Clinical and cost-effectiveness of left ventricular assist devices as destination therapy for advanced heart failure: systematic review and economic evaluation

Eligibility criteria

Searches

The following databases were searched initially from inception until 20 May 2020: Cochrane Library (CENTRAL), MEDLINE and EMBASE via Ovid. For any relevant systematic reviews Epistemonikos, the Cochrane Library of Systematic Reviews and MEDLINE and EMBASE were searched. Searches incorporated free text and index terms related to population and intervention, with no restriction by study design. All database search strategies are available (see Appendix 2). The search term combinations in the example search strategy applied to the bibliographic databases were formulated in the standard way for a review and then augmented to ensure the strategy was sensitive to capturing studies known to the reviewers while keeping the yield to manageable numbers of records.

There was no restriction by date or language of publication on searches. Reference lists of relevant systematic reviews and included primary studies were checked for additional primary studies. Grey literature (e.g. institutional reports) was sought from key organisations. Conference abstracts were included if published within the previous 3 years of the search date. Ongoing and recently completed trials were searched using the World Health Organization (WHO) clinical trials portal.

Data and reports published from relevant registries were also identified from our searches. Further targeted searching was performed to identify publications that were not found during the searches.

Search updates were carried out from April 2020 until 11 January 2022. Searches for registry reports via relevant website lists (e.g. www.uab.edu/medicine/intermacs/research/publications) were undertaken at the same time as the database searches.

Study selection

All records received from the literature searches were initially entered into EndNote X9 (Clarivate Analytics) to facilitate removal of duplicates. 42 Records were entered into Covidence for screening and selection. 43 Title and abstracts were screened for potential relevance using the study eligibility criteria.

Where it was not clear if DT patients were included in the study or if any DT data were reported from the abstract alone, the full text of the study was sought. Full texts were retrieved for any potentially relevant records and checked for eligibility.

All stages of the study selection were undertaken by two reviewers independently and disagreements were resolved by third reviewer or consensus. Reasons for exclusions were recorded via Covidence and within an Excel spreadsheet.

Search results for both the cost and clinical effectiveness reviews were combined within the same EndNote and Covidence databases. During screening and selection, appropriate tags were assigned to potentially relevant records to identify them as either relevant for the clinical or cost-effectiveness review, or both.

Data extraction

Risk-of-bias assessment

Risk-of-bias assessment was carried out by one reviewer and checked by a second. Tools appropriate for study design were used to assess risk of bias. For RCTs and non-RCTs, version 2 of the Cochrane risk-of-bias tool was used. 44 The randomisation domain was not applicable for non-randomised trials, and it was acknowledged that blinding is not possible in most surgery trials. Risk-of-bias assessment was carried out for the key outcomes of survival and QoL. These were considered as key outcomes as they were consistently reported across trials of this type and were important outcomes for the economic evaluation. Results of the risk of bias were presented in tabular format.

Data synthesis

Consideration, data extraction and reporting of the evidence were based on a hierarchical approach. RCTs and controlled non-randomised trials were considered in the first instance. Registry reports and uncontrolled observational studies were used to supplement findings where gaps were evident. Inclusion of studies with overlapping patient data was avoided where possible. However, all studies, regardless of design, were included and reported in the review.

Clinical trial participants are not eligible to be included in LVAD registries (such as INTERMACS and IMACS). To assess whether patients in trials differed from those in registries, exploratory analyses were undertaken comparing key population differences between the two. This was carried out where the data were available for key outcomes including survival and QoL. Additionally, changes over time in each population were assessed.

Overlap of participants between single, multicentre observational studies and registries was also considered. Many of these centres in the USA also contribute data to registries, including INTERMACS and by default, IMACS. To avoid overlap of participant data reported, data were only considered from studies that clearly did not contribute to the INTERMACS registry (as defined by the list of participating centres on the INTERMACS website www.uab.edu/medicine/intermacs/pedimacs/participating-centers-pedimacs).

Data were tabulated and analysed in a narrative approach in the first instance, firstly with comparative data stratified by device (e.g. HM3) and INTERMACS scores (where available), and then by outcome within this. Following this, non-comparative data were presented stratified by device type and INTERMACS scores (where available), and by outcome within this.

Due to large clinical and methodological heterogeneity expected in the data, no meta-analyses were performed.

Forest plots without pooled estimates were created for all data stratified by device type for each outcome category (survival, QoL, hospitalisations, major events and complications).

While all recent devices (e.g. HeartMate XVE™, HeartMate II, HeartWare HVAD, HM3) were included and analysed, priority was given to the HM3 device given the availability of devices in the UK and the recent withdrawal of the HeartWare HVAD device in June 2021.

Selection

Searches for both clinical and cost-effectiveness studies identified 12,153 articles. Following removal of duplicates, 9006 articles remained. There were 982 articles found to be potentially relevant following title and abstract screening, and full texts were sought and checked against the eligibility criteria. Following this, 240 articles from 134 studies met the criteria for the clinical effectiveness review and were included.

There were 6 trials (1 non-randomised, 5 RCTs), 86 observational studies, reports from 5 registries, 5 ongoing studies and 32 systematic reviews (included for citation checking). It should be noted that some of these studies were from the same centres and reports from the same registries were considered as one single study. There were 24 relevant articles included following citation checking, 21 of which were relevant to the clinical effectiveness review. Seven full-text articles could not be retrieved. A summary of the selection process is given in Figure 4. A list of full-text articles excluded with reasons is available in Appendix 3.

FIGURE 4.

Preferred Reporting Items for Systematic Reviews and Meta-Analyses flow diagram showing selection of studies. a, Only studies considered for the clinical effectivenessreview were excluded based on sample size.

Types of included studies

Studies included in the review were categorised as comparative trials, registry reports or single and multicentre observational studies.

HeartMate 3 data

The HM3 is currently the only option for LVAD-eligible patients in the UK following the withdrawal of the HeartWare HVAD. This section will describe the available data pertaining to the HM3 device.

Trial data

There was one RCT (MOMENTUM 3) undertaken in the USA that compared the HM3 device to the older HeartMate II device. There were 1028 participants included in the study: 515 randomised to the HM3 (317 of these DT) and 505 randomised to the HeartMate II (307 DT). The mean age for all included DT patients was 63 (SD 12) and the majority of included participants were INTERMACS level 3 (Table 1). The following section reports data by outcome from this RCT.

Survival

Survival in DT patients implanted with the HM3 device was 84% at 12 months and 77% at 24 months of follow-up (Figure 5). This is higher than any previously reported survival data in earlier generation devices at 24 months (see Appendix 5). Survival in HM3 DT patients was higher than the HeartMate II patients in the MOMENTUM study at 24 months (77% vs. 59%, respectively) with a hazard ratio (HR) of 0.87 (95% CI 0.63 to 1.2), though this was not considered a statistically significant difference. However, survival free of disabling stroke or reoperation to replace or remove a malfunctioning device at 24 months was significantly higher in patients in the HM3 group compared to the HeartMate II group (73% vs. 57%), with a HR of 0.61 (95% CI 0.46 to 0.81).

FIGURE 5.

Survival data in the HeartMate 3 at all reported follow-up time points.

Quality of life

While clear improvements in QoL were reported in DT patients with the HM3 device in both the KCCQ and the EQ-5D at 12 and 24 months compared to baseline, similar QoL improvements were also seen in the HeartMate II group. The mean visual analogue scale (VAS) summary score for the KCCQ improved from 40 at baseline to 69 at both 12 and 24 months in the HM3 group (Figure 6), compared to 39 at baseline increasing to 70 at 12 months and 68 at 24 months in the HeartMate II group. Furthermore, the EQ-5D VAS score improved from 48 at baseline in the HeartMate II group to 74 at both 12 and 24 months (compared to an increase from 51 at baseline to 77 at 24 months in HM3). Improvements in QoL appeared to peak at 12 months and remain stable at 24 months. Scores remained similar to those at 24 months at the end of study follow-up.

FIGURE 6.

Mean QoL scores in the HeartMate 3 at all reported follow-up time points.

Hospitalisations

Rehospitalisation days were reported to be fewer in HM3 patients compared to those with the HeartMate II device (median duration 15 vs. 22 days) over 24 months. The event rate of rehospitalisation per 100 patient-years was found to be significantly lower in the HM3 group versus the HeartMate II group (212 vs. 243, HR 0.88 95% CI 0.81 to 0.96, Figure 7 Hospitalisation rate per person-year in the HeartMate 3 at 24 months follow-up).

FIGURE 7.

Hospitalisation rate per person-year in the HeartMate 3 at 24 months follow-up.

Major events

Stroke rates were reported to be lower than seen in the literature previously in the HM3 and lower than the HeartMate II group over 24 months. Regarding any stroke, there were 8 events per 100 patient-years in the HM3 group (Figure 8) (11% had a stroke at 24 months, Figure 9) compared to 19 events in the HeartMate II group (RR 0.42 95% CI 0.29 to 0.62). This was similar for disabling stroke: 4 events per 100 patient-years for HM3 versus 7 events per 100 patient-years for HeartMate II with a RR of 0.59 (95% CI 0.34 to 1.03). There were no significant differences in rates of RHF or RHF requiring RVAD between the device groups.

FIGURE 8.

Event rate per 100 person-years in the HeartMate 3 at 24 months follow-up.

FIGURE 9.

Proportion of patients with major events in the HeartMate 3 at all reported follow-up time points.

Complications

Rates of pump thrombosis were found to be significantly lower in the HM3 group compared to the HeartMate II group with 1 event versus 12 events per 100 patient-years, respectively, and a RR of 0.1 (95% CI 0.04 to 0.24). On the other hand, DIs were common and not found to be different between the two arms.

While bleeding events were still an issue in MOMENTUM 3, they occurred less frequently in patients implanted with the HM3 compared to the HeartMate II (70 events vs. 103 events per 100 patient-years) with a RR of 0.68 (95% CI 0.59 to 0.78). More specifically, bleeding requiring surgery and GIB rates were both significantly lower in the HM3 group. The proportion of patients with complications and the complication rate are shown in Figures 10 and 11.

FIGURE 10.

Complication rate per 100 person-years in the HeartMate 3 at all reported follow-up time points.

FIGURE 11.

