3 The manufacturer's submission

The appraisal committee (appendix A) considered evidence submitted by the manufacturer of ivabradine and a review of this submission by the evidence review group (ERG; appendix B).

3.1

The manufacturer conducted a systematic literature search and identified only 1 randomised controlled trial that assessed the effect of ivabradine in people with heart failure, known as SHIFT (systolic heart failure treatment with the If inhibitor ivabradine trial). SHIFT was an international, multicentre, randomised, double-blind, placebo-controlled trial comparing ivabradine with placebo for the treatment of moderate to severe heart failure and left ventricular systolic dysfunction. The trial was carried out in 625 centres in 37 countries and lasted from 12 to 36 months in the active double-blind treatment period, extended to a maximum duration of 52 months. The clinical-effectiveness evidence presented in the manufacturer's submission was based on this trial alone, but results were also provided for the SHIFT patient-reported outcomes (SHIFT-PRO) study. SHIFT-PRO was carried out to evaluate the effects of ivabradine compared with placebo on health-related quality of life in a representative sample of the main trial population.

3.2

Patients with symptomatic heart failure with a left ventricular ejection fraction of 35% or lower who were in sinus rhythm with a heart rate of 70 bpm or more and were receiving stable background treatment for heart failure were considered eligible for participation in SHIFT. After screening, 6,505 patients were randomised to receive either ivabradine or placebo in addition to ongoing optimal therapy (standard care) for heart failure (as assessed by the investigator responsible for the patient). All patients received 5 mg of ivabradine or matching placebo twice daily at day 0. This dose was maintained, or increased to 7.5 mg twice daily or reduced to 2.5 mg twice daily depending on resting heart rate and tolerability. All analyses were based on intention to treat even though a total of 1,190 patients died, withdrew consent or were lost to follow-up.

3.3

The trial groups in SHIFT were well balanced in patient baseline characteristics. The mean age was 60.4 years, 76% of the patients were men and mostly white. Mean heart rate was 79.9 bpm and mean left ventricular ejection fraction was 29%. Heart failure was ischaemic in 68% of the patients and patients were equally distributed between NYHA class II, III or IV. Alcohol consumption and smoking status were also similar between the trial groups, with less than 20% of the patients being current smokers in both groups. The background therapies were also similar in both arms (ACE inhibitors or angiotensin receptor blockers: 91%; diuretics: 84%; beta-blockers: 89%; aldosterone antagonists: 61% and cardiac devices [implantable cardioverter defibrillators: 3% and cardiac resynchronisation therapy: 1%]).

3.4

Subgroups were predefined in terms of age, sex, beta-blocker intake at randomisation, primary cause of heart failure, NYHA class, presence of diabetes, presence of hypertension and heart rate above and below the median of 77 bpm. The manufacturer stated in its submission that another subgroup was identified after the committee for Medicinal Products for Human Use recommended identifying the heart rate threshold at which there is a statistically significant mortality benefit. This subgroup consisted of people with a baseline heart rate of 75 bpm or more (n=4,150) and was identified post hoc. Data from this subgroup were used to identify the population to be covered by the marketing authorisation. The manufacturer's economic model was also based on this post hoc subgroup. Other post hoc subgroups identified were based on age (75 years or older and 70 years or older).

3.5

The baseline characteristics of the subgroup with a baseline heart rate of 75 bpm or more (the population covered by the marketing authorisation) were similar to the main trial population. The mean age for this subgroup was 59.6 years and, like the main trial population, they were mostly men (77%) and mostly white. There were no baseline differences between the treatment groups in this population including mean heart rate (84.5 bpm) and distribution of NYHA class. The background therapies received were also similar to the main trial population for both treatment groups (ACE inhibitors or angiotensin receptor blockers: 90%; diuretics: 84%; beta-blockers: 88%; aldosterone antagonists: 62% and cardiac devices).

