Chapter  144:  Comparison of Endovascular and Open Surgical Repairs for Abdominal Aortic Aneurysm

Objectives and Key Questions

The Minnesota Evidence-based Practice Center (EPC) was asked to answer the following questions related to elective treatment of nonruptured AAA that were nominated by America's Health Insurance Plans (AHIP).

• 1

  What are the comparative effectiveness and adverse effects of treatment options of AAA including active surveillance, open repair, and endovascular repair?

• 2

  What is the relationship of volume, both hospital and physician, to the benefits and harms of endovascular procedures to repair AAA?

• 3

  How do the characteristics of the aneurysm (size/location/shape) and the patient (age/gender) affect the benefits and harms of endovascular and open-surgical repair?

• 4

  What are the costs-benefits for each of the procedures?

Literature Search

For questions 1 and 3 we searched PubMed®, Cochrane Library, Food and Drug Administration (FDA), and other electronic websites from 1990 to October 2005. The following terms were included: abdominal aortic aneurysm, endovascular repair, surveillance, and surgery. Titles and abstracts of identified references were reviewed. For question 2, MEDLINE® was searched for publications about volume-outcome relationships for procedures to repair AAA published after 1990. Key words included abdominal aortic aneurysm, volume, outcome, and process assessment of health care. For question 4 we used the terms abdominal aortic aneurysm combined with economics, nursing economics, pharmaceutical economics, cost, pharmacoeconomics, cost analysis, cost allocation, cost-benefit analysis, cost control, cost savings, cost of illness, cost sharing, “deductibles and coinsurance,” medical savings accounts, health care costs, direct service costs, drug costs, employer health costs, hospital costs, health expenditures, capital expenditures, hospital economics, hospital charges, hospital costs, medical economics, or medical fees. Reference lists and contacts with content experts were used to identify additional reports.

Study Selection Extract Data

For questions 1 and 3, studies were eligible if they were randomized controlled trials or multivariate natural history studies involving AAA, open repair, endovascular repair, or active surveillance and reported clinical outcomes. We included and updated a systematic review of EVAR vs. open repair published by the U.K. National Institute of Clinical Excellence (NICE) as well as nonrandomized U.S. trials or national registries.⁶ To assess volume-outcome relations (question 2), articles that met the following criteria were reviewed. The report had to be an original analysis of U.S. data representing repair of unruptured AAA beginning in 1990; the sample had to represent variation between hospitals or surgeons in a community or larger geographic area and present sample statistics representing the relationship between a measure of hospital or physician volume and a beneficial or harmful outcome of AAA repair; the analysis had to attempt to make adjustments for other known risk factors. For question 4, studies had to include at least 50 AAA procedures in each arm, provide data on costs, charges, or reimbursement rates, or be a cost-effectiveness analysis.

Quality Assessment and Strength of Evidence

The methods of Schulz et al. were used to assess the quality of randomized controlled trials (RCT).⁷ We determined whether intention-to-treat analysis was utilized and whether published RCT used EVAR devices similar to those available in the United States. We evaluated whether effectiveness of interventions varied according to patient, aneurysm, or device characteristics including: age, race, gender, EVAR device manufacturer, or aneurysm size (small [<5.5 cm] vs. large aneurysm). Data from nonrandomized studies were included to assess relevance and consistency to practice in the United States. Analyses of administrative databases to examine relationships between provider volume and outcomes were evaluated by their ability to represent current practices in the United States, the reliability and validity of ‘volume’ and outcome measures, and the thoroughness of the regression analyses, including control of key risk factors. Quality of cost studies were based on whether they included relevant costs, used appropriate time frame, and evidence of effectiveness and adverse effects from randomized controlled trials.

