Introduction

Open repair of AAA was first introduced by Dubost et al. in 1951.⁷⁵ In the 1990s the less invasive EVAR was introduced and the first multicentre randomised trial of EVAR compared with OR was started in 1999 (EVAR-1), in the UK.¹⁰⁰ This was soon followed by other multicentre trials in Europe (the DREAM and ACE trials) and the OVER trial in the USA.^98,99,104

Each of the randomised trials of EVAR compared with OR recruited patients (suitable for either OR or EVAR) with slightly different entry characteristics with respect to age, sex, aneurysm morphology and other demographics. The EVAR-1, DREAM and OVER trials all showed an early survival benefit for EVAR, whereas the ACE trial did not. This early survival benefit for EVAR was lost within 1–3 years for EVAR-1 and the DREAM trial, but not until later for the OVER trial.^94,98,104 This ‘catch-up’ in mortality has been noted in many other studies, including a Cochrane review and analyses of the Medicare database,^108,139 but no satisfactory explanation for this phenomenon has emerged (the Cochrane review was limited by not being able to report aneurysm-related mortality or subgroup analyses).¹³⁹ Each trial individually has been too small to investigate the reasons for this ‘catch-up’ mortality in the EVAR groups or to answer the much-discussed question of whether or not younger and fitter patients (or other subgroups) should be offered OR, which is considered more durable than EVAR.^140,141 This ‘catch-up’ in mortality needs to be avoided if EVAR is to outperform OR in the longer term. To try to address some of these issues, the four randomised trials agreed to pool their data for an IPD meta-analysis.

Methods

In July 2013, MEDLINE, EMBASE and clinical trial databases were searched for randomised trials comparing OR and EVAR of AAAs. The search terms used were AAA, aneurysm, endovascular, stent, open repair and randomised trial. From 275 reports, we identified four eligible trials reporting mid-term follow-up. The methods for these four multicentre trials included in this meta-analysis have been published previously.^{94,102,105,106} EVAR-1 randomised 1252 patients (91% male), with an aneurysm diameter of > 5.5 cm, between September 1999 and August 2004 in the UK. The DREAM trial randomised 351 patients (92% male), with an aneurysm diameter of ≥ 5 cm, between November 2000 and December 2003 in the Netherlands and Belgium. The OVER trial randomised 881 patients (99% male), with an aneurysm diameter of ≥ 5.0 cm, between October 2002 and April 2008 at Veteran Affairs hospitals in the USA. The ACE trial randomised 306 patients (99% male), with an aneurysm diameter of > 5.0 cm, between March 2003 and March 2008 in France (seven patients withdrew consent before discharge). All patients were considered fit for open surgery under general anaesthesia and all trials used approved devices for EVAR, predominantly within the manufacturers’ IFU, and followed up patients for a minimum of 3 years. Summary baseline characteristics for patients by trial are shown in Table 18.

TABLE 18

Baseline and post-randomisation characteristics of patients in the four trials

The four data sets were merged based on fields available in the case record forms of the largest trial (EVAR-1), range checks were conducted and queries resolved with the individual trial co-ordinating centres. The common baseline variables across the trials were age, sex, history of smoking, diabetes, coronary artery disease (defined as previous stable or unstable angina or MI), BMI, maximum aneurysm diameter, proximal aortic neck length and diameter, ankle–brachial pressure index (ABPI) and creatinine level, used for estimated glomerular filtration rate (eGFR),¹⁴² but without information on ethnicity. Each trial also contained some data about hypertension, which was included in a modified Wilkins cardiovascular survival risk score¹⁴³ (Table 19). The reporting of both drug use (including antiplatelet and lipid-lowering agents) and reinterventions was very different in the four trials (particularly intestinal and wound-related reinterventions following OR). The postoperative surveillance protocol was identical for both randomised groups in all trials, except for the DREAM trial, in which, after 2 years, surveillance was relaxed for the OR group. However, for complications, only endoleaks after EVAR were reported similarly across the trials; laparotomy-related complications were not.

TABLE 19

Derivation of a cardiovascular risk score based on a modified version of Wilkins et al.