3.6

The primary outcome in the SHIFT main trial population was a composite endpoint of first event of cardiovascular death or hospital admission for worsening heart failure. This was carried out using a survival analysis based on time-to-first event estimated by the Kaplan-Meier method. Secondary and other efficacy outcomes included mortality, hospital admission, change in heart rate, change in NYHA class, change in global assessment of heart failure symptoms and efficacy in patients aged 70 years or older (post hoc analysis in the main trial population).

3.7

In the SHIFT-PRO study (n=5,038), which studied a subset of the main SHIFT population, health-related quality of life was estimated using the EuroQol-5 dimensions (EQ-5D) questionnaire and 'Kansas City cardiomyopathy questionnaire' (KCCQ). Analysis in this study was also performed according to the same predefined subgroups specified in the main trial population, with the exception of presence of diabetes and hypertension. An additional subgroup was specified according to whether or not patients had received at least half the target dose of beta-blockers at randomisation. The manufacturer's submission noted that there were no relevant differences in baseline demographics and disease characteristics among the main trial population, the population covered by the marketing authorisation and the population in the SHIFT-PRO study.

Main SHIFT population

3.8

In the main trial population, the primary outcome of first event of cardiovascular death or hospital admission for worsening heart failure was analysed using a Cox proportional hazards model adjusted for beta-blocker intake at randomisation. The hazard ratio (HR) estimate was 0.82; (95% confidence interval [CI] 0.75 to 0.90, p<0.0001), representing a statistically significant relative risk reduction of 18% for ivabradine compared with placebo. This composite endpoint was driven more by the rate of hospital admission for worsening heart failure (HR 0.74; 95% CI 0.66 to 0.83) than by the rate of cardiovascular death (HR 0.91; 95% CI 0.80 to 1.03) because people are often admitted to hospital before they die.

3.9

Further analysis was carried out by the manufacturer to assess the impact of baseline beta-blocker dose on the efficacy of ivabradine in the main SHIFT population. For the primary composite endpoint, the relative effects of ivabradine compared with placebo for the 5 categories of beta-blocker intake were:

  • HR 0.71; 95% CI 0.55 to 0.93, p=0.012 (no beta-blocker)

  • HR 0.74; 95% CI 0.59 to 0.92, p=0.007 (less than 25% of target dose)

  • HR 0.81; 95% CI 0.68 to 0.98, p=0.029 (25% or more but less than 50% of target dose)

  • HR 0.88; 95% CI 0.72 to 1.07, p=0.193 (50% or more but less than 100% of target dose) and

  • HR 0.99; 95% CI 0.79 to 1.24, p=0.913 (100% or more of target dose).

    There were similar trends in efficacy for ivabradine compared with placebo across the beta-blocker categories for the component outcomes of hospital admission for worsening heart failure and cardiovascular death. The manufacturer noted that this could be a result of lower doses of beta-blockers being associated with higher heart rate because beta-blockers primarily reduce heart rate. There were no statistically significant differences across the beta-blocker categories. These findings suggest that the efficacy of ivabradine is primarily driven by heart rate and not by beta-blocker dose.

Population covered by the marketing authorisation

3.10

In the subgroup with a baseline heart rate of 75 bpm or more, the incidence of the primary composite endpoint was statistically significantly lower in the ivabradine group than in the placebo group (26.6% and 32.8% respectively, p<0.0001). The hazard ratio showed a clinically and statistically significant reduction of 24% in the risk of the composite endpoint for ivabradine compared with placebo (HR 0.76; 95% CI 0.68 to 0.85). This was in line with the predefined subgroup analysis on median heart rate, which revealed that baseline heart rate modified the treatment effect of ivabradine.

3.11

There was a statistically significant improvement in all secondary outcomes for the population covered by the marketing authorisation, unlike for the main SHIFT population in whom some of the secondary outcomes were not statistically significant. There were statistically significant reductions in all mortality outcomes in the ivabradine group compared with placebo as follows:

  • cardiovascular death (HR 0.83; 95% CI 0.71 to 0.97, p=0.0166)

  • heart failure death (HR 0.61; 95% CI 0.46 to 0.81, p=0.0006)

  • all-cause death (HR 0.83; 95% CI 0.72 to 0.96, p=0.0109).

