Contrast-enhanced ultrasound and/or colour duplex ultrasound for surveillance after endovascular abdominal aortic aneurysm repair: a systematic review and economic evaluation

Brief statement describing the health problem

Endovascular abdominal aortic aneurysm repair (EVAR) was undertaken for the first time by a Ukrainian surgeon, Nicholas Volodos, in 19871 and introduced into wider clinical practice by Juan Parodi in 1991. 2 Since then, EVAR has become the preferred treatment option for abdominal aortic aneurysm (AAA). 3 A typical EVAR device consists of a stent covered with graft material to prevent the leakage of blood out of the device. 4 Although it is less invasive than open surgery, and has a lower perioperative mortality rate, EVAR is associated with complications in the follow-up period, such as different types of endoleaks, stent–graft migration, distortion or kinking of the stent–graft, structural disintegration of the stent–graft and aneurysm expansion, all of which could potentially lead to failure of treatment in the form of an aneurysm rupture. 5–8 Therefore, all patients receiving EVAR are placed on surveillance with a view to identifying complications in time to allow for remedial secondary interventions.

Data from The EUROpean collaborators on Stent–graft Techniques for abdominal aortic Aneurysm Repair (EUROSTAR) registry of 2846 patients treated with EVAR from December 1999 to December 2004 showed a cumulative incidence of secondary interventions of 6.0%, 8.7%, 12% and 14% at 1, 2, 3 and 4 years, respectively. 9,10 It is therefore necessary that patients receive lifelong surveillance following EVAR. The main purpose of surveillance is to detect clinically significant complications, which are often asymptomatic, and to prevent aneurysm rupture. 11 Endoleaks are the most common complications that occur after EVAR. 12–14

Classification of endoleaks

An endoleak, which can be defined as a persistent blood flow within the aneurysm but outside the stent–graft, is the most frequent complication after EVAR, and is noted in approximately 20% of patients at some point during follow-up. Endoleaks vary in size, direction and the rate of blood flow, and they have variable origins. 15 Five categories of endoleaks have been described in the literature in accordance with the source of blood flow (Table 1).

Treatment and prognosis depends on the type of endoleak. Type I endoleaks, which have been reported to occur in as many as 10% of patients after EVAR,17 have blood flow from the stent–graft attachment site as a result of sealing failure and are associated with increased pressure in the aneurysm sac. Type I endoleaks are usually treated at the time of the index operation and require urgent treatment if they present later. The risk of intraoperative as well as a secondary late type I endoleak is higher in anatomically difficult situation. 17,18 Type II endoleaks, which are characterised by retrograde blood flow into the aneurysm, are the most common type of endoleaks after abdominal EVAR and could be noted in as many as 20–30% of patients at 30 days, 18.9% of patients at 1 year and 10% of patients after 1 year. 17 Most of the type II endoleaks run a benign course and hence are dealt with a ‘wait and see’ follow-up approach. In some patients, surveillance monitoring may be increased. Treatment is required if the aneurysm increases in size; often a > 5 mm increase is deemed to be clinically significant. 15,18 Type III endoleaks result from structural defects arising in the stent–graft or modular disconnection, and always require immediate treatment. Structural failure of the device is more likely to happen over time as arterial pulsations and other factors cause repetitive stress on the device. Tears or holes in the fabric of the graft can be hard to detect, but modular disconnections are usually well seen with computed tomography angiography (CTA) and on plain radiography (stent–grafts have radio-opaque markers to allow for the diagnosis of modular distraction or dissociation on radiological examinations). The incidence of type III endoleaks is usually low (with an estimated incidence of 4% beyond 1 year). 17 Type IV endoleaks occur perioperatively or in the early postoperative phase (defined as being within 30 days) as a result of graft fabric porosity. However, with the advent of low-porosity graft fabrics, this type of endoleak is now observed less frequently. An endoleak detected on follow-up imaging should not be considered a type IV endoleak. Type IV endoleaks usually resolve once the coagulation profile returns to normal after the EVAR procedure. Treatment of type IV endoleaks is not usually required, but care should be taken to exclude other types of endoleaks at the point of diagnosis. 15,18–20 Type V endoleaks are a diagnosis of exclusion when no endoleak is actually demonstrable. This refers to the phenomenon of endotension, defined as the persistent or recurrent pressurisation of an aneurysm, which is identified by the continued expansion of the aneurysm sac. Although the exact cause of endotension is not always elucidated, possible causes include slow blood flow that is not visible on current imaging techniques, ultrafiltration of blood through the stent–graft, seroma, infection and the transmission of pressure through the thrombus in seal zones. Type V endoleaks are managed on an individual basis. 15,18

Epidemiology of abdominal aortic aneurysm

Abdominal aortic aneurysm represents a significant health risk in the older population. Studies conducted in the 1990s in Europe and the USA indicated an overall prevalence of 2–4% for men and 1–2% for women. 10,21,22 A prospective population-based study conducted in Oxfordshire, UK, between 2002 and 2014 showed an annual incidence rate per 100,000 population of 55 in men aged 65–74 years; the incidence increased to 112 in men aged 75–84 years and to 298 in those aged ≥ 85 years. 23 Similarly, a systematic literature review published in 2014, which estimated the global and regional incidence and prevalence of AAA in 21 world regions, reported that in 2010 the age-specific annual incidence rate per 100,000 population ranged from 0.83 [95% confidence interval (CI) 0.61 to 1.11] in the 40–44 years age group to 165 (95% CI 152.20 to 178.78) in the 75–79 years age group. 24

In the USA, even though the total number of AAAs remains stable at 45,000 cases per year, the overall use of EVAR has risen sharply in the past 10 years (from 5.2% to 74% of the total number of AAA repairs). 25 In the UK, the 2016 report of the National Vascular Registry (NVR), which was based on information on AAA repairs from 98 NHS organisations (82 in England, five in Wales, nine in Scotland and two in Northern Ireland), showed an increasing trend in the proportion of EVAR procedures, growing from 54% in 2009 to 66% in 2013. This trend appears to have stabilised over the last few years, with EVAR procedures accounting for 69% of the elective AAA repairs in 2015. The total number of elective EVAR repairs submitted to the NVR in 2015 was 2882. The majority of the EVAR procedures performed were in men (89%) and in people aged > 65 years (86%). Similarly, the UK 2015–16 record of the Hospital Episode Statistics indicates that there were 2975 hospital admissions for endovascular insertion of a stent–graft for AAA in England. Of these, 2650 were admissions of male patients and 382 were emergency admissions. The mean age of admitted patients was 76 years.

Current post-endovascular abdominal aortic aneurysm repair surveillance: variation in services and uncertainty about best practice

Surveillance following EVAR is now universally accepted and recommended, even though there are currently no standard regimens. 26 Post-EVAR surveillance should include a measurement of the aortic aneurysm, the identification and classification of endoleaks and the detection of stent–graft deformation and thrombus build-up within the graft. 27,28 The ideal frequency of surveillance is not defined and heterogeneous strategies exist between centres. 8,15,29 A web-based survey of UK surveillance practice conducted among the members of the British Society of Interventional Radiologists (BSIR) in 2011 indicated that imaging protocols comprise routine CTA imaging at 1 month, 6 months, 12 months and annually thereafter. 29 CTA is still considered to be the current reference standard for monitoring aneurysm size and migration and for the detection of endoleaks. 26 CTA scanning, however, does not provide information on the direction of blood flow associated with an endoleak and its frequent use has the disadvantage of exposing the patient to cumulative doses of ionising radiation with a potential lifetime cancer risk, as well as exposing the patient to contrast medium-induced nephrotoxicity. 30–32 The risks associated with the repeated use of CTA have led some investigators to consider revising the current surveillance protocols in order to minimise the radiation dose and to eliminate unnecessary CTA examinations. 12,33–35 The results of the 5-year follow-up of the US Zenith (Cook Inc., Bloomington, IN, USA) trial suggest, for example, that, in patients without an early endoleak, the 6-month surveillance can be safely omitted from the surveillance schedule. 36 Moreover, it has been observed that only 1.4–9% of patients require reintervention as a result of surveillance-detected abnormalities, whereas the majority of reinterventions occur in symptomatic patients with previously normal surveillance assessments. 11,26,37–39 Colour duplex ultrasound (CDU) and, more recently, contrast-enhanced ultrasound (CEU) have been proposed as possible safer alternatives to CTA. 40–43 Some investigators have suggested that CDU/CEU might have a role in situations when CTA is equivocal or when endotension is suspected. 44 It has also been suggested that CDU/CEU could replace CTA for annual surveillance for patients who have not experienced endoleaks or an increase in aneurysmal sac size in the first year after EVAR. 19,36,45,46 It is debatable whether or not CDU or CEU can currently replace CTA in the immediate post-EVAR surveillance period, as complications are more likely in the early postoperative period and CTA provides more precise evaluation of aneurysm morphologic changes, sac diameter, graft anchorage and integrity. 18 A significant increase in aneurysm size, the detection of a new endoleak or cases in which CDU is non-diagnostic because of obesity, gas or the lack of a suitable window, may also prompt further imaging with CTA for clarification. 3,12,36

A survey conducted in 2010 among the 41 clinical centres enrolled in the UK EVAR trial 147 showed that 12 out of 41 centres used CTA as the primary surveillance modality, 14 out of 41 centres used CDU as the primary surveillance modality and 15 out of 41 centres used a combination of CTA and CDU. Similarly, the recently published 15-year follow-up of the UK EVAR trial 148 demonstrated a shift in contemporary practice towards CDU.

Although the original EVAR trial 1 protocol was for annual follow-up using CTA, which was used in the early stages of the trial, in the later stages, many EVAR patients were followed up with CDU. 48 The change from CTA to CDU was partly influenced by the growing concern about the risks associated with radiation exposure. 49

Relevant clinical guidelines

Although there is currently no consensus on the best place for CDU/CEU in the care pathway of surveillance after EVAR, some clinical guidelines allude to a possible role of CDU/CEU within the existing imaging care pathway. In the USA, the Society for Vascular Surgery practice guidelines, published in 2009,19 recommend contrast-enhanced computerised tomography (CT) imaging at 1 month and 12 months during the first year after EVAR. If at 1 month the CT imaging identifies an endoleak or other abnormalities of concern, postoperative imaging at 6 months should be considered to further evaluate the proper exclusion of an aneurysm. If neither an endoleak nor an aneurysm enlargement is detected during the first year of surveillance after EVAR, colour duplex ultrasound may be regarded as a reasonable alternative to CT imaging for postoperative surveillance. The presence of a type II endoleak should initially prompt continued CT surveillance to ascertain whether or not the aneurysm is increasing in size. However, if the aneurysm is shrinking in size or is stable, follow-up with CDU may be an option.

Despite some existing algorithms and guidelines,19,29 there is currently no consensus on the optimal surveillance strategy after EVAR. Current surveillance paradigms in the UK are considerably heterogeneous, with each centre performing its own protocol, which varies in both the timing and the modality of imaging.

Summary of interventions

Purpose of this assessment

The purpose of this appraisal is to assess the current evidence for the clinical effectiveness and cost-effectiveness of imaging strategies using either CDU or CEU alone or in conjunction with plain radiography compared with CTA for the surveillance of EVAR.

Methods for assessing the outcomes arising from the use of the intervention

We conducted an objective synthesis of the evidence for the clinical effectiveness of imaging strategies using either CDU or CEU alone or in conjunction with plain film X-ray compared with CTA for the surveillance of EVAR. The evidence synthesis was carried out in accordance with the general principles of the Centre for Reviews and Dissemination (CRD)’s guidance for undertaking reviews in health care,58 the recommendations of the Cochrane Handbook for Systematic Reviews of Interventions Version 5.1.0. 59 and the National Institute for Health and Care Excellence (NICE)’s guidance on the methods of technology appraisal,60 and it is reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA). 61

Identification of studies

Comprehensive electronic searches were conducted to identify reports of published randomised trials and cohort studies. Highly sensitive search strategies were designed, including appropriate subject headings and text-word terms, to combine the search facets for endovascular aneurysm repair, the imaging modalities under consideration and the study design. The searches were initially undertaken on 25 January 2015 and updated on 5 September 2016, and these included studies published from 1996 in order to reflect the introduction of CEU into clinical practice. There were no language restrictions, but non-English-language reports were excluded because the evidence base containing English-language reports was sufficiently large. Full details of the search strategies are reported in Appendix 1. The databases searched were Ovid MEDLINE Epub Ahead of Print, In-Process & Other Non-Indexed Citations, Daily and Ovid MEDLINE (1946 to 5 September 2016), EMBASE (1996 to week 36 2016), Science Citation Index (1997 to 5 September 2016), Scopus’ Articles-in-Press (inception to 5 September 2016), Cochrane Central Register of Controlled Trials [(CENTRAL) issue 3 2016], Cochrane Database of Systematic Reviews [(CDSR) issue 3 2016], Database of Abstracts of Reviews of Effects [(DARE) inception to 25 January 2016] and the Health Technology Assessment (HTA) database (inception to 25 January 2016). The reference lists of all of the included studies were perused for further evidence. Members of the advisory group were contacted for details of additional reports.

Eligibility criteria

Studies fulfilling the following criteria were included in this assessment.

Exclusion criteria

Studies not fulfilling the prespecified criteria and the following types of reports were excluded:

• preclinical and biological studies

• case reports

• reports investigating technical aspects of the imaging modalities used for surveillance after EVAR

• editorials and opinions.

Data extraction and management

Two reviewers (PS and MS or CR and MS) independently screened the titles and abstracts of all citations identified by the search strategies. Full-text copies of all of the potentially relevant studies were retrieved and assessed independently by the two reviewers for eligibility using a screening form developed ad hoc for the purpose of this assessment (see Appendix 2). Any disagreements during study selection were resolved by discussion or in consultation with a third reviewer (MB).

A data extraction form was specifically designed and piloted for the purpose of this assessment (see Appendix 2). Detailed information on study design, characteristics of the participants, settings, characteristics of the interventions and outcome measures was extracted. Data extraction was carried out by three reviewers (PS, CR and MS). One reviewer completed the data extraction and a second reviewer cross-checked the extracted data for errors or inaccuracies. There were no disagreements between reviewers.

Quality assessment strategy

The methodological quality of the included studies was independently assessed by two reviewers (PS, CR or MS). Disagreements were resolved by consensus or arbitration with a third reviewer (MB). Studies were not excluded on the basis of their methodological quality. We assessed the risk of bias of non-randomised studies using a 17-item checklist that we developed for NICE through the Review Body for Interventional Procedures [(ReBIP) see Appendix 3]. The ReBIP checklist was adapted from several sources, including the NHS CRD guidance for conducting or commissioning systematic reviews,58 Verhagen et al. ,63 Downs and Black64 and the Generic Appraisal Tool for Epidemiology (GATE). 65 The four italicised questions of the checklist used to evaluate the risk of bias of comparative studies were disregarded for all but the two included comparative studies. 66,67 Individual items within the checklist were rated as ‘yes’, ‘no’ or ‘unclear’ so that a rating of ‘yes’ denoted the optimal rating for methodological quality. We did not assess the quality of abstracts, as the word limit for abstracts is usually insufficient to make informed judgements about the potential risk of bias of the reported study.