Proportion of patients with complications in the HeartMate 3 at all reported follow-up time points.

Functional status

The mean six-minute walking distance increased in HM3 patients in MOMENTUM from 137 m at baseline to 320 m at 24 months. However, the proportion of patients with NYHA Class I or II remained similar at baseline (78%) and 24 months (78%).

Indirect comparison of HeartMate 3 and medical management – survival

This section discusses the results of the NMA carried out to indirectly compare HM3 and MM for survival. Direct comparison data were taken from several of the RCTs described above (MOMENTUM, HeartMate II DT and REMATCH). The network diagram Figure 12 illustrates the comparisons available to enable the NMA.

FIGURE 12.

Network diagram.

Assessment of the transitivity assumption

Baseline patient characteristics in the three RCTs included in the NMA are shown in Table 9. While most patient and treatment characteristics appeared similar, it shows that IV inotropic drugs (and INTERMACs 1–3) were required by 68% in REMATCH, compared to 87% in MOMENTUM with HeartMate II in the middle at 79%. If the baseline INTERMACS level is an effect modifier for any of the comparisons made in the network, then this would break the transitivity assumption and introduce bias.

TABLE 9 - Table of characteristics of included intervention studies

HMII, HeartMate II; NR, not reported.

Study ID (no. publications)	Study design	No. participants (no. DT)	Intervention	Comparator	Mean age (SD)	Sex (n, % male)	INTERMACS Profile (n, %)	Length of follow-up (months)	Notes
REMATCH 200154 (10)	RCT	129 (All DT)	HeartMate XVE LVAD (n = 68)	Medical management (n = 61)	HeartMate 3 66 ± 9.1 Medical management 68 ± 8.2	HeartMate 53 (78) Medical management 50 (82)	NR	24 months (1 patient in LVAD group still alive at 30 months)
HMII DT 200955 (10)	RCT	200 (All DT)	HMII LVAD (n = 134)	HeartMate XVE LVAD (n = 66)	HMII 62 ± 12 HeartMate XVE 63 ± 12	HMII 108 (81) HeartMate XVE 61 (92)	NR	24 months	Seven papers are retrospective analyses of both the HMII DT and BTT trials and include HMII single arm only
ENDURANCE DT 201756 (22, all ENDURANCE papers)	RCT	445 (All DT)	HeartWare HVAD (n = 297)	HMII LVAD (n = 148)	HeartWare 63.9 ± 11.6 HMII 66.2 ± 10.2	HeartWare 227 (76.4) HMII 122 (82.4)	HeartWare 1: 10 (3.4); 2: 86 (29.0); 3: 120 (40.4); 4: 59 (19.9); 5–7: 22 (7.4). HMII 1: 5 (3.4); 2: 46 (31.1); 3: 60 (40.5); 4: 27 (18.2); 5–7: 10 (6.8)	24 months
ENDURANCE DT 2 201857 (22, all ENDURANCE papers)	RCT	465 (All DT)	HeartWare HVAD (n = 308)	HMII LVAD (n = 157)	HeartWare 63.3 ± 11.4 HMII 64.2 ± 11.1	HeartWare 252 (82) HMII 125 (80)	HeartWare 1: 12 (3.9); 2: 101 (32.8); 3: 133 (43.4); 4–7: 62 (20) HMII 1: 4 (2.5); 2: 51 (32.5); 3: 68 (43.3); 4–7: 34 (21.7)	12 months
MOMENTUM 3 201958 (17)	RCT	1028 (624 DT)	HM3 LVAD (n = 317 DT)	HMII LVAD (n = 307 DT)	Reported for all DT patients only: 63 ± 12	Reported for all DT patients only: 513 (82.2)	Reported for all DT patients only: 1: 12 (1.9); 2: 187 (30.0); 3: 325 (52.1); 4: 89 (14.3); 5–7: 7 (1.1); not provided: 4 (0.6)	24 months
ROADMAP 2015 (8)	Multicentre, prospective observational	200 (All DT)	HMII LVAD (n = 97)	MM (n = 103)	HMII median 64 (range: 55–70) MM median 66 (range 54–74)	HMII 75 (77) MM 71 (69)	HMII: 4: 63 (65); 5: 21 (22); 6: 10 (10); 7: 0 (0). MM: 4: 35 (34); 5: 29 (28); 6: 35 (34); 7: 2 (2)	24 months

Network meta-analysis results

Analysis of survival

The results of the NMA are shown in Table 10.

TABLE 10 - Results table from the NMA for the outcome of all-cause mortality at 24 months

The contrast of interest is HM3 vs. MM (highlighted in bold).

Comparator 1	Comparator 2	Evidence type	RR	95% CI
HM3	HeartMate II	Direct58	0.88	0.64 to 1.22
HM3	HeartMate XVE	Indirect	0.48	0.27 to 0.84
HM3	MM	Indirect	0.25	0.13 to 0.47
HeartMate II	HeartMate XVE	Direct56	0.54	0.34 to 0.86
HeartMate II	MM	Indirect	0.28	0.16 to 0.49
HeartMate XVE	MM	Direct54	0.52	0.43 to 0.78

The direct evidence results reported by the three included trials were used to derive the indirect estimates using the methods described in Network meta-analysis. These data were used to produce a RR of death of 0.25 (95% CI 0.13 to 0.47) in the HM3 compared to MM at 24 months. This translates to a 75% reduction in the risk of death in patients with a HM3 compared to those on MM within 24 months. However, the results should be treated with caution due to the wide uncertainty levels placed around the RR and questions relating to the transitivity assumption as described above.

Other outcomes and observational data

Data to carry out the NMA were only available for the survival outcome. There were insufficient data to allow for indirect comparisons of any other outcomes (QoL, major events, complications, etc.).

Concerning other comparative data, this was also considered when carrying out the NMA. The ROADMAP study was considered a non-randomised comparative intervention study. However, there were not sufficient relevant data to use in a NMA. No comparative observational studies compare MM to HVAD.

There were no other easily accessible data for direct comparative estimates from studies, even when considering observational studies.

Other registry data by outcome

This section reports the remaining non-HM3 specific data that are available from registry reports. It includes data from registries such as INTERMACS, IMACS and ITAMACS. The majority of registry data are reported for a group of LVADs (e.g. continuous flow devices) or do not differentiate by device type at all. There were 47 registry report articles; 37 from INTERMACS, 4 from IMACS, 2 from ITAMACS and the remaining reports were from ELEVATE and the older Thoratec DT registry. However, seven of the INTERMACS reports were only included for completeness, but their data were not included in the analysis due to the age of the data and the duplicate reporting of outcomes from similar reports over the same implant periods.

Survival

Survival was reported in 15 INTERMACS reports at various time points, as well as for other registries including IMACS (n = 3) and ITAMACS (n = 2). 31,38,46,59–74 Survival appeared to increase over the calendar time of the data set, as expected with the introduction of newer, more effective devices (Figure 13). This is also in line with data reported in the trials. Recent data from INTERMACS74 suggested survival was 80% in DT patients at 12 months in a cohort of patients implanted between 2014 and 2017, using either the HVAD or HeartMate II device. The longest participant follow-up time reported was at 60 months from the INTERMACS registry in a cohort of patients implanted between 2014 and 2018 (this included 14.8% HM3 patients). Survival here was reported to be 43% in DT patients. 46 Follow-up data beyond 24 months is not reported in any of the trials; therefore registry data and observational studies are currently the only sources for this longer-term data.

FIGURE 13.

Survival data of destination therapy patients from registry reports over time on any device at 12 months.

Quality of life

Quality of life was reported in a small number of registry reports. Kirklin 2012 (INTERMACS) reported QoL with the EQ-5D VAS at various points in a cohort of patients implanted with continuous flow devices between 2006 and 2011. 69 Scores improved from 44 at baseline to 72.2 at 12 months of follow-up. This is similar to scores reported in other trials. A more recent report from INTERMACS continuous flow patients (implanted 2008–13) reported improvements in the KCCQ overall score from 33.6 at baseline to 67.1 at 24 months. 75 No further QoL data have been reported.

Hospitalisations

Hospitalisations were not reported widely across registry reports. An INTERMACS analysis of US patients reported that rehospitalisation’s occurred in 75% of patients in transplant centres and 77% of patients in non-transplant centres at 12 months of follow-up (cohort implanted 2012–4). 62 Differential late re-admissions to hospital between shock (190 per 100 patient-years) and non-shock (181 per 100 patient-years) have been reported from the INERMACS registry in a more recent study. 74

Major events

Nine papers reported major events from registry reports (eight from INTERMACS, one from IMACS). 37,59,60,62,68,69,74,76,77 Acharya reported that strokes occurred in 12% of DT patients (implanted 2012–5) in 35 months of follow-up. 59 In the Brinkley report, 12% of transplant centre patients suffered a stroke (censored for death, transplant or explant) at 12 months, and 8.2% in non-transplant centre patients. 62 In both groups, 19% of patients had RHF at 12 months. In the earlier Kirklin analysis, there were 20.8 events of RHF per 100 patient-years in 24 months of follow-up. 69 Pump exchange was reported in 6% of patients at 12 months in Aleksova. 60

Complications

Seven articles reported complications including bleeding events, infections and device complications. 60,62,69,74,78–80 The Kirklin analysis reported 143 bleeding events per 100 patient-years at 24 months of follow-up. 69 There were also 13.8 device malfunction events and 97.1 infections per 100 patient-years. On the other hand, the more recent Michelis analysis reported 5.6 device malfunction events per 100 patient-years at 12 months as well as 8.7 DI events and 2.2 pump thrombosis events per 100 patient-years. 80 In the Brinkley report, 45% of transplant centre patients had bleeding events at 12 months, and 43% in the non-transplant centres. 62 It also reported device infections at 14% and 10% and device malfunctions at 16% and 14% in transplant and non-transplant centre patients, respectively, at 12 months.

Non-INTERMACS observational studies

Data from 21 observational studies thought to not overlap with patients in the INTERMACS registry are presented in this section.

Interagency registry for mechanically assisted circulatory support observational studies

The final category of studies included were observational studies from centres that also contributed data to INTERMACS or other registries. There were 65 studies. These studies were not included in the full analysis to avoid double counting of patient data where possible Table 30 (see Appendix 6) summarises the characteristics of the studies. Data from this category were only considered if key data were unavailable from the previous study types. However, sufficient data were found in the trials, registries and observational studies not overlapping with registries.

