3 The manufacturer's submission
The Appraisal Committee (appendix A) considered evidence submitted by the manufacturer of apixaban and a review of this submission by the Evidence Review Group (ERG; appendix B).
3.1 The manufacturer's submission compared apixaban with enoxaparin, a low molecular weight heparin (LMWH), using direct evidence from randomised controlled trials, and with dabigatran etexilate, rivaroxaban and fondaparinux, using evidence from randomised controlled trials that had been incorporated into an adjusted indirect comparison and a mixed treatment comparison. Outcomes analysed included mortality, incidence of venous thromboembolism and adverse reactions to treatment. All of these were specified in the decision problem for this appraisal. The clinical evidence submitted by the manufacturer did not include analysis of outcomes such as joint infection, length of hospital stay, complications after deep vein thrombosis or health-related quality of life because these were not available from the clinical trials of apixaban. However, the manufacturer included complications after deep vein thrombosis and health-related quality of life in its economic submission.
3.2 The manufacturer identified four randomised controlled trials comparing apixaban with enoxaparin for the prevention of venous thromboembolism. ADVANCE 1 ('Apixaban dosed orally versus anticoagulation with enoxaparin'), ADVANCE 2 and APROPOS ('Apixaban prophylaxis in patients undergoing total knee replacement surgery') recruited patients having total knee replacement surgery. ADVANCE 3 recruited patients having total hip replacement surgery.
3.3 ADVANCE 1 (n = 3195) and ADVANCE 2 (n = 3057) were multicentre parallel-group randomised controlled trials. In ADVANCE 1, apixaban was given at a dosage of 2.5 mg twice daily for 12 days and enoxaparin at a dosage of 30 mg twice daily for 12 days. Treatment in both arms was started 12–24 hours after surgery. In ADVANCE 2, apixaban was given at a dosage of 2.5 mg twice daily for 11 days and enoxaparin at a dosage of 40 mg once daily for 11 days. Apixaban was started 12–24 hours after surgery and enoxaparin 9–15 hours before surgery. APROPOS (n = 305) was a dose-finding clinical study in which patients were randomised to receive one of several doses of apixaban (5 mg, 10 mg, 20 mg, once or twice daily), enoxaparin 30 mg twice daily or warfarin. ADVANCE 2 used the UK dosing regimen for enoxaparin (40 mg once daily) whereas ADVANCE 1 and APROPOS used the US dosing regimen (30 mg twice daily). The manufacturer considered ADVANCE 2 to be the most relevant study for the prevention of venous thromboembolism after total knee replacement surgery in the context of UK clinical practice because it was the only study that compared apixaban with the UK licensed dose of enoxaparin.
3.4 ADVANCE 3 (n = 5407) was a multicentre parallel-group randomised controlled trial comparing apixaban with enoxaparin for the prevention of venous thromboembolism after total hip replacement surgery. Apixaban was given at a dosage of 2.5 mg twice daily for 32–38 days and enoxaparin at a dosage of 40 mg once daily for 32–38 days. Apixaban was started 12–24 hours after surgery and enoxaparin 9–15 hours before surgery.
3.5 The primary efficacy end point for ADVANCE 1 and ADVANCE 2 was the composite of all incidences of venous thromboembolism (pulmonary embolism, symptomatic and asymptomatic deep vein thrombosis), and death from any cause during the intended treatment period. The primary safety end point was bleeding and included confirmed major bleeding events, a composite of confirmed major bleeding events and confirmed non-major bleeding events, and all bleeding events. The primary efficacy end point for ADVANCE 3 was the composite of symptomatic or asymptomatic deep vein thrombosis, non-fatal pulmonary embolism, and death from any cause during the intended treatment period. The primary safety end point was bleeding during treatment or within 2 days of the last dose of study medication. The primary efficacy end point for APROPOS was the composite of symptomatic or asymptomatic deep vein thrombosis, non-fatal pulmonary embolism and death from any cause. The primary safety end point was major bleeding.
3.6 The primary efficacy analysis dataset included all randomised patients who had a bilateral venogram that was evaluable, venous thromboembolism, or who died from any cause. The manufacturer stated that the intention-to-treat analysis assumed that no readable venogram represented no event, which potentially underestimated the number of venous thromboembolic events in the intention-to-treat population. The remaining efficacy and safety analyses were conducted on the intention-to-treat population.
