Is transcatheter aortic valve implantation (TAVI) a cost-effective treatment in patients who are ineligible for surgical aortic valve replacement? A systematic review of economic evaluations

Abstract

Objectives:

Health Technology Assessment (HTA) agencies often undertake a review of economic evaluations of an intervention during an appraisal in order to identify published estimates of cost-effectiveness, to elicit comparisons with the results of their own model, and to support local reimbursement decision-making. The aim of this research is to determine whether Transcatheter Aortic Valve Implantation (TAVI) compared to medical management (MM) is cost-effective in patients ineligible for surgical aortic valve replacement (SAVR), across different jurisdictions and country-specific evaluations.

Methods:

A systematic review of the literature from 2007–2012 was performed in the MEDLINE, MEDLINE in-process, EMBASE, and UK NHS EED databases according to standard methods, supplemented by a search of published HTA models. All identified publications were reviewed independently by two health economists. The British Medical Journal (BMJ) 35-point checklist for economic evaluations was used to assess study reporting. To compare results, incremental cost effectiveness ratios (ICERs) were converted to 2012 dollars using purchasing power parity (PPP) techniques.

Results:

Six studies were identified representing five reimbursement jurisdictions (England/Wales, Scotland, the US, Canada, and Belgium) and different modeling techniques. The identified economic evaluations represent different willingness-to-pay thresholds, discount rates, medical costs, and healthcare systems. In addition, the model structures, time horizons, and cycle lengths varied. When adjusting for differences in currencies, the ICERs ranged from $27K–$65K per QALY gained.

Conclusions:

Despite notable differences in modeling approach, under the thresholds defined by using either the local threshold value or that recommended by the World Health Organization (WHO) threshold value, each study showed that TAVI was likely to be a cost-effective intervention for patients ineligible for SAVR.

Keywords::

• Aortic stenosis
• Aortic valve disease
• Cost benefit analysis
• Percutaneous valve therapy
• Systematic review
• TAVI

Introduction

Aortic stenosis (AS) is a common valvular heart disease affecting ∼26% of individuals aged >65 years, with 2–3% of individuals in this age group experiencing symptomatic AS1,2. If untreated, AS will progress, resulting in functional decline and death3. The current standard of care for symptomatic AS is surgical aortic valve replacement (SAVR), which has been shown to both reduce symptoms and improve survival4. However, comorbidities are common in this population, and not all patients are considered eligible for surgery based upon a prohibitively high estimated risk from conventional surgery. Historical treatment options for patients ineligible for SAVR were largely palliative, and the prognosis poor5,6.

Transcatheter aortic valve implantation (TAVI) is a less invasive treatment alternative for individuals who are ineligible for SAVR4. The randomized Placement of Aortic Transcatheter Valves (PARTNER) trial demonstrated that, in high-risk inoperable patients (Cohort B), TAVI resulted in improved survival compared with medical management (MM) at 1 year (hazard ratio [HR] = 0.55; 95% confidence interval [CI] = 0.40–0.77; p < 0.001)7, 2 years (HR = 0.56; 95% CI = 0.43–0.73; p < 0.001)8, and 3 years (HR = 0.53; 95% CI = 0.41–0.68; p < 0.0001)9 of follow-up. Significant improvements in health-related quality-of-life (HRQoL) compared to MM were also observed at each of these time points7–9. These results led to a class I, level B recommendation for this patient group in the European Society of Cardiology/European Association of Cardio-Thoracic Surgery guidelines10.

TAVI, however, is just one of many interventions competing for healthcare funding in a given country. Medical coverage decisions are increasingly made on the basis of not only clinical but also cost-effectiveness outcomes. A formal assessment of cost-effectiveness (i.e., an economic evaluation) can be defined as ‘the comparative analysis of alternative courses of action in terms of both their costs and their consequences’11. Cost-effectiveness thresholds are then used to represent the maximum amount society is willing to pay for additional improvements in health outcomes. The UK National Institute for Health and Care Excellence uses an explicit threshold of £20,000 per Quality Adjusted Life Year (QALY) gained in all reimbursement decisions, and will consider technologies up to and above £30,000 if certain conditions are met12. In jurisdictions where no publically stated explicit threshold value exists, an alternative approach to evaluating cost-effectiveness is to use the World Health Organization (WHO) stated approach based on the GDP of a particular country (less than 1 × GDP: highly cost-effective, 1–3 × GDP: cost-effective, greater than 3 × GDP: not cost-effective)13.

Differences in clinical practices as well as cost and payment structures across countries influence the outcome of cost-effectiveness analyses and present a challenge in interpreting and comparing the results of economic analyses across countries14,15. These issues led to a debate on the validity of different studies assessing the clinical and cost-effectiveness outcomes of TAVI16–19.