Statistical analysis

The primary analyses considered the groups ‘as randomised’ within each trial. Mortality after randomisation was assessed at 30 days, in hospital and then at three defined time periods: 0–6 months, 6 months–4 years and > 4 years after randomisation. Aneurysm-related mortality included death from (1) primary aneurysm rupture, (2) within 30 days of aneurysm repair or any reintervention; and (3) rupture after repair. Given the different times between randomisation and aneurysm repair in the four trials (Table 20), aneurysm-related mortality was also assessed at 30 days, 31 days–3 years and > 3 years after aneurysm repair. Kaplan–Meier survival curves by randomised group were generated from the combined data from all four trials and the restricted mean life-years up to a certain time estimated by the area under the curve up to that time. Logistic regression was used to compare operative (30-day) and in-hospital mortality among patients who underwent repair, and Cox proportional hazards regression was used to compare total and aneurysm-related mortality and time to reintervention. A two-stage IPD meta-analysis was performed. First analyses were conducted separately within each trial and then pooled time-period-specific estimates were calculated using random-effects meta-analysis with between-study heterogeneity estimated using the method of DerSimonian and Laird.¹⁴⁴ The proportion of between-trial variability beyond that expected by chance was quantified using the I²-statistic.¹⁴⁵ All analyses were then repeated adjusting for the following baseline covariates: age, sex, maximum aneurysm diameter and log-creatinine level.

TABLE 20

Unadjusted and adjusted HRs for total mortality by time since randomisation

The subgroups age, sex, eGFR, coronary artery disease, ABPI, the modified Wilkins cardiovascular survival risk score, maximum aneurysm diameter, proximal aneurysm neck diameter and neck length were assessed for differences in the effect of the EVAR and open strategies by including an interaction term between the subgroup and randomised group in a Cox regression model. Except for sex and coronary artery disease, all measures were entered as continuous variables to assess effect modification. Each interaction term was pooled across the trials using random-effects meta-analysis and its statistical significance assessed using a Wald test, taking the 5% level as significant. For presentation purposes only (and not for assessing significance), HRs are shown by dichotomising continuous measures at chosen cut-off points: age (72 years), eGFR (68.4), ABPI (0.9), cardiovascular risk score (2 major), maximum aneurysm diameter (5.9 cm), neck diameter (2.3 cm), neck length (2.5 cm), estimating the HRs within each subgroup and pooling these across studies.

The hazard of reintervention following aneurysm repair was analysed using an Anderson–Gill multiple failure time model.¹⁴⁶

All analyses were performed using Stata statistical software, version 13.

Results

A total of 2783 patients, with 14,245 person-years of follow-up, were included in this meta-analysis: their baseline characteristics are shown in Table 18, with significant intertrial differences in all variables. Patients in EVAR-1 were older and had larger aneurysms than patients in the other trials. Nearly all patients in the OVER and EVAR-1 trials had a history of smoking, compared with about half the patients in the DREAM and ACE trials. Nearly all patients in the OVER and EVAR-1 trials had a history of smoking, compared with about half the patients in the DREAM and ACE trials, and patients in the OVER trial had the highest BMI and highest proportion of patients with diabetes. Summary post-randomisation characteristics of the trials also are given in Table 18. The median follow-up was 6.0, 6.0, 5.4 and 3.1 years, for EVAR-1, the DREAM, OVER and ACE trials, respectively, 5.5 years for the pooled data and compliance with randomised allocation was 93% or higher in all trials.

Total mortality

Over the follow-up of all four trials there were 481 deaths in the EVAR groups and 482 in the OR groups. Kaplan–Meier curves by randomised group for total mortality across all four trials are shown in Figure 13. Overall, there was no difference in total mortality over the follow-up period of the trial (pooled HR 0.99, 95% CI 0.87 to 1.13) (Figure 14a and see Table 20). Between 0 and 6 months mortality was lower for the EVAR groups, with 46 deaths, than for OR, with 73 deaths (HR 0.61, 95% CI 0.42 to 0.89), with no evidence of heterogeneity between the trials. The early survival advantage of EVAR in the first 6 months was largely attributable to the lower 30-day operative mortality for EVAR than with OR (unadjusted pooled odds ratio 0.40, 95% CI 0.22 to 0.73) (Figure 14b). After this, the early advantage of the EVAR group was lost and the HRs moved (non-significantly) in the direction of OR. Adjusted HRs were similar. By 5 years the estimated survival was 73.6% (95% CI 71.1% to 75.9%) in both the EVAR and OR groups with an expected 0.06 additional life-years in the EVAR group, corresponding to 23 days (95% CI –16 to 61 days; p = 0.246). The causes of death by each time period are shown in Table 21.