Method of analysis/synthesis

The summary results and baseline characteristics from eligible studies have been described, tabulated and demonstrated by graphs using methods that are appropriate for the types of measurements reported by the included studies. We had planned a formal meta-analysis and metaregression of outcome data from the included studies; however, this was not possible, owing to the lack of comparative studies. The outcome data have been summarised descriptively.

Quantity and source of the evidence

The original primary searches and subsequent updates retrieved a total of 3249 records after deduplication. After reviewing the titles and abstracts, 456 records were subsequently excluded. Full-text copies of 483 potentially relevant reports were obtained and screened for inclusion, of which 27 were deemed to be eligible for inclusion. This comprised 24 full-text papers (two non-randomised comparative studies and 22 cohort studies) and three abstracts (all cohort studies). Figure 1 shows the flow diagram of the study selection process. Appendix 4 lists all of the studies included in this assessment, and Appendix 5 lists the studies excluded after full-text scrutiny together with the reasons for their exclusion. Studies were excluded if they failed to meet one or more of the specified inclusion criteria with regard to study design, participants, intervention or outcomes.

FIGURE 1.

Flow diagram of the study selection process.

Quality assessment of included studies

Non-randomised comparative cohort studies

The results of the methodological quality assessment for the two non-randomised comparative cohort studies66,67 indicated that the Chisci et al. 66 study was of moderate methodological quality, whereas the Nyheim et al. 67 study was of poor methodological quality, mainly because over half of the ReBIP checklist items were rated as having an ‘unclear’ risk of bias. 67 In particular, it was unclear if the patients were taken from a representative sample – at a similar point in their disease progression – or selected consecutively, and if the study groups were comparable. 67 The study groups in the Chisci et al. 66 study were comparable, but we noted that the length of the follow-up period was not similar between the study groups. In both studies, it was unclear whether or not the outcomes were assessed blindly or if the authors had adjusted for confounding factors. Figure 2 summarises the results of the methodological assessment of the two non-randomised comparative studies.

FIGURE 2.

Risk-of-bias assessment of the two non-randomised comparative studies. 66,67

Cohort studies

The 22 cohort studies published in full were of mixed quality (Figure 3). 40,41,68–87 The individual study-level results are detailed in Appendix 6. For the majority of studies, over half or more of the ReBIP criteria were not met, or the information provided in the studies was insufficient to determine if the criteria were met, and were, therefore, judged as being of low or moderate quality. Three cohort studies were deemed to be of higher quality as they met all the ReBIP criteria. 76,84,87 Three of the included cohort studies were published only as abstracts88–90 and therefore were not quality assessed.

FIGURE 3.

Quality assessment for non-comparative cohort studies. N/A, not applicable.

Participants were selected consecutively and were a representative sample in just over one-third of the 22 cohort studies. Similarly, the majority of studies undertook prospective data collection (see Appendix 6), clearly defined the intervention and clinical setting and considered long-term outcomes. The majority of the cohort studies (85%) were not clear in their reporting of participant dropouts and withdrawals and over one-third did not clearly report their inclusion/exclusion criteria.

Study characteristics of all included studies

Details of all of the included studies, including baseline characteristics of participants, description of the adopted surveillance strategy (imaging modality and frequency) and clinical outcomes, are described in the subsequent text and Table 2, and are presented in Appendix 8.

Assessment of outcomes and follow-up

Of the 27 included studies, > 90% reported data on reinterventions and on the incidence and type of clinical complications, 85% reported mortality data and 37% reported changes in aneurysm diameter. In total, 20 studies reported the incidence of type I endoleaks,40,41,66–68,73–77,79,82–85,87–91 18 studies reported the incidence of type II endoleaks,40,41,66–68,73,75–77,79,80,82–85,88–90 10 studies reported the incidence of type III endoleaks,66–68,74–76,79,83,85,86 10 studies reported the incidence of limb occlusion40,41,66,69,72,74,78,82,84,85 and 12 studies reported the rate of aneurysm rupture41,66,74,76,77,80,82,83,85–87,90 (see Appendix 9). Other complications reported in the included studies were thrombosis (seven studies68,71,75,76,79,86,90), infection (seven studies66–68,70,74,79,85), stenosis (five studies41,74–76,79), migration (six studies66,67,72,76,77,79), ischaemia (five studies66,68,72,88,92) and kinking (three studies41,72,78). All of the outcomes were measured at different time points after EVAR and during surveillance using various imaging modalities. When reported, the definitions of complications varied among the included studies.

Results of individual studies

The results of the included studies in terms of type and rate of EVAR-related clinical complications, reintervention rates and types of secondary procedure performed, changes in aneurysm diameter and mortality rates are presented in Appendices 10–13.

Results of non-randomised comparative cohort studies

The two non-randomised comparative studies included a total of 750 participants. Both studies used CTA along with CDU, but the timing of imaging varied between the two studies. The results from two comparative studies are shown in Table 4 and described in the text below.

The study by Chisci et al. 66 compared CTA and CDU surveillance at 1 month after EVAR and every 6 months thereafter (protocol I; 376 participants) with CTA and CDU at 1 month after EVAR and CDU and radiography every 6 months thereafter (protocol II; 341 participants) and reported outcomes on reintervention rates, clinical complications, mortality and aneurysm diameter. The proportion of participants who required reintervention was similar between the two protocols (18.1% vs. 16.4%; p = 0.625). There was no evidence of a difference between the two protocols with regard to early reintervention and late reintervention rates. Similarly, the incidence of type Ia and Ib endoleak, type II endoleak, type III endoleak, graft migration, limb occlusion, limb ischaemia, aneurysm rupture, graft infection and bowel ischaemia was similar between the two protocols (Table 5). A higher proportion of graft kinking was picked up by protocol II compared with protocol I (3.0% vs. 1.3%; p = 0.050). Mortality was similar between the two protocols (2.1% vs. 1.8%; p = 0.932) and there was no evidence of a difference in the proportion of participants with permanent (> 30% over baseline) renal impairment (8.8% vs. 8.5%; p = 0.997).

The study by Nyheim et al. 67 compared a conventional surveillance protocol consisting of CTA, CDU and plain radiography at 1, 6 and 12 months and annually thereafter (participant numbers not reported), with a simplified surveillance protocol of CDU and plain radiography at 6–8 weeks, CTA/CDU/plain radiography at 1 year and CDU and plain radiography annually thereafter (56 participants), but failed to provide suitable comparative data. Data on reintervention rates, mortality rates and aneurysm diameter were available for the simplified protocol only. The number of participants who died (16%) or required reintervention (25%) was fairly high. In general, the rate of complications picked up by a surveillance protocol based on CDU soon after EVAR, CTA/CDU at 1 year and CDU annually thereafter was higher than that in the study by Chisci et al. 66

Results of cohort studies

Non-randomised comparative studies

The two non-randomised comparative studies assessed a total of 750 participants (694 participants in one study and 56 participants in the other), and compared a CTA and CDU surveillance protocol with a simplified protocol based on the use of CDU for long-term surveillance after EVAR. The timing of imaging varied between studies, and one of these did not provide suitable data for statistical comparisons. It is worth noting that the largest comparative study, which assessed a total of 694 participants, reported that there was no evidence of a difference between the two surveillance groups in terms of reintervention rate, clinical incidences, mortality and adverse effects, including renal impairment.

Cohort studies

Twenty-five cohort studies assessed a total of 7196 participants. There was considerable heterogeneity in terms of frequency of imaging modalities, duration of the follow-up period, outcome measures, definition of outcomes used (e.g. the definition of decreased aneurysm size) and the time points at which the outcomes were assessed. Owing to the observed heterogeneity between studies, it was not deemed to be appropriate to provide a statistical summary of the outcomes considered. We decided to group the studies according to their similarity in terms of type and frequency of imaging modalities for surveillance after EVAR and created six different surveillance categories (see Table 5). We tabulated and narratively summarised the results for each of these categories.

All but one of the included studies included CDU as part of their surveillance protocols. The remaining study followed up patients using CEU and/or CTA. In the majority of the studies (n = 10), surveillance after EVAR was based on the use of CTA and CDU throughout the follow-up period. Eight studies used CDU for long-term surveillance after EVAR and CTA and/or CDU for early and mid-term surveillance. Two studies used CTA for the long-term surveillance after EVAR (CTA at discharge, CDU at 6 months and then CTA at 12 months and annually thereafter). Two studies included CEU, together with CTA, as part of their surveillance strategies, and three studies adopted a surveillance protocol based exclusively on the use of CDU.

Overall, in the assessed cohort studies, the proportion of participants requiring reintervention after EVAR ranged from 1.1% during a mean follow-up period of 24 months to 23.8% in a cohort that included high-risk patients with hostile neck anatomy who were followed up for a mean length of 32 months. For the cohort studies that provided information on the type of complications requiring treatment, a reintervention was required mainly for the treatment of limb occlusion (< 1–7.2% of participants), thrombosis/stenosis (< 1–5.6% of participants), type II endoleaks (< 1–3.6% of participants), type I endoleaks (< 1–3.1% of participants) and type III endoleaks (< 1–1.6%). The studies that used a protocol based on assessments with CTA and/or CDU throughout the follow-up period showed the highest proportion of participants (range 2.9–23.8%) who required reintervention for complications after EVAR, including type I endoleaks, type II endoleaks, type III endoleaks, thrombosis, limb occlusion, infection and aneurysm rupture. It is worth noting that the study that reported the highest proportion of participants requiring reinterventions (23.8%) focused on high-risk patients, some of whom presented with features of hostile neck anatomy and had the longest follow-up period (mean length 54.8 months). Only limited data were available from studies using CEU as part of their surveillance protocol or studies based exclusively on CDU.

Across the included studies, all-cause mortality ranged from 2.7% (during a mean follow-up length of 24 months) to 42% in a cohort that included a proportion of high-risk patients with hostile neck anatomy (during a mean follow-up length of 54.8 months). Aneurysm-related deaths occurred in < 1% of the participants in four studies. All-cause mortality was generally higher among surveillance strategies that used CTA for early and long-term surveillance after EVAR. One study based on long-term CDU surveillance (median follow-up length of 68 months; range 1–144 months) reported a higher mortality rate and a higher proportion of participants who required reintervention.

The current evidence from the literature assessing the effect of surveillance after EVAR does not show a consistent paradigm. The type of imaging modalities, frequency of imaging and length of follow-up vary considerably between surveillance protocols. Therefore, no firm conclusions can be drawn with regard to the optimal surveillance strategy after EVAR.

The Endurant Stent Graft Natural Selection Global Postmarket Registry

The Endurant Stent Graft Natural Selection Global Postmarket Registry (ENGAGE) is a prospective, multinational, long-term post-market study of the real-world use of the Endurant stent–graft system (Medtronic, Santa Rosa, CA, USA) for infrarenal AAA repair. The registry, which used only minimal selection criteria to obtain a more realistic representation of the current clinical practice, commenced in March 2009 and ended in January 2017. Patients with unruptured infrarenal AAAs who underwent elective EVAR were recruited from 79 clinical centres in 30 different countries. A minimum of five consecutive patients were enrolled from each centre. An EVAR surveillance protocol was carried out in accordance with the standard practice at each clinical site, with the exception of the requirement for 30-day and 1-year imaging.

Five publications reporting data from the ENGAGE registry were identified in the literature93–97 (Table 10). The study by Tang et al. ,93 which compared the 12 outcomes after repair of AAA with bifurcated versus aorto-uni-iliac configuration of the Endurant stent–graft, used data collected in the ENGAGE registry from March 2009 to August 2010. Among the total of 1172 participants in this study, 1089 (92.9%) received bifurcated device stent–graft repair and 83 (7.1%) were treated with an aorto-uni-iliac femorofemoral bypass. The study by Stokmans et al. 94 reported data from 1266 participants from March 2008 to April 2011. Both of these studies reported similar proportions of participants requiring secondary intervention at 1 month (1.5%94 and 0.9%93) and at 12 months (4.6%94 and 4.9%93). Reinterventions were needed for the repair of type I and type III endoleaks in 1.2% of participants in the Stokmans et al. study94 and in 0.6% of participants at 12 months in the Tang et al. study. 93 In the Stokmans et al. study,94 at 12 months, secondary procedures were performed in 2.0% of participants for occlusion/stenosis/kinking and in 0.6% of participants for persistent type II endoleaks. Overall, at 1 month, the detection rates of type I endoleaks, type II endoleaks, type III endoleaks, type IV endoleaks, graft occlusion, graft kinking and graft stenosis were similar in both of these studies (see Table 10). The occurrence of other complications, including bowel ischaemia, myocardial infarction, renal failure and stroke, was similar in both studies. In both studies, the all-cause mortality rate was 1.3% at 1 month and ≈8.5% at 1 year. In the Tang et al. study,93 the proportion of participants who died from aneurysm-related causes was 1.2% at 1 month and 1.5% at 1 year. 93 In the Stokmans et al. study,94 the 1-year assessment showed an overall survival rate of 91.6% (SD 1.4%) and an aneurysm-related survival rate of 98.8% (SD 0.5%).

The study by Karthikesalingam et al. 95 used data from the ENGAGE registry on reintervention and engraft complications at 3 years to predict whether patients would be at a low risk or a high risk of complications after EVAR based on the international validated St George’s Vascular Institute score. 95 Overall, there were 107.6 type I endoleaks, 209.8 type II endoleaks and 86.3 type III endoleaks per 100 patient-years of follow-up (affecting 4.5%, 19.6% and 0.4% of participants, respectively). Aneurysm expansion that was greater than 5 mm was observed in 90.1 participants per 100 patient-years of follow-up (affecting 14.8% of participants). The study by Faure et al. 96 reported 97.9% freedom from limb occlusion at 2 years.

The EUROpean collaborators on Stent–graft Techniques for abdominal aortic Aneurysm Repair registry

The EUROSTAR registry was established in 1996 to collate and analyse data from patients who underwent endovascular treatment for AAAs. The EUROSTAR standardised case record forms were used to collect data. Information on patients with AAA enlargements but without detectable endoleaks (known as endotension), patients who had an elective treatment for AAAs and patients with suitable vascular anatomy for implantation of a stent–graft were collected from various centres in different European countries. The EUROSTAR registry is no longer active (patient enrolment was closed in November 2006).

