3 Evidence

The appraisal committee (section 7) considered evidence submitted by Novartis and a review of this submission by the evidence review group (ERG; section 8).

Clinical effectiveness

3.1 The pivotal clinical evidence presented in the company's submission was taken from the PARADIGM‑HF trial, which compared sacubitril valsartan with enalapril (an angiotensin‑converting enzyme [ACE] inhibitor). The company also did a network meta‑analysis to compare sacubitril valsartan with angiotensin II receptor‑blockers (ARBs) for people who cannot have an ACE inhibitor. Finally, the company provided supplementary evidence in its submission from the TITRATION trial (a multicentre, randomised, double‑blind, parallel group, phase II study in clinically stable outpatients or hospitalised patients), which evaluated the safety and tolerability of sacubitril valsartan at increasing doses.

3.2 PARADIGM‑HF was a randomised, double‑blind, controlled, phase III trial comparing sacubitril valsartan (n=4,187) with enalapril (n=4,212). Both treatments were given in combination with standard care (including beta blockers and aldosterone antagonists). The trial included people with symptomatic heart failure – that is, New York Heart Association (NYHA) class II to IV – with left ventricular ejection fraction (LVEF) of 35% or lower. The company chose enalapril as a comparator in the trial because it is the ACE inhibitor that has been studied in the largest number of trials in this population.

3.3 The trial comprised 4 stages:

  • Screening for inclusion and exclusion criteria. Eligible patients were on a stable dose of an ACE inhibitor or an ARB equivalent to enalapril 10 mg per day for 4 weeks or more before screening visit.

  • Enalapril run-in (2 weeks): eligible patients were switched from current medication (ACE inhibitor or ARB) to single‑blind treatment with enalapril (10 mg twice daily).

  • Sacubitril valsartan run-in (4 to 6 weeks): patients were eligible if they had no unacceptable side effects in the previous stage. Eligible patients were switched to single‑blind treatment with sacubitril valsartan at a dose of 100 mg twice daily, which was increased to 200 mg twice daily during the run‑in stage. The 2 run‑in stages were sequential, with only a brief (around 36 hours) washout period, and both included all eligible patients.

  • Main trial: patients with no unacceptable side effects after taking target doses of the 2 study medications were randomly assigned (1:1) to double‑blinded treatment with either sacubitril valsartan (200 mg twice daily) or enalapril (10 mg twice daily).

3.4 Although the inclusion criteria specified people with NYHA class II to IV, some people had an improvement in their NYHA class between screening and randomisation, so nearly 5% of randomised patients were NYHA class I. The LVEF entry criterion was initially 40% or lower but was subsequently reduced to 35% or lower (961 patients were randomised who had LVEF greater than 35%) in order to ensure an adequate event rate in the study population. Raised plasma B‑type natriuretic peptide (BNP) level of at least 150 pg per ml (or N‑terminal pro‑brain natriuretic peptide [NT‑proBNP] of at least 600 pg per ml) at screening was also an entry criterion. If patients had been hospitalised for heart failure in the past 12 months, then a slightly lower cut‑off for BNP level (100 pg per ml) or an NT‑proBNP level (400 pg per ml) was accepted.

3.5 The company stated that at baseline, most characteristics were balanced between the treatment groups, including age, geographic region, NYHA class, standard care or background therapies received, and medical histories. It also commented that patients in the trial were younger (only 49% were 65 years or older) and more likely to be men (just 22% were women) than the general population seen in clinical practice in England. The company reported standard care and background therapies to be comparable to those in clinical practice in England; in the trial, at baseline, 93% of patients had beta blockers and 56% had aldosterone antagonists. About 78% had previously had an ACE inhibitor, and 23% had previously had an ARB. About 30% of people in the trial had been diagnosed with heart failure within the last year, 38% between 1 and 5 years previously, and 32% more than 5 years previously.

3.6 Results were presented based on the full analysis set, which consisted of all patients except those who did not meet the eligibility criteria or did not have a single dose of the study drug. These data were used for the efficacy outcomes (n=8,399). The primary end point was a composite of death from cardiovascular causes or a first hospitalisation for worsening heart failure, assessed at every study visit (0, 2, 4 and 8 weeks, 4 months, and then every 4 months). The composite primary end point significantly favoured sacubitril valsartan compared with enalapril (hazard ratio [HR] 0.80; 95% confidence interval [CI] 0.73 to 0.87, p<0.001).

3.7 The secondary outcomes included:

  • all-cause mortality (assessed at all study visits)

  • change from baseline to 8 months in the clinical summary score on the Kansas City Cardiomyopathy Questionnaire (KCCQ); patient scores were assessed at baseline/randomisation visit (visit 5), at 4, 8 and 12 months (visits 8, 9 and 10), at 24 and 36 months (visits 14 and 17), and at the end of study visit.

    Sacubitril valsartan compared with enalapril showed a significantly reduced risk for all‑cause mortality (HR 0.84; 95% CI 0.76 to 0.93, p<0.001), first all‑cause hospitalisation (HR 0.88; 95% CI 0.82 to 0.94, p<0.0001), and first cardiovascular hospitalisation (HR 0.88; 95% CI 0.81 to 0.95, p<0.0008). The KCCQ patient scores were reduced for both sacubitril valsartan and enalapril, but this reduction was less with sacubitril valsartan (by 2.99 points) than with enalapril (by 4.63 points).

3.8 Patients were stratified at randomisation by a number of factors, which included region, NYHA class, systolic blood pressure, LVEF, prior ACE inhibitors, prior ARBs, prior aldosterone antagonists, and prior hospitalisation for heart failure. Sacubitril valsartan treatment reduced the risk of the primary composite end point when compared with enalapril, independent of all predefined subgroups, although not all were statistically significant.

3.9 The company stated that age, gender, and NYHA class were important factors because the baseline characteristics of patients in the trial were different from those seen in clinical practice in England. The primary composite outcome was statistically significant in favour of sacubitril valsartan compared with enalapril across all subgroups, except in people aged 75 years and older (HR 0.86, 95% CI 0.72 to 1.04), and people with NYHA class III or IV heart failure (HR 0.92, 95% CI 0.79 to 1.08).