FIGURE 13

Kaplan–Meier survival curves for overall total mortality, by randomised group, for all four trials combined.

FIGURE 14

(a) Total mortality, overall and at 0–6 months, 6 months–4 years and > 4 years since randomisation, unadjusted HRs; and (b) mortality within 30 days of operation, showing odds ratio.

TABLE 21

Causes of death by randomised group and time since randomisation

Aneurysm-related mortality

The findings for aneurysm-related mortality were similar in direction, with relative benefit for the EVAR groups 0–6 months after randomisation (25 aneurysm-related deaths vs. 55 in the OR groups; pooled unadjusted HR 0.44, 95% CI 0.26 to 0.76). In later time periods the results move in the direction of OR, 6 months–4 years and > 4 years, with pooled HRs of 1.43 (95% CI 0.61 to 3.34) and 2.29 (95% CI 0.49 to 10.85), respectively (Figure 15). For those who received aneurysm repair, analysis by time from repair showed a strong relative advantage for the EVAR group in the first 30 days, between 30 days and 3 years there was no difference between the groups, but after 3 years there was a significant relative advantage for the OR group, with three aneurysm-related deaths compared with 19 deaths in the EVAR groups (HR 5.16, 95% CI 1.49 to 17.89; p = 0.010) (Table 22).

FIGURE 15

Aneurysm-related mortality, overall and at 0–6 months, 6 months–4 years and > 4 years since randomisation. Unadjusted HRs.

TABLE 22

Unadjusted and adjusted HRs for aneurysm-related mortality by time since operation for those who underwent surgery

Total mortality by subgroups

There was no significant effect of age or sex on the relative effectiveness of EVAR in preventing deaths in any time period, including the first 6 months following randomisation (Figure 16). There were two subgroups of patients who appeared to have no early benefit (to 6 months) under EVAR compared with OR: patients with moderate renal dysfunction and those with coronary artery disease. For those with above-median eGFR, the pooled HR of 0.42 (95% CI 0.21 to 0.84) was significantly in favour of EVAR compared with the less favourable and non-significant pooled HR of 0.68 (95% CI 0.43 to 1.08) for those with worse renal function; therefore, the interaction between eGFR measure and treatment group was significant (interaction p = 0.024) (see Figure 16).

FIGURE 16

Unadjusted HRs for total mortality by subgroups of age, sex and eGFR: overall and at 0–6 months, 6 months–4 years and > 4 years since randomisation. Interaction p-value for age and eGFR calculated using the continuous measures (more...)

Similarly, patients with coronary artery disease gained no early advantage of being in the EVAR group, in comparison with patients without prior coronary artery disease (interaction p = 0.047) (Figure 17). None of the morphological aneurysm characteristics, smoking, diabetes or BMI was associated with mortality (Figures 18 and 19).

FIGURE 17

Unadjusted HRs for total mortality by subgroups of history of angina or MI, ABPI and cardiovascular risk score: overall and at 0–6 months, 6 months–4 years and > 4 years since randomisation. Interaction p-values for ABPI (more...)

FIGURE 18

Unadjusted HRs for total mortality by subgroups of maximum AAA diameter, neck diameter and neck length: overall and at 0–6 months, 6 months–4 years and > 4 years since randomisation. Interaction p-values for maximum AAA (more...)

FIGURE 19

Unadjusted HRs for total mortality by subgroups of history of diabetes, BMI and smoking status: overall and at 0–6 months, 6 months–4 years and > 4 years since randomisation. Interaction p-values for BMI calculated using (more...)

