Dronedarone for the treatment of non-permanent atrial fibrillation

The manufacturer's submission presented the use of dronedarone in two positions in the rhythm control treatment pathway for paroxysmal and persistent atrial fibrillation. According to NICE's guideline on the management of atrial fibrillation (updated by NICE's guideline on atrial fibrillation: diagnosis and management), beta-blockers (in addition to anticoagulation) should be the initial treatment option for people with symptomatic paroxysmal atrial fibrillation and people with persistent atrial fibrillation in whom an antiarrhythmic drug is needed to maintain sinus rhythm after cardioversion. For those in whom standard beta-blockers are contraindicated, not tolerated or fail to suppress symptoms, the guideline states that amiodarone, sotalol or a class 1c drug should be used (that is, as a second-line treatment). The choice of amiodarone, sotalol or a class 1c drug depends on the type of atrial fibrillation (persistent or paroxysmal) and the presence or absence of structural heart disease, left ventricular dysfunction or coronary heart disease. The manufacturer's submission considered the use of dronedarone as an alternative to amiodarone, sotalol and class 1c drugs for people in whom one of these antiarrhythmic drugs is indicated. The submission also considered the use of dronedarone as part of first-line treatment in addition to standard baseline therapy (usually including beta-blockers and anticoagulation), but only for people with a CHADS₂ score of 4 or more. The CHADS₂ score is used to estimate the risk of stroke in people with atrial fibrillation to determine whether they need treatment with anticoagulation therapy. The score is calculated by giving one point each for the presence of congestive heart failure, hypertension or diabetes mellitus, and being aged 75 years or older. Two points are given if people have already had an ischaemic stroke or transient ischaemic attack.

The main clinical evidence on dronedarone in the manufacturer's submission was based on four randomised controlled trials comparing dronedarone with placebo and one comparing it with amiodarone:

• EURIDIS and ADONIS (n=1,237, 12-month follow-up), phase 3 multicentre, parallel, randomised, double-blind, placebo-controlled trials. EURIDIS and ADONIS were European and non-European trials of the same design and the results were combined and reported together. Both dronedarone and placebo were given in addition to standard first-line therapy, which included beta-blockers (in about 60% of participants) and anticoagulation (in about 70% of participants).

• ATHENA (n=4,628, mean follow-up: 21 months), a phase 3 multicentre, parallel, randomised, double-blind, placebo-controlled trial. Both dronedarone and placebo were given in addition to standard first-line therapy, which included beta-blockers (in 71% of participants) and anticoagulation (in 60% of participants).

• DIONYSOS (n=504, 6-month follow-up), a phase 3 multicentre, parallel, randomised, double-blind trial comparing dronedarone with amiodarone.

The EURIDIS and ADONIS trials included people with paroxysmal or persistent atrial fibrillation or atrial flutter, who had an episode of atrial fibrillation in the 3 months before study entry but were in sinus rhythm on entry into the study. Both trials excluded people with NYHA class 3 and 4 heart failure. The trial population across the two trials had a mean age of 63 years, and 69% were male. Eleven per cent of people had atrial flutter, 41% had structural heart disease and 17% of people had congestive heart failure. The results showed that fewer people in the dronedarone arm had atrial fibrillation recurrence at 12 months than in the placebo group (64% and 75% respectively, hazard ratio 0.75, 95% confidence interval [CI] 0.65 to 0.87, p<0.001). The median time to atrial fibrillation recurrence was 116 days in the dronedarone group and 53 days in the placebo group. In the dronedarone group, the mean ventricular rate during the first adjudicated atrial fibrillation recurrence was 103 beats per minute compared with 117 beats per minute in the placebo group (p<0.001). Most adverse events were similar between the study groups (numbers not reported), although there was a lower incidence of hyperthyroidism in the dronedarone group (8.4% and 14.1% respectively, p=0.002) and a higher incidence of serum creatinine elevation (2.4% and 0.2% respectively, p=0.004). A post-hoc analysis showed that 23% of people in the dronedarone groups had been admitted to hospital or had died at 12 months versus 31% in the placebo groups (hazard ratio 0.73, 95% CI 0.57 to 0.93, p=0.01).