Sixteen publications reporting data from the EUROSTAR registry were identified in the literature. 5,9,91,98–110 Post-EVAR protocols varied considerably between centres. In most centres, CTA examinations were used at 1, 6, 12, 18 and 24 months after EVAR and annually thereafter. Other imaging modalities used for EVAR surveillance were CDU, CEU and magnetic resonance imaging. It is worth noting, however, that these publications were based on data collected from 1996 to 2006, when CDU and CEU were not in much use in clinical practice. The results from each EUROSTAR report are outlined in Table 11.

The Kaiser Permanente Endovascular Stent Graft Registry

The Kaiser Permanente Endovascular Stent Graft Registry (KPSGR) is a prospective registry that makes use of electronic medical records to track device utilisation and to appraise short- and long-term EVAR outcomes. The data collection started in 2000 and is ongoing. Patients with endovascular repair of AAAs were identified from a retrospective review of EVARs performed at 17 Kaiser Permanente Northern California medical centres in the USA.

Five publications reported data from the KPSGR. 111–116 No standardised post-EVAR surveillance protocol existed during the data collection period. In general, patients received CTA at the 1-month follow-up and then every 6–12 months depending on the clinical scenario.

In three studies, the proportion of participants requiring reintervention was 10.3%,115 10.8%112 and 15% (median of 32.2 months’ follow-up), respectively. 116 The study by Hye et al. 112 reported an overall reintervention rate of 10.8%. Of the reinterventions, 4.6% were for endoleaks, 1.7% were for stenosis, 1.5% were for thrombosis, 1.3% were for occlusion, 1.6% were for device malfunction, 1.3% were for haematoma/seroma, 0.6% were for pseudoaneurysm, 0.6% were for abdominal compartment syndrome, 0.5% were for infection and 0.4% were for rupture. In the study by Walker et al. ,114 aneurysm rupture occurred in 1.2% of participants during a median follow-up length of 32.2 months (IQR 14.2–52.8 months). Aneurysm-related mortality was 0.6% at 1 month and 0.8% at 1 year, whereas all-cause mortality was 14.3% at 1 year. In the study by Anthony et al. ,115 all-cause mortality was 1.2% at the 1-month follow-up.

Australian Safety and Efficacy Register of New Interventional Procedures – Surgical

The Australian Safety and Efficacy Register of New Interventional Procedures – Surgical (ASERNIP-S) is a national collection of data for the evaluation of EVAR. An audit containing information on patients who had a Zenith graft repair for AAA between November 1999 and May 2001 was managed and published by the ASERNIP-S.

No standardised EVAR surveillance protocols were specified. Postoperative follow-up was carried out at 30 days, 3 months, 6 months, 12 months and annually thereafter.

The 787 Zenith graft patients enrolled in the audit were followed up until 2008. Technical success was 93.5% at 30 days. During the 7-year follow-up period, reinterventions were required in 13.5% of participants. Overall, 4.2% of participants developed type I endoleaks, 14% developed type II endoleaks, < 2% experienced kinking, stenosis, migration or thrombosis and < 1% developed type III endoleaks or infection.

All-cause mortality was 0.5% at 1 month, 32% at 5 years and 44% at 7 years. During the follow-up period (7 years after EVAR) 4.4% of participants (35/787) died from aneurysm-related causes. Ten of these deaths (1.5%) were due to ruptured aneurysms.

A recent publication by Fitridge et al. ,117 which combined data from a total of 1647 patients from two ASERNIP-S audits of EVAR (from 1999 to 2001 and from 2009 to 2013), reported a 1-year survival rate of 93.7% (1544/1647) and a 30-day survival rate of 98.4% (1620/1647).

Vascunet database

The Vascunet registry collected data from national and regional vascular registries in Australia, Denmark, Finland, Hungary, Italy, Norway, Sweden, Switzerland and the UK on primary AAA repairs performed between 2005 and 2009.

A total of 31,427 intact AAA repairs were assessed. The overall perioperative mortality rate (in-hospital or within 30 days) was 2.8% and was stable over time. The perioperative mortality rate varied from 1.6% (95% CI 1.3% to 1.8%) in Italy to 4.1% (95% CI 2.4% to 7.0%) in Finland. A total of 7040 ruptured AAA repairs were identified. The overall perioperative mortality rate was 31.6% (95% CI 30.6% to 32.8%), which decreased over time. 118

Registry of Endovascular Treatment of Abdominal Aortic Aneurysms

The Registry of Endovascular Treatment of Abdominal Aortic Aneurysms was established in January 1996 to collect data from 41 centres that initially undertook EVAR in the UK. 119,120 The data for the first 1000 cases submitted to the registry were published in 2005 by Thomas et al. 119 Overall, the mortality rate was 11% at 1 year and 8% at 5 years. The cumulative risk of rupture was 2% at 5 years. Complications related to the aneurysm or device occurred in 13% of participants at 1 year and in 16% of participants at 5 years. The most common complications were endoleaks or graft migration during a mean follow-up length of 3.1 years (range 30 days to 5 years). The cumulative freedom from endoleak was 88% at 1 year and 68% at 5 years. The cumulative freedom from secondary procedures was 87% at 1 year and 62% at 5 years.

Lifeline Registry of Endovascular Aneurysm Repair

The Lifeline Registry of Endovascular Aneurysm Repair was established in 1998 to evaluate the long-term outcomes of endovascular treatment for patients with AAAs. Four publications have reported outcomes of EVAR based on the data submitted to the Lifeline Registry of Endovascular Aneurysm Repair from 1999 to 2004. 121–124 Outcome data from 2664 endograft patients were published in 2005. 124 The overall survival rate was 74% at 4 years, 66% at 5 years and 52% at 6 years. A survival analysis conducted using 6-year data revealed freedom from aneurysm rupture in 99% of patients who had undergone EVAR, freedom from aneurysm-related death in 98% of patients and freedom from surgical conversion in 95% of patients. Most secondary interventions (85%) were performed < 30 days after EVAR. Freedom from secondary interventions was 84% at 1 year and 78% at 5 years.

Anaconda Registry

The Anaconda Registry was a prospective database of clinical outcomes of 106 consecutive patients who underwent endoluminal repair of AAAs using the Anaconda endograft (Vascunet, Inchinnan, UK) in three hospitals in the west of Scotland between 2005 and 2009. Four publications based on data from the Anaconda Registry were identified in the literature. 76,86,92,125 Three of these publications were included in the review of clinical effectiveness evidence. 76,86,92 During a mean follow-up period of 2 years, 9.4% of participants died from causes other than aneurysm. There were no aneurysm-related deaths. Type II endoleaks were detected through CTA scanning at 1, 6, 12, 24, 36 and 48 months in 8.4%, 4.8%, 7.2%, 7.8%, 11.1% and 0% of patients, respectively. There were no type I, III or IV endoleaks. Five cases of endograft limb thrombosis were observed during follow-up. Four of these cases were treated by femorofemoral crossover grafting without any further complications. Follow-up CTA detected hypogastric artery occlusion in three other patients. All three patients remained asymptomatic with no further intervention required. 125

Identification of studies

The literature searches for the clinical effectiveness review were sufficiently broad that they retrieved nine relevant diagnostic test accuracy reviews. 3,62,127–133 Therefore, specific searching to identify additional reviews was more focused but included appropriate subject headings and text word terms. To combine the search facets for endovascular aneurysm repair, the imaging modalities under consideration and diagnostic reviews, MEDLINE and EMBASE were searched from 1996 until March 2016, whereas the CDSR and DARE were searched on 29 March 2016 without date restrictions. The search strategies are reproduced in Appendix 1.

Data extraction and management

Two reviewers (PS and CR) independently screened the titles and abstracts of all citations identified by the search strategies. Full-text copies of all potentially relevant studies were retrieved and assessed for eligibility independently by the two reviewers. Any disagreements during study selection were resolved by discussion or in consultation with a third reviewer (MB). A data extraction form was specifically designed and piloted for the purpose of this assessment (see Appendix 2). Detailed information on study design, participant characteristics, study settings, characteristics of the index tests and reference standard and estimates of accuracy was extracted. One reviewer completed the data extraction form (CR) and a second reviewer (MS) cross-checked the extracted data for possible errors or inaccuracies. There was no disagreement between the reviewers.

Quality assessment strategy

The risk of bias of included reviews was assessed using both the Assessment of Multiple Systematic Reviews (AMSTAR) tool for the assessment of the methodological quality of systematic reviews134 and the recommendations of the York CRD. 58 The included reviews were independently assessed by two reviewers (PS, CR or MS). Disagreements were resolved by consensus or arbitration with a third reviewer (MB). We did not assess the quality of the abstracts, as the word limit for abstracts is usually insufficient to make informed judgements about the potential risk of bias of reported reviews.

Quantity and quality of the evidence

Quality assessment of the diagnostic test performance reviews

The eight reviews of diagnostic test accuracy published in full were of mixed methodological quality. 3,62,127–130,132,133 The majority of the included reviews were considered to have searched major relevant bibliographic data sources, conducted hand-searching of references and provided example of key text words,3,62,127,128,130,132 specified their inclusion/exclusion criteria62,127,128,130,132,133 and provided sufficient information on the characteristics of included studies. 3,62,127,130,132,133 However, only one study provided information on the inclusion of grey literature and on a priori design,127 two reported on duplicated selection and extraction,62,127 two provided a list of excluded studies,130,133 one assessed the presence of publication bias62 and only two used the results of the risk-of-bias assessment to draw conclusions. 62,132 A potential conflict of interest was assessed in half of the included reviews. 62,130,132,133 Of the eight reviews published in full, one, by Cantisani et al. ,129 was rated as being at a high risk of bias because, for the majority of the AMSTAR and CRD criteria, the information was unclear or not reported (Figures 4 and 5). Four reviews were considered to be of good methodological quality. 62,127,130,132 The three remaining reviews were considered to be of moderate quality. 3,128,133

FIGURE 4.

Risk-of-bias assessment using the CRD criteria. N/A, not applicable.

FIGURE 5.

Risk-of-bias assessment using the AMSTAR criteria. N/A, not applicable.

Methods for review of the cost-effectiveness studies

Results of the review of cost-effectiveness studies

After deduplication of the records, 278 abstracts were screened for suitability. Seven studies were selected for full-text assessment, with only five of these studies meeting the inclusion criteria. 41,45,66,137,138 Two studies were conducted in the USA45,137 and three studies were conducted in Europe (in Italy,66 Ireland138 and the UK41). All of these studies considered the effects of crossing from CTA to CDU plus plain radiography as first-line surveillance test. The studies estimated the difference in the number of CTAs required by the new surveillance strategy and the replaced strategy.

Beeman et al. 45 attempted to determine the cost savings and outcome differences of moving from both CTA and CDU imaging at 2 weeks, 6 months and 12 months after discharge and yearly thereafter (group 1 – before 1 July 2004) to CDU imaging as the sole surveillance test after the 2-week scans (group 2 – after 1 July 2004), with CTA being conducted only if any problem was detected by CDU. The authors analysed data on 82 patients for group 1 and on 117 patients for group 2. The clinical outcomes included the number of endoleaks detected and the measurement of the AAA sac diameter. The average length of the follow-up period was 3.5 years (range 0–9 years) and 1.6 years (range 6 months–4 years) for groups 1 and 2, respectively. The authors used hospital charges for CTA and CDU and noted that the decreased charges of US$1595 per patient per year (2008 prices) or US$198 per patient per year using Medicare reimbursements were realised by eliminating CTA surveillance in group 2. Moreover, the sensitivity of CDU for detecting endoleaks was 0.71 and the specificity was 0.99, whereas the sensitivity of CTA was 0.731 and the specificity was 0.991. The authors could not find any difference in aneurysm sac diameters measured by CDU and CTA when these scans were performed within 1 month of each other in group 1. Although the authors do not advocate eliminating CTA from the surveillance protocols, they state that its use should be limited to those circumstances in which it could provide other details about problems first detected by the ultrasound examination.

Bendick et al. 137 evaluated the use of an ultrasound contrast agent to enhance imaging for stent–graft surveillance and compared the costs of this technique with those of CTA. Data on the first 40 patients received in their vascular laboratory were analysed. The follow-up examinations ranged from 1 to 24 months after graft placement, with a mean follow-up time of 13.7 ± 6.1 months. Clinical outcomes included type I or type II endoleaks and costs were calculated using hospital charges with average costs per study that were equal to US$2779 for CTA, US$525 for CDU (including contrast) and US$147 for plain film abdominal radiography. No details of the price year were stated. The authors reported a sensitivity to the presence of any endoleaks of 53% (8/15) for CDU, 93% (14/15) for CEU and 73% (11/15) for CTA. The average 3-year charges per patient were US$22,232 and US$5376 for CTA-based surveillance and surveillance using CDU (including contrast) plus radiography, respectively (a saving of US$16,856 per patient over 3 years). The authors concluded that the technique of duplex ultrasound with an ultrasound contrast agent should become the method of choice for stent–graft surveillance if the promising early results shown in their present series can be demonstrated in a larger patient population.

Chisci et al. 66 evaluated whether or not the imaging modality of surveillance influenced the detection of these conditions affecting the rate of asymptomatic secondary interventions (i.e. endoleaks, AAA expansion, migration, graft infection, graft thrombosis, conversion to open repair, postoperative renal impairment, bowel ischaemia and myocardial infarction). The authors followed a cohort of individuals for whom the follow-up protocol was changed at a given date. Protocol I, performed from January 2003 to December 2006, consisted of CDU plus CTA at 1 month after the procedure and every 6 months thereafter. Protocol II, performed from January 2007 to June 2010, included CDU plus CTA at 1 month after operation, CDU plus plain radiography every 6 months thereafter and CTA carried out during follow-up only for specific conditions. The authors analysed data for 376 individuals in protocol I and 341 individuals in protocol II with a mean follow-up of 1148 days (range 1–3204 days) and 942 days (range 1–1512 days), respectively (p < 0.001). On the 3-year analysis, the authors reported that protocol I cost approximately €3000, whereas protocol II cost approximately €1000; this was a threefold reduction in overall costs for protocol II (p < 0.0001). However, there were no details of the costing method used or the cost categories included in this analysis. The authors concluded that the detection rate of asymptomatic secondary interventions following EVAR is not affected by the type of surveillance imaging and that a surveillance schedule based primarily on CDU and radiography appears to be justified.