Data Extraction

Data regarding study, patient, and device characteristics as well as outcomes were extracted. Outcomes were described as effectiveness (initial clinical success ≤30 days or during initial hospitalization; short-term (30 days to 6 months), midterm (6 months to 5 years), and long-term >5 years) and adverse effects based on recommendations for reporting of outcomes for open and EVAR by the Society for Vascular Surgery.⁸ The primary outcome was midterm all-cause mortality. Additional outcomes included initial all-cause mortality, AAA mortality, AAA rupture, quality of life, and technical measures of EVAR and OSR success. Adverse events and complications included treatment-related mortality and morbidity such as endoleaks, graft rupture, migration, kinking, and need for additional interventions, including OSR.

For question 2 the main outcome that has been studied in relation to provider volume is short-term mortality. All published analyses used regression models of this outcome on retrospectively defined volume measures. Information was abstracted to characterize (1) the study population, (2) volume measures, (3) outcome measures, (4) regression model including covariates used for risk adjustment and handling of clustering of patients within hospitals or surgeons, and (5) adjusted and unadjusted estimates of volume effects. For question 4 we assessed charges, costs, reimbursement rates, hospital and intensive care unit length of stay, incremental cost effectiveness ratios, and quality adjusted life years.

Data Analysis

Descriptive characteristics of patient populations were presented. For results from RCTs we calculated odds ratios (ORs) for categorical variables and weighted risk differences for continuous variables with the corresponding 95 percent confidence intervals. Because of the paucity of RCT, results from each study were evaluated and described and the corresponding statistical parameters for effect size and significance were described. We reported on tests of interaction for predefined subgroups. The preponderance of evidence for relationships between the volume of procedures performed by hospitals and physicians to repair AAA and outcomes has been obtained by secondary regression analysis of administrative databases. Methodological differences, including the definition of volume and outcomes measures, make it difficult to compare results from the different studies. Use of variables to control for pre-procedure risk (case-mix), physician training, and health care system attributes varied. Analytical methods varied as well. Therefore, we did not attempt to do a meta-analysis to combine estimates from different retrospective observational studies. The summary of studies of the volume-outcome relationship was limited to tabulation of individual study characteristics and results. Cost analysis was limited to a review of published findings and discussion of study limitations and strengths.

Comparative Effectiveness of Treatment Options

The strongest known predictor of rupture is AAA diameter. AAA enlargement rate is not consistently independently associated with rupture rate. The annual rate of rupture for AAA <5.5 cm is 1 percent or less. Among individuals refusing or medically unfit for surgical intervention, the 1 year rupture risk may exceed 10 percent in AAA >6 cm and for AAA of >8 cm, the risk may exceed 25 percent at 6 months.

For patients with small AAA (<5.5 cm), two high-quality RCT (n=2,226), the Aneurysm Detection and Management study (ADAM)⁹ and the United Kingdom Small Aneurysm Trial (UKSAT)¹⁰ demonstrated that early/immediate OSR of AAA did not reduce all-cause mortality compared with active surveillance and delayed OSR. After a mean followup of approximately 5 years, the overall relative risk (RR) and hazard ratio (HR) for all-cause mortality for ADAM and UKSAT were 1.21 [95 percent confidence interval (CI) 0.95 to 1.54] and 0.94 [95 percent CI 0.75 to 1.17, p=0.56], respectively. Outcomes did not differ between treatment groups according to age, gender, or AAA diameter at baseline. After a mean followup of 8 years in the UKSAT study, no significant difference in mean survival was found between groups (6.5 years for surveillance versus 6.7 for early surgery, p=0.29) although total mortality was lower in the early/immediate OSR (adjusted HR was 0.83 [95 percent CI 0.69 to 1.00, p=0.05]). AAA related mortality was not reduced by early/immediate OSR in the ADAM study (3.3 percent vs. 3.4 percent). In the UKSAT trial, AAA-related deaths (combined ruptured AAA, secondary AAA rupture, and AAA repair deaths) accounted for 19 percent of all deaths in the surveillance group compared to 15 percent for the early OSR group at a mean followup of 8 years. Differences in quality of life measures, when they existed, were small.