3.10 For the subgroups based on region, the primary composite outcome was statistically significant in favour of sacubitril valsartan compared with enalapril across all regions, except for the Western European subgroup (HR 0.89, 95% CI 0.74 to 1.07) and the Asia/Pacific and Other subgroup (HR 0.85, 95% CI 0.69 to 1.04). In the subgroup of patients who had not previously had an ACE inhibitor (n=1,867), sacubitril valsartan showed an improvement in the primary composite outcome of death from cardiovascular causes or a first hospitalisation for worsening heart failure, but this was not statistically significant (HR 0.92, 95% CI 0.76 to 1.10).

3.11 The NICE scope specified the comparator for people who cannot have an ACE inhibitor to be an ARB in combination with standard care. Because there is no head-to-head evidence comparing sacubitril valsartan with ARBs, the company conducted a network meta‑analysis to inform the economic model with estimates of the effectiveness of sacubitril valsartan compared with ARBs, as well as the effectiveness of ARBs compared with ACE inhibitors.

3.12 The network meta-analysis was based on data from 28 randomised controlled trials and provided comparative evidence for all‑cause mortality (28 trials, 4 treatment comparisons), cardiovascular mortality (13 trials, 4 treatment comparisons) and all-cause hospitalisations (28 trials, 4 treatment comparisons). The company commented that the network meta‑analysis reflected the approach taken by Heran et al. (2012) in a Cochrane review, which assessed ACE inhibitors against ARBs with regard to morbidity and mortality irrespective of concomitant treatment with standard care therapies.

3.13 The company's network meta‑analysis categorised treatment by class (angiotensin receptor neprilysin inhibitor [ARNI; sacubitril valsartan], ACE inhibitors, ARBs and placebo), assuming equal efficacy across all treatments in each class. To justify the class‑effect assumption of ACE inhibitors, the company referenced a systematic review and network meta‑analysis by Chatterjee et al. (2013) which found that 'there is currently no statistical evidence in support of the superiority of any single agent over the others'. The company cited the Cochrane review by Heran et al. (2012) to support the assumption of a class effect for ARBs.

3.14 The company used a Bayesian framework for its network meta-analysis. The Bayesian network meta‑analysis random effects model outcomes included all‑cause mortality, cardiovascular mortality and all‑cause hospitalisations. The results of the network meta‑analysis presented by the company were designated academic in confidence and cannot be reported here. However, the results demonstrated that:

  • ARBs and ACE inhibitors were broadly equivalent

  • sacubitril valsartan was superior to ARBs with regard to all‑cause and cardiovascular mortality and broadly equivalent with regard to all‑cause hospitalisation outcomes

  • sacubitril valsartan was superior to ACE inhibitors with regard to all‑cause and cardiovascular mortality and superior with regard to all‑cause hospitalisation.

3.15 The overall safety profile of sacubitril valsartan was comparable to that of the ACE inhibitor, enalapril, during the double‑blind trial period of PARADIGM‑HF. Compared with the enalapril group, fewer patients in the sacubitril valsartan group experienced 1 or more treatment‑related adverse events, 1 or more serious adverse events, death or discontinued as a result of an adverse event. Treatment with sacubitril valsartan was associated with higher rates of hypotension. The company noted this was a result of sacubitril valsartan's greater vasodilator effect, and that there was no increase in the rate of discontinuation because of possible hypotension‑related adverse effects. Fewer patients having sacubitril valsartan experienced renal adverse events compared with those having enalapril, which was driven by a lower incidence of renal impairment and renal failure in the sacubitril valsartan group (10.14% and 2.66% respectively) compared with the enalapril group (11.52% and 3.41% respectively). Other adverse events that were more frequent in the enalapril group than in the sacubitril valsartan group were hyperkalaemia, cardiac failure, cough, dyspnoea, hypertension, hyperuricemia and constipation.

Cost-effectiveness evidence

3.16 The company submitted a 2‑state Markov economic model with health states defined as 'alive' and 'dead'. In the base case, the model included all‑cause mortality, all‑cause-hospitalisation rates, EQ‑5D and adverse event rates. The company stated that models with similar structures have been published previously, including the model submitted to NICE as part of the technology appraisal guidance on ivabradine for treating chronic heart failure. In the company's primary base‑case analysis, patients entered in the model in either the sacubitril valsartan or the enalapril treatment arms to reflect the company's anticipated first‑line positioning of sacubitril valsartan in the heart failure treatment pathway. The company also developed a secondary base‑case model that included patients who cannot have ACE inhibitors; patients entered this model in either the sacubitril valsartan or ARB treatment arms. The ARB considered in the economic analysis was candesartan, and a class effect for ARBs was assumed.

3.17 The company's base-case analysis used individual patient‑level data from the PARADIGM‑HF trial, such that the model was run the same number of times as the number of patients included in the analysis (8,399). Actual model outcomes were obtained by averaging across the individual patients' outcomes. The model used a cycle length of 1 month, and a half‑cycle correction was applied to all calculations. The model was conducted over a lifetime horizon (equivalent to 30 years). Both costs and benefits were discounted at a rate of 3.5% and the perspective adopted was that of the NHS and personal social services. Deterministic and probabilistic sensitivity analyses were also done to explore parameter uncertainty in the model.

3.18 The model population characteristics were based on the full analysis set population of PARADIGM‑HF (see section 3.6). Baseline characteristics were used as covariates in the regression models to estimate mortality, hospitalisation and quality of life in the economic analysis.

3.19 In both treatment and comparator arms of the model, a proportion of patients had standard care (and other background therapies) in addition to sacubitril valsartan or enalapril (or candesartan). Standard care was defined as beta blockers and aldosterone antagonists. Additional background therapies consisted of diuretics, digoxin, anticoagulants, aspirin, adenosine diphosphate antagonists and lipid‑lowering drugs.

3.20 The company's primary base-case analysis for sacubitril valsartan compared with enalapril modelled the likelihood of a patient experiencing a hospitalisation event using a negative binomial regression model. Predicted all‑cause hospitalisation rates were determined by the treatment the patient had (sacubitril valsartan or enalapril) and patients' baseline characteristics, taken from the PARADIGM‑HF trial. These were used to inform the number of hospitalisations occurring in the initial period of the economic analysis, but also allowed for extrapolation beyond the follow‑up of the PARADIGM‑HF trial. The rate of hospitalisation was assumed constant over time, therefore assuming that hospitalisation was not related to disease progression over time.

