Endovascular stent grafting and open surgical replacement for chronic thoracic aortic aneurysms: a systematic review and prospective cohort study

Pathology

Most aneurysms develop chronically, over several years. In the majority of patients, these are asymptomatic but occasionally they give rise to episodic chest or back pain. Severe pain or cardiovascular collapse may be a sign of rupture or dissection. Aneurysms can also occur acutely (i.e. within days) in the context of dissection of a normal-sized aorta, infection or trauma, but such situations fall outside the remit of this project.

Anatomical definitions and classifications

As shown in Figure 1, CTAAs are subdivided into ascending aortic, aortic arch and DTA aneurysms. Aneurysms may extend across both thoracic and abdominal segments, in which case they are classified as thoracoabdominal aortic aneurysms. The ETTAA study focuses on CTAAs of ≥ 4 cm in diameter in the arch, descending or thoracoabdominal aorta. Ascending aortic aneurysms are excluded because there is an established body of evidence supporting surgical repair, and no established endovascular treatment for comparison. 4

FIGURE 1.

Diagram illustrating sections of the thoracic aorta. ST, sinotubular.

Aneurysms do not always exhibit defined abrupt proximal and distal ends but instead have dilated aortic segments that extend into neighbouring segments. In some cases, most of the aorta or the entire aorta may be dilated to some degree, described as ectatic. Aneurysms in which the aorta has dilated evenly around its circumference are described as fusiform, whereas those in which the dilatation is predominantly on one side of the aorta are described as saccular. Thus, aortic dilatation can have diverse patterns, for various reasons. The different pathological features of the aneurysm influences treatment decisions and, therefore, patient outcomes.

Aetiology

Chronic thoracic aortic aneurysm occurs because the aortic wall is in some way abnormal and cannot withstand the normal stresses exerted by the blood pressure. In approximately 80% of CTAAs, the abnormality in the aortic wall is secondary to atherosclerotic plaques, which in turn are associated with smoking, hypertension, hyperlipidaemia and obesity. 5 In the remaining 20% of cases there is a genetic defect in some structural component of the aortic wall, so that it is weaker and dilates in response to normal stresses. Examples include mutated FBN1 (fibrillin 1 gene), causing Marfan syndrome, and mutated TGFBR2 (transforming growth factor beta receptor 2 gene), causing Loeys–Dietz syndrome. 6,7 The underlying cause of CTAA is important, as this can direct treatment. For example, in cases of genetically mediated aneurysm, current opinion is that the ‘normal’ aortic wall above and below the aneurysm, which are landing zones for the stent graft, will continue to dilate and lead to migration of the stent and endoleak, and so surgery is generally preferred. As individual genetic disorders are rare, we grouped them under the umbrella term ‘connective tissue disorders’ (CTDs).

Treatment options

In the case of patients with smaller aneurysms, those who are less fit to undergo chest surgery or endovascular intervention or those who reject such interventions, treatment is confined to lifestyle modification advice (smoking cessation and dietary management) and medical management of hypercholesterolaemia and hypertension. In the ETTAA study, these non-intervention patients are described as either watchful waiting (WW) or conservative management (CM) patients (see Chapter 3). WW patients are monitored with serial (usually annual or biennial) CT or MRI scans with a view to future intervention should the aneurysm grow beyond the intervention threshold. CM patients either are considered to be at very high risk of life-threatening complications from intervention, despite having an aneurysm that is above the guideline thresholds for intervention, or have rejected potential interventions. In such patients aneurysm monitoring may decrease or stop altogether.

As aneurysms grow and the risk of fatal complications such as rupture or dissection increases, two main interventions are available: endovascular stent grafting (ESG) and open surgical repair (OSR). 10

Open surgical replacement describes procedures in which the chest cavity is opened in order to replace the diseased aortic segment, usually with woven prosthetic tube grafts. Most cases of OSR require cardiopulmonary bypass to support or reroute blood flow while the aorta is operated on. The arch and DTA give rise to important branch arteries to the head, neck, upper limbs and spine and so OSR carries a risk to these organs when the blood supply is interrupted. This risk is usually mitigated by the use of hypothermia and cardiopulmonary bypass. As the mainstay of treatment for CTAAs for over four decades, OSR has been demonstrated to reduce mortality and can be performed reproducibly in cardiac surgical centres. 11,12 Techniques have improved, but OSR still carries a risk of mortality of around 5% and of paraplegia of around 10%. 13

In ESG, a covered stent or frame (stent graft) is inserted into the arterial system at a peripheral access point (usually the femoral artery in the groin) and guided using X-rays to the aneurysm site. At the target segment, the stent springs open and fixes to the normal aorta above and below the aneurysm, so that the aneurysm is sealed and blood flows through the stent graft. The aneurysm outside the stent graft has no flowing blood and clots; thus, it is excluded from the circulation and the risk of rupture is very low. ESG for CTAAs is a more recent and less invasive technique, with reported risk of in-hospital mortality of 2–10%. It is technically feasible in many patients and excludes the aneurysm from the circulation, with shrinkage or stabilisation of the aneurysm sac in most cases, at least initially. Unfortunately, the procedure cannot be performed in all patients because it has specific anatomical requirements. It is resource intensive, as the stent grafts themselves are expensive, and it requires a hybrid theatre and an appropriate theatre team, but it does lead to a shorter length of stay and usually faster recovery. In the long term, patients who receive ESG need to be monitored as there may be leakage around/between stent components that requires urgent reintervention.

For some patients a hybrid procedure, including both stent components and surgical components, is necessary. For example, a minor surgical procedure may be completed in the patient’s neck to protect the cerebrovascular blood supply after arch stenting, thereby facilitating the endovascular placement of a stent into the arch of the aorta. When the ETTAA study began, these hybrids were most prevalent, and such patients were intended for the ESG arm, as the stent graft was considered (clinically) the predominant intervention. Alternatively, an endovascular stent may be placed into the DTA at the same time as the arch is replaced in an open surgical procedure. During the ETTAA study, these approaches became more prevalent when a hybrid graft became available. The hybrid graft is half surgical prosthesis–half stent and is inserted during open surgery, with the patient usually undergoing median sternotomy. In such cases the surgery is the predominant procedure, so these hybrid patients were placed in the OSR arm of the study.

Treatment decisions

Endovascular stent grafting and OSR are considered for arch/DTA aneurysms when the risk of rupture/dissection or death exceeds the risks associated with intervention. Risk of rupture is mostly determined by aneurysm size, measured using CT or MRI. 14,15

When the ETTAA study launched in 2014, clinical practice was informed by the 2010 American Heart Association guidelines for the diagnosis and management of thoracic aortic disease16 and the 2014 European Society of Cardiology Guidelines on the diagnosis and treatment of aortic diseases. 17 These guidelines were written by international panels of clinical experts, based on the available evidence. The American guidelines advised that operative treatment was reasonable for patients at low operative risk who had an arch aneurysm of > 5.5 cm in diameter. The level of evidence supporting this recommendation is classified as ‘B – multiple non-randomised studies of surgery versus conservative management’ in the hierarchy of evidence. 18 The European guidelines advised that an endovascular procedure ought to be considered in patients for whom it was technically feasible, but that, if surgery (OSR) were the only option, then it should be planned after the aortic diameter reached 6.0 cm. The level of evidence for the European guidelines was classified as ‘C – consensus of opinion of the experts and/or small studies, retrospective studies, registries’ in the hierarchy of evidence. 18 The main body of evidence considered by both committees came largely from data obtained from Yale University publications,19,20 alongside data from the American International Registry of Acute Aortic Dissections (IRAD) database21 and, to a lesser extent, the European German Registry for Acute Aortic Dissection Type A (GERAADA) data set22 and the International Aortic Arch Surgery Study Group's ARCH projects. 23 Analysis of the Yale registry of acute aortic dissections and ruptures indicated that the risk of life-threatening dissection or rupture dramatically increases at a diameter of 6 cm and outweighs the 5–10% risk of death, stroke or paralysis from procedural intervention. 19,20,23

Taking the two guidelines together, and especially considering that many patients requiring arch surgery may not be considered at ‘low surgical risk’, the general practice in the UK in 2014 was to consider intervention with OSR or ESG when an arch/DTA aneurysm became > 6 cm in diameter. 16,17 Clinical opinion at the start of the ETTAA study is explored more comprehensively in Chapter 2, but it is useful to state here that the risks of intervention are influenced by a variety of factors and are different for every patient. At present, there are no national or international guidelines addressing patient selection for WW, CM, ESG or OSR. In the absence of such guidelines, specialist multidisciplinary teams (MDTs) play a key role in assessing the risk–benefit profile of each patient before recommending a treatment pathway (Figure 2). Ultimately, the choice of treatment is made by the patient after appropriate explanation and discussion.

FIGURE 2.

Multidisciplinary team decision-making process for treatment for CTAAs.

Systematic review of outcomes

As part of the ETTAA study, a systematic review was conducted in January 2016 to assess the available evidence regarding the effectiveness of ESG compared with OSR for CTAA in the aortic arch or DTA. The review protocol was registered in the PROSPERO database (CRD42017054565)31 and details can be found in the published article. 32

Briefly, no randomised studies comparing ESG and OSR have been published. Cohort studies and case–control studies matched on key outcomes were included if patients had elective treatment for arch/DTA aneurysms and some attempt had been made to adjust for selection bias. Five comparative cohort studies met the inclusion criteria, reporting a total of 3955 ESG and 21,197 OSR patients. In accordance with the ROBINS-I (Risk Of Bias In Non-randomized Studies – of Interventions) tool,33 one study was judged to be at moderate risk of bias34 and the remaining four were judged to be at severe risk of bias because of the potential for confounding. 35–39 Early mortality rates (30 days or to discharge) ranged from 3.1% to 6.1% after ESG and from 1.5% to 20% after OSR, with extreme rates arising from small studies. The meta-analysis of unadjusted short-term all-cause mortality favoured ESG [odds ratio 0.75, 95% confidence interval (CI) 0.55 to 0.1.03]. Adjusting for heterogeneity between small and large studies, the odds ratio did not change substantially (0.71, 95% CI 0.51 to 0.98). Meta-analysis of long-term all-cause mortality could not be carried out owing to differences in how results were reported. Overall, the long-term mortality was higher for ESG in larger studies and higher for OSR in smaller studies. For example, in von Allmen et al. ’s study36 of 618 patients, the hazard ratio (HR) for ESG relative to OSR up to 5 years, adjusting for age and sex, was 1.45 (95% CI 1.08 to 1.94; p = 0.013). Conversely, the Gore TAG study of 234 patients reported identical survival of 63% to 5 years for the two groups (log-rank p = 0.625),35 and the study of 28 patients by Piffaretti et al. 39 reported higher (non-significant) long-term survival with ESG. Freedom from reinterventions in the long-term also favoured OSR.

Overall, studies reporting short-term non-fatal complications suggested fewer events following ESG, although limited data prevented meta-analysis (Table 1). 35–39 However, Hughes et al. ’s 2014 study38 of 8967 patients reported lower odds of neurological complications (odds ratio 0.48, 95% CI 0.26 to 0.86; p = 0.015), pulmonary complications (odds ratio 0.38, 95% CI 0.21 to 0.67; p = 0.001) and cardiac complications (odds ratio 0.24, 95% CI 0.15 to 0.37; p < 0.001) for ESG patients. The Gore TAG study35 reported a substantial endoleak rate for ESG patients of 8.5%.

Although this systematic review was, to our knowledge, the first to consider evidence from non-randomised studies directly comparing ESG and OSR for treatment of elective arch/DTA aneurysms in CTAA patients, it identified increasingly dated evidence only, and this was limited by either small size or severe risk of bias. The conflicting evidence reinforced the need for updated evidence on UK practice and comparisons of short- and long-term outcomes of ESG and OSR.

Additional important studies reporting outcomes

Five relatively large, recent cohort studies40–44 reported clinical and cost outcomes but were not eligible for the systematic review because of the inclusion of a heterogeneous cohort. All five studies acknowledged important differences between ESG and OSR cohorts, which resulted in biased comparisons. ESG tended to be chosen for older patients with more comorbidity, many of whom may have been unsuitable for OSR. Despite this, the risk of death, paraplegia or other complications appeared to be lower after ESG than after OSR. Conversely, the need for reintervention was greater after ESG as a result of technical failures of the stent over time, and with each reintervention there was an added risk of complication owing to either the increased complexity of the procedure or the deteriorating health of the patient. One UK44 and one US41 study compared the costs of ESG against those of OSR procedures in the context of CTAAs. Both studies found open surgery to be more expensive by approximately US$6700 and £1650, but they considered in-hospital cost only, excluding reintervention. No formal economic evaluation has been performed. Therefore, there is a lack of economic data to guide decision-makers in allocating the scarce resources available.

Preparation of resources

An initial design period involved production of study resources, including assembling an expert panel, defining clinical criteria and designing the study questionnaire. Thereafter, the consensus study was carried out in two rounds, combining features of both the Delphi survey technique and the nominal group technique. 47 The Delphi technique uses questionnaires and anonymised responses from experts to identify consensus where it exists. The nominal group technique allows further refinement of consensus in a face-to-face meeting of the panel (the nominal group), where experts discuss reasons for their decisions with the group and have the opportunity to revise their decision. The two rounds are described in greater detail below. The initial Delphi survey was conducted during autumn 2015 and the nominal group technique meeting was held in January 2016.

Round 1: Delphi survey

For each clinical scenario, participants scored the ‘appropriateness’ of all four management options by indicating how strongly they considered each treatment option to be ‘appropriate’ on a scale of 1–9, where one represented ‘not at all appropriate’, five represented ‘just appropriate’ and nine represented ‘most appropriate’ (see Report Supplementary Material 1 for an example of a completed Delphi survey). Experts were guided to score for all four treatment methods for each clinical scenario and to avoid just marking the most ‘appropriate’. When an expert recorded ‘most appropriate’ for more than one management approach, this suggested equipoise for that expert.

