Endovascular stents for abdominal aortic aneurysms: a systematic review and economic model

Study characteristics

The characteristics of the included RCTs are summarised in Tables 4–7.

Results by outcome

30-day mortality

All four RCTs comparing EVAR with open repair in patients with unruptured AAAs (DREAM,41 EVAR trial 142 and the studies by Cuypers et al. 44 and Soulez et al. 45) reported 30-day operative mortality (Figure 2). Results from a later analysis of EVAR trial 1 based on a larger sample size gave an odds ratio (OR) of 0.38 (95% CI 0.18 to 0.80). 23 The pooled estimate of effect suggested a significantly lower rate of 30-day mortality in the EVAR group: pooled OR 0.35 (95% CI 0.19 to 0.63).

FIGURE 2.

EVAR vs open repair: meta-analysis of RCTs for 30-day mortality.

The small Soulez et al. trial45 did not contribute to this analysis and exclusion of the less relevant data from the Cuypers et al. trial44 (i.e. a pooled analysis including only the DREAM and EVAR trial 1) produced an almost identical measure of effect: pooled OR 0.35 (95% CI 0.19 to 0.64) (Figure 3).

FIGURE 3.

EVAR vs open repair: meta-analysis of the DREAM and EVAR trial 1 RCTs for 30-day mortality.

Abdominal aortic aneurysm-related mortality

The two small RCTs failed to provide information on AAA-related mortality at mid-term follow-up. The DREAM trial and EVAR trial 1 had similar definitions of AAA-related mortality, i.e. death within 30 days of the original procedure or a reintervention. DREAM was originally designed to detect differences in a primary end point of short-term mortality and complications and so its power to detect differences at longer-term follow-up is unclear. The mean duration of medium-term follow-up was about 22 months in DREAM compared with a median of about 35 months (2.9 years) in EVAR trial 1. Maximum follow-up in DREAM was 42 months whereas 24% of patients in EVAR trial 1 were followed up for 4 years or more. Longer-term data for AAA-related mortality were not available.

Both RCTs reported lower rates of AAA-related mortality in patients treated with EVAR than in those undergoing open repair. In DREAM, 3/173 patients in the EVAR group (2.1%) and 9/178 in the open repair group (5.7%) died of aneurysm-related causes. The estimated hazard ratio (HR) was 0.27 (95% CI 0.07 to 1.00, p = 0.05). In EVAR trial 1 there were 19/543 and 34/539 deaths in the EVAR and open repair groups respectively. The unadjusted HR was 0.55 (95% CI 0.31 to 0.96, p = 0.04); HRs adjusted for primary and secondary covariates were similar. Results from a later analysis of EVAR trial 1 based on a larger sample size gave an HR of 0.60 (95% CI 0.35 to 1.02). 23

The pooled estimate for the HR across the two trials was 0.49 (95% CI 0.29 to 0.83, p = 0.007), confirming a statistically significant benefit of EVAR over open repair for this outcome (Figure 4).

FIGURE 4.

EVAR vs open repair: meta-analysis of RCTs for abdominal aortic aneurysm-related mortality at follow-up.

In a post hoc analysis, follow-up was divided into the first 6 months after randomisation and the period beyond 6 months. The HR for the first 6 months was 0.42 (95% CI 0.21 to 0.82), a statistically significant difference favouring the EVAR group. For the later period the HR was 1.15 (95% CI 0.39 to 3.41), i.e. there was no significant difference between groups; the wide confidence interval reflected the small number of AAA-related deaths during this period.

All-cause mortality

Of the four relevant RCTs, only DREAM and EVAR trial 1 provided useful information on all-cause mortality at follow-up (2 years in DREAM and 4 years in EVAR trial 1). The trial by Soulez et al. 45 reported only one death during a mean follow-up of 29 months for the EVAR group and 27 months for the open repair group. In the trial by Cuypers et al. 44 patients were only followed up for 30 days.

The two main RCTs reported no significant differences in medium-term (35 and 42 months, respectively) mortality in patients treated with EVAR compared with those treated with open repair. In DREAM, 20/173 patients in the EVAR group and 18/178 in the open repair group died of any cause. The estimated unadjusted HR was 0.94 (95% CI 0.50 to 1.79, p = 0.86). In EVAR trial 1 there were 100/543 and 109/539 deaths in the EVAR and open repair groups respectively. The unadjusted HR was 0.90 (95% CI 0.69 to 1.18, p = 0.46); HRs adjusted for primary and secondary covariates were similar. Results from a later analysis of EVAR trial 1 based on a larger sample size gave an HR of 0.93 (95% CI 0.74 to 1.18). 23

In a post hoc analysis, follow-up was divided into the first 6 months after randomisation and the period beyond 6 months. The HR for the first 6 months was 0.55 (95% CI 0.33 to 0.93), a statistically significant difference favouring the EVAR group. For the later period the HR was 1.10 (95% CI 0.80 to 1.52), i.e. there was no significant difference between groups.

A pooled analysis of the two trials confirmed that there was no statistically significant difference between EVAR and open repair for all-cause mortality at medium-term follow-up (Figure 5).

FIGURE 5.

EVAR vs open repair: meta-analysis of RCTs for all-cause mortality at follow-up. (Academic-in-confidence information on EVAR trial 1 has been omitted.)

Study characteristics

One RCT47 compared EVAR and open repair in patients with ruptured AAAs. Only 32 patients were randomised compared with a planned sample size of 100 and so it is difficult to draw firm conclusions from the trial. Compared with RCTs of elective EVAR, the patients were similar in age but had larger aneurysms and the proportion of women was slightly higher. Non-commercial stent grafts were used in patients receiving EVAR. Other study characteristics are shown in Tables 11–14.

Results

Study characteristics

EVAR trial 246 is the only published RCT in this patient group. This UK RCT compared EVAR (n = 166) with non-surgical management (n = 172) in patients judged to be unfit for open repair. The trial met all quality criteria. The primary end point was all-cause mortality and secondary end points were aneurysm-related mortality, HRQoL, postoperative complications and hospital costs. A total of 14 patients randomised to EVAR died before operation (including six from AAA rupture), and 47 patients assigned to non-surgical management received surgical aneurysm repair (including 12 who received open repair despite having been classified as unfit for this procedure). These factors complicate the analysis and interpretation of the trial. Tables 15–18 give details of patient, intervention and comparator characteristics.

Results

Clinical expertise

NVD and RETA did not specify entry requirements for centres to be eligible for inclusion in the registries. It is therefore unclear what level of expertise the surgical teams had with performing EVAR and open repair, which makes it difficult to compare patient outcomes for the different registries and to ascertain whether there may be an association between surgical experience and outcomes. Because procedures carried out by specialist teams with a high level of experience in EVAR result in lower mortality rates and fewer adverse events that lead to secondary interventions,53 EUROSTAR specifies that centres must have a throughput of at least 10 patients undergoing stent graft procedures for AAA per year if they are to be included in the registry.

Data collection

EUROSTAR data were collected using a case record form, which included an informed consent form for signing by the patient. Only surgeons from participating centres who had sufficient expertise (i.e. involvement in a series of at least 10 stent graft procedures for AAA) submitted data to the registry.

Submission of data to the NVD was on a voluntary basis, with almost half of the members of the Vascular Society contributing to the database at the time of the report. However, to gain a true picture of the outcomes of vascular surgery (e.g. AAA repair) throughout Great Britain and Ireland, inclusion of all surgeons performing such operations is needed, but at the time of the report external validation to ensure accuracy and completeness of data had not been undertaken. 16

Data collection for RETA was also on a voluntary basis and the UK centres submitted cases as they were performed. However, the majority of endovascular repairs in the UK at the time were performed as part of the EVAR trials and cases submitted to RETA at the time of their report were cases performed outside the trial (usually early on in a centre’s experience to allow entry into the EVAR trials), and as such the full RETA data set of all cases submitted was less representative of UK practice at the time of the report.

It is unclear whether all participants undergoing EVAR or open repair were included in the registry, but as only certain surgeons were submitting cases, potential sample bias cannot be ruled out.

Dates of procedures

Patients were registered and treated between 1999 and 31 March 2004 for NVD, up to June 2006 for EUROSTAR, and between January 1996 and March 2000 for RETA. Data from the RETA registry are therefore very out of date, which suggests that the results may not be relevant to current practice. This highlights the importance of the data provided by the EUROSTAR registry. The relevance of the data is also reflected in the use of ‘older’ types of devices by the RETA registry. The latest report from the EUROSTAR registry explicitly excluded any data relating to ‘older’ devices and included only those patients treated with the newer generation of endografts in current use.

Procedure details

The report from the EUROSTAR registry identified nine devices, with the Zenith, Talent and Excluder devices being the main ones in use (39.6%, 28.3% and 13.9%, respectively), all of which are still in current use. By comparison, the RETA data includes 14 devices (four of which were ‘home made’), the main ones in use being AneuRx, Zenith and Gianturco-Dacron (home made) (25.4%, 14.4% and 12.3%, respectively). However, as mentioned above and in Chapter 1 (see Description of technology under assessment), many of the devices included in RETA are no longer in current use. ‘In-house’ (homemade) uni-iliac stents were once the most often used type of graft but have now been superseded by commercially available and CE-marked devices,56 such as those included in the EUROSTAR registry.

Bi-iliac grafts were the most prevalent form of graft type used by EUROSTAR and RETA (89.8% and 70.4%, respectively). This reflects the increasing use of bi-iliac grafts for EVAR, which appear to be superseding other types of graft such as the aortic tube, the use of which fell because of the number of distal endoleaks associated with this type of device. 56 This again highlights the importance of the EUROSTAR data and its greater relevance to current practice as the RETA registry includes the use of aortic tube grafts and a smaller percentage of patients received bi-iliac grafts compared with patients in the EUROSTAR registry.

General anaesthesia was reported to be used most often by all three registries (Table 22).

Patient characteristics

Full details of patient characteristics are given in Tables 23 and 24.

To be eligible for inclusion in the EUROSTAR registry, patients were required to meet specific entry criteria: age greater than 21 years and presenting for elective AAA operation without symptoms of rupture or expansion. Patients were excluded if their aneurysms measured < 3 cm, and patients with aneurysms measuring 3–4 cm were only included if they were associated with iliac aneurysms. The mean aneurysm diameter for patients included in the registry was 5.84 cm, ranging between 3 and 17.2 cm. The majority of patients were male (93.2%) and the mean age was 72.5 years, ranging between 34 and 100 years. Approximately half of the patients had a history of smoking (51.1%), and a high proportion reported a history of heart disease (78.8%) and hypertension (65.5%). Almost half had a history of pulmonary disease (42.3%), a quarter were classed as unfit for open repair and a quarter were considered obese (Table 24).

In the RETA data, details for gender were available for only 51.4% of patients; however, 90% of this population was male. The median age reported was 73 years, ranging between 44 and 93 years. The health status of patients was unclear from the registry data as no details were provided for comorbidities. However, almost half of patients presented with aneurysms > 6 cm and fitness scores indicated that almost a quarter of patients (22.7%) were classified as unfit for open repair. Incomplete reporting of details was one of the shortcomings of this registry as it is difficult to then make comparisons.

In the NVD the majority of patients were male (84.4%), although this figure is almost 10% less than the corresponding figure in the EUROSTAR population. Mean age was 72.5 years, which was comparable to that in the EUROSTAR population. The mean aneurysm diameter was not reported but sizes ranged from < 5 cm to > 9.9 cm and there was a 1-cm difference between the majority of ruptured and the majority of unruptured AAAs. Only one patient characteristic of interest was reported, which indicated that almost half the population (44.2%) had a history of heart disease.

EUROSTAR54

EUROSTAR presented outcomes for short-term (30-day) and long-term (96 months/8 years) mortality, with 190 (2.3%) deaths occurring within 30 days and 789 (9.5%) during the follow-up period. It is unclear from the report whether patients died from aneurysm-related or other causes. Kaplan–Meier survival analysis reported the cumulative number of deaths as 979 and a mortality rate of 39%. It should be noted, however, that for the 30-day outcome 4543 patients were observed out of 5515 expected; 90 patients were observed out of 326 expected for 84 months’ follow-up; and only 20 patients were observed out of 77 for 96 months’ follow-up. In total, 111 patients (1.3%) were lost to follow-up, but this will have been included as censored data and accounted for by the Kaplan–Meier survival analysis.

RETA56

RETA reported outcomes for short-term (30-day) mortality,58 aneurysm-related mortality at follow-up and all-cause mortality at follow-up (5 years/60 months)56 (return rates for follow-up data are reported in Table 25). In total, 58 patients (5.8%) died within 30 days and 9 patients were reported to have died from fatal rupture (aneurysm-related mortality) at follow-up [6 (0.8%) at 1 year and 3 (0.8%) at 2 years]. A cumulative rate of all-cause mortality was not reported, although figures were presented for each year of follow-up: 11.9% mortality in year 1 and 10%, 8% and 7.9% at 2, 3, and 4 years post procedure respectively.