Gray et al. 138 retrospectively reviewed the CDU and CTA scans of all 145 patients who underwent EVAR at the Mater Misericordiae University Hospital, Ireland, from 1 June 2003 to 1 July 2010 and compared their results for endoleak detection and determination of residual sac size. The authors’ aim was to assess the cost savings obtained if CDU was employed as a first-line surveillance tool following EVAR and to compare CDU and CTA in terms of efficacy. A total of 484 scans (68%) from 114 patients (78.6%) were available for comparison. The hospital protocol for patients after EVAR included CDU and CTA scans of the aorta within 7 days of surgery. After discharge, all patients underwent CDU at 1 month and then CDU and CTA at 6 months, 12 months and annually thereafter, provided that there was no documented endoleak on either the CDU or CTA. The costs of CTA (€500 per scan) and of radiography (€85) were considered in the costing calculations (expressed in 2010 prices). However, no details of the unit cost sources were reported. The authors found that CDU was 100% sensitive and 95.7% specific in the detection of endoleaks, with a positive predictive value of 28.7% and a negative predictive value of 100%. Furthermore, no statistically significant difference between the two imaging modalities was detected for the determination of residual AAA sac diameter. The authors hypothesised that a reduction in costs resulted from a change in protocol for the year 2010. Adopting a protocol with CDU and abdominal radiography as the first-line surveillance tool would result in a reduction in the number of postoperative CTA scans from 235 to 36. This would equate to a reduction in expenditure from €117,500 to €34,915 (a saving of €82,585). The authors concluded that CDU combined with plain abdominal radiography could safely replace CTA as the primary long-term imaging modality, resulting in a significant cost saving without the loss of scan accuracy.

The only study conducted in the UK41 was a retrospective review of a prospectively maintained database of all patients undergoing elective, standard EVAR at a large tertiary referral centre (Royal Liverpool University Hospital). As with the other studies, the authors assessed the efficacy of a modified post-EVAR surveillance protocol based, primarily, on CDU and radiography, with CTA triggered only by significant findings on the CDU scan or radiography. The study included patients who had their EVAR operation between 1 August 2005 and April 2009, for whom at least 1 year’s post-surgical follow-up data were available. The primary outcome measure considered was aneurysm rupture, whereas the secondary outcome measures included the requirement for secondary intervention and the number of CTA scans avoided, from which the radiation dose reduction and cost savings were calculated. The costs were expressed in 2010–11 prices and NHS tariffs were used to cost the tests (radiography, €35.71; CDU, €187.47; CTA, €269.61; exchange rate: £1 = €1.18). 41 The authors analysed data on 194 patients who underwent a total of 606 sets of surveillance imaging: 194 sets at 1 month (radiography, CDU and CTA) and 412 per protocol sets thereafter (radiography and CDU). No patient presented with ruptures or aneurysm-related complications that were not identified by the modified surveillance protocol. The authors obtained the number of tests performed in the modified protocol and compared this with those that would have been performed for the same group had the protocol not been modified: 412 tests would have been conducted for the follow-up period that would have costed €14,711 for radiography, €77,326 for CDU and €111,078 for CTA. With the modified protocol, the number of tests and costs for radiography and CDU remained the same. However, only 71 CTA scans were conducted, with a cost of €19,142, which was a saving of €91,936. The authors concluded that follow-up after EVAR primarily based on CDU and radiography was feasible and safe, and reduced the use of CTA substantially, with consequent reductions in exposure to ionising radiation or an intravascular contrast medium, and costs.

Summary

Five studies met the inclusion criteria. All of these compared a surveillance strategy based on CDU or CEU with a CTA-based strategy. All of the studies assessed the reduction in costs as a result of the smaller number of CTA scans performed in the modified protocol. Only cohort studies were identified in the searches. However, the studies by Beeman et al. 45 and Chisci et al. 66 compared cohorts before and after the protocol changes took place. In the other studies, an economic analysis was conducted on the basis of the resources required (i.e. the number of CTA scans performed) if a hypothetical alternative protocol were to be used. Although all of the studies41,45,66,137,138 fairly agree on the clinical outcomes of interest (i.e. endoleaks, AAA size and the need for secondary interventions), the reporting of costs and the cost method was disparate; although one study gave details of the cost categories, the unit cost used, the sources and the price year used, another study reported only the final cost calculations.

The only study from the UK that could inform NHS policy was the study conducted by Harrison et al. 41 However, neither this nor any of the non-UK studies used a preference-based measure of effectiveness. Moreover, judging from the number of scans used in the authors’ cost calculations, the follow-up period considered was just over 2 years. This time horizon might not be long enough to consider all of the costs and consequences that are relevant for the question posed. As such, a new economic model was developed to assess the relative efficiency of CDU or CEU in the surveillance of individuals after EVAR. This is reported in the next section.

Methods

The economic model

A Markov model approach was selected for the decision-analytic model exercise. 135 Markov models have Markov states in which individuals spend a period of time, which is named a ‘cycle’. At the end of each cycle, the individuals can remain in their current Markov state or move to another state. The probabilities of moving to other Markov states or remaining in the current state are named ‘transition probabilities’. Individuals in the model would accrue costs and benefits (e.g. life-years) depending on the time spent in each Markov state and the interventions and/or events modelled within each Markov state. Markov models are particularly suitable to model recurrent issues and chronic diseases. They allow the incorporation of health states to reflect the movement of patients during surveillance, further diagnosis and treatment. In the current study, model states reflect the underlying condition (e.g. post-EVAR state with known or unknown complications), together with the decision on treatment (e.g. reintervention after EVAR). In all of these models, an absorbing state is included, in which all individuals would end up if the model was run for a sufficiently long period of time (e.g. Markov death state).

Description of the Markov model and the model structure

The model overall state-transition diagram is reported in Appendix 14. A simpler, schematic representation of the Markov model is shown in Figures 6–9. In these figures, circles represent the Markov states, whereas squares represent an event that occurs within a Markov state (e.g. an emergency procedure). Arrows show the direction of the possible transitions in the model. Unless specified, individuals can remain in a Markov state for more than one cycle. Eight Markov states are considered in the model:

1. normal (no residual EVAR complication)

2. abnormal Ia (intervention required)

3. abnormal Ib (intervention required – endoleak)

4. abnormal II (no intervention required)

5. enhanced follow-up (normal)

6. elective surgery (one cycle, temporary state)

7. enhanced follow-up (abnormal II)

8. death.

FIGURE 6.

Schematic diagram of the surveillance for EVAR Markov model; underlying condition.

FIGURE 7.

Schematic diagram of the surveillance for EVAR Markov model (from abnormal Ia, intervention needed Markov state). FN, false negative; TN, true negative; TP, true positive.

FIGURE 8.

Schematic diagram of the surveillance for EVAR Markov model (from abnormal II, no intervention needed Markov state). FN, false negative; TN, true negative; TP, true positive.

FIGURE 9.

Schematic diagram of the surveillance for EVAR Markov model (from normal, no residual EVAR complications Markov state). FP, false positive; TN, true negative; TP, true positive.

Figure 6 shows the four Markov states that reflect the underlying condition but are yet to be detected (‘undiagnosed’ states 2–4 above). Figures 7–9 show one of these four Markov states representing the underlying condition together with the Markov states that individuals can move to (e.g. those states that result from a diagnosis – the ’diagnosed’ side of the figure – correctly or not). The performance of a surveillance strategy in this model is given by the correct identification of those individuals with an abnormal condition and the corresponding transfer of those individuals into the true-positive states on the right sides of Figures 7–9.

All individuals start in the ‘normal (no residual EVAR complication)’ health state and can develop abnormalities as the model runs (see Figure 6). The surveillance strategies aimed to detect a variety of conditions and complications . These were generically described as ‘abnormalities’ and were divided into two categories: conditions that trigger an elective intervention and conditions that, on clinical assessment, necessitate closer follow-up (e.g. additional 6-month CTA scans). The first category was subdivided into two: abnormal 1a includes non-endoleak-triggered interventions (e.g. limb occlusions, graft infections) and abnormal 1b counts for the endoleak-prompted interventions (e.g. type I, III and IV, type II or endotension with sac expansion > 5 mm). Examples of patients with conditions within the close follow-up group (abnormal II) are those with type II endoleaks with a sac expansion of < 5 mm in 6 months or with limbs with kinking or partial thrombosis.

These abnormal Markov states are tunnel states. In tunnel states, individuals go through the health states in a defined sequence. The rationale behind these tunnel states for individuals with an abnormality is to count the length of time for which the individuals have had the abnormality. Thus, the probability of experiencing a rupture as a result of, for instance, a type I or III endoleak will increase as length of time for which the individual has had the endoleak increases (i.e. the length of time that the individual remains in the undetected health state). Finally, individuals can move from any health state to the absorbing death state (presented, for simplicity, at the side of the figures).

Abnormal Ia

Once an abnormal condition has been identified, individuals move to another Markov state, in which they can be treated (see Figure 7) or followed up more closely (see Figure 8). Figure 7 shows the pathway of individuals who developed a non-endoleak abnormality that would require elective surgery. If the condition is undetected through surveillance (a false-negative result), the person remains in the abnormal Ia state. This person can experience an event and, as a result, go through an emergency intervention. If the situation is resolved, the person would move back to the original normal state. If the underlying condition is detected through surveillance (a true-positive result), the individual would move to the elective surgery state. This is a temporary state and individuals can remain within this state only for one cycle (to have surgery and subsequently move on to another health state). Again, once the surgical intervention is successful, the individual moves back to the original normal state. Moreover, individuals can move to the absorbing death state from any of the other health states as a result of an emergency (e.g. rupture), surgery (e.g. hospital mortality) or other comorbidities (general population mortality). A similar structure was followed for individuals who developed an endoleak abnormality that required an elective intervention (abnormal Ib Markov health state).

Figure 8 shows the pathway for the abnormal II health state. Undetected individuals (a false-negative result) will remain in this state (i.e. represented by the back arrow from the state). These people have the risk of experiencing an event that would trigger an emergency intervention. If the abnormality is detected (a true-positive result), the individuals will move to enhanced follow-up, which is defined as 6-month visits at which a CTA-based assessment is conducted. If the patient is stable and no further interventions (e.g. elective surgery) are decided after 2 years (four model cycles), the patient returns to the surveillance pathway (e.g. annual check-ups based on the original surveillance test – CDU, CEU or CTA).

Abnormal II

The model also includes a false-positive state for those individuals without an abnormal condition (see Figure 9) but with a positive test result [e.g. enhanced follow-up (normal)]. If the individual developed an abnormality while under enhanced follow-up, they would either move straight to elective intervention (e.g. for abnormal Ia or Ib) or remain under enhanced follow-up but within a different Markov state (e.g. for abnormal II).

Normal state

Individuals can suffer an event between surveillance visits. This is considered in the model to be an event within a Markov state and is shown in Figures 6–9 as a square (e.g. emergency procedure). The model assumes that individuals who survive an unsuccessful second intervention could undergo a third intervention. However, a pragmatic assumption based on small probabilities was made and individuals can either have a successful third intervention or die.

Parameter estimates used in the economic model

The parameter estimates required to populate the economic model were obtained from the results of the clinical effectiveness search, which was supplemented by structured and focused searches (e.g. of EVAR trials with a longer follow-up period). When no suitable data resulted from these searches, expert opinion was sought. Probabilities gives details of the probabilities, unit costs and utility weights used in the model. Also provided within this section are details of the probability distributions used for the probabilistic sensitivity analysis.

Probabilities

The model starts with the whole cohort in the normal state, and therefore no prevalence data were necessary. The annual incidence of abnormalities was developed from data reported by Tang et al. ,93 based on ENGAGE (Table 15). The authors report 1-year data, excluding the first 31 days after EVAR. From a total of 325 patients, 25 abnormalities were reported (20 endoleaks, one stent–graft occlusion, three stent–graft stenoses and one other event related to the stent–graft). These data were initially used to obtain the proportions of cases within each of the model abnormal subgroups (graft occlusion for abnormal Ia, types I and III endoleaks for abnormal Ib, and type II endoleak for abnormal II). However, in consultation with the experts in the project advisory group, these figures were revised, as it was believed that a higher proportion of abnormalities Ia and Ib are currently seen in UK practice (Professor Srinivasa Rao Vallabhaneni, and Dr Russell Jamieson, NHS Grampian, 2017, personal communication). Beta distributions were used to assess the uncertainty around the central point estimates.

TABLE 15 - Incidence and mortality

Variable	Value	Probability distribution	Source
Annual incidence of abnormal cases	0.04	Beta(13,312)	Tang et al.93
Proportion of abnormal Ia subgroup from all abnormal individuals	0.15	Beta(15,85)	Based on expert opinion
Proportion of abnormal Ib subgroup from all abnormal individuals	0.15	Beta(15,85)	Based on expert opinion
Proportion of abnormal II subgroup from all abnormal individuals	0.70	Beta(70,30)	Based on expert opinion

Unfortunately, the systematic review of clinical effectiveness could not identify any studies that directly compared the performance of alternative CDU and CEU surveillance strategies. Therefore, test performance data were used to feed the model. Test sensitivity and specificity (Table 16) were obtained from the systematic review by Karthikesalingam et al. 132 Alternative data were available (see Diagnostic performance of imaging modalities for surveillance after endovascular abdominal aortic aneurysm repair in Chapter 2). However, Karthikesalingam et al. 132 is the only review reporting sensitivity and specificity for all three tests (CDU, CEU and CTA), and the quality assessment of all of these diagnostic review studies resulted in the Karthikesalingam et al. 132 review being deemed to be of higher quality. Beta distributions were used to address uncertainty around the central parameter values.

TABLE 16 - Test sensitivity and specificity

Variable	Value	Probability distribution	Source
CDU sensitivity	0.74	Beta(48.8,17.2)	Karthikesalingam et al.132
CDU specificity	0.94	Beta(164.5,10.5)	Karthikesalingam et al.132
CEU sensitivity	0.96	Beta(28,1.2)	Karthikesalingam et al.132
CEU specificity	0.85	Beta(64.3,11.3)	Karthikesalingam et al.132
CTA sensitivity	0.70	Beta(26.1,11.2)	Karthikesalingam et al.132
CTA specificity	0.98	Beta(81.1,1.7)	Karthikesalingam et al.132
Proportion of individuals adhering to surveillance visits	0.93	Uniform(0.5,1)	Assumption based on expert opinion

The probability of having a reintervention and the risk of rupture and mortality are reported in Table 17. The probability of having a reintervention was developed from Tang et al. 93 Disregarding the first month post EVAR, 13 individuals out of 319 had a secondary procedure. The proportion of successful secondary procedures was based on the proportion of individuals free of a secondary intervention in the EVAR 1 trial – 15-year follow-up data. 48 The model allows for individuals with an unsuccessful surgery to go to a third procedure. We found no data to inform the model parameters related to the third intervention (e.g. the probability of having a third intervention and the proportion of successful interventions) and therefore the data as for the second intervention were applied.