Subgroup analysis

Some subgroups were considered important to explore to determine if any differences arose in outcomes in a particular group of patients, such as certain age groups or INTERMACS classes. However, there were no data specific to HM3 available for any relevant subgroups. Data from the HeartMate II device stratified by INTERMACS class were reported in the ROADMAP study (INTERMACS 4 vs. INTERMACS 5–7) but in no RCTs. Seven reports59,60,69,72,74,76,86 from registries reported some outcomes by age group including survival, major events and QoL and three reported by INTERMACS class. 63,69,74 One observational study reported survival by age group (see Appendix 6). 83

Risk of bias and quality assessment

Summary of findings

There is a large volume of evidence analysing and summarising the use of LVADs for DT in patients with end-stage HF, extending from RCT data to single-centre observational data. However, much of this evidence is now redundant due to devices no longer being used or having been withdrawn from the market, as well as very few comparative studies being carried out.

Findings from the NMA in this review demonstrate that LVADs are effective alternatives to transplant and may offer survival rates of nearly 77% at 24 months with the HM3 device, which is the only currently available device in the UK (based on evidence from MOMENTUM, HeartMate II DT and REMATCH). However, comparative study evidence for the HM3 is limited to one RCT, with no direct comparisons made to MM. Indirect comparisons to MM demonstrated a clear, significant benefit for HM3 when considering risk of mortality, though with a wider ranging CI and some concerns about transitivity. The SweVAD trial results and further device-specific data at longer follow-up times would increase the certainty of the findings. 48 Complications and events such as stroke and pump thrombosis may still remain an issue.

Strengths and weakness of the review

There were several strengths to this review, which include the large body of evidence included, with this review being the most comprehensive of its kind to date. This review has considered effectiveness for important subgroups including INTERMACS level and age. The review also considered the difficulties of overlapping patient populations (e.g. within registries and observational studies) and tried to address this by including data in a hierarchical approach. The trials were assessed for risk of bias and the limitations of registry data were considered in the context of the results. Furthermore, decisions on review methods were discussed extensively with clinical experts, as well as a wider steering group of independent clinicians, specialists and LVAD patients.

While every effort was taken to minimise risk of bias during the review process, some issues remain. Due to the large volume of evidence, a pragmatic approach was required to manage the review. This involved a hierarchical approach to analysis of the evidence, meaning that some studies were only considered in the analysis if they did not overlap with other patient populations. However, this also ensured there was no duplication of patient data. To further manage the large volume of evidence, studies with < 50 DT patients were excluded, which may have resulted in missed data. However, the impact of excluding these studies was assessed and found to be minimal (approximately 4.8% of patients) and should not cause bias. Finally, to limit the search hits to a manageable volume, terms for LVAD indication were included as requirements in the strategy (e.g. DT, BTT). This could potentially have resulted in missed studies, but it is likely that any important articles were found via citation checking of key publications and searches of INTERMACS publications via the Society of Thoracic Surgeons’ INTERMACS website.

Strengths and weaknesses of the evidence

One of the most significant limitations of the included evidence is the lack of direct comparisons between the HM3 and MM. No studies have been completed that compare the HM3 and MM, either randomised trials or observational studies. This means that any comparisons made between these two interventions had to be made indirectly, which requires assumptions (though the SweVAD study will address this in the near future). Much of the evidence identified is no longer relevant as devices have been withdrawn or have been superseded by newer versions.

Furthermore, there are currently no studies in DT patients in the UK due to the current guidelines for LVADs being for BTT only. Therefore, findings from studies carried out elsewhere may not be representative of the UK population.

However, there are also several strengths of the evidence that is included. The evidence is wide ranging in terms of study design, meaning that there is both clinical trial data and real-world data from comprehensive registries. This allows for consideration of the differences in these data. The registry data are extensive and INTERMACS is considered the gold standard for patient data registries in health care. However, reporting of HM3 specific data in DT patients is still limited outside of MOMENTUM.

Evidence in context

This review summarises the existing evidence on LVADs for DT. Several evidence reviews have previously explored the effectiveness of LVADs for DT from various countries. A Belgian health technology assessment (HTA) report reviewed evidence up to 2015 and included the REMATCH and ENDURANCE trials, as well as several INTERMACS reports. 90 They found that survival and QoL improved in LVAD patients, though complications (e.g. bleeding and stroke) were an issue. Another HTA report from Canada drew similar conclusions but reiterated the high costs of the devices and surgery. 91 However, these reports were not as inclusive as this review and were carried out before approval of the HM3 so do not reflect the current device availability.

There was another NMA which previously assessed LVADs for DT. 92 This NMA included HM3, HeartWare HVAD, HeartMate II, HeartMate XVE, and MM. It included four RCTs and four observational studies. The primary analysis included the trials only. There are a number of limitations identified in this NMA. It is unclear where the incidence rate ratios (IRRs) come from, for example for the outcome of death for MOMENTUM the IRR per 100 person-years was 0.61 in the paper, whereas the HR in MOMENTUM for death was 0.87. Furthermore, the NMA appeared to only include one year of follow-up from each study.

Implications for stakeholders/future research

This systematic review has demonstrated that the HM3 LVAD is an effective alternative for patients who are not eligible for HT. The current evidence indicates a clear survival benefit in LVADs compared to MM based on early trials and this is maintained with the HM3 when indirectly compared to MM via NMA. The HM3 also shows reduced stroke and other complications and events when compared to other devices. This information may be important for stakeholders in the UK who may consider recommending LVADs and, more specifically, the HM3 for DT in the future.

However, there are still missing data that may be key in determining the true effectiveness of LVADs when compared to MM. The lack of comparative studies, whether randomised or observational, directly comparing the HM3 with MM reduces the certainty of the effect, but publication of the SweVAD trial results could be significant evidence for future decision-makers. Furthermore, the cost-effectiveness of this device will be an important consideration for stakeholders, and this will be explored in the coming chapters.

Finally, research and development of LVADs continues. The new centrifugal flow EVAHEART 2 device has a pump and impeller design, which allows for gentle blood circulation while retaining pulsatility and is currently being trialled in the USA. The FDA-approved RCT is currently comparing the EVAHEART 2 device with the HM3 and aims to enrol approximately 400 patients. 47 This could lead to the possibility of an alternative to the HM3 device for both BTT and DT patients in the future.

Inclusion and exclusion criteria

Papers were included if they met the following criteria:

• Participants – AHF patients (age > 16 years) who are ineligible for HT and are receiving a LVAD intended as DT.

• Intervention – Any LVAD irrespective of type, mechanism or generation. Studies of participants with biventricular assist devices, or RVADs were not eligible for inclusion.

• Comparators – MM or different generation or type of devices or no comparator.

• Outcomes – QoL, cost or incremental cost-effectiveness ratios (ICERs).

To capture as many studies as possible, no restrictions were placed on study design, year of publication or language. Any formal economic evaluations or studies of effectiveness with an assessment of costs or QoL studies were included. Conference proceedings published in the last 3 years (from the date of the searches) were included. The economic evaluations could take the form of a cost-consequence analysis, cost–benefit analysis, cost-effectiveness analysis (CEA) or cost–utility analysis (CUA). Conference proceedings, editorials, reviews or studies that reported on the use of technology for interventions unrelated to AHF were excluded.

Search strategy

A comprehensive search of six electronic databases was conducted from inception until 20 May 2020 in the first instance. Search updates were carried out on 19 May 2021 and 11 January 2022. The electronic databases included three general databases: Cochrane Library (CENTRAL), MEDLINE and EMBASE via Ovid; and three specialist economics databases: EconLit, CEA registry and the NHS Economic Evaluation Database (NHS EED).

Combinations of keywords with the Boolean logic terms ‘OR’ and ‘AND’ were used. Search terms were refined using the MESH library. Search terms for each search strategy are listed in Appendix 2.

The reference lists of key papers were hand-searched to identify additional papers.

Study screening and selection

Following the database search, removal of duplicates was facilitated with EndNote X9 and study selection was performed independently by two reviewers in two stages with the use of Covidence systematic review software. 42,43 Titles and abstracts were screened, and full texts obtained for studies that potentially met the inclusion criteria. The full texts were then checked against the inclusion criteria to assess their eligibility for inclusion in the review. All discrepancies were resolved by discussion between the two reviewers or by a third reviewer.

Quality assessment

Quality assessment of the economic evaluations was conducted with appropriate tools. For trial-based studies, the Consensus on Health Economic Criteria (CHEC) tool was applied; this tool is widely used for the appraisal of economic evaluations. 97 For model-based studies, the Philips Checklist was used. 98 The Philips Checklist is specifically designed for the assessment of modelling studies and is recommended by both the National Institute for Health and Care Excellence (NICE) and the Cochrane Collaboration. 99

Data extraction

The selected studies were read carefully to identify data important to the systematic review. A data extraction template was developed based on the study objectives and subsequent planned analysis. Data extraction was performed independently by two reviewers (TA and CO) using a standardised form. Information was extracted from each paper on the study background and the condition that was studied.

Given the review’s objectives, information on the types of models used, and especially the range of health states used in the models were extracted. Model inputs such as resource use and QoL measures were also sought.

A narrative synthesis was undertaken as is recommended when the methodologies of the included studies are heterogeneous (Centre for Reviews and Dissemination 2008). 95 The quality appraisal was undertaken to inform the analysis rather than to exclude studies. As part of the analysis, approaches to economic evaluation, model structures, time horizons, cycle lengths and parameter inputs were compared and contrasted. This included an assessment of assumptions and the validity of model inputs and the sources of the costs and utility values.

Selection

There were 19 studies from 20 articles that were relevant and included in the cost-effectiveness review. Three of these studies were identified by hand searching. The identified studies included 5 cost analyses and 14 economic evaluations. The PRISMA diagram is provided in Chapter 1, Figure 4.

Included studies

The main characteristics of the included studies are summarised in Table 12. The majority of the studies were conducted in the USA (n = 9) and some in the UK (n = 4). Most of the studies aimed to compare the health and cost outcomes of LVADs with MM, except one study that used HT as the comparator. 100 The study population was patients who received LVADs intended as DT because of ineligibility for a HT.

