A systematic review and economic evaluation of the clinical effectiveness and cost-effectiveness of aldosterone antagonists for postmyocardial infarction heart failure

Myocardial infarction

A myocardial infarction (MI) occurs when an area of the myocardium is exposed to prolonged ischaemia, usually as a result of a failure of the blood supply from one or more coronary arteries; the affected myocardium necroses and heals leaving a non-contractile scar. Depending upon the size of the infarcted area, heart function can be impaired leading to varying degrees of contractile function known as ‘systolic dysfunction’. 1 For many years, the commonest criteria used to diagnose MI were those of the World Health Organization (WHO), which required the presence of any two of the following three criteria: ischaemic symptoms, electrocardiographic changes and elevated creatine kinase-MB (CK-MB) level. 1 In 2000, the American College of Cardiology and the European Society of Cardiology redefined MI. 1 The new definition combined typical changes in biochemical markers, preferably the cardiac-specific troponins T or I (which show a rise and gradual fall), or CK-MB (which shows a more rapid rise and fall), with ischaemic symptoms, electrocardiographic (ECG) changes and/or coronary intervention. 1,2 Applying the new criteria to patients presenting with suspected cardiac chest pain leads to a 26% increase in the number of events diagnosed as MI (11% increase in the number of patients diagnosed with MI); approximately 66% of the additional MI would previously have been diagnosed as unstable angina. 3 This increase in the frequency of diagnosis of MI gives an indication of the impact of changes in diagnostic criteria on clinical practice.

Postmyocardial infarction heart failure

A large MI stimulates adaptations in cardiac structure and function (usually those of the left ventricle) known as ventricular remodelling in an attempt to compensate for reduced overall cardiac performance. Ventricular remodelling occurs rapidly immediately postMI, and more slowly thereafter. 4 The initial stage of ventricular remodelling is the thinning of the wall of the ventricle in the area of the infarct, and the dilatation of the ventricular chamber. This is followed by hypertrophy and fibrosis including lengthening of the non-infarcted part of the myocardium. 4 Initially ventricular remodelling preserves stroke volume and pump function of the left ventricle, however, these changes become maladaptive over time. The hormone aldosterone facilitates ventricular remodelling by promoting the development of myocardial fibrosis, leading to the progression of heart failure (HF) and its symptoms. 5 Left ventricular systolic dysfunction (LVSD) is the most common cause of postMI HF. 4 Other causes of HF after MI are papillary muscle dysfunction and mitral regurgitation or arrhythmias (for example atrial fibrillation), or rarer complications such as ventricular rupture or formation of ventricular septal defects; HF may also develop after an MI in the absence of any of these problems. 6 However, over time these adaptations can become counterproductive with dilatation and hypertrophy of the left ventricle eventually leading to impaired cardiac function. This reduced cardiac output leads to symptoms of breathlessness and fatigue, the syndrome known as HF. 4

The risk of developing HF postMI varies with the site, severity and type of MI, and the presence of comorbid conditions. Predictors of HF/LVSD after an MI have been reported as higher blood pressure,4,7,8 previous HF,4,7,8 left bundle branch block,7 anterior MI,4,7,8 diabetes,4,8 higher heart rate,4,7, older age,4,7–9 and larger infarcts (indicated by peak creatine phosphokinase level). 4,10 Patients who develop HF after an MI have been reported as having an increased risk of cardiovascular (CV) events including recurrent acute MI, stroke, atrioventricular block, ventricular arrhythmias cardiac rupture, and unexpected cardiac arrest8 and death. 8,11–13 Patients with impaired left ventricular ejection fraction (LVEF) have been shown to have worse outcomes than those with preserved LVEF. 14,15

Heart failure classification

There are two clinical subgroups of HF patients; either a reduction in the efficiency of the left ventricle to relax and fill with blood (diastolic HF), or to contract and pump blood (systolic HF), resulting in the heart becoming unable to meet the demands of the body. 16 Patients with ventricular dysfunction, whether diastolic or systolic, exhibit the same signs and symptoms; shortness of breath (dyspnoea) particularly when lying down (orthopnoea), fatigue and oedema. Cardiac remodelling occurs with both diastolic and systolic HF. Patients with diastolic HF develop concentric hypertrophy, with a normal or reduced heart cavity size, an increase in wall thickness, and a high mass : cavity ratio, end-systolic and diastolic volumes are generally normal and LVEF is usually normal or elevated. 16,17 In contrast, patients with systolic HF have an increase in cavity size, decreased or unchanged wall thickness, normal or reduced mass : cavity ratio, elevated systolic and diastolic volumes, and a lower LVEF. 16,17 Left ventricular (LV) dilatation always occurs with systolic dysfunction; however, with diastolic dysfunction, LV dilatation only occurs when additional injury is sustained such as an MI. 16 The differential diagnosis of diastolic and systolic HF can require the measurement of LVEF once the diagnosis of the presence of HF has been made; LVEF ≤ 45% is considered to be indicative of LVSD, indicating impaired contractile function, and a diagnosis of systolic HF can be made. 16

Coronary heart disease, hypertension, diabetes, increasing age and obesity are all risk factors for both diastolic and systolic HF. 16,17 Hypertension is a more common risk factor with diastolic HF because it increases LV afterload, which results in delayed relaxation time, elevated LV filling pressure and reduced end-diastolic volume. 16,17

The most commonly used systems for classifying patients with HF are the New York Heart Association (NYHA) and Killip classification systems; the classification system of the American College of Cardiology/American Heart Association (ACC/AHA) is also used. The NYHA classification has four classes based upon physical function and symptoms and the ACC/AHA provides definitions of different stages of HF. These two systems are compared in Table 1. 18 The Killip classification is used to stratify postMI patients into risk groups. 19

• Killip class I: No clinical signs of HF.

• Killip class II: Rales or crackles in the lungs, an S₃ gallop (an additional heart sound), and elevated jugular venous pressure.

• Killip class III: Acute pulmonary oedema.

• Killip class IV: Cardiogenic shock.

Burden on the NHS

The incidence of postMI HF is increasing in the UK because of the shifting age distribution of the population and increased survival after acute MI. 4,20 Approximately 707,000 people in the UK aged 45 years and older are thought to have HF (393,000 men; 314,000 women), with prevalence increasing steeply with age (1% under 65 years; 6–7% 75 to 84 years; 12–22% 85 years and older). 9 Approximately 40% of patients that experience an acute MI suffer HF. 20 The number of people with postMI HF in the UK for the year 2000 was estimated to be between 130,000 and 202,000, with associated annual costs to the NHS in the region of £125M to £181M. 21

Spironolactone

Spironolactone (Aldactone, Pharmacia Ltd, Sandwich, Kent, UK; alternative names include Novo-Spiroton, Spiractin, Spirotone, Verospiron and Berlactone) is licensed for use for HF in the UK. It inhibits the effect of aldosterone by competing for intracellular aldosterone receptors in the collecting ducts in the kidney, increasing water and sodium excretion and decreasing the excretion of potassium. Spironolactone is available alone or combined with hydroflumethiazide (Aldactide, Pharmacia Ltd), hydrochlorothiazide (Aldactazide, Pfizer Ltd) or furosemide (Lasilactone, Sanofi-Aventis). Spironolactone is indicated in patients with moderate to severe HF (NYHA class III and IV) caused by LVSD. 22,23

According to the Summary of Product Characteristics (SPC), the advised dose of spironolactone (Aldactone) in adults with congested cardiac failure is 100 mg/day, gradually increased up to 400 mg/day if required, with a maintenance dose of 25–200 mg/day once oedema is controlled. 24 However, high doses of spironolactone are rarely used, with most patients receiving between 12.5 and 50 mg/day. The SPC states that adverse drug reactions associated with spironolactone include electrolyte disturbances, hyperkalaemia, leukopenia, thrombocytopenia, malaise, gastrointestinal disturbances and drowsiness, rashes and abnormal hepatic and renal function. As a result of the effect on androgen receptors and other steroid receptors, gynaecomastia, testicular atrophy, sexual dysfunction and menstrual irregularities may occur. Being a mineralocorticoid antagonist, spironolactone may reduce the effectiveness of antidepressant drugs in the treatment of major depression, presumably by interfering with normalisation of the hypothalamic–pituitary–adrenal axis in patients receiving antidepressant therapy.

Eplerenone

Eplerenone (Ispra; Pfizer Ltd) is a selective aldosterone antagonist used as an adjunct in the management of chronic HF. It is specifically indicated for the reduction of risk of CV death in patients with HF and LV dysfunction within 3–14 days of an acute MI, in combination with standard therapies and as treatment for hypertension.

Eplerenone is similar to spironolactone, but has a greater affinity for the mineralocorticoid receptor and as a result is thought to have fewer side effects, in particular gynaecomastia. According to the SPC, the usual starting dose of eplerenone is 25 mg once daily, increasing to 50 mg/day after approximately 4 weeks. 24 The SPC state that common adverse drug reactions associated with eplerenone include hyperkalaemia, hypotension, dizziness, altered renal function and increased creatinine concentration.

Canrenone

Canrenone is the active metabolite of both spironolactone and potassium canrenoate. Neither canrenone nor potassium canrenoate are licensed as human medicines in either Europe or the USA so they will not be considered in this assessment. Available data on canrenone and potassium canrenoate may be used where appropriate to complement the evidence base for spironolactone and eplerenone.

Search strategy

Inclusion and exclusion criteria

Data extraction

Data were extracted by one reviewer and checked by a second reviewer, using a standardised data extraction form for effectiveness data, and extracting as reported in the studies for adverse events data. Discrepancies were resolved by discussion. Non-English language studies were extracted by one reviewer with a native speaker. Authors were contacted to obtain additional information on the baseline characteristics and outcome measures for ischaemic (postMI) subgroups from trials of general HF populations, where subgroup data were not reported in the published papers. Data from multiple publications of the same study were extracted as one study. The data extracted included: study characteristics (e.g. author, year, number of participants, countries and study centres, and duration of follow-up), intervention (type of aldosterone antagonist, dose and dosing regimen), comparator (placebo or optimal medical management), population characteristics (e.g. NYHA class included, proportion of population that were ischaemic, time from MI, LVEF, age, gender, using a standardised data extraction form), and data for the outcomes of interest as described above.

Quality assessment

The methodological quality of the included RCTs was assessed in terms of randomisation, allocation concealment, blinding, reporting of withdrawals, completeness of follow-up and the use of a power calculation and an intention-to-treat (ITT) analysis. Review-specific criteria included the requirement of baseline characteristics of the ischaemic population, the duration of follow-up and whether the population recruited was representative of the postMI HF population seen in clinical practice. Observational studies of adverse events data were assessed in terms of the use of a control group and the comparability of these groups at baseline, the method of data collection, the potential for selection bias and generalisability of the patient spectrum, length of follow-up, attrition and the appropriateness of the analysis. Full results of the quality assessment are presented in Appendix 4. One reviewer initially assessed study quality and this was checked by a second reviewer. Disagreements were resolved by discussion.

