Dual-chamber pacemakers for treating symptomatic bradycardia due to sick sinus syndrome without atrioventricular block: a systematic review and economic evaluation

Study design

Three RCTs with a parallel-group design19,37,41 and three crossover RCTs34,35,43 were identified as relevant and were included in this review.

The follow-up period varied greatly among the included studies. Of the parallel-group RCTs, Albertsen et al. had a set follow-up of 12 months,37 the DANPACE trial had a follow-up of up to 10 years with an average of 5.4 years (SD 2.6 years),19 and, in Nielsen et al. the follow-up ranged from 6 days to 5.3 years (mean 2.9 years, SD 1.1 years). 41

The follow-up in the crossover trials was shorter than in the parallel studies. In Lau et al. 35 and Schwaab et al. ,43 patients spent 4 weeks and 3 months, respectively, before crossing over to the other pacing mode. Gallik et al. 34 studied the immediate effects of pacing mode during exercise; haemodynamic effects were measured during bicycle exercise first in one pacing mode and after 0.5–1 hour’s rest, the exercise was repeated in the other pacing mode.

Intervention and comparator

The three parallel RCTs randomised patients to receive single- or dual-chamber pacemakers. 19,37,41 In the crossover trials, all patients were implanted with a dual-chamber pacemaker and then randomised to a pacing programme of dual-chamber or single-chamber atrial pacing, followed by the alternative pacing mode. 34,35,43

Most trials randomised patients before pacemaker implantation, including the trials randomising patients by device (parallel RCTs),19,37,41 and two of the studies randomising by pacing programme. 35,43 The remaining trial, by Gallik et al. ,34 randomised patients who had recently had a dual pacemaker implanted.

The single and dual pacemakers used in the included trials were from several different manufacturers including Medtronic (Minneapolis, MN, USA), ELA Medical Inc. (Arvada, CO, USA), Boston Scientific (Marlborough, MA, USA); St Jude Medical (St Paul, MN, USA), Guidant (Indianapolis, IN, USA), Pacesetter (St Jude Medical; St Paul, MN, USA); Cardiac Pacemakers Inc. (St Paul, MN, USA), Telectronics Pacing Systems (Englewood, CO, USA) and Intermedics Inc. (Angleton, TX, USA).

The included trials compared DDD(R) with AAI(R) pacing. However, Nielsen et al. 41 included two DDDR trial arms with different programmed AV delay: DDDR-s with a short AV delay (< 150 milliseconds) and DDDR-l with a fixed long AV delay (300 milliseconds). Data for these two study arms have been combined in analyses in this review. However, for each outcome, either the impact of combining the study arms has been explored in a sensitivity analysis or data from each study arm have been presented separately.

The DANPACE19 study was the only trial that specifically stated that programmable features prolonging or eliminating the AV interval, in order to minimise VP, were not permitted in the trial.

In all the included studies, all or the majority of patients within each study received pacemakers programmed with the rate-adaptive function activated. The rate-adaptive function was activated in all patients in Albertsen et al. ,37 DANPACE,19 Gallik et al. ,34 Lau et al. 35 and Schwaab et al. 43 In Nielsen et al. ,41 all but two patients had the rate-adaptive function active.

The programmed AV delay in the dual-chamber pacing mode differed greatly across the studies and between study arms, as shown in Table 4. The studies had, for each study arm with dual-chamber pacing, an AV delay that was set at a specific value,34,41 in a range19,35,37 or optimised according to a programmed algorithm. 43 Gallik et al. ,34 Lau et al. 35 and the DDDR-s arm in Nielsen et al. 41 employed relatively short AV delays, up to 150 milliseconds. By contrast, the DDDR-l arm in Nielsen et al. 41 had an AV delay of 300 milliseconds. The AV delay in Albertsen et al. 37 and the DANPACE trial19 was around 220 milliseconds.

The mode switch function was active in all three parallel-group RCTs. 19,37,41 In Schwaab et al. ,43 mode switch was activated in some patients, but the number of patients was not specified. Gallik et al. 34 and Lau et al. 35 did not report mode switch settings; however, mode switching may not have been available at the time of these trials.

Population

Most of the parallel and crossover RCTs included patients with symptomatic bradycardia or SSS in combination with certain ECG criteria, for example indicating normal AV conduction.

Schwaab et al. 43 had slightly different inclusion criteria: patients had to have chronotropic incompetence, have experienced at least two documented episodes of atrial tachyarrhythmia, be on antiarrhythmic medication for prevention of atrial flutter or AF, as well as being eligible for a dual-chamber pacemaker for symptomatic bradycardia.

The parallel RCTs19,37,41 had similar exclusion criteria, excluding patients if they had chronic AF, AV block, carotid sinus syndrome, vasovagal syncope, bundle branch block, surgery, a short life expectancy, dementia or cancer. Lau et al. 35 did not report specific exclusion criteria, and Gallik et al. 34 excluded patients with evidence of AV node disease or who were unable to exercise.

Summaries of the characteristics of the study populations in the included RCTs are presented in Table 5 (parallel RCTs) and Table 6 (crossover RCTs). More detailed baseline characteristics can be found in Appendix 2, Data abstraction.

The parallel RCTs19,37,41 varied in size from 50 to 1415 randomised patients. The crossover studies34,35,43 were smaller, with between 12 and 21 participants. The RCTs all included patients with SSS or sinus node dysfunction (SND). The parallel RCTs, Albertsen et al. 37 and Nielsen et al. ,41 reported the breakdown of pacing indication of the participants for sinus arrest/SA block BTS and sinus bradycardia, with some imbalances between the trial arms; most notably, there were more people with BTS in the two dual-chamber pacing arms than in the AAIR arm in Nielsen et al. 41

Mean age was similar across the three parallel RCTs,19,37,41 and between study arms (72–74 years). The participants in the crossover trials34,35,43 had a slightly lower mean age, of 61–70 years. Only the DANPACE trial1 reported previous history of AF, with around 44% of the participants having a history of AF in each trial arm. 3 Previous stroke was captured in Albertsen et al. and the DANPACE trial. In the smaller study by Albertsen et al. ,37 the number of patients with prior stroke was low but with a notable difference between groups in the proportion of people with prior stroke (5 out of 24 patients in the AAIR arm and only 1 out of 26 in the DDDR arm). In the DANPACE trial, there was no statistically significant difference between the trial arms in the percentage of patients having experienced a stroke at trial entry (7.5% and 8.6%, respectively). 19 In the parallel RCTs that reported NYHA class at baseline19,37,41 the majority of participants were NYHA class I or II (96%) with no or mild symptoms of HF.

Outcomes

The outcomes of interest to this review that were reported in the included studies are listed in Table 7. For several of the outcomes, the trials had used different scales or measurements, which have been reported separately. These include HF, exercise capacity and HRQoL.

Randomised controlled trial quality

This section describes the trial designs and methodology employed in the trials, which may give rise to an increased risk of bias in terms of selection, detection, performance, and attrition bias. Additionally, other potential sources of bias, such as statistical methods used, are also assessed. Tables 8 and 9 summarise the results of critical appraisal of the included parallel and crossover RCTs, respectively. A more detailed description of the quality assessment of the trials can be found in Appendix 3, Quality assessment.

Selection bias

None of the full publications of the included trials described how the randomisation sequence had been generated. However, based on correspondence with the researchers for Albertsen et al. ,37 randomisation was performed in a 1 : 1 ratio. 37 Each patient was asked to draw one envelope, containing the allocation, from a batch of 10. Albertsen et al. ,37 the DANPACE trial19 and Schwaab et al. 43 gave some details about how the allocation sequence had been concealed from staff involved in the enrolment and assignment of participants. In these studies, the allocation was performed using sealed envelopes before pacemaker implantation19,37 or programming of the first pacing mode. 43 Nielsen et al. ,41 Gallik et al. 34 and Lau et al. 35 did not describe allocation concealment.

Performance and detections bias

Participants and investigators were blinded to the pacing mode in Lau et al. 35 and Schwaab et al. 43 Albertsen et al. ,37 Nielsen et al. 41 and Gallik et al. 34 did not describe the trial design regarding blinding. Based on correspondence with the researchers of the DANPACE trial,1 it was confirmed that this study was an open-label trial with participants, researchers and outcome assessors aware of the type of pacemaker and pacing mode in each patient. 19

In Nielsen et al. ,41 physical examinations and echocardiography were carried out unblinded, in contrast to the study by Albertsen et al. ,37 in which echocardiographic analyses were carried out blinded to the pacing mode. 37 However, blinding of echocardiography would have had only limited impact on the outcomes of interest captured in Albertsen et al. :37 HF, exercise capacity and adverse events. 37 In the DANPACE trial,19 a committee adjudicated stroke and thromboembolic events unaware of the assigned pacing mode. Gallik et al. 34 did not specify the blinding status of any outcome assessors.