NVD16

Mortality rates following open repair were reported for the 30-day period only, with an overall crude mortality rate of 14.8% (95% CI 13.7 to 16.0%). Crude mortality rates for ruptured and unruptured AAAs were 41% (95% CI 37.7 to 44.3%) and 6.8% (95% CI 5.9 to 7.8%) respectively (Table 25).

EUROSTAR54

Some form of major adverse event was experienced by 11.1% of patients, with cardiac, pulmonary and renal events being the most significant. By the end of 96 months’ follow-up only a very small proportion of patients (0.5%) experienced rupture and device migration (1.8%), whereas 15.9% of patients experienced endoleak. In total, 9% of patients required some form of reintervention at 84 months, increasing to 19.2% at 96 months.

RETA56

RETA reported a very small percentage of ruptures during stent deployment (0.35%)57 and a cumulative rate of 2% at 5 years’ follow-up. 56 A total of 132 cases of endoleak (13.2%) were reported at 30 days,56 with a cumulative rate of 68% free from endoleak at 5 years’ follow-up. 56 A small number of device migrations were reported: 9 (0.9%) at 30 days (requiring conversion to open repair) and 14 over the 4-year follow-up. 57

Conversion to open repair within 30 days occurred in 3.3% of cases. Kaplan–Meier totals for cumulative rates of reintervention were not clearly reported; however, the rate at 5 years’ follow-up was reported as 62%. 56 This 5-year figure reflects a much higher reintervention rate than that reported by EUROSTAR at 8 years’ follow-up (19.2%). Only small numbers of cardiac events and stroke were reported at 30 days (4.2% and 1.5%, respectively57), but overall 27.8% were reported as having experienced some form of complication (including technical complications and renal failure56).

NVD16

No data were reported in the NVD registry for occurrence of endoleak and device migration as these complications cannot occur with open repair. No data were presented for rupture rates, reintervention rates or major adverse events, which limits analysis of the data and prevents comparison with the EVAR registries.

Study results

Details of the risk scores used in the three studies and the results are summarised in Tables 28 and 29. One study also assessed other risk factors and results for these factors are discussed later in this chapter (see Studies investigating specific risk factors).

Age

In total, 24 studies60,61,63–67,69,71–74,77–79,81,82,84,86–91 investigated the role of age in relation to adverse outcomes after EVAR (Figures 8–12). Age was either treated as a continuous variable or dichotomised, for example into under 80 years and octogenarians. 71

FIGURE 8.

Age and 30-day mortality. Studies to the left of the vertical line found age to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 9.

Age and aneurysm-related mortality. Studies to the left of the vertical line found age to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 10.

Age and all-cause mortality. Studies to the left of the vertical line found age to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 11.

Age and reintervention. Studies to the left of the vertical line found age to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 12.

Age and endoleak. Studies to the left of the vertical line found age to be an independent risk factor (IRF), whereas those to the right did not.

The evidence showed age to be a risk factor for 30-day mortality (Figure 8). For the outcome of aneurysm-related mortality evidence was mixed (Figure 9). For all-cause mortality all nine studies in this group correctly identified increasing age as an independent risk factor (Figure 10). Results for reintervention were almost all analyses of EUROSTAR data and most, but not all, studies concluded that age was not a risk factor (Figure 11). On balance the mainly EUROSTAR-based evidence indicates that age is an independent risk factor for type II endoleak or all types of endoleak (Figure 12).

Varying interpretations of old age and the way that data were handled may affect findings and may explain some of the inconsistency in the results in this section.

Gender

A total of 11 studies60,61,63,64,69,79,81,82,86,87,89 investigated the role of gender in relation to adverse outcomes after EVAR (Figures 13–17).

FIGURE 13.

Female gender and 30-day mortality. Studies to the left of the vertical line found gender to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 14.

Female gender and aneurysm-related mortality. Studies to the left of the vertical line found gender to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 15.

Female gender and all-cause mortality. Studies to the left of the vertical line found gender to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 16.

Female gender and reintervention. Studies to the left of the vertical line found gender to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 17.

Female gender and endoleak. Studies to the left of the vertical line found gender to be an independent risk factor (IRF), whereas those to the right did not.

The results of the very large recent US-based study and the smaller, older EUROSTAR study provide contradictory results regarding the association between female gender and 30-day mortality (Figure 13). However, given the small number of female patients in most series the very large study is likely to be more reliable. Therefore, there may be a link between female gender and 30-day mortality. There is no indication of any link between female gender and aneurysm-related or all-cause mortality (Figures 14 and 15). The evidence suggests that gender is not an independent risk factor for reintervention (Figure 16). There is contradictory evidence regarding association with endoleak (Figure 17).

Pre-existing conditions

In total, 19 studies60,64,65,68,69,72–76,78,79,81–83,88–91 investigated the role of pre-existing conditions in relation to adverse outcomes after EVAR. The studies assessed the role of a range of pre-existing conditions such as pulmonary insufficiency, diabetes, chronic heart failure, obesity, anaemia and hypertension (Figures 18–22).

FIGURE 18.

Pre-existing conditions and 30-day mortality. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 19.

Pre-existing conditions and aneurysm-related mortality. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 20.

Pre-existing conditions and all-cause mortality. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 21.

Pre-existing conditions and reintervention. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 22.

Pre-existing conditions and endoleak. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

The available analyses of EUROSTAR data indicate that cardiac status, high blood pressure and obesity are not independent risk factors for 30-day mortality. The results regarding diabetes and pulmonary impairment as predictors of 30-day mortality are inconsistent (Figure 18).

The analyses for aneurysm-related mortality showed inconsistent results for pulmonary status. However, the evidence suggested that diabetes is not a risk factor for aneurysm-related mortality and based on one US study hypertension was not found to be a risk factor for aneurysm-related mortality. Evidence on other pre-existing conditions was lacking for this outcome (Figure 19).

There were inconsistent results regarding cardiac disease and all-cause mortality after EVAR. The majority of studies found that pulmonary status/chronic obstructive pulmonary disease (COPD) was an independent risk factor for all-cause mortality after EVAR in both the EUROSTAR and the US populations. The findings for diabetes as a risk factor for all-cause mortality were inconsistent. In one EUROSTAR and one US study hypertension was not found to be a risk factor for all-cause mortality. Evidence was lacking on other risk factors (Figure 20).

The available analyses suggest that diabetes is not a risk factor for reintervention/conversion to open repair. One Australian study concluded that the higher the number of pre-existing conditions the greater the rates of reintervention whereas all EUROSTAR studies found that pre-existing conditions did not tend to predict reintervention (Figure 21).

The studies consistently found that pre-existing conditions were not risk factors for endoleak (Figure 22).

Renal function

A total of 11 studies60,61,65,73,78,79,82,83,88,89,91 investigated renal function/renal impairment as a potential risk factor for adverse outcomes in multivariable modelling (Figures 23–27). Although all outcomes were considered, the outcome of re-intervention was only investigated in one study.

FIGURE 23.

Renal function and 30-day mortality. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 24.

Renal function and aneurysm-related mortality. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 25.

Renal function and all-cause mortality. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 26.

Renal function and reintervention. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 27.

Renal function and endoleak. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

There was consistent evidence from a small number of studies that renal impairment affects 30-day mortality after EVAR (Figure 23) but inconsistent evidence of its effects on aneurysm-related mortality (Figure 24). The balance of evidence suggests that renal impairment is an independent risk factor for all-cause mortality (Figure 25). Analyses of EUROSTAR data indicate no link between renal dysfunction and reintervention (Figure 26) or endoleak (Figure 27) after EVAR.

Fitness for open procedure

Six studies investigated whether patients’ fitness for open procedure determined adverse outcomes after EVAR (Figures 28–32). Five of these were based on EUROSTAR data64,65,73,82,87 whereas one was based on a national Australian audit. 60

FIGURE 28.

Fitness for open procedure and 30-day mortality. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 29.

Fitness for open procedure and aneurysm-related mortality. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 30.

Fitness for open procedure and all-cause mortality. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 31.

Fitness for open procedure and reintervention. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 32.

Fitness for open procedure and endoleak. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

There was inconsistent evidence linking fitness and 30-day mortality but the more recent analysis with a larger cohort suggested there might be an association (Figure 28). On balance, analyses indicate that fitness for open procedure is linked to aneurysm-related mortality (Figure 29). Evidence was lacking to link fitness for open procedure and all-cause mortality (Figure 30), and, on balance, fitness was not an independent risk factor for reintervention (Figure 31) or endoleak (Figure 32).

ASA status

In total, 12 studies60,62–65,69,77,81,82,88–90 investigated the role of patients’ ASA status in relation to adverse outcomes after EVAR (Figures 33–37). The majority of the studies were based on EUROSTAR data.

FIGURE 33.

ASA class and 30-day mortality. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 34.

ASA class and aneurysm-related mortality. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 35.

ASA class and all-cause mortality. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 36.

ASA class and reintervention. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 37.

ASA class and endoleak. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

According to EUROSTAR data, ASA classes III and IV are predictive of statistically significantly worse 30-day mortality (Figure 33). Evidence for aneurysm-related mortality was inconsistent (Figure 34). With the exception of a large US study, all analyses found ASA to be an independent risk factor for all-cause mortality (Figure 35). On balance, ASA status was not found to be a significant independent risk factor for reintervention (Figure 36) or endoleak (Figure 37).

Smoking status

Seven studies60,64,66,76,80,81,89 investigated smoking status as a risk factor for adverse outcomes after EVAR.

The evidence suggests that smoking status is not associated with adverse outcomes after EVAR (Figures 38–42). However, the evidence investigating smoking and mortality after EVAR is very limited.

FIGURE 38.

Smoking status and 30-day mortality. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 39.

Smoking status and aneurysm-related mortality. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 40.

Smoking status and all-cause mortality. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 41.

Smoking status and reintervention. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 42.

Smoking status and endoleak. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

Aneurysm size

In total, 19 studies60,61,63,64,67,69,72,73,76,78,79,81,82,84,85,87–90 investigated aneurysm size as a potential risk factor for adverse outcomes in multivariable modelling (Figures 43–47).

FIGURE 43.

Aneurysm size and 30-day mortality. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 44.

Aneurysm size and aneurysm-related mortality. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 45.

Aneurysm size and all-cause mortality. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 46.

Aneurysm size and reintervention. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 47.

Aneurysm size and endoleak. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

More recent and larger cohort analysis demonstrates that aneurysm size is an independent risk factor for 30-day mortality (Figure 43). Evidence also suggests that it is an independent risk factor for aneurysm-related mortality (Figure 44) and all-cause mortality (Figure 45). Evidence for aneurysm size as an independent risk factor for reintervention (Figure 46) and endoleaks (Figure 47) is inconsistent.

Aortic neck and aneurysm angle

Eight studies60,64,70,74,76,81,82,87 investigated aortic neck and aneurysm angle as potential risk factors for adverse outcomes in multivariable modelling (Figures 48–52). With the exception of one Australian study60 all were based on EUROSTAR populations.

FIGURE 48.

Aortic neck/aneurysm angle and 30-day mortality. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 49.

Aortic neck/aneurysm angle and aneurysm-related mortality. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 50.

Aortic neck/aneurysm angle and all-cause mortality. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 51.

Aortic neck/aneurysm angle and reintervention. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 52.

Aortic neck/aneurysm angle and endoleak. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

The balance of evidence suggests no effect of aortic neck and aneurysm angle on 30-day mortality (Figure 48), aneurysm-related mortality (Figure 49) or all-cause mortality (Figure 50). Evidence with regard to reintervention (Figure 51) and endoleak (Figure 52) was mixed allowing no firm conclusions to be drawn.

Aortic neck length

Nine studies60,66,67,74,76,81,82,87,89 investigated aortic neck length as a potential risk factor for adverse outcomes in multivariable modelling (Figures 53–57). With the exception of one Australian study60 all were based on EUROSTAR populations.

FIGURE 53.

Aortic neck length and 30-day mortality. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 54.

Aortic neck length and aneurysm-related mortality. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 55.

Aortic neck length and all-cause mortality. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 56.

Aortic neck length and reintervention. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 57.

Aortic neck length and endoleak. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

There was limited evidence available for 30-day mortality (Figure 53). Evidence with regard to aneurysm-related mortality was inconsistent but none of the EUROSTAR analyses found it to be an independent risk factor (Figure 54). Evidence for all-cause mortality was limited but suggestive of no effect (Figure 55). Evidence regarding reintervention rates was mixed (Figure 56) as was the evidence for endoleak (Figure 57) with possible differences with type of endoleak.

Graft configuration and device type

In total, 10 studies60,61,64,67,69,81,82,84,87,89 investigated the roles of graft configuration and device type in adverse outcomes after EVAR (Figures 58–62).

FIGURE 58.

Graft configuration/device type and 30-day mortality. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 59.

Graft configuration/device type and aneurysm-related mortality. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 60.

Graft configuration/device type and all-cause mortality. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 61.

Graft configuration/device type and reintervention. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

FIGURE 62.

Graft configuration/device type and endoleak. Studies to the left of the vertical line found pre-existing conditions to be an independent risk factor (IRF), whereas those to the right did not.