Data analysis

As a result of the clinical heterogeneity observed between studies, the results are presented in a narrative synthesis. The synthesis will explore the exhangeability between the drugs on a pharmacological basis, and the trials in relation to the population recruited. Although the focus of the review is the postMI population, only eplerenone has been investigated directly in this population. Results of subgroups of patient with chronic HF who have a history of cardiac ischaemia will be investigated in an attempt to provide some indication as to how spironolactone may perform in the postMI population.

Quantity and quality of research available

Effectiveness

The searches for studies of clinical effectiveness identified 1616 papers; 37 were considered potentially relevant and were retrieved as full papers for assessment. Five trials (across 10 publications) met the inclusion criteria; two papers provided adverse events data only. The flow of studies is shown in Figure 1. Three trials evaluated spironolactone,31,48,49 one evaluated eplerenone27 and one evaluated canrenone. 50 Of the five trials, three were conducted specifically in postMI HF patients,27,48,49 and two in a general HF population with a subgroup of patients in whom the cause of HF was ischaemic. 31,50 The authors of the two trials in the general HF population were contacted for data for the ischaemic subgroup; additional data were denied by the manufacturer of the drug for one trial,31 and were provided by the triallists for the second. 50

FIGURE 1.

Flow of studies through the review obtained from the searches for clinical effectiveness studies.

All five studies were described as randomised trials; however, none provided details of the randomisation method. Three trials reported allocation concealment, double blinding (although no details as to who this referred to were provided), and the use of a power calculation. 27,31,50 All five studies had groups that were comparable at baseline, clearly described eligibility criteria, had at least 12 months of follow-up, and used an ITT analysis. Of the three larger trials,27,31,50 only one had at least 90% follow-up. 50 Two of the studies were very small, poorly reported and were of poor quality; one was published in Polish48 and the other in Chinese. 49 Results of the quality assessment are given in Table 2, with guidelines for scoring each criterion provided in Appendix 4.

TABLE 2 - Results of the quality assessment of the RCTs of clinical effectiveness

Y, yes; N, no; U, unclear; ITT, intention to treat.

RALES,31 EPHESUS,27,51,52 AREA-IN-CHF,32,50,53,54 Ruta et al. ,48 Tu and Chen. 49

	RALES	EPHESUS	AREA-IN-CHF	Ruta (2006)	Tu (2003)
Number randomised reported	Y	Y	Y	Y	Y
Randomisation method used reported	U	U	U	U	U
Allocation concealment implemented	Y	Y	Y	N	U
Groups comparability at baseline	Y	Y	Y	Y	Y
Study reported as double blind	Y	Y	Y	N	N
Patient blinded specifically stated	U	U	U	U	U
Outcome assessors blinded specifically stated	U	U	U	U	U
Care givers blinded specifically stated	U	U	U	U	U
Power calculation used	Y	Y	Y	N	N
Eligibility criteria clearly described	Y	Y	Y	Y	Y
Baseline characteristics of ischaemic population provided	N	Y	N	Y	Y
At least 12 months of follow-up	Y	Y	Y	Y	Y
Representative sample recruited	N	Y	N	N	U
ITT analysis used	Y	Y	Y	Y	Y
Losses to follow described	Y	Y	Y	Y	N
At least 90% follow-up	N	N	Y	Y	Y

Adverse events

Initially, information relating to adverse events was sought from reference sources such as Martindale,33 Meyler’s,37 AHFS,38,39 FDA medical reviews and label information,40,41 and the BNF. 42 Little information was provided in these sources (see section Adverse events, p. 21), and therefore additional searches were conducted to identify studies reporting adverse events using the inclusion criteria stated previously (see Inclusion and exclusion criteria). From these additional searches, 459 papers were identified. Of these, 49 were considered potentially relevant for full paper review. Two papers were unobtainable. 55,56 Of the 47 papers screened, 15 met the inclusion criteria; one was only available as an abstract,57 and two papers were results from the same consecutive sample. 58,59 Two dose-ranging studies described only in the FDA eplerenone medical review were also included. 40 With three studies identified from the search for studies of clinical effectiveness, 18 studies (across 23 publications) were finally included in the review of adverse events. The flow of studies is shown in Figure 2. Results of the quality assessment for studies included only for adverse events are given in Table 3 and Table 4, with guidelines for scoring each criterion provided in Appendix 4.

TABLE 3 - Results of the quality assessment of additional RCTs reporting adverse events (results for EPHESUS, RALES and AREA IN-CHF are provided in Table 2)

Y, yes; N, no; U, unclear; ITT, intention-to-treat.

	Dose ranging: 01140	Dose ranging: Japanese40
Number randomised reported	Y	Y
Randomisation method used reported	U	U
Allocation concealment implemented	Y	Y
Groups comparability at baseline	U	U
Study reported as double blind	Y	Y
Patient blinded specifically stated	U	U
Outcome assessors blinded specifically stated	U	U
Care givers blinded specifically stated	U	U
Power calculation used	U	U
Eligibility criteria clearly described	Y	Y
Baseline characteristics of ischaemic population provided	N	N
At least 12 months follow-up	N	N
Representative sample recruited	U	N
ITT analysis used	Y	Y
Losses to follow described	N	N
At least 90% follow-up	N	N

TABLE 4 - Results of the quality assessment of observational studies reporting adverse events

Y, yes; N, no; U, unclear; NA, not available.

Lawson et al. (1982);60 Greenblatt and Koch-Weser (1973);61 Shah and Gottlieb (2006);62 Smith (1980);63 Hauben et al. (2007);64 Juurlink et al. (2004);65 Nassiacos and Meloni (2005);57 Witham et al. (2004);66 Sligl et al. (2004);67 Svensson et al. (2003);58,59,68 Anton et al. (2003);69 Tamirisa et al. (2004);70 Williams et al. (2006). 71

Control group was patients not discontinuing spironolactone and therefore not considered a control group in this assessment.

	Reference numbera:
	60	61	62	63	64	65	57	66	67	58, 59, 68	69	70	71
Control group recruited?	N	N	N	N	N	N	Y	N	N	N	N	Nb	N
Was the data obtained prospectively?	N	Y	N	Y	N	Y	Y	N	Y	Y	N	Y	N
Consecutive patients?	U	Y	U	U	U	Y	Y	Y	U	Y	U	Y	U
Inclusion criteria clearly reported	Y	Y	Y	Y	Y	Y	N	Y	Y	Y	Y	Y	Y
Study size explained	N	N	N	N	Y	Y	N	N	N	N	N	N	N
Similar baseline characteristics?	NA	NA	NA	NA	NA	NA	U	NA	NA	NA	NA	NA	NA
Conducted in patients with post-MI HF?	N	N	N	N	N	N	N	N	N	N	N	N	N
Confounders identified and accounted for?	Y	Y	N	N	N	Y	U	Y	Y	Y	Y	Y	Y
Losses to follow-up accounted for?	NA	NA	Y	NA	NA	NA	U	Y	Y	Y	Y	Y	Y
All recruited participants included in the final analysis?	Y	Y	N	Y	Y	Y	U	N	Y	Y	N	Y	Y
Outcomes measured at least 6 months after initiation of treatment?	U	U	N	Y	U	U	Y	Y	Y	Y	Y	Y	U
Adverse events were a primary outcome?	Y	Y	Y	N	Y	Y	Y	Y	Y	Y	Y	Y	Y
Attrition rate less than 90%?	NA	NA	U	Y	NA	NA	U	U	Y	Y	U	Y	Y

FIGURE 2.

Flow of studies through the review obtained from the searches for adverse events data.

Result of the review of clinical effectiveness

Exchangeability between RALES, EPHESUS and AREA IN-CHF

If a reliable indirect comparison is to be made between eplerenone, spironolactone and canrenone the trials need to be comparable, both in terms of the intervention evaluated and the population recruited. Given the small size and poorer quality of the trials by Ruta et al. 48 and Tu and Chen,49 clinical effectiveness can only be reliably assessed using the three main trials: EPHESUS (eplerenone),27,52 RALES (spironolactone),31 and AREA IN-CHF (canrenone). 50 The exchangeability between the the drugs and the three trials are discussed in the following sections.

Pharmacology of spironolactone (and canrenone) and eplerenone

Increased plasma aldosterone is associated with adverse CV effects in HF. Originally, this was thought to be because of its impact on sodium and water retention and potassium excretion, by binding to the mineralocorticoid receptor in epithelial tissues, such as the kidney. 72,73 Subsequent research has shown aldosterone to also act on non-epithelial tissues, such as the heart, brain and vasculature, and therefore to have a role in the pathophysiology of CV disease beyond ion transport,72,73 and to contribute to a number of non-renal mechanisms that progress disease such as thrombogenesis, cardiac fibrosis and cardiac remodelling. 72 The role of aldosterone in the pathophysiology of CV disease is still not fully understood. 72,73

Spironolactone (7α-acetylthio-3-oxo-17 α-pregn-4-ene-21,17β-carbolactone; C₂₄H₃₂O₄S; molecular weight 416.58) (Figure 3) is a competitive non-selective antagonist of the aldosterone receptor. The term non-selective is used because of spironolactone’s moderate affinity to other steroid receptors, including the progesterone and androgen receptors. 72–74 This affinity for these steroid receptors results in progestational and antiandrogenic side effects, such as gynaecomastia, disruptions to the menstrual cycle and impotence. 33,37,38,75,76 The peak plasma concentration of spironolactone is reached approximately 2.6 hours after oral administration. 24 Spironolactone is quickly metabolised by the liver into a number of metabolites. 74 The two main active metabolites of spironolactone are canrenoate and canrenone, which have elimination half-lives of approximately 13.8–16.5 hours. 41,72,73 These metabolites are mainly excreted in urine, but also in bile. 76

FIGURE 3.

Molecular structure of eplerenone and spironolactone.

Eplerenone [pregn-4-ene-7,21-dicarboxylic acid, 9,11-epoxy-17-hydroxy-3-oxo-,γ-lactone, methyl ester, (7α,11α,17α); C₂₄H₃₀O₆; molecular weight 414.50] is a derivative of spironolactone, and is therefore also a competitive antagonist of the aldosterone receptor; the 17α-thioacetyl group of spironolactone is replaced with a carbomethoxy group (Figure 3). 72,75,77 Eplerenone has a very low affinity for steroid receptors in vitro; however, it has an increased selectivity for the mineralocorticoid receptor, resulting in a comparable bioavailablity with spironolactone in vivo. 72,73,75 With a low affinity for other steroid receptors, eplerenone has fewer progestational and antiandrogenic side effects than spironolactone. 72,74,75,77,78 The mean peak plasma concentration of eplerenone is reached approximately 1.5 hours after oral administration. 41,77 Eplerenone is metabolised to inactive metabolites, primarily to the 6β-OH metabolite SC71597, predominantly by cytochrome P450 3A4 metabolism (CYP3A4). Most of the metabolites are excreted in urine and faeces (67% and 32%, respectively); less than 5% of eplerenone is excreted unchanged. 41,79–81 The elimination half-life of eplerenone is shorter than those of the active metabolites of spironolactone (4–6 hours versus 13.8–16.5 hours). 41,73 As a result, exposure to eplerenone after administration is less prolonged than with spironolactone metabolites and therefore the risk of hyperkalaemia is potentially reduced with eplerenone. 74

A study investigating the binding specificity and the ability to compete for binding sites using mineralocorticoid receptor–glucocorticoid receptor chimeras, showed that the amino acids 804–874, thought to affect the overall shape of the binding pocket of the mineralocorticoid receptor, are essential for the binding of eplerenone, spironolactone and aldosterone. 82 The study showed that the binding determinants of eplerenone and spironolactone were very similar, attributed to their very similar structure – a lactone ring at the C₁₇ position and similar C₇ side chains. 82 This demonstrates the similarity in structure and mode of action between these drugs, and therefore their pharmacological exchangeability.