3.21 In the company's primary base-case analysis, transition probabilities between the alive and dead health states were taken from all‑cause mortality data from PARADIGM‑HF in the base case. All‑cause mortality was estimated with survival regression analysis. The company chose the Gompertz distribution for its base case, noting that its clinical experts considered it to be clinically plausible, that it provided the most conservative (shortest) estimate of survival benefit, and that it was used in NICE technology appraisal guidance on ivabradine for treating chronic heart failure. Predicted all‑cause mortality was determined by the treatment the patient had (sacubitril valsartan or enalapril) and patients' baseline characteristics, taken from the PARADIGM‑HF trial. The mortality model was run using the full analysis set population of the PARADIGM‑HF trial and the model outputs provided daily hazard rates. These were used to model the probability of patients dying in the initial period of the economic analysis but also allowed for extrapolation of mortality beyond the end of the PARADIGM‑HF trial for the remainder of the modelled time horizon. In an alternative analysis, the company derived transition probabilities between the alive and dead health states from cardiovascular‑related mortality. The Gompertz distribution was also used for this analysis.

3.22 The company used a linear mixed regression model based on EQ‑5D trial data from PARADIGM‑HF to predict the utility scores for patients in the base‑case analysis. Since the economic model did not explicitly include mutually exclusive health states (other than the alive and the dead states), mean utility values over time were calculated for each patient profile. Predicted EQ‑5D scores were based on which treatment the patient had, baseline characteristics (including baseline EQ‑5D), and risk of hospitalisation and adverse events.

3.23 A small but statistically significant EQ‑5D treatment effect in favour of sacubitril valsartan was assumed after controlling for the effects of hospitalisations and adverse events. This was assumed to persist for the duration of the time horizon. EQ‑5D scores were assumed to decline at a constant rate of −0.008 per year over the modelled time horizon (30 years), which was based on data from PARADIGM‑HF and a longitudinal study by Berg et al. (2015) which reported an annual decline in EQ‑5D of −0.006. The rate of decline was not dependent on baseline characteristics.

3.24 The company applied utility decrements when a patient was hospitalised, with a decrement of −0.105 during days 0 to 30, and −0.054 during days 30 to 90. The company also applied adverse event utility decrements for hypotension (−0.029) and cough (−0.028) over an average duration of 64.9 days and 73.3 days respectively. The effect of serious adverse events that needed hospitalisation on quality of life was assumed to be captured in the utility decrements associated with hospitalisation.

3.25 For the comparison of sacubitril valsartan with ARBs in the company's secondary base‑case analysis, all‑cause mortality and all‑cause hospitalisation models used the network meta‑analysis results to estimate the effectiveness of sacubitril valsartan compared with candesartan. For the all‑cause hospitalisation model the company applied a hazard ratio of 0.90 for ARBs compared with ACE inhibitors (that is, candesartan was assumed to be 10% more effective than enalapril in preventing hospitalisations). Utility values in the ARB treatment arm of the model were assumed to be equivalent to the ACE inhibitor treatment arm as modelled in the primary base‑case analysis.

3.26 Adverse events included in the base‑case model were based on the full analysis set population rather than the safety analysis set. The company stated this was to ensure consistency with the modelling of clinical and quality of life outcomes, which were also based on the full analysis set population. The company modelled the adverse events by assuming a constant probability of a specific adverse event occurring each cycle. It assumed that all‑cause hospitalisation included all the relevant serious adverse events, including the associated costs and how it affects patients' quality of life. Adverse events in the trial that were designated non‑serious were modelled independently from hospitalisation. These were hypotension, elevated serum creatinine, elevated serum potassium, cough and angioedema. Adverse events in the secondary analysis in the ARB treatment arm of the model were assumed to be equivalent to the sacubitril valsartan treatment arm.

3.27 Resource use and costs considered in the model included:

  • intervention and comparator costs (including background therapies)

  • treatment initiation costs

  • hospitalisation costs

  • heart failure management costs

  • adverse event costs.

3.28 The company based the daily costs of ACE inhibitors and sacubitril valsartan on observed doses from PARADIGM‑HF. The cost of hospitalisation was based on healthcare resource groups mapped from physician‑reported diagnoses, surgeries and interventional procedures that could be classified, and medical management hospitalisations with more than 30 instances considered. Typical costs of standard care (including beta blockers and aldosterone antagonists) and background medications were based on recommended doses. Estimates of background resource use, including emergency department referrals, outpatient contacts and GP visits, were taken from relevant national sources. A Clinical Practice Research Datalink (CPRD) analysis commissioned by the company in order to characterise the burden of illness in the UK for patients with heart failure was used as the main source for resource use in the base case.

3.29 The primary base-case deterministic incremental cost-effectiveness ratio (ICER) for sacubitril valsartan compared with ACE inhibitors was £17,939 per quality‑adjusted life year (QALY) gained (representing incremental costs of £7,514 and incremental QALYs of 0.42), and the probabilistic ICER was £18,818 per QALY gained. The probabilities of sacubitril valsartan being cost effective at the maximum acceptable ICERs of £20,000 and £30,000 per QALY gained were 64% and 93% respectively.

3.30 Deterministic one-way sensitivity analysis showed that for the comparison with ACE inhibitors the ICER was most sensitive to all‑cause mortality, with the greatest effects on the ICER coming from the treatment effect of sacubitril valsartan on all‑cause mortality, the baseline risk of all‑cause mortality, and age (as a result of its effect on expected survival). Variables which had a modest effect included the improvements in health‑related quality of life and reduction in hospitalisations.

3.31 The company carried out deterministic scenario analyses for the comparison of sacubitril valsartan with ACE inhibitors. The scenarios associated with ICERs over £30,000 per QALY gained were if the sacubitril valsartan treatment effect were assumed to persist for less than 5 years and if the modelled time horizon was reduced to less than 5 years.