In order to assess the cost-effectiveness from an international, pan-jurisdiction perspective, we undertook a systematic review of all published economic evaluations of TAVI compared with MM. The objective was to assess the variability of results across models and healthcare systems, and, if possible, to make some form of consensus statement based on the results.

Methods

Study identification

In November 2012, we systematically searched MEDLINE and MEDLINE In-Process, EMBASE, and the National Health Service Economic Evaluation Database. The search was supplemented with a review of HTA reports through the Centre for Reviews and Dissemination database. Search strategies combined indexed and free text terms for the disease (‘aortic valve stenosis’) and treatments (‘heart valve prosthesis implantation’, ‘TAVI’, ‘TAVR’, ‘SAVR’). In MEDLINE and EMBASE, cost-effectiveness filters were used. No restrictions were applied to populations of interest or jurisdiction. Publications prior to 2007 were not considered relevant to the research question and excluded. Publications retrieved were reviewed independently by two reviewers (JE and SM).

To be eligible for inclusion, publications had to present economic evaluations of TAVI vs MM. In case of overlapping publications within the HTA literature, we used data from peer-reviewed journal articles and supplemented this with HTA informational reports. To enable international comparisons, data extracted Incremental Cost-Effectiveness Ratios (ICERs) had to be reported as either cost per QALY gained or as cost per life year gained. Data were extracted into pre-defined extraction tables by one reviewer, and validated by a separate reviewer.

Assessment of study reporting

The reporting quality of all identified papers was assessed using an internationally recognized checklist developed by the British Medical Journal (BMJ)20. The checklist comprises 35 questions divided into three sections which assess study design (seven questions), data collection and model parameters (14 questions), and the analysis performed and interpretation of results (14 questions). Where reports were available from more than one source, the assessment of study reporting was based on the publications in peer reviewed journals. Each study was assessed by one reviewer and checked by another. We derived a simple point scoring system to assist in assessing the reporting standards by providing one point for each ‘yes’ in the BMJ checklist and zero for all other possible answers (i.e. ‘no’, ‘unsure’, ‘not clear from text’, ‘not applicable’). A lower score does not explicitly mean that reporting standards are lower as some questions, for example question 19 ‘Details of currency of price adjustments for inflation or currency conversion are given’ may not apply to all studies.

Standardization of all model outputs

We anticipated that the models would reflect different healthcare systems, different base years for the purpose of cost allocation, and/or publication years. To facilitate a fair comparison of model outputs, a purchasing power parity (PPP) based approach was used. PPP estimates how much money would be needed to purchase the same goods or services in different countries and thereby eliminates the differences in currency and price levels between them21. Reported ICERs were first converted to US dollars for the year of cost collection and then, depending on the cost year, inflated to 2012 values using the US Bureau of Labor Statistics inflation calculator (accessed November 2012)22. Although this approach is mathematically the most appropriate it does not account for all the differences in the jurisdictions.

Results

Search results

Our search identified 171 potentially relevant articles, of which five underwent formal data extraction: Watt et al.16, Reynolds et al.23, Gada et al.24, Neyt et al.25, and Sehatzadeh et al.26. Subsequently, the analysis by Sehatzadeh et al. was published by Doble et al.27. The supplementary search of HTA reports identified three cost-effectiveness models of TAVI in Belgium28, Canada26, and Scotland29. The publication by Neyt et al.28 overlapped with the Belgian HTA report and the report by Sehatzadeh et al.26 had been previously identified.The model presented by the Scottish Health Technologies Group (SHTG)29 was the only de novo analysis identified in the search of HTA reports.

The six evaluations included in this manuscript are summarized in Table 1. They were all published between 2010–2012 and represented five distinct healthcare reimbursement systems: Belgium, Canada, England and Wales, Scotland, and the US.

Table 1. Overview of cost-effectiveness model design and cost inputs.