Baseline ABPI was not available for the ACE trial. In the other trials, patients with peripheral arterial disease (low ABPI of < 0.9) had a similar early survival advantage from being in the EVAR group as those with a ABPI of ≥ 0.9. However, in the 6 months to 4 years time period, for those with a ABPI of < 0.9, the OR group had the survival advantage (HR 1.67, 95% CI 1.12 to 2.49), in comparison with patients with a ABPI of > 0.9 (interaction p = 0.022). During the 6 months to 4 years time period, for those with a ABPI of < 0.9, total mortality was 9.6 and 5.7 per 100 person-years in the EVAR and OR groups, respectively, compared with 5.1 and 5.8 per 100 person-years, respectively, in the higher ABPI group. The cause of death in the two ABPI groups by time period is shown in Table 23. The operative mortality by subgroup shows that the highest mortality was in those with low ABPI (Table 24) and this group has higher aneurysm-related mortality throughout. Finally, a cardiovascular risk score was not discriminatory at any time period.

TABLE 23

Causes of death by categorisation of baseline ABPI and time since randomisation

TABLE 24

Operative mortality by subgroup level, for individuals who underwent an operation

Complications and reinterventions

Complications (apart from endoleaks after EVAR) and reinterventions were reported heterogeneously across the four trials. The overall rates of reinterventions reported were higher in the EVAR group than in the OR group for all trials (see Table 18). The risk of reintervention by time period following aneurysm repair is shown in Figure 20, with substantial heterogeneity between trials for reinterventions recorded between 31 days and 3 years.

FIGURE 20

Hazard ratio for any reintervention by time period following aneurysm repair for 2718 subjects undergoing aneurysm repair (seven subjects missing follow-up for reinterventions after aneurysm repair). Note that the number of reinterventions reported here (more...)

There was no indication that complications following EVAR decreased with the year in which the trial commenced: these rates, together with types and numbers of complications, are reported in Table 25. The most common reported complication after EVAR was type II endoleak, which overall was reported 435 times in 325 of 2783 (12%) patients, with corrective reintervention being performed in 99 of 435 (23%) detected type II endoleaks. The second most common type of complication was type I endoleak, for which 79 of 120 (66%) patients received an early reintervention. Similarly, early correction of other serious EVAR-related complications was attempted only in less than two-thirds of cases. Secondary sac rupture was reported in 37 patients – 33 in the EVAR randomised groups (2.4% of patients) and four in the OR groups (0.3% of patients); although all four of these patients were treated with EVAR, for those with secondary rupture following treatment with EVAR, the median time to rupture was 3.5 years (Figure 21). Of these patients, 11 (30%) had a type I endoleak, of which seven were treated, seven (19%) had a type II endoleak, of which three were treated, two (5%) had a type III endoleak, of which one was treated, and nine had known graft migration. Nineteen (51%) patients had no endoleaks detected before secondary sac rupture and one further patient had a thoracic endograft for proximal aortic dissection 3 days earlier. The mean time between detection of the first endoleak and rupture was 1.8 years. The 30-day mortality rate following rupture was 62% (n = 23).

TABLE 25

Complications by trial, focusing on EVAR-related complications

FIGURE 21

Time from repair to secondary rupture, in those who were treated with EVAR.

Discussion

These four randomised trials, in Europe and the USA, provide the best evidence for the early survival advantage offered by EVAR rather than OR. Patients prefer the less invasive method of AAA repair, EVAR, which has been widely adopted, and the majority of elective repairs are now performed using EVAR.^108,120,147 This meta-analysis, over a 5-year time horizon, confirms that overall there is an early survival advantage under EVAR, which is lost within 3 years of randomisation, so that the life-years saved from EVAR over a 5-year time period are minimal. Between 0 and 6 months after randomisation, total and aneurysm-related mortality were lower for the EVAR group, mainly because of the 2.5-fold lower operative mortality in this group. However, after this time period, the early EVAR group advantage was eroded progressively. By 3 years after aneurysm repair, aneurysm-related mortality was five times higher in the EVAR group (at this stage mainly because of secondary rupture or reinterventions), and this is likely to contribute to the ‘catch-up’ in mortality.