3.32 For the company's secondary base-case analysis of sacubitril valsartan compared with ARBs, the deterministic ICER was £16,481 per QALY gained (representing incremental costs of £8,513 and incremental QALYs of 0.52) and the probabilistic ICER was £17,599 per QALY gained. The probabilities of sacubitril valsartan being cost effective at the maximum acceptable ICERs of £20,000 and £30,000 per QALY gained were 60% and 77% respectively. Results of the one-way deterministic sensitivity analysis were consistent with the analysis comparing sacubitril valsartan with ACE inhibitors, except the all‑cause mortality hazard ratio for ARB compared with ACE inhibitors from the network meta‑analysis was the most influential parameter. This parameter was subject to a high degree of uncertainty as a result of the wide credible intervals generated by the network meta‑analysis (see section 3.39).

3.33 The company also presented results obtained using cardiovascular mortality (rather than overall mortality in the base case). In this case, the deterministic ICERs for sacubitril valsartan were £16,678 per QALY gained compared with ACE inhibitors and £16,569 per QALY gained compared with ARBs.

ERG critique of the company's submission

Clinical effectiveness

3.34 The ERG commented that the PARADIGM‑HF trial was well conducted and that most patients in the trial were taking beta blockers as concomitant therapies, which reflected UK clinical practice. However, the ERG had the following concerns:

  • The ERG noted that the population from the trial had a mean age of 63.8 years and that 32% of patients were younger than 55 years old. It stated that in routine clinical practice average age would be much higher, at between 76 years (men) and 80 years (women). The ERG also noted that the trial included a lower proportion of women (about 22%) than in UK clinical practice. The ERG was advised by its clinical experts that these patient characteristics were associated with improved outcomes, although it also noted that this effect would be observed across both treatment arms of the trial.

  • The ERG was advised by its clinical experts that the cardiac device use observed at baseline in the trial was lower than is typical in UK clinical practice.

3.35 The ERG was advised by its clinical experts that the dose of valsartan (in sacubitril valsartan) in the PARADIGM‑HF trial was higher than that typically prescribed in UK clinical practice. The ERG noted that the target dose of sacubitril valsartan was 200 mg twice daily, of which 103 mg is valsartan, which is equivalent to a 160 mg dose of valsartan given alone. The ERG noted that this dose was, according to the summary of product characteristics, the maximum dose allowed in clinical trials for valsartan monotherapy. According to clinical expert opinion provided to the ERG it is uncommon for patients to tolerate such high doses of valsartan in UK clinical practice. The ERG noted several factors that were likely to have contributed to the increased tolerability of valsartan in the trial:

  • At baseline, around 78% of patients were taking an ACE inhibitor and around 23% of patients were taking an ARB.

  • Around 70% of patients had been diagnosed with heart failure for over 1 year.

  • The minimum tolerability inclusion criterion in the PARADIGM‑HF protocol defined a minimum tolerable dose of valsartan (160 mg daily), which appears to be higher than the average dose tolerated by patients in UK clinical practice.

    Patients in the trial did not have any serious comorbidities and death was included as a reason for discontinuation in both the trial and the CPRD analysis.

    The ERG stated that the higher dose of valsartan tolerated by patients in the trial affected the observed discontinuation of study drugs, which it suggested was likely to be higher in UK clinical practice than it was in the trial.

3.36 The ERG also had concerns over the comparison with enalapril because it was not representative of UK clinical practice. The company stated that enalapril was chosen because it is the ACE inhibitor that has been studied in the largest number of trials of patients with heart failure and it has a well‑documented mortality benefit. However, the ERG's clinical experts advised that, in the UK, the most commonly used ACE inhibitor is ramipril. The ERG analysed the CPRD data commissioned by the company which showed that ramipril is the most commonly used ACE inhibitor in the UK. Therefore, the ERG stated that comparing sacubitril valsartan with enalapril did not reflect UK clinical practice.

3.37 The ERG considered the Western Europe population to be the most representative of the UK (24% of patients in PARADIGM‑HF were from Western Europe). Clinical expert opinion sought by the ERG suggested that heart failure can have different causes across different geographical regions. The ERG also noted that the place of care was likely to have an effect on the use of medical devices; for example, implants are more likely to be seen in Western Europe and North America than in Latin America. In response to the clarification questions, the company provided the baseline characteristics of patients in the trial who were part of the Western European population (n=2057). The ERG noted that sacubitril valsartan was associated with a favourable but non‑statistically significant difference in the Western Europe subgroup for the primary composite outcome, as well as in terms of both cardiovascular and all-cause mortality. It considered that this may be because people in this subgroup have lower blood pressure, less severe heart failure and more intensive 'standard care' (as indicated by a slightly higher consumption of ACE inhibitors). The ERG concluded that the effect of sacubitril valsartan observed in the trial population may not be observed when used in clinical practice in the UK.

3.38 The ERG considered the results from the PARADIGM‑HF and TITRATION trials in relation to the company's proposed positioning of sacubitril valsartan in the treatment pathway. The ERG's clinical experts indicated that based on the PARADIGM‑HF trial design, population and outcomes, the evidence best supported sacubitril valsartan as a second‑line treatment option for patients who are still symptomatic despite taking an ACE inhibitor. The ERG did not agree that the company's first‑line positioning of sacubitril valsartan was reflected in the clinical trial evidence base for several reasons:

  • The ERG felt the trial population did not reflect a newly diagnosed population because 70% of patients had been diagnosed for more than 1 year, and almost all patients were already having either ACE inhibitors or ARBs before entry in to the trial.

  • The ERG commented that the mortality in the PARADIGM‑HF trial portrayed a scenario representative of the use of sacubitril valsartan in patients whose disease is established. It noted that less than 10% of patients in the trial had died by the end of year 1 and 20% were dead in both treatment arms by the end of year 2. The ERG contrasted this with the prognosis in NICE's guideline on chronic heart failure in adults: diagnosis and management that 30% to 40% of patients diagnosed with heart failure die within a year. The ERG stated that this reinforced its view that the evidence presented in the company submission was most applicable to the use of sacubitril valsartan as a second-line treatment option, given to patients who are still symptomatic despite taking an ACE inhibitor.