Attrition bias

As mentioned previously, the DANPACE trial19 and Nielsen et al. 41 study participants were followed up for a variable length of time. In both studies, patients were followed up from enrolment to death or end of study, with no loss to follow-up. In Albertsen et al. ,37 one patient randomised to single-chamber atrial pacing was lost to follow-up, which has been accounted for as a treatment failure in the Technology Assessment Group (TAG)’s analyses.

Despite the small number of patients lost to follow-up, the number of patients changing pacemaker or pacing mode from the one to which they were randomised was relatively high and uneven between the trial arms in all three parallel RCTs. 19,37,41 In all three trials, the number of patients in the single-chamber atrial pacing arm who switched to (predominantly) DDDR pacing was higher than the number of patients in the dual-chamber pacing arm who switched to another pacing mode.

Among the crossover trials, three patients in Lau et al. 35 and two patients in Schwaab et al. 43 were excluded from the trials. The reasons for exclusion in Lau et al. 35 were pacemaker failure (n = 2) and patient non-compliance (n = 1), and, in Schwaab et al. ,43 development of chronic AF (n = 1) and death (n = 1). As expected, the crossover trials had to exclude participants who did not complete both intervention periods.

Reporting bias

In an early publication of the DANPACE trial results,19 outlying the protocol for the study,38 one of the secondary end points listed was a QoL evaluation, comprising elements from the general health questionnaire SF-36. However, no result of this outcome was published in either of the identified references linked to this study. 1,40

All three crossover trials34,35,43 reported results for each pacing mode separately, with mean and SE or SD. Exact p-values were not provided for the within-patient difference for any of the outcomes: the p-value was not reported, was described as non-significant or was reported to be less than a certain value. Lau et al. 35 reported IPD for general well-being [as measured by a visual analogue scale (VAS)] and Gallik et al. 34 reported IPD for exercise time, which were used to calculate the within-patient difference for these outcomes. The lack of reporting of p-value for the paired t-test for other outcome data in the crossover trials rendered the data unsuitable for meta-analysis.

Statistical analysis

Both the DANPACE trial19 and Nielsen et al. 41 were suspended before reaching the target number of participants and are consequently underpowered to show a statistically significant difference in the primary outcome: all-cause mortality in the DANPACE trial19 and changes in left atrium size and left ventricle (LV) size and function in Nielsen et al. A total of 450 patients were to be included in Nielsen et al. ,41 but recruitment was stopped after randomisation of 177 patients because recruitment for the DANPACE trial19 had started. However, recruitment for the DANPACE trial19 was also stopped early, at 1415 randomised patients, compared with the target of 1900 patients. 1 This was as a result of the increasing use of dual-chamber pacemakers with features that prolong or eliminate the AV interval to minimise VP in patients with SSS, which were not permitted in the trial and which therefore led to a decrease in the recruitment rate. In addition, a planned interim analysis showed that no statistically significant difference could be reached with respect to the primary outcome of all-cause mortality even with the planned 1900 patients.

Overall trial quality

Overall trial design and methodology were appropriate in the included trials; however, detailed descriptions of randomisation and allocation concealment were sparse. The parallel RCTs were either open label or it was unclear if and how patients, trial personnel and outcome assessors were blinded to the pacing modes. Blinding is likely to have a limited effect on the result of objective outcomes such as mortality, stroke and adverse events; however, for more subjective outcomes, including patient-reported outcomes such as QoL and HF questionnaires and exercise capacity, there is an increased risk of introducing bias into the results. The two crossover RCTs reporting results on QoL were both described as double blind. The risk of attrition bias was generally low as few patients were lost to follow-up in the parallel RCTs and the crossover trials excluded a small number of patients from the analyses. However, the number of patients in the parallel RCTs who changed pacing mode during the follow-up period was uneven between the pacing modes, which may lead to a conservative estimate of the effect of pacing mode.

Change in pacing mode

Several patients in the parallel RCTs19,37,41 changed pacing mode during the study from the one to which they were randomised. Among the 857 patients randomised to DDDR and 785 to AAIR in the three trials, significantly more people with single-chamber atrial pacing than with dual-chamber pacing changed pacemaker and pacing mode (OR 0.50, 95% CI 0.37 to 0.67; Figure 2). There was no statistical heterogeneity in the meta-analysis of the three trials and only modest uncertainty; however, the result was mainly driven by the largest and longest trial, DANPACE.

FIGURE 2.

Results from analysis of change in pacing mode.

Most patients who changed from AAI(R) changed to DDD(R). The primary reasons for the implantation of a dual-chamber pacemaker in patients with a single-chamber atrial pacemaker were development of a high-degree AV block, or Wenckebach block during implantation. However, there were also a small number of patients who switched from AAIR to VVI. Patients randomised to DDD(R) who changed pacing mode during implantation or during follow-up primarily changed to VVI pacing because of development of persistent AF. One patient was lost to follow-up in the AAIR arm of Albertsen et al. ,37 and has been included in this analysis as changing pacing mode.

Per cent atrial and ventricular pacing

The rate of atrial pacing and VP (%) varied greatly among the studies, study arms and pacing modes (Table 10). Differences between studies in the rate of paced atrial or ventricular beats may be associated with differences in other outcome measures. VP has been associated with an increased incidence of AF. 44

In the DANPACE trial19 and Nielsen et al. ,41 the percentages of atrial pacing and VP were calculated using the mean of the number of paced beats at each follow-up, which was captured by the pacemaker event counters. Schwaab et al. 43 used stored pacemaker histograms to capture the percentage of paced beats in the atrium and ventricle. Albertsen et al. 37 and Lau et al. 35 did not describe how the percentages of atrial pacing and VP were captured, and Gallik et al. 34 did not report data on the percentages of atrial pacing or VP.

Schwaab et al. 43 had the highest rate of atrial pacing and VP, with patients being paced in both the atrium and ventricle for almost every beat. The amount of atrial pacing was balanced between the trial arms in Schwaab et al. 43 (95–96%) and in the DANPACE trial19 (58–59%). By contrast, in Albertsen et al. 37 the percentage of atrial pacing was higher in the DDDR (62%) than in the AAIR group (53%), and in Nielsen et al. 41 there were similar amounts of atrial pacing in the AAIR (69%) and DDDR-l (67%) group but less in DDDR-s (57%). Lau et al. 35 did not report the percentage of atrial pacing.

The variation between the trials in percentage of VP was even greater than for atrial pacing. The VP in the dual-chamber pacing arm was 64–66% in Albertsen et al. ,37 the DANPACE trial1 and Lau et al. 35 However, the dual-chamber pacing arm with the long AV delay in Nielsen et al. 41 (DDDR-l) had only 17% VP compared with a VP percentage of above 90% in the dual-chamber pacing arm with short AV delay (DDDR-s) in the same trial. VP was also above 90% in Schwaab et al. 43 The programmed AV delay varied between the included studies, which may explain some of the variation in the percentage of VP.

All-cause mortality

The DANPACE trial19 and Nielsen et al. 41 reported all-cause mortality. With 831 people randomised to DDDR pacing and 761 patients to AAIR pacing in total, there were fewer deaths among patients with dual-chamber pacing than single-chamber atrial pacing, but the difference was not statistically significant (OR 0.97, 95% CI 0.67 to 1.41; Figure 3).

FIGURE 3.

Results from analysis of all-cause mortality.

The large DANPACE trial,19 which stopped recruitment before reaching the planned 1900 patients, was not powered to detect a difference in mortality between the two pacing modes. However, from a planned interim analysis of the DANPACE trial results,19 it was calculated that no statistically significant difference in all-cause mortality would have been observed even if all 1900 patients had been recruited. As the meta-analysis of the DANPACE trial19 and Nielsen et al. considers only 1592 patients, it is unlikely to have sufficient power to identify a statistically significant difference. The breakdown of the number of deaths in the two DDD trial arms in Nielsen et al. is shown in Table 11. 41

The DANPACE trial,19 in which the primary outcome was all-cause mortality, presented this outcome as a HR. The HR presented in the full publication was in line with the meta-analysis of mortality OR of the two included trials. The unadjusted HR for AAIR pacing compared with DDDR pacing was 1.06 (95% CI 0.88 to 1.29; p = 0.53). The HR after adjustment for baseline variables [age, sex, prior history of AF, prior myocardial infarction, left ventricular ejection fraction (LVEF) < 50%, and hypertension] was 0.94 (95% CI 0.77 to 1.14; p = 0.52) for AAIR pacing versus DDDR pacing. The all-cause mortality incidence was similar in all predefined subgroups (age > or ≤ 75 years; sex; hypertension; LVEF < or ≥ 50%; history of AF; previous myocardial infarction; PQ-interval > or ≤ 180 milliseconds; diabetes; NYHA class I or II–IV), with the smallest p-value for interaction of 0.45 (Figure 4).