The ATHENA trial included people with paroxysmal or persistent atrial fibrillation or atrial flutter with at least one of the following additional risk factors for a major cardiovascular event: aged 70 years or older; hypertension needing treatment with at least two antihypertensive drugs of different classes; diabetes; previous stroke, transient ischaemic attack or systemic embolism; left atrial diameter of greater than or equal to 50 mm; left ventricular ejection fraction of less than 40%. During the course of the trial, overall mortality figures were lower than expected. As a consequence the inclusion criteria were changed so that people were required to be aged 75 years or older to be eligible, or aged 70 to 75 years with at least one of the previously specified risk factors, and people younger than 70 years of age were no longer eligible. People were excluded from the ATHENA trial if they had permanent atrial fibrillation, an unstable haemodynamic condition, NYHA class 4 congestive heart failure or an acute myocardial infarction.

The ATHENA trial population had a mean age of 72 years, and 53% were male. The primary outcome was a composite of first unplanned hospitalisation because of a cardiovascular event and death before hospitalisation. This outcome occurred in 31.9% of the dronedarone group, of whom 29.3% had a first unplanned hospitalisation because of a cardiovascular event and 2.6% died before hospitalisation. The primary outcome occurred in 39.4% of the placebo group, of whom 36.9% had a first unplanned hospitalisation because of a cardiovascular event and 2.5% died before hospitalisation. The hazard ratio for the primary composite outcome in the dronedarone group was 0.76 (95% CI 0.69 to 0.84, p<0.001). There was no statistically significant difference in all-cause mortality between the dronedarone and placebo groups (5% and 6% respectively, hazard ratio 0.84, 95% CI 0.66 to 1.08, p=0.18); however there were significantly fewer deaths from cardiovascular causes in the dronedarone group than the placebo group (2.7% and 3.9% respectively, hazard ratio 0.71, 95% CI 0.51 to 0.98, p=0.03). A post-hoc analysis of the ATHENA trial found that dronedarone was associated with a statistically significant reduction in the risk of stroke compared with placebo (hazard ratio 0.66, 95% CI 0.46 to 0.96, p=0.027). Another post-hoc analysis reported that dronedarone was associated with a statistically significant reduction in all-cause mortality compared with placebo in people with a CHADS₂ score of 4 or more (hazard ratio 0.53, 95% CI 0.31 to 0.91, p=0.022).

In the ATHENA trial, treatment was stopped early in 30.2% of the dronedarone group compared with 30.8% of the placebo group (statistical significance not reported). The main reasons for discontinuation were: adverse events (12.7% in the dronedarone group and 8.1% in the placebo group, p<0.001), participant's request (7.5% in each group), and other reasons (9.4% in the dronedarone group and 14.4% in the placebo group). Gastrointestinal events (including diarrhoea and nausea) were the most common adverse events in both groups, but were more frequent in the dronedarone than the placebo group (26.2% and 22.0% respectively, p<0.001). The dronedarone group also had higher incidences of bradycardia (p<0.001), QT-interval prolongation (p<0.001), rash (p=0.006) and serum creatinine elevation (p<0.001) than the placebo group. There was no statistically significant difference between the dronedarone and placebo groups in the number of serious treatment-emergent adverse events (456 and 489 in each group respectively).

The DIONYSOS trial included people with paroxysmal or persistent atrial fibrillation or atrial flutter in whom cardioversion and antiarrhythmic treatment were indicated and who were also receiving anticoagulation. The trial population had a mean age of 64 years, and two-thirds were male. The primary composite outcome of first incidence of either recurrence of atrial fibrillation or premature study discontinuation because of intolerance or lack of efficacy occurred in 75.5% of the dronedarone group and 58.8% of the amiodarone group (hazard ratio 1.59, p<0.0001). This difference was mainly because of the higher incidence of recurrence of atrial fibrillation in the dronedarone group than in the amiodarone group (63.5% and 42.0% respectively). The main safety endpoint was defined as the incidence of thyroid-, hepatic-, pulmonary-, neurological-, skin-, eye- or gastrointestinal-specific events, or early study drug discontinuation after any adverse event. This endpoint occurred in 39.3% of the dronedarone group after 12 months of treatment compared with 44.5% in the amiodarone group (hazard ratio 0.80, 95% CI 0.60 to 1.07, p=0.13). Dronedarone was associated with lower incidences of adverse drug reactions including thyroid dysfunction, significant bradycardia and effects on the central nervous system. There were two (0.8%) deaths in the on-treatment period in the dronedarone group compared with five (2.0%) in the amiodarone group (the causes of the deaths were provided but marked academic in confidence).

