a lack of data

  • people with DVT: the ERG disagreed with the exclusion of this subgroup. Because trials in the network for rivaroxaban, warfarin and edoxaban included data for people with DVT, the ERG was able to conduct an analysis for this population.

3.16

The ERG noted that although health‑related quality‑of‑life data had been collected in the trial, the small number of respondents severely limited the usefulness of this information.

3.17

The ERG considered the results of the trial, noting that recurrent VTE was more frequent for edoxaban in the first 30 days of the trial (21 events in the edoxaban group compared with 15 events in the warfarin group, in the DVT subgroup; and 9 events in the edoxaban group compared with 7 events in the warfarin group, in the PE subgroup). However, this difference was not statistically significant. Overall, the trial demonstrated statistically significant non‑inferiority compared with warfarin for VTE recurrence. The company reported that edoxaban was statistically significantly superior (p=0.004) for the outcome of major and non‑major bleeding. However, the ERG noted that this difference was driven by non‑major bleeding events (major bleeding events were not statistically significantly different for edoxaban compared with warfarin).

3.18

The ERG considered the approaches taken by the company for different aspects of the network meta‑analysis. It stated that:

  • the approach to reduce heterogeneity by including only trials with 6‑month data was reasonable

  • the qualitative approach to considering heterogeneity was pragmatic, but the ERG would have preferred it if the company had performed 2 separate analyses based on the patient's initial diagnosis (DVT or PE)

  • the use of a fixed‑effects model was reasonable because of the small number of studies included in the network and the lack of direct comparison of newer oral anticoagulants with any treatment other than warfarin

  • the quality assessments of the trials included in the network were mostly adequate

  • the indirect comparisons from the network meta‑analysis were consistent with the direct evidence from the individual trials.

3.19

The ERG stated that any differences in outcomes between the treatments in the network were small and more likely to be as a result of random chance. This is because the non‑inferiority design of the original studies limited the opportunity for any treatment in the network to have any statistically significant differences. Overall, there were no large differences in treatment effects for the outcomes and the odds ratios had wide credible intervals that approached or crossed '1' (that is, there was no difference in effect), indicating a large amount of uncertainty. The ERG noted that results for the subgroups of DVT (conducted by the ERG because the company had not included this) and PE showed results that were consistent with the main population.

Cost effectiveness

3.20

The company identified 12 studies related to the comparators in the scope in its systematic review of cost‑effectiveness analyses. The company found no studies evaluating the cost‑effectiveness of edoxaban for treating and preventing VTE, and developed a new economic model.

3.21

The company developed a Markov model that compared edoxaban with warfarin, rivaroxaban and dabigatran etexilate for the treatment and secondary prevention of an acute VTE event. The model included 12 states representing treatment status (on‑treatment or off‑treatment health states), adverse events (post‑thrombotic syndrome, heparin‑induced thrombocytopenia, VTE recurrence, major bleeding, clinically relevant non‑major bleeding, chronic‑thromboembolic pulmonary hypertension, long‑term chronic thromboembolic pulmonary hypertension, stroke, and post‑stroke) and death. Each model cycle was 2 weeks long; the company stated that this cycle length was used to accurately model the effects of initial heparin for those treatments that needed it (edoxaban and dabigatran etexilate), and to more accurately model the costs and utilities associated with various adverse events represented in the model (which often occur within a short period of time in clinical practice). The company conducted the analysis from the perspective of the NHS and personal social services, and discounted costs and health effects at an annual rate of 3.5%. The model had a lifetime time horizon (maximum 50 years) with 5 key time points (0 to 2 weeks; 2 weeks to 3 months; 4 to 6 months; 7 to 12 months, and 12 months onwards); this allowed the use of different transition probabilities over time in order to capture the change in risk of having an event.

3.22

Patients entered the model in the 'on treatment' health state where they had initial anticoagulation treatment. While having treatment, patients were at risk of having an adverse event and moving to the associated health state. In the adverse event health states of chronic‑thromboembolic pulmonary hypertension and stroke, patients experienced the event and accrued costs and utility values for 1 cycle only. After this, patients moved to post‑chronic thromboembolic pulmonary hypertension and post‑stroke health states. Patients could also only experience heparin‑induced thrombocytopenia, VTE recurrence, and bleeding for 1 cycle. After this, most patients moved back to the on‑treatment health state (that is, restarted their initial anticoagulation treatment), but a proportion of patients moved to the off‑treatment health state. The off‑treatment health state captured patients who had stopped anticoagulation treatment either as a result of an adverse event or because they had reached the end of a specific treatment duration (after 12 months in the base case). Patients could move to a death state at any point in the model.

