3.1 The company submission presented clinical effectiveness data from 2 trials: AMPLIFY and AMPLIFY‑EXT. AMPLIFY was a randomised, active‑controlled, parallel‑group, double‑blind, triple‑dummy study carried out in 28 countries including 14 in Europe (but not the UK). The aim of AMPLIFY was to determine if apixaban was non‑inferior to the low molecular weight heparin (LMWH) enoxaparin followed by a vitamin K antagonist (VKA; in this case warfarin) for the composite end point of confirmed recurrent symptomatic non‑fatal venous thromboembolism (VTE) or VTE‑related death over 6 months of therapy. The criteria to demonstrate non‑inferiority were an upper boundary of the 95% confidence interval surrounding the relative risk of less than 1.8 and surrounding the risk difference of less than 3.5%. Patients were randomised 1:1 to apixaban (n=2691) or enoxaparin/warfarin (n=2704). Apixaban was dosed at 10 mg twice a day for 7 days followed by 5 mg twice a day for the remainder of the study. Patients in the enoxaparin/warfarin arm had 1 mg/kg subcutaneous enoxaparin twice a day for at least 5 days and warfarin to achieve an international normalised ratio (INR) of between 2.0 and 3.0: enoxaparin was stopped when the target INR was achieved. The median duration of enoxaparin treatment was 6.5 days (interquartile range 5.0 to 8.0). Patients were treated for 6 months and were followed‑up for 30 days after they stopped treatment.
3.2 The mean age of patients in AMPLIFY was 57 years and 58% were men. Most patients in the study (65%) had been randomised following a deep vein thrombosis (DVT), 25% had a pulmonary embolism (PE) and 9% had both a DVT and PE (qualifying diagnosis for entry into the study could not be evaluated in the other patients). Around 90% of patients had a VTE that was considered to be unprovoked. Sixty six (2.5%) patients in the apixaban arm and 77 (2.9%) patients in the enoxaparin/warfarin arm had active cancer.
3.3 In AMPLIFY, the efficacy population was defined as the intention‑to‑treat population for whom the outcome status at 6 months was documented (this comprised 2609 patients in the apixaban arm and 2635 patients in the enoxaparin/warfarin arm). There were 59 patients (2.3%) in the apixaban arm and 71 patients (2.7%) in the enoxaparin/warfarin arm who had a recurrent VTE or died because of a VTE (relative risk 0.84, 95% confidence interval [CI] 0.60 to 1.18, p<0.001 for non‑inferiority).
3.4 In AMPLIFY, fewer people had a major bleed in the apixaban arm than in the enoxaparin/warfarin arm (15 [0.6%] compared with 49 [1.8%]; relative risk 0.31, 95% CI 0.17 to 0.55; p<0.001). Three patients (0.1%) in the apixaban arm and 6 (0.2%) in the enoxaparin/warfarin arm had an intracranial bleed, and 7 patients (0.3%) in the apixaban arm and 18 (0.7%) in the enoxaparin/warfarin arm had a major gastrointestinal bleed. One patient in the apixaban arm and 2 in the enoxaparin/warfarin arm died because of their bleed. The company noted that the rates of major bleeds across the prespecified subgroups were consistent with the full population. Fewer people had a clinically relevant non‑major bleed with apixaban (103 [3.8%]) than with enoxaparin/warfarin (215 [8.0%] relative risk 0.48, 95% CI 0.38 to 0.60). In terms of events, 67.1% of patients in the apixaban arm and 71.5% in the enoxaparin/warfarin arm had an adverse event, and similar proportions of patients in both arms had a serious adverse event (15.6% and 15.2% respectively).
3.5 AMPLIFY‑EXT was a randomised, parallel‑group, double‑blind, placebo‑controlled study carried out in 28 countries including 7 centres in the UK. The aim of the study was to determine if 2.5 mg or 5 mg twice‑daily apixaban was superior to placebo for the composite end point of symptomatic recurrent non‑fatal VTE or all‑cause death in people who had a proximal symptomatic DVT or symptomatic PE, and who had completed 6–12 months of anticoagulant therapy for this index event. The study included patients for whom there was uncertainty about the need for continued anticoagulation treatment (termed 'clinical equipoise'); patients who definitely needed further anticoagulation were excluded from the study. Patients were randomised 1:1:1 to 2.5 mg apixaban twice daily (n=840), 5 mg apixaban twice daily (n=813) or placebo (n=829). Treatment was given for 12 months and patients were followed‑up for 30 days after they stopped treatment. The company presented only the results for the 2.5 mg dose of apixaban compared with placebo, because it is the licensed dose if anticoagulation with apixaban is continued beyond 6 months (see section 2.1).