Most economic evaluations (n = 12) used a CUA approach and only two conducted a CEA. Markov-based modelling was applied in eight studies. The perspective was stated as the service provider in most studies, just two evaluations reported adopting a societal perspective and in one study the perspective was not explicitly stated. 101–103 The cost data were generally based on studies with small sample sizes. Comparability across studies was very low due to the substantial differences in methodology. The findings and limitations of the studies are discussed below.

Health outcomes

The studies considered outcomes such as survival probability, initial hospitalisation and the probability of hospitalisation per year after the initial discharge. Economic evaluations conducting a CUA (n = 12) estimated incremental quality-adjusted life-years (QALYs) as well as survival and hospitalisations. Table 13 provides the survival and hospitalisation rates and health utilities used in the included studies.

The economic evaluations mostly used clinical data from HeartMate II DT and REMATCH trials or the INTERMACS registry. However, the utilisation of data differed substantially. Only four studies that conducted model-based CUAs explicitly stated how hospitalisations after the LVAD implantation were incorporated.

Generally, very few studies included utility values that were specific to DT patients rather than BTT patients. Only three studies applied utility values that were specific to DT patients. 110,115,116 Schueler et al. converted utility values derived from a US population to UK values, using the Dolan algorithm. 116 One study used the utilities reported by Moskowitz et al., which were estimated amongst 29 bridged patients. 118 Long et al. 111 applied the values for BTT patients by Sharples et al. 102 for the first month. 119 The other utility sources used in the economic evaluations did not distinguish between DT and BTT patients. Patients’ monthly NYHA classifications were used in three studies to estimate health utilities. 106,114

Costs

Most studies reported direct medical costs, which included the device cost, initial hospitalisation, re-admissions and outpatient visits. Seven studies also considered the cost of a LVAD replacement in the case of a device failure. 101,106,110,111,114–116 In addition, the analysis by Neyt et al. 102 considered the travel and ‘social work’ costs, although no definition was provided for ‘social work’. One study reported adopting a societal perspective but only the direct costs were incorporated into the analysis. 101

The cost estimates were usually based on a small number of patients from a single centre, and this caused variability. Most studies used standard discounting rates based on setting. Four studies used cost data for BTT patients since the LVAD was not part of the standard care for DT patients in the Netherlands and the UK. 102,109,111,116,120

There were significant variations in the cost estimates, depending on the setting, device and methodology of the studies (e.g. included costs, time horizon). The largest cost component was the initial cost of the LVAD implantation, and in 2019 prices it varied from £83,567 in Belgium and £91,162 in the UK to £220,176 in the USA (Figure 14). 100,113,116 It is important to note that the device and implantation costs might have included different components, considering the differences in healthcare delivery across countries. The follow-up cost estimates ranged from £33,873 for 2 years to £402,309 over a lifetime. 101,112 The estimated cost of MM varied from £6009 for 1 year to £264,271 over 6 years. 101

FIGURE 14.

Index cost of LVAD implantation in 2019 prices.

The cost-effectiveness of a left ventricular assist device as destination therapy

The review identified 14 studies that estimated the cost-effectiveness of a LVAD as DT. The majority of the economic evaluations concluded that LVADs were not cost-effective compared to MM (n = 8) or HT (n = 1) as DT, while three recent evaluations, including two from the UK, found favourable results. 110,115,116 The overall conclusion was not clear in two evaluations. 103,105

The main outcome was the incremental cost per QALY gained in 14 studies and the incremental cost per life-year gained in two studies. It is important to note that Neyt et al. and Clegg et al. applied different discount rates for costs (4% and 6%, respectively) and outcomes (1.5%), while all others used the same rate for costs and outcomes. 90,108

The cost-effectiveness was estimated over a time horizon of 5 years or longer in nine studies (Figure 15), and the incremental cost per QALY ranged between £46,207 and £238,401 in 2019 prices. 108,116 These studies were comparable because all had a payer’s perspective except one, which included transport costs. 90 Despite the differences in the cost estimates, there was a downward trend over time in the ICER per QALYs reported. Baras Shreibati et al. calculated the incremental cost per QALY to be £152,953 or £124,904, based on two different mortality risk estimates. 101 Only one study considered the impact of severity on the cost-effectiveness, based on the INTERMACS profiles. 110 In a subgroup analysis, it was found that the ICER per QALY was lower for the patients with an INTERMACS profile 1–3 (£45,616) and higher for those with an INTERMACS profile 4–7 (£65,018), compared to the base-case estimate (£47,361).

FIGURE 15.

Incremental cost per QALY estimates in economic evaluations with a time horizon of 5 years or longer.

The incremental cost of a LVAD as DT per QALY was found higher than the standard thresholds in all evaluations except two studies, which estimated an ICER that was just below NICE’s threshold of £50,000 per QALY for end-of-life treatments. Both were recently conducted studies from the UK health service perspective, with one based on the now withdrawn HeartWare HVAD device (£46,207), and the other on the HM3 device (£47,361). 110,116

Economic evaluations with Markov models

Risk of bias

The quality of the economic evaluations was assessed using the CHEC and Philips criteria and overall judged as poor to moderate (see Appendix 7). Some important issues were identified in terms of the quality of the studies. For example, none of them justified the sources of parameters except two, which used systematic reviews to identify the mortality risk and health utilities within the first 2 years. 110,116 Additionally, some studies did not provide all the parameters used in the estimates. For instance, two studies did not provide the estimated mortality risks beyond 2 years despite using extrapolated data. 106,110 Similarly, Silvestry et al. did not provide the number of patients in different stroke groups generated based on the mRS. 115 Furthermore, studies did not fully explain how the complications were incorporated, although many included the hospitalisation costs.

Some studies did not consider all important cost components, for example, some omitted hospitalisation costs or the cost of device replacement in the case of a device failure. Similarly, some relevant variables, such as improvement in QoL, were not subjected to sensitivity analysis in some studies. Only Rogers et al. explored the alternatives to the assumptions about the estimated mortality risks beyond the available data in a sensitivity analysis. 114 The studies used data from small numbers of patients from a single centre to identify the cost inputs. Moreover, some studies used data from BTT patients in the absence of cost estimates specific to DT patients. 111

Summary of findings

This review aimed to identify existing economic evidence concerning the use of LVADs in patients with AHF who are ineligible for a HT. The study identified 19 economic analyses focusing on the cost implications based on the screening criteria. Among these 19 studies, 14 were full economic evaluations, assessing both the health and cost impacts of LVADs. Most economic evaluations (n = 8) concluded that LVADs were not cost-effective compared to MM for patients with AHF who are ineligible for a HT. On the other hand, two UK-based evaluations reported favourable findings. The studies had some limitations, such as limited consideration of time horizon and the omission of some clinically significant adverse events.

Strengths and weakness of the review

The review was conducted based on a preregistered protocol (CRD42020158987) and the CRD and PRISMA guidelines were followed. The review was comprehensive as there was no restriction on the database search concerning dates or languages. Two independent reviewers conducted the study selection and quality assessment.

There are some limitations to be acknowledged. In line with the study objectives, the quality assessment focused on economic evaluations only. Thus, the quality of the cost analysis studies was not evaluated.

Strengths and weaknesses of the evidence

The cost-effectiveness outcomes were consistent in the published studies, given that the majority found that a LVAD was not a cost-effective alternative to medical therapy in patients with AHF who were not eligible for heart transplantation. The only two studies reporting favourable outcomes estimated ICER per QALYs to be just below the £50,000 threshold, defined as the end-of-life criteria in the UK.

However, the data inputs and methods used in the studies varied widely. For example, the index cost of LVAD implantation differed considerably from one study to another. It is difficult to explain this variation because studies from the same countries reported considerably different figures. In contrast, the variation in the ICER per QALY estimates was low within the same country in the studies published over the last 10 years. The studies from the European countries were more likely to report lower ICER per QALY estimates compared to the US-based studies.

The evidence on the cost-effectiveness of LVADs for patients with different INTERMACS profiles was not conclusive since only one study conducted an analysis based on the INTERMACS profiles, and the model inputs used in this study were not provided. 110

The economic evaluations had some significant limitations regarding the data used. For example, very few studies used DT-specific health utility data and the others relied on utility estimates for BTT patients. The impact of LVADs on health utilities might be different in DT patients compared to BTT patients. Similarly, the cost data were usually based on a small number of patients and there was limited consideration of the ongoing costs such as outpatient costs. In most studies, these were not addressed, for example, by conducting a sensitivity analysis. There were also some methodological limitations in the economic evaluations included in the review. For instance, only three studies adopted a lifetime perspective, estimating long-term health and cost impacts. Additionally, there was a lack of data on outpatient costs and palliative care, especially for the MM patients.

Additionally, most economic evaluations used clinical data based on HeartWare HVAD. However, HeartWare HVAD has recently been withdrawn by the producing company due to safety concerns. Thus, the findings of these studies should be interpreted with caution.

Evidence in context

Only two UK-based studies found LVADs a cost-effective treatment compared to MM in patients with AHF who are ineligible for a HT, both reporting ICER per QALY estimates just below the £50,000 threshold. 110,116 However, the evidence was not conclusive since these evaluations had some limitations. Firstly, one of these studies focussed on HeartWare HVAD, which has recently been withdrawn. 116 Secondly, these studies did not consider the impact of adverse events on life expectancy, utilising two health states in the economic models. Additionally, the cost inputs used in these two models varied considerably. For example, the cost of LVAD implantation was £91,162 in the study by Schueler et al. 116 and £108,223 in the study by Lim et al.;110 however, as these are two different devices, this could account for some of the cost differences. Additionally, the latter study estimated the additional cost of a LVAD compared to MM at £113,552. Thus, the key difference between the two treatment options was the device cost, estimating only a small cost for the cost of adverse events (£5329). However, the corresponding figure in the study by Scheuler et al. was £35,070.

Implications for stakeholders/future research

There were some limitations in the economic evaluations regarding their methodology and data inputs. Additionally, discrepancies were found between two recent UK-based models, which reported ICERs just below £50,000. Thus, the existing evidence is not sufficient to make commissioning decisions in the UK.