FIGURE 4.

Subgroup analyses of all-cause mortality in the DANPACE trial. 1 MI, myocardial infarction.

In Albertsen et al. ,37 which did not report mortality as an outcome, one patient was lost to follow-up in the AAIR group, and may have died within the follow-up period.

Heart failure

Heart failure was reported in all three parallel RCTs. 19,37,41 However, the outcome measures varied between the studies (Table 12). In the three trials, HF was captured as NYHA class at the end of follow-up; number of patients taking diuretics; HF leading to hospitalisation; number of cases of new HF (defined as new NYHA class IV or new NYHA class III with the presence of oedema and/or dyspnoea); number of patients with an increase in consumption of diuretics; and number of patients with an increase of at least one NYHA class.

All the outcome measures for HF with a reported measure of uncertainty consistently showed no statistically significant difference between dual-chamber and single-chamber atrial pacing (see Table 12). However, because of low event rates or relatively small sample sizes the uncertainty was large around the HF outcome measures reported by Nielsen et al. 41 (patients with increased consumption of diuretics, patients with an increase of at least one NYHA class) and HF leading to hospitalisation reported by the DANPACE trial. 19

Pre-defined subgroup analyses in the DANPACE trial19 showed a statistically significant difference between single-chamber atrial pacing and dual-chamber pacing for patients aged ≤ 75 years and patients aged > 75 years in the number of patients developing new HF. In the younger subgroup (≤ 75 years), patients with AAIR were at a lower risk than those with DDDR of developing HF (HR 0.72, 95% CI 0.53 to 1.00), and in the older subgroup (> 75 years) patients were at a higher risk when on AAIR (HR 1.34, 95% CI 1.00 to 1.80). All other subgroup analyses were non-significant (sex; hypertension; LVEF < or ≥ 50%; previous myocardial infarction; PQ interval > or ≤ 180 milliseconds; NYHA class I or II–IV; diuretics, p > 0.31).

Atrial fibrillation

The DANPACE trial19 and Nielsen et al. 41 reported results on the incidence of AF. In both studies, DANPACE19 and Nielsen et al. ,41 AF was diagnosed by standard 12-lead ECG at planned follow-up visits. In the DANPACE trial,19 AF was defined as either paroxysmal (the first diagnosis of AF detected on ECG and verified by the pacemaker telemetry at a planned follow-up visit) or chronic (AF at two consecutive follow-up visits and at all subsequent follow-up visit). The results of paroxysmal and chronic AF have been combined to simplify the comparison with the results from Nielsen et al. (Table 13), but they are also reported separately (Table 14).

Nielsen et al. found that the risk of developing AF with dual-chamber pacing was significantly higher than with single-chamber atrial pacing (OR 3.19, 95% CI 1.05 to 9.67; see Table 13). 41 A sensitivity analysis of the DDDR-l and DDDR-s trial arms analysed separately similarly shows a larger proportion of patients developing AF in either dual-chamber pacing arms than in the trial arm paced with a single-chamber atrial pacemaker (see Table 13). In the sensitivity analysis, the single-chamber atrial pacing group has been split in two, so as to avoid double counting of patients.

In contrast to the results in Nielsen et al. ,41 in the DANPACE trial,19 the risk of developing paroxysmal AF was significantly lower with dual-chamber pacing than with single-chamber atrial pacing (OR 0.75, 95% CI 0.59 to 0.96; see Table 14). By contrast, no statistically significant difference between dual-chamber and single-chamber atrial pacing was identified when focusing on the number of patients who developed chronic AF (OR 0.96, 95% CI 0.68 to 1.33; see Table 14); substantial uncertainty was identified in this analysis.

The HRs for paroxysmal and chronic AF reported in the DANPACE trial19 (unadjusted and adjusted for age, sex, prior history of AF, prior myocardial infarction, LVEF < 50% and hypertension), comparing single-chamber atrial pacing with dual-chamber pacing, support the analyses (Table 15).

There are several possible reasons for the disparity in the result of AF between the DANPACE trial19 and Nielsen et al. ,41 including differences in baseline characteristics of patients enrolled in the studies and differences in pacemaker programming. Various hypotheses have been put forward around factors that may have an effect on the incidence of AF:

• Previous history of AF: in the DANPACE trial19 the strongest predictor of paroxysmal AF was prior history of AF (HR 3.23, 95% CI 2.59 to 4.03; p = 0.001). However, the DDDR and AAIR pacing arms were well balanced for previous AF at baseline. Nielsen et al. 41 did not report previous history of AF; however, they did report a breakdown of the underlying pacing indications including BTS, in which the tachyarrhythmia often is AF. In Nielsen et al. 41 there was an imbalance in the number of patients with BTS, with a larger proportion among patients in the two dual-chamber pacing arms than in the single-chamber pacing arm. Nielsen et al. 41 found a correlation between BTS at implantation and an increased risk of AF (relative risk 3.3, 95% CI 1.3 to 8.1; p = 0.01).

• PQ-interval: the result of a subgroup analysis of 650 patients in the DDDR group in the DANPACE trial19 indicates that a longer baseline PQ-interval (> 180 milliseconds) is associated with an increased risk of AF (p < 0.001). There was a slight difference in PQ interval at baseline between the studies; however, the PQ interval was well balanced between the different trial arms within each study (Table 16). 1,41

• Programmed AV interval and percentage VP: both DDDR and AAIR preserve AV synchrony. However, in AAIR normal ventricular activation pattern is preserved, whereas DDDR causes some degree of unnecessary VP with changes to the ventricular activation and contraction pattern, which has been associated with an increased risk of AF. 44,45 The programmed AV delay is closely linked to the resulting VP percentage; the DDDR-l arm in Nielsen et al. 41 had a programmed AV delay of 300 milliseconds and just 17% VP, whereas the DDDR-s arm had an AV delay of 150 milliseconds and 90% VP. In the DANPACE trial19 the patients in the DDDR arm had an AV delay and VP percentage in the middle of the range observed in Nielsen et al. 41 [225 milliseconds (SD 39 milliseconds) and 65% (SD 33%), respectively; see Table 10]. However, in Nielsen et al. ,41 there were significantly more patients with AF in both the DDDR-l and the DDDR-s arms than in the AAIR group, despite having low and high VP percentages, respectively. A subgroup analysis of 650 patients with a DDDR pacemaker in the DANPACE trial19 showed no statistically significant association between percentage VP or the length of the AV delay and risk of AF. 39

Both studies, DANPACE19 and Nielsen et al. ,41 seem to be of good quality, although there are some differences in the methods (e.g. programmed AV interval) and in the baseline characteristics of the patients in the two studies. However, the DANPACE trial19 is almost 10 times the size of Nielsen et al. 41 and it has a longer mean follow-up [5.4 years (SD 2.6 years) compared with 2.9 years (SD 1.1 years), respectively]; thus, it is reasonable to have more confidence in the results from the DANPACE trial1 than from the Nielsen et al. study.

Subgroup analyses of paroxysmal AF in the DANPACE trial19 showed a statistically significant difference between the subgroups of patients with and without a prior history of AF, body mass index (BMI) ≥ or < 25 kg/m², and a left atrial diameter > or ≤ 39 mm at baseline (Figure 5). 1 In these three subgroups the incidence of paroxysmal AF was lower with DDDR than with AAIR pacing in patients without a prior history of AF, a higher BMI and a dilated left atrium at baseline (p < 0.05). The subgroup analysis of patients with different PQ-interval > or ≤ 180 milliseconds indicated a lower risk of paroxysmal AF with DDDR than AAIR pacing in patients with a longer PQ-interval (p = 0.084). The p-values for all other interaction were > 0.34.

FIGURE 5.

Subgroup analyses of paroxysmal AF in the DANPACE trial. 19 MI, myocardial infarction.