Three high-quality multicenter trials compared the outcomes of EVAR with OSR of large AAA (≥5.5 cm) (n = 1,489) in patients judged medically fit for OSR.¹¹–¹³ Two trials provided data at 2 years or longer (n = 1,413).^{12, 13} None of the studies were conducted in the United States. All began recruitment prior to 2000 and some EVAR devices used may not be currently available in the United States EVAR reduced postoperative 30-day mortality compared to OSR (EVAR = 1.6 percent vs. OSR = 4.7 percent RR = 0.34 [95 percent CI 0.17 to 0.65]) and hospital length of stay. Reduction in all-cause mortality seen with EVAR disappeared by 2 years. There was a persistent 3 percent reduction in AAA-related mortality. Outcomes did not differ significantly by treatment for either all-cause or AAA mortality according to age, gender, aneurysm diameter, or renal function though few women were enrolled. In EVAR-1, postoperative complications, included primarily EVAR graft related ruptures, infections, endoleaks, thrombosis, or other surgery required and re-exploration of OSR. These were five times more common with EVAR as with OSR (17.6 vs. 3.3 per 100 person years). The Dutch Randomized Endovascular Aneurysm Management (DREAM) trial reported nearly identical rates of severe (systemic, local-vascular/graft complications or local nonvascular complications) adverse event-free survival in the two groups (16.9 percent [EVAR] versus 19.4 percent [OSR] at year 2). The rate of survival free of moderate or severe complications was similar at two years (65.6 percent for EVAR and 65.9 percent for OSR). In EVAR-1, reinterventions occurred three times as often in the EVAR group, exceeding 20 percent at 4 years. DREAM showed a similar pattern in the first 9 months but then reintervention rates were roughly parallel.

DREAM study data on quality of life and sexual functioning favored EVAR in the early postoperative period, but by 6 months, scores in the OSR group equaled or surpassed those in the EVAR group. In EVAR-1, physical component summary scores in the Short Form-36 (SF-36) were lower with OSR to 3 months, with no differences thereafter. There were no differences at any point in the mental component summary scores. Differences at 3 months did not achieve a level previously determined detectable by patients. EVAR costs were higher than OSR.

The British EVAR-2 study was a methodologically high-quality study and the only RCT that evaluated EVAR versus no intervention for patients with large AAA (≥5.5 cm) and deemed medically unfit for OSR (n = 338). There was a 6 percent higher all-cause mortality in patients receiving EVAR compared with no intervention that did not achieve statistical significance (45 percent vs. 39 percent; HR = 1.21; [95 percent CI 0.87 to 1.69]). AAA rupture occurred in 4 percent of EVAR patients and 12 percent of individuals in the no intervention group (rupture rate in the no intervention group = 9 per 100 person years.) There was no difference in AAA related mortality (12 vs. 13 percent) (HR = 1.01 [0.55 to 1.84]), although nine ruptures and six AAA deaths occurred in the EVAR group prior to elective repair (median time from randomization to EVAR = 57 days). There were no significant differences in mortality for EVAR compared with no intervention according to age, sex, aneurysm diameter, or renal function. The 30-day EVAR mortality was higher in the sicker EVAR-2 patients than those receiving EVAR who were judged medically fit for surgery in EVAR-1 (9 percent versus 2 percent; p <0.0001). If only elective EVAR cases were considered in EVAR-2 the 30-day EVAR mortality was 7 percent. Compared to EVAR-1 patients, there was a greater need for internal iliac artery embolization, blood products, renal dialysis, and longer hospital stay.

Approximately 55 percent of individuals with large AAA screened for trial enrollment in EVAR-1 and 2 were judged to be anatomically suitable for EVAR. The most commonly used EVAR devices in these two trials were Zenith (33–59 percent of devices) and Talent (21 to 33 percent). Over 90 percent of EVAR devices used in these trials were aortobiiliac device systems commercially available in the U.K. at the time of the RCT. Aorto-aortic (tube) grafts were used in the majority of OSR. Selection for EVAR and OSR grafts was based on the discretion of the surgical team. There are no data to determine whether outcomes varied according to device type. The exact make and model of EVAR devices used is not known, and some may not be approved for use or currently available in the United States. Refinements in device, delivery systems, and provider experience may be associated with different outcomes in the United States than reported in these RCTs.