3.23

Subgroup analyses varied the treatment length (from 12 months in the base case to 3 months, 6 months or lifelong [50 years]) or varied the initial VTE event (PE with or without DVT). The company did not include an analysis for people with active cancer, for whom warfarin would be unsuitable, because these patients were not included in studies of edoxaban.

3.24

The company modelled the interventions using the dosage described in the marketing authorisations. The company used the following sources to estimate data for the model: Hokusai‑VTE (VTE recurrence [time dependent], bleeding, adverse events, and VTE‑related mortality, for warfarin only; and heparin‑induced thrombocytopenia, stroke, probability of discontinuation after adverse event, and mortality as a result of an adverse event, for all treatments); network meta‑analysis (odds‑ratios of edoxaban, dabigatran etexilate, and rivaroxaban compared with warfarin for VTE recurrence and bleeding); published literature (risk of initial and long‑term chronic thromboembolic pulmonary hypertension, probability of death as a result of VTE recurrence, transition probabilities while off‑treatment, death as a result of heparin‑induced thrombocytopenia, and risk of post‑thrombotic syndrome).

3.25

The company conducted a systematic literature review (and further additional targeted searches) to identify sources for utility values for people with VTE. It identified 6 sources that were used to derive utility data for the model. Other than for the initial VTE event, it did not use the Hokusai‑VTE trial quality‑of‑life data, because it stated that the sample size was too small and therefore the utility values were not reliable. At entry into the model, for the first cycle only, all patients had a utility value that reflected the disutility of the initial VTE (derived from Hokusai‑VTE data, using the European population only). For all subsequent cycles, all patients in all treatment groups were assigned age‑dependent baseline utility values from the general population without illness. When patients experienced adverse events in the model, the company applied a health‑state‑related utility decrement that was deducted from the baseline utility value. For the health states 'heparin‑induced thrombocytopenia', 'VTE recurrence', 'bleeding', 'chronic‑thromboembolic pulmonary hypertension' and 'stroke' disutility was applied for 1 cycle only; post‑thrombotic syndrome, post‑stroke and long‑term chronic thromboembolic pulmonary hypertension were assumed to accrue the disutility for the lifetime of the model. A utility decrement was also applied to all patients having treatment with warfarin, to capture disutility associated with warfarin treatment such as frequent INR monitoring. No treatment‑related disutility was assumed for other treatments. The company adjusted the modelled utility values over time to reflect increasing age, with most decrements taken from a population aged 55 to 64 years.

3.26

The company used drug costs from the BNF (edition 68; 2014) and costs of hospitalisation from NHS reference costs 2013/14 in its model. It also conducted a systematic review of the literature for other costs such as those for treating stroke. For all treatments, the costs associated with the first cycle of treatment (that is, treatment for the initial VTE) were calculated independently of the costs of the subsequent cycles; the first cycle included the drug and administration costs associated with low‑molecular‑weight heparin (for patients who had warfarin, edoxaban and dabigatran etexilate). In the health states of heparin‑induced thrombocytopenia, VTE recurrence, bleeding, chronic‑thromboembolic pulmonary hypertension, and stroke, patients could accrue costs for 1 cycle only. For warfarin, the company included monitoring costs of £24.95 per cycle (a weighted average cost for INR testing of £27 from NHS reference costs, and a monitoring frequency of 0.9 per cycle from NICE's technology appraisal guidance on rivaroxaban for PE and recurrent VTE, which assumed 9 visits in the first 3 months, then 5 visits each quarter). No monitoring costs were assumed for any other treatment.

Company's base‑case results and sensitivity analysis

3.27

The company's base-case results for edoxaban (based on a model updated with corrections advised by the ERG, including costs and utilities) were presented incrementally, and also as edoxaban compared with each comparator, and each comparator compared with warfarin. In the incremental analysis, edoxaban, and all other comparator treatments, were dominated by rivaroxaban (that is, all treatments were more expensive and less effective than rivaroxaban). Edoxaban had an incremental cost‑effectiveness ratio (ICER) of £2,451 per quality‑adjusted life year (QALY) gained (incremental costs £45.37, incremental QALYs 0.0185) compared with warfarin, and was dominant (that is, more effective and less costly) compared with dabigatran etexilate.