The evidence regarding graft configuration/device type and 30-day mortality (Figure 58) and all-cause mortality (Figure 60) was too limited to draw conclusions. The balance of evidence suggests that there might be a link between device type and aneurysm-related mortality (Figure 59). Evidence regarding graft configuration/device type and reintervention (Figure 61) and endoleak (Figure 62) was inconsistent.

Summary statements

A large number of studies have modelled the risk of mortality and other adverse outcomes after EVAR. We do not have definitive evidence on all of the risk factors and outcomes explored. The firmest evidence supports the following conclusions.

Overview

This study was designed to determine whether EVAR is a cost-effective alternative to open surgery in the treatment of AAAs. The base case was defined as 70-year-old men with an AAA of 5 cm in diameter. This study was conducted before the publication of trial results for either EVAR trial 1 or DREAM.

The authors developed a Markov decision model to compute lifetime QALYs and costs for a hypothetical cohort of patients who underwent either EVAR or open surgery. In the model, once a patient has undergone a procedure, either EVAR or open surgery, the outcomes include a successful repair or any of a number of complications. Effectiveness, resource use and cost data were derived from the literature. Figure 63 provides a schematic for the Markov model developed by the authors of this study.

FIGURE 63.

Schematic of model from Patel et al. ,105 with permission from Elsevier.

Schematic of model from Patel et al. , with permission from Elsevier.

Summary of effectiveness data

Effectiveness data were derived from the literature with preference given to data derived from large multicentre studies. For open surgery a large Canadian study was used to derive mortality and morbidity rates. For EVAR, because there were no RCTs at the time that this study was published, mortality rates were taken as an average of those in the three largest trials and the occurrence of long-term morbidity was estimated from other sources. It should be noted that their review found EVAR to have lower stroke, myocardial infarction, major amputation and dialysis-dependent renal failure rates than open repair during primary admission. For reinterventions, probabilities for both initial procedures were derived from the literature. The study also considered the immediate and late conversions of EVAR to open surgery.

A quality adjustment factor was assigned for each year of survival of a patient who had a major morbidity, for example a quality adjustment factor of 0.4 was used for a patient who had had a major stroke. Quality adjustment for temporary conditions was achieved through subtracting disutilities from the overall QALY estimate. Although it is not made clear in the study, it would appear that patients who are experiencing no complications are assigned a utility of 1. This would appear inappropriate given the age and general ill health of the patients being considered. QALYs were discounted at a rate of 3% per annum. Table 32 presents some of the key effectiveness parameters used in the model.

Summary of resource utilisation and cost data

The costs were derived from the cost accounting system at New York Presbyterian Hospital, as well as from the literature. For calculating the costs of the open surgery and EVAR procedures, the major resources consumed were identified and the costs calculated based on the average resource use reported in the literature. Fees for surgeons and radiologists were derived from the Medicare reimbursement rates for the appropriate current procedural terminology codes. The immediate and long-term costs of major long-term morbidities, such as stroke, dialysis-dependent renal failure and myocardial infarction, were derived from the literature. For EVAR rigorous postoperative surveillance was also conducted, with CT scanning at 1 week, 3 and 6 months, 1 year and annually thereafter. The study assumed that there was no follow-up surveillance for those patients who underwent open repair. Costs were discounted at a rate of 3% per annum. Table 33 presents values for some of the key cost parameters used in the model.

Summary of cost-effectiveness

For a hypothetical cohort of 70-year-old men with an AAA of 5 cm in diameter, EVAR produced more QALYs than open surgery (7.95 versus 7.53, respectively) at a higher lifetime cost (US$28,901 versus US$19,314). This yielded an ICER of US$22,826 per QALY.

A wide range of sensitivity analyses were also undertaken. It was found that the ICER was sensitive to changes in mortality and morbidity rates of open surgery or EVAR, initial hospitalisation costs of EVAR or open surgery, and the conversion rate of EVAR to open repair during primary procedure. For example, it was found that the mortality rate of open surgery had a large effect on the ICER, such that halving the mortality rate of open surgery from 4.8% to 2.4% (and keeping the operative mortality rate of EVAR constant) increased the ICER to US$43,408 per QALY; similarly, if the mortality rate of the EVAR procedure was doubled from 1.2% to 2.4% (keeping the operative mortality rate of open repair constant) the ICER increased to US$30,064 per QALY.

Table 34 presents ICERs for the base case and some of the sensitivity analyses performed in the study.

Comments

Overview

Bosch et al. 108 performed a cost–utility analysis comparing lifetime costs and QALYs for treatment with EVAR or treatment with open repair. The aim of the study was to evaluate the cost-effectiveness of EVAR compared with open repair.

The authors developed a Markov decision model comparing lifetime costs and QALYs for EVAR and open repair in a cohort of 70-year-old men with AAAs of between 5 and 6 cm in diameter. The clinical effectiveness data for the study were derived from the published literature, which at the time did not include the two largest RCTs (EVAR trial 143 and DREAM40). The authors focused on studies with large patient series and cases of both EVAR and open surgery. Resource use and cost estimates were derived from various sources, which will be discussed further.

Figure 64 provides a schematic of the model used by the authors.

Summary of effectiveness data

Because of the lack of RCTs a meta-analysis of the short-term results of studies comparing patients who underwent EVAR with matched patients who underwent open surgery was undertaken. The meta-analysis allowed calculation of the operative mortality rates for each procedure as well as the rates of complications in the short term (however, it is unclear from the study what time period is considered the short term). The meta-analysis found that the most commonly reported systemic and remote complications at 30 days were cardiac, cerebral, renal and pulmonary. It was assumed that these complications had a long-term effect, which resulted in decreased HRQoL and added long-term costs. They estimated that the probability of systemic/remote complications was considerably lower with EVAR than with open surgery (a probability of 0.13 for EVAR versus 0.32 for open repair). Following the initial treatment period no new systemic complications could occur except when a patient underwent emergent surgical repair. In the long term an annual average rupture rate of 0.01 for EVAR was used, with no rupture after open repair; for long-term reintervention a rate of 0.08 per year was used for EVAR and 0.01 per year for open surgery. Long-term life expectancy was calculated based on age- and sex-specific mortality rates from life tables for the US general population; however, this would seem inappropriate given the general ill health of the patient population being considered. For patients with major systemic complications, survival was adjusted with an excess mortality rate.

Quality of life weights before treatment and after recovery from either treatment were set similar to those in the general population. To show the effect that the treatments had in the short term on quality of life, a 10% reduction in the first month following EVAR was assumed, and a 30% reduction for 2 months following open surgery was assumed. Long-term quality of life adjustments were also made for patients with cardiac, cerebral, renal or pulmonary complications. QALYs were discounted at a rate of 3% per annum.

Table 35 presents values for some of the key parameters used by the authors.

Summary of resource utilisation and cost data

The authors included costs for procedures (including patient time productivity costs), morbidity and mortality, and imaging in follow-up. All costs were converted to year 2000 US dollars. Procedure costs included those of the hospital, physician and patient for EVAR, open surgery, percutaneous treatment and emergent surgical repair of rupture. The hospital cost and physician fees were derived from Medicare reimbursement rates by using diagnosis-related groups. Patient costs were determined by multiplying the daily wage rate by the number of days spent in hospital. It should be noted that when considering patient costs the authors of the study do not appear to have accounted for other sick days that did not involve hospital stays. For costs for morbidity and mortality, if a major systemic/remote complication occurred during surgery then extra costs were added to the procedure costs. Costs of follow-up included physician visit costs, imaging costs and patient costs. In the model, patients who underwent EVAR were imaged at 3, 6 and 12 months, and annually thereafter. All costs were discounted at 3%. Table 36 presents some of the key resource costs used in the model.

Summary of cost-effectiveness

In the base case it was found that EVAR resulted in more QALYs than open repair (6.74 versus 6.52, respectively) and also more costs (US$39,785 versus US$37,606, respectively), resulting in an ICER of US$9905 per QALY. A wide range of sensitivity analyses, including both one- and two-way analyses, were also performed. These found that the results were highly sensitive to the uncertain parameters in the model, such as the systemic complication rate, long-term failure rate and rupture rate. For example, if the annual rate for procedures in follow-up in the EVAR arm was increased from 8% to 12% then the ICER increased to US$56,630 per QALY, and if the rate exceeded 12% then the ICER was more than US$100,000 per QALY. Table 37 presents the ICERs for the base case and for some of the sensitivity analyses performed in the study.

Comments

Overview

This study evaluated the cost-effectiveness of EVAR compared with open repair in patients fit for surgery (RC1) or with conservative management in those unfit for surgery (RC2) (this section of the study will be discussed later in this chapter; see Cost-effectiveness studies focusing on the EVAR type 2 population). The aim of the study was to determine an optimal strategy for the use of EVAR based on the best available evidence at the time.

Effectiveness and resource use data were based on recent RCTs (EVAR trial 142 and DREAM40) as well as on a systematic review of the literature. The study was conducted after the short-term (30-day) operative mortality results were published from these trials but before the mid-term results were available. The authors developed a Markov model and used it to consider two separate ‘reference cases’, one of which was similar to the EVAR trial 1 population. They considered fit 70-year-old patients with an AAA of 5.5 cm in diameter for which the choice of treatment was between EVAR and open surgery (RC1). The primary outcome measure for the cost-effectiveness analysis was the incremental cost per QALY gained. The authors used a 10-year time horizon. The evaluation was undertaken from the perspective of the NHS.

Figure 65 represents the Markov decision model for RC1.

FIGURE 65.

Schematic of model from Michaels et al. 107 © British Journal of Surgery Society Ltd. Reproduced with permission, granted by John Wiley & Sons on behalf of the BJSS Ltd.

© British Journal of Surgery Society Ltd. Reproduced with permission, granted by John Wiley & Sons on behalf of the BJSS Ltd.

Summary of effectiveness data

Short-term operative mortality probabilities were taken from the EVAR trial 142 and DREAM trial. 40 The probabilities of reintervention and complications were derived from a previously conducted systematic review. General mortality was taken from standardised mortality tables for England and Wales (it should be noted that it is not stated whether these have been adjusted for the poorer health of patients with aneurysms). Aneurysm-related mortality was calculated from a previous modelling study.

Utility estimates were based on published figures derived from the EQ-5D tariff values for men aged 65–74 years. To account for the lower HRQoL initially following surgery, a reduction in keeping with that seen after major surgery was applied for the first 4 weeks after open surgery and for the first 2 weeks after EVAR. QALYs were discounted at a rate of 3.5% per annum.

The key effectiveness parameters for the model are reported in Table 38.

Summary of resource utilisation and cost data

Most costs were based on NHS reference costs for 2003–4123 with the mean cost being the point estimate. For the probabilistic sensitivity analysis a normal distribution was assumed with standard deviation based on the assumption that 50% of observations were within the published interquartile range. The additional incremental cost of EVAR was estimated from data collected at the Sheffield Teaching Hospital NHS Trust. Follow-up costs for EVAR were based on NHS reference costs with the assumption that on average an EVAR patient will have two outpatient visits and two CT scans per year. After open repair the average cost of a reintervention in the EVAR arm again used NHS reference costs123 but was based on the case mix of reinterventions as recorded in the EUROSTAR registry. 124 All costs have been discounted at a rate of 3.5% per annum. The key resource cost parameters for the model are reported in Table 39.

Summary of cost-effectiveness

Comments

Overview

The study evaluated the cost-effectiveness of EVAR compared with open surgery in a patient population of 74-year-old men with a diagnosed AAA of diameter ≥ 5.5 cm. It should be noted that several of the authors of this report were authors of this study.

The authors constructed a Markov decision model to estimate the lifetime costs and QALYs of male patients aged 74 years with an AAA of diameter ≥ 5.5 cm. Effectiveness and resource use data used to populate the Markov model were largely drawn from an RCT, EVAR trial 1. The model includes the risks of death from aneurysm and other cardiovascular and non-cardiovascular causes, secondary reinterventions and non-fatal cardiovascular events.

ICERs were reported for the base case as well as for a number of sensitivity analyses (e.g. for different starting ages). The probability that EVAR is cost-effective at a threshold of £20,000 per QALY and £40,000 per QALY was also reported, based on probabilistic sensitivity analyses.

Figure 66 provides a schematic of the model used in the study.

FIGURE 66.

Schematic of model from Epstein et al. 106 © British Journal of Surgery Society Ltd. Reproduced with permission, granted by John Wiley & Sons on behalf of the BJSS Ltd.

© British Journal of Surgery Society Ltd. Reproduced with permission, granted by John Wiley & Sons on behalf of the BJSS Ltd.