TABLE 17 - Reintervention, rupture and mortality

Professor Srinivasa Rao Vallabhaneni and Dr Russell Jamieson, personal communication.

Variable	Value	Probability distribution	Source
Probability of having a reintervention	0.020	Beta(13,312)	Tang et al.93
Probability of not having a third intervention	0.860	Beta(541,85)	Patel et al.48
Probability of rupture (type I or III endoleak)	0.0102	Beta(3,294)	Developed from Buth et al.91
Probability of rupture (type I or III endoleak) after 1 year	0.1	Uniform(0.1,0.2)	Moll et al.31
Probability of rupture (type I or III endoleak) after 2 years	0.3	Uniform(0.3,0.3)	Moll et al.31
Probability of rupture (type II endoleak)	0.0018	Beta(3.5,1971.5)	Developed from Buth et al.91
Probability of rupture (no endoleaks)	0.0018	Beta(3.5,1971.5)	Developed from Buth et al.91
Mortality
Elective surgery	0.007	Beta(12,598)	Based on expert opiniona
Emergency surgical intervention (rupture)	0.32	Beta(36.6,77.9)	Antoniou et al.139
Rupture	0.16	Beta(30,160)	Antoniou et al.139
Age- and gender-specific general population mortality for the UK	Various	Not applicable	National Life Tables, UK146 2013–2015. Office for National Statistics

The risk of rupture for undetected endoleaks was based on an analysis of early data from the EUROSTAR registry. Buth et al. 91 conducted two cohort analyses comparing a cohort of people with type I and type III endoleaks (n = 297) with those who had never experienced an endoleak (n = 1975), and a cohort of people with type II endoleaks (n = 320) with those who had never experienced an endoleak (n = 3275). The authors state that the cumulative rate of rupture from type I and type III endoleaks was 4% at 2 years compared with 0.7% for those who had never experienced an endoleak. Moreover, the number of late ruptures in patients with type II endoleaks was not significantly different from the number of late ruptures in those who had never experienced an endoleak. The risk of rupture for individuals with type I and type III endoleaks after 1 year was adjusted based on data reported by Moll et al. 31 The authors report the risk of rupture according to aneurysm size, based on population studies. To calculate the risk in the model, an aneurysm size of 55 mm at baseline was assumed, together with a growth rate of 5 mm per cycle. These rupture risks were applied to the undetected abnormal Ib group in the model only.

Table 17 also reports the mortality data assumed in the model. Age- and gender-specific general population mortality rates were applied to the cohort. The risk of surgical death in an elective setting was based on expert opinion (Professor Srinivasa Rao Vallabhaneni and Dr Russell Jamieson, personal communication). The main event that surveillance is trying to avoid is the aneurysm rupture and its associated high mortality rate. The risk of death from a rupture was calculated based on the systematic review and meta-analysis of late ruptures by Antoniou et al. 139 The authors identified 11 studies (case series) that reported a total of 190 ruptures: 30 patients were managed with palliative care or died before surgery. Moreover, the authors reported a perioperative mortality rate of 32% (95% CI 24% to 41%).

Costs

Unit costs were obtained from NHS Reference Costs 2015 to 2016140 (Table 18). The unit costs for ultrasound tests (with and without contrast) of < 20 minutes’ duration reported in the NHS reference costs are surprisingly similar. Moreover, the stated average unit cost for an ultrasound with contrast was lower than an ultrasound without contrast. Therefore, the unit cost for a vascular ultrasound scan was used to cost CDU and CEU tests. The clinical experts noted that clinical staff (i.e. a consultant radiologist) should be present to administer the contrast agent for CEU. In addition, CEU includes a contrast agent [i.e. sulfur hexafluoride or perfluorocarbon encapsulated by a phospholipid shell (SonoVue)] with an associated cost of £46 for a 10-vial box (Mr Craig Rore, Grampian Medicines Information Centre, 2017, personal communication). Therefore, the unit cost of CEU was adjusted to add the cost of 30 minutes (Professor Srinivasa Rao Vallabhaneni, personal communication) of a medical consultation (based on a cost per hour of £135)141 plus £4.60 for the cost of the contrast. Furthermore, the cost of a CT scan of one area, with pre and post contrast, was used for CTA. Notably, NHS Reference Costs 2015 to 2016140 does not report the unit cost for a plain radiography. The cost of a plain radiography is absorbed within other cost categories (e.g. outpatient visit) because of its relatively high volume and low cost. As plain radiography was considered in all of the strategies in a similar manner (on an annual basis), an executive decision was made and the unit cost of plain radiography was not incorporated in the model. If a surveillance test outcome was indeterminate, a further assessment was assumed (i.e. with CTA) and the cost of a visit was added to the cost of the subsequent test (i.e. non-admitted face-to-face attendance, follow-up – vascular surgery).

TABLE 18 - Unit costs

PSSRU, Personal Social Services Research Unit.

Variable	Cost (£)	Probability distribution	Source
CDU	57.53	Gamma(6.23,0.11)	RD47Z – vascular ultrasound scan – NHS Reference Costs 2015 to 2016 (main schedule)140
CEU	57.53	Gamma(6.23,0.11)	RD47Z – vascular ultrasound scan – NHS Reference Costs 2015 to 2016 (main schedule)140
CTA	118.53	Gamma(13.67,0.12)	RD22Z – CT scan of one area, with pre and post contrast – NHS Reference Costs 2015 to 2016 (main schedule)140
Additional cost of CEU	72.10	Uniform(49.6,94.6)	30 minutes of medical consultation (based on an hourly cost of £135 – PSSRU hospital-based doctors with qualifications) plus contrast agent at £46 for 10 vials (one vial used per test per person). Probability distribution based on assumption141
Further assessment visit	140.21	Gamma(8.38,0.06)	WF01A – non-admitted face-to-face attendance, follow-up – vascular surgery – NHS Reference Costs 2015 to 2016140
EVAR intervention (elective)	11,925.16	Gamma(5.99,0)	Weighted average codes YR04Z and YR03Z (AAA endovascular and complex endovascular elective repair) – NHS Reference Costs 2015 to 2016140
Other surgical procedures (elective)	12,707.99	Gamma(4.03,0)	Percutaneous transluminal embolectomy or thrombolysis (weighted average codes YR23A and YQ11B plus YR12Z – percutaneous and open procedures plus stent) – elective inpatient – NHS Reference Costs 2015 to 2016140
EVAR intervention (emergency)	20,675.86	Gamma(4.3,0.0002)	AAA endovascular and complex endovascular repair (weighted average categories YR04Z and YR03Z – non-elective) and VB01Z (emergency medicine) and ASS02 (ambulance) – NHS Reference Costs 2015 to 2016140
Other surgical procedures (emergency)	18,681.87	Gamma(3.65,0.0002)	Percutaneous transluminal embolectomy or thrombolysis (weighted average for codes YR23A and YQ11B plus YR12Z – percutaneous and open procedures plus stent) – non-elective inpatient and VB01Z (emergency medicine) and ASS02 (ambulance) – NHS Reference Costs 2015 to 2016140

Endovascular AAA repair reintervention was costed as a weighted average of the unit costs for elective EVAR repair (complex and non-complex). The cost of percutaneous transluminal embolectomy or thrombolysis was used as the cost of other procedures for the abnormal Ia group. Emergency procedures were costed assuming non-elective categories plus the cost of emergency medicine (i.e. any investigation with category 5 treatment) and ambulance service (i.e. see and treat and convey).

Utility weights

Population-based utility weights for patients aged ≥ 74 years were assumed for individuals after EVAR (Table 19). These utility weights were calculated using the equation provided by Ara and Brazier142 [i.e. EuroQol-5 Dimensions (EQ-5D) = 0.9508566 + 0.0212126 × (1, if males, or 0, if females) – 0.0002587 × age – 0.0000332 × age2]. 142 The rationale behind this is that the condition is mostly asymptomatic and, as such, would have no effect on the individual’s quality of life. Interestingly, this utility weight is of a similar value to the one reported by Brown et al. 143 on the EVAR 1 RCT for baseline EQ-5D score [mean 0.75 (SD 0.22); mean age 74 years]. The utility decrement for those going into any reintervention was developed from the EVAR 1 trial. 143 This decrement was calculated as a proportional reduction from baseline until the first year post EVAR, when patients are assumed to be back to the population-based utility weight.

TABLE 19 - Quality-of-life weights

QALY, quality-adjusted life-year.

Variable	Value	Probability distribution	Source
All health states – at the start (74-year-old male cohort)	0.77	Not applicable	Age- and gender-specific general population EQ-5D score142
Surgery QALY weight decrement	2.22	Beta(4.7,209.4)	Developed from Brown et al.143 (EVAR 1 RCT) as difference from baseline

Base-case analyses

The model base-case analysis was run for a cohort of 74-year-old men for a lifetime time horizon. A 6-month cycle length was defined. The analysis was conducted from the NHS and Personal Social Services perspective. Costs were expressed in 2015–16 Great British pounds and effectiveness was expressed in quality-adjusted life-years (QALYs). Costs and QALYs were discounted at an annual rate of 3.5%. 60 The cost-effectiveness analysis results are reported using ICERs. 60 ICERs are calculated as the ratio of the difference in expected costs between two alternative strategies to the difference in expected QALYs. This ratio measures the additional costs that would have to be paid in order to obtain an extra unit of effectiveness (i.e. an extra QALY). A probabilistic sensitivity analysis was conducted, in which 10,000 iterations were run. The stability of results was verified by examining the probabilistic results for a lower number of iterations (e.g. 1000). The probabilistic analysis results are reported using cost-effectiveness acceptability curves (CEACs). 144,145 These curves show the probability that a particular strategy is cost-effective at alternative values of willingness to pay for an extra QALY.

Assessment of uncertainty (sensitivity analysis)

A number of sensitivity analyses were conducted to address the uncertainty in this economic evaluation (one-way, two-way, threshold, scenario and probabilistic sensitivity analyses).

Approximately 90% of EVAR procedures in the UK are conducted in males (see Epidemiology of abdominal aortic aneurysm). Therefore, the base-case analysis was run for a male cohort. Gender-specific data were not available and the only differing data for men and women were general population mortality rates and utility weight. Female utility weight for 74-year-olds is lower (0.75) than that for males (mean 0.77), but mortality data show a longer life expectancy that could result in a longer time for benefits, but also costs. Therefore, a further analysis was conducted using these data, to observe the effect of longer life expectancy in the model results. In addition, one-way sensitivity analyses were conducted on all cost categories (e.g. cost of tests, visits and surgery), test diagnosis sensitivity and specificity, incidence of abnormalities, adherence to surveillance and mortality as a result of an unexpected event (rupture) and emergency intervention. Ranges to run these analyses were informed by the lower- and upper-unit cost quartiles published in NHS Reference Costs 2015 to 2016140 (cost variables), the 95% CI reported by Karthikesalingam et al. 132 (sensitivity and specificity) and, for those variables for which there were no published data available, by expert opinion (e.g. adherence to surveillance).

Given the base-case and sensitivity analyses results, a threshold analysis was conducted for the cost of CEU, which explored the value that would make CEU cost-effective. Two scenario analyses were developed; the first assumes that the information from the CEU test is perfect, that is, sensitivity and specificity are equal to 1, plus no indeterminate or inconclusive results. This scenario corresponds to the notion that CEU could be the present reference standard.

A further scenario analysis was implemented, which assumed a cohort with a higher proportion (50%) of individuals belonging to abnormal Ib group together with a higher overall incidence for any abnormality (up to 10% per 6-month cycle). This scenario explored the effects of monitoring only those individuals at high risk of developing abnormalities.

The base-case and selected sensitivity analyses results are presented in Results. The full sensitivity analysis results are reported in Appendix 15.

Results

In Table 20, the base-case analysis results are reported. Annual follow-up with CDU only is the strategy with the lowest expected cost, followed by CTA only and CEU only. The strategies with higher expected costs are those that use CDU or CEU in conjunction with CTA at the start. In Table 21, the strategy expected costs are disaggregated into three cost categories: costs of surgical procedures, costs of surveillance visits and costs of tests. Consistently throughout the alternative strategies, surgical costs represent a higher proportion of the total costs. For the strategies involving CTA, the costs of the test represent over 30% of the total costs. The costs of visits were considered only in the case of a reassessment, and therefore these represent the lowest proportion in all of the surveillance strategies (i.e. between 6% and 13%).

TABLE 20 - Base-case cost-effectiveness results: men

Strategy	Cost (£)	Incremental cost (£)	QALYs	Incremental QALYs	ICER (£)
CDU	3791		6.5532
CTA	3828	37	6.5517	–0.00150	Dominated
CEU	4709	919	6.5594	0.00622	147,626
CDU and CTA, then CDU	4732	22	6.5543	–0.00510	Dominated
CEU and CTA, then CEU	5644	935	6.5598	0.00032	2,912,177

TABLE 21 - Base-case cost results: disaggregated

Strategy	Cost of, £ (%)			Total cost, £ (%)
	Surgery only	Visits only	Test only
CDU	2423 (64)	477 (13)	890 (23)	3791 (100)
CTA	2400 (63)	229 (6)	1198 (31)	3828 (100)
CEU	3146 (67)	585 (12)	978 (21)	4709 (100)
CDU and CTA, then CDU	2430 (51)	441 (9)	1861 (39)	4732 (100)
CEU and CTA, then CEU	3148 (56)	550 (10)	1945 (34)	5644 (100)

The CTA-only strategy produces the lowest number of expected QALYs (see Table 20). This can be explained by the relatively low sensitivity of CTA that was assumed in the model. As such, the CDU-only strategy dominates the CTA-only strategy (i.e. CDU has lower expected costs and a higher number of expected QALYs than CTA only). Moreover, adding CTA to CDU or CEU at the start results in more QALYs than using only one imaging modality. However, either these strategies are dominated (i.e. CDU and CTA, then CDU) or the incremental cost for an additional QALY is well above the often-accepted cost-effectiveness threshold [£30,00060 (i.e. CEU and CTA, then CEU)]. Furthermore, CEU-based strategies result in a higher number of expected QALYs than all of the other strategies, although the ICER to adopt any of the CEU-based strategies is well above the £30,000 threshold. 60

Figure 10 shows the cost-effectiveness plane for the base-case analysis. For ease of interpretation, square data markers were used for CDU-based strategies, triangle data markers were used for CEU-based strategies and dots were used for CTA-only strategies. It can be clearly observed that the CEU-based strategies produce more QALYs, but at higher expected costs, than the CDU-based strategies. Furthermore, it is of note that the CDU plus CTA and then annual CDU strategy is dominated by the CEU-only strategy. However, the ICERs to move to any of these strategies (either CDU and CTA, then CDU or CEU only) from CDU only are well above the usual cost-effectiveness threshold.