    Results similarly favoured ivabradine compared with placebo for:

  • hospital admission for cardiovascular problems (HR 0.79; 95% CI 0.71 to 0.88, p<0.0001)

  • worsening heart failure (HR 0.70; 95% CI 0.61 to 0.80, p<0.0001)

  • hospital admission for any cause (HR 0.82; 95% CI 0.75 to 0.90, p<0.0001).

3.12

In the population covered by the marketing authorisation, heart rate decreased in the ivabradine and placebo groups by 17.4 bpm and 5.7 bpm at day 28 and 14.5 bpm, and 5.8 bpm at the last visit respectively. The manufacturer noted that the greater decrease in heart rate in the population covered by the marketing authorisation was consistent with a higher mean baseline heart rate of 84 bpm in this subgroup compared with 80 bpm in the main trial population. This was confirmed to be in line with previous ivabradine trials, which showed that greater reductions in heart rate are associated with higher resting heart rate. In this subgroup there was a statistically significant improvement in NYHA class in the ivabradine group compared with the placebo group.

3.13

Using the SHIFT-PRO study data, 3 types of quality-of-life analyses were performed. The first (main analysis) used '0' as the last post-baseline value for deceased patients, the second (an analysis of surviving patients) used the last post-baseline value for deceased patients, and the third used the change from baseline to month 12 from the main analysis. For the EQ-5D index score measure, quality of life worsened from baseline to the last assessment in the ivabradine group and the placebo group in the main analysis. However, there was an improvement in quality of life from baseline to the last assessment for the analysis of surviving patients in the 2 groups, with a greater improvement in the ivabradine group. The quality of life improvement from baseline to month 12 in both groups was higher in the ivabradine group. The manufacturer suggested that this was because there were fewer deaths during the first 12 months than during the whole study.

3.14

A mixed regression model was used to estimate quality of life using EQ-5D index scores with UK population tariff values. This showed that quality of life improved in the ivabradine group for the population covered by the marketing authorisation. The KCCQ disease-specific measure was also used and it showed a statistically significant difference of 2.6 (95% CI 0.7 to 4.5, p=0.008) for ivabradine compared with placebo for the 12-month analysis, which was also similar to the main analysis and the analysis of surviving patients.

3.15

The safety population (n=6,492 main trial cohort; n=4,141 population covered by the marketing authorisation) was the population who received at least 1 dose of any study treatment. The adverse events that occurred on treatment (between the first study drug intake and last intake plus 2 days) were analysed in this safety population. The following adverse events occurred more frequently with ivabradine than with placebo in the population covered by the marketing authorisation: symptomatic bradycardia (4.1% and 0.7% respectively), atrial fibrillation (7.9% and 6.8% respectively) and phosphenes (2.8% and 0.5% respectively). There were similar results for the main trial population. However, other serious adverse events and fatal events were higher in the placebo group in the 2 populations. The manufacturer noted that the tolerability of ivabradine was not affected by baseline heart rate because there were no differences in the adverse events leading to withdrawal between the main trial population and the population covered by the marketing authorisation.

3.16

After a request from the ERG during the clarification stage, the manufacturer provided the absolute numbers for the primary composite outcome and key secondary outcomes for the subgroups of the population covered by the marketing authorisation according to their beta-blocker category, age and NYHA class (details of the analyses are in section 3.22). The manufacturer also provided separate scenario analyses of the impact of using a regression model for NYHA progression adjusted for patient baseline characteristics, using updated standard care drug costs and different assumptions for modelling mortality. In addition, the manufacturer provided details of the patients who experienced symptomatic bradycardia and atrial fibrillation, and follow-up data on the reduction in heart rate at various time points for the population covered by the marketing authorisation.