Stroke

The studies, DANPACE19 and Nielsen et al. ,41 captured the number of patients suffering a stroke as an outcome. In Nielsen et al. ,41 the diagnosis of stroke was given when neurological symptoms of presumably cerebral ischaemic origin persisted for more than 24 hours, or if patients died within 24 hours from an acute cerebrovascular event. The definition of stroke in the DANPACE trial19 was similar: the sudden development of focal neurological symptoms lasting more than 24 hours. As for several other outcomes, the number of events was low, the uncertainty considerable and no statistically significant difference was shown (OR 0.93, 95% CI 0.60 to 1.45; Figure 6).

FIGURE 6.

Results from analysis of stroke.

The DANPACE trial19 reported an unadjusted HR for stroke of 1.13 (95% CI 0.72 to 1.80; p = 0.59) for patients with single-chamber atrial pacing compared with dual-chamber pacing. The HR when adjusted for age, sex, prior history of AF, hypertension and prior stroke was similar (HR 1.11, 95% CI 0.70 to 1.77; p = 0.65).

The breakdown of the number of patients suffering a stroke in the trial arms in Nielsen et al. 41 is shown in Table 17.

Exercise capacity

Exercise capacity was measured in the parallel RCT Albertsen et al. 37 and the crossover trials Gallik et al. 34 and Schwaab et al. 43 Albertsen et al. 37 used the 6-minute walking test (6MWT) to test exercise tolerance/capacity. 37 The 6MWT measures the distance an individual is able to walk over a total of 6 minutes on a hard, flat surface. In Gallik et al. ,34 exercise capacity was tested through an upright bicycle exercise. The initial workload was 200 kilopond metres (kpm), which was increased incrementally by 200 kpm every 3 minutes. The aim was to achieve a peak heart rate ≥ 85% predicted by age, and the outcome measure was exercise time. Schwaab et al. 43 used bicycle ergometry by incremental exercise test to exhaustion, using workload increments of 15 W/minute. 43 Outcome measures included total exercise duration and maximum workload.

Gallik et al. 34 presented IPD for exercise duration, whereas Schwaab et al. 43 presented only data for the individual treatment periods, but the results of paired t-tests of the within-patient difference for both studies were only reported as significant or not, without the numerical details of p-values.

In Albertsen et al. ,37 there was no statistically significant difference between the trial arms in the 6MWT at baseline, but, at 12 months’ follow-up, patients with a single-chamber atrial pacemaker walked significantly further than patients with a dual-chamber pacemaker (Table 18). 37 Although the result was statistically significant and the mean difference just reached the minimal clinically important difference of 54–80 m,46 there was substantial uncertainty around this value. One patient in the single-chamber atrial pacing group was lost to follow-up during the study, which may have had a small impact on the overall result.

Schwaab et al. 43 also showed a significantly better exercise capacity with single-chamber atrial pacing than with dual-chamber pacing for bicycle exercise duration and workload. However, Gallik et al. 34 did not detect a statistically significant difference between the pacing modes for a similar bicycle test. It is noteworthy that Gallik et al. 34 evaluated pacemakers over a markedly shorter test period, with both pacing modes tested on the same day with 0.5 to 1 hour’s rest between modes, which may partly explain the difference in result between Gallik et al. 34 and Schwaab et al. 43 However, as with the 6MWT in Albertsen et al. , there was substantial uncertainty around the result of the exercise testing in both Schwaab et al. 43 and Gallik et al. 34

Further surgery

One of the outcomes in the DANPACE trial19 was pacemaker reoperation during follow-up. The need for pacemaker reoperation was decided by the physician in charge of follow-up. Significantly fewer participants in the DDDR arm than in the AAIR arm needed a reoperation (OR 0.48, 95% CI 0.36 to 0.63; Table 19) during the relatively long-term follow-up in the DANPACE trial [5.4 years (SD 2.6 years)]. 19 The difference in reoperations between the pacing modes was statistically significant with only modest uncertainty around the result. Need for surgical change of pacing mode was the only reason for reoperation for which the difference between the pacing modes was statistically significant (Table 20). The reported unadjusted and adjusted HRs for AAIR compared with DDDR pacing are also listed in Table 19.

Adverse effects of pacemaker implantation

Although all three parallel-group RCTs randomised patients by device, only two of the trials reported data on adverse effects linked to pacemaker implantation: Albertsen et al. 37 and DANPACE. 19 Albertsen et al. looked at complications around implantation; in the 50 randomised participants there were no lead displacements, infections or haematomas at pacemaker implantation (Table 21). 37 No other peri- or post-operative adverse effects of pacemaker implantation were reported. One patient in the single-chamber atrial pacing group was lost to follow-up during the study and could have had an adverse event after withdrawing from the trial.

The DANPACE trial19 did not report adverse effects at implantation at randomisation; however, the indications for reoperations during follow-up were detailed. Of 1415 patients, 240 underwent one or more reoperations during the follow-up period (see Table 19). The more frequent indications for reoperation were battery depletion, lead complications and need for change of pacing mode (see Table 20). Less common adverse effects leading to reoperation were surgical or mechanical complications, infection, skin erosion or device failure. The only indication that was significantly different between the dual-chamber and single-chamber atrial pacemaker arm was surgical change in pacing mode.

Health-related quality of life and symptoms

Quality of life was studied in the crossover trials of Lau et al. 35 and Schwaab et al. 43 Both studies used different instruments to measure symptoms, QoL and functional status. Lau et al. 35 used VAS for general well-being; the SAS functional questionnaire for physical capacity (described in Chapter 2, Stakeholder’s submissions); the 12-item General Health Questionnaire (12-GHQ), symptom questionnaire; and the somatic symptoms inventory (SSI) adapted for local use from the Bradford Somatic Inventory. The 12-GHQ is a measure of current mental health. Each item is rated on a four-point scale (less than usual, no more than usual, rather more than usual or much more than usual). The symptom questionnaire assessed the incidence and frequency of dyspnoea, palpitations, dizziness, chest pain, sleep disturbance and neck pulsations, rated between 1 (all the time) and 5 (never). The SSI measures adequacy of daily life activities, emotional adjustment, social interactions (frequency, range and quality), work adjustment, sleep, fatigue and appetite.

Schwaab et al. 43 used three different self-administered questionnaires relevant to this review: VAS for general well-being, physical, emotional and cognitive functioning; the VAS Karolinska questionnaire including 16 questions on cardiovascular symptoms relevant to pacemaker patients; and the SAS functional questionnaire.

Lau et al. 35 presented IPD for general well-being; however, for all other outcomes of interest, Lau et al. 35 and Schwaab et al. 43 presented data for only the individual treatment periods. The results of paired t-tests of the within-patient difference were reported only as significant or not, without the numerical details of p-values.

Caro et al.50

Caro et al. 50 estimated the cost–utility of dual-chamber pacemakers compared with single-chamber ventricular pacemakers in people with bradycardia as a result of SND or AV block. The analysis was carried out from a UK NHS perspective, using costs from 2003 discounted at 6% per annum over a 5-year time horizon; benefits, namely quality-adjusted life-years (QALYs), were discounted at a rate of 1.5% per annum.

Castlenuovo et al.49

The review and economic evaluation carried out by Castelnuovo et al. 49 informed NICE’s TA88, an MTA of which this review is in part an update. Consequently, the scope of the review reported by Castelnuovo et al. 49 was broader than the decision problem that is the focus of this MTA. Castelnuovo et al. 49 considered the clinical effectiveness and cost-effectiveness of dual-chamber versus single-chamber pacemakers for the management of bradycardia as a result of SSS and/or AV block. The analysis was carried out over 5- and 10-year time horizons from the perspective of the UK NHS. Costs (from 2003) were discounted at a rate of 6% per annum and benefits (QALYs) were discounted at a rate of 1.5% per annum, according to the NICE reference case of the time.

Model structure and assumptions

A series of Microsoft Excel-based (Microsoft Corporation, Redmond, WA, USA) Markov models were used to assess the cost-effectiveness of three different treatment choices:

1. dual-chamber pacemakers versus single-chamber ventricular pacemakers in people with AV block

2. dual-chamber pacemakers versus single-chamber ventricular pacemakers in people with SSS

3. dual-chamber pacemakers versus single-chamber atrial pacemakers in the SSS population.

The patient population of each analysis were assumed to be homogeneous, that is, either all patients had SSS or all had AV block. A simplified outline of the structure of the Markov models is displayed in Figure 8. Health states were designed to reflect disease course and potential outcomes following pacemaker implantation. People receiving a dual-chamber pacemaker were initially exposed to the risk of perioperative and subsequent complications and over the longer term to the risk of AF, HF or stroke. People who developed AF were exposed to a higher risk of HF or stroke than people without AF. In addition, people with a dual-chamber device (either initially or following upgrade) who went on to develop AF had their device reprogrammed to act as a single-chamber ventricular device. Mortality was also accounted for within the model, with people exposed to the risk of death from any cause, perioperative mortality, death from stroke or death from HF. The subsequent disease pathways for people experiencing HF or stroke were not explicitly modelled; broad assumptions regarding the ongoing cost and utility associated with these health states were made.