The manufacturer carried out direct and indirect analyses and a mixed treatment comparison of dronedarone compared with amiodarone, sotalol and class 1c drugs (flecainide and propafenone combined) and of each antiarrhythmic drug compared with placebo. The analyses were conducted for five outcomes: atrial fibrillation recurrence, all-cause mortality, treatment discontinuation, treatment discontinuation because of adverse events and stroke. The meta-analysis demonstrated that atrial fibrillation recurrence was significantly lower with all antiarrhythmic drugs compared with placebo and that dronedarone was the least effective of all antiarrhythmic drugs (odds ratios were marked academic in confidence). The mixed treatment comparison indicated that dronedarone was associated with a lower risk of all-cause mortality than amiodarone (odds ratio 3.19, 95% CI 1.16 to 8.76, p=0.032) and sotalol (odds ratio 5.05, 95% CI 1.84 to 13.87, p=0.009). There was no statistically significant difference in all-cause mortality between dronedarone and placebo and not enough evidence to compare dronedarone with class 1c drugs. The mixed treatment comparison showed that dronedarone was associated with a lower risk of stroke than placebo (odds ratio 1.44, 95% CI 1.19 to 1.76, p=0.015), but there was no difference between dronedarone and sotalol or amiodarone. However, this analysis was based on limited data.

The manufacturer provided information on two trials of dronedarone that were outside its licensed indication and that therefore did not form the main evidence base for the submission. These were the ERATO trial, in people with permanent atrial fibrillation, and the ANDROMEDA trial, in people with severe heart failure. In addition, the manufacturer submitted results of an analysis of safety using pooled data from five placebo-controlled dronedarone trials (ATHENA, EURIDIS and ADONIS plus two additional trials that did not meet inclusion criteria for the main clinical-effectiveness review: ERATO and DAFNE). The analysis included a total of 6,285 people and the average duration of dronedarone exposure was 12 months. The main adverse events associated with dronedarone were diarrhoea, nausea or vomiting, serum creatinine elevation, rash, and cardiac events (bradycardia and QT prolongation). The incidence of serious adverse events was similar in the dronedarone and placebo groups (18.0% and 19.7% respectively) and these were mainly related to infections and infestations, gastrointestinal disorders, and cardiac disorders. There were more early discontinuations in the dronedarone group than the placebo group (11.8% and 7.7% respectively) and the most common reason for stopping dronedarone was diarrhoea (statistical significance not reported).

The model used in the manufacturer's cost-effectiveness analysis was a discrete event simulation that predicts a person's course if they are treated with dronedarone compared with the predicted course with alternative treatment pathways. The manufacturer stratified people depending on their type of atrial fibrillation and baseline risk factors into five groups in accordance with NICE's guideline on the management of atrial fibrillation (updated by NICE's guideline on atrial fibrillation: diagnosis and management): paroxysmal atrial fibrillation without structural heart disease, paroxysmal atrial fibrillation with coronary heart disease, paroxysmal atrial fibrillation with left ventricular dysfunction, persistent atrial fibrillation without structural heart disease, and persistent atrial fibrillation with structural heart disease. For each of these subgroups, the manufacturer evaluated the cost effectiveness of dronedarone at two positions in the care pathway for atrial fibrillation (see section 3.1). When dronedarone was evaluated as part of a first-line treatment for people with a CHADS₂ score of 4 or more (in addition to standard baseline therapy), the comparator was standard baseline therapy alone (including beta-blockers [excluding sotalol] and anticoagulation). When dronedarone was evaluated as a second-line treatment option, the comparators were the antiarrhythmic drugs amiodarone, sotalol and class 1c drugs, depending on the type of atrial fibrillation and baseline risk factors described above.