Comparability of the populations in RALES, EPHESUS and AREA IN-CHF

There are several important ways in which the HF populations in these three trials differ from each other, and from current clinical practice.

Beta-blocker usage

Following an MI, beta-blockers decrease myocardial oxygen demand by decreasing heart rate and blood pressure, and increasing coronary artery flow by prolonging diastole. 83,84 They also decrease ventricular arrhythmias by blocking the activation of the sympathetic nervous system. 83 When beta-blockers were first used in HF patients, they were prescribed at full dose, and became contraindicated because of negative inotropic properties (weakening myocardial contractility). 85 Large RCTs conducted in the 1990s of bisoprolol, metoprolol and carvedilol demonstrated reductions in mortality and morbidity in patients with chronic HF with the use of beta-blockers when slow individualised upward titration was applied. 85 This altered method of administration resulted in beta-blockers no longer being considered contraindicated in patients with chronic HF, but an essential component of the medical management of these patients. 85 The number of prescriptions for beta-blockers has risen steeply in the UK, from 14,282,000 in 1991 to 26,810,000 in 2007 in England alone. 86 Even in the relatively short time between RALES and EPHESUS, beta-blocker prescriptions increased in England from 14,375,000 in 1996 to 20,439,000 in 2001. 86

Over the years, beta-blockers have been developed, and their pharmacological properties have changed:83,87

• first-generation (propanolol; timolol): non-selective antagonists, with equal affinity for beta-1 and beta-2 receptors

• second-generation (atenolol; metoprolol; bisoprolol): beta-1 selective

• third-generation (carvedilol): non-selective activity against alpha-1, beta-1, and beta-2 receptors.

Propanolol was the first beta-blocker licensed in the UK, for angina in 1965 and then hypertension in 1969. 88 Atenolol became available in 1976,88 metoprolol in 1997,24 and carvedilol in 2001. 89 The later, thrid-generation beta-blockers have vasodilatory effects,85 and are more effective at lowering aortic pressure than conventional beta-blockers. 87 Carvedilol also has antioxidant,90,91 antiarrhythmic,91,92 and anti-inflammatory91 properties. Currently the most commonly used beta-blockers for HF are metoprolol, bisoprolol and carvedilol. 85

Prior to the standard use of thrombolytics, ACE inhibitors and aspirin, beta-blockers produced a 20–30% reduction in mortality in postMI patients with LVSD; however, it was unclear whether this benefit would be sustained after the introduction of these treatments. 83,93 Retrospective analyses showed continued significant reductions in mortality and progression to severe HF in patients receiving beta-blockers alongside captopril94 or ramipril95 compared with those not receiving beta-blockers. A more recent prospective study showed carvedilol to reduce all-cause mortality by 23%, CV mortality by 25% and non-fatal MI by 41%; of the 1959 patients included, 98% were on ACE inhibitors, 86% were on aspirin and 45% received reperfusion treatment. 96 From these studies it can be seen that despite the introduction of thrombolytics, ACE inhibitors and aspirin into the standard treatment for postMI HF, beta-blockers continue to have a significant benefit in terms of mortality and morbidity and their use is now part of standard management postMI.

As a result, the use of beta-blockers in only 10.5% of the population in RALES, compared with 75% in EPHESUS and 80% in AREA IN-CHF, makes the exchangeability of the populations in these three trials questionable. In addition, it is likely that the type of beta-blocker would vary between RALES and the more recent trials EPHESUS and AREA IN-CHF. With RALES being conducted in 1995–6, patients recruited would not generally have been prescribed beta-blockers, and those that were, would not have been receiving third-generation agents. By the time EPHESUS was conducted in 1999–2001, third-generation beta-blockers would have been available and may have been prescribed to some patients. The use of third-generation beta-blockers was likely to have been common in the AREA IN-CHF trial. The use of third-generation beta-blockers would afford the patients receiving these drugs additional benefits over patients receiving first- or second-generation beta-blockers.

It is also likely that the type of patients prescribed beta-blockers will differ between RALES and the other two trials. Although higher-risk patients gain most from beta-blocker therapy,97 it is unlikely that the 10.5% of patients in RALES who received beta-blockers are those with more severe illness because at the time of the trial, beta-blockers tended to be used on patients with milder disease. In EPHESUS and AREA IN-CHF, beta-blocker use probably encompasses a wide range of disease severity; to confirm this, access to patient-level data would be required. There is also evidence that some patients are more likely to receive treatment with beta-blockers than others, over and above the severity of illness. One study showed that from 7106 patients with non-ST-elevation MI eligible for beta-blocker treatment, early beta-blocker therapy (within 24 hours of admission) was initiated in 76%; patients prescribed a beta-blocker were more commonly younger, and had a previous MI, hypertension or hyperlipidemia. 97 In RALES, patients likely to do well, young men with mild HF, were more likely to receive beta-blockers.

The time between MI and recruitment into the trial

The patients in EPHESUS were randomised between 3 and 14 days postMI. In RALES, patients had HF for at least 6 weeks, and in AREA IN-CHF for at least 3 months. Therefore, the ischaemic patients in RALES and AREA IN-CHF were further ahead in their postMI recovery and were a more stable chronic HF population compared with the EPHESUS population. Evidence shows that the risk of death and CV events is greatest in the first 24 to 48 hours after an MI. This declines over the subsequent 3 to 5 days, with survival curves becoming relatively flat after this period. 98 This would imply that the patients enrolled in EPHESUS would have an increased baseline risk of death and CV events early in the trial because of their recent history of MI, compared with the ischaemic subgroups of RALES and AREA IN-CHF.

The primary mode of action of aldosterone antagonists may differ between acute and chronic HF populations. Therefore, differences between the trials in terms of response may be attributed to the population being treated, rather than the drug administered. Reanalysis of the data from EPHESUS and RALES were reported in the FDA eplerenone medical review. 40 It was noted that approximately 66% of the difference in deaths between eplerenone and placebo occurred within the first 30 days with little continued divergence of the survival curves during the remainder of the trial [Figure 4(A)]. The reductions in deaths in patients taking eplerenone were considered to be primarily the result of a reduction in sudden death, with reductions in deaths due to recurrent MI and HF also prominent. In RALES, the survival curves did not start to diverge until approximately 3 months, with the divergence continuing over the course of the trial [Figure 4(B)]. The major contributors to the reduction in deaths with spironolactone were considered to be reductions in sudden death and the progression of HF.

FIGURE 4.

Cumulative mortality in (A) EPHESUS and (B) RALES. 40

The diagnosis of MI

In 2000, the European Society of Cardiology and the American College of Cardiology produced a consensus document that changed the way in which MI is diagnosed. A study comparing the patient diagnoses using the old and new systems showed that the new definition increased the number of events diagnosed as MI by approximately 26% (an 11% increase in the number of patients diagnosed with MI); two-thirds of the additional events diagnosed as MI having previously been diagnosed as unstable angina. 3 Therefore EPHESUS and AREA IN-CHF may have a number of patients diagnosed with MI that would not have received that diagnosis in RALES.

The severity of LVSD

The inclusion criteria relating to LVEF varied across the three trials. In RALES, patients had to have an LVEF of 35% or less, compared with 40% or less in EPHESUS, and 45% or less in AREA IN-CHF. A subgroup of patients with severe LSVD in EPHESUS with LVEF < 35% has been investigated. 52 Baseline LVEF was approximately 25% in RALES, 33% in EPHESUS and 38% for the ischaemic group from AREA IN-CHF; the range of LVEF in the ischaemic group from AREA IN-CHF was 16–58%, therefore some patients were recruited who did not have LVSD despite the inclusion criteria.

Other comparability issues

The proportion of patients on ACE inhibitors is similar in both trials; however, the proportion prescribed a diuretic (100% in RALES, 60.5% in EPHESUS and 63% in AREA IN-CHF) and aspirin (36.5% in RALES, 88.5% in EPHESUS and 71% in AREA IN-CHF) varied greatly. Digoxin is indicated for chronic HF; the proportion of patients prescribed digoxin was high in RALES (73.5%), and low in the AREA IN-CHF trial (18%) despite patients having HF for at least 3 months (digoxin use not reported for EPHESUS).

There has been a substantial increase in the use of percutaneous coronary interventions over recent years, making AREA IN-CHF the most likely to reflect current practice. 99

Some outcomes of interest in this review were described differently across the trials. Deaths and hospitalisations were reported as cardiac in RALES and AREA IN-CHF, whereas in EPHESUS they were cardiovascular, which is a broader term including non-cardiac and CV outcomes (such as stroke). As a consequence, the results for these outcomes are not comparable.

The mix of NYHA classifications of the populations varied across the three trials. In RALES most patients were NYHA class III, with the majority of the remaining patients class IV (only 0.5% were class II). In EPHESUS most patients were NYHA class II, with few patients in class IV; 28% were NYHA class I. In AREA IN-CHF, all patients were NYHA class II. The usefulness of this classification system when applied to hopsitalised patients is questionable, as much of the assessment is based on the ability to undertake physical activity and the exacerbation of symptoms, which are less easily assessed in hospitalised patients.

Summary

Although the drugs being evaluated could be considered pharmacologically exchangeable, the populations in which they have been studied are not. In terms of beta-blocker and percutaneous coronary intervention usage, RALES is not comparable to either EPHESUS or AREA IN-CHF, and RALES and EPHESUS are not comparable to current practice. The patients recruited into EPHESUS are early postMI, whereas in RALES and AREA IN-CHF the patients had their ishaemic event at least 6 weeks and 3 months before randomisation, respectively. The patients recruited into RALES were a more chronic HF population than in the other two trials, with HF newly developing in patients in EPHESUS who had only just experienced an MI. EPHESUS and AREA IN-CHF had patients with less severe HF compared with RALES in terms of LVSD and NYHA class. Most patients in RALES were NYHA class III, whereas in EPHESUS most patients were NYHA class II, with few patients in class IV and 28% in class I, and in AREA IN-CHF, all patients were NYHA class II. In addition, not only were there differences in the rate of beta-blocker usage, which was much higher in EPHESUS and AREA IN-CHF compared with RALES, there were likely to be differences in the types of beta-blockers prescribed, with an increase in the more effective third-generation beta-blockers over time.