A new economic evaluation, which considers all the important adverse events with more recent data over a lifetime horizon, is needed. Considering the scarce data available regarding the ongoing costs of both MM and a LVAD, it would be valuable to identify the key parameters that have a significant impact on the ICER per QALY estimates and demonstrate how they influence the cost-effectiveness findings. This would guide future research.

Chapter summary

The evidence identified in this systematic review suggests that the estimates of the cost-effectiveness of a LVAD as DT has improved over time. This may be explained by the increased life expectancy associated with newer generation devices and a reduction in adverse events and device costs. However, the estimated incremental cost per QALY has tended to remain higher than the accepted thresholds in most studies.

The existing evaluations have important limitations, such as not considering a number of important complications that can occur after the LVAD implantation. In addition, none of the existing studies considered the impact of INTERMACS profiles and age at the time of implantation. In terms of setting, there were only two UK-based economic models, but one study evaluated a device that has since been withdrawn due to safety concerns. Hence, the findings of this review suggest that a novel and more comprehensive economic evaluation of using LVADs as DT in patients who are not eligible for a transplant is needed.

Model description

Economic modelling was required to estimate the long-term implications and to be able to incorporate all the key information collected in the different trials into an economic model. A Markov model was designed to evaluate the cost-effectiveness of a LVAD as DT for patients who are ineligible for a HT, compared to MM. A Markov model was appropriate for this analysis due to the chronic nature of the condition under consideration. 123

Figure 16 presents the simplified overall model structure. In the intervention group, HT ineligible patients received a LVAD, while the comparator group received MM (representing usual care for DT patients in the UK).

FIGURE 16.

Simplified model structure. a, Major event: stroke, GIB, DI, LVAD failure as relevant.

The cycle length for the model was 1 month, and at the end of the first cycle, patients could be either alive without any major event, alive with major events, or dead. The health states for patients receiving LVADs and MM are demonstrated in Figures 17 and 18, respectively.

FIGURE 17.

Health states for LVAD recipients.

FIGURE 18.

Health states for medical management patients.

A major event was defined as any health condition that substantially increases long-term mortality risk, and these were modelled as separate health states. The major events identified for patients receiving LVADs were stroke, RHF and AR. For the MM group, the major event included was stroke. In addition, complications with no or limited impact on long-term mortality were incorporated in the model for LVAD patients to estimate their QoL impacts and the costs. These conditions are GIB, DI, PI, PE and arrhythmia.

In the base case, it was assumed that patients would not be eligible for HT at all. This was to reflect the central aim of the analysis, which focused on patients who were ineligible for transplant. However, given that some patients may become eligible for transplant at a later date, a sensitivity analysis was conducted to explore this impact on the results. For the sensitivity analysis, two additional health states were added to the model structure, BTT and HT. These states are shown in Figure 17 for completeness.

Mortality risk

Mortality rates published by the Office for National Statistics were used to obtain the age-standardised mortality risks. 124 The age at the time of the implant was assumed to be 65 years in the base case. The overall mortality risks for DT and MM patients were adjusted for the probability of death due to the major events, applying the below formula where ME stands for major events. 125

As the systematic review of the clinical literature found no trial comparing contemporary LVADs to MM directly, estimating the mortality risks in the LVAD and MM arms required some assumptions. Four different potential methods were identified:

1. Non-comparative, net weight estimates: The mortality risks reported for LVAD recipients in the MOMENTUM trial and MM patients in the REMATCH trial were utilised to obtain monthly probabilities in the model. 54,126 Thus, it was assumed that the profiles of patients in the MOMENTUM trial matched perfectly to the MM patients in the REMATCH trial, and the clinical effectiveness of standard care had not changed over the last 20 years.

2. Non-comparative, weighted estimates for MM: The mortality risks for LVAD recipients in the MOMENTUM trial were used. The mortality risk in the MM arm was obtained based on a weighted average, using data from the REMATCH and MEDAMACS trials and the proportions of INTERMACS 2 and 3 and INTERMACS 4 and 5 patients in the MOMENTUM trial, respectively (Rose, Mehra, Ambardekar). 54,126,127 Thus, it was assumed that the mortality risks reported in the REMATCH trial included MM patients who could be considered to have a similar disease progression to INTERMACS 2 and 3, and that the MEDAMACS data included MM patients who could be considered to have a similar disease progression to INTERMACS 4 and 5 patients.

3. Comparative estimates mapped to MM in REMATCH: The mortality risks for MM patients in the MM arm were used. The mortality risks for the LVAD patients were obtained based on the RR estimated in the NMA reported in Chapter 3 and the mortality risks for MM patients. Thus, it was assumed that the profiles of patients in the MOMENTUM trial matched perfectly to the MM patients in the REMATCH trial, and the other trials that were used in the NMA reported in Chapter 3. It was also assumed that the mortality risk estimates for the MM patients in the REMATCH trial were more reliable than the mortality risk estimates for the LVAD recipients in the MOMENTUM trial.

4. Comparative estimates mapped to the LVAD in the MOMENTUM trial: The mortality risks for LVAD patients in the MOMENTUM trial were utilised. The mortality risks for the MM patients were obtained based on 1/RR estimated in the NMA reported in Chapter 3 and the mortality risks for MM patients. Thus, it was assumed that the profiles of patients in the MOMENTUM trial matched perfectly to the MM patients in the REMATCH trial and the other trials that were used in the NMA. It was also assumed that the mortality risk estimates for the LVAD recipients in the MOMENTUM trial were more reliable than the mortality risk estimates for the MM patients in the REMATCH trial.

It was deemed important to explore the impacts of incorporating the estimates from these four options into the evaluations, since all included different assumptions. The estimated mortality risks based on the four options are provided (see Table 33 and Appendix 8).

In the absence of mortality data beyond 2 years, it was necessary to extrapolate the limited trial data to estimate cost-effectiveness for a longer time horizon. The model assumed that the mortality risk after the 24th month would be the same as the risk reported for the months between 13th and 24th.

Major events

The major events were modelled separately, and every major event was represented in a health state (Figure 19).

FIGURE 19.

Transition between major events.

Complications

The complications incorporated into the model could be experienced by all patients receiving LVADs.

Health utilities

To estimate QALYs, baseline health utilities were first applied to all the patients in the model. The baseline utility value reported in the MOMENTUM trial was used for the MM patients in the model (0.51). Similarly, the utility values reported in the MOMENTUM trial were utilised in the LVAD arm (0.76 and 0.77, Mehra). 126

Utility decrements

Utility decrements were applied to those who experienced a major event or complication. It was assumed that major events would cause a permanent utility loss, while complications would result in a reduction in utility during the cycle in which the complication occurs.

Costs

The costs and resource use associated with the intervention and comparator were estimated using a range of sources. To identify the one-off cost inputs, for example the costs associated with stroke, NHS reference costs were used for the operation costs while the number of hospital days in the intensive care unit (ICU) and in a cardiac ward were estimated through discussions with two practicing heart surgeons, working in the NHS. The ongoing cost inputs, for example outpatient costs for LVAD recipients, were identified from the systematic review (see Chapter 3). All the costs are presented in 2019 prices and the future costs and benefits were discounted at 3.5% as per NICE guidelines.

For LVAD recipients experiencing early RHF, the usual practice in the UK is to make some adjustments to the existing LVAD setting (in addition to the treatment with inotropes) or insertion of a temporary RVAD rather than implanting a RVAD into the right ventricular, and thus the cost of the device is expected to be much lower.

The cost inputs used in the model are provided (see Appendix 8 and Table 35).

Modelling assumptions

A range of assumptions were required for the analysis due to the limitations associated with the data available and for computational practicalities. The assumptions were as follows:

• Any patient could experience only one major event within a given month. Patients experiencing a major event were assumed not to experience any of the complications within the same month.

• Disabling stroke was assumed to comprise 4.6% of all stroke cases (mRS > 4), both in LVAD and MM recipients. 57,136

• Patients who experienced AR or RHF could experience non-disabling stroke unlimited times. This is because those patients would transition back to the AR or RHF states in the following cycle, unless they died.

• The probability of a disabling stroke amongst patients with a prior non-disabling stroke experience was estimated based on the study by Kirklin et al. who reported that 17% of those who experienced non-disabling stroke had a second stroke within 3 years, which was 0.002 monthly. 137 The probability of death amongst patients experiencing a second stroke was assumed to be the same as for the first stroke (0.25). Similarly, it was assumed that of the patients who survived a second stroke, 4.6% would have a mRS > 4, as in the first stroke.

• Apart from the relationship between stroke and disabling stroke, previous experience of major events and complications was assumed not to have an impact on the probability and outcomes of a major event or a complication. This was to reflect the published evidence, for example Truby et al. found that experiencing AR did not have a significant impact on the risk of stroke, bleeding and arrhythmia, conditional on survival to 1 year. 38

• For model simplicity, it was assumed that valve replacement did not have any impact on QoL or life expectancy.

• It was assumed that 57% of arrhythmia cases would be ventricular and a higher QALY loss (0.06) was applied for these patients. 112,138

Sensitivity analysis

A range of deterministic and probabilistic sensitivity analyses were conducted to explore the uncertainties around the model parameters. In the first sensitivity analysis, the impact of incorporating the probability of transitioning to BTT and then HT for LVAD patients on the model outcomes was estimated. Secondly, key parameters were varied and the impacts on the model outcomes were presented on tornado diagrams. Finally, PSAs were conducted to estimate the uncertainties around the model outcomes. The details of these analyses are provided below.

Transition to bridge to transplant and heart transplant

In order to explore the impact of LVAD patients becoming eligible for transplant, a sensitivity analysis was conducted to incorporate a small proportion of LVAD recipients moving to the BTT state (0.06/month), starting 6 months after the implant and until 3 years after the implant. 130 In the next cycle, some BTT patients (0.028/month) would have the probability of receiving a HT until 3 years after the implant. 137 BTT patients could experience a major event, have a HT, or die. BTT patients experiencing a major event became ineligible for a HT and moved back to the DT state or died. The model inputs used for this analysis are provided (see Table 35 and Appendix 8).