3.6 The mean age of the patients in AMPLIFY‑EXT was also 57 years and 57% of the population were male. Qualifying diagnosis for inclusion in the study was DVT in 65% and PE in 35%. In most patients (92%) the VTE was considered to be unprovoked. Fifteen patients (1.8%) in the 2.5 mg twice‑daily apixaban arm and 18 patients (2.2%) in the placebo arm had active cancer.
3.7 In AMPLIFY‑EXT, all efficacy outcomes were analysed in the intention‑to‑treat population. Patients lost to follow‑up were counted as having had a primary outcome event. There were 13 patients in the 2.5 mg twice‑daily apixaban arm (1.5%) and 19 in the placebo arm (2.3%) who were lost to follow‑up. The results showed that 32 patients (3.8%) in the 2.5 mg twice‑daily apixaban arm and 96 patients (11.6%) in the placebo arm had recurrent VTE or died by 12 months (relative risk 0.33, 95% CI 0.22 to 0.48; p<0.0001).
3.8 In AMPLIFY‑EXT similar proportions of patients in the 2.5 mg twice‑daily apixaban arm (71.0%) and the placebo arm (73.4%) had an adverse event, although a higher proportion of patients in the placebo arm (19.1%) had a serious adverse event than in the 2.5 mg twice‑daily apixaban arm (13.3%). DVT was classed as an adverse event. Approximately 3% of patients in both the 2.5 mg twice‑daily apixaban and placebo arms had major or clinically relevant non‑major bleeding. There was no statistically significant difference between the 2 study arms because the confidence interval around the calculated relative risk crossed 1.
3.9 The company did not identify any head‑to‑head trials comparing apixaban with rivaroxaban or dabigatran etexilate for the treatment or secondary prevention of VTE. It therefore carried out 2 network meta‑analyses. The first (NMA 1) included trials which assessed anticoagulant therapy for the treatment of an initial VTE event, and the second (NMA 2) included trials which assessed extended anticoagulant therapy in patients who had already had treatment for a VTE event and were having continued anticoagulant treatment for secondary prevention.
3.10 NMA 1 was carried out to estimate the relative treatment effect and safety of apixaban compared with rivaroxaban and dabigatran etexilate for treating an initial VTE event. It included the following trials:
AMPLIFY: comparing apixaban (10 mg twice daily for 7 days followed by 5 mg twice daily) with LMWH (enoxaparin)/warfarin. The intention‑to‑treat dataset was used for efficacy analyses and the on‑treatment population was used for safety analyses.
RE‑COVER and RE‑COVER II: 2 trials, identical in design, comparing unfractionated heparin (UFH) or LMWH/dabigatran etexilate with UFH or LMWH/warfarin. A modified intention‑to‑treat dataset was used for efficacy analyses in which patients who did not have any study drug were excluded. The on‑treatment population was used for safety analyses.
EINSTEIN DVT and EINSTEIN PE: 2 trials that differed by the index event of the trial population (DVT or PE). Both trials compared rivaroxaban (15 mg twice daily for 21 days followed by 20 mg once daily) with LMWH/vitamin K antagonist (either warfarin or acenocoumarol). The company used a pooled data set from these trials. The intention‑to‑treat dataset was used for efficacy analyses and the on‑treatment population was used for safety analyses.
The company noted that most trials in the network used a modified intention‑to‑treat analysis (patients from that population who had no outcome data were excluded from the analysis). The company presented results using a fixed‑effects model, because there were few studies in the network.
3.11 There were no differences in the number of recurrent VTE or VTE‑related deaths with apixaban compared with LMWH/VKA LMWH/dabigatran etexilate or rivaroxaban. There were lower rates of bleeding (the composite outcome of major or clinically relevant non major bleeding, major bleeding assessed separately and clinically relevant non‑major bleeding) with apixaban compared with LMWH/VKA, LMWH/dabigatran etexilate and rivaroxaban. In response to clarification, the company re‑ran the meta‑analysis using different statistical modelling assumptions as requested by the Evidence Review Group (ERG). These results resulted in marginal differences to the company's base‑case results. The company carried out 3 further sensitivity analyses in which it used a modified intention‑to‑treat population, used pooled results from RE‑COVER and RE‑COVER II (rather than using the results from each trial separately) and excluded the dabigatran etexilate trials from the meta‑analysis. These sensitivity analyses also showed only a marginal effect. The company has stated that the exact results of NMA1 are academic in confidence and so are not reported here.