FIGURE 8.

Overview of model structure used in NICE’s TA88. 19 AVB, atrioventricular block.

People receiving a single-chamber pacemaker were exposed to the same risks as people receiving a dual-chamber pacemaker. However, those receiving a single-chamber ventricular pacemaker were also at risk of developing pacemaker syndrome, which could result in upgrade to a dual-chamber pacemaker. Similarly, people receiving a single-chamber atrial pacemaker were at risk of developing AV block, necessitating upgrade to a dual-chamber pacemaker.

Clarke et al.51

Clarke et al. 51 carried out a retrospective follow-up of patients implanted with single-chamber atrial pacemakers in a single centre between 1992 and 1996. The aim of the study was to ascertain the rate of development of AV block and estimate potential cost-saving from use of single-chamber atrial pacing instead of dual-chamber pacing in people with SND and no AV block. A retrospective analysis of case notes identified 81 patients implanted with a single-chamber atrial pacemaker between 1992 and 1996. Of these, eight patients died during the analysis period and case notes were unobtainable for five patients. Fifteen (22%) of the 68 patients for whom case notes were available required a revision procedure. Of these, 10 patients (67%) required revision as a result of complications or manufacturer recall and four (5.8%) patients required revision following the development of AV block. Based on these data and on the cost of implantation (£2885 and £3844 for single-chamber and dual-chamber pacemakers, respectively), Clarke et al. estimated cost-savings of £103,000 a year associated with the use of single-chamber atrial pacemakers instead of dual-chamber pacemakers in people with SND and no AV block.

Osman et al.55

The aim of the study reported by Osman et al. 55 was to assess the safety and potential cost savings associated with same-day procedures (as opposed to procedures followed by an overnight stay) for implantation of new pacemaker devices (i.e. first pacemaker implants). The study used retrospective safety and cost data from a single centre to assess the level of complications associated with a same-day procedure in patients scheduled for a new pacemaker implant.

Sutton et al.48

Sutton et al. 48 carried out a comparison of costs and benefits associated with atrial (AAI/DDD) versus ventricular (VVI) pacing in people with SSS or AV block. The analysis was carried out over a 10-year time horizon; however, no discounting was applied to costs or benefits. A generic unit of currency, based on 1991 UK prices, was used to inform all estimates of cost. Furthermore, the perspective of the analysis was not explicitly stated.

Oddershede et al.54

Oddershede et al. 54 considered the cost–utility, from a Danish health-care system perspective, of dual-chamber versus single-chamber atrial pacemakers in people with SSS and preserved AV conduction. Costs and benefits (QALYs) were discounted at a rate of 3.5% per annum over a period of 60 years to estimate lifetime costs and effects. A Markov model comprising four health states (well, first stroke, second stroke and dead) was used to analyse cost-effectiveness. Patients without a history of stroke entered the model in the health state well, whereas those with a history of stroke entered the model in the health state first stroke. The model had monthly cycles and allowed patients to develop up to seven strokes.

The authors estimated cost-effectiveness using three approaches: an adjusted and an unadjusted approach using data from the DANPACE trial19 and an adjusted pooled approach using data from the DANPACE trial19 pooled with two other Danish clinical trials. 41,45,83 Patients were divided according to predicted survival probability into three groups to account for heterogeneity, with group 1 categorised as the group at highest risk of death and group 3 at the lowest risk of death. A Cox proportional hazards model was used to estimate the characteristics of the three groups. Survival probability was reported to change with:

• age

• sex

• previous myocardial infarction

• history of AF

• proportion of patients entering the model at the health state first stroke.

The cost-effectiveness of DDDR was assessed by calculating net monetary benefit, which combines lifetime costs and QALYs. Therefore, a net monetary benefit greater than zero indicated that DDDR was cost-effective. A probabilistic sensitivity analysis was carried out to test the robustness of the results.

Deniz et al.52

Deniz et al. 52 considered the cost–utility, from an Italian government perspective, of dual-chamber versus single-chamber ventricular pacemakers in people with bradycardia as a result of SND or AV block. Costs and benefits (QALYs) were discounted at a rate of 3% per annum over a 5-year time horizon. The authors adapted a discrete event simulation that was originally developed to assess the cost–utility of dual-chamber pacemakers for the management of bradycardia as a result of SND or AV block in the UK50 to consider an Italian government perspective. The model structure used in the analysis reported by Deniz et al. 52 is identical to that described by Caro et al. 50 Similarly, outcome data used in the economic analysis reported by Deniz et al. 52 are identical to those used to inform the economic analysis reported by Caro et al. 50

Mahoney47

The study reported by Mahoney47 compares (without the use of modelling) costs and outcomes associated with single-chamber ventricular (VVI) pacing versus single-chamber atrial (AAI) pacing and dual-chamber (DDD) pacing; the patient population is not specified. The study purports to assess the long-term costs of care for patients receiving each type of pacing considered. However, the time horizon of the analysis is not stated and no discounting is applied to costs or benefits.

O’Brien et al.53

O’Brien et al. 53 carried out an economic evaluation alongside the CTOPP. The CTOPP considered the effects of physiological (dual-chamber or single-chamber atrial) pacing compared with VP in people without chronic AF indicated for initial pacemaker implantation for the management of symptomatic bradycardia. 62 Resource use and cost data were collected from a subset (n = 1058) of patients enrolled in CTOPP (n = 2568) and adjusted for censoring using methods described by Lin et al. 85 Life expectancy and the number of AF episodes by type of pacing (physiological vs. ventricular) were estimated from the full trial population of CTOPP.

Rinfret et al.56

Rinfret et al. 56 assessed the cost–utility of dual-chamber (DDDR) versus single-chamber ventricular (VVIR) pacemakers in people paced for SSS. Analyses was carried out from a US societal perspective across a within-trial time horizon (4 years) and a lifetime time horizon, with costs and benefits discounted at 3% per annum.

Wiegand et al.58

Wiegand et al. 58 considered the costs and benefits of single-lead VDD pacemakers compared with DDD pacing in patients with AV block and normal sinus node function. The analysis was carried out based on clinical data gathered in a single-centre prospective study, over an average time horizon of 42 months. No discounting was used and HRQoL was not considered.

Ray et al.57

The aims of the study carried out by Ray et al. 57 were twofold: first, to assess the impact of the 1990 BPEG guidelines on clinical practice and, second, to assess the impact of full guideline adherence on cost. An audit of patients undergoing first pacemaker implant for the period of March 1990 to August 1991 was carried out and these data were used to assess changes in clinical practice and average device costs.

Probabilities of reoperation

Nielsen et al. 19 reports the number of patients requiring reoperation for various indications, including battery depletion, need for surgical change of mode of pacing, lead complications, surgical or mechanical complications, infection, skin erosion and device failure. Overall, reoperation is statistically significantly (single-chamber atrial vs. dual-chamber pacing, adjusted HR 2.00, 95% CI 1.54 to 2.61; p < 0.001)19 different between treatment arms, with a higher rate of reoperation in people receiving a single-chamber atrial pacemaker (see Model structure).

As discussed in Model structure, of conditions requiring reoperation, a statistically significant difference between pacemaker types was identified only for surgical change of mode of pacing, with a significantly larger percentage of people in the AAIR treatment arm requiring reoperation compared with the DDDR treatment arm (9.3% AAIR vs. 0.6% DDDR; p < 0.001) over an average follow-up period of 5.4 years.

For people implanted with single-chamber atrial devices, the need to change pacing mode is predominantly a result of the development of AV block, which requires an upgrade to a dual-chamber device. For the model, the difference in event rates for reoperation was used to estimate the risk of patients with a single-chamber atrial device developing AV block per patient per month, and this was applied as a constant risk for the time horizon covered by the model.

To derive the probabilities required for the economic model, the difference in monthly event rates in the single-chamber and dual-chambers arms was calculated using:

where r = event rate, n_s = number of events in the single-chamber atrial pacemaker arm, N_s = number of patients receiving a single-chamber atrial pacemaker, n_d = number of events in the dual-chamber pacemaker arm and N_d = number of patients receiving a dual-chamber pacemaker.

This monthly rate was then converted into a monthly probability, using a standard formula:

where p = monthly probability, r = event rate and t = time (months).