When completing the scores, experts were asked to represent the opinion of their local multidisciplinary team as far as possible and to follow established definitions for clinical attributes. For each scenario, experts could assume that WW, CM, ESG and OSR were available, with no anatomical/morphological contraindications. ‘Hybrid’ interventions that included a component of conventional surgery as well as an endovascular stent graft were classified as OSR if they involved opening a body cavity (e.g. visceral artery bypass, re-implantation of innominate artery origin) and otherwise as endovascular repair (e.g. carotid to subclavian bypass through neck incision). No specific stipulations were given regarding methods of assessing the risk of intervention, but panel members were asked to follow local standard clinical practice.

Data analysis: round 1

The anonymised results of round 1 were summarised as medians, interquartile ranges and whole ranges using box-and-whisker plots. For example, Figure 3 depicts the scoring for a clinical scenario with a median ‘appropriateness’ score of eight for OSR (range 4–9, interquartile range 7–9).

FIGURE 3.

Example of round 1 results for one clinical scenario: aneurysm size of 7.1–8 cm in diameter, in the aortic arch, in a patient with no connective tissue disorder, aged < 65 years and of low surgical risk.

To assess disagreement and appropriateness (and, thus, define consensus) the Research ANd Development (RAND)/University of California Los Angeles appropriateness method was used. 48 This considers the dispersion of individual scores and identifies scenarios in which expert responses are clustered at either end of the 9-point Likert scale, so that consensus is evident. Fitch et al. 48 argue that in cases when agreement is good, the distribution of responses should be narrow, and in cases where there is disagreement, the distribution should be wider. The width of the distribution is measured by the range between the 30th and 70th percentile, known as the interpercentile range (IPR). However, Fitch et al. 48 found that, in general, the IPR required for disagreement was smaller when responses were symmetrical than when they were asymmetrical, with respect to the middle of the distribution. To overcome this, they developed the asymmetry-adjusted IPR (IPRAS), which includes a correction factor for asymmetry. In this method, disagreement between experts is concluded for the ith scenario if the IPR > IPRAS for that scenario, or, conversely, IPR ≤ IPRAS indicates consensus. Clinical scenarios for which consensus was demonstrated in round 1 were noted and these were not taken into round 2.

Round 2: nominal group technique

Round 2 was completed at a face-to-face meeting of the expert panel moderated by one of the investigators (SRV). For each clinical scenario entering round 2, ‘appropriateness’ scores of the treatment options from round 1 were displayed as box-and-whisker plots, as demonstrated in Figure 3. The expert panel was given 60 seconds to study each summary, after which a brief discussion was held in which individual experts explained the reasons for their treatment choice. The role of the moderator was to clarify ambiguities, ensure a balanced discussion, give everyone a chance to express their opinion to the group and explore reasons for divergent views. It was decided prospectively that each expert would be given a maximum of 60 seconds of uninterrupted time to express opinions. Experts were concise, discussions were constructive and curtailment by the moderator was never required. After each discussion, individual experts were asked to select their single most ‘appropriate’ management strategy or to indicate more than one treatment method if there was equipoise between them.

Round 1: Delphi survey

Twenty experts from 13 centres returned round 1 scores. The expert panel consisted of an anaesthetist, an interventional radiologist, five cardiac surgeons and 13 vascular surgeons. Among the 360 scenarios considered, consensus was reached in 167 (46%) and the remaining 193 were discussed in round 2. The consensus achieved in round 1 was predominantly that WW was most appropriate for cases involving smaller aneurysms (< 6.0 cm in diameter), OSR was most appropriate for arch aneurysms in low-risk scenarios and ESG was most appropriate for DTA aneurysms in low- or medium-risk patients without CTDs.

Round 2: nominal group technique

Twelve experts, nine vascular surgeons and three cardiac surgeons took part in round 2, during which consensus was reached for a further 80 (22%) scenarios and equipoise between two different treatment modalities was noted for 34 (9%) scenarios, leaving neither consensus nor equipoise for a total of 79 scenarios (22%). Outcomes at the end of round 2 are presented in Tables 3 (aneurysms of ≤ 6.0 cm in diameter) and 4 (aneurysms of > 6.0 cm in diameter).

Inclusion and exclusion criteria

Setting

All NHS hospitals that treat or manage patients with CTAA in a MDT setting or specialist clinic were eligible to participate in the study.

Patient and centre recruitment

Centres were recruited after completion of a feasibility questionnaire confirming that they treated patients with CTAA by ESG, OSR or both. In addition, hospitals that cared for patients using WW or CM were eligible to participate if they referred patients for intervention to a centre also participating in the ETTAA study. The initial intention was to recruit 8–10 large centres but, owing to slow recruitment, it became necessary to open the study to 30 hospitals.

Patient eligibility for the study was assessed either in a MDT setting or in a specialist clinic at participating centres. Eligible patients were enrolled, and consent to collect and retain the patient’s data was taken by local research personnel. Consent was obtained face to face or by post/telephone with the consent form posted to the research team.

Study groups

Patients were divided into four groups, depending on the planned management at the time of recruitment:

1. WW – patients with smaller aneurysms at low risk of rupture who were not expected to undergo a surgical or endovascular procedure as part of the current management plan, but for whom these interventions may be a future option should the aneurysm expand.

2. CM – patients with aneurysms of a size where risk of rupture is significant, who were not expected to undergo a surgical or endovascular procedure as part of the current or future management plan due to patient choice, comorbidities or procedural risk.

3. ESG – patients for whom the risks around intervention were considered lower than the risks of rupture, who were referred to a vascular surgeon for aneurysm repair.

4. OSR – patients for whom the risks around intervention were lower than the risks of rupture, who were referred to a cardiac surgeon for aneurysm repair.

During the study some patients transferred between groups, particularly from WW to active intervention (ESG or OSR), so that the final analysis was based on the numbers of patients in each group at end of the study period, with the exception of the analysis of aneurysm growth rates (see Chapter 4 for more details).

Management and interventions

Patient group allocation at recruitment

Once consented, patients took one of two typical pathways depending on whether or not an intervention had been planned and a date of procedure fixed. At recruitment, patients were assigned to CM by the recruiting centre if future management was not expected to involve an intervention, or to WW if future management could include an intervention but further imaging was to be completed before any procedure was planned. These patients entered a non-intervention period, during which baseline characteristics, medical history and HRQoL were recorded at the point of consent and at follow-up visits planned at 3, 6, 12, 18, 24, 36 and 48 months while the patient remained in the non-intervention period. Each patient had either a CT or MRI scan to measure aneurysm size at baseline, and this was then repeated according to local management protocols, expected to occur approximately once per year.

Patients who had a known date of intervention at recruitment were assigned to the ESG or OSR group by the recruiting centre. For patients in the OSR or ESG group an additional assessment was undertaken at 1 month post intervention, with all other measurements taken at the same stages as for WW and CM patients. Procedure data, important complications, subsequent hospital admissions and serious adverse events (SAEs) were recorded by the participating centres as they occurred.

Crossover between groups

Owing to the observational nature of the study, patients switched groups according to local centre management protocols. Patients switching from WW to an intervention group were analysed as part of the non-intervention period up to the date of the procedure. Thereafter, patients entered a post-intervention period, with the timing of follow-up reset so that time zero was the date of the procedure to align with those who went straight to procedure.

Amendments to the work packages

Work package 6 was not completed as initial analysis showed that the database included only 23 scans in 11 patients who satisfied the ETTAA study inclusion and exclusion criteria. This work package is not discussed further. For other work packages, aneurysm measurements were completed in sufficient numbers at only four locations (ascending, arch, descending thoracic and thoracoabdominal). Other locations are not reported in this monograph. See Report Supplementary Material 2 for a full list of protocol amendments and Report Supplementary Material 3 for a list of departures from the original protocol.

Outcomes

The primary and secondary outcomes for work packages 1–5 are listed briefly below with definitions.

1. Primary: aneurysm diameter in the aortic arch and DTA. Secondary: survival, time to intervention, clinical complications and HRQoL. Exploratory: aneurysm diameter in the ascending aorta and the thoracoabdominal aorta.

2. Primary: survival. Secondary: clinical complications, reinterventions, re-admission and HRQoL.

3. Primary: survival. Secondary: complications, reinterventions, re-admission, length of stay and HRQoL.

4. Primary: incremental cost per QALY gained. Secondary: HRQoL.

5. Primary: expert opinion on best treatment for any given theoretical patient scenario.

Definitions of outcome measurements

Demographic and baseline variables collected

The baseline variables explored as predictors in modelling and for propensity score development are listed below.

Data collection

At baseline, medical history was taken by research personnel to identify a patient’s eligibility. Procedure details and related complications, clinical outcome data and EQ-5D-5L questionnaires were collected prospectively until the study ended, either during hospital attendances or by post/telephone and from hospital databases.

Aneurysm imaging using CT or MRI was undertaken in accordance with local practice, and anonymised copies of scans from the time of diagnosis were sent to the study team.

For patients who transferred from WW to ESG or OSR, a reassignment form was completed at the time the clinical decision was made. Following reassignment, assessment visits were scheduled relative to the reassignment date. For patients waiting over 3 months for surgery, pre-procedure follow-up assessments were completed every 3 months.

Follow-up visits had to be conducted within the following windows:

• 1- and 3-month follow-up: ± 1 week

• 6-, 12-, 18- and 24-month follow-up: ± 2 weeks

• 36-, 48- and 60-month follow-up: ± 4 weeks.

Data were transferred into ETTAA’s electronic database by the local principal investigator (PI) or a delegated researcher.

Original planned sample size and revision in October 2016

Although the ETTAA study was an observational study, we provided sample size estimates based on comparing survival between ESG and OSR. From UK registry data, 360 elective operations and stents were performed each year in the UK for arch and DTA aneurysms. 2 Based on log-rank tests, assuming proportional hazards and uninformative censoring, we calculated the smallest possible effect sizes that would be statistically significant at (two-sided) 5% error rate, with 80% power, assuming a fixed sample size and a range of predicted incidence of events in the OSR group.

Table 5 gives the range of the minimum HRs detectable for a given expected event incidence in OSR. The first set of estimates uses pre-study predictions of the final sample size (ESG, n = 293; OSR, n = 147). The second set uses October 2016 predictions of sample size (ESG, n = 170; OSR, n = 112). With these numbers, moderate to large effects (HR > 0.5) could be detected, providing that the event incidence (e.g. deaths) during the study in the OSR group was at least 30%.

In response to requests from the National Institute for Health Research (NIHR), we also estimated the power of the study to detect the minimum clinically important difference (MCID) in HRQoL. Assuming a MCID of 0.1 in the EuroQoL utility measure,59 with 5% significance and a sample size of 170 ESG and 112 OSR the power was > 90% for either two-sided t-test or Wilcoxon-Mann–Whitney U-test (the latter would account for potential ceiling effects in HRQoL). The more sophisticated modelling methods to be used, adjusting for confounders, mean that 90% is a lower bound for power. Given this number of procedures, we expected 81 CM and 730 WW patients to be recruited during the study.

Data quality assurance

The assessment of data quality was complicated by the observational nature of the ETTAA study and the movement of patients between groups over time. Data completeness was assessed using summaries of individual case report forms (CRFs) returned. Data checks broadly followed recommendations in Kirkwood and Sterne. 60 Outliers in continuous variables were detected using ranges and plotting distributions within each group. Unresolved outliers that were extreme and separated from the distribution of a variable were removed and considered missing. Categorical variables were tabulated and unexpected values were queried with centres. Consistency checks between two or more variables were performed (e.g. bivariate plots, cross-tabulations). Dates were checked against planned timing assessments and interventions, as well as relative to other assessments in the same person. All queries were checked with centres and amended in the ETTAA database.

Data summaries

Detailed methods are provided in each chapter; here we give a brief overview of the descriptive methods. Throughout, variables were summarised as the total participants per group and overall, with means and standard deviations (SDs) if normally distributed, or median and interquartile range otherwise. Categorical data were presented as frequencies and proportion in each level. Time-to-event data were summarised as the actuarial survival probability or incidence during the non-intervention and post-intervention periods using Kaplan–Meier estimates. Post-intervention survival was also calculated separately for deaths within 30 days of an intervention. To assess whether or not patients allocated to different management groups were comparable, baseline variables were compared across the four groups using one-way analysis of variance, Pearson’s chi-squared test or a generalisation of Fisher’s exact test, as appropriate. 61

Multiple-centre issues

For the analysis of aneurysm growth and HRQoL over time, clustering by centre was investigated using normal random effects in a three-level hierarchy (scan within patient within hospital) (see Chapter 4). For time-to-event outcomes in work package 1, gamma-distributed frailty terms for centres were investigated. For all other analyses, between-centre variation in outcomes could not be assessed owing to the small number of patients contributed by most centres.

Missing data

The extent of missing data per variable was quantified as the number of cases divided by the number of patients who were in the study at the point of assessment. All essential variables for work packages 1–3 are expected to be complete or to have low missing rates (< 8%). Variables for which > 25% of data were unavailable or missing were not used in modelling but were summarised and reported.