  • Because the patients in PARADIGM‑HF were symptomatic despite having ARBs and ACE inhibitors, the ERG noted that the effect of continuing these patients on ACE inhibitors was likely to misrepresent what would happen in patients who have not previously had ACE inhibitors or ARBs. It further stated that, in principle, the ACE inhibitor treatment regimen has been demonstrated to not improve these patients' symptoms, and therefore randomising them to the same treatment regime is unlikely to show any improvements. The ERG suggested that this has an effect on the observed relative effectiveness of sacubitril valsartan, which may be overestimated in the trial population when compared with patients who have not previously had ACE inhibitors or ARBs.

3.39 The ERG noted that the company used methods for the network meta-analysis that were in line with the NICE decision support unit's technical support document 2. It also noted that, across all outcomes (all‑cause mortality, cardiovascular mortality and all‑cause hospitalisation) there were no hazard ratios from the network meta‑analysis in which the credible intervals could be considered statistically significant. The ERG commented that the wide range of drug doses used to manage heart failure and the differences in NYHA classification of patients recruited to the trials in the network meta‑analysis were sources of clinical heterogeneity which may have resulted in the wide credible intervals. Overall, the ERG regarded the results of the network meta‑analysis conducted by the company to be uncertain because of the clinical heterogeneity in the trials underpinning the network.

3.40 The ERG discussed the Cochrane systematic review by Heran et al. that the company had referenced in its assumption of a class effect for ARBs. It noted that the Cochrane review included some trials in which the population studied was not within the scope issued by NICE (for example, because the patients included had heart failure with preserved ejection fraction). The ERG noted that there were similar results observed between the company's network meta‑analysis and the meta‑analysis from the Cochrane review, and stated that this gave some reassurance that the results were valid. However, it commented that the results needed to be interpreted with caution because of the inclusion in both meta-analyses of populations that were not within the scope issued by NICE.

3.41 Based on the ERG's concerns regarding the company's positioning of sacubitril valsartan as a first‑line treatment, the ERG considered that the clinical effectiveness of sacubitril valsartan compared with ARBs in newly diagnosed patients with heart failure remained uncertain.

Cost effectiveness

3.42 The ERG stated that the formulae within the economic model were generally sound and that the economic model was a good predictor of the PARADIGM‑HF trial outcomes. It also commented that the company had conducted scenario and subgroup analyses that were not requested in the NICE final scope but which added value to the submission. The ERG stated that the company did not provide a clear justification for choosing a patient‑level approach when modelling clinical effectiveness.

3.43 Since the model population was based on the population of the PARADIGM‑HF trial, the ERG reviewed how well the population reflected UK clinical practice (the ERG also reviewed this in its critique of the clinical effectiveness evidence for sacubitril valsartan; see sections 3.34, 3.35 and 3.37):

  • The ERG considered mean age at baseline, and noted that NICE's guideline on chronic heart failure in adults: diagnosis and management states that 30% to 40% of people diagnosed die in the first year, but thereafter the mortality is less than 10% per year. Based on this, the ERG suggested that the starting age of patients in the economic analysis was a key factor. The ERG constructed hypothetical survival curves for mortality based on patients entering the model at 64 years old or 75 years old. Comparing the difference in the areas under the superimposed survival curves, the ERG demonstrated that there were considerable survival gains over time for the younger population, and this had implications for the costs and benefits collected during that time.

  • The ERG was uncertain if the effectiveness of sacubitril valsartan in preventing hospitalisation differed across different age groups. The ERG discussed a study by Jhund et al. (2015) which concluded that the effect of sacubitril valsartan compared with enalapril was consistent across age groups, even though hazard ratios were non‑statistically significant in older groups. The ERG suggested that the non‑statistically significant result in older people was consistent with expert opinion advising that for patients who are around 80 years old, clinicians expect treatment to improve patients' quality of life but not mortality. The ERG commented that this was particularly relevant to the UK given that the average age of patients seen in clinical practice is between 75 and 80 years.

  • Although the company positioned sacubitril valsartan as a first‑line treatment in newly diagnosed patients, the population in the PARADIGM‑HF trial was not reflective of newly diagnosed patients with heart failure seen in UK clinical practice.

  • The target dose of valsartan (in sacubitril valsartan) in the PARADIGM‑HF trial was the maximum dose allowed for valsartan. However, the ERG stated that it seems to be uncommon for patients to tolerate such high doses of valsartan in clinical practice (see section 3.35). This affects the observed discontinuation of study drugs, which is likely to be higher in UK clinical practice than in the trial.

  • The ERG's clinical experts advised that heart failure can have different causes across different geographical regions. The ERG also noted that there is likely to be variation in medical device use across regions (see section 3.39). The ERG's clinical experts also advised that differences in mortality across North America, Western Europe and the UK could be expected given that the UK has previously used fewer implantable cardioverter-defibrillators than the rest of Europe or North America.

  • The ERG's clinical experts advised that the cardiac device use observed at baseline in PARADIGM‑HF was lower than what would be expected in UK clinical practice and that the use of devices at baseline is an important prognostic factor for heart failure.

3.44 The ERG discussed the modelled treatment regimens. It stated that these broadly reflected the PARADIGM‑HF trial, even though there was some inconsistency in the chosen treatment doses (see section 3.52). The ERG was concerned with how representative the modelled treatment regimens were of clinical practice. It noted that the modelled dose of sacubitril valsartan of 400 mg per day was unlikely to accurately represent the average dose of valsartan tolerated typically observed in clinical practice (see section 3.35). The ERG also noted that the dose of the ARB, candesartan, modelled in the economic analysis (32 mg daily) was different to the average dose reported in the CPRD analysis (around 10 mg per day during the follow‑up period) and the observed daily mean dose of candesartan in UK clinical practice (around 16 mg according to clinical experts). Finally, the ERG noted a discrepancy in the observed average daily dose for enalapril of 18.9 mg in the PARADIGM‑HF trial compared with the CPRD data of about 16.5 mg. The ERG stated that the difference in intervention doses compared with clinical practice had an effect on the observed discontinuation of study drugs. The ERG noted that the base‑case economic model did not consider drug discontinuation, but the company had carried out a scenario analysis in which the inclusion of discontinuation over the lifetime time horizon had only a modest effect on the ICER.

3.45 The ERG reiterated that the modelled population did not reflect patients typically observed in clinical practice or a newly diagnosed population, both of which would affect mortality in the model. The ERG did not run any additional analyses to try and replicate the mortality of newly diagnosed patients because too many assumptions would have had to be made to approximate a treatment‑naive population.