Company scenarios

One‑way sensitivity analyses
3.28

The company conducted 1-way deterministic sensitivity analyses using upper/lower 95% confidence intervals for transition probabilities for warfarin, probabilities of complications while on warfarin or newer oral anticoagulant treatment, probability of death, hazard ratio for VTE recurrence compared with warfarin, and utility values and utility decrements. It also varied costs by plus or minus 20%. Compared with warfarin, the company stated that most ICERs for edoxaban were similar to its base case of £2,451 per QALY gained. The ICER increased to around £22,500 per QALY gained when it used high values for the probability of chronic thromboembolic pulmonary hypertension (between 3 to 12 months) or stroke with newer oral anticoagulants, and a low value for the probability of stroke with warfarin. When the company used the low value for the probability of chronic thromboembolic pulmonary hypertension between 3 months and 12 months with warfarin, the ICER for edoxaban increased to £10,377 per QALY gained. Compared with dabigatran etexilate, the scenarios that had the largest impact on ICERs were the higher value for the odds ratio of VTE recurrence for edoxaban compared with warfarin within 3 months (£180,870 per QALY gained) and the lower value of the same odds ratio of VTE recurrence for dabigatran etexilate compared with warfarin (£45,755 per QALY gained). However, the company stated that most scenarios had a limited impact on the ICERs. Rivaroxaban dominated edoxaban in all scenarios.

Probabilistic sensitivity analyses
3.29

The company conducted probabilistic sensitivity analyses using 2,000 simulations. Key parameters (including event rates, costs, risks, utility values and population characteristics) were varied simultaneously by sampling from various probability distributions. Compared with warfarin, edoxaban was dominant in 42% of simulations and dominated in 10% of simulations; it was more effective and more costly in 46% of simulations. Compared with rivaroxaban, edoxaban was dominated in 86% of simulations and more effective and more costly in 14% of simulations. Compared with dabigatran etexilate, edoxaban was dominant in 69% of simulations and less effective and less costly in 31% of simulations. The probability of edoxaban being cost effective at a maximum acceptable ICER of £20,000 per QALY gained compared with warfarin, rivaroxaban and dabigatran etexilate was approximately 70%, 8% and 75% respectively.

Evidence review group's comments on the company's cost‑effectiveness model

3.30

The ERG noted several concerns about the model structure, but it stated that most of these issues did not have a substantial impact on the cost‑effectiveness estimates. The ERG's concerns included:

  • If patients had a recurrent VTE, they returned to their original therapy. However, clinical experts advised the ERG that patients would switch to a higher dosage or change treatment because their original treatment had not prevented recurrence.

  • If shorter cycles are used (in this instance a 2‑week cycle was used) then the use of health states that have a fixed cycle duration needs to be carefully considered, because this may incorrectly estimate the impact of certain events on quality of life and costs. To overcome this, the company could have increased the length of the cycle, modelled the event without a fixed cycle duration, or used more post‑event states (as it had done for chronic thromboembolic pulmonary hypertension).

  • Patients were at risk of post‑thrombotic syndrome when on or off treatment, but only at risk of chronic thromboembolic pulmonary hypertension and stroke when on treatment. However, clinical experts advised the ERG that these are all VTE‑related complications. The ERG therefore stated that patients should be at risk of these outcomes at all times in the model.

  • The inclusion of the stroke health state was methodologically flawed, so the ERG conducted a scenario analysis without this. Flaws noted were that:

    • the company had not included several equally relevant cardiovascular events; for example, myocardial infarction

    • it included both intracerebral haemorrhages and ischaemic stroke, which meant that intracerebral haemorrhages (also included in the major bleeding health state) were double counted

    • clinical experts advised the ERG that the type of stroke would affect whether treatment would be stopped (patients would temporarily stop after an ischaemic stroke, but permanently stop after intracerebral haemorrhage).

  • In the major bleeding health state, the impact of intracerebral haemorrhage (which has a 40% mortality risk and a 60% disabling risk) on quality of life was underestimated.

3.31

The ERG considered the clinical‑effectiveness estimates used in the model. It noted that the odds ratios derived from the 6‑month meta‑analysis for VTE recurrence used in the model had not shown any statistically significant differences between treatments. To consider this, the ERG conducted a scenario analysis that assumed there were no differences between treatments in effectiveness for this outcome (that is, the odds ratios were set to '1'). The ERG also conducted an analysis setting the HR to '1' for bleeding.