One-way sensitivity analysis

There was limited or conflicting evidence on some of the model parameters in the literature, such as the monthly ongoing costs for MM and LVAD patients. Different values were used for these parameters to estimate the impact on the model outcomes and to identify the parameters with the greatest impact. The model inputs used for this analysis are provided in Appendix 8, Table 37.

Probabilistic sensitivity analyses

A PSA involves running thousands of different versions of the model where parameters are varied randomly alongside the prespecified distributions around parameter values. 140 The distributions were defined based on the nature of the parameters, with beta distributions used for binomial data and gamma distributions for costs. 94 Additionally, the difference in sampling method was applied for the parameters, which had different probabilities at different time points. 141 This method takes the relationship between the probabilities at different time points into account, by estimating the mean and variance of the logit distribution for each parameter to obtain the probabilistic values. The independent random sampling method was utilised for the remaining probabilities that did not require such adjustment and for the cost inputs.

Additionally, a cost-effectiveness acceptability curve was produced, to allow estimation of the cost-effectiveness outcomes at different willingness to pay (WTPs) thresholds. 142

Primary analysis

To be comprehensive, two base-case analyses were undertaken for the primary analysis, using two separate estimates of mortality risks and utilities. Based on the systematic review of the clinical literature, four different methods were defined to estimate mortality risks and health utilities. Economic evaluations were conducted for all four of these options, keeping all the remaining parameters constant, to identify the base case. Based on the face validity of the model outcomes and the assumptions associated with each option, it was decided to present two base cases: non-comparative, net weight estimates and comparative estimates mapped to LVAD in MOMENTUM. The justifications for not including the other two options are provided below and the model outcomes for these options are provided (see Appendix 8).

1. Non-comparative, net weight estimates: Taken forward as one of the two base cases and the outcomes are provided below.

2. Non-comparative, weighted estimates for MM: This option required combining data from three different trials and the modelling outcomes were similar to the non-comparative, net weight estimates. Thus, it was decided to present the outcomes as an appendix.

3. Comparative estimates mapped to MM in REMATCH: The model outcomes were not realistic for the LVAD recipients. It suggested that patients receiving a LVAD would have 2.86 LYs and 2.02 QALYs on average. This was deemed too low, underestimating the benefits of LVADs compared to the findings of recent trials. Thus, it was decided to present the outcomes as an appendix.

4. Comparative estimates mapped to LVAD in MOMENTUM: Taken forward as one of the two base cases and the outcomes are provided below.

Deterministic model outcomes

The deterministic lifetime outcomes provided in Table 19 showed that using the non-comparative, net weight estimates and the comparative estimates (mapped to LVAD in MOMENTUM), LVAD would be expected to produce additional QALYs of 2.86 and 2.51 per patient, respectively, compared to MM. The incremental costs were £152,735 and £146,275, respectively. Therefore, although the QALY and cost estimates in the MM arm were higher using the comparative estimates mapped to LVAD in MOMENTUM (compared with the non-comparative net weight estimates), the ICERs produced were similar at £53,496 and £58,244, respectively.

Table 20 provides the model outcomes at the end of 2 years. These estimates showed that by the end of the second year, LVADs would reduce expected deaths by 66% using the non-comparative, net weight estimates and by 44% using the comparative estimates mapped to LVAD in MOMENTUM. The ICERs at 2 years were estimated to be £131,593 and £151,101, respectively.

Table 21 provides the model outcomes at the end of 5 years. These estimates showed that by the end of the first 5 years, LVADs would reduce expected deaths by 46% using the non-comparative, net weight estimates and by 39% using the comparative estimates mapped to LVAD in MOMENTUM. The ICERs at 5 years were estimated to be £67,997 and £77,775, respectively.

Severity weighted incremental cost-effectiveness ratio estimates

The updated NICE guidelines recommend using QALY weights based on the QALY shortfalls a certain population faces as a result of an illness. 122 The absolute QALY shortfalls for the study population were estimated as 10.38 using the non-comparative, net weight estimates and 9.95 using the comparative estimates mapped to LVAD in MOMENTUM, while the proportional QALY shortfalls were 0.96 and 0.92, respectively. The incremental QALY gains estimated by the model were weighted according to the proportional QALY shortfalls, as recommended.

The appropriate weighting was 1.7 in the non-comparative, net weight estimates, and 1.2 in the comparative estimates mapped to LVAD in MOMENTUM. The weighted ICER per QALY estimates were £31,468 and £48,537, respectively (Table 22). Therefore, although the QALY-weighting reduced the ICERs, they were still above the upper bound of NICE’s recommended cost-effectiveness threshold (£30,000 per QALY).

Sensitivity analyses

One-way and probabilistic sensitivity analyses were conducted to address the structural and parametric uncertainties in the model estimates.

Transition to bridge to transplant and heart transplant

The impact of incorporating the probability of transition from DT to BTT states and from BTT to HT states was estimated in a sensitivity analysis. The analysis showed that the ICER was reduced slightly to £52,762 and £57,470, respectively (Table 23).

One-way sensitivity analyses

Figures 20 and 21 show the results of the one-way sensitivity analyses, conducted by varying one parameter at a time to understand their impact on the ICER estimates. The input values were chosen from the studies identified in the systematic review reported in Chapter 3 and from two previous UK studies. 110,116 The analysis showed that the outpatient costs for LVAD recipients had the greatest impact on the ICER estimates. This was followed by the outpatient costs for patients on MM and the mortality risk amongst LVAD patients.

FIGURE 20.

One-way sensitivity analysis for non-comparative, net weight estimates. OMM, optimal medical management.

FIGURE 21.

One-way sensitivity analysis for comparative estimates mapped to LVAD in MOMENTUM.

Since the deterministic sensitivity analysis showed that outpatient costs had a substantial impact on the ICER estimates, further analyses were conducted to explore how the results changed when different values reported in the literature were utilised for these specific parameters. The parameters used in these analyses and the impact on results are provided in Tables 24 and 25, and the details of these values are summarised in Table 37 (see Appendix 8).

The findings showed that the ICER estimates ranged between £36,623 and £94,394, but remained above £50,000 except for four cases where the input values were significantly different from the others reported in the literature.

Additional analyses were also conducted to vary the mortality risk for the LVAD after 24 months to understand the impacts of incorporating the values used in the most recent economic evaluations conducted in the UK (Table 26). The estimated ICERs ranged between £43,380 and £58,844 using the non-comparative, net weight estimates, and between £45,807 and £59,846 in the comparative estimates mapped to LVAD in MOMENTUM.

When the mean of all the values identified for these three parameters (ongoing costs for the LVAD and MM and mortality risk in LVAD recipients after 24 months) were entered into the model simultaneously, the ICER values were £60,272 and £61,675 using the non-comparative, net weight estimates and the comparative estimates (mapped to LVAD in MOMENTUM).

Probabilistic sensitivity analysis

The PSA showed some uncertainty around the model outputs based on 10,000 iterations. The mean estimates and the 95% CIs are provided in Table 27 and the ICER estimates are shown in Figures 22 and 23. The uncertainty around incremental QALY gains was greater using the comparative estimates mapped to LVAD in MOMENTUM compared to the non-comparative, net weight estimates, although the uncertainty around cost differences was similar.

FIGURE 22.

Probabilistic sensitivity analysis based on 10,000 iterations for non-comparative, net weight estimates.

FIGURE 23.

Probabilistic sensitivity analysis based on 10,000 iterations for comparative estimates mapped to LVAD in MOMENTUM.

The analysis also found that the probability of cost-effectiveness at a WTP threshold of £30,000 was 0%.

The probability of cost-effectiveness of a LVAD as DT at different WTP thresholds were explored in a cost-effectiveness acceptability analysis (Figures 24 and 25). This analysis showed that the probability of cost-effectiveness was 21% and 11% at a WTP threshold of £50,000 per QALY using the non-comparative, net weight estimates and the comparative estimates mapped to MM in REMATCH, respectively. The probability of cost-effectiveness reached 100% at a WTP threshold of £75,000 and £91,000 per QALY using the non-comparative, net weight estimates and the comparative estimates mapped to LVAD in MOMENTUM, respectively.

FIGURE 24.

Cost-effectiveness acceptability curve for non-comparative, net weight estimates.

FIGURE 25.

Cost-effectiveness acceptability curve for comparative estimates mapped to LVAD in MOMENTUM.

Summary of findings

This study involved an economic evaluation of LVADs as DT for AHF patients ineligible for a HT compared to MM from the NHS/PSS perspective. A decision-analytic model with a lifetime horizon and one-month cycles was developed based on the systematic reviews in Chapters 3 and 4.

The economic evaluation found that a LVAD would increase life expectancy by 3.73 and 3.06 years, produce an additional 2.86 and 2.56 QALYs per person and the incremental costs to the NHS would be £152,735 and £146,275 per person using the non-comparative, net weight estimates and the comparative estimates mapped to LVAD in MOMENTUM, respectively. Thus, at a WTP threshold of £50,000 per QALY, LVADs were not cost-effective compared to MM for AHF patients ineligible for a HT. The deterministic sensitivity analysis showed that inclusion of the probability of becoming eligible for a HT did not change the findings, while the outpatient costs for LVAD recipients had a significant impact on ICER estimates. The PSA showed some uncertainty around the model outcomes, especially in terms of the incremental QALY gains. The probability of cost-effectiveness at a WTP threshold of £30,000 was 0% based on 10,000 iterations. According to the exploratory subgroup analysis, a LVAD was not cost-effective for any specific INTERMACS group evaluated (£84,800 for INTERMACS 1, £65,458 for INTERMACS 2 & 3 and £56,256 for INTERMACS 4 & 5).

Decision-making based on severity weighted bridge to transplant estimates

The updated NICE guidelines recommend using QALY weights defined based on the shortfall between the expected QALYs in the same age and sex group of general population and the expected QALYs in the study population in the absence of the novel intervention being evaluated. This study estimated the expected QALYs for MM patients as 0.46 using the non-comparative, net weight estimates and 0.85 using the comparative estimates mapped to LVAD in MOMENTUM, and the corresponding proportional QALY shortfalls were 0.96 and 0.92. The recommended QALY weights were 1.2 if the proportional QALY shortfall was between 0.85 and 0.95, and 1.7 if the shortfall was above 0.95. Thus, there was a substantial gap between the weights recommended for the two base-case analyses in this study, although the proportional QALY shortfalls were close.