3.7 ADVANCE 2 showed that apixaban was statistically significantly superior to enoxaparin in terms of the primary composite end point of all venous thromboembolism and death from any cause (relative risk [RR] 0.62, 95% confidence interval [CI] 0.51 to 0.74), as well as in terms of major venous thromboembolism (RR 0.5, 95% CI 0.26 to 0.97) and all deep vein thrombosis (RR 0.6, 95% CI 0.5 to 0.72). For major bleeding events and all bleeding events, the relative risks were 0.65 (95% CI 0.28 to 1.49) and 0.83 (95% CI 0.65 to 1.06) respectively. ADVANCE 1 and APROPOS both used the US dosing regimen for enoxaparin and neither reported significant differences for any of the outcomes reported.
3.8 ADVANCE 3 showed that apixaban was statistically significantly superior to enoxaparin in terms of the primary composite end point of all venous thromboembolism and death from any cause (RR 0.36, 95% CI 0.23 to 0.56), as well as in terms of major venous thromboembolism (RR 0.4, 95% CI 0.19 to 0.83) and all deep vein thrombosis (RR academic in confidence). For major bleeding events and all bleeding events the relative risks were 1.22 (95% CI 0.65 to 2.26) and 0.93 (95% CI 0.81 to 1.08) respectively.
3.9 In the absence of direct evidence comparing apixaban with dabigatran etexilate, rivaroxaban and fondaparinux, the manufacturer presented results of an adjusted indirect comparison using enoxaparin as the common comparator. The manufacturer did not include an assessment of apixaban compared with other LMWHs in the indirect comparison. It considered enoxaparin to be the most widely used LMWH in the UK. Moreover, enoxaparin was the comparator used in the apixaban registration trials (ADVANCE 2 and 3). The manufacturer identified 15 randomised controlled trials for inclusion in the indirect comparison. Of these, nine compared the treatment of interest with enoxaparin at the UK dosage of 40 mg once daily. The remaining six studies compared the treatment of interest with enoxaparin 30 mg twice daily.
3.10 The manufacturer reported adjusted indirect comparisons using pooled data from studies using the UK dosage of enoxaparin alone, US dosage of enoxaparin alone and combined UK and US dosages of enoxaparin as the common comparator. Results were expressed as odds ratios for apixaban versus the other treatments of interest. For the outcome of composite venous thromboembolic events, and other end points of any deep vein thrombosis, asymptomatic deep vein thrombosis and major venous thromboembolism, results from the primary efficacy population were reported. For symptomatic deep vein thrombosis, pulmonary embolism, any bleeding, major bleeding, clinically relevant non-major bleeding and minor bleeding, results were reported from the intention-to-treat population.
3.11 The manufacturer stated that it considered the adjusted indirect comparisons of apixaban 2.5 mg twice daily against other anticoagulants of interest using the UK dosage of enoxaparin to be the most relevant to UK clinical practice. The results of such analyses for hip replacement showed that there were no significant differences for apixaban when compared with rivaroxaban for total venous thromboembolism and death from any cause, any deep vein thrombosis, major venous thromboembolism, pulmonary embolism, any bleeding or major bleeding. These results were the same for total knee replacement, except for the number of pulmonary embolic events, which was statistically significantly reduced with rivaroxaban. When compared with dabigatran etexilate, total venous thromboembolism, death from any cause and any deep vein thrombosis were statistically significantly reduced with apixaban; there were no significant differences for the other main outcomes (major venous thromboembolism, pulmonary embolism, any bleeding and major bleeding). These results were the same for total hip replacement and total knee replacement. When compared with fondaparinux in patients with total hip replacement, apixaban showed no significant differences for any deep vein thrombosis, pulmonary embolism or major bleeding. Other main outcomes (total venous thromboembolism and death from any cause, major venous thromboembolism and any bleeding) were not reported using indirect comparisons. There were no randomised controlled trials comparing fondaparinux with the UK dosage (40 mg daily) of enoxaparin in patients having total knee replacement surgery, therefore an adjusted indirect comparison was not possible. The results of the adjusted indirect comparisons using data with US dosages of enoxaparin and pooled data with UK and US dosages of enoxaparin were considered to be academic in confidence by the manufacturer and therefore cannot be presented.