FIGURE 10.

Base-case cost-effectiveness results: men.

Table 22 shows the probabilistic sensitivity analysis results for the base case. Annual follow-up with CDU only has the highest probability of being cost-effective for any value of willingness to pay for an extra QALY below £50,000 (i.e. a probability of > 58%). Surveillance with CTA only has a probability of between 32% and 42% of being cost-effective at values of willingness to pay for an extra QALY of between £10,000 and £50,000. Adding CTA to CDU or CEU has a zero probability of being cost-effective at any willingness-to-pay values. Finally, as surveillance with CEU only produces more expected QALYs, this strategy has a growing probability of being cost-effective at increasing willingness-to-pay values. However, at £50,000, its chance of being cost-effective is just 4.1%. Figure 11 presents the CEACs. It is worth noting that the probability of CDU being cost-effective stabilises at around 60% for high values of willingness to pay for a QALY. At higher values than those reported in the figure, surveillance with CEU increases its chances of being cost-effective compared with CDU- and CTA-only strategies (i.e. 27% at £100,000 – data not shown). Finally, the ICERs calculated with the probabilistic analysis were lower than the deterministic base-case analysis reported in Table 20 (i.e. ICER for CEU with respect to CDU: £129,700 and for CEU and CTA; then CEU strategy with respect to CEU only: £2,479,000). However, these ICERs are all well above the usual threshold used in the UK (e.g. £30,000).

TABLE 22 - Probabilistic analysis results

Strategy	Proportion cost-effective at alternative willingness-to-pay for a QALY threshold (%)
	£10,000	£20,000	£30,000	£40,000	£50,000
CDU	58.0	60.5	62.6	63.5	63.8
CTA	42.0	39.2	36.6	34.6	32.1
CEU	0.1	0.2	0.8	1.9	4.1
CDU and CTA, then CDU	0.0	0.0	0.0	0.0	0.0
CEU and CTA, then CEU	0.0	0.0	0.0	0.0	0.0

FIGURE 11.

Cost-effectiveness acceptability curves, base case: men.

Sensitivity analysis results

Base-case analysis for women

Table 23 shows the results of the base-case analysis for women. General population mortality data146 for women were used for this analysis, as well as the utility weight for women > 74 years of age (i.e. mean 0.75), using the methods provided by Ara and Brazier 2010. 142 Expected costs and QALYs for women are generally higher than for males, reflecting the longer life expectancy of women. Overall, the results are very similar to those of the base-case analysis for men, with CDU having the lowest expected cost, followed by CTA only and surveillance with CTA only being dominated by surveillance with CDU only. CEU-based strategies have ICERs that are well above the often-used willingness to pay for an extra QALY threshold. 60 Owing to the similarity of these results to those for the males model run, all other sensitivity analyses were conducted using data for only males.

TABLE 23 - Base-case analysis: women

Strategy	Cost (£)	Incremental cost (£)	QALYs	Incremental QALYs	ICER
CDU	4327		7.1460		–
CTA	4367	40	7.1442	–0.0018	Dominated
CEU	5372	1046	7.1536	0.0075	138,707
CDU and CTA, then CDU	5393	20	7.1473	–0.0063	Dominated
CEU and CTA, then CEU	6431	1059	7.1539	0.0004	2,926,141

One-way sensitivity analysis results for the unit cost of the CDU test are reported in Table 24. The upper quartile for a vascular ultrasound in NHS Reference Costs 2015–16140 is £70. For this reason, a range up to £80 was used in an attempt to include other plausible values. The base-case unit cost for a CDU test was £58; thus, for any values below this, the base-case results hold. At a CDU unit cost of £80, CDU is more costly than CTA, and, therefore, CTA becomes cost-effective. CEU only improves its cost-effectiveness compared with CDU as the unit cost for CDU increases. However, the ICERs for CEU-based strategies are still above the £30,000 threshold, at a unit cost of £80 for a CDU test.

TABLE 24 - One-way sensitivity analysis for the unit cost of a CDU test

Value (£)	Strategy	Cost (£)	Incremental costs (£)	QALYs	Incremental QALYs	ICER (£)
55	CDU	3769		6.5532
	CTA	3828	58	6.5517	–0.0015	Dominated
	CEU	4709	940	6.5594	0.0062	151,067
	CDU and CTA, then CDU	4710	1	6.5543	–0.0051	Dominated
	CEU and CTA, then CEU	5644	935	6.5598	0.0003	2,912,177
60	CDU	3812		6.5532
	CTA	3828	16	6.5517	–0.0015	Dominated
	CEU	4709	898	6.5594	0.0062	144,267
	CDU and CTA, then CDU	4753	43	6.5543	–0.0051	Dominated
	CEU and CTA, then CEU	5644	935	6.5598	0.0003	2,912,177
70	CTA	3828		6.5517
	CDU	3896	69	6.5532	0.0015	45,611
	CEU	4709	813	6.5594	0.0062	130,667
	CDU and CTA, then CDU	4837	128	6.5543	–0.0051	Dominated
	CEU and CTA, then CEU	5644	935	6.5598	0.0003	2,912,177
75	CTA	3828		6.5517
	CDU	3939	111	6.5532	0.0015	73,755
	CEU	4709	771	6.5594	0.0062	123,867
	CDU and CTA, then CDU	4879	170	6.5543	–0.0051	Dominated
	CEU and CTA, then CEU	5644	935	6.5598	0.0003	2,912,177
80	CTA	3828		6.5517
	CDU	3981	153	6.5532	0.0015	101,899
	CEU	4709	728	6.5594	0.0062	117,067
	CDU and CTA, then CDU	4922	212	6.5543	–0.0051	Dominated
	CEU and CTA, then CEU	5644	935	6.5598	0.0003	2,912,177

Table 25 presents the results for the one-way sensitivity analysis for the unit cost of CEU. The lower quartile reported in NHS Reference Costs 2015 to 2016140 for a vascular ultrasound was £39; thus, a lower value was used in order to consider other lower plausible values. In addition to the £72.10 for the contrast agent and the extra staff involved (who remained unchanged for the present one-way sensitivity analysis), the ultrasound cost for the CEU base-case analysis was £58, so any values above this would result in CEU being less cost-effective. The base-case results are robust to changes in the cost of the CEU test. A threshold analysis was conducted, and, owing to CEU being more sensitive but less specific, CEU needs to be slightly cheaper than CDU in order to become cost-effective.

TABLE 25 - One-way sensitivity analysis for the unit cost of a CEU test

Value (£)	Strategy	Cost (£)	Incremental costs (£)	QALYs	Incremental QALYs	ICER (£)
20	CDU	3791		6.5532
	CTA	3828	37	6.5517	–0.0015	Dominated
	CEU	4394	603	6.5594	0.0062	96,889
	CDU and CTA, then CDU	4732	338	6.5543	–0.0051	Dominated
	CEU and CTA, then CEU	5328	935	6.5598	0.0003	2,912,607
30	CDU	3791		6.5532
	CTA	3828	37	6.5517	–0.0015	Dominated
	CEU	4478	687	6.5594	0.0062	110,408
	CDU and CTA, then CDU	4732	254	6.5543	–0.0051	Dominated
	CEU and CTA, then CEU	5412	935	6.5598	0.0003	2,912,493
40	CDU	3791		6.5532
	CTA	3828	37	6.5517	–0.0015	Dominated
	CEU	4562	771	6.5594	0.0062	123,927
	CDU and CTA, then CDU	4732	170	6.5543	–0.0051	Dominated
	CEU and CTA, then CEU	5496	935	6.5598	0.0003	2,912,378
50	CDU	3791		6.5532
	CTA	3828	37	6.5517	–0.0015	Dominated
	CEU	4646	855	6.5594	0.0062	137,446
	CDU and CTA, then CDU	4732	86	6.5543	–0.0051	Dominated
	CEU and CTA, then CEU	5581	935	6.5598	0.0003	2,912,264

Test performance

Two-way sensitivity analyses were conducted for sensitivity and specificity for each compared test. Figure 12 presentd the results for CDU. The figure shows the strategy with the highest net benefit at £30,000 per QALY according to alternative values of sensitivity and specificity for CDU. The value ranges used were broader than the 95% CIs reported by Karthikesalingam et al. 132 (i.e. sensitivity, 95% CI 0.62 to 0.83; specificity, 95% CI 0.90 to 0.97), in order to explore the effects of alternative plausible figures. At low levels of sensitivity and specificity for CDU, the imaging strategy that has the highest net benefit is CTA. Surveillance with CDU only has the highest net benefit at 93% specificity (or higher), regardless of the CDU sensitivity.

FIGURE 12.

Two-way sensitivity analysis: CDU test sensitivity and specificity (net benefit, willingness-to-pay threshold of 30,000).

Perfect information from contrast-enhanced ultrasound

There was an indication from the clinical experts that the CEU test might have become the reference standard. A scenario analysis was conducted, assuming perfect information from CEU. That is, it was assumed that sensitivity and specificity were equal to 100% and that no indeterminate or inconclusive results were possible. Moreover, a threshold analysis was conducted to explore the difference in cost between CEU and CDU that would make CEU cost-effective. The results of this analysis are reported in Table 26 and show that, if the test performance from CEU is assumed to be perfect, a cost difference of up to £55 between CEU and CDU could make CEU cost-effective at a threshold value of willingness to pay for a QALY of £30,000. Larger cost differences would shift the ICER above the frequently used cost-effectiveness threshold (i.e. £30,000).

TABLE 26 - Scenario analysis assuming perfect information from CEU. Value refers to the cost difference between CEU and CDU

Note

‘Value’ refers to the cost difference between CEU and CDU.

Value (£)	Strategy	Cost (£)	Incremental costs (£)	QALYs	Incremental QALYs	ICER (£)
40	CDU	3791		6.5532
	CTA	3828	37	6.5517	–0.0015	Dominated
	CEU	3943	152	6.5614	0.0082	18,682
	CDU and CTA, then CDU	4732	788	6.5543	–0.0070	Dominated
	CEU and CTA, then CEU	4947	1004	6.5614	0.0000	> 29 million
50	CDU	3791		6.5532
	CTA	3828	37	6.5517	–0.0015	Dominated
	CEU	4027	237	6.5614	0.0082	28,978
	CDU and CTA, then CDU	4732	704	6.5543	–0.0070	Dominated
	CEU and CTA, then CEU	5032	1004	6.5614	0.0000	> 29 million
55	CDU	3791		6.5532
	CTA	3828	37	6.5517	–0.0015	Dominated
	CEU	4069	279	6.5614	0.0082	34,126
	CDU and CTA, then CDU	4732	662	6.5543	–0.0070	Dominated
	CEU and CTA, then CEU	5074	1004	6.5614	0.0000	> 29 million
60	CDU	3791		6.5532
	CTA	3828	37	6.5517	–0.0015	Dominated
	CEU	4111	321	6.5614	0.0082	39,274
	CDU and CTA, then CDU	4732	620	6.5543	–0.0070	Dominated
	CEU and CTA, then CEU	5116	1004	6.5614	0.0000	> 29 million
70	CDU	3791		6.5532
	CTA	3828	37	6.5517	–0.0015	Dominated
	CEU	4195	405	6.5614	0.0082	49,569
	CDU and CTA, then CDU	4732	536	6.5543	–0.0070	Dominated
	CEU and CTA, then CEU	5200	1004	6.5614	0.0000	> 29 million

High-risk patient group

A relatively more sensitive test can be beneficial when a higher proportion of individuals have the condition under study. A scenario analysis was considered in which half of the abnormal individuals belonged to the abnormal Ib group (e.g. types I and III endoleaks, plus other conditions necessitating elective intervention). Table 27 presents the results from a one-way sensitivity analysis for the incidence of any abnormality. The base-case analysis assumed circa 4% incidence per 6-month cycle. In the present analysis, this incidence of abnormalities implies that 2% of abnormalities are type Ib. Alternatively, the percentages in Table 27 could be broadly interpreted as the annual incidence of Ib abnormalities. To facilitate the comparison between CDU- and CEU-only strategies, in Table 27 the ICER for the CEU-only strategy has been calculated with respect to the CDU-only strategy and not the strategy with an immediately lower cost (i.e. CDU and CTA, then CDU). The results show that CEU is more cost-effective than CDU when the incidence of group Ib abnormalites is > 6% per year (3% incidence per 6-month cycle, which corresponds to 6% in Table 27).

TABLE 27 - Scenario analysis assuming that 50% of abnormalities are Ib (e.g. types I and III endoleaks)

Note

CEU only Incremental cost and QALYs and ICER are calculated with respect to CDU only strategy.

Value	Strategy	Cost (£)	Incremental costs (£)	QALYs	Incremental QALYs	ICER (£)
4%	CDU	5966		6.5018
	CTA	5991	25	6.4961	–0.0057	Dominated
	CDU and CTA, then CDU	6918	952	6.5057	0.0038	248,084
	CEU	6929	963	6.5258	0.0240	40,126
	CEU and CTA, then CEU	7875	946	6.5270	0.0011	849,127
5%	CDU	6902		6.4818
	CTA	6919	17	6.4751	–0.0068	Dominated
	CDU and CTA, then CDU	7854	952	6.4866	0.0047	200,797
	CEU	7910	1008	6.5105	0.0287	35,167
	CEU and CTA, then CEU	8852	942	6.5119	0.0014	683,732
6%	CDU	7748		6.4635
	CTA	7757	9	6.4557	–0.0078	Dominated
	CDU and CTA, then CDU	8701	953	6.4691	0.0056	169,359
	CEU	8801	1053	6.4964	0.0329	31,965
	CEU and CTA, then CEU	9740	939	6.4980	0.0016	573,576
7%	CDU	8516		6.4465
	CTA	8517	2	6.4378	–0.0087	Dominated
	CDU and CTA, then CDU	9470	954	6.4530	0.0065	146,967
	CEU	9612	1096	6.4833	0.0368	29,756
	CEU and CTA, then CEU	10,548	936	6.4852	0.0019	494,974
8%	CTA	9209		6.4212
	CDU	9215	6	6.4308	0.0095	642
	CDU and CTA, then CDU	10,172	957	6.4381	0.0073	130,221
	CEU	10,353	1138	6.4712	0.0404	28,159
	CEU and CTA, then CEU	11,287	934	6.4733	0.0021	436,083
9%	CTA	9840		6.4059
	CDU	9853	14	6.4162	0.0103	1316
	CDU and CTA, then CDU	10,813	959	6.4244	0.0082	117,233
	CEU	11,031	1178	6.4599	0.0437	26,964
	CEU and CTA, then CEU	11,964	932	6.4623	0.0024	390,326
10%	CTA	10,417		6.3916
	CDU	10,437	21	6.4026	0.0110	1884
	CDU and CTA, then CDU	11,400	963	6.4116	0.0090	106,873
	CEU	11,655	1217	6.4494	0.0467	26,045
	CEU and CTA, then CEU	12,586	931	6.4520	0.0026	353,758

Summary of cost-effectiveness

This chapter reported on a systematic review of economic evaluations and a model-based economic evaluation of alternative strategies to monitor individuals after an EVAR intervention. Similarly to the systematic review of clinical effectiveness, only cohort studies were identified in the systematic review of economic evaluations. Five studies met the inclusion criteria. Although two studies45,66 compared outcomes before and after a surveillance protocol took place, the three remaining studies hypothesised the resource use implications (i.e. the number of CTA scans performed) of moving to surveillance strategies, using ultrasound as first-line test. Only one study gave full details of the cost calculations. Moreover, none of the studies assessed the relative efficiency of CDU and CEU, which is addressed by the current assessment. As such, the studies identified were unsuitable to fully inform the study question posed and therefore unlikely to help decision-making in the UK. Thus, a new model was developed following UK methodological guidelines. 60

The developed model included five strategies. Three of these were CTA, CDU or CEU used on an annual basis, and the two other strategies considered CTA in addition to CDU or CEU for the first surveillance visit, with CTA scans being conducted afterwards only if further investigations were needed. Plain radiography was included in all of the strategies as part of the surveillance assessment.