Evidence review group comments – population covered by the marketing authorisation

3.17

The ERG stated that the literature search conducted by the manufacturer was appropriate, all relevant studies had been identified and that SHIFT, on which the manufacturer's submission was based, was relevant to the decision problem in its analysis. The ERG was satisfied that SHIFT was a well-designed randomised controlled trial with a robust method of randomisation. However, it highlighted that only 12 patients (0.2%) in the study were recruited from the UK, but noted the manufacturer's comment about the difficulties gaining study approval in the UK. The ERG also stated that the low UK patient numbers may have resulted from the difficulty in identifying eligible patients if patients were attending heart failure centres and had good titration of beta-blocker therapy. It also noted that the population covered by the marketing authorisation was younger, included a higher proportion of men and patients with more severe heart failure than a typical UK heart failure patient population, but it recognised that the baseline characteristics of the population covered by the marketing authorisation were similar to those reported for other key heart failure studies. However, the ERG considered that the results of SHIFT were robust and generalisable to a UK population because there was evidence to suggest that the patients in the trial received standard treatments.

3.18

The ERG noted that the clinical-effectiveness evidence for ivabradine was based on a post hoc subgroup of patients with a resting heart rate of 75 bpm or more without prior stratification based on resting heart rate, but in line with ivabradine's marketing authorisation. Therefore it considered that the evidence presented should be interpreted with a level of caution because there is likely to be an imbalance between the groups in terms of heart rate and potential unknown confounders. However, the ERG acknowledged that the baseline characteristics were well balanced between the treatment groups in the main trial population and the population covered by the marketing authorisation.

3.19

The ERG was aware that only approximately 26% of the main trial population and the population covered by the marketing authorisation were each treated with the recommended target dose of beta-blocker, and 55.4% of the trial population covered by the marketing authorisation were treated with 50% or more of the recommended dose of beta-blocker despite the recommendations in the SHIFT protocol. It was concerned that the patients who were not treated with the target dose of beta-blocker may not have been optimally treated. The ERG also noted the low use of cardiac devices in SHIFT and considered that this could have resulted from the exclusion of patients with pacemakers from the trial.

3.20

The ERG noted that the greatest benefit of ivabradine compared with placebo was in reducing heart failure deaths (HR 0.61; 95% CI 0.46 to 0.81, p=0.0006), which supports the observation that the results were generally driven by the cause-specific endpoints of hospital admission for heart failure and heart failure deaths in both populations. The ERG noted that ivabradine was associated with an improvement in NYHA class in the population covered by the marketing authorisation at their last visit compared with their baseline classification and that it had little impact on the proportion of patients with worsening NYHA classification.

3.21

The ERG noted that treatment-related adverse events occurred more frequently in the ivabradine group (17.8%) than in the placebo group (8.3%) in the main trial population. It felt that this was likely to be the same for the population covered by the marketing authorisation because the most common adverse events were the same as in the main trial population. The ERG highlighted that the reported adverse events (apart from inadequate blood pressure control) were similar to those reported in the BEAUTIFUL trial (10,917 randomised patients), which assessed the effects of ivabradine plus standard care in patients with coronary artery disease and left ventricular systolic dysfunction.

3.22

The ERG carried out an exploratory analysis of the data provided by the manufacturer after the clarification request on the primary and secondary outcomes of the population covered by the marketing authorisation according to their beta-blocker dosage at randomisation (that is, no beta-blocker, less than 25% of target beta-blocker dose, 25% or more but less than 50% of target beta-blocker dose, 50% or more but less than 100% of target beta-blocker dose and 100% or more of target beta-blocker dose).The ERG highlighted that their exploratory analyses suggest that there is uncertainty around the benefit of ivabradine plus standard care for patients with a resting heart rate of 75 bpm or more and who are receiving at least 25% of beta-blockers. The ERG also explored the efficacy of ivabradine according to NYHA class and in patients aged 70 years or older. It noted that the analysis in the NYHA class IV subgroup was based on small numbers, creating uncertainty about the benefit of ivabradine observed in this subgroup. Because the input data used in the exploratory analyses were marked as academic-in-confidence by the manufacturer, the results have also been marked as confidential and so cannot be shown here. However, the ERG emphasised that these analyses are speculative and based on subgroups of subgroups and should be interpreted with caution.