Forty-eight reports of lower methodologic quality than RCT (case series, nonrandomized controlled studies, national registries, and U.S. FDA reports) assessed clinical outcomes of EVAR. While it is not possible to make direct comparisons with RCT, most reports explicitly stated patients were candidates for OSR. Baseline patient characteristics, AAA diameter, 30 day and 2 year overall, and AAA mortality as well as EVAR conversion rates and secondary interventions for included patients were similar to those from EVAR-1 and DREAM. None of these reports assessed EVAR in patients with AAA ≥5.5 cm and considered “medically unfit for OSR” and direct comparisons cannot be made with EVAR-2. One report evaluated outcomes of EVAR and OSR in a “high surgical risk” subgroup of patients entered into any of five nonrandomized multicenter Investigational Device Exemption (IDE) studies leading to U.S. FDA approval of EVAR devices. Patients with AAA ≥5.5 cm were retrospectively categorized as “high surgical risk” based on age >60 years and having at least one cardiac, pulmonary, or renal comorbidity. Inclusion criteria for the IDE studies required that patients were candidates for OSR, though in one study patients were prospectively defined as being poor candidates for OSR or at high risk due to pathophysiologic conditions including creatinine >2.0 mg/dL, disabling COPD, ejection fraction <25 percent, stroke with residual deficit, or myocardial infarction within the past 6 months. Three-quarters of “high-risk” patients had only one comorbid category and less than one percent had all three categories.¹⁴ After 4 years, deaths categorized as due to AAA were similar between EVAR and OSR patients (4.2 percent and 5.1 percent respectively (p = 0.58). Overall-survival in EVAR treated patients was 10 percent lower compared to OSR (56 percent vs. 66 percent) though this difference was not statistically significant (p = 0.23).

Patient treatment preference according to AAA size and medical fitness for surgical intervention is difficult to ascertain. How the results from published RCTs influence future patient and provider treatment preferences in the United States is not known.

Volume Outcomes

Our search did not find any adequate published studies of the relationship between hospital or physician volume and any outcome of EVAR.

There were several historical studies of the relationship between the volume of OSR done by hospitals and physicians to repair AAA and short-term mortality. Many of these studies represented practices in the 1990s when endovascular repair procedures were being developed and later tested in randomized controlled trials. Studies of the volume-outcome relationship for OSR have not been reported since the widespread adoption of EVAR in the United States. Nevertheless, historical studies of OSR of AAA can serve to inform the design of future studies of EVAR. Studies of OSR have not determined why hospital volume appears to be related to short-term mortality. Surgeon volume and surgeon specialty may explain a large portion of the commonly observed inverse relationship between the volume of OSR done by hospitals and short-term mortality. Very low-volume hospitals often appear to have higher mortality; however, data from low-volume providers is highly variable and statistically imprecise. No previous study carefully controlled all the key risk factors that might confound these observational studies, especially the characteristics of the aneurysms that will be important in studies of EVAR. The type of endograft might be another important factor in studies of EVAR outcomes. The one study of OSR that was able to control for preoperative clinical measures did not find a significant association between hospital volume and mortality. Very low-volume hospitals often appear to have higher mortality; however, data from low-volume providers is highly variable and statistically imprecise. Thus, future studies of EVAR volume in relation to outcomes must strive to take all important covariates into account, and consider the imprecision of outcome data for low-volume providers. Two studies suggested that the volume effect for OSR was relatively consistent in subgroups of patients with different preoperative risks of death.^{15, 16} However, studies of EVAR should not assume that any volume-outcome relationship will be constant across subgroups of patients. Arbitrary cutpoints for hospital volume that were associated with lower adjusted mortality ranged from 17 to 100 AAA repairs per year in studies of OSR. Investigators used a variety of measures of volume and mortality and did not attempt to analyze their data to find a threshold(s) for the volume effect. If studies of volume-outcome relationships for EVAR are going to be used by payers or policymakers to define volume thresholds for preferential referral, studies should be designed with standardized measures of volume and outcomes, and the nature of relationships between hospital and physician volume and outcomes needs to be characterized more completely.