Summary of effectiveness data

The effectiveness data were largely taken from EVAR trial 1 although this has been supplemented by other data sources. Mortality from the initial procedure was calculated from EVAR trial 1. It was assumed that if an EVAR patient converted to open repair during the primary admission then they would have the same long-term prognosis as an individual who had originally been allocated to EVAR. Mortality rates after the initial admission were estimated as three competing risks: (1) death from an AAA cause, (2) death from a cardiovascular cause other than AAA and (3) death from a non-cardiovascular cause. Patients were also at risk of a non-fatal cardiovascular event or a readmission for a second AAA procedure, all of which were associated with higher costs and lower utilities. The model assumed that the initial operative mortality benefit of EVAR compared with open repair was eroded after 2 years by additional deaths from cardiovascular causes after EVAR, based on the results of EVAR trial 1 and the DREAM trial showing that there was no difference in mid-term survival between the treatments. The model also assumed that there would be a small but persistent difference in late aneurysm-related deaths between the treatments.

It has been assumed that the baseline utility of these patients is the same as that of the age-specific UK general population estimates. There is an initial loss of utility for 1 month post surgery with open repair resulting in a larger loss (a reduction of 0.094 compared with 0.027 for EVAR). 43 There is also a 1-month loss of utility for a non-disabling stroke or myocardial infarction, and a permanent utility decrease following a disabling stroke. In the base case all QALYs were discounted at a rate of 3.5% per annum.

Table 41 provides a summary of some of the key effectiveness parameters used in the model.

Summary of resource utilisation and cost data

Resource utilisation and cost data for the initial EVAR or open repair surgery, a conversion to open repair during primary EVAR, and a secondary readmission for an AAA have all been taken from the EVAR trial. Costs for non-fatal cardiovascular events have been taken from Jones et al. 125

All patients in the EVAR group have been assumed to require hospital outpatient attendances and CT to monitor their aneurysm repair. In the base case it was assumed that two surveillance visits would be required in the first year and then one annually thereafter. The costs for these visits and scans have been taken from NHS reference costs. 123 In the base case all costs were discounted at a rate of 3.5% per annum. Table 42 summarises some of the key cost parameters used by the authors in the model.

Summary of cost-effectiveness

In the base case EVAR was more costly than open repair by £3800 per patient but also produced fewer lifetime QALYs than open repair (mean –0.020 QALYs). Therefore, under the base-case assumptions EVAR was dominated by open repair.

The base-case assumptions were varied in a series of secondary analyses to reflect alternative evidence and opinions about some of the key parameters in the model. In only one case was the EVAR ICER found to be under £30,000 per QALY. This occurred when the age of the initial cohort was increased from 74 to 82 years (with a greater absolute difference in operative mortality between the treatments) and the lower long-term rate of cardiovascular death after open surgery was replaced with the assumption that there is no difference in the rate of cardiovascular death after open repair or EVAR.

ICERs for the base case and some of the sensitivity analyses conducted are presented in Table 43.

Comments

Overview

The authors conducted a cost-effectiveness analysis of a multicentre randomised trial of EVAR compared with open repair in patients with AAAs of ≥5 cm in diameter. The analysis is conducted for up to 1 year after the original procedure. All of the effectiveness and resource use data were taken from the DREAM trial (therefore relevant data from EVAR trial 1 has been excluded).

Summary of effectiveness data

HRQoL was assessed using the EQ-5D questionnaire. Questionnaires were filled in by the trial patients at baseline (upon randomisation) and at 3 and 6 weeks and 3, 6 and 12 months postoperatively. By using linear interpolation for periods between measurements, quality-adjusted survival time was calculated up to 1 year after inclusion. A small and non-significant benefit of open repair compared with EVAR was found. Table 44 summarises the QALY outcomes from the two trial arms over a 1-year period.

Summary of resource utilisation and cost data

Costs associated with treatment and follow-up until 1 year after inclusion were calculated by multiplying individual patient resource use recorded in the trial by unit costs. All costs were calculated in 2003 euros.

The costs of lost productivity were also calculated. These took account of sick leave and travel, as well as other costs incurred by the patients and their families. Table 45 summarises the average total cost in each trial arm based on a bootstrap estimate.

Summary of cost-effectiveness

The authors found that patients in the EVAR group experienced less QALYs than those in the open repair group (0.72 QALYs versus 0.73, respectively) whilst incurring more costs (€18,179 versus €13,886, respectively). Thus, EVAR was dominated by open repair with a 1-year time horizon.

The authors also conducted a non-parametric bootstrapping approach to evaluate the joint uncertainty in outcomes and costs.

Table 46 presents the key cost-effectiveness results for the study.

Comments

Overview

In this study the authors conducted a cost–utility analysis comparing EVAR with open repair. The patient population considered was that of EVAR trial 1, i.e. patients with an unruptured infrarenal AAA of at least 5.5 cm in diameter who are considered fit for open surgery. The average age of the population was 70 years and 90% of patients were men.

The authors developed a two-stage model to estimate the lifetime costs and QALYs of EVAR and open repair in this patient population: first, a decision tree for the first 30 days post surgery; second, a Markov model from 30 days post surgery until death.

Figure 67 represents the short-term decision tree for the first 30 days post surgery. At the end of the first 30 days patients in the EVAR arm will end up in one of four states: (1) successful EVAR with no complications, (2) EVAR with complications, (3) conversion to open surgery or (4) death. Conversely, those in the open repair arm initially may end up in one of three states: (1) open repair with no complications, (2) open repair with complications or (3) death.

FIGURE 67.

Schematic of decision tree from Medtronic submission,127 reproduced with permission from Medtronic.

Schematic of decision tree from Medtronic submission, reproduced with permission from Medtronic.

Once patients enter the 30 days post surgery Markov model (Figure 68) then they must be in one of four health states: (1) no complications requiring secondary intervention, (2) technical complications requiring secondary intervention, (3) systemic complications (split into a first-year phase and then subsequent years phase) or (4) death.

FIGURE 68.

Schematic of Markov model from Medtronic submission,127 reproduced with permission from Medtronic.

Schematic of decision tree from Medtronic submission, reproduced with permission from Medtronic.

Summary of effectiveness data

The effectiveness data used to parameterise the model were largely drawn from EVAR trial 143 but were supplemented with data from additional sources.

For the short-term model, mortality estimates and the need for secondary intervention were based on data from EVAR trial 1. However, the risk of conversion from EVAR to open surgery was based on clinical expert opinion rather than on the trial (the authors used a probability of conversion of 0.2%, which is much lower than that used in other studies, e.g. Epstein et al. ,106 in which the probability was four times larger at 0.8%, taken from EVAR trial 143).

The baseline risk of systemic complications (myocardial infarction, temporary and permanent renal failure, and disabling and non-disabling stroke) for EVAR patients in the first 30 days was estimated from the EUROSTAR data set. The relative risk of systemic complications for open surgery versus EVAR was taken from a meta-analysis of observational studies and one RCT (DREAM). The authors have assumed that the incidence rates for systemic complications follow the same pattern as for all-cause mortality, i.e. that open repair patients have a higher incidence in the first 30 days post surgery whereas EVAR patients have a higher incidence from 30 days to 18 months post surgery. Therefore, it has been assumed that over the first 18 months the number of events that occur in the two groups is equal. The authors have achieved this in the model by using the incidence rate of myocardial infarction and stroke in the general UK population for the open repair group from 30 days to 18 months. Then the relative risk for EVAR was calculated such that the number of events was equal at 18 months. However, it should be noted that the numbers of events in each arm were only equal for myocardial infarction and stroke and not for renal failure. This was considered to be closely related to the intervention itself and therefore could only occur in the first 30 days, hence there was a higher prevalence of renal failure in the open repair arm. The authors have then assumed that no new systemic complications occur from 18 months onwards.

Long-term risks of mortality and secondary interventions were also based on data from EVAR trial 1. 43 The authors considered two scenarios for estimating the difference in late mortality (after 30 days). First, they considered a relative risk for EVAR compared with open repair for late mortality (for any cause) of 1.055, applied for 4 years. The authors stated that this relative risk was calculated from EVAR trial 1 but it was not clear exactly how this was carried out. In the second scenario the difference in mortality between the two treatments is only due to AAA mortality, and the authors estimate a relative risk of 1.18, again based on EVAR trial 1 results for late aneurysm mortality. It should be noted that EVAR trial 1 reported HRs (EVAR relative to open surgery) for aneurysm-related mortality of 0.42 (95% CI 0.21 to 0.82) for the first 6 months and 1.15 (95% CI 0.39 to 3.41) from 6 months to 4 years; the corresponding HRs for total mortality in the two periods were 0.55 (95% CI 0.33 to 0.93) and 1.10 (95% CI 0.80 to 1.52) respectively. The authors assumed that from 1 month post surgery until 4 years the patients experience this as a constant relative risk of death. At 4 years it has been assumed that patients in both arms experience a risk of death that is similar to that of the background UK population with adjustments for increased incidence of cardiovascular death in an AAA population after surgery.

The risk of patients requiring a secondary intervention was derived from EVAR trial 1 and then supplemented with data from other sources. The total number of secondary interventions was taken from the EVAR trial; the secondary intervention rate was 1.72% per month from 2 to 6 months post surgery in the EVAR arm and 1.03% in the open repair arm, whereas for post 6 months the rate in the EVAR arm was 0.27% and there were no secondary interventions post 6 months in the open repair arm. The percentages of these secondary interventions that were transabdominal, extra-anatomic or transfemoral has then been derived from other sources, notably EUROSTAR for the EVAR group and expert clinical opinion for the open repair group. Both patient groups have a constant risk of secondary intervention from the operation until 6 months post operation. Post 6 months it is assumed that open repair patients are no longer at risk of secondary interventions and that the rate of reintervention for EVAR patients remains constant over time. The authors have also assumed that patients do not experience disutility from secondary interventions and that they have the same prognosis as other patients after the reintervention.

Utility scores for health states have been taken directly from EVAR trial 1. In the first 3 months post surgery, those in the open repair arm had a lower utility than those in the EVAR arm (0.67 versus 0.73). From 24 months onwards it was assumed that utility was equal in both arms (although it was age dependent). Disutility scores for the systemic complications have been drawn from several sources.

Table 47 presents the values used for some of the key parameters in the model.

Summary of resource utilisation and cost data

(Commercial-in-confidence information has been removed.)

The authors also assumed that 50% of follow-up scans are now duplex ultrasound and 50% are CT, with the same frequency of monitoring as in the EVAR trial 1 protocol. As duplex ultrasound is cheaper than CT this reduced the overall cost of monitoring in the EVAR arm.

The costs for secondary interventions were drawn from NHS reference costs. 123 The costs for the same type of intervention (e.g. transabdominal intervention) are assumed to be the same for each treatment arm. However, the percentage of each type of intervention as a proportion of the total number of secondary interventions differs between the two treatments, with open repair having the highest proportion of the most costly procedures, making the average cost per secondary intervention higher in the open repair group. These percentages are not based on trial evidence but instead on the EUROSTAR registry in the case of the EVAR arm and on clinical opinion in the case of the open surgery arm. Table 48 presents some of the values used for key cost parameters in the model.

Summary of cost-effectiveness

The authors found that in the base case patients treated with EVAR were expected to receive more QALYs than those treated with open surgery but at a higher cost (commercial-in-confidence information has been removed). This resulted in an ICER of £15,681 per QALY for EVAR compared with open repair.

The authors also conducted univariate sensitivity analyses for all of the parameters in the model, using the values for the lower and upper confidence limits of each parameter. They found that the ICER was most sensitive to the short-term relative risk of operative mortality.

Table 49 presents ICERs for the base case and the sensitivity analysis in which the short-term relative risk of mortality was varied.

Comments

Overview

The authors developed a decision-analytic model to evaluate the costs and QALYs associated with EVAR and open surgical repair. The model relates to 70-year-old male patients with AAAs of 5.5 cm diameter who are considered medically suitable to undergo either open surgical repair or EVAR over a period of 13 months. The first 30 days are modelled using a decision tree and the following 12 months are modelled using a Markov decision model. The Markov models for the following 12 months for those patients who received EVAR and those patients undergoing open surgical repair were similar, with conversion to open surgical repair and the endoleak states removed.

Summary of effectiveness data

Several sources of data were used to parameterise the model, including data from a non-randomised field evaluation conducted by the authors and results from a systematic literature review. Most of the probabilities used in the model were derived from the literature review and meta-analyses. These included the EVAR 1 and DREAM trials as well as other non-randomised sources. It should be noted that it is not clear which sources were used for the meta-analysis for probability of death. All-cause mortality was derived from the life tables of Statistics Canada. The utility values assigned to different states in the model were based on adjusted estimates of the EQ-5D scores reported in the field evaluation conducted by the authors.

Table 50 presents some of the key effectiveness parameters from the study.

Summary of resource utilisation and cost data

Data on costs and resource use were derived from a variety of sources. The costs of the initial hospitalisation for each treatment arm were derived from the costs observed in the field study conducted by the authors. This field study was a non-randomised prospective study that aimed to compare EVAR patients at high risk for open surgical repair with patients receiving open surgical repair with either low or high surgical risk. The study prospectively collected clinical outcomes, resource utilisation data and quality of life information.

Costs of major complications during initial hospitalisation (e.g. myocardial infarction, stroke, etc.) were based on mean hospital costs for each condition found in the Ontario Case Costing Initiative database. The 1-year follow-up costs for EVAR and open surgical repair were also derived from the field study. Costs were also derived from the literature.