This resulted in a monthly rate of 0.142% and a monthly probability of 0.142%.

The decision to use this approach was based on the need to capture the uncertainty associated with reoperation for one-way sensitivity analysis (OWSA) and probabilistic sensitivity analysis (see Approach to uncertainty). However, as the Kaplan–Meier plot presented by Nielsen et al. 19 suggests a non-linear decline in reoperation, as opposed to a constant rate, an alternative approach using reoperation as a time-dependent parameter was explored as a structural sensitivity analysis (see Approach to uncertainty).

At 96 months post implantation, based on an 8-year battery life, all patients who had not yet experienced reoperation or developed AF were assumed to receive a replacement dual-chamber device.

Fleischmann et al.94

Fleischmann et al. 94 assess the impact of AF on the QoL and functional status of patients enrolled in the MOST. 68 Patients enrolled in the MOST had SSS and were randomised to either DDDR or VVIR pacing. The average age of patients was 73 years, 52% were male, approximately 45% had paroxysmal AF and 20% had some form of AV block. HRQoL was assessed using the SF-36 general health survey, and utilities were elicited using standard TTO methodology. Functional status was assessed using the SAS. Measurements were collected at baseline and 3 and 12 months post implantation in 1841 patients, who were subdivided into:

• those without AF (n = 1737)

• those who developed paroxysmal AF, but not chronic AF (n = 75)

• those with chronic AF (n = 29).

The changes observed in each measure (SF-36, TTO and SAS) between baseline and 12 months and between 3 months and 12 months of follow-up were analysed, with adjustments for age, sex, history of AF, history of HF, treatment arm and baseline QoL score. In addition, to avoid confounding as a result of crossover, the last observed scores of patients who crossed over from VVIR to DDDR pacing were carried forward for the remainder of the analytical time frame. A summary of the results reported by Fleischmann et al. 94 is presented in Table 43; statistically significant differences in utility, as assessed using TTO methodology, were identified.

Based on the analyses conducted, the authors concluded that ‘AF was not a major determinant of most QoL measures’. 94 However, statistically significant changes in the physical component of the SF-36 score and in the SAS measure of functional capacity indicate that the presence of AF may impair the physical improvement associated with pacemaker implantation.

Fleischmann et al.92

In this analysis of serial QoL measures elicited from people enrolled in the MOST,68 Fleischmann et al. 92 consider the impact of pacemaker implantation and pacemaker mode (DDDR vs. VVIR) on QoL. SF-36, SAS and TTO measures were used to assess QoL at baseline, and after 3 months and 12 months of follow-up, followed by yearly estimates; last observations were carried forward in people who crossed over from VVIR to DDDR pacing. Table 44 displays the TTO utilities, adjusted for age and sex, by pacing mode as presented by Fleischman et al. 92

The authors conclude that, irrespective of sex, the presence of HF or level of comorbidity, pacemaker implantation was associated with statistically significant improvements in QoL. Furthermore, although a small but measurable effect of pacemaker mode was noted with the SF-36, no significant differences were observed in TTO utility estimates.

Shukla et al.108

Shukla et al. 108 assessed the impact of pacemaker sensor type (accelerometer, piezoelectric or blended) on the QoL of patients enrolled in MOST. 68 The SF-36, SAS, 0–100 scale and TTO measures were used to elicit QoL and, in the case of TTO, to value QoL. Measures were taken at baseline, 3 months post transplant and yearly thereafter; adjustments were made for age, sex, pacing mode (DDDR vs. VVIR), follow-up time and baseline QoL. Patients implanted with a blended sensor device reported statistically significantly worse physical function at (physical function, p = 0.009; physical summary score, p = 0.039; and physical role function, p = 0.08) than patients with accelerometer or piezoelectric sensors. However, no other statistically significant differences in QoL were identified.

The authors concluded that patients implanted with blended sensor devices had lower physical function and absolute QoL scores. However, the authors considered that these observations may be a result of ‘clinical selection of the most sophisticated sensor for the most ill patient’. 108

Link et al.107

A subset (18.3%) of patients enrolled in the MOST68 and randomised to VVIR pacing went on to develop pacemaker syndrome according to pre-specified criteria. That is, they developed ‘either congestive signs and symptoms associated with retrograde conduction during VVIR pacing or a ≥ 20 mmHg reduction of systolic blood pressure during VVIR pacing, associated with reproducible symptoms of weakness, light[-]headedness, or syncope’. 107 Link et al. report the QoL (SF-36, SAS, TTO and 0–100 score) measured in these patients at baseline, prior to crossover and after crossover. Significant decrements in QoL were observed in six of the 10 SF-36 scales from baseline (i.e. pre implantation) to pre crossover (Table 45). Utility was also lower than at baseline, but the difference was not statistically significant. After crossover, statistically significant improvements were seen in all measures of QoL (see Table 47).

The authors concluded that ‘quality of life, as assessed by a variety of metrics, decreased at the time of diagnosis of pacemaker syndrome and improved after the pacemaker was reprogrammed to a physiologic mode. (However, a placebo effect cannot be truly ruled out, because neither patients nor physicians were blinded to the crossover status)’. 107

Lamas et al.68

Lamas et al. 68 report the results, including HRQoL, of the MOST. QoL was assessed with the SF-36, SAS and TTO measures. Table 46 presents the changes in QoL measures from baseline as reported by Lamas et al. 68

Similar to the conclusions of Fleischmann et al. ,92 the authors noted that, compared with VP, dual-chamber pacing provided significant improvements in six of the 10 SF-36 subscales, including the physical and mental component summary. However, no significant differences in utility, as assessed by TTO, were observed by pacing mode.

Lopez-Jimenez et al.10

Lopez-Jimenez et al. 10 report the results of the PASE study, which is a RCT with the primary endpoint of HRQoL. Patients enrolled in the PASE study were in sinus rhythm and indicated for permanent pacing as a result of bradycardia, and were randomised to VVIR or DDDR pacing mode. The average age of patients was 76 years, 59% were male, 57% had AV block and 28% had NYHA class III or IV HF. HRQoL was assessed using the TTO, SF-36 and 0–100 scoring systems. In addition to HRQoL, patients’ functional status was assessed with SAS. Measurements were collected at baseline and at 3, 9 and 18 months after enrolment. Table 47 summarises the estimates of HRQoL over time obtained by Lopez-Jimenez et al. 10

In order to assess the validity of the TTO measure, the authors carried out ‘known group validity tests’ in patients with and without CHF and in patients with and without stable angina. As expected, people with a history of CHF had a significantly lower utility than people with no history of CHF (0.64 vs. 0.78; p < 0.001). Furthermore, people with NYHA class III or IV had a significantly lower utility than people with NYHA class I or II (0.62 vs. 0.80; p = 0.0001).

Utility estimates were not adjusted for potential covariates such as age or sex. However, subgroup analyses suggest that improvements in utility following pacemaker implantation were consistent regardless of implantation diagnosis, pacing mode, sex, age, employment status or history of angina.

The authors concluded that pacemaker implantation improves HRQoL to a mean level close to that of the general population.

Device and implantation costs

The procedure costs (including hardware cost) associated with implantation of a single-chamber or dual-chamber device were obtained from a weighted average of episode costs associated with relevant HRG codes (NHS Reference Costs 2012–201395). Table 49 summarises the HRG codes used to inform each procedure cost used within the TAG’s economic model. Upgrade procedures were assumed to cost the same as an initial implantation of a dual-chamber device. Spell-level (rather than episode-level) data for each HRG code were used in sensitivity analysis (see Approach to uncertainty).

Device and implantation costs are applied at two points in the model at first implantation and at upgrade (i.e. to patients in the implant single-chamber atrial pacemaker or implant dual-chamber pacemaker health states).

Monitoring costs

Following pacemaker implantation, patients receive follow-up checks from a cardiologist (WF01C, cardiologist non-admitted, non-face-to-face attendance, follow-up, £86). 94 Based on expert clinical opinion, initial follow-up is assumed to be 1 week after implant, with a second follow-up carried out at 2 months post implantation and subsequent annual visits. 110 Therefore, within the model, the cost of a follow-up visit is applied on entry into the implant single-chamber atrial pacemaker, implant dual-chamber pacemaker, with single-chamber atrial pacemaker and with dual-chamber pacemaker health states. The cost of a follow-up visit is also applied annually to all patients in the with single-chamber atrial pacemaker and with dual-chamber pacemaker health states.