3.12 NMA 2 was carried out to compare apixaban with rivaroxaban or dabigatran etexilate for secondary prevention of VTE. It included the following trials:
AMPLIFY‑EXT: comparing 2.5 mg apixaban twice daily with placebo for 12 months after initial treatment of 6–12 months.
EINSTEIN‑EXT: comparing 20 mg rivaroxaban once daily with placebo over 6–12 months after an initial treatment of 6–12 months.
RE‑SONATE: comparing 150 mg dabigatran etexilate twice daily with placebo over 6 months after an initial treatment of 6–18 months.
RE‑MEDY: comparing 150 mg dabigatran etexilate twice daily with warfarin (INR 2.0–3.0) over 6–36 months following an initial treatment of 3–12 months.
LAFIT and PREVENT: trials comparing warfarin with placebo over 24 months (LAFIT) or a mean of 2.1 years (PREVENT) after an initial treatment of 3 months. In LAFIT the target INR for people taking warfarin was 2.0 to 3.0; in PREVENT it was 1.5 to 1.9.
WODIT DVT, WODIT PE: comparing VKA continuation with VKA discontinuation 3–9 months after an initial 3‑month treatment.
WARFASA, ASPIRE: comparing 100 mg aspirin once daily over 2–4 years after an initial treatment of 6 weeks to 18 months.
3.13 The company noted that the network of studies included a mixture of open‑label and double‑blind studies, as well as differences in the proportions of patients who had an unprovoked VTE rather than a VTE which could be attributed to a specific cause. Other differences between the studies were the proportion of patients with active cancer, treatment duration with anticoagulants before entering the trials and study follow‑up. The company also noted that there may have been differences in clinical judgement regarding the need for continuation of anticoagulation across the trials, and the patients in the trials may have had different baseline characteristics. The company tested for heterogeneity of the studies included in the network meta‑analysis using the I2 statistic and found little evidence for heterogeneity. The meta‑analysis was done using a random‑effects model.
3.14 There were no differences between apixaban, LMWH/dabigatran etexilate, rivaroxaban or LMWH/VKA in the rates of recurrent VTE or VTE‑related death because the 95% credible intervals crossed 1. Apixaban was associated with fewer recurrent VTE or VTE‑related deaths than aspirin or placebo. Apixaban was associated with statistically significantly fewer major or clinically relevant non‑major bleeding events (assessed as a composite outcome) than the comparators. When major bleeding and clinically relevant non‑major bleeding were assessed as separate outcomes, patients taking apixaban had statistically significantly fewer bleeds of either severity than those having LMWH/VKA or rivaroxaban, but the rates were not statistically significantly different between apixaban and LMWH/dabigatran etexilate. In response to clarification, the company re‑ran the meta‑analysis using different statistical modelling assumptions as requested by the ERG. The results of these analyses were broadly consistent with the company's base case but the likelihood of a major bleed was no longer statistically significantly lower with apixaban than with rivaroxaban. The company did 3 further sensitivity analyses: using the intention‑to‑treat population from the trials, excluding the WODIT DVT/PE trials and excluding the dabigatran etexilate trials. These sensitivity analyses gave similar results to the company's base case, although the sensitivity analysis using the intention‑to‑treat population resulted in statistically significantly fewer recurrent VTE events or VTE‑related deaths with apixaban than LMWH/VKA. The company has stated that the full results of NMA 2 are academic in confidence and so are not reported here.
3.15 The ERG noted that the patients in both AMPLIFY and AMPLIFY‑EXT were younger than those seen in clinical practice and that relatively few people were older than 75 years (14.3% in AMPLIFY and 13.3% in AMPLIFY‑EXT). The ERG stated that a UK cohort study had found that the mean age of people having an unprovoked PE was 64 years and that 47% were men. The ERG further commented that the proportion of men in the UK cohort study was smaller than the proportion of men in AMPLIFY and AMPLIFY‑EXT (58% and 56% respectively).