The values for some key resource cost parameters are reported in Table 51.

Summary of cost-effectiveness

In the base case EVAR was both more costly than open surgical repair (C$32,079 versus C$17,503) and more effective (0.863 QALYs versus 0.772 QALYs) over a period of 13 months. As such the incremental cost per QALY was found by the authors to be C$160,176 (Table 52). Because of concerns that the authors had over some of the non-randomised results used from the systematic review to parameterise the model, particularly with regards to complications, the authors performed a secondary analysis in which complication costs and rates were taken from the field study. Because of the low number of complications observed in the EVAR arm of the field study the ICER was lower than that found in the base case (C$22,528 per QALY versus C$160,176 per QALY; Table 52).

Comments

Overview

A cost-effectiveness analysis was performed based upon a field evaluation conducted by the authors of this report. The analysis focused on high-risk patients and had a time horizon of 1 year. As this study only considers data from the authors’ non-randomized field study and does not incorporate other evidence (e.g. from the available RCTs), its relevance to the UK is considered to be limited. Therefore, only a brief review of the study is presented here.

Summary of effectiveness data

The authors estimated the life-years gained over 1 year using results from the field study and Kaplan–Meier survival curves. The life-years were then converted into QALYs by combining the survival curves with utility estimates, also derived from the field study, over time.

The authors also conducted sensitivity analysis on the time horizon considered in the study using several assumptions regarding the extrapolation of survival curves (no convergence, convergence after 10 years, convergence after 5 years and convergence after 3 years).

Summary of resource utilisation and cost data

The authors collected resource use in their field study over 1 year and used these data to inform the costs. The authors also calculated the costs due to productivity losses.

Summary of cost-effectiveness

In the analysis over 1 year the authors found that for high-risk patients EVAR was both less costly and more effective than open surgical repair and as such EVAR dominated open surgical repair.

When a longer time horizon was specified EVAR does result in more costs than open surgical repair under all four assumptions regarding the survival curves. However, even when there is convergence after 3 years the incremental cost per QALY is still only $18,616 and as such EVAR still appears cost-effective.

Comments

Overview

This study evaluated the cost-effectiveness of EVAR compared with open repair in patients fit for surgery (RC1) or with conservative management in those unfit for surgery (RC2). The aim of the study was to determine an optimal strategy for the use of EVAR based on the best available evidence at the time. The study was published before the results of the EVAR trial 2 were available.

Effectiveness and resource use data were based on recent RCTs (EVAR and DREAM) as well as on a systematic review of the literature. The authors developed a Markov model and used it to consider two separate ‘reference cases’, one of which, RC1, was discussed in the previous section. In this section we will consider their modelling of 80-year-old patients with an AAA of 6.5 cm diameter who were considered unfit for open surgery and for whom the choice of treatment was between EVAR and conservative management (RC2). The primary outcome measure for the cost-effectiveness analysis was the incremental cost per QALY gained. The authors used a 10-year time horizon. The evaluation was undertaken from the perspective of the NHS.

The model for RC2 is similar to that of RC1 (Figure 65) with the surgery arm being replaced by a conservative management arm. Conservative management in this model excludes the option for elective surgery.

Summary of effectiveness data

Short-term operative mortality probabilities were taken from the EVAR42 and DREAM40 trials. However, it should be noted that these trials were conducted in patient populations who were considered fit for open repair and thus the mortality probability of EVAR found might not be applicable to the less healthy patient population considered here. The probabilities of reintervention and complications were derived from a previously conducted systematic review. General mortality was taken from standardised mortality tables for England and Wales. Aneurysm-related mortality was calculated from a previous modelling study.

Rupture rates for conservative management were based on three published studies130–132 and are a function of aneurysm size. Expansion rates were also taken from other studies. 132–140 It should be noted that these studies are dated (all were published before 1992) and may not accurately reflect the current natural history of untreated aneurysm.

Utility estimates were based on published figures derived from the EQ-5D tariff values for men aged 65–74 years. To account for the lower HRQoL initially following surgery, a reduction in keeping with that seen after major surgery was applied for the first 2 weeks after EVAR. QALYs were discounted at a rate of 3.5% per annum.

The key effectiveness parameters for the model are reported in Table 54.

Summary of resource utilisation and cost data

Most costs were based on NHS reference costs for 2003–4123 with the mean cost being the point estimate. For the probabilistic sensitivity analysis a normal distribution was assumed with standard deviation based on the assumption that 50% of observations were within the published interquartile range. The procedure cost of EVAR was assumed to be the average national NHS reference cost for open surgery plus an additional incremental cost of EVAR estimated from data collected at the Sheffield Teaching Hospital NHS Trust. Follow-up costs for EVAR were based on NHS reference costs with the assumption that on average an EVAR patient will have two outpatient visits and two CT scans per year. It is not clear from the published paper if patients in the no surgery arm received continuing surveillance. The average cost of a reintervention in the EVAR arm again used NHS reference costs but was based on the case mix of reinterventions as recorded in the EUROSTAR registry. 124 All costs have been discounted at a rate of 3.5% per annum. Table 55 summarises the key resource cost parameters from the study.

Summary of cost-effectiveness

Comments

Overview

EVAR trial 2 investigated whether EVAR improved survival compared with no intervention in patients who were considered unfit for open repair. Although it was not explicitly a cost-effectiveness study, we review it in this section because the study reported life expectancy and costs, and there have been no other cost-effectiveness analyses published in the light of the results of this trial. The mean age of patients in the EVAR arm was slightly higher than in the no intervention arm (76.8 years versus 76.0 years, respectively). The mean AAA diameter was also marginally larger in the EVAR arm than in the no intervention arm (6.4 cm versus 6.3 cm, respectively).

In the trial patients were followed up over 4 years and data on mortality, HRQoL (measured by the EQ-5D and the SF-36) and resource use were collected over this period.

Summary of effectiveness data

The EVAR trial 2 found that the 30-day operative mortality rate for the EVAR group was 9%. The no intervention group was found to have a rupture rate of 9.0 per 100 person-years. By the end of the 4 years, overall mortality was around 64% and this did not significantly differ between the two trial arms. The trial also found no significant difference in aneurysm-related mortality between the two trial arms.

HRQoL data was collected from the EVAR arm patients at 1, 3 and 12 months after the operation, whereas for the no intervention arm it was collected from the patients at 2, 4 and 13 months after randomisation (this was based on the assumption that it would take 1 month following randomisation for the EVAR procedure to be performed). No clear and consistent differences in HRQoL between the two trial arms were found.

Summary of resource utilisation and cost data

Resource use and cost estimations were calculated using the same methods as those used in EVAR trial 1 (e.g. data on resource use were collected using case report forms, which were then multiplied by unit costs to calculate total costs). Resources considered included, among others, initial procedure resource use, hospital stay, secondary AAA procedures, outpatient visits and surveillance using CT.

The study found that the EVAR arm had considerably greater mean hospital costs per patient than the no intervention arm (£13,632 versus £4983 respectively).

Summary of cost-effectiveness

The study found that EVAR did not improve HRQoL over the follow-up period, had a high 30-day operative mortality rate, had no 4-year survival benefit and had considerably higher costs than in the no intervention arm. Therefore, in the patient group considered (patients of around 66 years of age with an AAA of approximately 6.5 cm in diameter) it appeared that EVAR may be dominated by the no intervention arm (i.e. EVAR has higher costs and worse outcomes).

Comments

Overview

The model compares a strategy of open repair with that of EVAR for patients with a diagnosed AAA of at least 5.5 cm in diameter and considered fit for open repair. The perspective of the model is that of the UK NHS. The measure of health benefit is expected QALYs over the patient’s lifetime. The price year is 2007 and all costs are measured in UK pounds. Costs and health benefits in future years were discounted at a rate of 3.5% per year. 15 The model is closely based on a previously published model undertaken by some of the assessment team . 106 The main difference is that this model extends the analysis for patients of different ages, fitness levels and aneurysm sizes at the time of the decision to undertake surgery. 143 The base-case model assumed that these factors influenced baseline risks but that the effect of treatment on operative mortality (OR of EVAR versus open repair) was constant for all patient groups.

The analysis seeks to provide estimates of the cost-effectiveness of management options for all patients in the relevant AAA populations. However, it should be emphasised that most RCT and registry data on EVAR relate to men (see Chapter 3). The cost-effectiveness of EVAR versus open repair in women is explored in a secondary analysis, given the limited data available. Furthermore, untreated rupture rates may differ between men and women and the implications of this are discussed in the model comparing surgery with watchful waiting.

Model structure

The model starts after the decision to operate has been made. The model structure is shown in Figure 69. Patients enter the model and have a primary aneurysm repair procedure (i.e. either EVAR or open repair). Following this, patients may die, convert to open repair or survive the procedure. Survivors pass into a Markov cohort model to estimate lifetime costs and QALYs. It has been assumed that patients who convert from EVAR to open repair during the primary admission have the same long-term prognosis as patients initially undergoing open repair. Unlike the model shown in Epstein et al. ,106 this model does not estimate the incidence of cardiovascular complications such as stroke and myocardial infarction, as the clinical review (Chapter 3) found no evidence that the incidence of these events differed between treatments in the short or long term.

FIGURE 69.

Model structure, once a decision to treat has been made.

Parameter estimation

Estimation of the rate of non-aneurysm-related deaths more than 30 days after aneurysm repair: equation 2

The rate of non-aneurysm-related deaths in the model after more than 30 days, h_Other(t), was in turn constructed from the product of three components: the rate of non-aneurysm-related deaths in the general population, h₀(t), multiplied by the relative risk in patients with a large AAA after aneurysm repair, HR_{LargeAneurysm}, multiplied by the relative risk after an EVAR procedure compared with open repair, HR_EVAR(t). Formally, this can be expressed as:

h Other ( t ) = h 0 ( t ) × HR LargeAneurysm × HR EVAR ( t )

These three components of non-aneurysm-related death after surgery in the model are illustrated in Figure 70. Mortality rates in the general population (h₀) were estimated from life tables,144 adjusting for aneurysm mortality. 145 The parameter HR_{LargeAneurysm} can be thought of as representing the general prognosis for survival free from non-aneurysm-related death after aneurysm repair for a person of that fitness and aneurysm size relative to the general population of that age. The review of risk factors in Chapter 3 found that aneurysm size at the time of the procedure was predictive of the probability of long-term survival after EVAR. This is thought to be primarily cardiovascular risk. Brady et al. 146 found a strong association between aneurysm diameter and the risk of non-aneurysm-related cardiovascular mortality after aneurysm repair in the UK Small Aneurysm Trial and Study: the relative risk of cardiovascular death increased by 31% for each standard deviation increase in aneurysm diameter on a log scale (about 0.8 cm on the natural scale), after adjusting for other risk factors. We estimated the relationship between risk factors and non-aneurysm-related deaths using a Cox survival regression based on the EUROSTAR data set, censoring on AAA deaths. The results of this analysis are shown in Table 60.

FIGURE 70.

Rates of mortality from non-aneurysm-related causes after EVAR and open repair used in the base-case model for a patient aged 75 years with an AAA of 5.5 cm and of poor fitness at baseline.

TABLE 60 - Results of Cox survival analysis of the rate of non-aneurysm-related deaths more than 30 days following aneurysm repair, from the EUROSTAR data 1994–2006

The hazard ratio is the exponential of the coefficient. The hazard ratio for age is not shown because this is included in the Cox analysis only as an adjustment factor. The model calculates relative risks of non-aneurysm-related mortality for patients with relevant risk factors compared with mortality rates in the general population of a given age.

	Coefficient	SE (coefficient)	Hazard ratioa
Per year of age over 74 years	0.043	0.004	–
Unfit for open surgery	0.396	0.076	1.49
AAA 5.1–5.4 cm	0.185	0.112	1.20
AAA 5.5–5.9 cm	0.290	0.113	1.34
AAA 6–6.4 cm	0.429	0.116	1.54
AAA 6.5+ cm	0.565	0.108	1.76
Older generation	0.141	0.070	1.15
Pulmonary condition	0.250	0.067	1.28
ASA III or IV	0.334	0.070	1.40
Renal condition	0.332	0.076	1.39

In the decision model, patients with small (< 5 cm) aneurysms and no other risk factors were assumed to have the same risk of non-aneurysm-related mortality as the general population of the same age and gender. As shown by the HRs in Table 60, a patient with a large aneurysm at surgery (5.5 cm) would expect a rate of non-aneurysm-related death 34% greater than that in the general population of the same age; this value would be 76% greater if the aneurysm were ≥ 6.5 cm at surgery.