Cost-effectiveness evidence

3.23

In a systematic review of the literature the manufacturer did not identify any study on the cost effectiveness of ivabradine for treating chronic heart failure. No cost-effectiveness data were presented for the main SHIFT population, and so the economic evaluation carried out by the manufacturer was based only on the post hoc subgroup of patients from SHIFT with a baseline heart rate of 75 bpm or more. The manufacturer stated that this subgroup reflected the marketing authorisation for ivabradine; that is, people with chronic heart failure NYHA class II to IV with systolic dysfunction, in sinus rhythm and whose heart rate is 75 bpm or more, who are being treated with ivabradine in combination with standard therapy including beta-blockers, or for whom beta-blockers are contraindicated or not tolerated.

3.24

The manufacturer developed a Markov cohort model consisting of 2 states (alive and dead). The difference in quality of life of patients was captured according to NYHA class in the 'alive' state of the model without modelling the NYHA classes as separate health states. The model has a lifetime time horizon consisting of monthly cycles, includes a half-cycle correction, and both costs and benefits were discounted at 3.5%. The analysis was performed from the perspective of the NHS and personal social services. Standard care was modelled in line with SHIFT because the use of heart failure medications in the trial was higher than current standard care treatment patterns in the UK. The regression equations for mortality, NYHA class distribution, hospital admission and quality of life used in the analysis were based on data from the entire SHIFT cohort rather than developing risk equations based solely on the population covered by the marketing authorisation. This was to avoid breaking randomisation and reducing the predictive power of the risk equations because of smaller sample size. However, the risk equations for mortality, hospital admission and quality of life were adjusted for baseline heart rate to predict estimates for the population covered by the marketing authorisation with a heart rate of 75 bpm or more.

3.25

The manufacturer estimated the risk of non-cardiovascular death based on age-adjusted and sex-adjusted UK national life table data from the Office for National Statistics rather than SHIFT data because it provided a larger, UK-specific data source. This risk was assumed to be the same across treatment groups and no treatment effect was modelled for this endpoint. The risk of cardiovascular mortality (both heart failure and other non-heart-failure cardiovascular death) for the within-trial period was estimated using a Gompertz parametric survival regression model based on the full SHIFT cohort in the base-case analysis. Survival models based on exponential and Weibull parametric distributions, and as Kaplan-Meier data were included as part of the sensitivity analyses. The cardiovascular mortality risk equation was estimated adjusting for a series of baseline patient characteristics (including age, sex, NYHA class, heart failure duration, body mass index, medical history, baseline use of heart failure medications) to generate different estimates of mortality. The Gompertz distribution was also used to extrapolate cardiovascular mortality beyond the trial period. Mortality was approximately 17% in the standard care group of SHIFT. Because of the uncertainty generated by using a small proportion to extrapolate mortality for the rest of the cohort, the manufacturer considered mortality data from an external data source (CARE-HF data; Cleland 2010) in the sensitivity analyses. The extrapolation assumed that 50% of the cohort would have died after 2,000 days (65 months).

3.26

The distribution of patients in each NYHA class over time was estimated from a generalised ordered regression (a proportional odds model) developed from SHIFT data. It predicted the distribution of NYHA class adjusting for treatment and time covariates but not patient baseline characteristics. By the third year the proportion of patients in class III and IV reduced from 40.2% to 36.9% in the ivabradine arm and from 44% to 40.6% in the standard care arm, whereas those in class II increased from 58.4% to 61.4% and from 54.9% to 58.1% in the ivabradine arm and standard care arm respectively. Because of the lack of any evidence to predict the distribution of patients by NYHA class beyond the trial period, the model assumed that the proportions remained fixed after the trial based on the last observation in the trial at 29 months (although the absolute numbers in each category were expected to vary according to the number of patients alive).