Discussion of the clinical evaluation

The original purpose of this review was to systematically evaluate the evidence in order to assess the exchangeability of the aldosterone antagonists spironolactone and eplerenone in patients with postMI HF. Additionally, we aimed to determine whether further research comparing these two drugs in this indication is warranted and potentially cost-effective.

Review of Tilson et al. (2003). Cost-effectiveness of spironolactone in patients with severe heart failure

Review of Glick et al. (2002). Economic evaluation of the randomised aldactone evaluation study (RALES): treatment of patients with severe heart failure

Review of Weintraub et al. (2005). Cost-effectiveness of eplerenone compared with placebo in patients with myocardial infarction complicated by left ventricular dysfunction and heart failure

Discussion

The review of existing economic evidence identified three main published studies evaluating the cost-effectiveness of aldosterone antagonists as an adjunctive therapy to standard care, for either severe chronic HF or for patients with HF after AMI. None of these studies attempted to undertake a direct comparison of the cost-effectiveness of spironolactone versus eplerenone in either of these populations. Two of the three studies considered were based on the results from RALES and both were conducted before results from EPHESUS were available. 118,120 Consequently, the exclusion of eplerenone can be justified on the basis of data availability at the time the studies were conducted. The third study by Weintraub et al. 29 evaluated the cost-effectiveness of eplerenone compared with standard care alone based on the results from EPHESUS. Although spironolactone was acknowledged by the authors as a potentially relevant comparator, the lack of any direct evidence comparing eplerenone to spironolactone led them to conclude that any comparison would be speculative. The methods and approaches reported by Weintraub et al. 29 have since been used within a series of related publications reporting country-specific estimates of cost-effectiveness for eplerenone compared with standard care derived from applying country-specific sources of resource utilisation and costs. These individual studies were not considered within this review given the significant overlap with the earlier publication by Weintraub et al. 29 and the absence of a UK-specific analysis.

Despite the different approaches and assumptions employed, all of the studies consistently reported that the addition of an aldosterone antagonist to standard care for patients with either severe chronic HF or for patients with postMI HF appeared to be highly cost-effective compared with standard care alone. In addition, the results in each of the studies appeared relatively robust to a range of alternative assumptions or inputs. However, despite the consistent findings emerging from these studies, a number of important uncertainties remain surrounding the potential generalisability of these results to the UK NHS. Furthermore, the lack of existing evidence on the potential cost-effectiveness of eplerenone compared with spironolactone presents an important limitation to current decision-making.

Two of the three studies were based directly on the results of RALES and EPHESUS. 29,118 While the use of RCT data as the main source of evidence provides a high level of internal validity, the external validity and generalisability of these findings to clinical practice remains uncertain. The study by Tilson et al. 120 attempted to evaluate the potential impact of spironolactone within the context of current clinical practice by combining evidence on the relative effectiveness of spironolactone from RALES with data on the absolute event rates associated with standard care reported in an Irish setting. However, data for standard care were derived from a single acute teaching hospital in Ireland and the assumptions employed to extrapolate this data from a 1-year time horizon to a 10-year time horizon were not considered to have been sufficiently justified. Issues concerning the generalisability of the clinical data applied within existing studies to an NHS setting, combined with the application of non-UK sources to key resource utilisation and cost assumptions, meant that none of the three studies reported results that were considered to be directly applicable to the UK NHS.

Although one of the studies118 provided an evaluation of cost-effectiveness over a similar time horizon to the follow-up period reported in RALES, the extrapolation of the within-trial results to a longer time horizon was central to the other two studies. The extrapolation was considered necessary to appropriately quantify the estimate of LYG due to the mortality differences from the within-trial results. When the primary outcomes of interest for cost-effectiveness are expressed in terms of life-years or QALYs gained, the consequences of dying (i.e. the subsequent loss in life-years and QALYs) during the trial period can only be appropriately quantified by considering the remaining life expectancy for a patient who survives until the end of the trial follow-up period. Inevitably this requires the extrapolation of survival estimates over a longer time horizon and preferably over a lifetime time horizon if differences in LYG are to be fully captured. For this extrapolation, both of the studies used observational data sources relevant to the specific settings considered within each study. However, the extrapolation approach employed by Tilson et al. 120 was considered to be subject to a number of potential limitations and associated uncertainties arising from employing an arbitrary 10-year time horizon and using relatively short-term mortality data (12 months) from a single acute hospital as the basis for informing the longer-term mortality rates. In contrast, the approach employed by Weintraub et al. 29 used several large observational datasets with long-term follow-up and used these datasets to estimate the remaining life expectancy over a lifetime time horizon.

A potentially important limitation of the extrapolation approaches employed by both Tilson et al. 120 and Weintraub et al. 29 is that neither study attempted to project longer-term mortality differences because of non-fatal events arising during the trial period within the calculations used to estimate LYG and QALYs. Hence, the cost-effectiveness results reported in both studies may be considered potentially conservative estimates. Although the inclusion of this element is unlikely to alter the cost-effectiveness conclusions, it may have an important effect on the uncertainty surrounding the cost-effectiveness results, which could have important implications when it comes to estimating the costs of decision uncertainty using VOI approaches.

Although the extrapolation approach employed by Weintraub et al. 29 incorporates estimates of the additional costs and quality of life resulting from the additional life-years saved, it is important to note that this approach does not explicitly model the longer term prognosis of a patient in terms of the likelihood of their experiencing further non-fatal events over their remaining lifetime. Instead, the approach employed assumes that the differences in the average costs and quality of life reported for each treatment, at between 2- to 3-years postrandomisation within EPHESUS, will remain at this level over the longer term. However, if either the incidence of non-fatal events or the severity of HF increases, then the approach employed by Weintraub et al. 29 will potentially underestimate the costs and overestimate the quality of life assigned to the additional life-years saved, and the resulting cost-effectiveness estimates may not be reliable.

A final limitation of the existing cost-effectiveness evidence is that none of the studies have considered the potential cost-effectiveness of alternative treatment durations with aldosterone antagonists. Although different approaches and assumptions were applied within the studies for the purposes of extrapolating the effectiveness data, all of the studies assumed that treatment with either spironolactone or eplerenone would not be continued beyond the follow-up period of the trials.

Although several potential limitations have been identified concerning the current cost-effectiveness evidence for aldosterone antagonists, there are two particular issues that are likely to represent major challenges to ensuring current NHS decision-making: (1) the lack of prior cost-effectiveness evidence for either eplerenone or spironolactone from a UK NHS perspective and (2) the absence of existing studies comparing the cost-effectiveness of eplerenone and spironolactone directly. Although epleronone and spironolactone are clearly relevant comparators for the treatment of postMI HF from an economic perspective, there are real concerns about the feasibility of such a comparison because of the limited data on spironolactone in this specific indication, and the lack of clinical data directly comparing the two drugs. 29,33,35

The next chapter presents the methods and results of a new decision analytic model that has been specifically developed to address both of these limitations more formally. Central to this model is the need to provide estimates that are more relevant to the UK NHS and the feasibility of undertaking a direct comparison between the alternative aldosterone antagonists.

Short-term model

The short-term model is structured as a Markov model131 as shown in Figure 5. Monthly cycles were used to reflect the events that occur in each of the first 3 months of the acute postMI period. The primary events of interest were all-cause mortality and hospitalisations for CV causes. Four mutually exclusive health states are defined: (1) index hospitalisation, (2) non-fatal CV event, (3) CV death, and (4) non-CV death (not shown on figure). Patients enter the index hospitalisation state at the point at which they have an AMI and remain in this state until they experience either a CV event (fatal or non-fatal) or non-CV death. These outcomes also represent states in the long-term model. The probabilities of the different end points during the acute postMI period are used to estimate the proportion of patients starting in the health states in the long-term model.

FIGURE 5.

Structure of the short-term model.

Long-term model

Any assessment of the cost-effectiveness of aldosterone antagonists must allow for the long-term cost and outcome implications of treatment. This ‘extrapolation’ is needed for two reasons. First, many patients who are treated for postMI HF will continue to consume health-service resources for the remainder of their life, and the effectiveness of aldosterone antagonists in the short-term may influence these costs. Second, to compare the cost-effectiveness of aldosterone antagonists with other uses of health-service resources (inside and outside cardiology), it is necessary to express the benefits of the drugs in terms of a generic measure of health gain that can be compared across treatment areas. The most frequently used generic measure for this purpose is the QALY. To provide a realistic estimate of the QALY impact of aldosterone antagonists, the long-term implications for survival and HRQoL need to be modelled.

The long-term model (Figure 6) is structured as a Markov model with yearly cycles used to model the progression of postMI HF over the longer term. Five health states are defined: (1) postMI HF (i.e. no additional CV events have incurred since the index hospitalisation), (2) 1st year postCV event, (3) subsequent years postCV event, (4) CV death, and (5) non-CV death (not shown on figure). Depending on progress through the short-term model, patients enter the model in the postMI HF state, the 1st year postCV event state, or the death states (CV or non-CV). The proportion of patients starting in each of these states will differ between the treatment strategies depending on their relative effectiveness during the acute period.

FIGURE 6.

Structure of the long-term model.

Patients who do not experience any events in the short term enter the long-term model in the postMI HF state. Individuals remain in this state until they experience a further event either CV in nature (e.g. AMI, progression of HF, stroke, ventricular arrhythmia, or CV death) or non-CV death. Patients who experience a fatal event exit the model. Patients who experience a non-fatal CV event, enter the ‘1st year postCV event’ state for 1 year. Once that year has elapsed they can experience another non-fatal CV event and re-enter the 1st year postCV event state for another year; have no additional events; or die and exit the model. Those patients who do not experience additional events within the Markov cycle move to the subsequent years postCV event state, where they remain until they experience another event (nonfatal CV event, CV death, or non-CV death).

The period after a non-fatal CV event is split into ‘1st year postevent’ and ‘subsequent years postevent’ to reflect the elevated risk of a subsequent event in the initial year following an event. Additional tunnel states (not shown in the figure) were also incorporated into the model to keep track of the number of rehospitalisations for recurrent CV events. Tunnel states are arranged so that they overcome the ‘memoryless’ feature of Markov models regarding where the patient has come from or the timing of the transition. It was necessary to record the number of rehospitalisations for CV events to correctly assign unit costs and quality of life adjustments to the period of these events (see Resource use and unit costs). A maximum of three recurrent CV events (four events in total) since the initial hospitalisation for MI was allowed because the utility estimates applied in the model relate to three or more rehospitalisations. Although the likelihood of having five or more CV events is not fully explored the difference in the results would be expected to be negligible.

From any of the states in the model where individuals are alive, there is a competing risk of a non-CV death.

Baseline events rates

The RCTs undertaken to evaluate the clinical effectiveness of eplerenone and spironolactone were mainly or wholly undertaken outside the UK. In many respects, treatment patterns and resource use in the UK can be expected to differ from those in the centres involved in the trials. One implication of this UK-specific practice is that the baseline event rates observed in the control group of the trials are unlikely to provide reliable estimates for UK practice. In addition, the limited follow-up reported in both EPHESUS and RALES and the generalisability of RALES to a postMI HF population, raise additional issues regarding their suitability for the purposes of long-term extrapolation.