3.32

The ERG raised several concerns with the disease‑specific mortality in the model. For example, the ERG noted that VTE recurrence data from the trial were used to model the initial event mortality, and not recurrence. For the recurrent‑VTE mortality rate, clinical experts advised the ERG that the rate used (13.7%) was overestimated, with the rate in practice being closer to 3%. The ERG concluded that the mortality estimates for both initial and recurrent VTE were not appropriate, and conducted 2 scenario analyses; 1 using trial data, and 1 using NICE's technology appraisal guidance on dabigatran etexilate for treating and secondary prevention of DVT and/or PE estimate of 3% that was more consistent with expert opinion. For stroke, the ERG noted that the health state included both intracerebral haemorrhage and ischaemic stroke, but the mortality was derived from ischaemic stroke only. Clinical experts advised the ERG that intracerebral haemorrhage has a mortality of around 40%, therefore the modelled mortality rate for stroke (3.85%) was an underestimate.

3.33

The ERG made some revisions to the quality‑of‑life estimates used in the model (see section 3.38).

3.34

The ERG considered monitoring costs, stating that they appeared to be overestimated for warfarin and underestimated for the other oral anticoagulants. For warfarin, the ERG stated that in clinical practice most visits for international normalised ratio (INR) monitoring occur within the first 3 months, but the company had used an average number of visits per cycle in the model, which did not capture this. Clinical experts also advised the ERG that after 3 months the monitoring schedule would be closer to 3 visits per quarter, not 5; after 12 months it would be approximately 10 visits per year. Follow‑up visits would likely be delivered by nurses. To consider the impact of this, the ERG conducted a scenario analysis to reduce the number of visits needed in the first year (for the base case) to 3 visits per quarter after the initial 3 months, and 10 visits per year after the first year when assuming lifelong treatment, with follow‑up visits based on non‑consultant‑led anticoagulation clinic attendance. For edoxaban, dabigatran etexilate and rivaroxaban, the ERG stated that in clinical practice there would be some monitoring costs expected beyond those considered standard monitoring for patients who have experienced VTE. The ERG conducted 3 scenario analyses to reflect different clinical opinions about monitoring of newer anticoagulants: a scenario in which patients have an annual visit where they have urea and electrolyte tests; a scenario in which patients have an annual appointment with their GP but they have no tests; and a scenario in which patients have an appointment twice a year where they have urea and electrolyte tests.

3.35

The ERG considered the 1‑way sensitivity analysis presented by the company, stating that it had concerns about its transparency, relevance and robustness. It stated that the company had not justified its choice of parameters and had presented only the 8 most influential parameters. The ERG stated that the analysis was rendered largely irrelevant because the company included clinically implausible scenarios. For example, the 3 inputs identified by the company as the key model drivers for edoxaban compared with warfarin (probability of stroke with warfarin, and probability of stroke and chronic thromboembolic pulmonary hypertension with the other oral anticoagulants) were based on the assumption that the probability of these events might be different depending on the type of treatment, which did not have face validity. For utility values and decrements, the company had estimated 95% confidence intervals as the high and low values. However, the ERG stated that the company had not reported how it obtained the standard deviations in the calculation. For costs, the company had varied overall health state costs by plus or minus 20%. However, the ERG noted that this meant it was not possible to identify specific costs or resource use within each health state. The ERG stated that the company should have explicitly varied the resource use associated with warfarin monitoring, because this was expected to be a key model parameter.

3.36

The ERG was concerned with the validity of the probabilistic results. It noted that the company had assigned a non‑parametric distribution to the odds ratios for newer anticoagulant treatment effectiveness (VTE recurrence and bleeding events) therefore they were not included in the probabilistic sensitivity analysis. The ERG stated that the exclusion of key clinical parameter estimates calculated in the network meta‑analysis considerably reduced the usefulness of the probabilistic results.