The next investigations focused on whether the early survival advantage was either maintained or lost in subgroups of patients categorised by their preoperative characteristics. Over a 5-year time horizon, there was no convincing evidence that being randomised to EVAR or OR resulted in differential survival between any subgroups of the population. This does not support the suggestion that younger and fitter patients with aortic morphology suitable for EVAR are likely to benefit from OR over 5 years.¹⁴⁰ However, differential effect modification was suggested in some subgroups (renal dysfunction, coronary artery disease) in the first 6 months and for those with peripheral arterial disease in the later 6 months–4 years time period, possibly caused by frailty effects.

Low ABPI was introduced as a measure of peripheral atherosclerosis¹⁴⁸ and is a marker of generalised atherosclerosis.¹⁴⁹ This subgroup had the highest pooled operative mortality, although the relative early advantage of EVAR was maintained. However, between 6 months and 4 years, fortunes reversed in favour of a survival advantage in the OR group, a pattern which indicates that those with low ABPI are another contributor to the ‘catch-up’ in mortality phenomenon.

All trials enrolled patients with evidence of moderate renal dysfunction, 35% of the overall enrolment having eGFR < 60 (chronic kidney disease stage 3 or above). For these patients, there was no evidence of a benefit from being randomised to EVAR (vs. OR), even in the first 6 months. These data are in agreement with an earlier observational study showing a high early postoperative event rate for patients with a low eGFR.¹⁵⁰ Whether or not renal function deteriorates more rapidly after either elective EVAR or OR of an AAA has been debated fiercely and less attention has been focused on improving the perioperative care.¹⁵¹ Similarly, patients with known coronary artery disease had no evidence of an early survival advantage from being randomised to EVAR. Given that EVAR is less invasive than OR, these findings for the moderate renal dysfunction and coronary artery disease are surprising. Perhaps the stress of EVAR in these subgroups has been underestimated. It also is possible that those randomised to EVAR received less stringent preoperative evaluations, resulting in better perioperative care for the OR group with these comorbidities.

This study has several limitations that restrict its scope. First, all endografts were implanted between 1999 and 2008 using general anaesthesia, and today newer endografts are used, often implanted under local anaesthesia. Second, reporting standards for baseline characteristics (e.g. smoking, drugs) differed between trials. Third, the reporting of complications and reinterventions (aneurysm related and other cardiovascular) was very heterogeneous across the trials. Moreover, today it is recognised that type II endoleaks with sac enlargement can be dangerous¹¹¹ and that a type II endoleak might even hide a type I or type III endoleak. Fourth, at the time these trials recruited (1999–2008), no trial had a clear policy for reintervention in the presence of endoleaks after EVAR and even serious complications such as type I endoleak were not always corrected quickly, which may have contributed to the increasing aneurysm-related mortality rate in the EVAR group at ≥ 3 years after aneurysm repair.

The reintervention rate was consistently higher in the EVAR groups (see Table 18), although the data are heterogeneous and the largest trial did not report incision-related complications after OR. It would be reassuring to learn that by using more recent EVAR devices within IFU, coupled with more rigorous surveillance, the continuing aneurysm-related mortality in the EVAR group could be attenuated to minimise the ‘catch-up’ in mortality. To rely entirely on the introduction of new devices to prevent aneurysm-related mortality in the EVAR group, especially without adequate surveillance, may be unwise. Recent analyses of the Medicare database support this caution.¹⁰⁸

In summary, this meta-analysis confirms the advantage of lower mortality in the EVAR group in the first 6 months and provides some new insight of how this early advantage of the EVAR group is eroded (aneurysm-related mortality and inclusion of those with peripheral arterial disease). Additionally, two subgroups were identified who do not have lower mortality under EVAR at any time, to suggest that these groups (moderate renal dysfunction and established coronary artery disease) may benefit from improved perioperative care, especially for EVAR. Surveillance must focus on reducing aneurysm-related deaths in the mid-term and longer term, particularly deaths resulting from reinterventions and secondary ruptures after EVAR.