3.27

The rate of heart failure, cardiovascular and all-cause hospital admission per person month was estimated using a Poisson regression model based on the entire SHIFT cohort and converted into a monthly transition probability in the economic model. The hospital admission endpoints were modelled separately to capture the appropriate resource use for each admission type and to permit sensitivity analysis on the treatment effect of ivabradine. However, the base-case analysis was based on all-cause hospital admission. Admission to hospital after the trial was modelled to be equivalent to the within-trial period and assumed to occur at a constant rate throughout the model irrespective of the ageing population.

3.28

The treatment effect of ivabradine on cardiovascular mortality (including heart failure death) compared with placebo was estimated as a hazard ratio of 0.90 (95% CI 0.80 to 1.03) from the parametric model to the underlying mortality risk in the standard care group. It was assumed that the treatment effect of ivabradine continues after the trial and is equivalent to that seen in SHIFT. To support this assumption, the manufacturer highlighted that the heart-rate-lowering effect of ivabradine was shown to be maintained throughout SHIFT and also over a 7-year study period for ivabradine in patients with angina. The treatment effect of ivabradine on the rate of admissions to hospital was estimated using a rate ratio of 0.83 (95% CI 0.78 to 0.93) derived from the Poisson regression model. The treatment effect was modelled on all-cause admission because cardiovascular and heart failure admissions were assumed to be implicitly captured in all-cause admission and ivabradine was shown to have a statistically significant effect on all-cause admission in the main trial and populations covered by the marketing authorisation. The length of stay associated with hospital admission was estimated using external data based on expert clinical advice. In the base-case model, the average length of stay was varied according to diagnosis on hospital admission (heart failure: 7.57 days, other cardiovascular: 3.97 days and non-cardiovascular: 5.13 days) and was based on a weighted average of elective and non-elective NHS reference cost data.

3.29

The utility values used in the model were derived from the SHIFT-PRO study, in which health-related quality of life was captured with the EQ-5D questionnaire. EQ-5D index scores were calculated using UK population tariff values and then analysed using a mixed regression model. Quality of life was modelled to reflect patients' baseline characteristics, severity of the disease over time by NYHA class, rate of hospital admission (which includes serious adverse events) and treatment group. The resulting utility scores by NYHA class without any hospital admission ranged from 0.82 in class I to 0.46 in class IV. Decrease in quality of life because of hospital admission was estimated as decreases in utility of 0.07, 0.03, 0.08 and 0.21 for NYHA class I, II, III and IV respectively. The effect of ivabradine on quality of life was modelled and showed a small utility increase in the ivabradine group compared with the baseline estimates used for the placebo (standard care) group. Treatment-related adverse events were assumed not to have any measurable impact on quality of life and the manufacturer indicated that they had been captured by the treatment covariate in the regression model. Quality of life was assumed to remain the same for each NYHA class in the post-trial period and in the base case and the model estimates were not based on an ageing population. This implies that the utility values for the patients in later cycles were higher than they should be and this was assumed to have favoured ivabradine because additional survival time was associated with greater quality-adjusted life year (QALY) benefits. In the sensitivity analysis, quality of life was adjusted for the increasing age of the modelled cohort by resetting the baseline age for each cycle.

3.30

The average monthly cost of ivabradine (£42.10; excluding VAT) used in the model was estimated according to the proportion of patients who received 2.5 mg (7%) and either 5 mg or 7.5 mg (93%) in the SHIFT study. The 5 mg and 7.5 mg tablets cost £40.17 per 56-tablet pack (excluding VAT; BNF 63), and the price of the 2.5 mg dose was assumed to be half the price of the 5 mg tablet. Average monthly standard care costs (£9.54) were estimated according to the proportion of patients using each standard care medication in SHIFT. The unit costs of the standard care drugs used such as beta-blockers, ACE inhibitors, diuretics, aldosterone antagonists, angiotensin receptor blockers and cardiac glycosides were also taken from the BNF. The manufacturer assumed that there were no extra costs in administering ivabradine and the standard care drugs. However, additional costs were included for ivabradine therapy titration (1 specialist visit) and an electrocardiogram (ECG). This increased the total monthly cost in the ivabradine group from £52 to £202 for the first month.