	Watt et al.16	Reynolds et al.23	Gada et al.24	SHTG29	Doble et al.27	Neyt et al.25
Model design
Publication year	2011	2012	2012	2010	2012	2012
Model design	Two interlinked Markov models	Trial based	Markov model	Decision tree and Markov model	Decision tree and Markov model	Markov model
Time horizon	Lifetime	Lifetime	Lifetime	Lifetime	20 years	Lifetime
Cycle length	Short-term: 30 days Long-term: 1 month	N/A	1 year	Short-term: 30 days Long-term: 1 year	Short-term: 30 days Long-term: 1 year	1 month
Country	England & Wales	US	US	Scotland	Canada	Belgium
Perspective	UK National Health service	National health care system	National health care system	Scottish Heath and social work	Third party payer	Healthcare payer
Discount rates	3.5% costs, benefits	0%, 3%, and 5% for costs and benefits	5% for costs and benefits	3.5% for costs and benefits	5% for costs	3% for costs and 1.5% for benefits
Model outputs: ICER	£/LYG. £/QALY gained	$/LYG. $/QALY gained	$/QALY gained	£/QALY gained	$/LYG. $/QALY gained	€/LYG, €/QALY gained
ICU stay	TAVI: ICU 2 days Non-ICU 7 days min hospital stay 5 days	TAVI: ICU 4 days Non-ICU 6.1 days Post-procedure 8.6 days Mean LOS 10.1 days	Not reported	Mean ICU stay: 2 days Mean highly dependent unit stay: 2 days Mean general ward stay: 6 days	Total hospital stay: 11 days	Not reported
Efficacy source	PARTNER trial	PARTNER B	PARTNER trial	Not reported	TAVI Registries for high risk patients	PARTNER trial
Utility inputs
	General population as per UK EQ5D Norms. Declining natural utility with age Baseline severe AS decrement based on NYHA class mix from the PARTNER study Time-dependent increase in HRQoL based on NYHA gain in PARTNER and an assumed 30 day recovery period (linear growth in utility)	Baseline EQ5D TAVI 0.59 MM 0.57 30 days EQ5D TAVI: 0.71 MM 0.64 12 months EQ5D TAVI 0.72 MM 0.62	Utility was age-adjusted, declining by 0.3%/year of age Valued in model: Heart failure: 0.67 Stroke: 0.46 MM: 0.57 Post-TAVI: 0.67 TAVI complication: one quality adjusted week removed from total	Published: Utility by NYHA class31 I: 0.85 II: 0.71 III: 0.57 IV: 0.43 From proportions: Utility of aortic stenosis class I: 1% II: 9%, III: 57% IV: 34% Average utility (MM) = 0.54 Utility of functioning valve replacement class 1: 59% 2: 28% 3: 10% 4: 3% Average utility (TAVI) = 0.77	Baseline health utilities by NYHA functional class estimated by Gohler et al.30 Published: Utility by NYHA class I: 0.86 II: 0.77 III: 0.67 IV: 0.53 Differences in the proportion of patients in each NYHA functional class between the respective comparators were based on data from the two cohorts of the PARTNER trial and carried forward for all four treatments for years 2–20 of the model Values at FU not explicitly stated in publication	Baseline EQ5D TAVI: 0.59 MM: 0.57 30 days EQ5D TAVI: 0.71 MM 0.64 12 months EQ5D TAVI: 0.72 MM: 0.62
Cost inputs
TAVI device cost	£14,800 ($23,727)	$30,000 ($31,696)	Not reported	£12,000 ($19,724)	$37,606 ($31,306)	€18,000 ($21,263)
Additional procedure fees	£950 ($1,523)	$7,406 ($7,825)	$374.35 ($383)	£2,630 ($4,323)	Not reported	€1,500 ($1,772)
TAVI annual follow-up	Not reported	Not reported	$336.41 ($345)	Not reported	$472 ($390) medications only	Not reported
TAVI total initial intervention	£25,167 ($40,347)	$78,542 (incl. physician fees) ($82,981)	$49,106 ($50,293)	Not reported	$45,865 ($37,853)	€40,057 ($47,319)
Stroke event	£4,155 ($6,661)	No cost reported	Acute treatment: $14,155 ($14,497) (Annual FU costs) = $14,561 ($14,913)	Major disabling stroke £11,450 ($18,820) (CVR and TAVI £343.50 ($565))	$7,994 ($6,597)	Minor €4,679 ($5,527) Major €12,493 ($14,758)
TIA event cost	Not reported	Not reported	Not reported	Not reported	Not reported	€3,946 ($4,661)
Repeat hospitalization	£2,176 ($3,489)	Not reported	Not reported	£3,316 ($5,450)	Not reported	€5,983 ($7,068)
MM annual cost	£461 ($739)	$53,621 ($56,652)	$10,832 ($11,094)	Not reported	$642 ($530) annually and $29,600 ($24,429) for first year including tests, BAV	€3,170 ($3,745)

N/A, not applicable; LYG, life year gained; QALY, quality adjusted life year; ICU, intensive care unit; TAVI, transcatheter aortic valve implantation; LOS, length of stay; FU, follow-up; TIA, transient ischemic attack; MM, medical management; EQ5D, Euro-quality of life 5 dimension questionnaire; NYHA, New York heart association.

NB: Standardized 2012 US$ costs presented in parentheses to side of published cost.