The clinical review in Chapter 3 found that there were several factors, such as renal insufficiency and ASA class, that were strongly associated with both operative death and long-term survival. The risk modelling shown in Tables 58 and 60 confirms these findings and furthermore finds that these factors are associated with late non-aneurysm-related deaths. This correlation between factors predictive of operative death and late non-aneurysm-related mortality lends support to the hypothesis that open repair is precipitating deaths in the most risky patients. As described earlier in this chapter (see Introduction), here we aimed to define fitness in a general way, rather than specifying results for every possible risk factor and combination of factors. However, we need to include the correlation between early and late mortality in the model in order to estimate life expectancy for a patient of a given operative fitness. The best way to estimate this correlation would be to calculate the risk of late non-aneurysm-related mortality associated with each level of a validated and generally accepted operative risk scoring system. As we do not have such a risk scoring system, we illustrate the model for groups with different levels of operative fitness as follows. We consider patients with renal insufficiency to represent a moderate fitness group, with about twice the odds of operative mortality (OR 1.97, Table 58) and a 40% greater risk of late non-aneurysm-related mortality (HR 1.39, Table 60), and patients with both renal insufficiency and ASA class III or IV to represent a group with poor fitness, with almost four times the odds of operative mortality (1.97 · 2.02 = 3.99, Table 58) and almost double the risk of late non-aneurysm-related mortality (1.39 · 1.40 = 1.95, Table 60). Further work will be needed to confirm these estimates of the correlation between early and late mortality in different populations using validated risk scoring systems.

Rate of convergence of survival curves for non-aneurysm-related mortality after EVAR and open repair (HR_EVAR)

Given that the EVAR trial 1 and DREAM trial found that the early survival advantage after EVAR was not maintained over the medium term (Chapter 3) it is necessary to estimate the rate of convergence of the survival curves after the primary admission (parameter HR_EVAR, see Figure 70). A large US matched-cohort study96 (see Chapter 3) found that both the initial difference in operative mortality of EVAR compared with open repair and the time taken for the survival curves to meet strongly depended on age. Younger patients (67–74 years) had an absolute difference in operative mortality of less than 2.5% but the proportion surviving at 18 months was the same after EVAR and open repair. On the other hand, 85-year-olds had an absolute reduction of 8.5% in operative mortality and the difference in survival was maintained between the groups until 4 years (Figure 71). These results suggest that, even though the process causing the survival curves to converge might be unknown, the phenomenon is observed in all patient fitness groups. Furthermore, the benefit of EVAR is prolonged in those patient groups with the greatest difference in operative mortality.

FIGURE 71.

Survival of patients undergoing EVAR or open repair of AAA, overall and according to age. From Schermerhorn et al. ,96 with permission from the Massachusetts Medical Society.

From Schermerhorn et al. , with permission from the Massachusetts Medical Society.

The EVAR trial 1 divided the follow-up into the first 6 months after randomisation and the period from 6 months (to allow for delays between randomisation and surgery) and calculated all-cause mortality HRs for the two periods to be 0.55 (95% CI 0.33 to 0.93) and 1.10 (95% CI 0.80 to 1.52) respectively (see Chapter 3). However, the HR for 6 months onwards is not directly useable in the model as we need the HR for non-aneurysm-related deaths occurring more than 30 days after the procedure. In the intention to treat analysis the EVAR trial 1 found 81 deaths (seven aneurysm related) occurring more than 30 days after the procedure in patients randomised to EVAR and 71 deaths (two aneurysm related) occurring more than 30 days after the procedure in those randomised to open repair. 43 In the base-case model we assume an HR of late non-aneurysm-related death of 1.072 (74 versus 69 deaths) as an estimate of HR_EVAR(t), given that the number of patients and mean length of follow-up in the groups were similar. This is assumed to apply to the EVAR group until the non-aneurysm-related survival curves converge, and for non-aneurysm-related deaths to be the same in both arms thereafter [HR_EVAR(t) = 1]. Sensitivity analysis explored other scenarios.

Estimation of the rate of late aneurysm-related death: equation 3

Hazard ratio for treatment effect for late aneurysm-related deaths

Chapter 3 found that the difference in aneurysm-related death between EVAR and open repair is maintained up to 4 years. Even if the rate of late aneurysm death is low, it is important to include it in the model if it is thought that there might be a persistent difference between the rates after EVAR and open repair. The HR (EVAR relative to open repair) for aneurysm-related mortality 6 months or more after randomisation estimated by the EVAR trial 1 was 1.15 (95% CI 0.39 to 3.41),43 with a wide confidence interval because of the few deaths included (see Chapter 3). However, this HR would seem to underestimate the difference in observed aneurysm-related deaths occurring more than 30 days after the primary procedure (seven after EVAR versus two after open repair in about 1250 patient-years of follow-up in each arm), perhaps because some of the deaths occurred in the first 6 months. For the base-case value in the model we estimated the HR (EVAR relative to open repair) from EVAR trial 1 for late aneurysm-related deaths occurring more than 30 days after the primary procedure, censoring on other causes of death. This was estimated to be 2.46 (95% CI 0.48 to 12.7). 106 However, this is not a randomised comparison because of different lengths of time from randomisation to surgery in the two arms. On the basis of clinical opinion we used a HR of 1.5 for the base case. Sensitivity analysis explored other estimates.

Baseline rate of late aneurysm-related deaths occurring more than 30 days after EVAR

Chapter 3 found that baseline aneurysm size was associated with aneurysm-related death after EVAR in most studies. We estimated the baseline rate of aneurysm-related death after EVAR occurring after 30 days using the EUROSTAR data set for patients enrolled between January 1994 and November 2006, censoring on other causes of deaths. 54 Table 61 shows the data stratified by the date of enrolment and AAA diameter at enrolment. The mean rate of late aneurysm-related death for recent patients was 0.4% per year in patients with large AAA (5.5–6.4 cm) and 1.2% per year in patients with very large AAA (> 6.5 cm). These rates are lower than those found in an earlier published analysis of the EUROSTAR data,82 but confirm the earlier finding that rates of late aneurysm-related mortality after EVAR are strongly associated with aneurysm size at the time of the procedure (see Chapter 3). The higher rate in earlier enrolments might indicate improvement in devices and procedures but could also arise because patients were followed up for a longer period, with more time for the aneurysm to expand, or because of more cautious patient selection.

TABLE 61 - Late aneurysm-related deaths, excluding operative mortality, in patients undergoing EVAR. EUROSTAR data stratified by enrolment date and baseline AAA diameter

Enrolment date	AAA size	n	Deaths by November 2006	Patient-years at risk	Mean follow-up (years) per patient	Mean rate per year
Before 1 January 2000	< 5.5 cm	1200	24	4753	3.96	0.5%
	5.5–6.4 cm	786	34	2977	3.79	1.1%
	≥ 6.5 cm	435	30	1410	3.24	2.1%
After 1 January 2000	< 5.5 cm	2296	10	4296	1.87	0.2%
	5.5–6.4 cm	2211	16	4116	1.86	0.4%
	≥ 6.5 cm	1340	28	2311	1.73	1.2%

It is uncertain whether, for any given patient, the risk of late aneurysm-related death is constant, increasing or decreasing with time from surgery. Peppelenbosch et al. 82 estimated that the risk of late aneurysm-related death tended to increase with time from surgery, using EUROSTAR data for patients enrolled from 1996 to 2002. For patients with large aneurysm (5.5–6.5 cm) the rate of late aneurysm-related death was 0.3% in the first 3 years rising to 2.1% after 4 years. For patients with very large aneurysm (> 6.5 cm) the rate was 1% in the first 3 years rising to 8% in the fourth year.

This apparent increase in the risk of death with time from EVAR may be confounded by evolution of devices and surgical technique, as those patients with the longest follow-up underwent EVAR with the oldest devices. We tried to adjust for this by estimating parametric survival models, including a variable representing the year that the device was fitted. Table 62 shows the results of the parametric survival regression using the EUROSTAR data (1994–2006) for a log-normal, a Weibull and an exponential (constant hazard) distribution. Figure 72 shows the rate of aneurysm-related death over time predicted by each regression model (for patients aged 74 years with an AAA of 5.5 cm). The Weibull and log-normal models estimate similar rates of aneurysm-related death during the first 5 years, the hazard increasing over time for these patients to a maximum of about 0.4%. The Weibull model then predicts that the hazard continues to gradually increase over time, whereas the log-normal model predicts that the rate after 5 years then gradually decreases over time. The average rate (exponential model) for this patient group was 0.33%.

TABLE 62 - Results of log-normal, Weibull and exponential survival models of the rate of aneurysm-related deaths occurring more than 30 days following aneurysm repair with EVAR, from the EUROSTAR data (equation 3 in Figure 69)

Number of observations	8182
Number of deaths	142
Patient-months (patient-years)	228,471 (19,039)
	Log-normal model		Weibull model		Exponential model (base case)
	Coefficient	SE	Coefficient	SE	Coefficient	SE
Per year of age over 74 years	–0.04	0.01	0.04	0.01	0.04	0.01
AAA size 5.5–5.9 cm	–0.48	0.25	0.52	0.26	0.52	0.26
AAA size 6–6.4 cm	–0.64	0.26	0.69	0.27	0.68	0.27
AAA size 6.5+ cm	–1.35	0.23	1.33	0.22	1.32	0.22
Older generation	–0.85	0.20	0.82	0.18	0.89	0.18
Unfit for open repair	–0.68	0.20	0.64	0.20	0.62	0.19
Intercept	9.16	0.46	–9.00	0.39	–8.53	0.22
Log sigma coefficient	0.81	0.07
Log shape coefficient			0.12	0.08
Log likelihood	–731		–728		–729

FIGURE 72.

Predicted rates of aneurysm-related death occurring more than 30 days following aneurysm repair with EVAR for a patient aged 74 years, with an aneurysm diameter 5.5–5.9 cm, with a recent EVAR device and fit for open surgery. Estimated from EUROSTAR data using an exponential model, a Weibull model and a log-normal model. Dashed lines indicate extrapolation beyond the maximum of 6 years of follow-up between 2000 and 2006.

Hence, there is considerable uncertainty about the relationship between late aneurysm-related death and time from surgery. The base-case model assumed that the rate of late aneurysm-related death was constant from 1 month after surgery (exponential survival model). Sensitivity analyses explored alternative scenarios. An increasing rate might correspond with a belief that the aneurysm continues to expand after EVAR, whereas a decreasing rate might correspond with a belief that patients at risk will be successfully identified by long-term surveillance and receive appropriate treatment.

llustration of differences in operative, aneurysm-related and non-aneurysm-related causes of death between EVAR and open repair in the base-case model

Figure 73 illustrates how predicted death rates for each cause differ in the base-case model between EVAR and open repair. The figure shows the differences in the cumulative rates of death between the treatments (in patients at risk up to time t) for all-cause deaths, AAA deaths and non-AAA death. The initial difference in favour of EVAR is due to a benefit in early operative mortality. There is a continuing difference in late aneurysm-related mortality between the treatments. There is also a difference in late non-aneurysm-related deaths because of there being a greater proportion of patients with poor fitness among survivors of EVAR than among survivors of open repair. This higher rate of mortality persists until the survival curves for late non-aneurysm-related death converge, in this case at about 4 years. Although there is a persisting difference in aneurysm-related deaths after the survival curves meet, this has only a small effect on all-cause mortality because of the relatively high competing risk of deaths from other causes (Figure 74).

FIGURE 73.

Illustration of the difference between the treatments in the cumulative rates of death (from aneurysm-related and non-aneurysm-related causes) estimated in the base-case decision model for a patient aged 75 years with a large AAA (6.5 cm) and poor fitness.

FIGURE 74.

Proportions surviving each year free from aneurysm-related death and all-cause death estimated in the decision model for a patient aged 75 years with a large AAA (6.5 cm) and poor fitness.

Resource use and costs

Costs are incurred in the model during the primary admission, during surveillance post surgery and if the patient is readmitted to hospital for an aneurysm-related complication.

Cost-effectiveness analysis

Standard decision rules were followed for the cost-effectiveness analysis using expected costs and QALYs. 151 When there are two options under comparison, the ICER is calculated if both the cost and the benefits of EVAR exceed those of open repair. If EVAR is more costly but less effective than the alternative then EVAR is dominated and no ICER is calculated.

The same decision rule can be expressed in terms of maximising ‘expected net benefit’. 152 Expected net benefit (NB) for a treatment option is defined as:

where λ is the threshold cost-effectiveness used by the decision-maker. This is the most convenient decision rule when there are three or more mutually exclusive strategies being compared, as is the case in the subsequent section in which we compare open surgery, EVAR and watchful waiting. Results are shown for thresholds of £20,000 and £30,000 per QALY.

The results of the decision model are shown (1) for the aggregate UK population who are considered suitable for aneurysm repair and (2) disaggregated according to age group, aneurysm size and operative fitness. Appendix 6 explains how the mean characteristics of the UK population were determined.

One-way sensitivity analyses were carried out by varying key parameters in the model. A probabilistic sensitivity analysis, based on the uncertainty in all of the parameters of the model, was undertaken to estimate the probability that EVAR is more cost-effective than open repair as a function of the threshold ICER. 153

Table 66 shows the uncertainty arising from measurement error in the estimates of each of the parameters used in the base-case model comparing EVAR and open repair. Some parameters have been estimated from regression equations (equations 1–4) and therefore there may be correlations between the coefficients of these equations. The Cholesky matrix was estimated for each risk equation and used to calculate the distribution of the linear predictor for these parameters, assuming that the coefficients of these equations follow a joint normal distribution.