Missing data patterns were explored (e.g. monotonic, intermittent). Missing data mechanisms, missing at random (MAR) and missing completely at random (MCAR) were investigated using standard statistical tests (log-rank, Student’s t-test, Mann–Whitney U-test, Pearson’s chi-squared test, Fisher’s exact test) to assess the associations between missing variable status (yes/no) and outcomes. To inform imputation models, associations between pairs of covariates and predictors of missingness were assessed using correlations and other standard statistical tests. Missing covariates were analysed together irrespective of the reasons (death, withdrawal, loss to follow-up, test not completed). For the analysis of aneurysm growth and HRQoL (work packages 1 and 2), all patients with at least two measurements were included in random-effects models. No adjustment was made for missing measurements as estimates from such models are unbiased provided that the data are MAR conditional on the observed data. 62 For work packages 2 and 3, the analysis found little evidence against the hypothesis that data were MCAR (see Appendix 2), so the complete-case analysis is presented throughout. Sensitivity analysis assuming MAR used multiple imputation with chained equations (MICE). Imputation models included the outcome variable as well as all important covariates from exploratory analysis. Each imputation model performed predictive mean matching to impute missing data. Values were simulated for each missing variable and the resulting models were combined using Rubin’s rules. 63

Recruitment

Centre recruitment

Between 24 March 2014 and 24 July 2018, 886 CTAA patients were recruited from 30 centres (see Appendix 3 for a list of participating centres). Studies covered the majority of England but did not recruit from the devolved nations (Figure 4). Although some centres specialised in either vascular or cardiac surgery, many centres recruited patients to all four management groups.

FIGURE 4.

Locations of the 30 centres participating in the ETTAA study.

The Consolidated Standards of Reporting Trials (CONSORT)-style flow charts in Tables 6 and 7 show the number of cases assigned to each group pre (see Table 6) and post intervention (see Table 7). A total of 112 (12.6%) patients were assigned to CM, of whom 46 died during follow-up. The remaining 774 (87.4%) patients were eligible and willing to receive the intervention. Of these 774 patients, 150 (19.4%) subsequently received ESG, 135 (17.4%) subsequently received OSR and 83 (10.7%) died before any intervention. Plots of actual against target recruitment are provided in Appendix 4.

TABLE 6 - Pre-procedure follow-up relative to date of recruitment, according to intention to intervene (N = 886)

Percentages have been rounded to nearest 0.1%, so the total percentage is not always exactly 100%.

Procedure follow-up relative to date	Number of patients	Percentage of total
Intervention or WW (N = 774)
3 months
Completed	525	67.8
Died	12	1.6
Procedure	182	23.5
Withdrew	8	1.0
Missing	47	6.1
Censored	0	0.0
6 months
Completed	482	62.3
Died	24	3.1
Procedure	212	27.4
Withdrew	11	1.4
Missing	45	5.8
Censored	0	0.0
12 months
Completed	418	54.0
Died	43	5.6
Procedure	244	31.5
Withdrew	15	1.9
Missing	54	7.0
Censored	0	0.0
18 months
Completed	323	41.7
Died	58	7.5
Procedure	256	33.1
Withdrew	19	2.5
Missing	51	6.6
Censored	67	8.7
24 months
Completed	251	32.4
Died	64	8.3
Procedure	268	34.6
Withdrew	21	2.7
Missing	56	7.2
Censored	114	14.7
36 months
Completed	159	20.5
Died	77	9.9
Procedure	278	35.9
Withdrew	23	3.0
Missing	30	3.9
Censored	207	26.7
48 months
Completed	66	8.5
Died	83	10.7
Procedure	284	36.7
Withdrew	23	3.0
Missing	29	3.7
Censored	289	37.3
60 months
Completed	1	0.1
Died	83	10.7
Procedure	285	36.8
Withdrew	23	3.0
Missing	14	1.8
Censored	368	47.5
CM (N = 112)
3 months
Completed	98	87.5
Died	6	5.4
Procedure	0	0.0
Withdrew	0	0.0
Missing	8	7.1
Censored	0	0.0
6 months
Completed	92	82.1
Died	9	8.0
Procedure	0	0.0
Withdrew	2	1.8
Missing	9	8.0
Censored	0	0.0
12 months
Completed	85	75.9
Died	15	13.4
Procedure	0	0.0
Withdrew	4	3.6
Missing	8	7.1
Censored	0	0.0
18 months
Completed	60	53.6
Died	23	20.5
Procedure	0	0.0
Withdrew	6	5.4
Missing	11	9.8
Censored	12	10.7
24 months
Completed	51	45.5
Died	28	25.0
Procedure	0	0.0
Withdrew	7	6.3
Missing	6	5.4
Censored	20	17.9
36 months
Completed	28	25.0
Died	40	35.7
Procedure	0	0.0
Withdrew	7	6.3
Missing	2	1.8
Censored	35	31.3
48 months
Completed	8	7.1
Died	46	41.1
Procedure	0	0.0
Withdrew	7	6.3
Missing	1	0.9
Censored	50	44.6
60 months
Completed	0	0.0
Died	46	41.1
Procedure	0	0.0
Withdrew	7	6.3
Missing	0	0.0
Censored	59	52.7

TABLE 7 - Post-procedure follow-up relative to procedure date (N = 285)

Percentages have been rounded to nearest 0.1%, so the total percentage is not always exactly 100%.

Procedure follow-up relative to date	Number of patients	Percetange of total
ESG (N = 150)
1 month
Completed	102	68.0
Died	9	6.0
Withdrew	0	0.0
Missing	39	26.0
Censored	0	0.0
3 months
Completed	113	75.3
Died	12	8.0
Withdrew	1	0.7
Missing	23	15.3
Censored	1	0.7
6 months
Completed	109	72.7
Died	15	10.0
Withdrew	1	0.7
Missing	21	14.0
Censored	4	2.7
12 months
Completed	100	66.7
Died	25	16.7
Withdrew	1	0.7
Missing	13	8.7
Censored	11	7.3
18 months
Completed	73	48.7
Died	29	19.3
Withdrew	1	0.7
Missing	18	12.0
Censored	29	19.3
24 months
Completed	44	29.3
Died	34	22.7
Withdrew	1	0.7
Missing	13	8.7
Censored	58	38.7
36 months
Completed	21	14.0
Died	40	26.7
Withdrew	1	0.7
Missing	11	7.3
Censored	77	51.3
48 months
Completed	5	3.3
Died	40	26.7
Withdrew	1	0.7
Missing	2	1.3
Censored	102	68.0
OSR (N = 135)
1 month
Completed	71	52.6
Died	15	11.1
Withdrew	2	1.5
Missing	47	34.8
Censored	0	0.0
3 months
Completed	82	60.7
Died	17	12.6
Withdrew	2	1.5
Missing	29	21.5
Censored	5	3.7
6 months
Completed	92	68.1
Died	21	15.6
Withdrew	2	1.5
Missing	17	12.6
Censored	3	2.2
12 months
Completed	81	60.0
Died	26	19.3
Withdrew	4	3.0
Missing	16	11.9
Censored	8	5.9
18 months
Completed	62	45.9
Died	28	20.7
Withdrew	4	3.0
Missing	17	12.6
Censored	24	17.8
24 months
Completed	49	36.3
Died	32	23.7
Withdrew	4	3.0
Missing	0	0.0
Censored	50	37.0
36 months
Completed	28	20.7
Died	35	25.9
Withdrew	4	3.0
Missing	4	3.0
Censored	64	47.4
48 months
Completed	11	8.1
Died	36	26.7
Withdrew	4	3.0
Missing	3	2.2
Censored	81	60.0

Characteristics of the cohort at recruitment

Both prevalent and incident aneurysms were included in the cohort. Among the 871 patients with a record, the median time between diagnosis and recruitment was 9.1 months (range 4.0 months to 21.6 years).

Details of the available scans

Reading of baseline scans was mandated for aneurysms in the arch and DTA, but it was restricted by resources for other locations in the thoracic aorta. Table 11 provides summaries of the measurements at recruitment for the four aortic segments with consistent data collection, along with the largest diameter at any of the four sites. Aneurysms in the arch and those extending into the ascending aorta were smaller in WW and ESG patients than in patients receiving CM and OSR. Overall, aneurysms in the DTA were similar in CM, ESG and OSR and smaller in the WW group, but significantly larger in all groups than those in the arch. Aneurysms that extended into the thoracoabdomen were approximately 2 cm smaller on average.

Details of the procedures

Between 9 April 2014 and 18 June 2019, 150 patients underwent ESG as the first (index) procedure and 135 underwent OSR. Because the co-ordinating centre is primarily a cardiac surgery centre and opened before other centres, OSR procedure dates began 5 months earlier than ESG procedures. Descriptive data for procedures are shown in Table 12. One ESG recipient had a hybrid procedure involving a carotid-to-carotid bypass; similarly, 12 OSR patients had a combined procedure in which a stent was inserted via a median sternotomy. Procedures were planned to be completed in stages for 39 ESG and 34 OSR patients; of these, 14 ESG and 2 OSR were started prior to recruitment.

Over 85% of patients were treated electively, with a minority having urgent or emergency procedures. OSR was almost always completed in theatre, whereas ESG was completed in theatre, a catheter laboratory or a hybrid theatre combining the traditional operating theatre with an interventional radiology suite. The predominant mode of access for OSR was a median sternotomy, although thoracotomy/thoracolaparotomy was reported for 34 patients. Thirty-seven OSR patients had concomitant procedures, including aortic valve replacements and bypass grafts, which could have been carried out only during open surgery.

A variety of endovascular stent grafts were implanted and, for some patients, branched, fenestrated and scalloped grafts were employed. The site of access for main stent body insertion was the femoral artery in the majority of ESG procedures (see Table 12). In 136 patients in whom the relationship with the left subclavian artery was recorded, 84 (61.8%) were landed distal to the subclavian artery, 36 (26.5%) underwent bypass and 16 (11.8%) had coverage of the left subclavian artery without bypass. One ESG patient had another endovascular procedure but no other details were provided by the time of data lockdown.

There were marked differences in aneurysm site between the two intervention groups. Aneurysms were more likely to be treated with OSR if they were in the aortic arch (102 vs. 37 patients) and less likely to be treated with OSR if they were in the DTA (82 vs. 139 patients). If aneurysms extended into the ascending aorta, then they were invariably treated by OSR. Both procedures were used for aneurysms extending to the thoracoabdominal aorta (see Table 12).

Some patients required reintervention during the study. Twelve patients who received ESG as the index procedure required a second ESG between 0.6 and 35.4 months after the first procedure. In addition, 25 OSR patients had a second procedure (21 ESG and 4 OSR) between 0.4 and 36.1 months after the first procedure; of these, three received a further ESG at 1.1, 11.5 and 11.9 months after the second procedure.

Further details of outcomes of surgical procedures, including complications and NHS resources used, are reported in Chapters 5 and 7.

Population

All patients who were recruited into the study were included in the analysis of pre-intervention outcomes. Patients contributed data from either the date of diagnosis (aneurysm growth) or the date of recruitment (HRQoL, survival, clinical events) until one of the following occurred: first intervention, death, major clinical event, withdrawal or end of the study.

Outcomes

The primary outcome of interest was growth in the CTAA diameter using measurements made at diagnosis and at all subsequent time points prior to intervention.

Secondary outcomes were survival, clinical complications and HRQoL trajectories over time as defined in Chapter 3.

Statistical analysis

Aneurysm growth from first scan

Descriptive analysis

A total of 1789 scans reporting measurements of aneurysm diameter for at least one section of the aorta were returned by 886 patients. After excluding scans after interventions (n = 10), scans using techniques other than CT or MRI (n = 11) and one scan whose follow-up date was over 8 years after the first scan, data from 1767 scans in 882 (99.5%) patients were analysed. These included 6433 measurements of aneurysm diameter: 1537 in the ascending aorta, 1698 in the arch, 1761 in the DTA and 1437 in the thoracoabdominal aorta. The time between first scan and subsequent scans ranged from 3 days to 7.35 years. Mean (SD) diameters at baseline for patients with at least two scans was 4.11 (0.87) in the ascending, 3.98 (0.85) in the arch, 5.26 (1.09) in the DTA and 3.48 (0.81) in the thoracoabdominal aorta. There was little difference between baseline measurements for those with and without repeat scans.

Figure 5 shows diameter measurements at the four aneurysm sites over time. Diameters at the first scan varied substantially between patients at all sites, with DTA aneurysms showing greater average diameter and more variation between patients; thoracoabdominal aneurysms were slightly smaller on average. Trajectories over time are difficult to disentangle, although on average diameters appeared to increase only slightly over time.

FIGURE 5.

Aneurysm diameters over time, by location in the thoracic aorta. (a) Ascending aorta; (b) suprarenal abdominal aorta; (c) descending thoracic aorta; (d) aortic arch.

Results of modelling

The final model describing the diameter measurement trajectories in the absence of treatment is shown in Appendix 7; the results are summarised below.

In the final model, at time zero all covariate values are zero if the aneurysm is measured using CT, the patient is of average age (70.9 years) and height (171 cm), has never smoked, and does not have a CTD, COPD or valvular heart disease. For these ‘zero’ aneurysms, the average diameter at first scan (time zero) was 4.14 cm in the ascending aorta, 4.13 cm in the arch, 5.29 cm in the DTA and 3.37 cm in the thoracoabdominal aorta (Table 13, first row).

TABLE 13 - Mean (95% CI) aneurysm diameter (cm) at first scan by site and baseline covariates

Male of average age and height, never smoked, with no comorbidities (CTD, COPD, valvular heart disease), measured by CT scan.