The model base-case results show that, once the primary EVAR surgical complications have been discarded, surveillance with CDU as a first-line test becomes the less expensive option. This strategy is less expensive and produces more expected QALYs than a strategy that uses CTA only. Adding CTA to CDU in the first surveillance visits is not worthwhile. Moreover, surveillance strategies based on CEU result in more expected QALYs, but are also more expensive, and the ICERs are well above the usual threshold used in the UK (i.e. £30,000). The our base-case probabilistic analyses show that the CDU-only strategy has a probability of being cost-effective of between 57% and 64%, depending on the cost-effectiveness threshold (e.g. 62% at £30,000).

Extensive sensitivity analyses were conducted (see Appendix 15), with the base-case results being robust for the great majority of these. The sensitivity analysis showed that a CEU-only strategy could become cost-effective at very high rates of test sensitivity and specificity (e.g. when it was assumed to produce perfect information – sensitivity and specificity of 100% and no indeterminate results). Even in this case, the cost difference between CDU and CEU should not be above £55 for CEU to be cost-effective at a £30,000 threshold of willingness to pay for an additional QALY.

As risk stratification of patients might become a feasible option, a further sensitivity analysis was conducted to explore the effect of using these surveillance strategies in a very high-risk group only. Incidence rates of > 2% per 6-month cycle were considered for the abnormal Ib group (i.e. types I and III endoleaks, together with type II endoleaks with sac expansion > 5 mm and other conditions commonly detected by non-radiography imaging modalities). At an annual incidence rate of 7% for this group, CEU-only surveillance becomes cost-effective with an ICER of £29,756 with respect to CDU-only surveillance. It is worth noting that, although an incidence of 7% for type I or type III endoleaks is unlikely to be observed in clinical practice, an incidence of 6% for type II endoleaks with sac expansion is perhaps possible.

We can conclude that the use of CDU as a first-line test for the surveillance of individuals after EVAR is cost-effective, with a probability of > 58% at the usual cost-effectiveness threshold used in the UK. 60 The analysis results are driven by the interplay of the test performance data (i.e. sensitivity and specificity) and the cost of the test. Lower unit costs together with higher specificity are needed for CEU to become cost-effective.

Clinical effectiveness

To our knowledge, this is the first assessment that considers the effectiveness and cost-effectiveness of CTA, CDU and CEU for surveillance after EVAR. The clinical evidence base for this assessment consists of two non-randomised comparative studies (with a total of 750 participants) and 25 observational cohort studies (with a total of 7196 participants), assessing various surveillance protocols after EVAR. The surveillance protocols were based on the use of either CDU and/or CEU in combination with CTA.

The majority of the included studies assessed EVAR surveillance protocols based on a combination of CDU and CTA. Three studies used CDU as the main imaging modality for surveillance after EVAR, two studies used CEU as the main imaging modality and one study used CEU in selected cases only. There were no studies that compared CDU surveillance with CEU surveillance.

The risk of bias was rated as being high or moderate for the majority of the included studies, with only three cohort studies rated as being at a low risk of bias according to the prespecified criteria for the risk-of-bias assessment (ReBIP checklist). 76,84,87

There was considerable heterogeneity among the included studies in terms of the surveillance protocols (imaging modality, frequency of imaging, duration of follow-up, reported outcomes, definition of clinical outcomes – for example, the definition of decreased aneurysm size, the axis of the diameter measured and the time points at which outcomes were assessed). Owing to the observed clinical heterogeneity, it was deemed to be inappropriate to perform a statistical synthesis of the reported outcomes.

We conducted a narrative synthesis of the main clinical findings and grouped studies according to their similarities in terms of modality and frequency of imaging. A combination of CTA and CDU was the most commonly implemented surveillance strategy. Studies that used a combination of CTA and CDU for surveillance after EVAR were published between 2001 and 2010. The second most common surveillance strategy involved CTA and/or CDU for early and mid-term assessments and CDU for long-term surveillance after EVAR. Studies assessing this type of strategy were published more recently, between 2009 and 2016.

This may indicate a growing trend towards a CDU-based surveillance. It is worth noting that one study that followed up 494 patients who underwent EVAR using CTA and CDU for early and mid-term imaging assessments and CDU for long-term surveillance reported the highest mortality and reintervention rates. 90 However, any comparisons between cohort studies are tentative, owing to the observed clinical heterogeneity. In this particular case, it is difficult to ascertain whether or not the reported high mortality and reintervention rates were observed because of the length of the follow-up period (12 years), the characteristics of the patient population or the imaging modalities used for surveillance.

Three of the included cohort studies were conducted in the UK. 41,78,82 Evidence from these studies was considered to be of moderate methodological quality, as the studies did not satisfy all of the criteria of the ReBIP checklist. One of these studies used CDU exclusively for surveillance after EVAR,78 whereas the other two studies used a strategy based on a combination of CDU and CTA. 41,82 In particular, Harrison et al. ,41 who followed up a total of 194 patients using a combination of CDU and CTA for early surveillance and CDU for long-term surveillance after EVAR, reported a mortality rate of 13% at 12 months. In contrast, a non-UK-based study that assessed 494 EVAR patients using a similar surveillance strategy reported 19.7% mortality during a median follow-up of 68 months. 90 In general, the proportion of patients requiring reintervention in the study by Harrison et al. 41 was similar to that reported by other non-UK-based studies that used a combination of CTA and CDU for early surveillance and CDU for long-term surveillance. The study by Karthikesalingam et al. 78 used CDU at 1.5, 3, 5, 9, 12 and 18 months and annually thereafter to assess the role of peak systolic velocity provided by CDU for the prediction of limb complications in a cohort of 478 EVAR patients. The authors found that serial increases in the peak systolic velocity recorded during CDU surveillance were associated with an increased risk of stent–graft limb complications. 78

The proportion of patients with type I endoleaks identified by a surveillance strategy based on early and mid-term CTA and/or CDU and long-term CDU was comparable to that identified by a surveillance strategy based on a combination of CTA and CDU throughout the follow-up period (range 0–7.9% vs. 0.8–8.3%). No type III endoleaks were reported in the eight cohort studies that used early and mid-term CTA and/or CDU and long-term CDU for the surveillance after EVAR. It is worth noting that all but one study76 were rated as being at a high or moderate risk of bias. The study by Freyrie et al. ,76 which was the only study that was rated as being at a low risk of bias in this surveillance category, reported two cases (1.1%) of type I endoleak, 23 cases (13%) of type II endoleak and no cases of type III or IV endoleak during a mean follow-up of 33 months.

Detection of limb occlusion was lower among cohort studies that used CDU for long-term surveillance after EVAR (range 0–1.1%) than cohort studies that used either CTA for long-term surveillance (range 3.1–3.7%) or a combination of CTA and CDU throughout surveillance (range 5.3–7.2%). This is, however, only an observation and not a causal association.

The study by Chaer et al. 40 evaluated the safety of long-term CDU for surveillance after EVAR. One hundred and eighty-four patients with shrinking or stable aneurysms who received CTA at 1 and 12 months after EVAR were followed up annually with CDU for up to 4 years. Freedom from endoleaks was 96% and freedom from secondary interventions was 95% at 4 years. The authors concluded that CDU-based surveillance after EVAR is safe in patients with stable aneurysms.

Similarly, the comparative study by Chisci et al. ,66 which compared CDU and CTA 1 month after EVAR and every 6 months afterwards (protocol I, 367 patients) with CDU and CTA 1 month after EVAR and CDU and radiography every 6 months afterwards (protocol II, 341 participants), reported no significant differences between the two surveillance strategies during the course of the study (3-year follow-up) in terms of reinterventions, clinical complications and mortality. The authors concluded that the current post-EVAR surveillance protocol could be simplified by adopting CDU as the main follow-up imaging modality and restricting the use of CTA to selective cases, when adverse events are suspected.

Cost-effectiveness

This assessment is the first model-based economic evaluation to consider the role of CDU, CEU and CTA for post-EVAR surveillance. Only the study by Bendick et al. 137 provided a head-to-head comparison of the three imaging modalities as first-line surveillance using data from a cohort of 40 individuals and considered both costs and clinical outcomes. All of the other economic evaluation studies identified in the systematic review compared the use of CTA as first-line monitoring with CDU only, with CTA being used after CDU for selected cases only, to provide further diagnostic information. Moreover, all of the studies assessed the reduction in costs as a result of the fewer number of CTA scans conducted in the ultrasound-based protocols. The model considered a broader measure of effectiveness based on a preference-based measure of utility in accordance with the UK economic evaluation methodological guidelines,60 as well as a lifetime time horizon with all relevant consequences from the NHS perspective. None of the retrieved studies attempted an incremental analysis of the costs and clinical outcomes. For this reason, any comparison between the study’s results and those of the earlier economic evaluations should be conducted with caution.

The model results show that a surveillance strategy based on CDU as the imaging modality of choice becomes the strategy with the lowest expected costs, in addition to producing more QALYs than a strategy based exclusively on CTA. By comparison, although a surveillance strategy based exclusively on CEU would generate more QALYs, it would be more expensive, and the ICER would be well above the usual threshold used in the UK (i.e. £30,000). In addition, the base-case probabilistic analysis shows that a CDU-only strategy would have a probability of being cost-effective of between 57% and 64%, depending on the cost-effectiveness threshold (e.g. 62% at £30,000). Adding CTA to CDU or CEU in the first annual surveillance visit is not worthwhile, as it generates more QALYs but at a very high cost per QALY.

The base-case results, which show that CDU is the least expensive option, are in general agreement with those of previous economic evaluations that reported savings because fewer CTA tests were conducted in ultrasound-based protocols. 41,45,66,137,138 However, although Bendick et al. 137 reported savings for a 3-year cost comparison between CEU- and CTA-based protocols, the model results indicate a higher expected cost for a CEU-only strategy than for a CTA-only strategy. This can be explained by the relatively lower specificity assumed for CEU in the economic model, which generates a higher proportion of false-positive results. These false-positive results will trigger further testing for a period of up to 2 years.

Extensive sensitivity analyses were conducted (see Appendix 15), with the base-case results being robust for the great majority of these. The sensitivity analysis showed that a CEU-only strategy could become cost-effective at very high rates of test sensitivity and specificity (e.g. when it was assumed to produce perfect information – sensitivity and specificity of 100% and no indeterminate results) and with a difference in the cost of CEU and CDU of < £55.

A further sensitivity analysis considered higher incidence rates for the abnormal Ib group (e.g. types I and III endoleaks and other abnormalities commonly detected by non-radiography imaging modalities). Annual incidence rates of 4% and above were used in this analysis. Compared with CDU-only surveillance, CEU-only surveillance becomes cost-effective, with an ICER of £29,756, when the annual incidence rate for this group is 7%. Although in clinical practice it is unlikely that an incidence rate of 7% for type I or type III endoleaks would be observed, an incidence rate of 6% for type II endoleaks with sac expansion is perhaps possible.

The interplay of sensitivity, specificity and unit costs of the test drives the results in the study’s model. For instance, a lower unit cost for CEU helps to make the CEU-based strategies relatively more cost-effective; however, cost on its own will not make a CEU-only strategy a cost-effective option. A higher specificity is also necessary in order to reduce the expected cost of the strategy as a result of the follow-up of individuals without an abnormality. In addition, a higher sensitivity triggers further interventions (e.g. secondary interventions for the abnormal Ia and Ib groups and further monitoring for the abnormal II group). As such, although these interventions might result in higher expected QALYs, they also add to the expected costs, with an uncertain final effect on cost-effectiveness.

The majority of patients in the cohort considered in the model will have no further need for subsequent interventions. Furthermore, in the base-case analysis, 70% of patients with an abnormality correspond to the abnormal II group. Because a large proportion of patients are elderly with multiple comorbidities, there are instances when a secondary intervention, which is considered to be technically indicated based on surveillance imaging, may not be carried out because the risk associated with the intervention is considered prohibitively high. Furthermore, in the cost–utility analysis, there is no benefit attributed to the reassurance that the abnormality is minor or from any information provided by the test. From the point of view of the economic model, following up individuals for whom no further interventions are possible just adds to the expected cost of the strategy, with no effects on QALYs. Future research should explore more broadly the effects of the information generated by the surveillance strategies and incorporate this within the economic analysis.

Clinical effectiveness

The clinical evidence identified for this assessment demonstrates that surveillance practice after EVAR is currently heterogeneous and the most effective method for surveillance has yet to be established.

Since the advent of EVAR, CTA has been the main imaging modality for long-term surveillance. A survey of current clinical practice after EVAR published by Uthoff et al. 147 in 2012, which involved 674 respondents from 52 countries worldwide, found that CTA was the imaging modality used most often for standard surveillance. A CTA scan at 1 year was scheduled by 64.5% of the respondents.