Long-term survival from EPHESUS and RALES was estimated by fitting parametric curves to the published empirical Kaplan–Meier survival data to predict the survival of individuals beyond the published follow-up. The trials provide survival curves for all-cause mortality up to a period of 27 months in EPHESUS and 33 months in RALES. A parametric Weibull distribution was fitted to the data to predict survival over a patient’s lifetime, as shown in Figure 7. The difference in survival between EPHESUS and RALES is largely the result of the differences in the trials regarding the time since AMI and the severity of HF. The EPHESUS population is immediately postMI (up to 14 days) and at the start of HF development (approximately 80% NYHA class I or II), whereas the RALES population is further into the progression of HF (NYHA class III or IV), and members of the population with ischaemic HF were at least 6 weeks postMI. It is also worth noting that the longer-term extrapolation of EPHESUS appears to underestimate mortality in the long term with over 50% of patients still predicted to be alive after 20 years.

FIGURE 7.

Extrapolated survival curves based on EPHESUS and RALES.

Given the concerns noted, the use of baseline data from EPHESUS or RALES was not considered appropriate for informing the long-term extrapolation. Instead, an alternative source of baseline event rates, specific for UK practice, was obtained from the linked Scottish Morbidity Record (SMR), which records all hospitalisations and subsequent deaths in Scotland. 132 These data are collated by the Information and Statistics Division, Scotland. Record linkage, using probability matching (with an accuracy of approximately interquartile range 98%), permits analysis of data at the level of the individual patient as well as the episode of care.

For this study, data from the SMR was obtained on all individuals with a ‘first’ discharge from hospital, between 1993 and 2003, with a principal diagnosis of HF [International Classification of Diseases (ICD) 9 425.4, 425.5, 425.9, 428.0, 428.1, 428.9, 402, ICD10 I50, I42.0, I42.6, I42.7, I42.9]. 133 A ‘first’ discharge was defined as one with an HF code in a primary diagnostic position, with no previous hospitalisation for HF (in any diagnostic position) for a minimum of 5 years before the index admission. For these individuals with a ‘first’ discharge for HF, data were also available on subsequent major CV events (MI, stroke, angina and other HF events) as well as deaths from CV and non-CV causes, until 31 December 2005.

Data on comorbid diagnoses (coded as a secondary diagnosis during the index hospitalisation or as the principal diagnosis during a previous hospitalisation within 5 years of the index hospitalisation) were also included. Data on MI as a comorbid diagnosis was used as a means of obtaining estimates for a postMI cohort. However, since the timing of the previous AMI was not recorded in the SMR data, it was considered that the SMR data might not adequately capture the initial risk of the acute episode. Given this concern, it was considered more appropriate to use the SMR data as part of the longer-term extrapolation and to use data from the control group of EPHESUS to characterise the initial acute period of 3 months.

A subset of the SMR data was used representing 39,307 individuals out of the overall dataset of 71,848. This subset of individuals has recently been used to estimate the progression of HF in patients who survived the initial episode and had stable HF. 134 Of the sample of 39,307 individuals, 23,150 (59%) had a primary non-fatal CV event, and 7509 (19%) had a fatal CV event as the primary (first) event. For those with non-fatal primary events, 6233 went on to have a subsequent non-fatal event, and 13,662 died of a subsequent CV event. The extrapolation is therefore based on a large and robust dataset. Figure 8 provides a comparison of the extrapolated survival curve from the SMR data compared with those estimates from EPHESUS and RALES.

FIGURE 8.

Extrapolated survival curves from Scottish Morbidity Register data.

Risk of non-cardiovascular mortality

The risk of dying from CV causes is described above using risk equations, but there is a competing risk of non-CV mortality. The age-dependent risk of other-cause mortality was estimated using UK age-specific and sex-specific mortality rates. 136 These rates were adjusted to exclude those deaths pertaining to CV mortality using a cause elimination approach. This involved eliminating deaths caused by CV disease (ICD10 code) from the UK lifetables according to standard methods used by the Office for National Statistics. 137

Relative treatment effect

In the absence of any direct head-to-head trials comparing aldosterone antagonists simultaneously in postMI HF, indirect evidence is required to establish the effectiveness of the interventions. The effect of spironolactone or eplerenone, when added to standard care, has been evaluated in RCTs with standard therapy as a control. The review reported in Chapter 3 identified a number of issues with regard to exchangeability of the trial evidence; heterogeneity between the trials with respect to different study populations (postMI HF versus HF more generally), timing of MI, beta-blocker usage and severity of HF, which made any formal ‘pooling’ of the individual trials inappropriate. A recent systematic review by Ezekowitz and McAlister138 that included studies of aldosterone antagonists in postMI without clinical HF as well as those in postMI HF and general HF, reinforces the concerns expressed in Chapter 3 that any pooling across the different populations (i.e. postMI HF and HF) would not be appropriate because of heterogeneity. Although the Ezekowitz and McAlister review138 has limitations in that many of the included trials were small, of poor quality and reported few (in some cases zero) events, it does show that pooling across studies within these populations may be less problematic given a lack of significant heterogeneity within the populations. In the context of the decision model the combined weight of evidence from all relevant trials and comparators is necessary to provide a useful aid to decision-making. To facilitate an indirect comparison of eplerenone and spironolactone, there was a need to extend the evidence ‘network’ considered in Chapter 3 to overcome the issues regarding exchangeability of trials and populations. The review by Ezekowitz and McAlister138 was important in this respect for two reasons.

• It provided additional RCT evidence reporting on the use of aldosterone antagonists (including canrenoate) in different populations. Although many of these studies were small, reported few events, were of unknown quality, and did not meet the inclusion criteria for the review of clinical effectiveness in Chapter 3, they provided evidence on the use of aldosterone antagonists for the treatment of MI and HF.

• It showed that there was no evidence of statistical heterogeneity within the different study populations despite differences that exist between the populations.

The additional evidence reported by Ezekowitz and McAlister138 was used to complement the RCT evidence reported in Chapter 3 to obtain a wider network of evidence on the use of aldosterone antagonists in different populations (including those beyond the scope of the decision problem). This wider network of evidence was used to facilitate an indirect comparison of eplerenone and spironolactone by examining the relationships that exist between the different treatments and study populations. A summary of the network of evidence for all-cause mortality is reported in Table 16.

A Bayesian evidence synthesis approach was employed which draws on the relationships that exist between treatments and populations while preserving differences that exist across populations. The approach essentially assumes that there is a correlation between the relative effectiveness of a treatment (spironolactone, eplerenone or canrenoate) in one study population (e.g. postMI HF) and another population (e.g. HF more generally). A metaregression analysis was used to examine the relationship between the treatments and the size of the treatment effect in the two populations (postMI HF and general HF) identified in the network of trials reported in Table 16. Distinct indicator variables were incorporated in the regression for the two populations (e.g. postMI HF assumes a value of 1, while general HF assumes a value of 0) and treatments to allow an estimate of the increment in relative risk for a particular treatment in a particular study population.

Formally, the regression takes the form:

where MI assumes the value 1 for postMI HF and 0 for general HF, EPL and CAN assume the value 1 for treatment with eplerenone and canrenoate, respectively, or 0 for treatment with spironolactone. The coefficient β₀ refers to the reference category, which estimates the log of the relative risk for treatment with spironolactone in general HF. The coefficient β₁ estimates the increment in relative risk for postMI HF relative to general HF. Similarly, the coefficients β₂ and β₃ estimate the increment in relative risk for treatment with eplerenone and canrenoate, respectively, relative to spironolactone. Details of the winbugs code are reported in Appendix 7. A fixed effects model was used in the analysis. A random effects model was investigated but resulted in problems with convergence because of the small amount of data available across the different populations and treatments.

The metaregression approach allows treatment specific estimates to be modelled in each population by drawing strength from the network of evidence, and assuming that all treatments share a similar reduction/improvement in efficacy in one population compared with another. The main advantage of this approach is that it provides predictions on efficacy for the treatments in the populations where there are limited or no data. For example, there are limited data to estimate the effect of spironolactone in postMI HF, or eplerenone in general HF, therefore informed predictions can be made by observing the relationships that exist between the treatments and between the populations. The inclusion of the canrenoate trials is essential to this synthesis because they provide sizeable studies in both populations.

The primary outcome reported across the trials is mortality. The relative risks with 95% credibility intervals (CrI) for spironolactone and eplerenone compared with standard care are shown in Table 17 for postMI HF and general HF. The estimates for the treatments in the populations where the evidence is more certain, i.e. eplerenone in postMI HF and spironolactone in general HF, are consistent with EPHESUS and RALES, respectively. The predictions for the treatments in the populations where there is a paucity of evidence (spironolactone in postMI HF and eplerenone in general HF) are more uncertain with wide 95% CrI.

The estimates for postMI HF are used within the model because this is the population of interest to the decision problem. The results, on average, suggest that eplerenone is associated with a significant reduction in mortality compared with spironolactone; however, there is high uncertainty around the estimate for spironolactone (95% CrI 0.575 to 1.652). The simulated output (10,000 Markov Chain Monte Carlo simulations implemented in winbugs139) is used directly in the model to maintain correlation between the estimates for the separate treatment strategies.

A secondary outcome of hospitalisations for CV events was not reported consistently across trials. Therefore meta-analytic approaches could not be used to derive a relative effect measure for this outcome. The relative risk of hospitalisation for CV events for eplerenone compared with standard care was informed by EPHESUS. Over a mean follow-up time of 16 months, the relative risk was 0.91 (95% CI 0.81 to 1.01). Within the model, this relative risk was incorporated as a log-normal distribution to allow for uncertainty in the parameter. The relative risk was assumed to remain the same over the individual’s lifetime. No information was available to inform the effect of spironolactone on the risk of hospitalisation for CV causes in postMI HF, therefore an assumption was required to estimate this measure. The model assumes that the increment in relative risk observed for mortality between eplerenone and spironolactone can be applied to hospitalisations. This relative increment was then applied to the relative risk of hospitalisations for eplerenone to obtain an estimate of the risk of hospitalisations for spironolactone.

The relative treatment effect measure for mortality and hospitalisations for CV causes is applied to the baseline event rates (see Baseline events rates) to obtain absolute event rates for the different treatment comparators. In separate scenarios the robustness of the results are explored by changing the network of evidence available to inform the evidence synthesis. Three scenarios are considered: (1) exclusion of HF trials reporting follow-up durations of < 3 months; (2) excluding canrenone studies; and (3) excluding lowest quality trials.