Episode costs

Over the course of the TAG’s economic model, patients are exposed to the risk of HF and stroke, with or without the presence of AF. In the absence of AF, the occurrence of HF or stroke is associated with a cost that is based on a weighted average of episode-level costs (spell-level costs are used in sensitivity analysis, see Approach to uncertainty) associated with relevant HRG codes (Table 50). In the presence of AF, however, the episode cost of stroke is assumed to be £11,275 based on evidence from the OXford VASCular (OXVASC) population-based cohort study reported by Luengo-Fernandez et al. 98 No evidence was identified indicating that the episode cost of HF would differ in the presence of AF.

The episode cost of HF is applied to patients entering the HF and the AF and HF health states. The episode cost of stroke is applied to patients entering the stroke health state and the episode cost of stroke following AF is applied to patients entering the AF and stroke health state.

In people with dual-chamber pacemakers, the onset of AF may be associated with a need to reprogramme the device to act as a ventricular pacemaker. In line with clinical expert opinion and assumptions made in NICE’s TA88,18 this cost comprises a cardiological consultation (WF01C, cardiologist non-admitted, non-face-to-face attendance, follow-up, £86)94 and electrocardiography (DIAGIMDA, simple echocardiography, 19 years and over, £41). 94 People with single-chamber atrial pacemakers who develop AF may also require VP, in which case the single-chamber atrial device will need to be replaced with a single-chamber ventricular device. The cost associated with device replacement is assumed to be equivalent to that of initial single-chamber implantation (£1875).

Based on expert clinical opinion, the cost of reprogramming and of device replacement is applied to one-third of people developing AF from the with dual-chamber pacemaker and with single-chamber atrial pacemaker health states. The impact of this assumption was tested in sensitivity analyses.

Long-term costs

Following the onset of HF, stroke or AF, patients are assumed to accrue costs over the long term, for example medication, hospitalisation and primary-care costs. For people with HF, these costs were determined from national prevalence and cost statistics reported in the British Heart Foundation’s Coronary Heart Disease Statistics 2012 publication. 96 The following data were extracted from the British Heart Foundation’s report:96

• total UK direct health-care costs of CVD, 2009: £8,680,892,000

• UK prevalence of HF, 2011: 0.90% in men and 0.70% in women, total 160,719 cases

• UK prevalence of CVD, 2007–10 (Table 51).

Based on the prevalence data presented in Table 51, the relative prevalence of HF as a percentage of CVD was calculated (7.53% for men and 6.39% for women; average 6.96%). Thereafter, the 2011 UK direct health-care costs of CVD (£8,680,892,000) and of HF (£632,586,584) were estimated, resulting in a per-person cost of HF of £3316 per annum (£276.34 per cycle) at 2013 prices. Full calculation details are available in Appendix 6, Calculation of long-term care costs associated with heart failure. The long-term costs associated with HF were applied monthly to people residing in the HF and the AF and HF health states.

For people experiencing stroke, the costs of hospitalisation estimated from the OXVASC population-based cohort study reported by Luengo-Fernandez et al. 98,109 were used. The costs of medication and primary care reported by Townsend et al. 96 were used to inform the base case. Table 52 summarises the unit costs used to inform long-term costs for people experiencing stroke.

The long-term costs associated with people with stroke and people with AF and stroke were applied monthly to people residing in the stroke and the AF and stroke health states, respectively.

In people with AF, long-term costs of primary care and hospitalisation were identified from a predictive study carried out by Stewart et al. ,99 who evaluated the UK health and social services cost of AF in 1995, and projected costs to 2000 based on epidemiological trends. Table 53 summarises the calculation of per-person primary care and hospitalisation costs, based on information presented by Stewart et al. 99

In addition to the costs of primary and hospital care, people with AF are assumed, in line with current clinical guidance,112 to receive effective anticoagulation therapy with apixaban, dabigatran etexilate (hereafter referred to as dabigatran), rivaroxaban or warfarin. Analysis of 2013 prescribing data indicates that, in primary care, the current market shares of apixaban, dabigatran, rivaroxaban and warfarin are 0.0004%, 0.47%, 0.15% and 99.38%, respectively. 113 However, data from the Hospital Prescribing Audit Index database indicate that, since 2012, the use of apixaban, dabigatran and rivaroxaban has changed by +179,873.2%, +132.4% and –21.5% , respectively. 114 Given that these therapies have recently been recommended for use in the prevention of stroke and systemic embolism in people with non-valvular AF, it is likely that market share will continue to change over the coming years. Therefore, although current market share estimates are used in the base case, the model is set up to have market share as a user input and different market share scenarios are assessed in sensitivity analyses (see Approach to uncertainty). Table 54 summarises the calculation of a per-person monthly cost of oral anticoagulation used in the TAG’s base-case model.

The long-term cost associated with people with AF and stroke was applied monthly to people residing in the AF, the AF and stroke and the AF and HF health states.

Probabilistic

The TAG economic model has been constructed probabilistically, that is, to account simultaneously for the impact of parameter uncertainty on the cost-effectiveness results. Probability distributions were assigned to parameters used within the model, from which values have been simultaneously sampled 1000 times. Table 55 summarises the type of distribution, and rationale for selection of the distribution, used to inform each group of parameters; full details of distributional specifications are provided in Table 36.

Deterministic

A series of OWSAs was carried out by using the lower and upper limits of the 95% CIs of the following parameters to assess their the impact on the ICER: age; efficacy values; utility values; costs; all-cause mortality; and HF hospitalisation. The estimates for the upper and lower 95% CIs are displayed in Table 36. Each key parameter was alternately assigned a low and high value and the deterministic cost-effectiveness results using this value were recorded.

Structural sensitivity analysis

The base-case analysis assumes a time horizon of 10 years, which requires an extrapolation of the data available from the pivotal RCT, DANPACE, beyond the typical duration that patients were followed up within the trial (5.4 years). 19 As a structural sensitivity analysis, the time horizon was reduced to 5 years to assess the degree of impact this extrapolation may have had on the base-case results.

In addition, as discussed in Treatment effectiveness, the risk of reoperation because of AV block was based on reoperation because of need for surgical change of mode of pacing, where it was found to be significantly higher in the AAIR treatment arm compared with the DDR treatment arm (9.3% AAIR vs. 0.6% DDDR; p < 0.001) over an average follow-up period of 5.4 years. However, the Kaplan–Meier plot (Figure 11) suggests a non-linear decline in reoperation, as opposed to a constant rate. In order to test the assumption of reoperation as a constant risk in the base case, a structural sensitivity analysis was undertaken based on the Kaplan–Meier data.

FIGURE 11.

Time-to-event curve for freedom from reoperation from the DANPACE trial. 19 Unadjusted p-value (log-rank test) shown.

In the structural sensitivity analysis, while residing in the with dual-chamber pacemaker and with single-chamber atrial pacemaker health states, patients are at risk of experiencing complications that require a reoperation, such as lead displacement or device failure. The risk of reoperation is time dependent and all patients requiring reoperation are assumed to be implanted with a dual-chamber device, regardless of the reason for reoperation or which pacemaker was originally implanted. The assumption that all patients requiring reoperation receive a dual-chamber device is based on reoperation data collected in the DANPACE trial. 19 As in the base case, only one instance of reoperation was permitted within the model time horizon.

The monthly probability of reoperation, by treatment arm, estimated from Kaplan–Meier data presented by Nielsen et al. 19 can be found in Appendix 7, Monthly probability of reoperation by treatment arm. These probabilities were estimated through digitisation of the Kaplan–Meier plot using the freely available online software WebPlotDigitizer (version 3.8, Ankit Rohatgi, Austin, TX, USA; http://arohatgi.info/WebPlotDigitizer/). The digitisation process provided monthly estimates of the ‘survival’ function S(t), from which the monthly probabilities of reoperation were calculated using the following formula:

where t = time (months) and p(re-operation)_t = probability of re-operation at time t.

Scenario analysis

Various assumptions have been made in the construction of the TAG’s base-case model. Where possible, these have been tested in scenario analysis. Table 56 lists the scenario analyses carried out by the TAG, the parameters used to inform these scenarios and the rationale for each analysis.

Probabilistic sensitivity analysis

Using a time horizon of 10 years, the results of the probabilistic analysis are presented in Figures 12 and 13. Probabilistic sensitivity analysis revealed that, in the majority (66.00%) of cases, implanting patients with dual-chamber pacemakers resulted in greater costs and greater QALYs than implanting single-chamber atrial pacemakers. Furthermore, dual-chamber pacemakers produced more QALYs at a lower cost in 24.1% of cases and were dominated by single-chamber atrial pacemakers in 9% of cases. At a WTP threshold of £20,000, the probability of dual-chamber pacemakers being cost-effective is 72.9%, which increases to 78.7% at a WTP threshold of £30,000.