The manufacturer's model used a lifetime time horizon and included four health states: normal sinus rhythm, permanent atrial fibrillation with uncontrolled symptoms, permanent atrial fibrillation with controlled symptoms and death. From the normal sinus rhythm state, people could move to any of the other states. From the two permanent atrial fibrillation health states, people could move between these states or to death. Transition between health states was determined by the following events: atrial fibrillation recurrence, acute coronary syndrome, stroke, congestive heart failure, treatment discontinuation, change in symptoms (for the permanent atrial fibrillation states) or death. The baseline risk of these events was taken from the ATHENA trial, extrapolated to a lifetime time horizon and adjusted for each treatment arm using odds ratios from the mixed treatment comparison. All-cause mortality was estimated using age-specific UK life tables (from the Government Actuary's Department) and adjusted for CHADS₂ score. The risk of death after stroke and congestive heart failure events was estimated using published sources.

The model included adverse events associated with each treatment. Adverse event rates for dronedarone were taken from a pooled analysis of the six dronedarone trials (DAFNE, ADONIS, ERATO, EURIDIS, ATHENA and DIONYSOS), for amiodarone they were taken from the DIONYSOS trial, and for sotalol and class 1c drugs they came from the SPCs. Utilities for the health states were taken from the AFTER cohort of the European Heart Survey on atrial fibrillation. The disutilities associated with adverse events were taken from a study undertaken by the manufacturer (n=127) using a time trade-off approach.

In the model, drug costs for comparators were taken from the BNF (edition 57). Doses were based on the recommended dosage stated in the SPCs. Drug administration costs were sourced from NHS Reference Costs 2007/08. For dronedarone these consisted of a specialist outpatient visit for treatment initiation and a GP visit for a day-7 creatinine test (£213). For comparators, it was assumed that hospitalisation was required for treatment initiation (£249) and 6-monthly GP visits and tests were required for monitoring (£58 to £76 depending on the treatment). Costs for the majority of health events occurring in the model were taken from published literature. Most events were assumed to incur a one-off cost; but for stroke and congestive heart failure, ongoing daily costs were assumed. Costs for adverse events came from NHS Reference Costs 2007/08. A proportion of adverse events were assumed to require hospitalisation (based on expert clinical opinion) and the rest were assumed to require an outpatient consultant visit. For short-term adverse events, a one-off cost at treatment initiation was incurred and for adverse events with lifetime effects, a 6-monthly GP visit was assumed to be required. Data on resource use were sourced from clinical opinion and published literature.

In the manufacturer's base-case analysis, the incremental cost-effectiveness ratios (ICERs) for the analysis of dronedarone if given in addition to standard baseline therapy (for people with a CHADS₂ score of 4 or more) compared with standard baseline therapy alone ranged from £6,757 to £7,890 per quality-adjusted life year (QALY) gained (incremental costs £3,053 and £3,307 and incremental benefits 0.45 and 0.42 QALYs for these two ICERs respectively). The ICERs varied depending on the type of atrial fibrillation and the presence of structural heart disease, coronary heart disease or left ventricular dysfunction. For the analysis of dronedarone as an alternative antiarrhythmic drug to amiodarone, the ICERs were £2,645 per QALY gained (incremental cost £3,528 and incremental benefit 1.33 QALYs) for paroxysmal atrial fibrillation with left ventricular dysfunction and £3,113 per QALY gained (incremental cost £3,986 and incremental benefit 1.28 QALYs) for persistent atrial fibrillation with structural heart disease. For the comparison of dronedarone with class 1c drugs, the ICERs were £20,003 per QALY gained (incremental cost £1,980 and incremental benefit 0.10 QALYs) for paroxysmal atrial fibrillation with no structural heart disease and £20,761 per QALY gained (incremental cost £2,069 and incremental benefit 0.10 QALYs) for persistent atrial fibrillation with no structural heart disease. For the comparison of dronedarone with sotalol, the ICERs ranged from £1,929 to £2,197 per QALY gained (incremental costs £3,986 and £4,384 and incremental benefits 2.07 and 2.00 QALYs for these two ICERs respectively; depending on the type of atrial fibrillation and the presence or absence of underlying heart disease).

The manufacturer conducted a number of sensitivity analyses including:

• subgroup analyses based on CHADS₂ scores and gender

• using alternative sources for the baseline distribution of CHADS₂ score

• varying the model time horizon

• assuming a minimum mortality benefit from dronedarone relative to its comparators by using the lower end of the 95% CI of the mortality estimate for comparators and the upper end of the 95% CI for dronedarone (rather than the point estimates), and vice versa (that is, assuming a maximum relative mortality benefit from dronedarone)

• using different curve fits for the modelled clinical events such as stroke and treatment discontinuations

• using different estimates for various parameters including mortality treatment effect, stroke treatment effect, treatment discontinuation, adverse event rate, costs of dronedarone and utilities.
  