Resource use and unit costs

Costs were incorporated into the Markov model by attaching a unit cost to each non-fatal CV event (AMI, HF, stroke, ventricular arrhythmia) which occurs over time. The 1st year postCV event state in the model represents three tunnel states, which record the number of rehospitalisations for CV events. Once a patient enters one of these states, it is assumed that there is a proportional risk that the event which occurred was AMI, HF, stroke or ventricular arrhythmia. The proportion of patients experiencing each of these events was informed by EPHESUS: 31.6% AMI, 53.9% HF, 7% stroke and 7.5% ventricular arrhythmia. The unit costs attached to each event were based on the National Schedule of Reference Costs 2007–08. 151 The reference costs are the average costs to the NHS of providing a defined service/resource in a given financial year. The costs are categorised into particular groups (Health Resource Groups; HRGs) according to episodes that are clinically coherent and consume similar resources. Table 18 presents the non-elective inpatient HRG costs for the CV events of interest in the model. The costs for each event are based on a weighted average of short-stay and long-stay visits, as well as the numbers of patients that incur complications. An annual background cost of HF relating to the routine management of HF, excluding event costs, was also applied for the length of time that a patient is alive within the model. This cost was based on British Heart Foundation statistics, which estimate that the annual cost of HF to the NHS was £628.6M in 2000. 25 Excluding inpatient hospitalisations, the annual cost is £250M. Assuming a prevalence of HF of 707,000 in the UK. 26 This equates to an annual background cost of HF of £354 per person (2000 prices). The price was uprated to the current year to give an annual cost of £462 per person per year alive.

The cost of spironolactone and eplerenone were obtained from the BNF. 42 The dosage of eplerenone was assumed to be 25 mg/day for the first 4 weeks and 50 mg thereafter. A 28-tab pack of 50-mg strength costs £42.72. This resulted in an annual cost of £557 incurred for each year that the patient continues to receive the drug. The dosage of spironolactone was assumed to be 100 mg/day. A 28-tab pack of 100-mg strength of generic spironolactone costs £3.55. This resulted in an annual cost of £46 for the duration of treatment. All patients, regardless of the treatment strategy were assumed to receive standard therapy.

Quality of life

In order to estimate QALYs, it is necessary to quality-adjust the period of time for which the average patient is alive within the model using an appropriate utility or preference score. Ideally, utility data are required to differentiate between the health status of patients in the different states of the Markov model using a generic measure of health such as the EQ-5D. 152 A recent study by Gohler et al. 153 addresses the limitations of currently available data with respect to providing utility estimates that can be linked to common proxy health states in HF. The study uses primary data from EPHESUS to estimate utility values for patients with HF according to NYHA classification and number of CV rehospitalisations. This study is therefore directly applicable to the current model. The 1st year postCV event state in the model represents three tunnel states, which record the number of rehospitalisations for CV events. A utility decrement, based on the Gohler et al. study,153 is applied to each of these rehospitalisations (Table 19) and a gamma distribution is used to characterise uncertainty in the mean values. The underlying baseline utility for a 64-year-old patient with postMI HF was also taken from Gohler et al. 153 (Table 19). To reflect the decreasing utility of patients as they age through the model, age and sex adjusted norms for the UK154 were adjusted downwards by approximately 5% to reflect the existence of HF. The ‘adjustment’ factor was estimated by comparing the baseline utility of Gohler et al. 153 to the average utility of a 64-year-old person in the UK, derived from a nationally representative UK sample using EQ-5D. 154

Cost-effectiveness

The model was developed in excel and is run probabilistically for 10,000 iterations incorporating all the estimates and assumptions described previously. Mean costs and QALYs for spironolactone, eplerenone and standard care alone are presented and the cost-effectiveness of each is compared using conventional decision rules, estimating ICERs as appropriate. 155 The ICER examines the additional costs that one strategy incurs over another and compares this with the additional benefits. When more than two interventions are being compared the ICERs are calculated using the following process:

• The strategies are ranked in terms of cost (least expensive to most costly).

• If a strategy is more expensive and less effective than any previous strategy, then this strategy is said to be dominated and is excluded from the calculation of the ICERs.

• The ICERs are calculated for each successive alternative, from the cheapest to the most costly. If the ICER for a given strategy is higher than that of any more effective strategy, then this strategy is ruled out on the basis of extended dominance.

• Finally, the ICERs are recalculated excluding any strategies that are ruled out by principles of dominance or extended dominance.

Uncertainty in the estimates of cost-effectiveness of the alternative strategies is reflected using cost-effectiveness acceptability curves (CEACs). 156 These show the probability that each strategy is more cost-effective than the others using alternative values for the maximum value that the health service is willing to pay for an additional QALY in these patients.

Value of information

Value of information analysis is used to quantify the cost associated with the decision uncertainty related to the use of aldosterone antagonists for postMI HF. This approach provides an explicit and quantitative method to value the consequences of decision uncertainty based on existing evidence. 157,158 The analysis can be used as the basis to inform future research priorities and can assist in establishing the potential value of a future head-to-head trial of spironolactone and epelerenone.

Using VOI approaches, the expected cost of uncertainty surrounding the adoption decision can be determined by the expected value of perfect information (EVPI). 159 The expected costs of decision uncertainty represent the consequences in terms of costs incurred and lost benefits that would have occurred should it later transpire that the optimal decision on cost-effectiveness grounds was not correct. The VOI analysis therefore involves establishing the difference between the expected value of a decision made on the basis of existing evidence and, following the arrival of further information and the subsequent resolution of uncertainty in the estimates, the expected value of a decision made on the basis of new evidence. The EVPI provides an estimate of the value of completely resolving all existing uncertainty, through the provision of perfect information, and provides a measure of the maximum return to further research. Consequently EVPI represents an upper bound to the amount a decision-maker should be willing to pay for additional evidence to support the current decision based on expected cost-effectiveness estimates. That is, if the EVPI exceeds the expected costs of additional research then it is potentially cost-effective to acquire more information by undertaking this research. 159

The overall EVPI surrounding a health-care policy decision depends upon the number of times that the decision is faced over the lifetime of the technology. This is termed the population level EVPI and is estimated by scaling up the individual (per-patient) EVPI estimates according to the number of people that would be affected by the information over the anticipated lifetime of the technology. Formally this can be expressed as:

where I is the incidence in the period, t is the period, T is the total number of periods for which information from research would be useful and r is the discount rate.

Based on published statistics it has been estimated that postMI HF accounts for approximately 20% of all HF cases. 21 The British Heart Foundation estimates that the prevalence of HF in the UK is 707,000 (393,000 men, 314,000 women) and the annual incidence of HF is estimated to be 68,000 cases (38,000 men, 30,000 women). 26 Together these sources suggest a prevalence of about 141,400 and an annual incidence of around 13,600 for postMI HF. Assuming a 10-year time horizon for the lifetime of the technologies being considered, this gives an effective population of 258,465 (prevalent population plus incident population per annum discounted at 3.5% over the lifetime of the technology) for the population EVPI calculations.

In addition to determining the EVPI surrounding the decision as a whole, VOI approaches can also be used for particular elements of the decision to direct and focus research towards the areas where the elimination of uncertainty has the most value. The partial EVPI can be calculated for individual or subsets of parameters.

On the basis of EVPI and partial EVPI calculations, the potential value of a future trial or other potential research designs can be evaluated. The VOI which could be acquired by conducting further research depends crucially on the number of future patients who could benefit from it and the time horizon over which the information would be useful. It has recently been shown that selecting a value for the time horizon is essentially a proxy for a more complex and uncertain process of future changes to the decision problem which impact on the EVPI. 160 One future change that can be anticipated in the short-term is the potential impact of the arrival of a generic version of epelerenone. The potential impact on EVPI estimates of this change was therefore an important consideration.

Scenarios

Cost-effectiveness and EVPI results are presented for a base-case analysis and a number of separate scenarios. All scenarios consider a cohort of postMI HF patients (mean age 65 years) over a time horizon of 40 years. The base-case analysis assumes that treatment with aldosterone antagonists will be given for a maximum of 2 years to ensure consistency with the available trial evidence for these treatments. The relative treatment effects for mortality and non-fatal events, derived from results of the evidence synthesis for spironolactone and eplerenone compared with standard care, are applied to the baseline risk prediction equations for the first 2 years only. After 2 years all patients share the same set of common risks from each health state in the decision model, regardless of their initial treatment assignment. However, since each strategy in the model will result in a different proportion of patients who reside in each of the separate health states of the model at 2 years, this means that the model will continue to account for the longer-term prognostic benefit associated with a reduction in non-fatal events incurred during the initial 2 year treatment period.

In addition to the base-case analysis a range of separate scenarios are explored to examine the robustness of these results to a range of alternative assumptions and parameter inputs. These include:

Cost-effectiveness and EVPI results

The results are presented for the base-case anaysis and for each of the separate scenarios. In each case, mean lifetime costs and QALYs are estimated for each strategy and ICERs are presented where appropriate. The probability that each intervention is cost-effective is also reported for three select values of the cost-effectiveness threshold representing a maximum willingness to pay between £10,000 and £30,000 per QALY. The value of the uncertainty surrounding a decision based on the expected ICER estimates is expressed using both individual (per-patient) estimates of EVPI and population EVPI assuming a 10-year time horizon.

Scenario 1: Lifetime treatment duration

The base-case analysis assumes that treatment with eplerenone and spironolactone will be discontinued after 2 years. This assumption was employed to be consistent with the average duration of follow-up reported in EPHESUS. In the absence of longer-term follow-up data, the effectiveness of continuing to treat patients with aldosterone antagonists over a longer time horizon remains highly uncertain. However, clearly the optimal duration of treatment with either spironolactone or eplerenone is an important issue. For chronic diseases, such as postMI HF, patients may continue to face an elevated risk of major clinical events for the remainder of their lifetime. The potential cost-effectiveness of maintaining treatment with aldosterone antagonists over the longer-term should therefore be explored. Thus, a separate scenario was undertaken to evaluate the potential cost-effectiveness of continuing treatment with aldosterone antagonists for the remainder of a patient’s lifetime. In this scenario the relative treatment effects applied to fatal and non-fatal events for spironolactone and eplerenone compared with standard care within the initial 2-year period were also applied to subsequent years.

Table 22 presents the cost-effectiveness results for the scenario assuming lifetime treatment with aldosterone antagonists. In common with the base-case scenario, the relevant ICER calculation was based on a comparison of eplerenone versus standard care. The ICER for lifetime treatment was £7893 per QALY compared with £4457 per QALY based on a treatment duration of 2 years. Hence, although the mean ICER of eplerenone compared with standard care becomes less favourable over a lifetime duration, the ICER estimate still remains well below conventional cost-effectiveness thresholds considered to represent value for money in the NHS.

In contrast to the base-case scenario, spironolactone was not dominated in the lifetime treatment duration scenario by standard care. Instead spironolactone was now extendedly dominated by eplerenone. The different findings between these scenarios may appear potentially counterintuitive because the same relative effect is being applied to spironolactone in both the base-case and lifetime treatment scenarios. However, the results clearly demonstrate the non-linear relationship between the model inputs and outputs. That is, although the mean of the relative risk estimates assigned to spironolactone are slightly higher than one (i.e. indicating that, on average, spironolactone is less effective than routine care) there is considerable uncertainty surrounding this estimate. As the model is non-linear, the simulations in which spironolactone is estimated to be more effective than standard care has a greater subsequent impact on the mean outputs (i.e. mean costs and QALYs) of the model in the lifetime treatment scenario than those in which spironolactone is estimated to be less effective. Hence, while the mean relative risk of spironolactone compared with standard care exceeds the value one, the expected values for QALYs derived from averaging across the entire 10,000 simulation are higher for spironolactone because of the non-linear relationship that exists in the model.