The BMJ checklist for each study is presented in Table 2. While no formal thresholds are defined in the scoring system, we assessed that all studies included in the present manuscript achieved a score of 25 or greater, indicating a good standard of reporting.

Table 2. BMJ Checklist Results.

		Watt (2011)	Reynolds (2012)	Gada (2012)	SHTG (2010)	Doble (2012)	Neyt (2012)
1.	The research question is stated.	•	•	•	•	•	•
2.	The economic importance of the research question is stated.	•	•	•	•	•	•
3.	The viewpoint(s) of the analysis are clearly stated and justified	•	•	•	•	•	•
4.	The rationale for choosing alternative programmes or interventions compared is stated.	•	•	•	•	•	•
5.	The alternatives being compared are clearly described	•	•	•	•	•	•
6.	The form of economic evaluation used is stated	•	•	•	•	•	•
7.	The choice of form of economic evaluation is justified in relation to the questions addressed.	•	•	○	•	•	•
8.	The source(s) of effectiveness estimates used are stated	•	•	•	•	•	•
9.	Details of the design and results of effectiveness study are given (if based on a single study)	•	•	•	•	•	•
10.	Details of the methods of synthesis or meta-analysis of estimates are given (if based on a synthesis of a number of effectiveness studies).	•	•	•	•	•	•
11.	The primary outcome measure(s) for the economic evaluation are clearly stated	•	•	•	•	•	•
12.	Methods to value benefits are stated	•	•	•	•	•	•
13.	Details of the subjects from whom valuations were obtained were given	•	•	•	○	○	•
14.	Productivity changes (if included) are reported separately	•	•	•	•	•	•
15.	The relevance of productivity changes to the study question is discussed	•	•	•	•	•	•
16.	Quantities of resource use are reported separately from their unit costs	•	•	•	•	•	•
17.	Methods for the estimation of quantities and unit costs are described.	•	•	•	•	•	•
18.	Currency and price data are recorded	•	•	•	•	•	•
19.	Details of currency of price adjustments for inflation or currency conversion are given	•	•	•	•	•	•
20.	Details of any model used are given.	•	○	•	•	•	•
21.	The choice of model used and the key parameters on which it is based are justified	•	○	•	•	•	○
22.	Time horizon of costs and benefits is stated	•	•	•	•	•	•
23.	The discount rate(s) is stated	•	•	•	•	•	•
24.	The choice of discount rate(s) is justified	○	•	○	•	○	•
25.	An explanation is given if costs and benefits are not discounted	•	•	•	•	•	•
26.	Details of statistical tests and confidence intervals are given for stochastic data	•	•	○	•	○	•
27.	The approach to sensitivity analysis is given	•	•	•	•	•	•
28.	The choice of variables for sensitivity analysis is justified	•	•	•	•	•	•
29.	The ranges over which the variables are varied are justified	•	○	•	•	○	•
30.	Relevant alternatives are compared	•	•	•	•	•	•
31.	Incremental analysis is reported	•	•	•	•	•	•
32.	Major outcomes are presented in a disaggregated as well as aggregated form	•	•	•	•	○	•
33.	The answer to the study question is given	•	•	•	•	•	•
34.	Conclusions follow from the data reported	•	•	•	•	•	•
35.	Conclusions are accompanied by the appropriate caveats	•	•	•	•	•	•
Total Score		30	27	28	28	25	28

• YES (+1) ○ NO (0) • UNSURE (0)

Overview of model structures

Five of the six studies described the underlying condition of the patient as a series of health states, with transitions between states based on a series of probabilities (i.e., they are Markov models)16,24–27,29. The remaining study was the pre-specified trial-based economic evaluation from the PARTNER B trial23.

All studies generated lifetime cost and benefit estimates. Three research groups26,27,29 also produced short-term models, with one using a separate short-term Markov model (time horizon 30 days, cycle length 1 day)16, and the other two a decision tree (time unit of 1 year)27,29.

The Markov models developed by Gada et al.24 and Doble et al.27 incorporated explicit health states for disease-specific and treatment-specific adverse events such as heart failure or myocardial infarction and stroke. SHTG produced a relatively simple structure using four health states: Valve-related event, functioning valve replacement (VR), AS/Failed VR or death29. Watt et al.16 took a different approach and represented the patient pathway using location of care as opposed to disease-related adverse events. Their long-term Markov model had only three possible health states: Home, Re-operation, or Dead, (‘re-operation’ refers to patients who required an additional TAVI-related intervention due to device failure. It was assumed in the base case analysis that the TAVI valves experienced a failure rate 50% greater than a surgical biological aortic valve). Neyt et al.25 combined elements of both approaches with health states corresponding to mortality, hospitalization, other events, and no event.