Variable	Ascending aorta	Aortic arch	Descending thoracic aorta	Thoracoabdominal aorta
All covariates zeroa	4.14 (4.03 to 4.25)	4.13 (4.02 to 4.24)	5.29 (5.18 to 5.40)	3.37 (3.25 to 3.49)
Average age + 10 years	4.33 (4.20 to 4.46)	4.22 (4.10 to 4.34)	5.46 (5.34 to 5.58)	3.41 (3.28 to 3.54)
Connective tissue disorder	4.12 (3.85 to 4.39)	4.14 (3.88 to 4.40)	5.52 (5.26 to 5.78)	3.93 (3.65 to 4.21)
COPD	4.19 (4.00 to 4.38)	4.09 (3.91 to 4.27)	5.56 (5.38 to 5.74)	3.35 (3.16 to 3.54)
Valvular heart disease	4.39 (4.21 to 4.57)	4.16 (3.99 to 4.33)	5.17 (5.00 to 5.34)	3.35 (3.17 to 3.53)
Current smoker	4.15 (3.95 to 4.35)	4.04 (3.84 to 4.24)	5.52 (5.33 to 5.71)	3.65 (3.45 to 3.85)
Ex-smoker	4.04 (3.90 to 4.18)	4.01 (3.88 to 4.14)	5.41 (5.28 to 5.54)	3.49 (3.35 to 3.63)

At the time of the first scan (time zero), older age, being taller, having comorbidities such as CTD, COPD and valvular heart disease and being a past or current smoker were all associated with larger aneurysms in one or more aortic sites, although these effects varied between sites. Table 13 shows the average diameters (95% CIs) in these subgroups. For example, at first scan, age was associated with bigger aneurysms in the ascending aorta and DTA, but less so in the arch and only slightly in the thoracoabdominal aorta. Aneurysms in the DTA and thoracoabdominal aorta (but not in the other sites) were larger on the first scan if the patient had CTD than if they did not. DTA aneurysms were larger on the first scan if the patient had COPD, but this effect was not evident in other vessels. Ascending aorta aneurysms on the first scan were larger if the patient had valvular heart disease. Again, this effect was not evident in other sites. Even after adjusting for these baseline variables, there was significant variation in aneurysm size at first scan between patients (random-effects SD 0.54 cm). That is, aneurysm diameters at time zero lie within ± 1.08 cm of the average for that site for 95% of patients, even after adjustment for baseline variables.

On average, aneurysms in the DTA grew by 0.07 cm per year (95% CI 0.03 to 0.12 cm per year), compared with 0.04 cm (95% CI –0.002 to 0.07 cm per year) in the arch and 0.10 cm (95% CI 0.06 to 0.14 cm per year) in the thoracoabdominal aorta. Average growth for aneurysms in the ascending aorta was –0.001 cm per year (95% CI –0.04 to 0.04 cm per year) during the study (see Appendix 7). Figure 6 summarises the average growth trajectories in each site, showing that aneurysms in the DTA were substantially larger throughout, and grew faster than aneurysms in the arch or ascending aorta (all else being equal). There was no evidence that aneurysm growth accelerated or decelerated during this study. At first scan, there was no difference between aneurysm diameters measured by CT and MRI, all other variables being equal. However, over time the average difference between diameters measured by the two modalities increased by 0.11 cm per year, with MRI measurements being smaller (see Appendix 7).

FIGURE 6.

Average model-predicted growth trajectories from first scan by aneurysm site, assuming assessment by CT and all other baseline variables set to zero. Adapted from Sharples et al. 64 © The Author(s) 2021. Published by Oxford University Press on behalf of the European Society of Cardiology. This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (https://creativecommons.org/licenses/by-nc/4.0/) which permits non-commercial re-use distribution and reproduction in any medium provided the original work is properly cited. The figure includes minor additions and formatting changes to the original figure.

Adapted from Sharples et al. 64 © The Author(s) 2021. Published by Oxford University Press on behalf of the European Society of Cardiology. This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (https://creativecommons.org/licenses/by-nc/4.0/) which permits non-commercial re-use distribution and reproduction in any medium provided the original work is properly cited. The figure includes minor additions and formatting changes to the original figure.

Survival, hospital admissions and aneurysm-related events

Survival: descriptive analysis

In this analysis, follow-up time was from the date of recruitment to the first of death, rupture, dissection, procedure or censoring date. Pre intervention, 129 patients died during a total of 1498.2 patient-years of follow-up, a rate of 8.6% per patient-year (Table 14). Of these, 83 were in the WW group (7.4% of deaths per patient-year) and 46 were in the CM group (20.0% of deaths per patient-year).

TABLE 14 - Number of patients with readmissions and clinical outcomes prior to any procedure

Adapted from Sharples et al. 64 © The Author(s) 2021. Published by Oxford University Press on behalf of the European Society of Cardiology. This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (https://creativecommons.org/licenses/by-nc/4.0/) which permits non-commercial re-use distribution and reproduction in any medium provided the original work is properly cited. The table includes minor additions and formatting changes to the original table.

	Patient subgroup
	WW (N = 489)	CM (N = 112)	ESG (N = 150)	OSR (N = 135)
Total time at risk (years)	1119.8	229.7	77.1	71.4
Deaths (rate per patient-year)	83 (0.07)	46 (0.20)	–	–
Aneurysm-related deaths (rate per patient-year)	39 (0.03)	25 (0.11)
All admissions, n (rate per patient-year)	243 (0.22)	71 (0.31)	36 (0.46)	13 (0.18)
People with at least one admission, n (%)	147 (30.1)	41 (36.6)	22 (14.6)	12 (9.0)
Admissions, definitely/probably aneurysm related, n (rate per patient-year)	25 (0.02)	11 (0.05)	14 (0.18)	2 (0.03)
Patients admitted, definitely/probably aneurysm related, n (%)	17 (3.5)	9 (8.0)	11 (7.3)	2 (1.5)
Non-fatal ruptured aneurysms	2	–	–	–
Non-fatal dissected aneurysms	4	1	2	–
Non-fatal neurological events	5	1	–	2

Aneurysm was the primary or contributory cause of death in 64 (49.6%) patients; 45 were ruptured, 11 were dissected, three were ruptured and dissected and five reported aneurysm-related cause of death but no specific event. Thirty-nine aneurysm-related deaths were in the WW group (3.5% per patient-year) and 25 were in the CM group (10.9% per patient-year). One person had a dissected aorta but died from sepsis and aplastic anaemia; this death was not classified as aneurysm related. The results changed only slightly in sensitivity analysis that included this death as aneurysm related (not shown).

The 1- and 3-year cumulative incidence rates for all-cause death were 7.6% (95% CI 5.9% to 9.8%) and 22.4% (95% CI 18.8% to 26.6%), respectively, and for aneurysm-related death were 3.6% (95% CI 2.4% to 5.2%) and 11.7% (95% CI 9.0% to 15.2%), respectively. CM patients had much higher overall and aneurysm-related death rates than patients assigned to WW (Figure 7). Note that 19 patients do not appear in these figures because they were consented on the day of the intervention.

FIGURE 7.

Kaplan–Meier cumulative incidence curves for death from any cause and aneurysm-related deaths by CM and non-CM management (labelled WW). (a) Kaplan–Meier deaths from any cause; and (b) Kaplan–Meier aneurysm-related deaths.

Survival: results of modelling

The variables associated with all-cause and aneurysm-related deaths were similar (Table 15). As expected, in univariable models the hazard for CM was over three times that for WW. Moreover, in univariable models, hazards were significantly higher for women, patients reporting formal/informal care and patients with previous cardiac interventions, COPD, higher NYHA classification, larger aneurysms, older age and smaller frame. The relationship between age and risk of death was not linear, with patients aged > 80 years at particularly high risk, so that age was categorised for analysis. It is likely that the higher risk with lower height and weight measurements reflects a loss of muscle mass due to ageing. Variables not significant at the 5% level in both survival outcomes are not reported.

TABLE 15 - Univariable Cox regression results for all-cause and aneurysm-related deaths: complete-case analysis

CABG, coronary artery bypass grafting; PCI, percutaneous coronary intervention.

Note

Adapted from Sharples et al. 64 © The Author(s) 2021. Published by Oxford University Press on behalf of the European Society of Cardiology. This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (https://creativecommons.org/licenses/by-nc/4.0/) which permits non-commercial re-use distribution and reproduction in any medium provided the original work is properly cited. The table includes minor additions and formatting changes to the original table.

Variable	All-cause deaths		Aneurysm-related deaths
	HR (95% CI)	z-test p-value	HR (95% CI)	z-test p-value
CM	3.05 (2.12 to 4.37)	< 0.001	3.55 (2.15 to 5.87)	< 0.001
Female sex	1.93 (1.36 to 2.72)	< 0.001	2.61 (1.59 to 4.29)	< 0.001
Use of formal/informal care	2.15 (1.40 to 3.29)	< 0.001	1.92 (1.02 to 3.61)	0.042
Previous cardiac interventions CABG/PCI	2.17 (1.43 to 3.29)	< 0.001	1.28 (0.63 to 2.59)	0.492
COPD	2.28 (1.56 to 3.34)	< 0.001	2.14 (1.24 to 3.69)	0.007
NYHA per class	1.47 (1.21 to 1.79)	< 0.001	1.35 (1.02 to 1.79)	0.037
Maximum aneurysm size per cm	1.94 (1.71 to 2.21)	< 0.001	2.16 (1.81 to 2.59)	< 0.001
Age (years) at consent		< 0.001		< 0.001
61–70	2.03 (0.76 to 5.43)		1.34 (0.42 to 4.29)
71–80	4.18 (1.68 to 10.41)		2.49 (0.87 to 7.11)
> 80	8.43 (3.35 to 21.23)		5.08 (1.75 to 14.74)
Height per 10 cm	0.66 (0.56 to 0.77)	< 0.001	0.60 (0.48 to 0.76)	< 0.001
Weight per kg	0.97 (0.96 to 0.98)	< 0.001	0.96 (0.94 to 0.98)	< 0.001
BMI per kg/m2	0.95 (0.91 to 0.99)	0.008	0.91 (0.86 to 0.97)	0.002

Many of the variables in Table 15 were correlated, and the final multivariable models for all-cause and aneurysm-related deaths are shown in Table 16. Apart from age and sex, aneurysm size was the strongest risk factor for all-cause and aneurysm-related deaths. Figure 8 shows the predicted survival for patients with different aneurysm sizes at baseline. This shows that, for example, the probability of survival to 1 year exceeds 95% for maximum aneurysm diameters of 4–5.5 cm and 90% for diameters up to 6.5 cm. The 1-year risk of death increases rapidly for aneurysms of > 6.5 cm in diamater. The 3-year survival probability exceeds 90% for small (4–4.5 cm) aneurysms and 80% for aneurysms of 5–5.5 cm in diameter. Predicted 3-year survival was 79% for aneurysms of 6 cm in diameter, falling to 42% for aneurysms of 8 cm in diameter.

TABLE 16 - Final multivariable Cox regression results for all-cause and aneurysm-related deaths: complete-case analysis

Adapted from Sharples et al. 64 © The Author(s) 2021. Published by Oxford University Press on behalf of the European Society of Cardiology. This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (https://creativecommons.org/licenses/by-nc/4.0/) which permits non-commercial re-use distribution and reproduction in any medium provided the original work is properly cited. The table includes minor additions and formatting changes to the original table.

Variable	All-cause deaths		Aneurysm-related deaths
	HR (95% CI)	z-test p-value	HR (95% CI)	z-test p-value
Female sex	1.79 (1.25 to 2.57)	0.001	2.67 (1.61 to 4.42)	< 0.001
NYHA per class	1.23 (1.00 to 1.52)	0.052
Maximum aneurysm size per cm	1.90 (1.65 to 2.18)	< 0.001	2.19 (1.81 to 2.65)	< 0.001
Age at consent (years)
61–70	2.50 (0.76 to 5.43)	< 0.001	1.30 (0.41 to 4.14)	0.0103
71–80	3.49 (1.26 to 9.66)		1.47 (0.51 to 4.23)
> 80	7.01 (2.50 to 19.62)		3.36 (1.15 to 9.87)

FIGURE 8.

Predicted overall survival (a) and time to composite event of death, ESG or OSR (b) by maximum aneurysm at baseline, using Cox regression, with all other variables set to their average. Adapted from Sharples et al. 64 © The Author(s) 2021. Published by Oxford University Press on behalf of the European Society of Cardiology. This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (https://creativecommons.org/licenses/by-nc/4.0/) which permits non-commercial re-use distribution and reproduction in any medium provided the original work is properly cited. The figure includes minor additions and formatting changes to the original figure.

Adapted from Sharples et al. 64 © The Author(s) 2021. Published by Oxford University Press on behalf of the European Society of Cardiology. This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (https://creativecommons.org/licenses/by-nc/4.0/) which permits non-commercial re-use distribution and reproduction in any medium provided the original work is properly cited. The figure includes minor additions and formatting changes to the original figure.

These predictions will underestimate the death rates if people undergoing procedures are at greater risk than people with similar characteristics who do not undergo procedures. Thus, we also provide predicted time to the composite outcome of death or intervention. This provides a worst-case scenario as it assumes that a patient would die on the day of the procedure if this were not performed.

There was weak evidence that increasing NYHA class was associated with increasing risk of all-cause death. In particular, CM and a range of comorbidity markers were not significant in the final model once age, sex, aneurysm size and NYHA class were included. The relatively small number of deaths meant that these models had limited power to detect small to moderate risk factors for death.

Frailty models showed that, given the small number of events, there was no evidence of variation between hospitals in the death rates, so this was not included in the models presented.

All survival models above used complete-case analysis. The results of repeating the analysis using multiple imputation for missing covariates were almost identical (see Appendix 2), showing that the results were not sensitive to missing data, provided that the assumption of MAR conditional on observed covariates holds. If there is some unknown missing data mechanism that depends on characteristics that were not measured in the ETTAA study, some bias in results could result. We consider this unlikely, given the comprehensive covariate adjustment in the imputation process.

We also refitted the final models with ESG and OSR treated as competing risks. This did not affect the main messages from this analysis, but the effect of baseline maximum aneurysm size was lower for both overall survival (HR 1.58, 95% CI 1.37 to 1.82; p < 0.001) and aneurysm-related survival (HR 1.81, 95% CI 1.51 to 2.17; p < 0.001).