3.12 The manufacturer also undertook a mixed treatment comparison that included 43 studies. Results were expressed as odds ratios and were for the same outcomes as for the adjusted indirect comparisons. The manufacturer reported mixed treatment comparisons using only the UK dosage of enoxaparin and also using pooled data for UK and US dosages. The results of the mixed treatment comparisons cannot be reported because the manufacturer considers the results to be academic in confidence.
3.13 The manufacturer considered the adjusted indirect comparison using the UK dosage of enoxaparin to be the most appropriate analysis for informing the relative efficacy and safety of apixaban compared with enoxaparin, rivaroxaban, dabigatran etexilate and fondaparinux. This was because results from the mixed treatment comparison were inconsistent with some of the head-to-head data from the randomised controlled trials. The manufacturer noted that the inconsistent results and wider credibility intervals may be a result of the large number of trials contributing to the enoxaparin 40 mg once daily node in the network analysis. These trials tended to be older, with fewer study criteria reported and small sample sizes, and compared enoxaparin 40 mg once daily with treatments other than the comparators in the scope. The manufacturer highlighted inconsistencies between the results of the mixed treatment comparison and the head-to-head trial for the following comparisons: apixaban 2.5 mg twice daily compared with enoxaparin 40 mg once daily, and rivaroxaban 10 mg once daily compared with enoxaparin 40 mg once daily for the primary composite end point (venous thromboembolism plus death from any cause) and on some of the secondary outcomes.
3.14 The ERG considered that the three clinical trials, ADVANCE 1, 2 and 3, which represent the main clinical efficacy evidence, were of reasonable methodological quality and measured a range of outcomes that were appropriate and clinically relevant. It stated that processes and validation of study screening and data extraction appeared to be appropriate. The ERG agreed with the chosen doses for each treatment included in the adjusted indirect and mixed treatment comparisons. The ERG commented that the statistical methods were explicitly described for the meta-analyses and indirect comparisons, and all relevant analyses were performed. In addition, the ERG commented that the manufacturer's conclusion that the mixed treatment comparison was less reliable than the adjusted indirect comparison seemed reasonable.
3.15 The manufacturer submitted an economic model assessing the cost effectiveness of apixaban compared with enoxaparin, dabigatran etexilate and rivaroxaban. A two-stage modelling approach was adopted, based on an approach previously used for 'Dabigatran etexilate for the prevention of venous thromboembolism after hip or knee replacement surgery in adults' (NICE technology appraisal guidance 157). A decision tree was used to model treatment in the prophylactic phase (from the time of surgery to 90 days after surgery) and a Markov model was used to model long-term events (90 days after surgery and beyond). The differential effects of treatment are only realised in the prophylactic phase of the model. The Markov model has a cycle length of 1 year and a maximum time horizon of 60 years (base case 35 years).
3.16 In the decision tree model, a hypothetical patient can experience a venous thromboembolic event (total venous thromboembolic events or death from any cause) or no event. When death is not a result of venous thromboembolism, death can result from a major bleed or other cause. Deaths from other causes refer to deaths not related to venous thromboembolism and deaths not related to anticoagulation during the prophylactic phase. A venous thromboembolic event can be a pulmonary embolism, symptomatic or asymptomatic deep vein thrombosis (both either distal or proximal). Pulmonary embolism may or may not result in death. A cohort of hypothetical patients surviving pulmonary embolism and a cohort with symptomatic deep vein thrombosis receive treatment and progress to the non-fatal bleeding events state of the model. A cohort of hypothetical patients with asymptomatic deep vein thrombosis progress to the non-fatal bleeding events state without treatment. A cohort of hypothetical patients without venous thromboembolic events also progress directly to the non-fatal bleeding state. Probabilities of bleeding are independent of what happened earlier in the model. A cohort of hypothetical patients experiencing an intracranial haemorrhage proceed immediately to the 'disabled' health state and remain there for the duration of the model or until they die. Alternatively, hypothetical patients can experience no bleeding, minor bleeding, a non-major clinically relevant bleed or a major bleed (other than an intracranial haemorrhage). In the period between the end of treatment and 90 days after surgery, hypothetical patients without symptoms can become symptomatic. Asymptomatic deep vein thrombosis that becomes symptomatic after prophylactic treatment is assumed to be of the same type (distal thrombosis remains distal and proximal thrombosis remains proximal).