3.31

The hospital admission costs used in the model were estimated using the NHS reference costs for heart failure admissions (general ward: £2,308 and cardiac ward: £3,295), cardiovascular admissions (general ward: £1,942 and cardiac ward: £1,730) and non-cardiovascular admissions (general ward: £2,644). It was assumed that there was an equal probability of being in a general ward or a cardiac ward. Serious adverse events were captured using these hospital admission endpoints, but non-serious adverse events were not included. The monthly cost of managing heart failure, including physician visits, outpatient procedures and diagnostic tests, was estimated to be £27 from British Heart Foundation statistics.

3.32

The base-case results of the economic analysis, which was based on the population covered by the marketing authorisation, was estimated by applying individual patient profiles from SHIFT to the risk equations sequentially, one at a time. It showed that the incremental costs and incremental QALYs gained from treating chronic heart failure with ivabradine plus standard care compared with standard care alone were £2,376 and 0.28 QALYs respectively. This gave an incremental cost-effectiveness ratio (ICER) of £8,498 per QALY gained.

3.33

The manufacturer highlighted that the deterministic, probabilistic and structural sensitivity analyses were performed using average covariate values in the regression equations to shorten analysis time and that this may have caused some loss in accuracy in the ICER estimates. The base-case ICER using this method was £7,743 per QALY gained. The one-way deterministic sensitivity analyses were performed on several model parameters using their 95% confidence intervals. The cost-effectiveness result was most sensitive to changes in cardiovascular mortality risk, with the resulting ICERs ranging from £5,655 to £40,638 per QALY gained. The base-case ICER also showed some sensitivity to changes in the rate of hospital admission (£6,384 to £10,424 per QALY gained) and treatment effect of ivabradine on quality of life (£6,283 to £9,253 per QALY gained). Changes in hospital length of stay and ivabradine treatment effect on NYHA class had much less impact on the ICER, £6,938 to £8,549 and £7,232 to £8,349 per QALY gained respectively.

3.34

The manufacturer's probabilistic sensitivity analysis indicated that ivabradine plus standard care would have a more than 95% chance of being cost effective compared with standard care alone if the maximum acceptable ICER was £20,000 per QALY gained.

3.35

The manufacturer carried out different scenario analyses to manage uncertainties about some of the assumptions in the base-case model. The scenario analyses explored the effect on the ICER of: varying the treatment duration of ivabradine; ivabradine's treatment effect stopping after 5 and 10 years; using alternative models to estimate the risk of cardiovascular mortality; increasing the median length of hospital stay based on the 'National heart failure audit' data; and excluding the costs of the titration visit and the ECG. The manufacturer also carried out other scenario analyses, including: using a within-trial time horizon; using external data to extrapolate cardiovascular mortality and utility values; including age-adjusted utility values; and assuming a 5% change in the distribution of NYHA classes (from I to II, from II to III and from III to IV) in the post-trial period. After a clarification request, the manufacturer also provided a scenario analysis in which a new regression equation was developed to predict NYHA class distribution. This was adjusted for treatment, time covariates and patient baseline characteristics, and drug prices were updated to those in BNF 63. These scenario analyses all gave ICERs below £9,000 per QALY gained except for the assumptions of the treatment effect of ivabradine stopping after 5 and 10 years and using the within-trial time horizon, which gave ICERs ranging from £13,964 to £15,200 per QALY gained.

3.36

The manufacturer carried out several subgroup analyses based on individual patient characteristics from the population covered by the marketing authorisation. These subgroups were based on age, NYHA class, beta-blocker doses, heart failure duration, level of left ventricular ejection fraction, and prior medical history (coronary artery disease and diabetes). The results showed that ivabradine plus standard care was still cost effective when compared with standard care alone. The estimated ICERs for the subgroups were all below £11,000 per QALY gained and ranged from £5,197 to £10,427 per QALY gained. The manufacturer also carried out additional subgroup analyses based on a population representative of a UK chronic heart failure patient group. This population was specified as western European men with a median age of 78 years, receiving at least half the target dose of beta-blockers. The ICER generated for this subgroup was £8,735 per QALY gained, and the ICER for a UK chronic heart failure patient group taking the target dose of beta-blockers was £9,185 per QALY gained.