3.16 The ERG discussed whether the population in AMPLIFY and AMPLIFY‑EXT was representative of people who would have apixaban for secondary prevention of VTE. The ERG commented that AMPLIFY‑EXT included only patients for whom there was uncertainty about the need for continued anticoagulation treatment (termed 'clinical equipoise'). People who definitely needed extended anticoagulation were not included. The ERG noted that in its submission the company had not discussed the extent to which the results of the AMPLIFY‑EXT trial are directly applicable to people who definitely need anticoagulation beyond 6 months. The ERG also noted that there were no clinical data for people who had a provoked index event but were not considered to be at risk of a recurrent event, because these people were not included in the trial.
3.17 The ERG commented that the company had stated that apixaban is not licensed for people with active cancer. It further noted that patients with active cancer for whom treatment with LMWH was planned were excluded from AMPLIFY, and that they were unlikely to meet the clinical equipoise criteria for AMPLIFY-EXT because patients with active cancer are treated with LMWH for extended periods of time.
3.18 The ERG considered that the characteristics of the trials included in NMA 1 were similar enough that combining the results was appropriate. However, it noted that the trials of rivaroxaban compared with LMWH/VKA included a higher proportion of people with PE (58%) than the other trials in the network of evidence (which ranged from 21.2% to 25.2%).
3.19 The ERG was concerned that there were substantial differences between the time spent on treatment (from 6 months in RE‑SONATE to 37.2 months in ASPIRE) and the follow‑up periods of trials (from 10 months in LAFIT to 37.2 months in ASPIRE) in NMA 2 (which included trials that had assessed anticoagulants for secondary prevention). The ERG did not consider it appropriate to combine data from these trials and did not agree with the company's assertion that the different treatment periods and follow‑up times would not have a substantial effect on the results. The ERG stated that there are likely to be more events in studies with longer treatment periods and follow‑up times. It concluded that because of this, the prevention network meta‑analysis (NMA 2) was not appropriate and only direct clinical evidence for apixaban from AMPLIFY-EXT could be used to assess the clinical effectiveness of apixaban for secondary prevention of VTE.
3.20 The ERG noted that the company had provided continuity correction factors for outcomes in which there were no events in 1 of the study arms in the trials. This had resulted in high estimates of relative risk for some outcomes, such as the relative risk of major bleeding with apixaban compared with rivaroxaban. However, the ERG was satisfied that the company's analyses provided in response to clarification resulted in less extreme estimates of the underlying treatment effect (that is, they were less likely to over‑ or underestimate the treatment effect).
3.21 The company developed a new Markov model with a 3‑month cycle length and lifelong time horizon (patients were assumed to live to a maximum age of 100 years). Patients entered the model following a PE or a DVT. In the model it was assumed that patients could have a recurrent DVT/PE, have a bleed (an intracranial bleed, a non‑intracranial major bleed or a clinically relevant non major bleed), discontinue treatment or die (either because of a recurrent DVT/PE, bleed or other reasons). Patients with an intracranial bleed were assumed to discontinue treatment permanently. The company assumed that approximately half of the patients who survived a non‑intracranial major bleed would discontinue treatment permanently. The others would have a 2‑week treatment break before resuming treatment (the company has stated that the precise proportion is academic in confidence and is not reported here). Patients who had a clinically relevant non‑major bleed were assumed to have a 2‑day treatment break then resume treatment. Patients who had a PE could develop chronic thromboembolic pulmonary hypertension (CTEPH), and patients who had a DVT could develop post‑thrombotic syndrome (PTS). The company only modelled the cost and effect on quality of life of severe PTS; it stated that mild and moderate PTS had little effect on a patient's utility or resource use.