3.17 At 90 days after surgery a cohort of hypothetical patients leave the decision tree model and enter the long-term Markov model. Hypothetical patients who have not experienced a venous thromboembolic event enter the Markov model in the 'well' state, whereas hypothetical patients who have asymptomatic deep vein thrombosis enter the Markov model in the untreated venous thromboembolic state. Hypothetical patients who have had a pulmonary embolism or a deep vein thrombosis or have made the transition from asymptomatic to symptomatic deep vein thrombosis enter the Markov model in the treated venous thromboembolic state. Hypothetical patients who have had an intracranial haemorrhage enter in the 'disabled' health state. A cohort of hypothetical patients who died while in the decision tree model enter the Markov model in the 'dead' state. In the long-term Markov model, a cohort of hypothetical patients can remain well, die, have a pulmonary embolism, a deep vein thrombosis, mild-to-moderate post-thrombotic syndrome (divided into year 1 and subsequent years) or a severe post-thrombotic syndrome (divided into year 1 and subsequent years). The same transitions are possible for treated and untreated patients. Once a hypothetical patient has a pulmonary embolism or deep vein thrombosis they make the transition to the treated venous thromboembolic state. There is no differential treatment effect in this long-term phase of the model.
3.18 Key assumptions in the economic evaluation included the assumption that during the prophylactic phase, deaths from pulmonary embolism and other causes occur at 35 days for total hip replacement and 14 days for total knee replacement in each treatment arm. It was assumed that during the phase after prophylaxis, deaths from pulmonary embolism occur at 63 days for total hip replacement and 52 days for total knee replacement. These are the midpoints of the post-prophylactic phase for each indication. It was also assumed that deaths from major bleeds occur at 35 days for total hip replacement and 14 days for total knee replacement, regardless of whether the bleeding rates were based on the duration of prophylaxis or 90 days.
3.19 The manufacturer modelled the efficacy (total venous thromboembolic events and all deaths) and safety (total bleeds) of the treatments in line with the corresponding end points in the two clinical trials of apixaban (ADVANCE 2 and 3) and in the indirect comparison of apixaban with rivaroxaban and with dabigatran etexilate. The manufacturer stated that because data were not available for an indirect comparison of apixaban with fondaparinux in patients having total knee replacement surgery, apixaban could not be compared with fondaparinux in the economic model. The manufacturer also stated that relative risks were used in the economic model rather than odds ratios because they can be directly applied to an absolute probability of an event to generate the absolute event rate for the comparator treatment. The manufacturer's original economic model did not distinguish between types of bleed and types of venous thromboembolism for each comparator individually, but assumed that they were all the same. Since this assumption may favour apixaban, the ERG asked the manufacturer to provide an adjusted model that allowed the differences in type of bleed and type of venous thromboembolism. This adapted model was provided by the manufacturer.
3.20 The probabilities of other clinical events in the decision tree element of the model were assumed to be treatment independent and assumed not to differ between apixaban, enoxaparin, rivaroxaban and dabigatran etexilate. Where possible, the probabilities for the post-event treatment-independent probabilities were obtained from a synthesis of all the trials on rivaroxaban and dabigatran etexilate. The manufacturer undertook a literature review to identify parameter estimates of long-term recurrent venous thromboembolism and post-thrombotic syndrome.
3.21 The manufacturer presented drug acquisition costs for a course of treatment depending on the treatment durations assumed for each treatment. The treatment durations applied were: for apixaban, total knee replacement 12 days and total hip replacement 34 days (mean duration in ADVANCE 2 and 3 trials); for enoxaparin, total knee replacement 12 days and total hip replacement 34 days (mean duration in ADVANCE 2 and 3 trials); for rivaroxaban total knee replacement 12 days and total hip replacement 33 days (mean duration in RECORD 1 and 3 trials); for dabigatran etexilate ,and total knee replacement 8 days and total hip replacement 32 days (median duration in RE-MODEL and RE-NOVATE). The costs per course of treatment for total knee replacement were £48.48, £52.97, £33.60 and £41.16 for enoxaparin, rivaroxaban, dabigatran etexilate and apixaban respectively. For total hip replacement the costs per course of treatment were £137.36, £145.68, £134.40 and £116.62 for enoxaparin, rivaroxaban dabigatran etexilate and apixaban respectively.