Health-related quality of life over time from recruitment

Descriptive analysis

During the study, 3732 pre-procedure HRQoL questionnaires were returned by 886 patients. After blank forms (n = 256), duplicate entries (n = 11) and incomplete forms (n = 35) were excluded, 3492 (93.6%) questionnaires remained. Overall, 855 of 886 (96.5%) patients completed between one and nine EQ-5D-5L questionnaires. Of these 855 patients, 179 completed a single questionnaire, leaving 676 (79.1%) who contributed to the longitudinal analysis. Figure 9 shows the HRQoL trajectories over time pre procedure for the four treatment groups. These plots show recognised ceiling effects (maximum health) and very wide variation between patients at recruitment. Patterns over time are difficult to unravel from the plots, but CM patients appear to have lower HRQoL at baseline and there is some sign of a general decline in all groups.

FIGURE 9.

The EQ-5D-5L utilities over time, according to management group: (a) WW; (b) CM; (c) ESG; and (d) OSR.

At least two pre-intervention EQ-5D-5L questionnaires were completed by 450 WW, 105 CM, 82 ESG and 64 OSR patients. Mean (SD) utilities at recruitment for these patients were 0.73 (0.23), 0.68 (0.25), 0.77 (0.24) and 0.76 (0.18), respectively. According to Szende et al. ,69 the population average in the UK is 0.785 for people aged 65–74 years and 0.734 for people aged ≥ 75 years, so the ETTAA cohort reported slightly worse HRQoL than the UK population of similar age. Future ESG and OSR patients had similar average HRQoL, despite the 5-year difference in average age. Patients allocated to intervention groups often left the pre-intervention phase before a second assessment had been made; ESG patients were less likely than other patients to complete baseline forms.

Results of modelling

The final model describing HRQoL trajectories in the absence of treatment is shown in the Appendix 7, and a brief summary is provided here.

At time zero, all covariate values are zero if the patient is male, is of average age (70.9 years), is in the WW group, has never smoked, does not receive formal/informal care and is in NYHA class I. For these ‘zero’ patients, average HRQoL at recruitment (time zero) was 0.85 (95% CI 0.82 to 0.88) (Table 17).

TABLE 17 - Model estimates of the mean HRQoL (95% CI) at time of consent to the ETTAA study by final management group and baseline covariates

Male of average age, no formal/informal care, never smoked and NYHA class I.

Variable	WW group	CM group	ESG group	OSR group
All covariates zeroa	0.85 (0.82 to 0.88)	0.83 (0.78 to 0.89)	0.86 (0.81 to 0.92)	0.82 (0.76 to 0.87)
Formal/informal care	0.64 (0.58 to 0.70)	0.71 (0.61 to 0.81)	0.65 (0.51 to 0.79)	0.82 (0.68 to 0.96)
NYHA class
II	0.76 (0.73 to 0.79)	0.71 (0.67 to 0.75)	0.70 (0.65 to 0.75)	0.76 (0.71 to 0.81)
III	0.67 (0.63 to 0.71)	0.58 (0.52 to 0.64)	0.54 (0.46 to 0.62)	0.71 (0.64 to 0.78)
IV	0.58 (0.53 to 0.63)	0.45 (0.35 to 0.55)	0.37 (0.25 to 0.49)	0.66 (0.54 to 0.78)

At recruitment (time zero), across all management groups, there was weak evidence that women had lower HRQoL (–0.029, 95% CI –0.55 to –0.003). The average age at recruitment was just under 71 years and HRQoL at recruitment actually increased slightly with age (0.013 per decade, 95% CI 0.000 to 0.025), possibly as a result of selection policies. In terms of patients not requiring formal/informal care, the management groups had similar average HRQoL at recruitment (all else being equal). There was weak evidence that current smokers had worse HRQoL (–0.047, 95% CI –0.091 to –0.004) at recruitment than ex-smokers or never-smokers in all management groups. Reported requirement for formal/informal care had a very large impact on HRQoL at recruitment in WW, CM and ESG patients, but HRQoL was not adversely affected by the reported need for care in OSR patients (see Table 17). In the WW group, HRQoL at recruitment was lower by –0.089 (95% CI –0.11 to –0.069) for each one-class increase in NYHA classification. The interaction between management group and NYHA class shows that the relationship with increasing NYHA class is even stronger in CM and ESG patients than in WW patients, but slightly less strong in OSR patients (see Table 17).

For the relatively fit patients with zero covariates at baseline, HRQoL did not change significantly over time (estimated decrease –0.010 per year, 95% CI –0.022 to 0.003 per year). However, the interaction between follow-up time and age showed that for two patients who differ in age by 10 years, the older patient has a faster decrease in HRQoL of –0.013 (95% CI –0.019 to –0.007) per year (all else being equal). As a result, the higher HRQoL (baseline, 0.013; change in the first year, –0.013) between age groups increased over time thereafter. Moreover, there was reasonably strong evidence that current smokers had a faster decline in HRQoL than non-smokers and ex-smokers (estimated regression parameter –0.034, 95% CI –0.057 to –0.01; p = 0.004) per year. Figure 10 shows the estimated trajectories for variables that affected rate of decline in HRQoL. This shows that, in this cohort, smoking has a much greater influence on HRQoL than a 10-year increase in age.

FIGURE 10.

Model-estimated HRQoL over time by smoking history and age (dark blue is average age of 70.9 years, light blue is average age + 10 = 80.9 years), with all other covariates set to zero (patient is WW, male, no formal/informal care, NYHA class I).

Significant random effects indicated that HRQoL varied significantly between patients both at recruitment and in the rate of decline over time, in addition to the variation that could be attributed to the variables in the model. There were also some differences between the management groups in how much patients varied around the average at recruitment.

Perhaps of greater interest were the variables that did not significantly affect HRQoL, including aneurysm size and comorbidities such as COPD, coronary heart disease and valvular heart disease, although age and the use of formal/informal care (reflecting frailty) were associated with these comorbidities and possibly acted as surrogates.

Population

All patients who underwent at least one ESG or OSR procedure were included in this analysis from the date of hospital admission for the index procedure to death, withdrawal or the end of the ETTAA study. Consistent with the observational study protocol, patients were analysed according to the treatment received, irrespective of management plans at recruitment.

Outcomes

The primary outcome was survival from the date of the index procedure to either death or censoring date. Secondary outcomes were clinical complications, death within 30 days of procedure, reinterventions, readmissions to hospital for aneurysm or cardiac causes, length of stay and HRQoL from procedure to end of follow-up, as defined in Chapter 3.

Statistical analysis

Baseline predictors

Baseline variables at consent for the two intervention groups were provided in Tables 8–10. These variables recorded before but closest to the procedure are summarised in the Appendix 8, although changes from baseline were small.

There were important differences between the two groups at the time of their procedures. Excluding missing covariates, compared with OSR patients, ESG patients were older (mean age difference 7.1 years, 95% CI 4.7 to 9.5 years; p < 0.0001), were smaller (mean height difference –3.7 cm, 95% CI –6.3 to –1.2 cm; p = 0.0041 and mean weight difference –5.3 kg, 95% CI –9.2 to –1.4 kg; p = 0.0075) and were more likely to be current or past smokers (75.8% vs 66.4%; p = 0.080). The differences in height and weight were partially explained by the higher proportion of CTD patients in the OSR group, although the differences remained significant when these patients were removed (mean height difference –2.7 cm, 95% CI –5.2 to –0.1 cm; p = 0.0381 and mean weight difference –5.0 kg, 95% CI –9.1 to –0.9 kg; p = 0.0172). ESG patients were more likely to have valve disease (89.9% vs. 71.6%; p < 0.0001), COPD (21.3% vs. 13.3%; p = 0.087) and stage III/IV NYHA (22.3% vs. 16.0%; p = 0.217). Patients with connective tissue disorders almost invariably underwent OSR (14.8% vs. 1.3%; p < 0.0001). A very high proportion of patients (90.7% ESG, 85.2% OSR) had hypertension; there was no difference between groups in the numbers who were prescribed antihypertensive medications (88.7% vs. 85.2%; p = 0.486). ESG patients were more likely to report use of statins (69.3% vs. 42.2%; p < 0.0001). Serum creatinine and haemoglobin measurements were missing for approximately 19% of cases in each group. For those with complete data ESG patients had similar renal function (mean serum creatinine difference –7.1 µmol/l, 95% CI –16.0 to 1.8 µmol/l; p = 0.1191) but lower levels of haemoglobin (mean haemoglobin difference –6.8 g/l, 95% CI –11.2 to –2.4 g/l; p = 0.0026) than OSR patients.

Despite differences in age and other covariates between the groups, mean (SD) HRQoL utilities before the procedures were the same, 0.73 (0.24) for ESG and 0.73 (0.26) for OSR.

Description of outcomes

Survival, hospital readmissions and aneurysm-related events

Survival post procedure: descriptive analysis

The primary outcome was overall survival after the procedure. During follow-up, 40 ESG and 36 OSR patients died, of whom 17 and 25, respectively, died from aneurysm-related causes; these included deaths within 30 days (Table 19). The 1-year overall death rate was 17.5% (95% CI 12.2% to 24.8%) after ESG and 20.7% (95% CI 14.6% to 28.9%) after OSR; the 3-year death rate was 34.7% (95% CI 26.1% to 45.1%) after ESG and 31.9% (95% CI 23.7% to 42.2%) after OSR (Figure 11a). For aneurysm-related deaths, the 1-year rate was 10.2% (95% CI 6.2% to 16.6%) after ESG and 12.3% (95% CI 10.9% to 24.1%) after OSR; the 3-year rate was 14.3% (95% CI 8.8% to 22.7%) after ESG and 20.6% (95% CI 14.3% to 29.3%) after OSR (Figure 11b). Despite the slightly higher aneurysm-related death rate for OSR patients in the first 30 days, there was no significant difference between the groups in time to death overall (log-rank p = 0.9918) or in aneurysm-related death (log-rank p = 0.1107).

TABLE 19 - Number of patients with re-admissions and clinical events after the index procedure

	Patient subgroup
	ESG (N = 150)	OSR (N = 135)
Total time at risk (years)	265.2	253.4
Deaths overall (rate/year)	40 (0.15)	36 (0.14)
Aneurysm-related deaths (rate/year)	17 (0.06)	25 (0.10)
All readmissions, n (rate/year)	111 (0.42)	87 (0.34)
People with at least one readmission, n (%)	61 (40.7)	43 (31.9)
Readmissions, definitely/probably aneurysm related, n (rate/year)	40 (0.15)	23 (0.09)
Patients readmitted, definitely/probably aneurysm related, n (%)	26 (17.3)	17 (12.6)
Non-fatal ruptured aneurysms	0	0
Non-fatal dissected aneurysms	3	1
Non-fatal neurological events	4	0

FIGURE 11.

Kaplan–Meier cumulative incidence curves for post-intervention death from any cause and aneurysm-related deaths by ESG and OSR groups. (a) Kaplan–Meier deaths from any cause; and (b) Kaplan–Meier aneurysm-related deaths.

Readmissions after discharge from the index procedure

Table 19 shows the readmissions after discharge from the index procedure. During a total of 265.2 years of follow-up, ESG patients were readmitted 111 times, a rate of 0.42 (0.35, 0.49) per patient-year (i.e. on average 42% of patients admitted to hospital per year of follow-up). This was higher than the rate for OSR patients (0.34, 95% CI 0.27 to 0.41) per patient-year. The relative hospital readmission rate (adjusted for time at risk) was 0.70 (95% CI 0.44 to 1.11; p = 0.133); further adjustment for age and sex increased the relative rate of readmission slightly, to 0.75 (95% CI 0.46 to 1.21; p = 0.237). Furthermore, 40.7% of ESG patients were readmitted during follow-up compared with 31.9% of OSR patients (p = 0.1398). A similar pattern was observed for aneurysm-related readmissions. No ruptures and only one dissection were reported as the reason for aneurysm-related readmissions.

Figure 12 shows the pattern of readmissions over time after procedure for the two groups. In the first 3 months, ESG patients were more likely to be readmitted, and thereafter the readmission rates were similar.

FIGURE 12.

Readmissions to hospital (per patient-year at risk) by group and by time after the index procedure.

Once readmitted, ESG patients spent a median of 6 days (quartiles 2 to 12) in hospital, compared with 5.5 days (quartiles 2 to 10) for OSR patients (Mann–Whitney U-test p = 0.8236). However, ESG patients were more likely to spend some of the inpatient stay in ICU (25 vs. 7; p = 0.0064).

Table 19 shows that 12 ESG patients underwent a second ESG during the study period at a median of 93.5 days after the index procedure (range 18–1076 days). Similarly, four OSR patients had a second OSR at 58, 149, 217 and 371 days post index procedure. Twenty-one OSR patients had a subsequent ESG; one was in the same admission, and the remaining 20 were a median of 173 days (range 11–1100 days) after the index procedure. Three OSR patients had ESG as a third procedure at 208, 908 and 1399 days after the index procedure. OSR patients had more complications during readmissions. All 40 reinterventions were reported as planned as part of a staged procedure. Complications after second and third procedures are summarised in Appendix 5.

Post-procedure health-related quality of life

Descriptive analysis

For this analysis the preoperative HRQoL assessment nearest to the date of the procedure was included as a potential predictor of postoperative HRQoL. Forty-three patients had no postoperative assessment of HRQoL; the remaining 242 patients completed between 1 and 12 postoperative questionnaires, resulting in a total of 1082 assessments.

Utility measurement trajectories for individual patients are plotted in Figure 13. Again, the ceiling effect at utility = 1 (maximum health) is observed, and there is evidence of wide variation in HRQoL at time zero. Patterns over time were not clear from the trajectories, but initial analysis suggested that both linear and quadratic terms for change over time should be considered for inclusion and that these may differ between the sexes. A marker variable of early postoperative assessment (first 6 weeks) was considered, as was an interaction between group and early postoperative assessment. Again, there was evidence that HRQoL at procedure was more variable in the ESG group.

FIGURE 13.