Evidence review group comments

3.37

The ERG was satisfied with the manufacturer's modelling approach, which was transparent, used patient-level data and was consistent with other published economic studies on heart failure treatments. The ERG stated that the manufacturer did not carry out an analysis in a patient population with a disease severity reflective of the UK population. However, it agreed with the manufacturer that using values for patient characteristics beyond the SHIFT population range may generate unreliable results. The ERG was satisfied that the standard care treatments used in SHIFT and the economic model reflected UK clinical practice.

3.38

The ERG accepted the manufacturer's use of Office for National Statistics UK life tables to provide estimates of non-cardiovascular mortality in the base case because this is standard practice in heart failure cost-effectiveness analyses. However, it noted that the risk of non-cardiovascular mortality was higher in SHIFT than in the UK life tables. The ERG noted that there were some uncertainties associated with the regression analyses performed for cardiovascular and heart failure mortality, which limited the potential of ivabradine to reduce the risks of these 2 outcomes. The treatment effect of ivabradine in the regression analysis was not statistically significant for cardiovascular mortality (p=0.38) and was borderline statistically significant (p=0.06) for heart failure mortality (although these results had been statistically significant for the population covered by the marketing authorisation only). By contrast, beta-blockers given at 50% or more of the target dose were associated with a statistically significant reduction in the risk of cardiovascular mortality for ivabradine compared with placebo and beta-blockers at any dose were associated with a statistically significant reduction in the risk of heart failure mortality for ivabradine compared with placebo. Because baseline heart rate was adjusted for in the regression analysis, the ERG thought that the risk reduction of ivabradine and beta-blockers was in addition to the attenuating effect of heart rate.

3.39

The ERG indicated that the regression model for health-related quality of life in the manufacturer's submission was clinically plausible and the disutility associated with hospital admission was likely to have captured any serious impact of adverse events on quality of life because hospital admission would be the main consequence of serious adverse events. The ERG noted that the impact of age adjustment for health-related quality of life was minimal (it increased the ICER by £216 per QALY gained). Therefore, it accepted the exclusion of age adjustment from the base-case analysis because of the time needed to re-run each cycle to adjust for age throughout the model's time horizon. The ERG was satisfied with the costing approach taken by the manufacturer in the economic analysis.

3.40

The ERG considered that the manufacturer's base-case ICER of £8,498 per QALY gained (incremental costs of £2,376 and incremental QALYs of 0.28) was likely to represent the expected cost effectiveness of adding ivabradine to standard care, although the ERG believed it was biased against ivabradine. The ERG was satisfied with the manufacturer's pragmatic approach of conducting the sensitivity analyses using average patient characteristics because of the longer analysis time needed to use individual patient profiles for the base case. It indicated that the reduced level of accuracy with this method was unlikely to alter any conclusions drawn from the evidence presented. The ERG was particularly interested in the cost-effectiveness results for the subgroups of patients at different doses of beta-blockers. It noted that the ICERs for these subgroups and all other subgroups analysed remained below £11,000 per QALY gained. However, the ERG noted that the regression equations used were based on the main trial population of SHIFT or the population covered by the marketing authorisation, rather than the particular subgroups of patients considered. It accepted that breaking randomisation and smaller patient numbers would compromise any analyses based on regression equations developed from subgroups. The ERG highlighted that the hazard ratios estimated from regression equations based on the main trial population of SHIFT or the population covered by the marketing authorisation may over (or under) estimate the effect of ivabradine treatment in particular patient populations.

3.41

Overall, the ERG considered the modelled results to be conservative because they underestimated the risk of cardiovascular mortality, the rate of hospital admission and the relative effect of treatment with ivabradine plus standard care compared with standard care alone. It stated that the sensitivity and subgroup analyses sufficiently addressed any areas of uncertainty.

3.42

Full details of all the evidence are in the manufacturer's submission and the ERG report.