  The analyses that had the greatest effect on the ICERs were using a 1-year time horizon (rather than a lifetime time horizon) and assuming a minimum mortality benefit from dronedarone relative to its comparators.

The ERG considered that all relevant trials of dronedarone had been included in the manufacturer's submission. It noted that the ATHENA trial included people who were older and had a higher risk of a major cardiovascular event than people in the other trials and that the application of this evidence to a lower-risk and younger population was uncertain. The ERG commented that the DIONYSOS trial was the only head-to-head trial of dronedarone versus an antiarrhythmic drug and therefore the relative efficacy of dronedarone compared with antiarrhythmic drugs other than amiodarone was unknown. It also noted that the DIONYSOS trial was short-term (minimum follow-up: 6 months). The ERG commented on a number of limitations of the meta-analyses and mixed treatment comparison in the manufacturer's submission. These included:

• a lack of consideration of clinical and statistical heterogeneity of the studies included in the analyses

• uncertainty about the validity of pooling the individual studies in the different analyses

• few events in the studies

• the use of outcomes that were neither pre-specified endpoints nor centrally adjudicated

• inconsistencies in the selection of studies across the different analyses

• the restriction of randomised controlled trials in the mixed treatment comparison.
  
  The ERG considered that the assumption that class 1c drugs have a similar effect on all-cause mortality to dronedarone and no effect on the risk of stroke (made because of a lack of evidence) might not be valid. It noted an inconsistency in the direction of effect between results of the direct and indirect analyses and the mixed treatment comparison for the outcome of treatment discontinuations because of any cause. The ERG considered that the existing clinical evidence across the antiarrhythmic drugs appeared most robust for the outcome of atrial fibrillation recurrence, but considerably more uncertain for the other major clinical endpoints such as stroke and all-cause mortality. The ERG also noted that although the marketing authorisation for dronedarone states that it should be used to lower ventricular rate, there was little evidence presented on this outcome.

The ERG considered that in general, the manufacturer's approach to the economic evaluation met the requirements of the NICE reference case, had an appropriate structure for the decision problem, and was of high quality, overall. However, the ERG noted a number of issues with the cost-effectiveness analysis, including concern over the pivotal assumption of mortality benefit:

• The treatment pathways evaluated by the manufacturer might not represent the full range of treatment strategies or sequences for dronedarone.

• The baseline data from the ATHENA trial, used in the model, might not be generalisable to people with atrial fibrillation in the NHS because they came from an older and higher-risk population.

• The results of the meta-analyses and mixed treatment comparisons, used in the model, might not be appropriate because of concerns about the methodology of these analyses.

• The lack of health-related quality-of-life data from any of the dronedarone studies.

• The assumption of lower initiation and monitoring costs for dronedarone compared with other antiarrhythmic drugs might not be appropriate.

• The uncertainty associated with modelling the benefits of dronedarone over the longer term because of the short duration of the trials.

The ERG made revisions to the manufacturer's model to correct coding errors (relating to adverse events costs and the length of time that mortality treatment benefits were applied). The revisions resulted in considerably lower ICERs than those reported in the manufacturer's base case for the comparisons of dronedarone with sotalol and amiodarone (ICERs ranged from £1,895 to £4,014 per QALY gained in the ERG's analysis applying a lifetime mortality benefit compared with £1,980 to £8,142 per QALY gained in the base case). The results for dronedarone compared with class 1c drugs were unaffected because both drugs were assumed to have the same mortality benefit.

The ERG stated that the manufacturer's base-case ICERs were based on the estimates of relative effectiveness of dronedarone compared with other antiarrhythmic drugs derived from the manufacturer's mixed treatment comparison. It had previously noted concerns about this mixed treatment comparison (section 3.16). The ERG therefore performed a number of analyses exploring the impact of assumptions about treatment effects on the ICERs. These included:

• assuming that dronedarone has the same effect on mortality across all CHADS₂ subgroups (for the comparison of dronedarone with standard baseline therapy)

• assuming that sotalol and amiodarone have no effect on mortality, but keeping the assumed mortality benefit of dronedarone

• assuming that class 1c drugs, sotalol and amiodarone have the same effect on mortality as dronedarone

• assuming that class 1c drugs have a more beneficial effect on mortality than dronedarone

• assuming that class 1c drugs have the same effect on stroke as dronedarone

• using effect estimates from a reanalysis of the mixed treatment comparison of all-cause mortality using a wider range of studies than that used in the manufacturer's analysis.