Results for aggregate population

Table 67 shows the results of the decision model for the average UK population.

In the base-case analysis, the total incremental lifetime cost of EVAR versus open repair is £2002. This can be approximately broken down as the additional cost of the initial procedure (£520), conversions to open repair (£250), late reinterventions (£820) and the additional cost of surveillance (the remainder, about £410). The total difference in lifetime QALYs between EVAR and open repair is estimated to be 0.041. The positive benefit of EVAR is maintained as a cumulative QALY gain of 0.056 up to about 3 years, but is subsequently offset by extra late aneurysm-related deaths and reinterventions in the long term after EVAR. The ICER for the base case is approximately £49,000 per QALY (calculated as 2002/0.041).

The model includes an excess hazard of late non-aneurysm-related death after EVAR until the survival curves converge. If the excess hazard is set such that the survival curves converge at 8 years (with other parameters as the revised base case) then the ICER is approximately £22,000 per QALY. If the excess hazard is twice that of the base case then the survival curves converge at 2 years and the ICER is approximately £96,000 per QALY.

The revised base case assumes that the hazard of late aneurysm-related death is 1.5 times greater after EVAR than after open repair, for the lifetime of the patient. If there is no difference (HR of 1) then the ICER is £29,000 per QALY. If the HR is 1.2, the ICER is approximately £37,000 per QALY. The HR may be time varying, for example there may be no excess hazard after 4 years. This scenario gives similar results to the revised base case, with an ICER of approximately £49,000 per QALY.

The original base case in the assessment report assumed that the HR of late reintervention was 6.7 for the lifetime of the patient, although the absolute rate of reintervention is declining over time and is low (about 2% per year) 4 years after EVAR (see Figure 75). If there is no difference between treatments (HR of 1) the ICER is approximately £27,000 per QALY. If the HR is 1.5 (corresponding to the relative hazard of reinterventions for any cause in Schermerhorn et al. 96) the ICER is £29,000 per QALY.

FIGURE 75.

Predicted rates of readmission per year over time (Weibull regression, Table 63, EVAR trial 1 data). Dashed lines show the rates of readmission predicted beyond the maximum 4-year follow-up available in EVAR trial 1.

The original base case assumed that one follow-up appointment per year with CT was required after EVAR. If the cost per year was half of this then the ICER is £44,000 per QALY. If there are no follow-up visits (with reinterventions and aneurysm deaths remaining unchanged) then the ICER is approximately £39,000 per QALY.

There may be a correlation between follow-up visits, reinterventions and late aneurysm-related deaths. In a favourable scenario, if there are fewer complications after EVAR, with less need for follow-up, fewer reinterventions and fewer aneurysm-related deaths, then the ICER is approximately £24,000 per QALY.

If the OR of operative death is 0.25 rather than 0.35 as used in the base case then the ICER is approximately £22,000 per QALY, with other parameters as in the revised base case. The survival curves take 4.5 years to converge, as the initial benefit from EVAR is larger.

The base case assumed that the EVAR procedure cost £523 more than open repair. We conducted a sensitivity analysis with a lower cost for the EVAR procedure, for example reflecting less use of intensive care than in the EVAR trial 1. If it is assumed that EVAR costs £623 less than open repair, the ICER is approximately £21,000 per QALY. If it is assumed that EVAR costs the same as open repair, the ICER is £36,000 per QALY. In a multivariate sensitivity analysis, if EVAR costs the same as open repair and the costs of intervention and follow-up are lower than in the base case then the ICER is £12,000 per QALY.

We also conducted a probabilistic sensitivity analysis using the distributions shown in Table 66. The results are shown in Table 67. The probability that EVAR is cost-effective in the base case is 0.261 at £20,000 per QALY or 0.424 at £30,000 per QALY. At a threshold willingness to pay of £20,000 per QALY, EVAR is more likely to be cost-effective than open repair under some scenarios: if the survival curves do not converge; if the HR of reintervention is 1.5 and the annual surveillance cost of follow-up is low; if the OR of operative mortality is 0.25; or if the EVAR procedure costs less than open repair. It may seem counterintuitive that in some scenarios the ICER (the measure of mean cost-effectiveness) is greater than £20,000 but the probability that EVAR is cost-effective is more than 50% (the measure of median cost-effectiveness) at a threshold of £20,000 per QALY. This occurs because the Markov model is non-linear, and the distribution of net benefits is not symmetrical over all of the simulations of the probabilistic sensitivity analysis.

Base-case results disaggregated by patient subgroup

Table 68 shows the cost-effectiveness results for EVAR compared with open repair by age, aneurysm size and fitness at baseline. The results of the base-case model suggest that EVAR is cost-effective for patients of relatively poor fitness at most ages and aneurysm sizes. For patients with relatively good fitness (i.e. no comorbidity) open repair is more cost-effective. Note that the result for patients with moderate fitness, aged 75 years and with an aneurysm of 6.5 cm corresponds to the average result shown in Table 67.

In general, the ICER of EVAR versus open repair is lower for older patients than for younger patients, for patients with a larger aneurysm size than with a smaller aneurysm size, and for patients of poorer fitness than of better fitness (Table 68). Older patients and those with larger aneurysms and poorer fitness face increased operative mortality. The model assumes that the relative treatment effect (OR) of operative mortality is constant across risk groups. Therefore, the absolute benefit of EVAR compared with open repair is greater in patients of poorer operative risk. Furthermore, there is a long-term risk of complications and reinterventions after EVAR. Patients with a longer life expectancy face a greater cumulative risk of complications, with additional costs, disutility and risk of late aneurysm-related mortality.

The ICER comparing EVAR and open repair does not decrease in a linear progression with age (Table 68). For example, for a patient with moderate fitness and aneurysm size of 6.5 cm, the ICER is £92,000 per QALY at age 70 years, £49,000 per QALY at age 75 years and £33,000 per QALY at age 80 years. This occurs because the ICER is a ratio of the difference in costs to the difference in health benefits and increases rapidly as the difference in the denominator approaches zero. The lifetime difference in costs is similar for all age groups (about £2000) whereas the lifetime difference in QALYs is only 0.025 in 70-year-olds but 0.052 in 80-year-olds, because the absolute difference in operative mortality is greater for older patients.

There is a non-linear relationship between aneurysm size and operative mortality. The EUROSTAR data predicted that the risk of late aneurysm-related death after EVAR increases with aneurysm size at the time of the procedure (equation 3, Table 62), confirming estimates made by studies on earlier EUROSTAR data sets. 82 The increased risk of late aneurysm-related death causes the ICER of EVAR versus open repair to be greater for patients with an aneurysm size of 6.5 cm than for patients with an aneurysm size of 6 cm (Table 68).

Table 68 shows that EVAR might also be cost-effective compared with open repair in older patients (80 years or more) with moderate fitness and very large aneurysms (≥ 7.5 cm). Although we define fitness in this analysis as ‘moderate’ if patients have few pre-existing conditions relative to other patients of their age, operative mortality would be high in absolute terms for these patients, estimated at 6.7% after EVAR and 17% after open repair. A policy of no surgery or watchful waiting should also be considered for patients with high expected operative mortality, compared with the risk of rupture without surgery. These polices will be evaluated in the model comparing immediate elective surgery, watchful waiting and no intervention.

The cost-effectiveness of EVAR compared with open repair in women

The risk equations (equations 1–4) estimated earlier in this section did not find gender to be a significant explanatory variable, and therefore the base-case model does not distinguish between male and female patients, other than to use life tables for men to estimate non-aneurysm-related mortality in the general population. However, Chapter 3 identified one large study86 that found an independent effect of gender on 30-day operative mortality (OR women versus men 1.46, 95% CI 1.26 to 1.68). A secondary analysis explored the cost-effectiveness of EVAR specifically in women, assuming greater 30-day operative mortality after EVAR in women as estimated by Timaran et al. 86, assuming that the average treatment effect (OR) of 30-day mortality of EVAR compared with open repair found by the RCTs in Chapter 3 applies to women and using the age-specific non-aneurysm-related mortality rates (life tables) for the female general population. Results were very similar to those of the base case.

Comparison of York model with Medtronic model of EVAR versus open repair

The Medtronic model comparing EVAR and open repair was described earlier in this chapter (see Systematic review of existing cost-effectiveness evidence). The main differences between the York base-case model and the Medtronic127 base-case model are:

• the difference between EVAR and open repair non-aneurysm-related mortality rates in the medium term

• the difference in late aneurysm-related mortality

• the hospital costs (intensive care and operating theatre time) of the EVAR procedure

• the relative rate of reinterventions.

The assumptions made by Medtronic127 are shown as scenario 18 in Table 67. In this scenario there is a slower rate of convergence of the survival curves than in the York base case, a lower relative cost of EVAR and no difference in late aneurysm-related deaths between EVAR and open surgery. The Medtronic model presented results for a patient aged 70 years with an aneurysm size of ≥ 5.5 cm. Fitness was unspecified in the Medtronic model (i.e. results were for the average level of fitness of patients in the EVAR trial 1). When these assumptions are used in the York model the ICER is about £15,000 per QALY, consistent with the results reported in the Medtronic report. 127

Figure 76 illustrates the differences between the assumptions for all-cause mortality made in the York model and in the Medtronic model. The figure shows the difference in cumulative deaths between EVAR and open repair predicted by the models. The York model assumed that the initial survival advantage of EVAR for operative mortality would be entirely offset within 3 years by a relatively higher non-aneurysm-related death rate after EVAR. This scenario is consistent with the results of EVAR trial 1, DREAM and a large US matched-cohort study. 96 The predicted survival curves are shown in Figure 77 for a patient aged 70 years with an aneurysm size of 5.5 cm. Furthermore, the York model assumed that there would be a small but persistent difference between the treatments in the rate of late aneurysm-related deaths. This scenario is consistent with the long-term EUROSTAR data. 82 In contrast, the base-case model of Medtronic assumed a more optimistic scenario, that the rate of convergence of the survival curves would be slightly slower and that the survival curves would not meet. There would be no further difference in deaths beyond 4 years and, therefore, a slight long-term difference in survival would be maintained in favour of EVAR over the lifetime of the patients (Figure 76). The predicted survival curves when the Medtronic assumptions are used in the York model are shown in Figure 78.

FIGURE 76.

A comparison of the difference in cumulative deaths between EVAR and open repair over time for a patient aged 70 years with an aneurysm size of 6.5 cm and moderate fitness predicted by (a) the York model and (b) the Medtronic base-case model.

FIGURE 77.

Predicted survival curves from the York model for a patient aged 70 years with moderate fitness and an aneurysm size of 6.5 cm. Survival curves converge by 3 years.

FIGURE 78.

Predicted survival curves using Medtronic assumptions in the York model for a patient aged 70 years with moderate fitness and an aneurysm size of 6.5 cm. Survival curves do not converge. There is a small but cummulatively important difference in all-cause survival between the treatments.

Comparison of York model with other published economic evaluations

The base-case model found that EVAR was not expected to be cost-effective on average, but was cost-effective for patients with poorer fitness. These results can be compared with the published models for similar patient groups. The systematic review of existing cost-effectiveness evidence presented earlier in this chapter found that the studies by Michaels et al. 107 and Epstein et al. 106 were the published economic evaluations most relevant to the current decision in England and Wales. JA Michaels, DM Epstein and MJ Sculpher are authors of one or both of these published papers and also of this report.

Both Michaels et al. 107 and Epstein et al. 106 concluded that EVAR was not expected to be cost-effective in patients eligible for elective sutgery. However, there were differences between these published models and the York model in the assumptions used to arrive at this conclusion. Michaels et al. found a greater long-term benefit in favour of EVAR than the York model, and a greater difference in costs, but this study was published before the mid-term results of the good-quality RCTs were available. The study by Epstein et al. was published after the mid-term results of the RCTs were available, and based on these trial results assumed that the survival curves for EVAR and open repair would meet by 4 years. The published model has been adapted for use in the York economic evaluation. The main difference in the parameter values between the models is that Epstein et al. assumed a greater difference in late aneurysm-related deaths (HR EVAR versus open repair = 6.0) than in the York model (HR = 1.5). The York model also used regression analysis to estimate baseline risks of operative mortality, late aneurysm-related mortality and non-aneurysm-related mortality in a wider range of patient groups than in Epstein et al.

The next section extends the York model to compare the cost-effectiveness of EVAR, open repair, watchful waiting and no intervention.

Introduction

The objective of this second model was to broaden the nature of the comparisons made using the first model. Specifically, the second model considers when surgery (with EVAR or open repair) might be cost-effective compared with no surgery or delaying the decision. The model brings together the sparse available evidence about natural history in untreated patients with evidence in treated patients to predict outcomes of a wide range of management policies in patients with diagnosed aneurysm. Given the uncertainties in these data, the model is intended to be exploratory and suggest areas for further research.