The models used a wide range of health state definitions. As a result, no consensus could be identified on how best to quantify the underlying disease pathway from the perspective of an economic model.

Sources of efficacy data

Two models (SHTG29 and Gada et al.24) used pre-PARTNER data to inform the clinical efficacy parameters. These studies entered the public domain after the publication of the PARTNER B 1 year data, but were developed beforehand. It was not clear from the publicly available information where the clinical efficacy data came from to inform the SHTG model29. Gada et al.24, on the other hand, used data from several national registries to inform their efficacy parameters. Of the other models, all used efficacy and adverse event data from the PARTNER B trial to inform the clinical parameters of the models7,16,23,25,27.

Extrapolation of survival data

Extrapolation of survival beyond the trial period was undertaken by all research groups, although there were major differences in the reported life years and QALYs gained, ranging from 0.6 QALYs to 1.6 QALYs16,23,25,27,29. For the models which used data from the PARTNER B trial, the differences in the estimated QALYs highlights the different approaches to extrapolation and consideration of the available data.

Of the four analyses which use data from the PARTNER B cohort, Neyt et al.25 combine unpublished data from the PARTNER B continued access trial with the main trial data from the PARTNER B publication, providing a weighted survival estimate at 12 months. Beyond 12 months, overall survival is then assumed to be constant and a mortality rate is calculated based on the monthly rate from months 2–12 in the first years survival. Doble et al.27 extrapolate beyond the 12-month trial data by applying standardized age-related mortality rates for the Canadian population to 1 year survival estimates from the PARTNER B trial. Both Watt et al.16 and Reynolds et al.23 use the PARTNER B data to inform a parametric survival analysis, the former applying a Weibull estimation and the latter an exponential distribution.

Methods for estimation of treatment-related adverse events

An increased risk of stroke has been observed in patients treated with TAVI7–9. A significant difference in all events in the PARTNER B cohort was observed at 30 days and 12 months (p = 0.03 and p = 0.04, respectively)7. However, differences in major and minor events were not significantly different at either 30 days (major event p = 0.06, minor event p = 0.62) or 1 year (major event p = 0.18, minor event p = 0.37)7. Despite the significance of the end-point, not all models specifically incorporated stroke. Three of the models24,27,29 provide a specific health state for stroke, and therefore assess the risk of stroke directly. Watt et al.16 and Neyt et al.25 incorporated strokes implicitly via alterations to costs and/or utilities. The trial-based economic evaluation does not make it clear in the text how stroke is incorporated23. Vascular complications were taken into account in four models16,24,27,29, and permanent pacemaker implantation and paravalvular leaks in three studies16,27,29. Interestingly, Gada et al.24 stated that

other less morbid conditions such as subsequent atrial fibrillation, perivalvular regurgitation, or the need for pacemaker placement […] considered within the costs and utilities of this state and [were] not included separately in our model.

Likewise, in the publication by Neyt et al.25, vascular complications and bleeding are not included separately to avoid double counting, such events were perceived to occur during either initial or repeat hospitalizations. Doble et al.27 accounted for paravalvular leaks, permanent pacemaker implantation, and vascular complications, respectively, via the inclusion of a cost to cover the expected number of lifetime events26.

Incorporation of resource use and costs

As might be expected, no standardized approach was identified to reporting cost components, with authors reporting overall procedure cost, overall initial hospitalization costs, and annual follow-up costs. The length of stay in intensive care units after TAVI ranged from 2–4 days, with overall hospital stay lasting ∼10 days (Table 1). Three of the research groups estimated a total cost for the TAVI intervention of between $37,000–$47,000, despite there being substantive differences of ∼$10,000 in device acquisition cost in Europe and North America. Taking into account differences in discounting and currencies, there was a wide range in predicted lifetime costs for patients treated with MM, ranging from $7900 (Watt et al.16) to $73,854 (Reynolds et al.23). Similarly, the cost inputs associated with TAVI ranged from $47,097–$158,203 (Table 1)23,29.

Modeling of HRQoL

Two evaluations, Neyt et al.25 and Reynolds et al.23, used the EQ-5D values observed in the PARTNER B trial and reported QALY gains of 0.74 and 1.29, respectively, the difference being attributed more to the different approaches in the estimation of lifetime survival rather than the modeling of HRQoL. The remaining research groups derived benefit measures using published New York Heart Association (NYHA) class specific utilities (Table 1) and published preference rates as a surrogate to the EQ-5D16,24,30,31. In a response to a critique of the Watt et al.16 publication by Neyt et al.25, a further analysis of the Watt model using the EQ5D estimates reported in the KCE resulted in a decrease to the QALY gain and increase in the reported ICER of ∼£3000, resulting in an ICER of £20,100 per QALY17.