For most analyses, the ICERs increased but remained below £20,000 per QALY gained. However, when sotalol and amiodarone were assumed to have the same effect on mortality as dronedarone, the ICERs increased to between £55,063 and £119,704 per QALY gained. When class 1c drugs were assumed to have the same effect on the risk of stroke as dronedarone, the ICERs approximately doubled (to about £38,000 per QALY gained) and when class 1c drugs were assumed to have greater mortality benefits than dronedarone, class 1c drugs had both higher effectiveness and lower costs than dronedarone. The ERG also explored the uncertainty around treatment initiation, monitoring costs and utility weights used in the model. The impact on the ICERs for all analyses and comparisons was marginal.

The ERG conducted exploratory analyses to identify the main drivers of the cost-effectiveness results. The first was to explore the effect on the ICERs when all treatment effects are excluded from the economic analysis except atrial fibrillation recurrence. The ICERs either increased to between £1,355,984 and £70,323,846 per QALY gained or dronedarone was shown to have both higher costs and lower effectiveness than the comparators. The ERG then explored the effect on the ICERs when the treatment effects on all-cause mortality were included in the analysis in addition to atrial fibrillation recurrence. The ICERs decreased to between £1,815 and £4,566 per QALY gained for the comparisons of dronedarone with standard baseline therapy, sotalol and amiodarone. For the comparison of dronedarone with class 1c drugs, the ICERs were either £370,690 or dronedarone was shown to have both higher costs and lower effectiveness than the comparators (because both drugs were assumed to have the same effect on mortality). Based on these analyses, the ERG concluded that the main driver of the cost effectiveness of dronedarone compared with standard baseline therapy, sotalol or amiodarone is the reduction in all-cause mortality associated with dronedarone. To explore the main driver of the cost effectiveness of dronedarone compared with class 1c drugs, the ERG conducted a further analysis including the treatment effects on stroke in addition to atrial fibrillation recurrence and mortality. This resulted in ICERs of £43,543 and £46,500 per QALY gained. The ERG noted that when treatment effects on adverse events were included in the analysis (as in the manufacturer's base-case analysis), the ICERs were around £18,000 per QALY gained. It therefore advised that the combined effect of reduced risk of stroke and fewer adverse events was the main driver of cost effectiveness for dronedarone compared with class 1c drugs.

After consultation on the first appraisal consultation document (ACD), the ERG conducted two further scenario analyses in which the treatment effects on all-cause mortality were varied. In the first analysis, dronedarone was assumed to have no effect on all-cause mortality compared with placebo, whereas amiodarone, sotalol and class 1c drugs were assumed to increase the risk of all-cause mortality (using effect estimates from the ERG's mixed treatment comparison). For the comparison of dronedarone with standard baseline therapy (in people with a CHADS₂ score of 4 or more), the ICERs increased from the ERG's revised base case (between £3,358 and £4,014 per QALY gained) to between £56,798 and £69,575 per QALY gained. For the comparisons with amiodarone and sotalol, the ICERs increased from between £1,692 and £2,349 to between £2,588 and £5,853 per QALY gained and for the comparison with class 1c drugs, the ICERs were lower (£11,648 and £12,760 per QALY gained) than in the ERG's revised base case (£18,206 and £18,955 per QALY gained). In the second analysis, amiodarone, sotalol and class 1c drugs were assumed to have no effect on all-cause mortality compared with placebo, and the effect of dronedarone compared with placebo was varied from an odds ratio of 0.84 (that is, a beneficial effect as in the manufacturer's model) to 1.0 (that is, no effect on all-cause mortality). This threshold analysis showed that when the odds ratio was 0.95 or lower (that is, when dronedarone was assumed to reduce all-cause mortality by at least 5%), the ICERs for all comparisons were between £9,323 and £20,689 per QALY gained.

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