3.46 The ERG noted that the company's decision to use a Gompertz distribution was based on this distribution presenting the most plausible survival time. The ERG noted the company had not tried other approaches than parametric curves, and suggested that different modelling options, such as spline models, would have been useful. Even though the Gompertz distribution produced the most plausible survival curves among the group of alternative distributions considered, the ERG considered that it could represent an overestimate of treatment effects compared with different approaches.

3.47 The ERG discussed the company's use of the all-cause mortality model in the base case, as opposed to the use of cardiovascular mortality. The company had chosen the all‑cause mortality model because it was considered the most conservative approach (that is, it produced the higher ICER). The ERG commented that the cardiovascular mortality approach was likely to have been more robust. It stated that there were issues in using an all‑cause mortality approach because it included non‑cardiovascular mortality observed in the trial. Clinical experts explained that non‑cardiovascular mortality was likely to be overestimated in the trial (compared with UK life tables) given that the trial included a considerable proportion of patients from countries where other causes of death, such as infection, are more prevalent than in Europe and North America.

3.48 The ERG commented that even though the modelled effect of age at baseline in cardiovascular mortality seems to be appropriate to capture the PARADIGM‑HF trial data, the unexpected shape of the mortality curve leads to other issues in the economic analysis, such as the lack of face validity of the predicted life expectancy in the model. The ERG highlighted that the mortality survival model made some implausible predictions, such as 21‑year olds having the same life expectancy as 87‑year olds, 72‑year olds having a much higher life expectancy than 18‑year olds. The ERG appreciated that this was a direct implication of the modelled effect of age at baseline on cardiovascular mortality, which in turn was a direct consequence of the PARDIGM‑HF trial data.

3.49 The ERG was concerned with the validity of the company's health‑related quality of life analysis. Firstly, the ERG could not be certain whether there was a baseline statistically significant difference in patients' EQ‑5D scores between the 2 treatment groups of sacubitril valsartan and enalapril. It suggested the statistical test performed by the company that found there was no statistically significant difference might not be appropriate. Secondly, the ERG stated that the trial and consequently the model outcomes could potentially be biased if there was a clinically significant difference in patients' disease severity and quality of life across the treatment groups. The ERG suggested that, assuming patients in a healthier state would have better outcomes, the potential imbalance in disease severity might have favoured the sacubitril valsartan group.

3.50 For the secondary base‑case analysis of sacubitril valsartan compared with ARBs, the ERG was concerned with the clinical heterogeneity in the trials underpinning the network meta‑analysis. It considered that the clinical effectiveness of sacubitril valsartan compared with ARBs in patients newly diagnosed with heart failure remained an unanswered question.

3.51 Regarding the company's use of CPRD data for estimating resource use, the ERG agreed with the company that such real‑world data was more robust and more reflective of the UK population than literature studies. However, the ERG was concerned with the appropriateness of using CPRD data to estimate the resource use for the patient profiles observed in the trial because there were differences in the 2 populations.

3.52 The ERG noted that the company's assumptions of daily drug doses were not consistent across different treatments. For some treatments, the doses were estimated as the average between the minimum and maximum dose; for others, the doses were based on maximum doses. The ERG carried out an exploratory analysis to reflect consistent drug dose assumption and using the cost of ramipril instead of enalapril. Based on advice from its clinical experts, it assumed a reduced cost for ramipril reflecting the fact that in clinical practice ramipril is given as a single daily dose, rather than as 2 daily doses.

3.53 The ERG stated the hospitalisation cost would be expected to depend on starting age and time. The ERG's clinical experts advised that the incidence of hospitalisation caused by renal failure in the trial appeared to be lower than expected, and that the cause could be as a result of the population being younger and healthier than in UK clinical practice. The ERG therefore had concerns that the starting age in the model affected the cost savings caused by the reduction in hospitalisations.

3.54 The ERG was concerned that the company had not appropriately accounted for parameter uncertainty in the economic analysis. The ERG stated that patients' baseline characteristics should have been included in the deterministic and probabilistic sensitivity analyses, given the concerns regarding the lack of generalisability of the PARADIGM‑HF trial population to clinical practice. The ERG also commented that the baseline characteristics were key parameters in the economic model given that these were included as prognostic factors of mortality, hospitalisation, quality of life and costs in the regression analyses.

ERG exploratory analyses

3.55 The ERG's scenario analyses were done in populations with a mean starting age of 64 years (as per the company's base case reflecting the clinical trial) and a mean starting age of 75 years (to better reflect the UK heart failure population). The ERG used cardiovascular mortality and average patient characteristics in each cohort (as opposed to all‑cause mortality and the use of patient‑level characteristics in the company's primary and secondary base case analyses).

3.56 The ERG's scenario analyses in the 64‑year old population included the following changes to the company's primary base‑case model:

  • The ERG explored a change in the cardiovascular mortality hazard ratio to reflect the Jhund et al. point estimate and confidence interval limits for the 55- to 64‑year category. The hazard ratio used was 0.79 (CI 0.64 to 0.98).

  • The ERG explored alternative baseline utility values, using a utility value of 0.72 reported by Berg et al. and, in another scenario analysis, using a utility value of 0.66 as reported by Austin et al.

  • The ERG explored the use of a simplified approach to modelling quality of life. Sacubitril valsartan's effect on patients' quality of life was linked to the incidence of adverse events and hospitalisation events and disease progression in both treatment arms. Therefore, the quality of life regression model was not used (although some estimates were taken from it because they had been validated by clinical experts). The effect of sacubitril valsartan alone on quality of life was also removed to reflect the lack of robust evidence to support a measurable improvement in patients' quality of life caused by sacubitril valsartan other than through hospitalisation, mortality and adverse events. The ERG assessed the treatments' effect on quality of life through:

    • adverse events and hospitalisation events (applying the same utility decrements used by the company to estimate the loss in quality of life due to the incidence of adverse events and hospitalisation)

    • disease progression (applying the same utility decrement used by the company to reflect the loss of quality of life as time progressed).

  • The ERG explored changing the drug doses used in the model to reflect a consistent approach to the estimation of drug costs.