Costs

Case series focusing on hospital costs generally found that EVAR costs more to perform than OSR, primarily due to the cost of the prosthesis. The high cost of the EVAR prosthesis is partially offset by reduced hospital and intensive care unit length of stay, operating time, and necessity for blood transfusion relative to open surgery. More comprehensive cost analyses noted the higher follow-up costs for EVAR.

None of the Markov cost effectiveness models had accurate data on the complication and reintervention rates associated with EVAR. Although the Michaels et al. study is based on the literature published through September 2004, several case series have come out since then, and data from midterm results of RCTs were not included.¹⁷ Results from ongoing RCTs, comparing EVAR and OSR for large AAA and EVAR versus surveillance for small AAA conducted in the United States have not yet been published.

Data from RCTs demonstrate that for small AAA (<5.5 cm) immediate OSR costs more and does not improve survival compared to active surveillance and delayed elective intervention. For large AAA (≥5.5 cm) among patients unfit for OSR, EVAR has both greater short- and long-term costs, does not improve overall survival or quality of life beyond 1 year, and is associated with complications, need for reintervention, and long-term monitoring compared to OSR or no intervention. EVAR is associated with shorter hospital and intensive care unit length of stay, reduced AAA mortality, and lower 30-day morbidity and mortality, compared to OSR.

The cost effectiveness of EVAR relative to OSR varies by whether an institutional or societal perspective was taken. Third party payers and hospitals each formulate their own institutional perspective on costs and effectiveness that depends on the extent and duration of their responsibility for the financing and/or provision of the patient's health care needs. For third party payers such as Medicare, hospital reimbursement levels are determined by a DRG that does not vary by the type of inpatient procedure (EVAR or OSR) or the cost of the prosthesis.

Literature Search and Review Strategy

A literature search was conducted using PubMed®, Cochrane Library, FDA, and other electronic websites from 1990 to October 2005, as well as handsearching of references from retrieved articles and contact with content experts. The following terms were included: abdominal aortic aneurysm, endovascular repair, surveillance, and surgery. Titles and abstracts of identified references were reviewed using standardized data abstraction sheets. The number of excluded studies and reasons for exclusion were noted. Studies meeting preliminary eligibility criteria were retrieved in full for further assessment and data extraction. (See Appendix B ^* for Exact Search Strings and Appendix C for Lists of Excluded References.) Data extraction forms included information related to study and patient characteristics, interventions, and outcomes.

Study Eligibility and Extract Data

Studies were eligible if they were randomized controlled trials or natural history studies involving AAA, open repair, endovascular repair, or active surveillance and reported clinically relevant outcomes (Figures 1 and 3 on pages 27 and 29). We also included and updated a systematic review of evidence of EVAR vs. open repair for AAA published online and in manuscript form by the U.K. National Institute of Clinical Excellence (NICE). Data from nonrandomized studies and device-specific data available on the FDA website as submitted by manufacturers to the FDA were included to assess relevance and consistency to the United States. This included a report submitted by the Society for Vascular Surgery and published online in April 2006.¹⁴

Outcomes

Outcomes were described as effectiveness (initial clinical success ≤30 days or during initial hospitalization), short-term (30 days to 6 months), midterm (6 months to 5 years), long term (>5 years), and adverse effects. These were based on recommendations for reporting of outcomes for open and EVAR by the Society for Vascular Surgery.⁸ The primary outcome was midterm all-cause mortality. Additional clinical outcomes included initial all-cause mortality, AAA mortality, AAA rupture, quality of life, and technical measures of EVAR and OSR success. Adverse events included treatment related mortality and morbidity (graft rupture, migration, and kinking) and need for additional interventions (including OSR). Incidences of endoleaks were also noted. Because data from RCT represent level one evidence, their results were emphasized and reported separately from reports using other study designs. Results from the United States that evaluated EVAR approved and available in the United States were highlighted for nonRCT data.