The use of CTA presents important drawbacks, including exposure to ionising radiation, which may result in an increased risk of cancer. 30 Moreover, the intravenous iodinated contrast medium used in CTA may damage renal function over time and increase the risk of contrast nephropathy. A study by Mitchell et al. ,148 published in 2010, found that the incidence of contrast-induced nephropathy was 11% among a cohort of 633 patients who received contrast-enhanced CT in the emergency department. Six patients with contrast-induced nephropathy developed severe renal failure and four (0.6%) died. 148 In most EVAR patients, these risks could be eliminated or reduced by modifying the surveillance protocol and limiting the number of CTA scans. 11,26,35,149

As the main purpose of surveillance is to identify complications and direct treatments, most surveillance protocols involve CTA scans at 1, 6 and 12 months and annually thereafter; however, some investigators have challenged the utility of the 6-month CTA in patients with a normal 1-month CTA. 46 The authors of the 5-year US Zenith multicentre trial have proposed a reduced surveillance protocol, with no 6-month CTA, for patients without early endoleaks. 36 According to the European Society for Vascular Surgery 2010 guideline,31 CTA and radiography should be used to categorise patients with and without endoleaks. Patients without an endoleak should be followed up with CTA at 12 months and with CDU and plain radiography thereafter, whereas those with a type II endoleak should receive CTA at 6 and 12 months and annual CTA and plain radiography thereafter. 31 Similarly, the Society for Vascular Surgery practice guidelines recommend CTA at 1 and 12 months during the first year after EVAR. CTA at 6 months is added to the surveillance schedule only if the 1-month CTA identifies an endoleak or other abnormalities of concern. 19 In the survey of current clinical practice after EVAR published by Uthoff et al. 147 in 2012, 48.6% of the 674 respondents agreed that, in the absence of an endoleak or AAA sac enlargement after initial CTA, no further CTA follow-up at 6 months is required.

Although the current trend is to reduce the frequency of CTA for early surveillance after EVAR or replace it with other imaging modalities, there is limited information on the optimal duration of long-term surveillance and if annual surveillance should continue indefinitely. A systematic review published by Nordon et al. 26 in 2010, assessed the rates of secondary interventions in 32 studies (17,987 EVAR patients) and reported the evidence in favour of a modified EVAR practice. The authors have observed a mean time to secondary intervention of approximately 1–1.5 years and have suggested that patients who have completed 3 years of surveillance without detection of endoleaks or sac enlargement can be discharged from standard surveillance. 26

Some investigators and clinical guidelines now recommend annual post-EVAR surveillance with CDU if the first annual CTA does not demonstrate an endoleak or residual sac enlargement. 12,19,73,150 Compared with CTA, CDU is less invasive, less expensive, easily available and less risky as it does not require the use of a contrast agent and does not expose the patient to repeated radiation. A number of studies and systematic reviews have confirmed the role of CDU in the evaluation of endoleaks. 42,62,73,78,129,151–154 CDU can be regarded as a feasible and safer alternative to CTA, especially in patients with a stable aneurysm. Indeed, the number of centres using CDU seems to be increasing. In the survey by Uthoff et al. 147 published in 2012, the use of CDU during surveillance was reported by 36.3% of centres. High-volume, experienced centres were more likely to opt for CDU surveillance after 1 year than less experienced centres with fewer cases. Moreover, centres with a lot of EVAR experience were more likely to favour ultrasound for the follow-up of type II endoleaks. 147 Similarly, a UK telephone survey administered to 41 centres with 10 years’ experience in EVAR showed that 14 out of 41 (34.1%) centres used CDU as the primary surveillance modality. 155

In general, the evidence identified for this assessment showed no significant differences in terms of reinterventions and clinical complications between strategies based on the use of CDU for long-term surveillance after EVAR and those based on the use of CTA or CTA and CDU; however, the identified studies were clinically heterogeneous and any attempt to compare surveillance strategies should be considered to be tentative.

Current evidence on the use of CEU is limited, and CEU technology has evolved considerably over the past decade. A number of studies in the literature have reported a high accuracy of CEU in comparison with single and biphasic CTA. 57,156,157 A systematic review published in 2015,130 which assessed the accuracy of CEU versus CTA for the detection of endoleaks during post-EVAR surveillance, concluded that, compared with CTA, CEU that utilises second-generation contrast agents is a highly sensitive modality for the detection of endoleaks and especially for the detection of type II endoleaks. Similarly, a study of 539 patients published by Millen et al. 43 in 2013, suggests that CEU may be useful for the resolution of clinical uncertainties that arise from conventional imaging modalities, especially in the classification of endoleaks.

There is growing evidence that the majority of reinterventions post EVAR are triggered by symptoms and are independent of standard surveillance. 39 Among the cohort studies included in this assessment, the proportion of patients requiring reintervention during surveillance ranged from 1.1% during a mean follow-up of 24 months40 to 23.8% in a cohort of high-risk patients who presented with hostile neck anatomy after a mean follow-up of 32 months,85 indicating that the risk of reintervention was not homogeneous and was related to patients’ characteristics. Karthikesalingam et al. ,39 who followed a cohort of 553 patients for a median follow-up period of 31 months (range 1–97 months) and assessed the extent to which surveillance after EVAR triggers reinterventions, found that 5.1% of asymptomatic patients underwent reintervention prompted exclusively by surveillance imaging, whereas 8.3% of patients presented symptomatically. Black et al. 37 assessed the number of secondary interventions after EVAR among 417 patients and reported that reinterventions were performed in 31 (7.4%), of whom only six (1.4% (6/417) had abnormalities that were detected by standard surveillance. Similarly, Dias et al. 38 found that the majority of follow-up CTA scans post EVAR did not lead to reintervention, and only 9.3% of asymptomatic patients (26/279) underwent a secondary procedure based on imaging findings detected by routine surveillance. The systematic review by Nordon et al. ,26 which assessed secondary interventions after EVAR from 32 papers and included a total of 17,987 cases, reported that surveillance practice alone initiated a secondary intervention in only 1.4–9% of cases.

It is possible that current surveillance practice is poorly targeted and that most patients do not benefit from an unstratified surveillance programme that does not take into account the individual risk of developing complications. 11 There is a growing interest in risk-stratified surveillance, whereby the frequency of imaging is directed by the preoperative risk of complications. Risk factors for early and late complications post EVAR have been documented. 158–160 Such an approach would allow more intense surveillance regimes to be targeted to those patients with greater risk, with less frequent surveillance in patients at low risk (Alan Karthikesalingham, St. George’s Vascular Institute, St. George’s University of London, 2016, personal communication).

Schanzer et al. 161 assessed a large population of US Medicare beneficiaries (19,962 patients) who underwent EVAR between 2001 and 2008 and found that 50% of patients were lost to annual imaging follow-up by 5 years post EVAR. For the subset of patients with 8 years of follow-up, substantive declines in imaging follow-up continued, with only 37% undergoing an imaging study between year 6 and year 8. 161 In the UK, Karthikesalingam and Holt,162 as part of the Multicentre Post-EVAR Surveillance Evaluation Study (EVAR-SCREEN), assessed 1539 patients who underwent EVAR in 10 EVAR-SCREEN collaborator centres. Five years after EVAR, 39.7% of patients remained compliant with the surveillance programme, whereas 21.4% were deliberately removed from surveillance. The authors reported that, compared with 131 compliant patients, non-compliant patients were more likely to undergo reintervention (5-year freedom from reintervention was 76.6% vs. 62.7% in compliant and non-compliant patients, respectively), but demonstrated similar all-cause mortality rates (5-year survival rate of 65.6% vs. 54.7% in compliant and non-compliant patients, respectively). 162 These findings suggest that patients who undergo EVAR should receive appropriate information and counselling about the lifelong risk of complications and the need for annual imaging surveillance. UK centres that have adopted a comprehensive informative approach towards EVAR patients have reported satisfactory compliance rates (Professor Srinivasa Rao Vallabhaneni, 2016, personal communication). These findings also highlight that the current challenge facing EVAR surveillance is the frequency/timing of imaging, which currently is not targeted to patients’ risk of developing complications. It is possible that current imaging surveillance is performed too frequently for low-risk patients and not frequently enough for high-risk patients.

Cost-effectiveness

With regard to the reported economic model and cost-effectiveness analyses, there are a number of limitations that are worth mentioning. Both the identification and the selection of the input data were challenging. In order to identify the relative effect of different surveillance strategies, it was necessary to model a baseline situation (e.g. the cost and consequences of a situation without surveillance). Unfortunately, data to model the natural history of the disease were scarce. For this reason, the attention was turned to studies that analysed registry data. When particular input data were available from more than one source, the newest study was selected in an attempt to capture the technical developments of the modalities under consideration. Moreover, in the economic model, it was assumed that the imaging modalities differentiate the same conditions, that is, the tests identify a proportion of ‘abnormalities’ according to test sensitivity and specificity data. Once the overall abnormality proportion was defined, this proportion was divided among the abnormal Ia, Ib and II groups in the same proportions, regardless of the original test performed. In addition, the performance data used (i.e. sensitivity and specificity) were based on the detection of endoleaks (all types) that was reported by Karthikesalingam et al. 132 Unfortunately, there were no sensitivity and specificity data available by type of abnormality and test, and therefore this is a limitation of the analysis.

A further assumption in the model is the perfect identification of certain abnormalities by plain radiography. In effect, all individuals developing a type Ia abnormality (e.g. non-endoleak needing elective intervention) are assumed to be correctly identified in the next surveillance visits through a plain radiography. Unfortunately, there were no data to inform the test performance for plain radiography in this context. Moreover, although plain radiography shows graft migration and kinking, the abnormal Ia group includes graft infection and limb thrombosis, which will not be seen on plain radiography. In fact, plain radiography in the model is always conducted alongside another imaging test. Therefore, the underlying assumption is that the conditions in the abnormal Ia group are perfectly identified by either plain radiography or the imaging test. A corollary of this is that the differences between the surveillance strategies are a result of the test performances for the abnormal Ib and II subgroups. This is in line with the project brief, as its main interest was related to the detection and management of endoleaks.

There are a number of structural assumptions in the study‘s model. First, no patients present with symptoms between surveillance visits. Individuals presenting with symptoms make surveillance less worthwhile. Hence, the assumption in the study’s model works in favour of surveillance. However, up to 8% of individuals could present with symptoms in a non-emergency situation (Professor Srinivasa Rao Vallabhaneni, 2016, personal communication), and the magnitude of the effect of this assumption on the model results is limited. Second, the model did not include a ‘do nothing’ alternative. Although this is recommended by a number of economic evaluation methodological guidelines,60 a no-surveillance strategy was regarded as unacceptable and unrealistic for the UK context. Third, none of the strategies considered a partial use of CEU in a search for further information if the results from the CDU test were inconclusive. This is a limitation of the present analysis and material for further research. Fifth, all strategies considered surveillance on an annual basis. This was agreed with the clinical advisors, as annual surveillance frequency was deemed to be the only acceptable option. Finally, there was no risk stratification of the cohort. An alternative model could consider different surveillance strategies with differing visit frequencies, as well as alternative test arrangements, depending on the patient’s risk of developing complications.

The model might overestimate overall survival for this patient group compared with the results of the EVAR 1 trial (EVAR trial arm). 48 A higher overall survival rate would make any surveillance programme relatively more attractive, as people would enjoy the benefits for a longer period of time. The EVAR 1 trial includes individuals who are under surveillance. As a result, the lower mortality rate in the study’s model might correspond to events and conditions that cannot be avoided through a surveillance programme. Therefore, our model might overestimate the overall expected costs and QALYs because of a higher overall survival rate; the relative effect of this issue on the modelled strategies is ultimately unclear, although it is believed to be small in magnitude.

Suggested research priorities

• Further research is needed to assess the value of targeted surveillance (i.e. patients with a greater risk of complications may receive more frequent surveillance, whereas those with uncomplicated EVAR may undergo less frequent assessments or be discharged from surveillance). A large, multicentre trial with an extended follow-up period (over many years) would be required to answer the question of the optimal surveillance strategy after EVAR; therefore, the identification of high-risk EVAR patients mandating close follow-up may be a more realistic recommendation.

• If surveillance is to be targeted, is ultrasound-based surveillance (CDU and/or CEU) satisfactory for all patient groups, or are there groups for which CTA is required to avoid excessive risk?

• The criteria for identifying patients who are at a high risk of complications (e.g. use of validated score systems, risk prediction models) require further investigation.

• There is a need to clarify the role of plain radiography as part of EVAR surveillance. If CTA is to be performed less frequently or avoided, should plain radiography be mandatory or reserved for patients with abnormalities on ultrasound imaging?

• Future research should explore more broadly the effects of the information generated by the imaging modalities used for surveillance and incorporate this within the economic analyses.

Various aspects of EVAR surveillance may also warrant further consideration, for example it would be useful if:

• some indication of patients’ compliance with surveillance could be documented across centres in order to identify best surveillance practice and ensure that surveillance protocols are engaging with patients

• national clinical registries and databases could consider recording data on complications and mortality after EVAR according to the imaging modalities used for surveillance.

Databases

EMBASE (1996 to week 36 2016), Epub Ahead of Print, In-Process & Other Non-Indexed Citations, Ovid MEDLINE(R) Daily and Ovid MEDLINE(R) (1946 to 5 September 2016).

Last date of search: 5 September 2016.

Databases

Science Citation Index (1997 to 5 September 2016).

Web of Knowledge ISI (http://wok.mimas.ac.uk/).

Last date of search: 5 September 2016.

Database

The Cochrane Library: Issue 3 2016 [CENTRAL, CDSR, DARE (www3.interscience.wiley.com/)].

Last date of search: 5 September 2016.

Database

Scopus’ Articles-In-Press (www.scopus.com/).

Last date of search: 5 September 2016.

Database

Clinical Trials (http://clinicaltrials.gov/ct/gui/c/r).

Date searched: 27 January 2016.

Database

EU Clinical Trials Register (www.clinicaltrialsregister.eu/).

Date searched: 27 January 2016.

Database

The World Health Organization’s ICTRP (www.who.int/ictrp/en/).

Date searched: 27 January 2016.

Databases

EMBASE (1996 to week 13 2016), Ovid MEDLINE(R) without Revisions (1996 to week 3 2016), Ovid MEDLINE(R) In-Process & Other Non-Indexed Citations (28 March 2016) Ovid multifile search (https://shibboleth.ovid.com/).

Date searched: 29 March 2016.

Databases

Database of Abstracts of Reviews of Effects and CRD (www.crd.york.ac.uk/CRDWeb/).

Date searched: 29 March 2016.

Database

The Cochrane Library Issue 1 2016 [CDSR (www3.interscience.wiley.com/)].

Date searched: 29 March 2016.

Databases

EMBASE (1996 to week 36 2016), Epub Ahead of Print, In-Process & Other Non-Indexed Citations, Ovid MEDLINE(R) Daily and Ovid MEDLINE(R) (1946 to 5 September 2016).

Last date of search: 5 September 2016.

Databases

NHS Economics Evaluations Database.

Centre for Reviews and Dissemination.

Date of last search: 9 September 2016.

Database

Health Technology Assessment Database [Canadian (www.crd.york.ac.uk/PanHTA/HistoryPage.asp)].

Date of last search: 9 September 2016.

Database

Research Papers in Economics (http://repec.org/).

Date of last search: 9 September 2016.