Utilising different QALY weightings increased the difference between ICER estimates in the two base-case analyses from £4748 (£53,496 and £58,244) to £17,069 (£31,468 and £48,537). Applying the higher weight (1.7) to ICER estimates for the comparative estimates mapped to LVAD in MOMENTUM would reduce the ICER from £58,244 to £34,261, and the weighted ICER was £31,468 for the non-comparative, net weight estimates. On the other hand, applying the lower weight (1.2) to the non-comparative, net weight estimates would reduce the ICER from £53,496 to £44,580, and the weighted ICER was £48,537 for analysis using the comparative estimates mapped to LVAD in MOMENTUM. From a decision-maker’s perspective, the subgroup analysis based on clinical characteristics suggested that severity did not make a difference to the overall results, given that the ICER remained above £30,000 in all estimates.

Strengths and limitations of modelling

The economic model was developed based on the best available evidence identified by the systematic reviews of the clinical and economic evidence. The model development was also informed by discussions with clinicians, commissioners and patient representatives.

In the absence of a direct comparison between the contemporary LVADs and MM, specific attention was paid to the parameters selected for mortality risks, and four different sets of mortality risks were defined. The impacts of these four options on modelling outcomes were explored and discussed with the clinicians in the research project team. Since each of these options were based on different assumptions, it was deemed appropriate to present two base-case estimates, namely using the non-comparative, net weight estimates and the comparative estimates mapped to LVAD in MOMENTUM.

The non-comparative, net weight estimates used life expectancy data on MM dated from 2001, thus the estimates assumed that the clinical effectiveness of standard care had not changed over the last 20 years Therefore, these estimates might be overestimating the benefits of LVADs, if the life expectancy of MM patients has improved over the last 20 years. Additionally, it was assumed that the profiles of patients in the MOMENTUM trial matched perfectly to the profiles of MM patients in the REMATCH trial. This was a reasonable assumption given that the MM patients in the REMATCH trial were on inotropes and the majority of LVAD recipients in the MOMENTUM trial were classified as INTERMACS 2 & 3. However, if this assumption is incorrect and MM patients in the REMATCH trial had worse health statuses compared to the baseline health statuses of the patients in the MOMENTUM trial, the model outcomes might be overestimating the benefits of LVADs.

In the comparative estimates mapped to LVAD in MOMENTUM, considering that trials might have chosen healthier patients to implant LVADs, estimating the mortality risks in the MM arm based on the mortality risks reported in the MOMENTUM trial and the RR estimated in Chapter 3. The clinical effectiveness of LVADs compared to MM as DT in AHF patients was reasonable. However, this assumed that the profiles of patients in the MOMENTUM trial matched perfectly to the MM patients in the REMATCH trial and the other trials used for the statistical analysis reported in Chapter 3. The model outcomes provided a higher life expectancy and QoL for the MM arm and a higher ICER estimate compared to the non-comparative, net weight estimates and the previous UK studies. 110,116 Therefore, the comparative estimates mapped to LVAD in MOMENTUM might be underestimating the benefits of LVADs.

Another consideration was the lack of mortality data beyond 2 years. This required making an assumption to extrapolate the mortality data. The model used the mortality risk reported for the months between 13 and 24 in the trial for the time points beyond 2 years. The sensitivity analyses showed that this was a conservative assumption and had a modest impact on the cost-effectiveness outcome. Therefore, the model might be underestimating the cost-effectiveness of LVADs if the impact on mortality beyond 2 years is greater than the values used in the model. On the other hand, the model might be overestimating the cost-effectiveness of LVADs if the impact on mortality beyond 2 years is lower than the values used.

Additionally, data on the ongoing cost for patients on MM were limited. There was only one UK-based estimate, but the details of this estimate were not clear and the reported value was deemed too high. Thus, estimates from other studies in countries with similar healthcare systems were utilised in the base case. Specific attention was paid to these parameters, and the impact of using different value inputs on the model outcomes was explored in the deterministic sensitivity analyses. When the mean of all the values identified in the literature were utilised, the ICER estimates were similar to the base-case estimates.

Since currently only one type of LVAD is available, the key model parameters were defined from the studies that included only this type of LVAD. However, device-specific data were not available for some major events and complications and data that included previous versions of the LVAD were used for these parameter estimates. This means that the model might be underestimating or overestimating the benefits of the LVAD, depending on the parameters chosen. However, the deterministic sensitivity analysis showed that none of these parameters had a substantial impact on the model results.

Another consideration, there was no trial that compared a LVAD to MM for defined INTERMACS profiles. Hence, the subgroup analyses were based on discrete data from different sources, and for some parameters assumptions were made based on expert views. It is difficult to speculate on the impacts of these assumptions on study findings in the absence of data, and therefore, the subgroup analyses should be considered exploratory.

Findings in context

The systematic review in Chapter 4 identified two recent UK-based economic evaluations. 110,116 One of these studies focused on a type of LVAD that was withdrawn from use in the course of this project. 116 The other study estimated an ICER of £47,361 per QALY gained. 110 The key difference between the two studies related to the mortality risks assumed for LVAD recipients beyond 24 months. Lim et al. extrapolated the values, averaging the changes in the last 6 months while in this study the risk for the 24th month was used for the remaining cycles. 110 When the extrapolated values used in that study were entered into the current model, the ICER per QALY estimates reduced from £54,295 to £49,120, as shown in the deterministic sensitivity analysis. In terms of the incremental costs, for the study by Lim et al. the cost difference between the LVAD and MM arm was only £113,552, which was mainly related to the cost of the device (£109,140). 110 The corresponding figure in the current analysis were £152,735 and £146,275. Thus, the study by Lim et al. assumed very little difference between patients on MM and LVAD in terms of adverse events and complications. 110

The systematic review in Chapter 4 as DT identified only one study that considered the impact of INTERMACS profiles on the cost-effectiveness estimates. 110 That study estimated a lower ICER for patients with INTERMACS 2 & 3 profiles and a higher ICER for the INTERMACS 4–7 group. This contradicts the findings of the current evaluation, which suggests that LVADs are less cost-effective in more severe patients compared to patients who are less unwell. It is not possible to explain the reasons behind this difference since the model inputs for the INTERMACS profiles-based analyses were not provided in the previous study.

Parameters with the greatest impact on incremental cost-effectiveness ratio per quality-adjusted life-year estimates

The deterministic sensitivity analyses indicated that two parameters had substantial impacts on the model results: outpatient costs for LVAD and MM patients. Thus, additional analyses were conducted to estimate the impact of using different value inputs for the parameters with the greatest impact. The value inputs for these analyses were taken from the studies identified in the systematic review reported in Chapter 3 and their applicability to the UK setting was unclear. Thus, these analyses were only exploratory and are not intended to guide decision-making.

The outpatient costs for LVADs had a substantial impact on the ICER estimates due to the wide range of values reported in the literature, ranging from £72 to £2598. The differences between the cost estimates from the published studies can be partly explained by the differences in healthcare provision across countries, given that the highest figure was reported in a study from the USA while the lowest figure was from a UK-based study. 110,115 Another potential reason might be the costs of different procedures delivered during an outpatient appointment. Depending on the values chosen, ICER estimates in the non-comparative, net weight estimates varied between £36,623 and £84,726, while they were between £38,714 and £94,394 in the comparative estimates mapped to LVAD in MOMENTUM. Different values were also reported for the outpatient costs in MM patients, varying between £72 and £2951. The highest cost estimate (£2951/month) resulted with ICER estimates of £44,579 in the non-comparative, net weight estimates and £39,773 using the comparative estimates mapped to LVAD in MOMENTUM.

Implications for stakeholders/future research

The economic evaluations reported in this chapter show that from the NHS/PSS perspective, LVADs as DT for AHF patients are not cost-effective compared to MM at a WTP threshold of £30,000 per QALY. This finding is unaltered by applying severity weightings to the QALY estimates as recommended by the updated NICE guidelines.

The sensitivity analyses indicate that varying the outpatient costs for patients on LVADs and MM had a significant impact on the results. Thus, future research focusing on defining these cost items might be helpful for decision-makers in the UK, although the VoI analysis suggested that the current evidence is sufficient for decision-making. The exploratory subgroup analyses conducted based on the best available evidence and expert views were consistent with the base-case analysis. However, the existing evidence was insufficient to reach a full conclusion on the cost-effectiveness of LVADs for patients with different INTERMACS profiles. Further research is needed to estimate the impacts of INTERMACS profiles on the cost-effectiveness estimates. Thus, these findings should be interpreted with caution.

Another consideration is the impact of applying severity weights as recommended by NICE on the study results. Although the severity weighted estimates did not change the cost-effectiveness findings of this evaluation, the increased differences between the outcomes of the two base-case analyses warrant attention. Health conditions that result in 0.95 QALY shortfalls would require a QALY weighting of 1.2, but the weight would jump to 1.7 for a QALY shortfall of 0.96. NICE defined the weight groups based on how the end-of-life criteria were applied in previous appraisals, and there is no evidence on whether this truly reflects societal preferences. In addition, the QALY shortfall cut-offs are higher than those used in other countries, such as the Netherlands (0.70). 146 Further research is needed to understand the appropriateness and potential implications of the severity-weightings on cost-effectiveness decisions for different health technologies and interventions.

Systematic review of clinical effectiveness

The systematic review summarised a large volume of evidence analysing the use of LVADs for DT in patients with end-stage HF with over 130 studies included. The withdrawal of the HeartWare HVAD during the review process meant that priority of reporting was given to data from HM3, the only available device in the UK. 32 One RCT (MOMENTUM 3) assessed the HM3 in comparison to the HeartMate II and reported survival of 76.7% at 24 months compared to 59% in the HeartMate II, the highest survival noted for a device at this follow-up point. 147 The HM3 device also demonstrated fewer bleeding and stroke events than earlier devices; issues that contributed to the withdrawal of the HeartWare HVAD. Furthermore, pump thrombosis could still be considered an issue in the HM3.