Quality of studies, reduction of bias, and strength of evidence. The methods of Schulz et al. were used to assess the quality of randomized controlled trials (RCT).⁷ We determined whether intention-to-treat analysis was utilized, whether results varied by intention-to-treat or per protocol analysis, and whether published RCT used EVAR devices similar to those approved and available in the United States. We evaluated whether effectiveness of interventions varied according to patient, aneurysm, or device characteristics including age, race, gender, EVAR device manufacturer and characteristics (name, tubular, mono or bi-iliac), aneurysm size (small [<5–5.5 cm] vs. large aneurysm) or medical fitness for OSR. Data from nonrandomized studies were included to assess relevance and consistency to practice in the United States. All data were extracted by trained data extractors onto standardized, piloted forms (Appendix D ^*).

Question 2. What is the relationship of volume, both hospital and physician, to the benefits and harms of endovascular procedures to repair AAA?

History of the relationship between hospital volume and operative mortality for OSR. Volume is believed to be a proxy for one or more structures or processes of care that influence outcomes. However, the structure and processes of care responsible for the volume-outcome relationship in cases of OSR of AAA have not been firmly established. Experience and greater availability of resources in higher volume hospitals could lead to better structure, patient selection, and pre-, intra-, and post-operative processes of care as well as more capable surgeons and support staff. The relationship between hospital volume and survival after OSR of AAA also might depend, in part, on surgeon volume and the type of surgeon (vascular, cardiothoracic, general). Some low-volume providers could perform operations only as necessary or in emergency situations that increase the risk, and a single death could greatly influence the mortality rate of low volume providers.

Conversely, hospitals that somehow achieve better outcomes or those that have a surgeon who specializes in vascular surgery could get more referrals. High volume itself might generate more referrals because information about volume may be more readily available to the community than outcome information. Some providers might be associated with a poor outcome and get fewer referrals. Thus, to some extent, outcome could drive volume rather than vice versa.^{44, 45} Selective referrals could be augmented by recent policies of purchasers such as the Leapfrog Group who encouraged preferred use of hospitals in urban areas that perform more than 30 repairs of AAA per year in an effort to improve outcomes.⁴⁶

Policymakers have already used studies of the volume-outcome relationship to set thresholds for preferential referral for OSR. The Leapfrog definition of ‘high’ volume was based on an investigation that was judged to be the single best report in the literature at the time.⁴⁷ This investigation analyzed 3,419 repairs of unruptured AAA done in 116 Veterans Affairs (VA) hospitals (>99 percent male patients) from 1991-1993.⁴⁸ Repairs of intact AAA were identified by Patient Management Category algorithms applied to ICD-9-CM codes recorded on discharge abstracts. The validity of this method of case identification wasn't reported. Unadjusted in-hospital mortality was 4.2 ± 3.5 percent in hospitals that performed 32 or more repairs versus 6.7 ± 7.8 percent in hospitals performing fewer than 32 repairs. The basis for selecting 32 as a cutoff was not reported, although it was the mean hospital volume. However, 22 so-called low-volume hospitals had zero mortality in 267 patients. In addition, the published report does not clearly state whether the volume categories were per annum as defined by Leapfrog or over the entire 3-year period of study. Furthermore, many of the operations might have been performed by surgical residents in training. Given these limitations, this study is not adequate for defining a hospital volume threshold for the United States. Investigators only recently demonstrated that hospital volume measured in one time period is related to outcomes in a subsequent period of time.⁴⁹ The Leapfrog criterion for a high volume hospital has been increased to 50 or more procedures per year.⁵⁰ It is not clear whether this count includes endovascular procedures to repair of AAA or whether the data used to define this threshold represent a period of time before FDA approval and widespread use of endovascular devices.