FIGURE 12.

Scatterplot of cost-effectiveness results for dual-chamber pacemakers vs. single-chamber atrial pacemakers using a time horizon of 10 years (dark-green line indicates threshold of £20,000 per additional QALY, light-green line indicates threshold of £30,000 per additional QALY). PSA, probabilistic sensitivity analysis.

FIGURE 13.

Cost-effectiveness acceptability curve for dual-chamber pacemakers vs. single-chamber atrial pacemakers using a time horizon of 10 years.

Structural sensitivity analysis

Table 58 presents the results of the structural sensitivity analysis in which the time horizon was reduced from 10 years to 5 years. The difference in costs in the deterministic results falls to £224.53, with a reduction in QALYs accrued to 0.02. This results in an increase in the deterministic ICER for the base case from £6506 to £14,261. Similarly, the probabilistic results demonstrate a fall in the incremental costs to £228.97 and a reduction in the incremental QALYs gained to 0.02, resulting in an increased ICER of £13,837. These results are perhaps to be expected; a halving of the time horizon results in roughly a halving of the difference in incremental QALYs and roughly a doubling of the resulting ICER.

The results of the probabilistic sensitivity analysis when the time horizon is reduced to 5 years are presented in Figures 14 and 15. At a WTP threshold of £20,000 the probability of dual-chamber pacemakers being cost-effective is 55.3%, which increases to 64.0% at a WTP threshold of £30,000.

FIGURE 14.

Scatterplot of cost-effectiveness results for dual-chamber pacemakers vs. single-chamber atrial pacemakers using a time horizon of 5 years (dark-green line indicates threshold of £20,000 per additional QALY, light-green line indicates threshold of £30,000 per additional QALY). PSA, probabilistic sensitivity analysis.

FIGURE 15.

Cost-effectiveness acceptability curve for dual-chamber pacemakers vs. single-chamber atrial pacemakers using a time horizon of 5 years.

Table 59 shows the results of the structural sensitivity analysis when risk of reoperation was taken from Kaplan–Meier data instead of using a constant risk. The incremental costs of dual-chamber pacemakers decreased to £136.43 compared with the deterministic base case whereas QALYs remained the same. This caused the ICER to decrease to £3425 per QALY.

One-way sensitivity analysis

As discussed in Approach to uncertainty, in addition to probabilistic sensitivity analysis, OWSA was carried out on the following parameters: age, clinical outcomes, health state utility values and all-cause mortality. The full results of these analyses are presented in Appendix 5, One-way sensitivity analysis. The TAG notes that many of the parameters tested in sensitivity analysis had minimal impact on the deterministic cost-effectiveness results and, therefore, the parameters that the ICER is most sensitive to are presented in Figure 16.

FIGURE 16.

Tornado diagram of parameters to which the cost-effectiveness of dual-chamber pacemakers vs. single-chamber atrial pacemakers is most sensitive. DC, dual chamber; SC, single chamber.

The ICER of dual-chamber pacemakers versus single-chamber atrial pacemakers was most sensitive to probability of HF in patients implanted with single-chamber atrial pacemakers. Dual-chamber pacemakers become dominated when the minimal probability is applied to single-chamber atrial pacemakers (i.e. when HF was less likely to occur in patients with single-chamber atrial pacemakers than in those with dual-chamber pacemakers). The ICER was also sensitive to implant/implantation costs for both dual-chamber and single-chamber atrial pacemakers.

Scenario analysis

As discussed in Approach to uncertainty, a series of scenario analyses was conducted to test the robustness of the results to alternative sources for parameter estimates or testing broader assumptions (e.g. reprogramming/device replacement in patients developing AF) within the model. The results of the scenario analyses are depicted in Table 60.

Assuming no difference in the risk of developing HF with the two types of implant almost quadruples the ICER, increasing it to £22,213 from a base-case value of £6506.

The only outcome from the meta-analyses conducted in Chapter 3, Results that could be implemented in the economic model was stroke. However, utilising the risk of stroke from the TAG’s meta-analysis (see Chapter 3, Assessment of effectiveness) had only a modest impact on the ICER, increasing it to £6438.

Using spell-level costs for pacemaker implantation increased the costs of dual-chamber pacemaker implantation more than the costs of single-chamber atrial pacemaker implantation. This resulted in a modest increase in the incremental costs of dual-chamber pacemakers, resulting in a slightly higher ICER (£7605 compared with £6506). The alternative cost of HF was substantially higher than in the base case and increased the incremental difference in costs between interventions. This resulted in a slightly higher ICER compared with the base case (£7140 compared with £6506). The lower alternative cost of stroke from NICE’s TA8818 had a more modest effect on the incremental cost of interventions and a relatively modest change in ICER compared with the base case (£3476 compared with £6506). Similarly, the alternative cost per episode of stroke from Saka et al. 116 had little impact on the incremental cost or the resulting ICER compared with the base case (£5513 compared with £6506).

The proportion of patients experiencing AF resulting in either reprogramming or replacement of their pacemaker was estimated as one-third in the base case, based on advice from clinical experts. This was tested in two extreme scenario analyses in which it was assumed that reprogramming/replacement was required either in no one or in 100% of patients. These two scenarios had a pronounced impact on the resulting ICERs compared with the base case; the ICER increased to £10,872 when it was assumed that reprogramming/replacement was required in 0% of people, and dual-chamber pacemakers became dominant when this was number was set to 100% (i.e. less costly and more effective).

Varying the discount rate from 3.5% in the base case to either 0% or 6% had a modest impact on the ICER. Although costs and benefits increased overall at a discount rate of 0%, the ICER was reduced to £5045. Similarly, although increasing the discount rate to 6% decreased the cost and benefits overall, the impact on the ICER was an increase to £6938.

The final individual scenario analysis undertaken was to increase the proportion of prescribing of the novel oral anticoagulants (NOACs) to a more even level with warfarin. This was achieved by setting the market share for warfarin to 55% and the NOAC market share to 45% (evenly distributed in three blocks of 15% to apixaban, dabigatran and rivaroxaban). This resulted in an overall increase in costs but a reduction in the incremental cost between the two interventions and a modest reduction in the ICER to £4942 compared with £6506 in the base case.

The results of the individual scenario analyses suggest that the base-case ICER is robust to changes in costs of pacemaker implantation, stroke, discount rate and market share of NOACs. In addition, although varying the costs of HF and the proportion of patients whose pacemaker had to be reprogrammed or replaced because of AF to extreme values had a more pronounced impact on the resulting ICER; it was still well below the NICE threshold value of £20,000. In only one scenario, in which it was assumed that there is no difference in the development of HF, did the ICER exceed £20,000, and then only by £2213.

A cumulative worst-case scenario is depicted in Table 61, in which all efficacy and cost scenario analyses found to increase the ICER beyond the base case have been combined. This results in an ICER of £48,738 compared with the base-case ICER of £6506 at 10 years. However, if it is again assumed that the proportion of patients whose pacemaker has to be reprogrammed/replaced because of AF is one-third (as in the base case), the cumulative impact of the other adjustments result in an ICER of £27,918 at 10 years.

Overall, the ICER increases beyond £20,000 but remains below £30,000 if, in the cumulative worst-case scenario, it is assumed either that reprogramming/device replacement because of AF is required in 0% patients or that the pacemaker has no impact on HF. It exceeds £30,000 only when both scenarios are included in the cumulative worst-case scenario.

The overall adjusted HR for risk of developing HF used in the base case indicates a non-significant increase in risk with single-chamber atrial pacemakers compared with dual-chamber pacemakers (HR 1.09, 95% CI 0.88 to 1.35). 40 Based on feedback from our clinical experts, and as was assumed in Oddershede et al. ,54 we conducted a scenario analysis assuming that there was no difference in risk of HF based on implanted device (i.e. HR 1.00).

The results of the scenario analysis and the OWSA highlight how sensitive the results are to risk of HF, with dual-chamber pacemakers being considered cost-effective or dominated by single-chamber atrial pacemakers depending on the data used. These results warranted further investigation into HF, for which we assessed the subgroups analysed from the DANPACE trial. 40

The subgroups identified as statistically significant in an analysis of risk of HF from the DANPACE trial19 were a result of age (p = 0.05); all other subgroups assessed were found to be statistically non-significant (p > 0.31). 40 We explored the impact of using the HRs for the subgroups based on age (patients > 75 years or patients ≤ 75 years) as additional scenario analyses. The results are depicted in Table 62.