Post-procedure EQ-5D-5L utilities over time by intervention group. (a) ESG; and (b) OSR.

Health-related quality-of-life modelling

The final model for postoperative HRQoL is provided in full in Appendix 10. Briefly, for a male, non-smoking ESG patient, in NYHA class I, with average preoperative utility of 0.73, the average postoperative HRQoL immediately post ESG was estimated to be 0.785 (95% CI 0.725 to 0.844). This decreased very slightly by –0.001 (95% CI –0.012 to 0.013) per year, with little evidence of acceleration or deceleration. There was a slight non-significant dip in HRQoL of –0.017 (95% CI –0.062 to 0.027) in the first 6 weeks. HRQoL between patients showed significantly more variation in the ESG group than in the OSR group. These results did not change substantially if the analysis excluded second and third stages of a staged procedure.

For OSR patients, there was a substantial, significant decrease in HRQoL in the first 6 weeks after the procedure of –0.160 (95% CI –0.199 to –0.121; p < 0.001). Otherwise, the difference between the two procedures during follow-up was not significant (all else being equal). Although female and male OSR patients had the same decrease in HRQoL in the first 6 weeks, the pattern over time was different; women had a slight increase in HRQoL over time, whereas men did not change after the first 6 weeks (Figure 14). Finally, in common with pre-procedure analysis, current smokers and those in higher NYHA classes had significantly lower HRQoL throughout, with decrements of –0.095 (95% CI –0.171 to –0.020) for current smokers and –0.034 per NYHA class (95% CI –0.066 to –0.003).

FIGURE 14.

Average estimated post-procedure EQ-5D-5L utility over time by sex and intervention group.

Population

All patients who underwent ESG or OSR procedures as defined in Chapter 5 were eligible. Patients were excluded if they had a zero or close to zero probability of having either of the procedures. Initially, patients were excluded based on variables that were found to be non-overlapping in the exploratory analysis. All remaining ESG patients were then assessed by a cardiac surgeon; anyone aged > 85 years, with BMI < 20 kg/m² or > 35 kg/m², with impaired mobility, assessed by the EQ-5D-5L item response severe difficulty walking/self-care, or of NYHA class IV was considered unfit to receive OSR and was excluded. CT/MRI scans for all except seven OSR patients were re-assessed by a vascular surgeon and patients with aneurysms with aortic morphology that could not be managed with ESG were excluded. Seven scans could not be retrieved and analysed by the end of follow-up and these patients were considered suitable for both procedures on the basis of the operation and clinical history CRFs and were included. Finally, because we were interested in the question of whether an open surgical or an endovascular procedure was the better approach for a new patient, any patients for whom the index procedure was the second or third part of a staged procedure were excluded.

Outcomes

Owing to the reduced sample size, outcomes were restricted to all-cause and aneurysm-related mortality, hospital readmissions and HRQoL, as defined in previous chapters.

Statistical analysis

Initially, the final statistical models developed in Chapter 5 were refitted for the comparison cohort only. Then, to reduce any residual bias caused by uncontrolled confounders, we completed a propensity score analysis based on methods described in Leyrat et al. 78 and Mitra and Reiter. 79 Propensity scores were calculated using binary logistic regression on the probability of OSR, including multiple predictors. 80 Theoretically, there is no limit on the number of predictors than can be used to predict propensity, but, owing to the small sample, we restricted the model to all variables related to the main outcomes (mortality and HRQoL) as well as major associates of treatment. The ability of a propensity score to balance the confounding variables between treatment groups was assessed and the model was refined by adding variables to reduce imbalance. 81,82 The final model included age at procedure, sex, height, need for formal/informal care (as a marker of frailty), COPD, NYHA class, diabetes, hypertension, smoking history, maximum aneurysm size, maximum aneurysm location and preoperative time in the ETTAA study. To assess the sensitivity of the results to different methods, we refitted the main outcome analyses by:

1. matching ESG and OSR participants in a 1 : 1 ratio using the nearest neighbour (with replacement) and excluding any patients without a match

2. inverse probability of treatment weighting (IPTW), in which the propensity score was used to weight individual responses as follows: w_i = z_i/ps_i + (1 – z_i)/(1 – ps_i), where z_i denotes treatment (1 = OSR, 0 = ESG) for individual i

3. IPTW excluding patients with extreme weights (w_i > 5) that conferred a large influence on results

4. including the propensity score in regression equations as a covariate. 83

For matching, we used a calliper width of 0.2 SDs of the logit(propensity score).

The analysis was further complicated by incomplete baseline measurements. Following Leyrat et al. 78 our strategy for accommodation of partially complete baseline measurements was as follows:

• Create 30 complete data sets using multiple imputation.

• Calculate the propensity score for each complete data set.

• For predictive mean matching, identify the set of matched patients over all imputed data sets, estimate treatment effects on these matched pairs and combine using Rubin’s rules. 62,63

• For IPTW or regression adjustment, estimate treatment effects for each imputed data set and combine using Rubin’s rules.

Variables in the imputation models were the same as those included in the propensity score, as well as serum creatinine. We used 30 imputation data sets because of the relatively large number of patients with missing creatinine.

Definition of no-contraindication cohort

Figure 15 shows how the final no-contraindication cohort was derived. Almost all (20/22) CTD patients and all patients who had either surgery in the ascending aorta or concomitant cardiac surgery had OSR. These violated the positivity assumption required for both comparability and propensity score methods (i.e. they were essentially unsuitable for ESG) and were excluded from the analysis before clinician review.

FIGURE 15.

Cohort of no-contraindication ESG and OSR patients.

Overall, 35 ESG patients were excluded from the comparison cohort because they had CTDs (n = 2), or the index procedure was stage 2 (n = 8) or 3 (n = 2) of a staged procedure, or the cardiac surgeon considered them unsuitable for OSR (aged > 85 years, n = 1; BMI < 20 kg/m² or > 35 kg/m², n = 11; severe mobility problems, n = 9; NYHA class IV, n = 2). The remaining 115 (76.7%) ESG patients were considered suitable for OSR in principle.

Of the 135 OSR patients recruited, 100 were excluded from the comparison cohort because they were not suitable for ESG due to CTDs (n = 20), ascending aorta surgery (n = 57) or concomitant cardiac surgery (n = 7), or based on morphology (n = 16). The remaining 35 (25.9%) were included in the no-contraindication cohort.

Despite the exclusion of patients with contraindications to one or other of the procedures, important differences in baseline variables between the groups remained. The difference between the groups in age at procedure increased to 10.5 years (95% CI 6.9 to 14.1 years), with the mean slightly reduced to 62.6 years for OSR patients and slightly increased to 73.1 years for ESG patients. ESG patients were smaller and were more likely to have diabetes, report a need for informal care and be treated with statins. However, ESG patients were treated quicker, were less likely to have raised serum creatinine and had slightly smaller maximum aneurysm size on average (Table 22).

Survival and hospital readmissions

During the follow-up of 150 no-contraindication patients, 31 ESG and 8 OSR patients died, of whom 14 and 5, respectively, died from aneurysm-related causes. Eight (7.0%) ESG and three (8.6%) OSR patients died within 30 days. In exploratory analysis (Figure 16), there were no differences between the groups in time to death overall (log-rank p = 0.6608) or aneurysm-related death (log-rank p = 0.7533).

FIGURE 16.

Kaplan–Meier cumulative incidence curves for post-intervention deaths from all-cause and aneurysm-related deaths by ESG and OSR groups (no-contraindication patients only). (a) Kaplan–Meier deaths from any-cause; and (b) Kaplan–Meier aneurysm-related deaths.

Post-procedure health-related quality of life

Planned analysis

Originally, and according to protocol, the aim of this analysis was to estimate the cost–utility of two alternative surgical methods of treating patients with arch/descending thoracic aortic aneurysm. In the economic evaluation, the costs and effects associated with ESG were to be compared with those associated with OSR to define the optimal management strategies for patients who could be eligible for both treatments and for whom, therefore, there was a choice.

The economic analysis aimed to compare the incremental cost per QALY gained with different threshold values for society’s willingness to pay for a QALY, including those commonly adopted by NICE. 85 Both a ‘within’-study patient-level analysis and a state-transition model to extrapolate findings over patients’ lifetimes were proposed. To facilitate this analysis, data were collected from individual study participants, participating centres and external sources in order to estimate costs. QALYs were to be estimated from serial responses to the EQ-5D-5L.

Resource utilisation and the associated costs were collected from an NHS and PSS perspective and concentrated on the micro-costing of the surgical procedures themselves in secondary care. The utilisation of any subsequent secondary, primary and personal social services during the follow-up period was also recorded. Data on resource use were captured on bespoke CRFs [see Report Supplementary Material 5 for CRFs (use of secondary care services, and incidence and frequency of cost-generating events, e.g. hospital readmissions)]. Further CRFs were completed at 3, 6, 12, 18, 24, 36, 48 and 60 months to capture the use of primary care and PSS. The unit costs of resources were obtained from standard sources such as NHS reference costs, HRG tariffs, manufacturers/suppliers and from the centres themselves. For each participant, measures of resource use were to be combined with unit costs to estimate the total cost for that participant.

In terms of QALY estimation, EQ-5D-5L scores were collected prospectively at baseline and at 1, 3, 6, 12, 18 and 24 months, and then annually until the study ended. The responses for each participant were converted into health state utilities using UK population tariffs via crosswalk mapping58 and used to estimate QALYs using the area-under-the-curve approach. 86

For the within-trial analysis, bootstrapping methods were to be used to estimate the imprecision around estimates of incremental costs, QALYs and incremental cost per QALY. For the model-based analysis, a probabilistic sensitivity analysis was to be conducted. The results of both the within-trial and the model-based analyses were to be presented as plots of incremental cost and QALY and cost-effectiveness acceptability curves.

Revised analysis

To ensure a fair comparison between OSR and ESG as a primary procedure, the cohort of patients who had a surgical procedure (presented in Chapter 6) were assessed by clinical experts to determine whether they would have been eligible to receive both procedures (see Figure 15). Chapter 6 describes reasons that patients who received one procedure would have been ineligible for the other and the numbers of participants excluded for that reason. In brief, the reasons why OSR patients were ineligible for ESG included aneurysm repair extending into the ascending aorta, concomitant cardiac procedures and unsuitable aortic morphology. The reasons why ESG patients were ineligible for OSR included BMI < 20 kg/m² or > 35 kg/m²; NYHA class IV dyspnoea; or age ≥ 85 years. In addition, 10 ESG patients had primary procedures that were the second or third stage of a planned staged procedure. These patients were excluded as they would have biased the analysis of costs and QALYs. Overall, 115 ESG patients and 35 OSR patients were judged potentially eligible for both procedures.

As there were only 35 no-contraindication participants in the OSR group, the results of a comparative cost–utility analysis were judged to be imprecise and potentially misleading. 87 Therefore, the aims of the economic evaluation were recast to:

• Provide a non-comparative, descriptive ‘within-study’ analysis of the available cost and QALY data for participants eligible to receive both OSR and ESG as the first (index) procedure, without formally comparing the two procedures statistically. Specifically, the following were estimated –

  • health-care costs of primary surgical procedures (average and median costs per surgical group)

  • health-care and PSS costs that were definitely or probably related to the index procedure/the aneurysm at all follow-ups including average total costs at 12 months by surgical group

  • utilities at all follow-up points and average total QALYs at 12-month follow-up by surgical group.

• Estimate a regression model to explore the predictors of NHS costs for both OSR and ESG that may aid future economic evaluation modelling studies.

• Estimate the costs and benefits of obtaining further information (about overall cost-effectiveness and costs and QALY individually) based on data obtained from the ETTAA study via value of information (VoI) analyses. 88

Given the potentially misleading results from conducting a comparative cost-effectiveness analysis, none was attempted. 85 Instead, we sought to utilise the limited information from the ETTAA study and conduct a VoI analysis in order to inform proposed future research into the relative cost-effectiveness of OSR compared with ESG and to identify potential sources of uncertainty. However, the results of this analysis should be seen in the context of a lack of a fair comparison in the ETTAA population, and are presented to aid any future research.

All health economic analysis has been designed and conducted to best practice conforming to the Consolidated Health Economic Evaluation Reporting Standards (CHEERS). 89

Health economic data collection

Valuation of NHS and informal caregiving resource use

Quality of life

The EQ-5D-5L questionnaire was administered at baseline and at 1, 3, 6, 12, 24, 36 and 48 months following a procedure. If a participant died, then they were assigned a zero score from the date of death. For each participant, utilities were collected across three distinct pathways:

1. Pre-procedure follow-ups after consent at 3, 6, 12, 18, 24, 36, 48 months.

2. After the index procedure, follow-up was ‘reset’ and occurred at 1, 3, 6, 12, 18, 24, 36 and 48 months after the index procedure.

3. After each additional procedure (second or third procedure), follow-up was ‘reset’ and occurred at 1, 3, 6, 12, 18, 24, 36 and 48 months after the additional procedure.

In some cases, however, the ‘resets’ did not occur as planned, resulting in misaligned follow-ups. To aid analysis and to use as many available data as possible, the time of all follow-ups was calculated from the date of primary procedure; that is, date of the primary procedure was taken to be time zero. Time of follow-up after the primary (index) procedure was then categorised and aligned to nominal time points 1, 3, 6, 12, 18, 24, 36 and 48 months after the index procedure. For example, 1-month follow-ups were those that actually occurred between 0 and 60 days after the index procedure. In addition, as in the analysis of clinical outcomes in Chapters 5 and 6, baseline was taken as the most recent utility measure prior to the index procedure. Appendix 11 provides the categories of the time post procedure assigned to the different nominal follow-up time points.

The mean (SD) and median (IQR) of the EQ-5D-5L utility scores were calculated at nominal measurement times (i.e. baseline and 1, 3, 6, 12, 24, 36 and 48 months) using the measurements available at those time points. Owing to the small sample size, especially in the OSR group, multiple imputation of missing outcomes was not undertaken.