Cost-effectiveness results (published)

The reported cost-effectiveness results are presented in Table 3. All included studies reported outputs in terms of ICERs per QALYs gained. Reported ICERs ranged from £16,700/QALY gained (Watt et al.16) to US$61,889/QALY gained (Reynolds et al.23). The willingness-to-pay thresholds used by all authors in their analyses were explicitly stated in all publications. Both Watt et al.16 and SHTG29 used a threshold of £30,000/QALY gained, and reported the probability of TAVI being cost-effective as 100% and 82%, respectively. At a lower threshold of £20,000/QALY gained, the probability of TAVI being cost-effective remained at 100% in Watt et al.16, but was not reported by SHTG29.

Table 3. Economic outputs across CE models.

	Watt et al.16	Reynolds et al.23	Gada et al.24	SHTG29	Doble et al.27	Neyt et al.25
Costs (MM)	£5,000 ($8,016)*	$69,903 ($73,854)*	Not reported	£13,971 ($22,963)*	$57,963 ($47,837)*	Not reported
Costs (TAVI)	£30,200 ($48,416)*	$149,740 ($158,203)*	$59,503 ($60,941)*	£28,654 ($47,097)*	$88,991 ($73,445)*	Not reported
Incremental costs	£25,200 ($40,400)*	$79,837 ($84,349)*	Not reported	£14,683 ($24,133)*	$31,028 ($25,608)*	€33,200 ($39,219)*
QALYs (MM)	0.8	0.73	Not reported	1.53	Not reported	Not reported
QALYs (TAVI)	2.36	2.03	1.78	2.18	Not reported	Not reported
Incremental QALYs	1.56	1.29	Not reported	0.65	0.6	0.74
ICER (COST/QALY)	£16,700	$61,889	$39,964	£22,603	CAN$51,324	€44,900
ICER (2012 dollar equivalent)	$26,773	$65,387	$40,930	$37,151	$42,358	$53,040

* ($) 2012 dollar equivalent.

Neyt et al.25 used Euro equivalents of the UK thresholds (£1 = €1.14 [25 August 2011]). The probability that TAVI was more cost-effective than MM at £20,000 and £30,000 per QALY gained was reported as 9.2% and 36.7%, respectively. Both Gada et al.24 and Reynolds et al.23 used a threshold value of US$100,000/QALY gained. Gada et al. did not report the probability of TAVI being cost-effective at this value, but Reynolds et al. concluded that TAVI was cost-effective in 100% of simulations. Reynolds et al. also assessed TAVI at a lower threshold value ($50,000/QALY gained) and noted that it was almost never cost-effective (3% of all simulations). The threshold value used by Doble et al.27 was CAD$49,000/QALY gained, with TAVI found to be cost-effective in 44.1% of all probabilistic simulations26.

Cost-effectiveness results (PPP adjusted)

The published cost-effectiveness results, when standardized to 2012 US dollar equivalents, are presented in Table 3. The values ranged from $26,773 (Watt et al.16) to $65,387/QALY gained (Reynolds et al.23). The only study generating an ICER above $60,000/QALY gained was the trial-based economic evaluation undertaken in the US by Reynolds et al.23.

Additional sensitivity analyses

Additional sensitivity analyses reported are outlined in Table 4. Gada et al.24 and SHTG29 did not undertake any deterministic sensitivity analyses comparing TAVI and MM. The other evaluations were most sensitive to changes in procedure cost, mortality, and short-term utility values. A particular analysis of interest was the sub-group analysis of medically inoperable and anatomically inoperable patients conducted by Neyt et al.25. This analysis reports an improvement in overall survival of 27.9% for anatomically inoperable patients, and 17% for medically inoperable patients.

Table 4. Sensitivity analyses.