Although the lifetime treatment duration assumption and the non-linear relationship between the model inputs and outputs do not alter the choice of strategy on cost-effectiveness grounds (i.e. eplerenone appears to be the most cost-effective strategy in both scenarios), these factors do appear to have marked effect on the decision uncertainty and the costs of this uncertainty. At a threshold of £10,000 per QALY, the probability that eplerenone is cost-effective is only marginally higher than the estimate for spironolactone. Although the probability that eplerenone is cost-effective increases at higher thresholds, the results are less certain than in the base-case scenario. Figure 10 reports the CEACs assuming a lifetime duration which demonstrates higher uncertainty across the range of cost-effectiveness thresholds considered.

FIGURE 10.

Cost-effectiveness acceptability curves (lifetime treatment duration).

The higher decision uncertainty in the lifetime treatment duration is reflected in the higher EVPI estimates for this scenario reported in Table 23. These estimates are between two and three times higher than the comparable estimates reported for a 2-year treatment duration. The estimates of total EVPI range between £820M and £1748M. The estimates reinforce the conclusions based on the base-case scenario that there appears to be significant value to the NHS in undertaking additional research to reduce the existing decision uncertainty.

Scenario 2: External calibration

Both the base-case and lifetime treatment scenarios are based on the risk prediction equations derived from the SMR data used to estimate the baseline event rates assigned to standard care after the initial acute episode. Although this approach was considered to provide a number of advantages for informing NHS decision-making, compared with using the event rates reported in the control arm from EPHESUS, there remain a number of potential uncertainties with this approach. First, although covariate adjustment based on the year of the cohort provided an approach to ensure that the risk predictions reflected the event rates for the most recent cohort possible (year 2002), important changes may have occurred within standard care since this time. Second, the identification of patients in the SMR dataset required patients to be hospitalised for their HF. Although information on the underlying severity of HF was not reported in the SMR dataset, it is plausible that this cohort may have more severe HF than the patients recruited within EPHESUS. To examine the robustness of the results to these sources of uncertainty, an attempt was made to calibrate and subsequently adjust the risk prediction equations applied to mortality using external sources.

The approach to calibration and subsequent adjustment of the mortality rates assigned to standard care was based on data from the Seattle Heart Failure Model. 161 This model, developed at the University of Washington, can be used to estimate survival rates up to 5 years after the diagnosis of HF. The risk calculator derived from the model can also be used to estimate survival rates according to particular patient characteristics and current medical management. Five-year mortality rates were therefore estimated from the Seattle model for a cohort of patients matched according to the baseline characteristics reported in EPHESUS and assumed to be receiving optimal medical management (excluding aldosterone antagonists). The 5-year mortality rates were approximately 30% lower than those derived from the SMR risk prediction equations. An adjustment was therefore applied to the SMR mortality prediction equations to provide comparable 5-year predictions.

The cost-effectiveness and EVPI results, based on the adjustments applied to the SMR mortality predictions are reported in Table 24 and Table 25, assuming a maximum treatment duration of 2 years. Although the adjusted estimates had an important effect on the absolute QALY estimates, both the ICER and EVPI estimates appeared robust to the adjustment applied. The ICER of eplerenone compared with standard care was £5010 per QALY compared with £4457 per QALY reported in the base-case scenario, which used the SMR risk prediction equations directly. The EVPI results did not appear significantly altered by the adjustment.

Table 26 and Table 27 report the cost-effectiveness and EVPI results assuming a lifetime treatment duration. Again both sets of results appeared robust to the adjustment applied, resulting in similar conclusions regardless of whether the baseline mortality rates were adjusted or not.

Scenario 3: Generic eplerenone

Although not currently available in the UK NHS, the arrival of a generic version of eplerenone appears imminent. The potential lower price of a generic version may have important implications for the cost-effectiveness and VOI estimates. In the absence of a current price, the price of generic eplerenone is subject to uncertainty. In general, generic prices are highly dependent on the policy environment162 and competition. Impact factors may include the following:163,164

• average revenue per brand name extended unit

• number of extended units sold before patent loss

• age of market in terms of time the brand-name product was sold

• time since the patent expired

• average revenue per generic extended unit.

Given that generic competition can be expected for eplerenone, three alternative scenarios were considered: (1) a price of 25% of the branded version, corresponding with the ratio of the average price of generic to branded drugs in the UK, (2) a price of 50% of the branded version, and (3) a price of 75% of the branded version.

The cost-effectiveness and EVPI results based on a 25% reduction are reported in Tables 28–31 for a 2-year treatment duration and lifetime treatment duration. Applying a 25% price reduction did not alter the ordering of the strategies for the ICER calculations and spironolactone was either dominated (2-year treatment duration) or extendedly dominated (lifetime treatment duration). The ICER of eplerenone compared with standard care alone was £3257 per QALY (2-year treatment duration) and £6080 per QALY (lifetime treatment duration). The corresponding ICERs based on the current branded price were £4457 per QALY and £7893 per QALY. As expected, the reduction in price improved the cost-effectiveness estimates for eplerenone in both scenarios, reinforcing the conclusion that eplerenone appears the most cost-effective strategy. The improved ICER estimates also resulted in a reduction in the uncertainty surrounding the decision itself and the subsequent EVPI estimates. However, the reduction in uncertainty appeared relatively marginal with significant uncertainty still apparent between eplerenone and spironolactone. The EVPI estimates ranged from £254M to £677M and from £699M to £1658M, across the cost-effectiveness thresholds considered assuming a 2-year and lifetime treatment duration, respectively.

The cost-effectiveness and EVPI results based on a 50% reduction are reported in Tables 32–35 for a 2-year treatment duration and lifetime treatment duration. Applying a higher price reduction resulted in further improvements in the ICER for eplerenone compared with standard care, although significant decision uncertainty still remained.

Given that there could be substantial generic competition for eplerenone, the cost-effectiveness and EVPI results based on a 75% reduction are reported in Tables 36–39 for a 2-year treatment duration and lifetime treatment duration. Again, the higher price reduction resulted in improvements in the ICER for eplerenone compared with standard care but significant decision uncertainty remains. A change in the cost of eplerenone has only a small effect on EVPI relative to a change in the effectiveness parameters.

Scenario 4: Network of evidence

Given the methodological uncertainty surrounding the evidence synthesis approaches applied in the main analyses, a range of alternative approaches were considered and the robustness of the results was explored. In particular, the evidence synthesis undertaken to inform the decision model incorporated a range of trials that were not considered within the main clinical effectiveness review. Although it was considered necessary within the modelling framework to link to a wider evidence network to address the problems noted in the clinical effectiveness review of making indirect comparisons based on RALES and EPHESUS, it should be recognised that this approach incorporated a number of additional studies that were beyond the specific patient population of interest, and had not been quality assessed or did not meet the inclusion criteria for the clinical review in Chapter 3. As a result, it is important to consider the robustness of the evidence synthesis to the inclusion of these additional studies.

The first scenario considered was the exclusion of HF trials reporting follow-up durations of < 3 months. As noted in Chapter 3, the survival curves in RALES did not start to diverge until approximately 3 months, with the divergence continuing over the course of the trial. Consequently, the inclusion of HF studies with short-term follow-up could potentially introduce a source of potential bias into the wider evidence network used to inform the decision model.

Tables 40–43 report the cost-effectiveness and EVPI results, excluding these trials, for a 2-year treatment duration and lifetime treatment duration. The results demonstrate that the results appeared robust to exclusion of HF trials reporting follow-up durations of < 3 months. Estimates of the ICER of eplerenone compared with routine care were very close to those obtained by including these studies. However, spironolactone was no longer dominated for the lifetime treatment duration scenario. In this scenario the ICER for spironolactone was £3684 per QALY compared with standard care. As spironolactone was not excluded from the ICER calculations, the ICER for eplerenone was now compared with spironolactone. The ICER for eplerenone was £8962 per QALY compared with spironolactone, demonstrating that eplerenone remained the most cost-effective strategy based on conventional cost-effectiveness thresholds. These results demonstrate the robustness of the main synthesis and provide an indication that the results from the shorter-term studies were not inconsistent with the other studies included in the wider network of evidence.

Although the exclusion of these trials did not alter the cost-effectiveness conclusions, there was marginally higher uncertainty surrounding this decision. The higher uncertainty translated into higher EVPI estimates. This provides an indication of the value derived from linking to the wider evidence network in the overall approach in terms of reducing the overall uncertainty surrounding the decision of interest.

Another important issue is the inclusion of canrenone studies in the overall network of evidence used to inform the cost-effectiveness analysis. As noted in Chapter 1 (see Canrenone), neither canrenone nor potassium canrenoate are licensed as human medicines in the UK. However, the wider network of evidence incorporated trial evidence for canrenone to strengthen the overall evidence base considered for the comparison of spironolactone and eplerenone. Clearly it is possible that the relationships assumed between the different drugs and the populations being considered may not be appropriate for each of the separate drugs. The inclusion of canrenone studies could introduce another source of potential bias into the synthesis results so the impact of excluding these studies was considered.

Tables 44–47 report the cost-effectiveness and EVPI results excluding the canrenone trials for a 2-year treatment duration and lifetime treatment duration. The impact of excluding these trials was similar to that of excluding the shorter-term trials in HF. That is, while the ICER results appeared robust to the exclusion of this evidence, the decision uncertainty increased accordingly. In addition to demonstrating the robustness of the main synthesis, the analysis provides further evidence that the canrenone studies did not appear to be inconsistent with the wider set of studies included in the overall network and that their inclusion was important in terms of reducing the overall uncertainty.

The final scenario considered, restricted the evidence network to the four trials that were considered to be of highest overall quality as determined by the clinical effectiveness review. 27,31,54,149 Tables 48–51 report the cost-effectiveness and EVPI results including only these four trials for a 2-year treatment duration and lifetime treatment duration. In contrast to the previous analyses, spironolactone was no longer ruled out either by dominance or by extended dominance in either the 2-year or the lifetime scenarios. Assuming a 2-year treatment duration, the ICER of spironolactone compared with standard care was £532 per QALY and the ICER of eplerenone compared with spironolactone was £1949 per QALY. Assuming lifetime treatment duration, the ICER of spironolactone compared with standard care was £43 per QALY and the ICER of eplerenone compared with spironolactone was £4798 per QALY. Hence, although spironolactone was not ruled out of the ICER calculations, the ICER of eplerenone compared with spironolactone demonstrated that eplerenone remained the most cost-effective strategy based on current cost-effectiveness thresholds.