The mean (SD) and median (IQR) QALYs at 12 months were estimated for each surgical group for those patients who had a follow-up at baseline, 1 month, 12 months and any follow-up point in between, including a score of zero if they had died during follow-up. QALYs for each participant were estimated as the area under the curve, constructed by interpolating between utilities at nominal measurement times. There was no comparison between total QALYs for each surgical procedure because of the limited number of data available. The analysis of the EQ-5D-5L and QALY data was conducted using R, version 6.2 (The R Foundation for Statistical Computing, Vienna, Austria).

Missing cost and quality-of-life data

The data available for comparative purposes were very limited, and there were considerable missing data for study participants (Table 26), particularly after 12 months’ follow-up. Every patient who provided utility data at each follow-up also provided a follow-up CRF for resource use and hence for costs. There was only a very small number of patients (n = 11) who provided resource use and not utility data for some of the follow-up points. Up to 12 months, there was a larger proportion of missing data in the OSR cohort relative to the ESG cohort for both resource use and quality-of-life data. An examination of the data highlighted that patients who were in hospital at 1 month in the OSR group had subsequent missing data points for both resource use and quality of life. The reasons for this are unknown but could include not being sent a questionnaire, being in hospital or being too ill to complete a questionnaire. As expected, there are more censored data at later follow-up points.

Given, the nature and extent of missing data, no imputation was attempted. The exceptions to this were that if the missing data were related to standard resources that are normally used during the treatment pathway, it was assumed that these resources were used and, therefore, costs were added. Furthermore, where length of theatre time and length of stay were missing, averages of similar events were used for each procedure type.

As specified previously, to maximise the data used to estimate EQ-5D-5L QALYs utilities at baseline, 1, 3, 6 or 12 months were utilised.

Cost analysis at all time points

The total average cost per recipient by each area of resource use for each of the three stages of the study is presented in Appendix 11. This shows that costs of theatre time and the corresponding staff time are higher on average for the OSR procedure, as are the costs of blood products. Stent costs are much higher on average for the ESG group relative to the average cost of grafts in the OSR group. The mean (SD) total costs of the primary OSR procedure was £17,239 (£8043) and for the primary ESG was £26,536 (£9877). The cost of the ESG procedure was largely driven by the stent costs, which accounted for 79% of the average total cost of an ESG procedure.

Appendix 11 shows the total average costs per patient postoperatively until discharge for each surgical procedure broken down into categories of resource use. With the exception of return-to-theatre costs, average costs were higher in all resource use categories in the OSR group relative to ESG. The largest cost driver for OSR procedures was ICU resource use, which accounted for 71%. Furthermore, a larger percentage of patients in the OSR group were transferred to other hospital settings for further treatment (e.g. rehabilitation).

The average total costs after discharge by follow-up period and resource use category are shown in Appendix 11. The average total NHS costs for the first 12 months following the primary procedure were higher in the ESG group. These were driven by the extra costs of patient admissions to hospital, including the costs of additional procedures (85% of total NHS costs).

There was no discernible pattern in the average total costs of formal and informal care. These costs may also be inaccurate.

The average total costs for the primary procedure, post procedure until discharge and follow-up at 12 months are presented in Table 27 for the participants who were suitable for either procedure. Particularly for the OSR group, mean costs are highly skewed to the right for most categories because of a small number of participants who incurred very high costs.

Analysis of quality of life

Owing to the very small numbers of patients reporting utility values at all time points, particularly in the OSR group, utilities [means (SD) and medians (IQR)] are reported for every patient who completed an EQ-5D-5L questionnaire at each time point, or were known to have died by a given time (Table 28). Patient numbers include all those who had completed a utility CRF at each time point, including a zero value for those who died. This excludes those who were censored for any reason. Utility values were lower at all follow-up points for OSR than for ESG. At 1 month post surgery, HRQoL dropped from baseline, with a fall of > 50% in HRQoL in the OSR group. HRQoL also reduced at 1 month in the ESG group, although this drop was not as large (15%). HRQoL in both groups increased until 6 months post procedure in the OSR group and until 12 months post procedure in the ESG group as people recovered from their primary procedure. Average utilities in both groups for the remaining follow-ups then started to decrease. This may be because of the impact of complications, readmissions and deaths (which are assigned a zero score). It must be noted that the sample sizes were small and so estimates were imprecise and potentially unreliable, particularly after the first 12 months. Table 28 shows that the estimated mean QALYs at 12 months were larger in the ESG group than in the OSR group, although this was based on very small (and possibly not representative) numbers of patients who had complete data up to 12 months.

Introduction and rationale

The aim of this part of the economic analysis was to identify the key determinants of costs to the NHS of an OSR procedure and, separately, of an ESG procedure up until hospital discharge. Therefore, a separate regression model was used for each surgical procedure. Regression models were used where the dependent variable was the NHS costs for an OSR or an ESG procedure up until discharge. Potential determinants of NHS costs up to discharge were included as independent (explanatory) variables. These analyses are shown to illustrate more fully what the main predictors of these costs were and also to provide a resource for future economic evaluation modelling.

Methods

Results

Current smoking status was the only variable associated with costs in the ESG model. The model estimated that current smokers have an increased cost on average of £10,447 (95% CI £2449 to £20,495) compared with non-smokers/past smokers. In the OSR cost model, there was no evidence of significant predictors. See Appendix 11 for model coefficients.

Methods

To inform the VoI analysis, we used a subset of patients for whom we had near-complete data about resources and quality of life from baseline to 12 months. Details of this patient subset are shown in Appendix 11 and the data definitions are the same as used throughout this chapter. We also developed a simple model to extrapolate the results to 36 months. This model used annual follow-up costs (ESG, £245; OSR, £337), 12–24 month unadjusted common mortality rate (5%), reintervention rates (yearly 3% OSR and 15% ESG) and 12-month utilities. Utility at 12 months was carried forward into years 2 and 3, and the mean cost of an aneurysm-related admission was estimated at £36,005.

Although we considered that cost–utility data would be of limited use and potentially misleading due to the limited data available for a comparative analysis, the principles of economic evaluation can be used to estimate the expected value of perfect information (EVPI) and the expected value of partial perfect information (EVPPI). Although these estimates suffer from the same limitations as the incremental cost–utility data on which they are based, they do show (for the small sample) the breakdown between the uncertainty associated with health outcomes and resource utilisations. Therefore, they are reported with the caveat that they are based on a small sample, but they may inform future research. EVPI and EVPPI were estimated using a bootstrapping technique whereby subpopulations of study participants were sampled with replacement from the overall study population. It is emphasised that the VoI analysis has been presented using the limited ETTAA data set purely to inform decision-making regarding future research and should not be interpreted as a definitive comparative analysis.

The EVPI for each patient is the net benefit of the decision made with perfect information about the uncertain parameters and the decision made based on existing evidence. For the study population, it is the net benefit of the overall optimal treatment choice across all bootstrap iterations less the net benefit lost when the overall optimal choice was not optimal in a particular iteration. The EVPI at the population level, the expected benefit accruing to all future patients, was calculated by multiplying the individual EVPI by the expected future population where there would be a choice about which of the two interventions to use. The annual population in the UK who might benefit from either treatment was assumed to be 100 (based on a plausible limit of the number of procedures the equipment could be used for in 1 year in a given centre), but sensitivity analyses were undertaken with eligible populations of 50 and 200. Initial analysis assumed that the consequences of intervention would last for 12 months. A second analysis assumed that the consequences would last for 3 years, with these second- and third-year impacts being determined using the data derived in the descriptive analysis presented above but incorporated into a simple Markov model predicting death and reintervention. The EVPPI was calculated to show the value of removing all uncertainty around two groups of parameters, namely costs and QALYs.

All VoI analyses were conducted in Stata 15.1 and used a threshold value of £20,000 per QALY. 85 The analyses used generalised linear models to estimate bootstrapped total costs and QALYs at the time points noted above. The bootstrapping took repeated random sampling with replacement of individuals from the data set and estimating a cost-effectiveness ratio for each sample. These samples capture the uncertainty in the overall estimate of cost-effectiveness and inform the VoI, each representing a possible decision about the marginal cost-effectiveness of the two surgical options.

The perspective of the analysis was that of the NHS (i.e. PSS and patient costs were excluded from this analysis). As the total costs and QALYs estimated extended beyond 1 year, each was discounted at 3.5%. 85 To estimate effect, each bootstrap iteration drew a set of matched patients for the analysis. Using multiple imputation, 45 sets of propensity-weighted patients were created. During each bootstrap iteration of the VoI analysis, the bootstrap sample was merged with a random (1 out of 45) set of propensity scores as estimated in Chapter 6 to estimate marginal costs and QALYs, and hence net benefits and the EVPI and EVPPI.

Results of the value-of-information analyses

For the VoI analyses, the population deemed eligible for either procedure and with complete resource use and EQ-5D-5L up to 12 months was used (OSR, n = 18; ESG, n = 65); 800 bootstraps were used for the estimation of EVPI and EVPPI at the 12-month follow-up, and a further 800 bootstraps were used for estimates of the EVPI and EVPPI at the 36-month follow-up.

The results for the propensity-scored primary analyses are shown in Table 29 and the accompanying 12-month and 36-month cost-effectiveness planes are shown in Figures 17 and 18. Each point on the cost-effectiveness plane shows the average difference in costs and effects from each bootstrap iteration. Points in the north-east and north-west represent circumstances in which OSR is more expensive than ESG, and points in the north-east and south-east quadrants indicate circumstances in which OSR has higher HRQoL compared with ESG.

FIGURE 17.

Cost-effectiveness plane for the 12-month VoI model. Cost-effectiveness plane OSR – ESG (generalised linear model). Incremental cost-effectiveness ratio: ESG dominates OSR.

FIGURE 18.

Cost-effectiveness plane for the 36-month VoI model. Cost-effectiveness plane OSR – ESG (generalised linear model). Incremental cost-effectiveness ratio: –£11,518.

Over 12 months, ESG dominates OSR in this exploratory analysis. It is, on average, associated with higher costs and statistically significantly greater QALYs. The EVPI of £84,200 reflects the distribution of points on the cost-effectiveness plane and that few exist outside the north-west and north-east quadrants. EVPPI confirms that most of the uncertainty derives from uncertainty about costs rather than about QALYs and is, therefore, less sensitive to willingness-to-pay thresholds.

Over 3 years, ESG no longer dominates OSR. OSR is on average less costly but less effective. The incremental cost-effectiveness ratio of £11,518 is the incremental cost per QALY gained for the comparison of the more effective but more costly ESG compared with the less costly and less effective OSR. In this analysis, the VoI increases, with EVPPI estimates for cost parameters increasing the most. Extending the time frame from 12 months to 10 years sees the total EVPI increase to £1,628,800.

Over 12 months, there is considerably more uncertainty around the net benefit associated with resource consequences than with quality of life. However, this relative gap closes in our simple 36-month model as reintervention costs of ESG offset the higher initial costs of OSR.

Without procedural intervention for chronic thoracic aortic aneurysm, what is the risk of aneurysm growth, dissection, rupture, permanent neurological injury or death, and how does health-related quality of life change over time?

Chronic thoracic aortic aneurysm patients were at high risk of death, had a range of comorbidities and had somewhat impaired HRQoL measured with EQ-5D-5L utility at the time of the intervention. Rate of aneurysm growth varied by location but was generally slow in this cohort of patients.

After diagnosis with arch/descending CTAA but before intervention (if planned), the death rate was high (8.6% per patient-year) compared with the general population with similar age distribution in England; for example, in 2015–17, the 1-year probability of death in England among people aged 71 years was 2.1% for men and 1.4% for women. 70 Thus, death rates for ETTAA patients were over four times the death rate for the general population of similar age. Approximately half of the deaths prior to a procedure in the ETTAA study were from aneurysm-related causes. The hazards of death in the absence of surgical intervention were significantly associated with aneurysm size, older age at presentation, female sex and (to a lesser extent) higher NYHA class. Other than aneurysm size and patient sex, these predictors of risk may simply be surrogates for comorbidity. Predictions from fitted models suggested that 3-year survival without treatment is > 90% for patients with small aneurysms (4–4.5 cm in diameter), falling to 88% for aneurysms of 5 cm, 79% for aneurysms of 6 cm, 63% for aneurysms of 7 cm and 42% for aneurysms of 8 cm. Together with the increased risk of poor post-intervention outcomes for longer waiting times, this indicates that intervention should be discussed once aneurysms exceed 6 cm in diameter, provided that operative risk is considered low/moderate.

Consistent with their older age, CM patients and those who went on to have ESG had more hospital admissions, including aneurysm-related admissions. Patients who were managed conservatively were judged to be sicker and less likely to benefit from intervention than other patients, so the higher admission and death rates in this group testify to the judgement of the clinicians. Non-fatal acute ruptures, dissections and neurological injuries were uncommon. Combining fatal and non-fatal cases, there were 69 ruptures or dissections, or both combined, in a combined total of 1489 years of patient follow-up pre intervention, a rate of 4.6% per patient-year at risk. This is broadly consistent with the benchmark study from Yale University, IRAD, GERAADA and reports of the Australasian experience. 19,21–23

Health-related quality of life as measured using the EQ-5D-5L varied substantially at presentation, but, on average, the mean was slightly lower than UK age- and sex-matched population estimates. Heterogeneity at presentation could be partially explained by age, sex, smoking and NYHA class, but there was substantial unexplained variation. Moreover, the decline in HRQoL was faster for older patients and current smokers, but there was also unexplained variation between patients in the rate of change over time. Patients destined for ESG had lower HRQoL at presentation, all other factors being equal, although a few patients waited more than 6 months before intervention, at least partly due to delays in production of custom-made stents.