3.22 The manufacturer presented base-case analyses for people having total hip replacement surgery and for those having total knee replacement surgery. In the base-case analyses a comparison was made between enoxaparin, apixaban, dabigatran etexilate and rivaroxaban. For both total hip replacement surgery and total knee replacement surgery, apixaban, dabigatran etexilate and rivaroxaban were less expensive than enoxaparin. The quality-adjusted life year (QALY) differences between the treatments were small. For people having total hip replacement surgery, total QALYs ranged from 9.520 for enoxaparin to 9.536 for rivaroxaban. For patients having total knee replacement surgery, total QALYs ranged from 9.023 for enoxaparin to 9.090 for rivaroxaban.
3.23 For patients having total hip replacement, apixaban, rivaroxaban and dabigatran etexilate all dominated enoxaparin, that is they were less expensive and provided more benefit than enoxaparin. Apixaban was the least expensive technology. Both apixaban and rivaroxaban were more effective and less costly, and thus dominant, compared with dabigatran etexilate and enoxaparin. Rivaroxaban generated more QALYs compared with apixaban. The incremental cost-effectiveness ratio (ICER) of rivaroxaban was £21,661 per QALY gained compared with apixaban. For people having total knee replacement surgery, apixaban was less expensive than dabigatran etexilate and enoxaparin, but more expensive than rivaroxaban. Apixaban was also more clinically effective than dabigatran etexilate and enoxaparin. Both apixaban and rivaroxaban dominated dabigatran etexilate and enoxaparin in the analyses for total knee replacement.
3.24 The manufacturer conducted deterministic one-way sensitivity analysis, which included an assessment of the impact of a 50% reduction in the cost of dabigatran etexilate. For patients having total hip replacement, apixaban remained dominant compared with enoxaparin and dabigatran etexilate for all changes in the sensitivity analysis. Rivaroxaban was cost effective compared with apixaban, except when the time horizon was 10 years or less, when the age at surgery was 80 years, or when a smaller relative difference in the risk of the primary end point (total venous thromboembolic events and death from any cause) was assumed. For patients having total knee replacement, apixaban remained dominant compared with enoxaparin and dabigatran etexilate for all changes in the deterministic sensitivity analysis, whereas rivaroxaban dominated apixaban for all changes.
3.25 The manufacturer also presented probabilistic sensitivity analysis, which showed that in total hip replacement apixaban had a 53% probability of being the most cost-effective drug at a maximum acceptable ICER of £20,000 per QALY gained and rivaroxaban had a probability of 47%. At a maximum acceptable ICER of £30,000 per QALY gained these probabilities were 47% and 53% respectively. For total knee replacement, at a maximum acceptable ICER of £20,000 per QALY gained, apixaban had an 11% probability of being the most cost-effective drug. For rivaroxaban this probability was 89%. At a maximum acceptable ICER of £30,000 per QALY gained, these probabilities were 10% and 90% respectively.
3.26 The ERG considered the modelling approach to be reasonable because it had followed previous economic models, including a previous submission to NICE for NICE technology appraisal guidance 157. The ERG considered that the health states modelled were appropriate. The ERG also considered it appropriate for enoxaparin to be restricted to 40 mg once daily, which is the dosage licensed in Europe. The ERG considered it reasonable that the manufacturer had used enoxaparin to represent LMWHs. The ERG noted that the model does not allow movement from mild-to-moderate post-thrombotic syndrome to severe post-thrombotic syndrome and that there are no bleeding events in the long-term Markov model.
3.27 The ERG noted that all parameter uncertainty was not reflected in the probabilistic sensitivity analyses. The ERG therefore commented that the probabilistic sensitivity analyses probably underestimate the total uncertainty.
3.28 Full details of all the evidence are in the manufacturer's submission and the ERG report.