3.22 In the model it was assumed that all patients with LMWH had enoxaparin and all patients with a VKA had warfarin. The risks of having a recurrent DVT/PE or bleeding for patients having enoxaparin/warfarin or apixaban in the first 6 months were derived from the absolute risks of these events in AMPLIFY; for those having apixaban or no treatment after this period, risks were derived from AMPLIFY‑EXT. The company used the estimates from the 2 network meta‑analyses of the risks of DVT/PE or bleeding relative to apixaban for rivaroxaban, dabigatran etexilate and aspirin, which it then applied in its model. The company noted that the clinical trial evidence showed that the risk of recurrent DVT/PE decreased over time. The company ran the model for 2 treatment durations: 6 months' treatment followed by no treatment for the rest of a patient's life, and lifelong treatment. In the 6‑month treatment analysis, the risks of recurrent DVT/PE for patients having no treatment after 6 months were derived from the rates in a prospective cohort of 1626 patients over 10 years. In the lifelong treatment duration analysis, the risks of recurrent VTE for patients having treatment were derived from AMPLIFY‑EXT (for 6–18 months after index event) and from the prospective cohort study (for 18 months to 10 years after index event). In the model, the risks of major bleeding were derived from the AMPLIFY trials and the network meta‑analyses. In the base case the risks of bleeding were unadjusted for aging. The company said that this was a conservative assumption because bleeding risks were lower for apixaban compared with the other comparators and the risk of bleeding would be expected to increase as the age of the modelled cohort increased. In the base case it was also assumed that 13.46% of major bleeds would be fatal and that of the remaining non‑fatal bleeds, 13.97% would be intracranial. These assumptions were based on a published pooled analysis of randomised trials in which patients had anticoagulant treatment for at least 6 months and were assumed to be consistent across different types of anticoagulant. The model assumed a constant risk of CTEPH and PTS over time and that the risk of these complications would be the same irrespective of treatment. Patients who had an intracranial bleed or CTEPH were assumed to have a higher mortality rate (hazard ratio [HR] 2.6, 95% CI 2.2 to 3.0 for intracranial bleed; HR 1.3, 95% CI 0.98 to 1.73 for CTEPH).
3.23 No quality of life data were collected in the AMPLIFY or AMPLIFY‑EXT trials and the company used utility values from published studies identified through a systematic review in its model. The baseline utility for the model population (0.825) was based on a UK population‑average score from Kind et al. (1999). Utility decrements associated with PE (0.32) and DVT (0.11) came from a study by Locadia et al. (2004). Patients with an intracranial bleed were assumed to have a utility value of 0.33 while they had acute care (for 91 days), after which their utility was assumed to increase to 0.61 during post‑acute care. A major non‑intracranial bleed was associated with a utility decrement of 0.30, meaning that a single patient's utility value in this case would be 0.5224. Clinically relevant non‑major bleeds were assumed to have a utility decrement of 0.0054. Patients who had CTEPH were assumed to have a utility value of 0.65; patients with PTS were assumed to have a utility decrement of 0.07. Taking enoxaparin/warfarin was associated with a utility decrement of 0.013. Taking apixaban, rivaroxaban, enoxaparin/dabigatran etexilate or aspirin was assumed to have a utility decrement of 0.002.
3.24 The company used the NHS list prices for apixaban, rivaroxaban, dabigatran etexilate and enoxaparin. It modelled the cost of enoxaparin and noted that the recommended dose for enoxaparin in the UK is 1.5 mg/kg, but the dose of enoxaparin in AMPLIFY was 1.0 mg/kg. The company used the cost of the lower UK dose in its model, but the efficacy estimates were based on the higher dose. For patients having enoxaparin it was assumed that 92% would self‑inject following a 1‑time training cost of £17. It was assumed that for the other 8% of patients, enoxaparin would be administered by a nurse at a cost of £8.78.
3.25 For patients having warfarin it was assumed that 6 monitoring visits would be needed in the first 3 months followed by 3 visits every 3 months thereafter. It was assumed that 66.45% of INR monitoring visits would be carried out in primary care and 33.55% would be in secondary care. It was further assumed that half the first INR monitoring visits conducted in primary care would be delivered by a GP with the remainder delivered by a nurse. The resulting annual cost of monitoring was £319.19 in the first year of the model and £252.52 in subsequent years.
3.26 It was assumed in the model that 69% of patients who had a DVT and 17% of patients who had a PE would be treated as outpatients. Longer‑term monitoring and post‑acute care was assumed to be done in primary care, whereas treating bleeds and CTEPH was assumed to be carried out in secondary care.