What factors affect aneurysm growth pre intervention?

The mean aneurysm size at presentation was partly related to known covariates such as age, smoking, CTD and COPD, with only location of the aneurysm related to both baseline diameter and growth rate.

The mean aneurysm size at study entry varied widely as a result of patients being identified at different stages in the evolution of their disease, and there was significant variation that was not explained by the variables measured in the ETTAA study. Almost all patients had documented hypertension, so the impact of hypertension on aneurysm growth (or any other outcome) could not be evaluated; hypertension is, however, an established causal factor in the development and outcome of aneurysms, and treatment should be optimised. 74 Given their older average age, poorer LV function, smoking history, COPD and prevalence of DTA aneurysms, patients who went on to have ESG are likely to have had faster-growing aneurysms before the procedure.

Differences in aortic pathology and morphology in different sections of the aorta meant that aneurysms in the DTA were significantly larger and more variable at presentation and grew faster than those in the arch or thoracoabdominal aorta; no growth was found in the ascending aorta, which largely results from previous interventions in this segment. Mode of surveillance (MRI or CT scans) was the only other variable associated with growth over time, which is likely to be a result of the selection for MRI of patients with slower-growing aneurysms and referral for CT more likely if growth appears to be accelerating. Slower-growing aneurysms are anticipated to require longer-term follow-up and, therefore, clinicians aim to reduce radiation exposure by choosing MRI scans for the patient. However, when an aneurysm is approaching an indication for surgery, CT images are superior for interventional planning. With these exceptions, there was no evidence of measured or unmeasured covariates affecting growth of the aneurysm, which partly results from the short-term follow-up in most patients and possible earlier intervention in patients who have signs of faster growth. Other studies suggested that growth is related to various risk factors including age, smoking, CTD, COPD and hypertension (see Chapter 1). The ETTAA study included surveillance imaging from diagnosis when growth may have been slower, so that ability to detect factors associated with growth was limited. Studies of growth in the infrarenal aorta have also failed to consistently find useful predictors of increased growth rate. 99

Detailed studies of the aneurysm/vasculature (patent or partially patent false lumen, larger false lumen diameter/saccule formation within the false lumen, number and location of tears around the arch, peak wall stress) could not be undertaken in this observational study and so we could not assess their relationship to size or growth. Moreover, treatment by indication bias may have hindered our ability to assess whether or not medical treatments slowed growth. It is important to note that patients leave this analysis when their aneurysms are fast-growing or become large enough for intervention, or when they die, so the pattern of growth outside this range is not known and extrapolation is not justified.

If a patient has endovascular stent grafting or open surgical repair, what is the risk of dissection, rupture, permanent neurological injury or death, and how does health-related quality of life change over time?

Differences in patient populations meant that ESG and OSR outcomes could not be directly compared. ESG patients had lower risk of death within 30 days (6.7% vs. 11.1%) but subsequent survival was comparable; incidences of non-fatal dissection, rupture and neurological injury was rare; and, after an initial dip after OSR, survivors in the two groups had comparable mean HRQoL.

Arch/descending CTAA often required complicated intervention, with most requiring ESG or OSR in more than one segment of the aorta. In addition, 11.2% required concomitant cardiac surgery, 14.0% required staged procedures and 4.6% required hybrid procedures. Although there was clear selection of younger, ‘fitter’ patients for OSR, outcomes for ESG and OSR patients were largely comparable. The invasive nature of OSR was associated with a slightly higher risk of death within 30 days, but few aneurysm-related procedural complications were reported. The main differences in clinical outcomes between interventions were higher readmissions in the short term for ESG patients, and the substantially higher incidence of postoperative complications, longer initial hospital stay and subsequent reinterventions in the OSR group. Although ESG patients were more likely than OSR patients to be readmitted to hospital early after discharge, their overall rate of readmission was not significantly higher, despite the 7- to 10-year age difference. This is at odds with the literature on infrarenal AAA repair, in which the complication rate and subsequent need for reintervention are higher. 100 This perhaps reflects the differences in natural history of the thoracic and abdominal aorta. After excluding patients who had a contraindication to one of the procedures, the results were qualitatively unchanged, although the smaller sample size resulted in more variable estimates of outcomes and low precision.

A large initial decrease in HRQoL followed OSR, a feature reported in other cardiothoracic surgery and vascular surgery trials,71,72 including the EVAR 1 trial,73 and is related to the longer hospital stay and post-surgical complications. Beyond this period, ESG and OSR had similar HRQoL in the full data analysis; conversely, the subset of OSR patients who were also eligible for ESG had significantly poorer HRQoL. The results for patients who were eligible for both ESG and OSR were based on small numbers and the difference between the groups from the full data analysis is more reliable. We also highlight that these analyses include only surviving patients and are not restricted to the nominal planned measurement times.

Can aneurysm- or patient-related predictors of treatment outcomes be determined?

Aneurysm location, age, sex, NYHA class and time before intervention were most closely related to survival after intervention, while previous HRQoL, smoking and NYHA class were related to HRQoL.

Patients receiving OSR had higher all-cause and aneurysm-related death rates overall, and this was more significant for DTA aneurysms than for arch aneurysms. Poorer survival after OSR in the DTA was partly influenced by treatment of CTD patients who have weaker vessel walls in this segment. Similar effects were noted in analyses that excluded patients with CTD (among other contraindications), even after the application of propensity score methods, suggesting that other factors were also in play. In observational studies it is difficult to delineate treatment effects from patient selection effects. As well as age at procedure, post-intervention survival was associated with a longer interval between entry to the study and intervention. The timing of intervention is challenging, our analysis showed that delays were associated with surgery in the DTA/thoracoabdominal aorta (especially OSR in these vessels), staged procedures, current smoking, previous CABG/PCI for coronary artery disease and concomitant valve disease, but most were not significant. It is likely that during the wait aneurysms will continue to grow and patients will continue to deteriorate over time, which may also contribute to this finding. Ultimately, it may be that vasculopathy of the DTA is simply a worse disease than vasculopathy of the arch, given the potential impact of the former aneurysms on spinal, renal, mesenteric and lower limb blood flow.

The strongest predictor of postoperative HRQoL was preoperative HRQoL, both of which were related to current smoking and NYHA grading of breathlessness. Recommending smoking cessation and treating causes of breathlessness may ease these symptoms to some extent and, therefore, optimise HRQoL. 74 This is supported by evidence that smoking cessation prior to other major procedures improves short-term outcomes. 75,76

What is the most cost-effective strategy in patients eligible for both endovascular stent grafting and open surgical replacement?

No comparative cost-effectiveness analysis was feasible. However, detailed micro-costing of ESG and OSR highlights that the ESG procedure itself is more costly than an OSR procedure, with the costs of the stents being the main driver. However, the total costs from admission to discharge on average were higher for the OSR procedure, largely as a result of increased ICU days.

Despite the rigorous work that was undertaken in developing this cohort of patients, relatively few participants were eligible to receive both treatments. We planned to complete a within-study economic evaluation and a sophisticated microsimulation model. However, owing to the limited data available, the analysis was restricted to describing the costs and QALYs for those who had initially received an ESG or OSR in the ‘no contraindication’ cohort, conducting a regression analysis to explore the key drivers of costs, and a VoI analysis. Standard methods of VoI analysis were used (see Chapter 7), and the analysis was based on the assumption that only 100 people would be eligible for both procedures per year.

Over 12 months’ follow-up, mean EQ-5D-5L scores were lower than median scores, indicating that some people in both groups had very low scores. For OSR patients, mean EQ-5D-5L score was much lower 1 month after surgery than at baseline, as patients were still recovering from surgery. Mean EQ-5D-5L scores increased thereafter but, on average, were still lower than baseline scores. This differs from the results of HRQoL trajectories in Chapters 5 and 6 because patients who died were assigned a utility value of zero in the economic analysis and the analysis did not include measurements outside the nominal measurement times. For the ESG group, even at 1 month, EQ-5D-5L scores were not much lower than at baseline. Over a 12-month follow-up period, mean and median QALYs were greater in the ESG group than in the OSR group. For the analysis underpinning the estimates of the EVPI and EVPPI, ESG appears to be the dominant intervention, being both cheaper and more effective than OSR at 12 months and with an incremental cost per QALY of < £12,000 for ESG compared with OSR at 36 months. The value of the EVPI and EVPPI is very modest, given the likely costs of any research project that might be conducted, especially as any research would remove only part of the uncertainty. These results should, however, be treated with caution (see Strengths and weaknesses).

How does this compare with existing literature and what does it add?

To our knowledge, this is the first analysis to scrutinise the clinical effectiveness and cost-effectiveness of elective treatments for chronic arch/descending aneurysms. Previous publications have identified the higher costs but shorter length of stay and reduced morbidity with ESG, but these reports conflated both acute and chronic pathologies. The ETTAA study has identified that clinicians are clear about which patients can benefit most from ESG and which can benefit most from OSR. There are also clear and objective technical parameters that determine whether or not an aneurysm can be treated with ESG at all. Finally, in many important aspects, the results of the ETTAA study are a straightforward extrapolation of the findings of studies of the management of AAAs. 100,101 With this in mind, the ETTAA study has established the costs and clinical outcomes associated with the two interventions and has also demonstrated the futility of directly comparing the two interventions.

The other contribution of the ETTAA study has been to describe the natural history of medically treated arch/DTA CTAA under surveillance. The 8.6% per patient-year risk of death and the 4.6% risk of fatal/non-fatal rupture or dissection are similar to the data from published literature. Having described the patterns of growth and risks associated with medical management, the ETTAA study will facilitate the future planning, location and organisation of specialist aortic centres to cater to the projected increase in demand. This is based on substantial literature showing that high-volume specialist centres improve clinical outcomes in complex aortic interventions. 102,103

What further research is required?

In priority order:

1. The prediction of aneurysm growth and the timing of intervention is difficult, and aneurysm diameter may not be the best indicator of whether or not rupture will occur. More detailed analysis of the diameter, length and volume of aneurysms, as well as other anatomical features in the ETTAA data set, may help to refine decisions around when and how to intervene. Joint analysis of aneurysm growth and acute clinical events (rupture, dissection and death) would provide valuable information on the timing of interventions.

2. The ETTAA database is now well positioned to describe, from real-world practice, what low, medium and high risk mean in terms of objective variables. This would facilitate a risk–benefit analysis and enable patients to be better informed at consent.

3. Combining post-procedure ETTAA data with longer-term routine electronic data sources would throw light on longer-term survival and hospital admissions to understand whether there is a divergence in survival and reintervention rates, as seen in other studies,36 and with a view to identifying factors that reduce the risk of these events.

4. For quality improvement, and to better understand the drivers of outcome in each group separately, it would be helpful to have a registry that records all CTAA patients, with a wider (but carefully chosen) set of associated variables recorded to the same protocol. This could augment the existing national cardiac and vascular surgical databases (National Institute for Cardiovascular Outcomes Research and National Vascular Registry) and would enable a more reliable assessment of variables affecting outcomes within each intervention group. 104,105 In particular, drug therapies and the value of optimisation prior to intervention could be investigated in a larger group of patients, with more adequate adjustment made for confounders. A registry that is maintained and adopted by the majority of specialist centres will also allow the longer-term follow-up of patients pre and post intervention.

Death

Recorded death by the local centre.

Myocardial infarction

The clinical suspicion of myocardial infarction together with elevated CK-MB (creatine kinase MB isoenzyme) or troponin and/or ECG/echo findings consistent with acute myocardial infarction.

Cardiac support

Support of myocardial pump function either by the use of intravenous/inhaled inotropic agents (e.g. adrenaline, noradrenaline, enoximone, dopamine, nitric oxide) or by the use of an intra-aortic balloon pump.

Prolonged ventilation > 48 hours

Support of respiratory or ventilatory function by means of a mechanical ventilator for more than 48 hours after (a) admission (for conservatively managed WW or CM patients) or (b) procedural intervention by means of ESG or surgery.

Renal support

Gastrointestinal complications

A new diagnosis of any of the following conditions as determined by the clinical history and standard investigations, interpreted and documented by a qualified physician: upper/lower gastrointestinal bleeding, intestinal ischaemia (small or large bowel), stoma formation, or others including (but not confined to) oesophagitis, duodenal ulcer (perforated or bleeding), erosive gastritis, pancreatitis, liver failure/necrosis and cholecystitis.

Neurological injury

Thromboembolic event (deep-vein thrombosis/pulmonary embolism)

Evidence of a venous thromboembolic event (e.g. deep-vein thrombosis, pulmonary embolism) by standard clinical and laboratory testing.

Infection

Infection pertaining to the operated segment of aorta (including periprosthetic abscess), vascular access site, surgical incision, lungs, pleural/peritoneal cavity or urinary tract; as diagnosed by an appropriately qualified physician according to standard clinical investigations.

Return to theatre

A secondary visit the operating/hybrid theatre for treatment or examination of suspected complications following but during the same admission as the index intervention by ESG or OSR.

Access vessel injury

New-onset intramural haematoma, pseudoaneurysm, dissection, avulsion, disruption, rupture or occlusion of any vessel used to provide vascular access for the delivery of an endovascular stent graft.

Endoleak

Aneurysm complication

Any direct complication localised to the operated segment of aorta, including (but not necessarily confined to) localised rupture, dissection, or pseudoaneurysm formation. This must be diagnosed and documented by an appropriately qualified physician (e.g. vascular/cardiothoracic surgeon or interventional radiologist) according to standard clinical and radiological investigations.

Fistula formation

Defined as an abnormal connection between the operated/stent-grafted segment of aorta and another epithelialised surface, and diagnosed according to standard clinical and radiological investigations.

Reintervention

Any intervention undertaken to preserve or restore the function of an endovascular stent graft (e.g. re-ballooning/additional stent/surgery) or surgically implanted aortic graft.