Data from the 5-year extended observational follow-up of MOMENTUM 3 were recently presented at the ESC congress. 148 While this was outside of our latest search update and therefore not included in the analysis, overall survival was reported at 54.8% in DT HM3 patients at 5 years, compared to 39.4% in those with the HeartMate II. This was higher than any survival figures reported in INTERMACS analyses of multiple device types, though these analyses did not include the HM3. 31,46,72 This reiterates the survival benefit of the HM3 over other devices and may be useful in future economic models.

While the HM3 has demonstrated clinical effectiveness, there is no direct evidence comparing the device to the standard treatment of MM. To determine this, a NMA was undertaken to establish an indirect comparison using data from previously conducted RCTs. The analysis demonstrated a significant benefit for HM3 compared to MM for risk of mortality, but there were concerns regarding transitivity, and CIs around this indirect estimate were wide.

Systematic review of cost-effectiveness

There was limited evidence pertaining to the cost-effectiveness of LVADs for DT. Most of the 19 studies included were conducted in the USA and were from the perspective of the service provider. The incremental cost per QALY estimates ranged from £46,207 to £238,401 in 2019 prices. Of the studies, 14 were full economic evaluations, which looked at the health and cost impacts of LVADs with 8 of these studies reporting that LVADs were not cost-effective when compared to MM for DT patients. However, two of the UK-based evaluations did report positive findings with the ICER per QALYs reported to be just below the £50,000 threshold, defined as the end-of-life criteria in the UK. Some studies did not include adverse events and there was often limited consideration of time horizon.

Economic evaluation

Based on the two base cases employed in the model, neither found LVADs to be cost-effective at a WTP threshold of £50,000 per QALY when compared to MM in DT patients. The non-comparative, net-weight estimates approach yielded an additional 2.86 QALYs per person with an ICER of £53,496. On the other hand, the comparative estimates mapped to LVAD in MOMENTUM approach produced an additional 2.51 QALYs per person with an ICER of £58,244.

Furthermore, subgroup analysis by INTERMACS profile did not demonstrate cost-effectiveness of LVADs for DT in any specific INTERMACS groupings. LVADs were closest to cost-effectiveness at the £50,000 threshold in patients with INTERMACS profiles 4 & 5 (ICER of £58,815), using the non-comparative, net-weight estimates approach.

Systematic review of clinical effectiveness

The systematic review included a large volume of evidence sourced using a comprehensive search strategy in key databases. Articles were also sought via citation checking and targeted searching of mechanical circulatory support registry reports (e.g. INTERMACS, IMACS). Input was sought from clinical experts as well as patients when considering aspects of the review such as the search strategy, outcome definitions and when planning the synthesis. Robust methods were used throughout to ensure the potential for bias was limited.

However, some limitations were evident. Due to the high volume of evidence found, studies with a sample size of < 50 DT patients were excluded in the clinical effectiveness review. While this may have resulted in the loss of some evidence, calculations (previously described) determined that this would result in < 5% loss of patients across the evidence base.

A pragmatic approach to searching was also applied to manage the huge database of hits that were produced with a less specific search strategy. Various terms for ‘DT’ as well as ‘bridge to transplant’ were used to reduce hits from over 20,000 to around 12,000 across all databases. This could potentially have resulted in the loss of some relevant evidence. Finally, approximately four single-centre observational studies could potentially have had comparative data on different devices available that were not reported. However, the authors of these papers were not contacted due to project time constraints and the likelihood of obtaining any valuable evidence.

Weaknesses were also noted within the included evidence. For example, there was very limited device-specific data reported outside of the trials, which made it difficult to make meaningful comparisons and analyses. Furthermore, subgroup data (by INTERMACS or age) were also sparse, which meant it was difficult to determine if devices were more or less effective in patients in different subgroups. No direct comparisons were made in the evidence between the HM3 and MM, which resulted in the need to carry out a NMA. However, there were concerns regarding the transitivity assumption and CIs around this indirect estimate were wide.

Systematic review of cost-effectiveness

Many of the strengths and limitations of the clinical effectiveness review are also applicable to the cost-effectiveness review, due to the use of the same search strategy, duplicate screening, selection and implementation of appropriate and valid risk-of-bias checklists and piloted data extraction. Further studies were found via systematic review citation checking and contact with clinical experts.

All modelling studies were quality assessed in duplicate. Furthermore, limitations with the evidence itself were also present. The overall quality of the included studies was not high and various issues were evident. For example, only two studies justified the sources of parameters, and many studies did not explain how complications were incorporated. Finally, all but two of the economic evaluations included were not UK-based.

Economic evaluation

Strengths of the economic model produced were clear. The model itself was comprehensive and developed based on the best available evidence produced from the systematic reviews. It considered clinical perspectives as well as device and class of INTERMACS. Clear sources for the parameters were presented and these were based on the best available evidence as well as input from clinicians, commissioners and patients. Decisions of which base cases to use were made following in-depth discussions and processes, and these were presented alongside clear justifications together with descriptions of all of the options considered.

Limitations were, however, unavoidable and mostly pertained to the costs of MM. These costs were not clear from the UK perspective and while attempts were made to acquire data, which more accurately reflected the MM costs, these were unsuccessful. Therefore, data from various sources were used, which could now be considered potentially inaccurate. Questions were also raised over other health-related costs that are not related to a LVAD and whether these should be included in the model.

While the base cases were discussed and considered at length, various assumptions had to be made for each. The non-comparative, net weight estimates approach assumed that the INTERMACS profiles across the MOMENTUM trial and the REMATCH trial were the same. This approach also used life expectancy data for MM from 2001, meaning the estimates assume that the clinical effectiveness of standard care has not changed over the last 20 years.

Finally, no direct relative effect measures were available between the HM3 and MM as there are currently no completed trials comparing these interventions. Data from patients with particular INTERMACS profiles were also limited, meaning that subgroup analysis carried uncertainties.

Previous systematic reviews of clinical and cost-effectiveness

Several systematic reviews and HTAs have been carried out which assessed both the clinical and cost-effectiveness of LVADs for DT. Reports from Canada, Belgium and Sweden all found LVADs to be beneficial for survival compared to MM, though there were some concerns over thromboembolic and device complications. 90,91,149 However, these reports were published before any HM3 data were available and therefore are not reflective of the current device market in the UK. The Canadian report also reiterated the high costs of devices and implant surgery, though once again these were not reflective of the current HM3 device.

A previous NMA assessed LVADs for end-stage HF, but this did not focus solely on DT patients but included patients with all LVAD indications. 92 The NMA included the HM3, HVAD, HeartMate II, HeartMate XVE and MM and focused on four RCTs and four observational studies. Results showed a RR for death of 0.62 in the HM3 when compared to HeartMate XVE; however, this only appeared to be for 12 months of follow-up. The paper also demonstrated further limitations as described earlier in Chapter 3.

Comparisons between the HeartMate 3 and medical management

Issues were highlighted in both systematic reviews due to the lack of available direct comparative evidence between the HM3 and MM. This meant using indirect data produced from the NMA (which relied upon various assumptions and resulted in uncertainty) for the economic model. While no results were published at the time of write-up of this report, an ongoing RCT is currently comparing the HM3 with MM in Sweden. The SweVAD trial aims to enrol 80 participants of DT indication who will be randomised to the HM3 or MM across seven Swedish University Hospitals. 48 The study will follow-up patients for a minimum of 2 years and is estimated to be completed in December 2023. Results of this trial will be important for updating any economic models with more robust comparative data between the HM3 device and MM.

Previous economic models

The systematic review of cost-effectiveness identified two previous UK economic models, both of which found LVADs to be cost-effective in comparison to MM in patients with AHF ineligible for transplant, in contrast with both base-case analyses presented here. 110,116 Both models reported ICER per QALY estimates at just below the NICE £50,000 threshold. However, one of these models was based on the now withdrawn HeartWare HVAD and is no longer relevant. Additionally, applying the severity weighting suggested by NICE (2022) did not change the cost-effectiveness outcomes.

The Lim et al. HM3 model found a ICER of £47,361 per QALY gained (compared to £54,748 and £59,840 in each of our base cases). 110 Differences were evident between the model presented here and the Lim model. Lim et al. extrapolated the mortality risk values, averaging the changes in the last 6 months while in this model, the risk for the 24th month was used for the remaining cycles. Differences in assumed rates of adverse events and complications were evident, with Lim et al. assuming very few differences between patients on MM and the HM3 compared to this model. The Lim model also used only two health states and did not consider the impact of adverse events on life expectancy.

Overall, while the model in this report did not find the HM3 to be cost-effective compared to MM when considering the current NICE thresholds, it does remain close to this cut-off and is not drastically different to the previously reported models.

Implications for practice

The systematic review of clinical effectiveness and NMA demonstrate a survival benefit for the HM3 compared to MM in end-stage HF patients ineligible for transplant. However, the economic model did not find the HM3 to be cost-effective when compared to MM at the £50,000 threshold. This would suggest that currently, though the clinical evidence supports the use of LVADs for DT, there may not be enough evidence to support the use in the UK setting in regard to cost-effectiveness. This remained the case when analyses were limited to particular INTERMACS groupings, suggesting that there is not enough evidence to support use in particular groups of patients either. However, data on survival and other outcomes in DT patients by INTERMACS profile were limited and therefore more data are required to produce more robust results. Future research may be key in refining the estimates of cost-effectiveness of the HM3 compared to MM in DT patients, allowing for clear recommendations to be made.

Recommendations for future research

The systematic reviews and economic evaluation have highlighted gaps and issues with the current available evidence. These issues may help to direct key areas for future research.

Patient and public involvement

Patient and public involvement were included throughout the project. The PPI group included both patient representatives as well as family members and carers to ensure a range of perspectives and experiences. The PPI group met several times throughout the study with members of the research team, including both clinicians and methodologists. The first meeting was held at the beginning of the research to convey the aims of the work and to allow PPI members to share their experiences to enable the non-clinical research team members to develop an understanding of living with HF and a LVAD. This meeting also offered the opportunity for PPI to comment on the proposed research and identify outcomes of importance regarding the systematic review of clinical effectiveness and aspects to be included in the economic evaluation. Two members were also invited to attend the wider steering group meetings for the project, to offer PPI insight and perspective.

Results of the research were conveyed to the PPI group during a meeting following submission of the report. Representatives also had the opportunity to comment on plain English summaries.