Adoption of endovascular procedures to repair AAA. The FDA approved the first devices for EVAR in September 1999. A separate International Classification of Diseases procedure code for EVAR was issued in October 2000. By 2001 approximately 36 percent of the repairs of intact AAA were coded as EVAR procedures in the National Inpatient Sample of 986 non-federal community hospitals in 33 states.⁵¹ Those undergoing EVAR rather than OSR tended to be older males with higher prevalence of hypertension and ischemic heart disease and less renal insufficiency or peripheral artery disease.⁵² Between May 2001 and September 2003 approximately 38 percent of AAA repairs were coded as endovascular procedures in the VA medical centers. Discharge data from 234 accredited hospitals in Illinois indicated that EVAR was used in approximately 32 percent of all (unruptured and ruptured) AAA cases in 2002.⁵³ The proportion of AAA repairs done by EVAR was less in women than men. In-hospital mortality for elective EVAR was 2.3 percent in this series of cases from 1995 to 2003, and higher in women than men (5.1 percent vs. 1.7 percent unadjusted for risk or length of stay).

Hospital discharge data representing non-federal hospitals in New York indicated that 40 percent of all repairs of intact AAA in 2001 were EVAR, increasing to more than 50 percent in 2002.⁵⁴ There was a corresponding increase of more than 20 percent in the number of hospitals performing EVAR. Compared to patients undergoing OSR, patients having EVAR were older and more likely to have coronary artery disease, hypertension, hyperlipidemia, and diabetes mellitus. In 2002, 60 hospitals in New York performed EVAR. Nearly half did five or fewer procedures, while 12 hospitals did 30 or more. In-hospital mortality rates were 1.9 percent and 0.8 percent respectively. This small difference in mortality was not statistically significant given the relatively small number of hospitals. Furthermore, risk adjustment was not done because the database did not contain patient or physician identifiers, information about the severity of comorbidities, or information about the aneurysms that were repaired.

Rationale for a volume-outcome relationship for endovascular repair of AAA. There is some evidence of a learning curve for EVAR of AAA. One practice at a tertiary facility analyzed 277 consecutive cases treated from 1994 to 1998 and estimated that approximately 55 cases were required before their technical success rate leveled off at 88 percent.⁵⁵ Custom-made endografts were used in two-thirds of these procedures prior to the availability of commercial bifurcated modular endografts for investigational use in 1997. The surgical team and patient selection criteria remained stable during the period of observation, although it is likely they were not trained to perform EVAR and learned as they were developing new procedures and devices. Thus, their experience should not be extrapolated to current EVAR practices where trained physicians use approved devices with less problematic designs. The same referral practice compared 30 procedures done as part of a clinical trial to 230 cases done after FDA approval of the device. Technical success was achieved in 97 percent and 98 percent of these cases. However, procedural complications such as endoleaks, contrast-induced azotemia, access site hematoma, and limb occlusions increased from 11 percent to 18 percent. The increase in complications was attributed not to poorer proficiency over time, but rather to a greater degree of anatomical and morphological difficulty of the cases being referred and treated after FDA approval.⁵⁶

European registry data were grouped according to cumulative experience in 93 hospitals from 1994 to 2000. The EVAR procedures (n=2,863) were grouped as the first 11 performed at a center, cases 12–37, 38–91, and 92 and higher. Unadjusted mortality after 30 days was the same (3–4 percent) in all four groups.⁵⁷ There were differences in smoking status, vascular morphology, and types of devices used in the four groups with initial patients appearing to have greater risk. Co-morbidity was not reported. The adjusted hazard of death during continued followup was lower in quartiles three and four (i.e., after 38 cases) compared to first quartile representing early experience. The adjusted hazard ratio for secondary interventions for endoleaks, migration, kinking, thrombosis, stenosis, and rupture was lower in all other quartiles compared to the most inexperienced case quartile. The amount of bias present in these comparisons due to differences in patient selection and censoring (initial cases followed much longer) is not clear. A more recent report based on this registry suggested that devices for earlier EVAR procedures might explain the influence of experience on midterm outcomes.⁵⁸ This observation highlights the importance of analyzing data that represent current devices and clinical practices in the United States rather than foreign data on devices that are not being used in the United States.