Current guidelines for the management of AAA were discussed in Chapter 1. Briefly, patients are observed until the aneurysm reaches 5.5 cm in diameter, after which surgical intervention is considered. 141 Patients considered fit for open surgery might be offered EVAR or open surgery; patients considered unfit for open surgery might be offered EVAR or no intervention. However, in practice there is a continuous range of probabilities of operative mortality and the optimum management policy should systematically weigh up all of the risks to the patient – that is, operative mortality and late mortality if treated versus the risks of rupture if untreated. 11 Furthermore, a publicly funded health-care system must also evaluate the use of health-care resources, which is not considered explicitly by the current clinical guidelines. This section presents a decision model to evaluate the cost-effectiveness of surgery, watchful waiting or no surgery for patients of different ages, operative fitness and aneurysm size.

Description of the watchful waiting strategy

At each consultation with the vascular surgeon the patient faces four options:

• immediate elective surgery with EVAR

• immediate elective surgery with open repair

• ruling out surgery entirely

• delaying the decision (watchful waiting).

We assume that the patient is evaluated every 6 months in the watchful waiting policy146 and that the patient attends all scheduled follow-up visits. In practice, a substantial risk of patient non-compliance would diminish the value of a watchful waiting strategy, although we do not model this scenario. We assume that surveillance is discontinued if a decision is made to rule out surgery and there are no subsequent monetary costs to the health-care service. In practice, the patient may return if the aneurysm becomes symptomatic, but we do not model this scenario. The benefits of delaying the decision are that it allows more information on the aneurysm growth rate to be assembled, and preserves the option to commence immediate surgery in the future should the patient’s health state (aneurysm size) worsen. 10 The costs of deferral are monetary costs (the monitoring costs of CT and outpatient attendance) but also an important opportunity cost: patients may die from rupture while waiting. Waiting is also a source of uncertainty: people prefer current benefits rather than future benefits. We represent the cost of this impatience by discounting delayed benefits. Because AAAs are usually asymptomatic we assume that patients have normal HRQoL for their age while under surveillance, although there is some evidence that patients with diagnosed untreated aneurysm suffer anxiety. 146

The approach used to model watchful waiting is as follows:

• first, the model previously described to evaluate EVAR versus open surgery was used to estimate the maximum expected net benefit of surgery in patients of a given fitness for a range of aneurysm sizes (4–8 cm, in increments of 0.5 cm) and ages (70–85 years, in increments of 6 months)

• second, another model was constructed to evaluate an option of no surgery (i.e. natural history, with no treatment and no surveillance) for the same patient groups; this model is described in the following section

• finally, a dynamic programme was constructed using these data to estimate the net benefit of a watchful waiting strategy and to calculate the optimum policy (EVAR, open repair, no surgery or watchful waiting) for each aneurysm size and patient age.

Model to estimate the natural history of untreated aneurysm

To estimate the natural history of untreated aneurysm a Markov cohort model is used. The aim of the model is to estimate QALYs over a patient’s lifetime if the patient is left untreated. As there is no surveillance and no surgery in this model there are no costs. The discrete health states are aneurysm size, from the size at diagnosis up to a maximum of 10 cm in diameter in increments of 0.5 cm. Rupture rate is conditional on aneurysm size. The mean growth rate and standard deviation and rupture rate were obtained from a review of the literature on the natural history of untreated aneurysm. It was assumed that the growth rate of aneurysms (g) is normally distributed with the mean and variance being a function of aneurysm size in the previous period. Markov transition probabilities for moving from one health state to another were derived from this continuous distribution as follows:

• Pr(aneurysm increases by 1 cm in a single 6-month period) = Pr(g ≥ 1)

• Pr(aneurysm increases by 0.5 cm in a single 6-month period) = Pr(0.5 ≤ g < 1)

• Pr(no change in aneurysm in the 6-month period) = Pr(g < 0.5)

• It is assumed that aneurysms do not diminish in size.

As the annual probability of rupture is estimated to be 90% for aneurysms of 10 cm in diameter11 it was assumed unnecessary to predict growth beyond this size. Consequently, when the aneurysm is 9.5 cm in diameter the above algorithm is amended to:

• Pr(aneurysm increases by 0.5 cm in a single 6-month period) = Pr(g ≥ 0.5)

• Pr(no change in aneurysm in the 6-month period) = Pr(g < 0.5).

In this model rupture is assumed to be fatal. Very few patients reach hospital alive after rupture and therefore the possibility of emergency surgery would be unlikely to affect the results. 11 Given the absence of evidence on how fitness might evolve over time, and the effect of fitness on aneurysm growth and rupture, it is assumed that fitness is constant over the duration of the model.

Parameter estimation for natural history model

Illustration of the predicted rate of mortality of surgical treatment compared with no treatment at each age and aneurysm size

Figure 79 illustrates the rate of mortality estimated over time in the model for a patient with a starting age of 70 years, with poor fitness and with an initial aneurysm diameter of 4 cm. Without treatment the aneurysm is predicted to grow exponentially and the risk of rupture increases according to aneurysm diameter. 11 Given these estimates of aneurysm growth, and the risk equation estimated from EUROSTAR (Table 58, equation 1), the expected operative mortality with EVAR would increase in relation to increasing age and aneurysm diameter, from about 1.5% at age 70 years to 10% after 7 years.

FIGURE 79.

Illustration of mean rates of operative mortality and rupture estimated over time in the model for a patient aged 70 years, with poor fitness and with an aneurysm of diameter 4 cm at diagnosis. The figure also shows the change in aneurysm size over time.

Cost-effectiveness analysis using dynamic programming

This section describes how dynamic programming was used to select the most cost-effective option (surgery, no intervention or watchful wait) for patients at each age and aneurysm size, using the results of the models for estimating the net benefits of surgery and the natural history described in the previous sections. The methods are closely based on the work of Driffield and Smith. 10

This section has three parts:

1. We explain the concepts of dynamic programming and how the method can be used to simplify the modelling of watchful waiting.

2. We calculate the optimal policy for each aneurysm size in the final time period.

3. We calculate the optimal policy for each aneurysm size in each of the previous time periods.

Concepts of dynamic programming

Figure 80 illustrates the management options. The benefits and costs of the strategies of immediate elective repair versus no surgery depend only on future chance events, such as whether the aneurysm grows slowly or quickly in each cycle, or the patient dies. The previous sections described the Markov models used to estimate patients’ lifetime expected benefits and costs for these two strategies, for any given starting age, aneurysm size and fitness. In principle, we could also use a decision tree or Markov model to calculate a watchful waiting strategy. However, this would quickly become intractable. The watchful waiting strategy is more complex than a decision simply to treat or not treat taken at diagnosis. At each future age and aneurysm size there are three decision options (immediate surgery, rule out surgery or watchful waiting) and an indeterminate time horizon. The problem is dynamic, i.e. the optimal strategy depends not only on future chance events but also on decisions made in the light of those events (Figure 80). This means that there are hundreds of possible strategies for watchful waiting (e.g. immediate EVAR, wait 6 months then surgery if the aneurysm is 0.5 cm larger, wait 1 year, etc.) for each starting age and aneurysm size.

FIGURE 80.

Management of a patient diagnosed with AAA, showing immediate elective repair, no surgery and watchful waiting strategies.

To reduce this complexity we use a dynamic programming formulation to calculate the optimum strategy for each age and aneurysm size. 165 Dynamic programming is based on a simple principle: that if a decision has a finite time horizon (N periods) and we know the optimal choices (payoffs) for each aneurysm size (model state) in period N+1 and we know the probabilities of transition between model states then we can work backwards to induce the optimal choices for each model state in the previous period (N, N–1 and so on) until the starting period (t = 1). Both surgery and a decision to discharge the patient are irreversible. Delaying a decision might have value because, as time progresses, the information regarding aneurysm size and growth rate is updated, resolving uncertainty and offering the option of changing the treatment policy. Continuing surveillance with an option to treat in the future if the aneurysm grows might give greater expected benefit than either of the irreversible decisions, which can be compared with the costs of obtaining this information. The output of the dynamic programme is a policy that states the optimal management option for each patient age and aneurysm size.

Calculating the optimal policy for each aneurysm size in each of the previous time periods

Given that we have calculated the optimum choices for each health state (aneurysm size) at period N + 1 (corresponding to age 79.5 years) we use backward induction to calculate the optimum choices for each health state in the previous period N. Figure 81 illustrates the numerical solution method with a segment of the decision tree for a patient of very poor operative fitness, for example ineligible for EVAR trial 1 or the DREAM trial. The willingness to pay threshold is £30,000 per QALY.

FIGURE 81.

Illustration of the dynamic programme decision tree for the periods N and N +  1 for a patient of very poor operative fitness. The values shown are the incremental net benefits of deferring the decision (watchful waiting) or of immediate treatment compared with abandoning watchful waiting (never offering surgical treatment). The dynamic programme is evaluated from right to left, that is, the cost-effective policies are identified for each state in period N + 1 and then for each state in period N, N –  1, etc. Note: The cost of an outpatient visit with CT scan is £191. The probability of the aneurysm growing from 7 to 8 cm in 6 months is 0.17 and of growing from 7.5 to 8 cm in 6 months is 0.52. Cost-effective options at a threshold of £30,000 per QALY are highlighted in bold.

The right-hand side of the figure shows the possible states of health in period N + 1 (age 79.5 years). The incremental net benefits of surgery (most cost-effective of EVAR versus open repair) compared with no surgery have been calculated using the decision models for treated and untreated patients previously described; the value for no surgical treatment is always shown as zero (relative to surgery). In the period N + 1 there is no option to defer and so the decision is merely whether the patient should receive treatment. This decision is straightforward (given the data), depending only on whether the incremental net benefits of surgery are positive (relative to no surgery). At age 79.5 years surgery would be offered only if the aneurysm is 8 cm in diameter.

Moving back to period N (age 79 years) there is now the additional option to defer treatment. In each state we compare three possible actions: deferral, treatment and abandonment. If the aneurysm is 7 cm, deferral is calculated as the expected net benefits from waiting another period (6 months), in which three possible states of health could occur: no growth, growth by 0.5 cm (to 7.5 cm) or growth by 1 cm (to 8 cm). The probabilities of these outcomes were calculated in Table 71 to be 0.31, 0.52 and 0.17 respectively. Delaying a decision might generate benefits but it also has costs. Delaying a decision might have value because it allows resolution of the uncertainty about whether the aneurysm will grow, which in turn would change the treatment decision, with greater benefit than either immediate treatment or never offering surgery. If delay is permitted in period N, the optimal strategy in period N + 1 is no surgery if the aneurysm does not grow, no surgery if the aneurysm grows to 7.5 cm and EVAR if the aneurysm grows to 8 cm. There are three sources of opportunity cost of delaying the decision. First, patients may rupture while waiting. Second, there is a time preference for current benefits and so future uncertain benefits are discounted. Third, there is a monetary cost of monitoring, which is assumed to be one outpatient visit with CT scan costing £191. These costs and benefits of delay are expressed in the following formula:

Present value at 7 cm in period N of future net benefit of deferral = ( 1 − Pr ( rupture ) ) × e − r / 2 × ( 0.31 × 0 + 0.52 × 0 + 0.17 × 1366 ) − 191 = 37

where Pr(rupture) is the probability of rupture for a patient with a 7-cm aneurysm, e is the exponential function and r is the discount rate (0.035 per year).

This can be compared with the counterfactual, which is the present value of immediate elective EVAR at 7 cm in the period N. This was calculated using the decision model for EVAR described in the previous section. For convenience, this value is shown relative to a policy of no surgery, so that ‘no treatment’ is always shown with a value of zero. If surgery has been carried out there is no monetary cost of continued surveillance, nor will the patient rupture from untreated aneurysm.

Therefore, if the aneurysm is 7 cm at 79 years the optimal decision is to continue waiting, as the net benefits of waiting are greater than those of treatment. For aneurysms < 7 cm the optimal decision is to discharge the patient, because at period N+1 it will never be cost-effective to treat, regardless of the aneurysm growth rate. For aneurysms > 7 cm it is not cost-effective to wait and the best strategy is immediate treatment.

The same algorithm is used to calculate the net benefits for each aneurysm size in period N–1, and the process continues by backward induction until period 1 is reached. In most dynamic programming applications the main interest is in the decision at period 1 and the future period calculations are performed merely to inform that decision. 10 In this application, however, we are also interested in the grid of policies for all ages and aneurysm sizes, as this indicates the most cost-effective policy for any patient at diagnosis.

Results of watchful waiting model for patients of very poor operative fitness

Table 72 shows the optimum policy for patients of very poor operative fitness at each age and aneurysm size, under base-case assumptions and at thresholds of £20,000 and £30,000 per QALY. Patients are similar to those who were eligible for the EVAR trial 2. The results show that, for patients of very poor fitness, EVAR might be cost-effective at £20,000 per QALY up to 77 years in patients with an aneurysm of 8 cm, up to 74 years in patients with an aneurysm of 6 cm and up to 71.5 years in patients with an aneurysm of 5 cm. Increasing the threshold to £30,000 per QALY increases the age at which EVAR is cost-effective by about 2 years.