3.27 The company presented deterministic pairwise and fully incremental results for 2 treatment durations: treatment over 6 months and lifelong treatment. If taken for 6 months the incremental cost effectiveness ratio (ICER) for apixaban compared with enoxaparin/warfarin was £2406 per quality adjusted life year (QALY) gained. Apixaban dominated (that is, was less costly and more effective than) rivaroxaban and enoxaparin/dabigatran etexilate. If taken lifelong, the ICER for apixaban compared with enoxaparin/warfarin was £16,676 per QALY gained. Rivaroxaban was dominated by enoxaparin/warfarin and enoxaparin/dabigatran etexilate was extendedly dominated by enoxaparin/warfarin and apixaban (a treatment is 'extendedly dominated' when its ICER is higher than that of the next, more effective, option – in this case apixaban – when compared with a common baseline). The company did not present probabilistic ICERs.
3.28 The company carried out 1‑way sensitivity analyses and scenario analyses. For 6 months' treatment, sensitivity analyses showed that the ICER for apixaban compared with enoxaparin/warfarin ranged from £1628 to £5330 per QALY gained. The highest of these was a result of decreasing the baseline utility value to 0.385. For lifelong treatment, the ICER for apixaban compared with enoxaparin/warfarin ranged from £2157 to £41,394 per QALY gained. Three sensitivity analyses resulted in an ICER above £30,000 per QALY gained; these were reducing the relative risk of major bleeding for enoxaparin/warfarin to be approximately the same as that for apixaban, setting the risk of bleeding to 0 for all treatments and reducing the baseline utility to 0.385. The company also tested the effect of over 30 scenarios. For 6 months' treatment, apixaban dominated rivaroxaban and enoxaparin/dabigatran etexilate in all scenarios. The ICER for apixaban compared with enoxaparin/warfarin was under £5000 per QALY gained in all scenarios. The scenarios that had the greatest effect on the ICER were assuming fewer warfarin monitoring costs and excluding the costs of enoxaparin. For lifelong treatment, assuming fewer warfarin monitoring visits (4 visits on initiation, 1 visit subsequent) increased the ICER for apixaban compared with enoxaparin/warfarin from £16,676 per QALY gained in the base case to £21,301 per QALY gained. The only other scenarios that increased the ICER for apixaban compared with enoxaparin/warfarin to over £20,000 per QALY gained were: assuming that the utility decrements for all treatments were the same as that assumed for warfarin (that is a utility decrement of −0.0013, which resulted in an ICER of £25,999 per QALY gained), and assuming an alternative distribution of fatal major bleeds and non‑fatal intracranial bleeds for patients who had a major bleed (which resulted in an ICER of £24,038 per QALY gained).
3.29 The ERG noted that the model used age‑specific mortality rates for all‑cause death but did not adjust the model parameters (such as the risk of VTE event or bleeding) by age or sex. The ERG further commented that the age‑specific mortality rates did not take into account that the ratio of men to women in the model cohort would be expected to change over time, because women tend to live longer than men. The ERG stated that the company's approach may have overestimated the mortality rates of the modelled cohort by up to 4% per year. The ERG further considered that the company's assumption surrounding baseline utility was flawed because the model did not account for the mean utility value of the modelled cohort decreasing as the age of the cohort increased over time.
3.30 The ERG commented that for lifelong treatment, the efficacy estimates of apixaban over the first 6 months were based on AMPLIFY; after this, data from AMPLIFY‑EXT were used. The ERG noted that the characteristics of the populations included in these 2 trials differed, and only a third of patients from AMPLIFY had then taken part in AMPLIFY‑EXT. The ERG noted that AMPLIFY‑EXT excluded patients who had a recurrent VTE event during earlier treatment of their index VTE event, and so at 6 months the characteristics of the modelled cohort effectively changes. The ERG suggested that 2 distinct decision models should have been developed, each based exclusively on a single trial: short‑term use of apixaban compared with comparators using AMPLIFY data and long‑term use of apixaban compared with no‑treatment using AMPLIFY‑EXT data.
3.31 The ERG commented that in the model, the cost of anti‑thromboembolic therapies for each 3‑month cycle was based on the average number of patients alive and on treatment over the course of the cycle. The ERG considered that this may underestimate the true costs, because oral medications prescribed at the start of a 3‑month treatment cycle could not be returned if they were not used.
3.32 The ERG commented that the cost of enoxaparin treatment in the model was based on a daily dose of 1.5 mg/kg, assuming a mean body weight of 84.6 kg (based on the mean weight of patients in AMPLIFY). However, the ERG considered this to be considerably higher than the mean adult weight of 77.4 kg reported in the Health Survey for England 2012.