  • The ERG included the cost of ramipril (using the ERG drug dose assumption of a single daily 5 mg dose) to reflect clinical practice in the UK.

  • The ERG used the option included in the company's economic model to run the ERG‑corrected model considering treatment discontinuation.

  • The ERG explored using the company's subgroup analysis results to run the ERG‑corrected model considering the Western European population.

3.57 The ERG's scenario analyses in the 75‑year old population included the following:

  • The ERG changed the cardiovascular mortality hazard ratio in the model to reflect the Jhund et al. point estimates and confidence interval limits for the ≥75‑year old category. This HR (0.84, 95% CI 0.67 to 1.06) was non‑statistically significant, so the ERG ran the model with a hazard ratio of 1.

3.58 The ERG noted its additional analyses for the 64‑year old and 75‑year old populations were consistent with the company's sensitivity analysis in showing that the model results were relatively robust but were most sensitive to changes in the mortality hazard ratio, with cardiovascular mortality the key model driver.

3.59 The ERG presented ICERs for sacubitril valsartan compared with enalapril assuming that sacubitril valsartan was used as a second‑line treatment in clinical practice. The ICERs estimated by the ERG were based on the PARADIGM‑HF population and clinical effectiveness results. The ERG used the following assumptions:

  • mean starting age of 75 years

  • baseline utility value taken from Berg et al.

  • using the cost of ramipril instead of enalapril to reflect clinical practice in the UK

  • using the effectiveness outcomes, costs, QALYs and population characteristics of the Western European subgroup analysis

  • using an alternative quality of life modelling approach (see section 3.61) and adjusted drug costs to reflect target doses consistently across the economic analysis.

3.60 The second-line ICERs estimated by the ERG are presented in table 1.

Table 1 ERG's estimated ICERs: second-line treatment

Results per patient

Sacubitril + SoC

Enalapril + SoC

Incremental value

Company's base case with ERG corrections

Total costs (£)

£22,961

£14,308

£8,653

QALYs

5.40

4.82

0.58

ICER

£15,026

Mean age at baseline of 75 years

Total costs (£)

£19,498

£12,562

£6,936

QALYs

4.43

3.99

0.44

ICER (compared with base case)

£15,843

ICER with all changes incorporated

£15,843

Change in baseline utility to reflect Berg et al utility (0.72)

Total costs (£)

£22,824

£14,299

£8,525

QALYs

5.11

4.55

0.55

ICER (compared with base case)

£15,407

ICER with all changes incorporated

£16,190

Change in QoL modelling approach

Total costs (£)

£22,961

£14,308

£8,653

QALYs

5.30

4.80

0.50

ICER (compared with base case)

£17,413

ICER with all changes incorporated

£19,697

Change in pharmaceutical costs to reflect drug target doses

Total costs (£)

£23,085

£14,430

£8,655

QALYs

5.40

4.82

0.58

ICER (compared with base case)

£15,030

ICER with all changes incorporated

£19,701

Change in pharmaceutical costs to reflect the cost of ramipril

Total costs (£)

£22,961

£14,257

£8,704

QALYs

5.40

4.82

0.58

ICER (compared with base case)

£15,115

ICER with all changes incorporated

£19,843

Western Europe subgroup

Total costs (£)

£24,182

£17,341

£6,841

QALYs

4.86

4.52

0.33

ICER (compared with base case)

£20,550

ICER with all changes incorporated

£29,478

Abbreviations: HR, hazard ratio; ICER, incremental cost‑effectiveness ratio; SoC, standard of care; QALYs, quality-adjusted life years; QoL, quality of life.

3.61 The ERG estimated a second-line ICER for sacubitril valsartan compared with ARBs of £30,140 per QALY gained. Noting its previous concerns, the ERG considered that its ICERs must be interpreted with caution because of uncertainty around the effectiveness of sacubitril valsartan compared with enalapril when analysed in the context of UK clinical practice. The ERG also presented further scenario analyses which demonstrated the variance in values when different hazard ratios and mortality approaches (cardiovascular or all‑cause) were taken. In these analyses, the ICERs for sacubitril valsartan compared with ACE inhibitors ranged from £14,942 per QALY gained to being dominated (that is, ACE inhibitors were both more effective and less costly).

Company's new evidence in response to consultation

3.62 In response to the consultation, the company requested permission to submit new evidence, which was granted by NICE. The new evidence was to support the clinical effectiveness and safety of sacubitril valsartan in people not included in the draft recommendations, specifically:

  • people with NYHA class IV symptoms (see section 3.63)

  • people who have not previously had ACE inhibitors or ARBs (see sections 3.64 to 3.67)

  • people with LVEF more than 35% (see section 3.68).

3.63 The company presented effectiveness data for a post‑hoc subgroup of patients in the PARADIGM‑HF trial with NYHA class IV symptoms at randomisation (n=60). The data showed that the primary end point was observed in 33.30% (10/33) of patients having sacubitril valsartan compared with 40.74% (11/27) of patients having enalapril (HR 0.71, 95% CI 0.30 to 1.69). Lower rates of cardiovascular deaths were also seen in this subgroup for patients having sacubitril valsartan (18.8%, 6/33 patients) than in those having enalapril (22.22%, 6/27 patients; HR 0.87, 95% CI 0.28 to 2.73). The company did not present data for the other component of the primary outcome for this subgroup (that is, first hospitalisation for worsening heart failure). The company also stated that 323 patients had NYHA class IV symptoms during the double‑blind phase of the trial but did not present any further information for these patients. The company also presented safety data for this subgroup that showed that for people having sacubitril valsartan, the proportions of patients with at least 1 adverse event (72.3%), hypotension (9.09%), renal impairment (6.06%) and hyperkalaemia (15.5%) were comparable to those in people with NYHA class II (82.11%, 17.39%, 9.79% and 11.69% respectively) and NYHA class III (78.53%, 18.71%, 11.04% and 12.07% respectively).