	Sensitivity analyses	Impact on results
Watt et al.16	PSA
	Use of pooled estimates for key model parameters	Insensitive to changes
	Truncation of time horizon	Only 2 year extrapolation of PARTNER required before TAVI becomes cost-effective
	Choice of baseline survival curve	Insensitive to choice of patient group
	Alternative modeling utility approaches	HRQoL improvement after TAVI can be immediate or time-dependent; Modest impact on ICER depending on utility values used
	Extensive deterministic one-way SAs altering parameters by ±10%	• Very sensitive to changes in short-term treatment effect and initial cost of TAVI • Robust to hospitalization costs and AE rates
Reynolds et al.23	PSA
	Altering discount rate (0% and 5%) Excluding non-cardiovascular or BAV costs	Insensitive to changes
	Using alternative time horizons, hazard functions for TAVI, group life expectancy projections, or variations in the annual cost of follow-up	Moderate sensitivity
Gada et al.24	PSA
	SAs largely performed between TAVI and SAVR. 2-way SA for the net health benefit of utility following TAVI vs MM	Insensitive to changes
SHTG29	PSA
	Other SAs performed between TAVI and SAVR
Doble et al.27	A number of 1-way SAs performed	• Model was most sensitive to changes in procedural costs and 1 year mortality rates. • The rates of paravalvular leaks and 30 day mortality were also sensitive to change.
	Scenario analysis of key modelling assumptions	• Altering long-term mortality rates had a marginal effect • Removing long-term complications favored TAVI
Neyt et al.25	PSA	PSA indicated the most important variables were mortality and utility at 12 months
	1-way SA	Indicated the importance of the extrapolation assumption.
	Sub-group analysis of anatomically inoperable patients and medically inoperable patients, for the results of the pivotal study only	Indicated that there was a difference in survival between the two sub-groups, favouring anatomically inoperable patients, resulting in ICERs for the medically inoperable and anatomically inoperable of €42,300 ($49,900) and €26,500 ($31,300), respectively.

PSA, Probabilistic Sensitivity Analysis; SA, Sensitivity Analysis; HRQoL, Health-Related Quality-of-Life; TAVI, Transcatheter Aortic Valve Implantation; ICER, Incremental Cost Effectiveness Ratio; AE, Adverse Event.

Discussion

Our review identified six economic evaluations spanning five different healthcare jurisdictions: Belgium, Canada, England and Wales, Scotland, and the US. There was a wide variation in the device costs and other costing methods. However, differences between the total TAVI-related HRQoL benefit were less pronounced. In terms of incremental QALYs gained, two broad groupings appear. Incremental QALYs derived from models using only the published PARTNER B mortality data were substantially higher than those derived from other sources. Despite this, excluding those of Reynolds et al.23, the ICERs generated were broadly comparable.

Choice of time horizon, and the different approaches to the calculation and extrapolation of overall survival, influenced the cost-effectiveness of TAVI. When viewing the results from all models ‘in the round’, the findings from our systematic review suggest that, for those models using the PARTNER B data, the drivers of differences in cost-effectiveness lies in the methods used to extrapolate overall survival to the time horizon of the analysis and the procedural costs of TAVI.

Determining whether a new technology is cost-effective is dependent on the willingness-to-pay threshold in individual jurisdictions. The values used in the evaluations identified ranged from £20,000–£30,000/QALY gained (England/Wales, Scotland, Belgium), to $100,000/QALY gained (US). To facilitate this cross-jurisdiction assessment of the cost-effectiveness of TAVI, we compared the results using PPP. When converted to 2012 US dollars the ICERs ranged from $26,773–$65,387/QALY gained.

In the absence of explicit cost-effectiveness thresholds for countries other than the UK, the World Health Organization (WHO) published a guide to cost-effectiveness analyses which included the section Choosing Interventions that are Cost Effective (WHO-CHOICE)13. WHO-CHOICE suggested that treatments in which QALYs were accrued at a rate of less than three times per capita GDP are cost-effective and, when QALYs are accrued at less than per capita GDP, highly cost-effective. In 2012, the International Monetary Fund reported that the GDP per capita (using PPP) in the European Union was US$32,24932. Three times this is US$96,747. If the WHO-CHOICE thresholds are applied, then all the economic evaluations identified show TAVI to be cost-effective, indeed, if we revise the WHO-CHOICE threshold to twice the GDP per capita, TAVI remains cost-effective. Using GDP per capita as the threshold results in a less favorable assessment, with only one study showing TAVI to be cost-effective16. Despite the different modeling approaches used and countries represented, the global picture and final conclusion of five of the six papers is that TAVI is likely to be a cost-effective treatment when compared to patients with AS ineligible for conventional surgery.

This study has a number of strengths and limitations. First, by performing a systematic review we can be confident that all relevant material was retrieved. This is, therefore, a comparison of all economic evaluations available to date of TAVI vs MM published in the era of randomized efficacy evidence. However, we were limited to published economic material and it is possible that key data may not have been included due to lack of access. We also did not have access to individual evaluations, so we could not make statements about the internal validity of the models.

It should also be noted that, in the most part, the efficacy parameters of these models have been informed by the PARTNER study of the Edwards SAPIEN device. At the time of preparation of this manuscript, the Medtronic CoreValve device was also available. Studies have demonstrated that these devices are associated with similar outcomes in this patient group, yet are known to differ with respect to pacemaker implantation rates33,34.