While restricting the evidence network to the four highest quality studies did not alter the cost-effectiveness conclusions, decision uncertainty again increased as a result of drawing on a more restricted set of trials. Consequently the EVPI results increased markedly compared with those estimated using the wider evidence network.

Scenario 5: Assuming equivalent efficacy

The final scenario considered an alternative approach to populating the effectiveness inputs reflecting the concerns that may be expressed regarding the robustness of the underlying assumptions of the evidence synthesis model. In this scenario a class effect was assumed for the aldosterone antagonists and it was assumed that the only differences considered related to the incidence of side effects. The class effect assumes that the relative effect on mortality and non-fatal events of eplerenone and spironolactone compared with standard care are the same. In other words, the relative efficacy (and the uncertainty surrounding it) of eplerenone compared with standard care is assumed for spironolactone. Therefore, the only difference between the two drugs is differences related to the incidence of adverse effects.

It was previously reported in Chapter 3 that the evidence relating to adverse events associated with aldosterone antagonists was sparsely reported, and where evidence was available it appeared that the overall rate of adverse events and the rate of hyperkalaemia may be similar across the drugs, but the rate of gynaecomastia could be higher with spironolactone. These data also indicate that although a large proportion of adverse events occur soon after the initiation of treatment, late events do occur, and the lack of long-term data may underestimate the rate of serious adverse events.

In the absence of robust data on the incidence of side effects, this scenario focused on the potential impact of incorporating higher rates of gynaecomastia for spironolactone. Two alternative scenarios were considered: (1) gynaecomastia would reduce overall compliance with spironolactone (i.e. patients would discontinue treatment if they had gynaecomastia) and (2) gynaecomastia would not alter compliance but would result in a decrement in overall quality of life for the entire duration of treatment with spironolactone. In both scenarios it was assumed that the incidence of gynaecomastia would be 10% for patients receiving spironolactone and 0% in the other strategies. The incidence of gynaecomastia ranged from 0.25% to 9% in Table 12. However, the evidence relating to adverse events associated with aldosterone antagonists was sparsely reported, and in the absence of robust data, 10% was considered a conservative estimate.

Tables 52–55 report the cost-effectiveness and EVPI results, assuming a 10% discontinuation rate each year in the spironolactone strategy for a 2-year treatment duration and lifetime treatment duration. As anticipated, assuming a class effect and considering potential differences in side effects only had a marked effect on the overall cost-effectiveness and EVPI results. In both the 2-year and lifetime duration scenarios, the ICER of eplerenone increased above conventional thresholds of cost-effectiveness (£44,134 per QALY and £70,509 per QALY for 2-year and lifetime durations, respectively). Despite incorporating an additional risk of discontinuation for spironolactone, this strategy appears the most cost-effective treatment with an ICER of £1046 per QALY (2-year treatment duration) and £1280 per QALY (lifetime treatment duration) compared with standard care alone.

Employing this particular set of assumptions has an equally marked effect on the EVPI estimates and the conclusions that could be drawn concerning the potential value of a future head-to-head trial. The results indicate that the probability that spironolactone is the most cost-effective strategy is almost one across the different cost-effectiveness thresholds. This results in estimates of the total EVPI of between £176,000 and £1.8M across the different treatment durations and thresholds, suggesting that a further trial would not appear to provide value for money.

Tables 56–59 report the cost-effectiveness and EVPI results assuming 10% of patients receiving spironolactone incur a decrement in quality of life because of adverse events, for 2 years of treatment or lifetime duration of treatment. A decrement in utlitity of 0.1 was applied in the model in this scenario consistent with the approach reported by Glick et al. 118 The results and conclusions were similar to those where a lower compliance was assumed.

Key findings from cost-effectiveness analysis

When the results of the evidence synthesis were used to inform the relative effectiveness estimates in the model, eplerenone consistently emerged as the most cost-effective strategy for the management of postMI HF. For 2-year treatment duration, the ICER of eplerenone compared to standard care was £4457 per QALY. This increased to £7893 per QALY assuming that treatment with eplerenone was continued over a patient’s lifetime. In both of these scenarios spironolactone was either dominated or extendedly dominated by eplerenone or standard care. The results were robust to a range of alternative assumptions including applying alternative baseline event rates and incorporating potential price reductions for a generic version of eplerenone. In each of the scenarios based on the evidence synthesis results, the ICER of eplerenone was consistently under the £20,000–£30,000 per QALY threshold of cost-effectiveness conventionally used to establish value for money in the NHS.

In the absence of direct head-to-head evidence and the inevitable uncertainties in the meta-regression analysis, an alternative scenario was also considered assuming a ‘class effect’ for the aldosterone antagonists in terms of major clinical events but allowing for potential differences in their side effect profiles. This scenario had an important effect on the cost-effectiveness results with spironolactone now appearing the most cost-effective strategy. Although eplerenone was not ruled out by either dominance or extended dominance, the ICER of eplerenone compared with spironolactone exceeded the £20,000–30,000 threshold.

Key findings from the value of information analysis

Bayesian VOI analysis was undertaken to determine the expected costs of decision uncertainty predicted by the model and the maximum value that can be placed on additional research aimed at reducing this uncertainty. The estimates of EVPI provide an upper boundary for the value of additional research and provide a necessary hurdle for determining the potential efficiency of further primary research. This analysis can therefore be used as the basis to inform policy decisions relating to future research priorities and study design issues in this area. The population EVPI estimates suggest that further research is likely to be of significant value. The results indicate a considerable range in the population EVPI estimates, between £820M (base-case) and £1265M (lifetime treatment duration scenario). This uncertainty was driven by the relative treatment effects of mortality between eplerenone and spironolactone, indicating that a future head-to-head RCT of these two treatments in a postMI HF population would be likely to be highly valuable. This conclusion remained robust to a range of alternative assumptions including the potential price reduction of a generic version of eplerenone. However, the results of the scenario analyses also demonstrated that the EVPI results appear highly sensitive to whether the cost-effectiveness analysis was based on the results of the Bayesian meta-regression or a class-effect was assumed. In the latter case the EVPI results indicate that undertaking further primary research would be unlikely to represent value for money.

Strengths

The decision model was developed to address a number of important evidence gaps identified concerning the use of aldosterone antagonists for the management of post-MI HF; most notably the lack of previous studies directly comparing the potential cost-effectiveness of alternative licensed aldosterone antagonists (eplerenone and spironolactone) and the absence of cost-effectiveness studies undertaken from the perspective of the NHS. The Bayesian synthesis incorporated evidence from a wider evidence network of aldosterone trials to address the problems noted with basing an indirect comparison on the results of the postMI HF trials alone. The main strength of this approach is that it provides more robust predictions on efficacy for the treatments in the populations where existing data are limited by ‘borrowing strength’ from the relationships that exist between the treatments and between the separate populations. A further strength is that the robustness of the results to the inclusion/exclusion of particular studies (e.g. studies considered to be of a lower quality) was explored and provided evidence that appeared to support consistency in the wider set of studies included in the Bayesian synthesis with those considered in the main clinical effectiveness review. Finally, a range of scenarios were explored to examine the robustness of the cost-effectiveness and EVPI results to a series of alternative assumptions.

Limitations

The cost-effectiveness and EVPI results are subject to several important limitations. The strength of conclusions that can be drawn based on the current set of cost-effectiveness results clearly depends on the validity of the Bayesian evidence synthesis approach employed. There remain a number of difficulties in making a reliable comparison between spironolactone and eplerenone based on evidence from either the main RCTs (RALES, EPHESUS) or from combining these with the results of the additional studies included in the Bayesian synthesis. The synthesis approach employed requires the trials to be exchangeable within, although not across, the HF and postMI HF populations. Given the relatively low number of studies and the small sample sizes for many of the studies considered, the statistical absence of heterogeneity within the separate populations may not provide sufficient evidence to support the assumption employed. Furthermore, it should be emphasised that the existing RCT evidence for spironolactone in a postMI HF population is extremely limited compared with that for eplerenone. While this is reflected in the wider confidence intervals for the treatment effect estimate applied to spironolactone, the lack of robust evidence specifically in the postMI HF population represents a major limitation in terms of informing current service provision in the NHS.

It should also be recognised that data on other key inputs into the model were less well reported in the RCTs. In particular, data on non-fatal events requiring hospitalisation were only reported in a few of the RCTs, which meant that it was not possible to undertake a robust synthesis of these data directly. Instead it was assumed that the relative difference between eplerenone and spironolactone estimated for all-cause mortality would also be similar for non-fatal events. There was also a lack of data on the adverse events associated with the aldosterone antagonists. Given the paucity of good quality evidence and the diversity of dosages and populations from which data were derived, it was not considered possible to draw reliable conclusions.

There also exist limitations with respect to the strength of existing evidence regarding the optimal duration of treatment with aldosterone antagonists. The base-case analysis assumed a maximum duration of treatment of 2 years consistent with the mean follow-up reported for EPHESUS. An alternative scenario was also considered which demonstrated that longer-term treatment could be potentially cost-effective. However, it should be noted that there is limited evidence on the longer-term effectiveness of aldosterone antagonists and hence the results of this scenario (based on assuming that the relative effect will remain constant over the longer term) should be seen as exploratory.

The use of long-term observational evidence from the SMR data addressed potential concerns over the generalisability of the baseline event rates reported in the trials to UK practice. However, the SMR data are subject to several potential limitations. First, the long-term follow-up available in the SMR data represents both a relative strength and a potential limitation because the data may no longer adequately reflect current NHS practice. Second, the population included from the SMR data is unlikely to precisely match the postMI HF population considered in the EPHESUS study. That is, details of the timing of previous MI were not reported in the SMR data and similarly no details were reported in relation to the severity of the HF. Consequently there exist a number of potential sources of uncertainty related to the use of these data as a source of baseline data within the cost-effectiveness analysis. However, it should also be noted that approaches were taken to explore the impact of these sources of uncertainty by using data from EPHESUS for the acute period and attempting to recalibrate the baseline mortality data to more closely reflect the baseline characteristics of the EPHESUS population and to incorporate changes in current practice. Importantly, the cost-effectiveness and EVPI results appeared robust to these alterations.

Finally, although the EVPI results demonstrated significant potential value in undertaking further research in the form of a head-to-head RCT, the EVPI results present an upper bound to further research and hence do not provide both a necessary and sufficient condition, even if the cost of a trial fell below this amount. This is because a trial will resolve only a proportion of the uncertainty and as such, the amount of uncertainty that is likely to be resolved would have to be assessed against the cost of the trial to ensure that any further research was considered an efficient use of resources. Additional work would therefore need to be undertaken to establish the efficiency and optimal design of a future RCT.

Search strategies to identify systematic reviews and guidelines of spironolactone and eplerenone for postMI HF or HF.

Search strategies to identify completed and ongoing clinical trials of spironolactone, eplerenone, canrenone and canrenoate potassium for postMI HF or HF

Search strategies to identify adverse effects information for spironolactone and eplerenone

Clinical effectiveness

Adverse events

Effectiveness

Adverse events