3.33 The ERG commented on the company's approach to discounting. It noted that the company had assumed a 3.5% discount rate, which is consistent with the NICE reference case. However, the ERG noted that the company applied discounting at a different rate for every 3‑month model cycle based on the time elapsed rather than using a more conventional approach of applying the discount every 4 cycles (that is, yearly) after the first year.
3.34 The ERG noted that the model structure (in which patients who have a non‑fatal VTE without a permanent adverse event return to the index DVT or PE health state after 3 months) led to an implicit assumption that the risk of a second or third recurrent VTE was the same as that of a first recurrent VTE. The ERG stated that there was no evidence to support this assumption and that a published study had suggested that the risk of a second recurrent VTE relative to a first recurrence of VTE was about 2‑fold (relative risk 2.1, p=0.02). The ERG stated that as a consequence the long‑term estimates of future VTE events (including deaths) were likely to be underestimated, meaning that the costs and disutility value of events would also be underestimated.
3.35 The ERG stated that there were differences in the proportion of patients who at 90 days had discontinued treatment in the model compared with the AMPLIFY trial results. The ERG further commented that it was unable to validate other model parameters against the trial data to determine whether a similar error had been made across the whole range of time‑to‑event model variables, because the Kaplan–Meier data it requested during the clarification process had not been provided.
3.36 The ERG carried out a number of exploratory analyses including the following:
Age and sex modelling: to assess the effect over time of the changing age and sex distribution of the modelled cohort on survival.
Treatment costs: the treatment costs were calculated using the full number of patients who began treatment at the start of each 3‑month cycle.
Age‑stratified utility values: incorporating the baseline utility values by 10‑year age band (under 25, 25–34 up to 65–74, and 75+) from the Measurement and Valuation of Health survey.
Applying the discount yearly rather than applying the discount per cycle as had been done by the company in its base case.
Body weight: assuming a mean adult body weight of 77.4 kg when calculating amount of LMWH administered to achieve a 1.5 mg/kg dose.
Rebase prevention model (lifelong duration): to address its concerns that the modelled assumption in the first 6 months of lifelong treatment (based on the AMPLIFY population) did not reflect the experience of those patients on whom lifelong treatment estimates were based (the population from AMPLIFY‑EXT), the ERG excluded the first 2 cycles from the model to determine lifelong treatment results. The ERG noted that this would reflect the third of patients who had 6 months' treatment in the AMPLIFY trial before joining AMPLIFY‑EXT.
Hazard ratios requested by ERG: using the results from NMA 2 which incorporated the changes to the meta‑analysis modelling as requested by the ERG during clarification (these included using an alternative vague prior for the trial effect and treating the trial effect as random rather than fixed).
Poisson hazard ratios: using Poisson distributions in the model for NMA 2, with and without Bayesian assumptions. The ERG carried out these analyses because the trials in NMA 2 have different follow‑up lengths. Using a Poisson assumption in the model relates the incidence of events to the length of time that patients are exposed to the risk of event, and so it limits the potential bias of different follow‑up times in the meta‑analysis.
3.37 Overall, the exploratory analyses had a small effect on the company's base‑case ICERs in the 6‑month treatment analyses. The ICER for apixaban compared with enoxaparin/warfarin remained under £3000 per QALY gained and apixaban dominated rivaroxaban and enoxaparin/dabigatran etexilate in all analyses. In the lifelong treatment analyses, most of the ERG's exploratory analyses had only a small effect on the company's base‑case ICERs. Only using data from the network meta‑analyses that incorporated Poisson assumptions increased the ICER for apixaban compared with enoxaparin/warfarin to over £20,000 per QALY gained.
3.38 The ERG carried out a further scenario analysis in which it assumed that the efficacy and bleeding risks of apixaban, rivaroxaban and dabigatran etexilate were the same over the secondary prevention period. In this scenario, the ERG also applied its preferred assumptions on age/sex modelling and utility values, treatment costs, discounting method and body weight (see section 3.36). In this scenario the ICER for apixaban compared with rivaroxaban increased from £809 per QALY gained to £21,798 per QALY gained. The ICER for apixaban compared with enoxaparin/dabigatran etexilate increased from £5058 to £9139 per QALY gained.
3.39 Full details of all the evidence are in the committee papers.