3.37

The ERG corrected technical errors in the company's model. After these corrections were applied the base case ICERs remained the same for edoxaban compared with warfarin, dabigatran etexilate and rivaroxaban. The ERG then provided scenario analyses based on this revised company base case. When compared with rivaroxaban, edoxaban was dominated in all scenarios. When compared with dabigatran etexilate, the key parameters were the odds ratio used to model the probability of VTE recurrence for dabigatran etexilate compared with warfarin (£9,678 per QALY gained) and the data used to model recurrent VTE‑related mortality (when using Hokusai‑VTE phase‑specific data the ICER was £15,111 per QALY gained and when using NICE's technology appraisal guidance on dabigatran etexilate for treating and secondary prevention of DVT and/or PE the ICER was £28,116 per QALY gained). When compared with warfarin, the main parameters were INR monitoring (assuming 2 less visits per quarter in the final 3 quarters of the year, the ICERs were £18,953 per QALY gained in the scenario base case and £40,359 per QALY gained in lifelong duration scenario) and the level of newer oral anticoagulant monitoring assumed (£4,780 to £7,315 per QALY gained depending on the level assumed).

3.38

The ERG presented exploratory analyses that compared edoxaban with warfarin, using data from Hokusai‑VTE rather than from the company's network meta‑analysis. The ERG stated that it considered this approach and the associated ICERs to be the most robust because they were derived from direct comparisons in a randomised controlled trial. The ERG presented ICERs for each individual change to the model, and the cumulative effect of each change (table 4), with a final ICER of £26,028 per QALY gained after all of the following changes were made:

  • used VTE recurrence and bleeding odds ratios from Hokusai‑VTE

  • used phase‑specific data for bleeding events and VTE‑related mortality (not time‑to‑event data)

  • increased the duration of the decrement in quality of life after recurring VTE and MB

  • removed the disutility associated with non‑major bleeding

  • reduced the number of INR monitoring visits for warfarin from 24 visits in the first year to 18 visits, to reflect 3 instead of 5 visits per quarter (after the initial 9 visits in the first 3 months)

  • revised the warfarin cost (£0.04)

  • changed the duration of heparin treatment to reflect the Hokusai‑VTE trial (7.5 days for edoxaban and 8.5 days for warfarin)

  • assumed that 30% of recurrent DVT cases and 50% of recurrent PE cases need hospitalisation

  • applied initial and recurrent VTE diagnostic costs to 100% of patients, according to the type of VTE event: DVT, £143.23; PE, £307.23

  • used the estimate from NICE's guidance on venous thromboembolism for the long-term chronic thromboembolic pulmonary hypertension cost (£1,280 at 2014 prices), combined with anticoagulation treatment costs

  • removed the stroke health state from the model

  • assumed that patients having oral anticoagulants (except warfarin) have an annual appointment where they have urea and electrolyte tests.

Table 4 ERG's exploratory analyses

Analysis

Incremental

Cumulative

Company base case using trial odds ratios

£1,958

£1,958

Phase-specific data warfarin bleeding

  • Cost: £46

  • QALYs: 0.022

  • ICER: £2,063

ICER: £2,063

Phase-specific data recurrent VTE mortality

  • Cost: £34

  • QALYs: 0.014

  • ICER: £2,433

ICER: £2,551

Increased QALY decrement duration recurrent VTE and major bleed

  • Cost: £43

  • QALYs: 0.022

  • ICER: £1,972

ICER: £2,574

Removed disutility for clinically relevant non-major bleed

  • Cost: £43

  • QALYs: 0.022

  • ICER: £1,968

ICER: £2,593

Reduced warfarin monitoring (18 visits year 1)

  • Cost: £349

  • QALYs: 0.022

  • ICER: £15,739

ICER: £21,505

Warfarin costs £0.02 per day

  • Cost: £36

  • QALYs: 0.022

  • ICER: £1,632

ICER: £21,057

Heparin duration 7.5 days edoxaban, 8.5 warfarin (Hokusai)

  • Cost: £78

  • QALYs: 0.022

  • ICER: £3,522

ICER: £23,324

VTE hospitalisation: 30% DVT and 50% PE

  • Cost: £45

  • QALYs: 0.022

  • ICER: £2,031

ICER: £23,352

VTE diagnostics: 100% (DVT, £143.23; PE, £307.23)

  • Cost: £43

  • QALYs: 0.022

  • ICER: £1,960

ICER: £23,352

NICE guideline on venous thromboembolism estimate for long-term chronic thromboembolic pulmonary hypertension (£1,280 a month)

  • Cost: £44

  • QALYs: 0.022

  • ICER: £1,997

ICER: £23,389

Stroke health state removed

  • Cost: £42

  • QALYs: 0.02

  • ICER: £1,869

ICER: £23,251

Non-warfarin treatments: annual monitoring

  • Cost: £88

  • QALYs: 0.022

  • ICER: £3,990

ICER: £26,028

  • National Institute for Health and Care Excellence (NICE)