Our comparison included information from five different jurisdictions, reflecting very different healthcare systems. This returns us to the original reason for carrying out this comparison: to compare the results of non-transferable economic evaluations developed from different perspectives in order to generate a ‘global’ cross-jurisdiction statement of cost-effectiveness. To enable such comparisons we used the PPP and applied inflation correction to present the results in 2012 US dollars. This type of conversion is useful for analytical purposes, but is subject to substantial limitations given the broad variations in healthcare provisions across jurisdictions such as different length-of-stays and levels of care, different insurance systems, physician’s incentives and price variations35. Such variations make international comparisons difficult, although the PPP and inflation correction allow us to make at least some inferences, despite these limitations12,36.

Furthermore, the scoring system to assess the quality of the cost-effectiveness studies is not validated. Some components such as, ‘17. Are the methods for the estimation of quantities and unit costs described?’ are arguably more important than others (‘2. Is the economic importance of the research question stated?’)20. This is a problem with many scoring systems37, and, therefore, we emphasize that our review is not intended as a critique of the studies included and as such studies should not be assessed on the reporting score.

As previously discussed, the results from the PARTNER B trial formed the basis for this comparison. Whilst elements of the PARTNER B trial have been criticized in the clinical literature18, it is the only published randomized clinical trial of TAVI vs MM. Given the positive clinical efficacy results of the PARTNER B study, it is unlikely that this type of study will be repeated. However, we do recognize the existence of the unpublished Continued Access study of the PARTNER Cohort B study and support the requests calling for the publication of this data, to reduce the uncertainty over the treatment effect of TAVI18. In the meantime, alternative options for modeling the cost effectiveness of TAVI in these patients remain limited by the paucity of good quality long-term survival data for AS patients ineligible for SAVR, as such it is highly likely that data from the MM arm of the PARTNER B trial will continue to be a central feature in cost-effectiveness analyses of TAVI for some time.

Our findings also highlight an important consideration for future health economic evaluations of TAVI and other interventions, namely, the need for a cross-jurisdiction agreement on appropriate methods for extrapolation of overall survival within an economic evaluation. In this review we see three approaches used to extrapolate survival from the PARTNER B trial, yielding very different results. Recent work by a number of researchers discuss methods available for the extrapolation of survival for both individual patient data and summary statistics38--40. In light of this review we would encourage and welcome further discussion or agreement on a systematic approach to survival analysis in cost-effectiveness analysis, to allow greater comparability across jurisdictions.

Whilst not approached in this review, the cost-effectiveness of TAVI vs SAVR in high-risk AS patients eligible for transfemoral implantation has been widely discussed, yet is still subject to significant debate41,42, and further work will be required in this area to reach a cross-jurisdiction statement on this comparison.

Conclusion

In conducting a systematic review of relevant studies, adjusting for PPP and inflation, and accepting the limitations of this exercise, we have demonstrated that it is possible to compare the results of similar evaluations across jurisdictions to support decision-making in health technology assessments.

By comparing the available published economic evaluations of TAVI vs MM we conclude that TAVI is likely to be a cost-effective procedure in AS patients who are ineligible for surgery, regardless of evaluation techniques used and jurisdiction represented. Despite the differences in model structure, diverse inputs and the potential that key unreported data has not been identified, our conclusion would be unlikely to change given the consistency of the results in the published evaluations.

Transparency

Declaration of funding

The study was sponsored by a consultancy agreement between Medtronic International Sàrl Limited and Oxford Outcomes an ICON Plc Company. No restrictions were placed by Medtronic International Sàrl Limited on the design of the study, the choice of included articles or the presentation of results. Publication of the manuscript was not contingent on sponsor approval or censorship of the contents.

Declaration of financial/other relationships

James Eaton, Stuart Mealing, and Juliette Thompson are employees of Oxford Outcomes, an international consultancy firm, and have undertaken similar analyses for a number of healthcare firms. Rachele Busca is an employee of Medtronic International Sàrl Ltd., Neil Moat and Nicolo Piazza have provided consultancy services to Medtronic International Sàrl Ltd. Pieter Kappetein and Ruben Osnabrugge had no competing interest. JME peer reviewers on this manuscript have no relevant financial relationships to disclose.

Author contributions: All authors are justifiably credited with authorship, according to the authorship criteria. In detail: JE, SM, and JT—conception, design, analysis, and interpretation of data, drafting of the manuscript, final approval given; RB—conception, design, final approval given; NM and NP—analysis and interpretation of data, final approval given; APK—critical revision of manuscript, final approval given; RLJO—interpretation of data, drafting of the manuscript, final approval given.