3.64 To demonstrate the clinical efficacy of sacubitril valsartan in people who have not previously had ACE inhibitors or ARBs, the company suggested that a subgroup of the PARADIGM‑HF trial comprising people who had been diagnosed less than 3 months before entering the trial could be considered a proxy. The company stated that for this subgroup analysis, the p value of statistical interaction was 0.2677. From this statistically non‑significant interaction, the company inferred that the treatment benefit of sacubitril valsartan over ACE inhibitors was independent of time since diagnosis. The company interpreted from this that the treatment benefit of sacubitril valsartan over ACE inhibitors would also be seen in people who have not previously had ACE inhibitors or ARBs.

3.65 The company stated that in the PARADIGM‑HF trial, the treatment benefit of sacubitril valsartan compared with enalapril for the primary composite end point and heart failure hospitalisations was evident in the first 30 days after randomisation. To support this, the company presented 2 graphs showing Kaplan–Meier estimates for the first 30 days after randomisation. These graphs demonstrated a decreased hazard for the first hospitalisation for heart failure (HR 0.60, 95% CI 0.38 to 0.94, p=0.027) and 'sudden death' (HR 0.80, 95% CI 0.68 to 0.94, p=0.08) in people having sacubitril valsartan within the first 30 days. The company did not define sudden death, but stated that it was the most common cause of death in the trial and attributed it to 36.23% of total deaths. The company did not present corresponding graphs and data for the primary end point or cardiovascular death.

3.66 To support the safety and tolerability of sacubitril valsartan in people who have not previously had ACE inhibitors or ARBs, the company stated that in the TITRATION study most of these patients were able to achieve and maintain the target dose of sacubitril valsartan.

3.67 The company also presented safety data in people who have not previously had ACE inhibitors or ARBs from the studies evaluating sacubitril valsartan for treating hypertension (a different indication than chronic heart failure). The company presented the percentages of patients with at least 1 adverse event, at least 1 serious adverse event and with adverse events leading to discontinuation. The data compared the safety profile of sacubitril valsartan with placebo, olmesartan, amlodipine and valsartan only in people who have not previously had ACE inhibitors or ARBs. The company did not present corresponding data for people who had previously had ACE inhibitors or ARBs. However, it concluded that these data demonstrated a comparable safety profile for sacubitril valsartan independent of previously having ACE inhibitors or ARBs.

3.68 To support the effectiveness of sacubitril valsartan in people with LVEF of more than 35%, the company presented 2 post-hoc analyses of the PARADIGM‑HF trial for the subgroups defined by ejection fraction at baseline. In the first analysis, the company defined the subgroups on the basis of LVEF at screening (less than 28%, 28% to 33%, and more than 33%). The company presented a forest plot that showed that for the primary end point, the point estimates in all 3 subgroups were comparable and that the p value for statistical interaction among subgroups was 0.9720. In the second analysis, the company defined 5 subcategories on the basis of LVEF at screening: 15% or less, 16% to 20%, 21% to 25%, 26% to 30%, 31% to 35%, and more than 35%. The company reported that there were no significant subgroup interactions for either the primary end point (p=0.9377) or cardiovascular death (p=0.3367), but did not present point estimates in these subgroups for either outcome.

3.69 The company commented on the ERG's exploratory analyses, highlighting that using the hazard ratio from a subgroup (Western Europe) instead of overall population (see sections 3.56 and 3.59) had no statistical basis because there was no subgroup-treatment effect interaction (p=0.373). The company also noted that the Western Europe subgroup was not powered to detect statistically significant differences in the primary end point. The company also commented that in the PARADIGM‑HF trial, point estimates for all subgroup analyses suggested a benefit in the sacubitril valsartan group and different treatment effect in subgroups should be interpreted with caution. Applying clinical effectiveness from the overall population on the ERG's preferred scenario, the company calculated an ICER of £19,843 per QALY gained for sacubitril valsartan compared with ACE inhibitors.

3.70 The company also commented that the ERG's alternative approach for modelling quality of life did not take into account the utility benefit of sacubitril valsartan (0.011), which was derived from the difference in the EQ-5D scores between the 2 treatment groups in the trial after controlling for the effects of hospitalisations and adverse events. The company did not agree with the ERG's rationale that there may have been a statistically significant difference in patients' EQ‑5D scores at baseline, which may have biased the EQ‑5D outcomes in favour of sacubitril valsartan; the EQ‑5D analysis was based on a repeated measures ANCOVA model that controls for random differences or imbalance in baseline scores. The company justified the assumption of utility benefit with sacubitril valsartan, noting that in PARADIGM‑HF more people having sacubitril valsartan than those having enalapril had symptomatic benefit, as evidenced by improvement in KCCQ scores and NYHA class. Applying a utility gain of 0.011 with sacubitril valsartan over all other ERG‑preferred assumptions, the company calculated an ICER of £25,607 per QALY gained for sacubitril valsartan compared with ACE inhibitors.

ERG's critique of company's new evidence

3.71 The ERG commented that the new evidence for sacubitril valsartan's effectiveness and safety in people with NYHA class IV heart failure at baseline (see section 3.63) was based on a small sample size (n=60) compared with the overall number of patients in PARADIGM‑HF: the full analysis set for the primary end point comprised 8,399 patients and 5,931 patients in the safety analysis who had a NYHA classification at baseline. The ERG commented that the lack of statistically significant difference in the treatment effect between sacubitril valsartan and enalapril in the NYHA class IV subgroup would increase the uncertainty in estimating the cost effectiveness of sacubitril valsartan in these patients. The ERG also noted that the company did not clarify why it chose only 3 adverse events (hypotension, renal impairment and hyperkalaemia) to compare the safety of sacubitril valsartan with that of enalapril in this subgroup.

3.72 The ERG noted that the company did not provide data to support its assumption that sacubitril valsartan would be as effective in patients who have not previously had ACE inhibitors or ARBs as those who have. The ERG queried the relevance of new evidence presented by the company (see sections 3.64 to 3.67) for people with chronic heart failure who have not previously had ACE inhibitors or ARBs.

3.73 The ERG acknowledged that the post-hoc analyses for subgroups based on LVEF at baseline (see section 3.68) supported the clinical benefit of sacubitril valsartan over enalapril across different subgroups. However, it did not consider that these analyses provided enough evidence to assume an equal effect in people with LVEF more than 35% and in people with LVEF of 35% or less. For the second analysis that considered 5 subgroups, the ERG commented that the company did not provide details of the analysis such as the number of patients and the estimated effects in each subgroup.