One should not assume there would be a relationship between EVAR volume and outcome just because there is a learning curve. When development of a new technique is fairly complete, teachers can greatly facilitate learning by experience. Measures of hospital and physician volume typically used in studies are not cumulative counts beginning with initial experience. In fact, volume is often measured during a specified time period ignoring prior experience. Therefore, typical volume measures may reflect a provider's ability to maintain rather than learn skills and effective processes of care. The volume needed to maintain processes of care, skills, and optimal outcomes after a learning curve levels off is not known.

Several randomized controlled trials have required that a certain number of EVAR be done by investigators prior to enrolling patients, but the threshold definitions vary. The EVAR trials required investigator centers to perform at least 20 AAA repair procedures before enrolling patients, whereas investigator teams in the DREAM trial had to have done at least five procedures.^{12, 59} The Veterans Administration Open Versus Endovascular Repair (OVER) trial requires a vascular surgeon or interventional radiologist who has performed at least 12 endovascular procedures.⁶⁰ Thus, investigator experience required for participation in clinical trials has varied. The bases for these volume criteria were not reported.

Analytical issues. Outcome measures that are based on a smaller number of cases (i.e., low volume) exhibit greater random variation. One is less confident that the outcome measured for low volume providers represents their true or long-term performance. A minimum number of cases per provider is sometimes required to help assure the outcome measurements aren't too unreliable. Regression analyses that weight data points by their variance can help avoid placing undue emphasis on outcomes measured in low-volume providers. Regression analyses generally assume that the variation in outcome measures is homogeneous across all levels of a predictor, which is generally not true when volume is used to predict outcomes. Data transformation, such as taking the logarithm of outcome measures, might make this assumption more tenable. Furthermore, the influence of each hospital or physician on the estimated volume-outcome relationship estimated by regression analysis should be examined.

Patients treated at the same hospital or by the same surgeon may be more likely to have similar outcomes than if the patients were treated by different providers. Thus, outcomes might not be truly independent of each other as is assumed by regression analyses that use the patient as the unit of analysis. This so-called clustering with hospitals and physicians reduces variation in outcomes compared to what would be observed with truly independent measurements.

General estimating equations can be used to calculate confidence intervals and p-values that take clustering into account.⁶¹ This analytical approach is difficult when the outcome data are clustered within more than one variable, e.g., hospital and surgeon volume, and when the size of the clusters, i.e., volume, is correlated with the outcome. In addition, the resulting estimate of relative effects such as an odds ratio is an aggregate or population averaged estimate specific to the mix of clusters in the analysis. Alternatively, random-effects regression models that consider clusters, i.e., hospitals or surgeons, as random variables can account for clustering in the data. Small clusters (low-volume providers) can complicate this approach. Clustering within surgeons and hospitals could be analyzed by hierarchical random-effects models if surgeons don't operate at more than one hospital. If a surgeon(s) represents most of the volume at a hospital, the surgeon and hospital volume will be redundant measures, making it difficult to estimate their separate effects on outcomes. Thus, different methods of analysis of the same data might not produce the same estimate of the effects of volume on outcomes. The method of analysis is an important consideration when comparing studies. Multiple methods of analysis could be employed to see if the results are sensitive to the method of analysis.

The adequacy of risk adjustment is questionable in retrospective studies of databases that did not prospectively collect data on known risk factors. Known risk factors for operative mortality for AAA repairs include the urgency of the operation, presence and severity of renal dysfunction, ischemic heart disease and its treatment with beta-adrenergic blockers and statins, heart failure and chronic obstructive pulmonary disease, advanced age, being female, and several characteristics of the aneurysm such as its diameter and juxtarenal position.^{52, 62} Unmeasured risk factors could be particularly important for low-volume providers. One unrecognized high risk case can greatly affect the outcomes, and if the variable that made the case high risk wasn't measured, the increased risk would not be controlled. Omission of influential covariates can bias estimates of volume-outcome relationships even if they are balanced across groups. Thus, it is important to examine the risk factors that were taken into consideration in analyses.

All studies of the relationship between provider volumes and outcomes used regression models to analyze the data. Therefore, it is important to know how well the model fits the data and how well the model discriminated the outcomes, such as death or survival. A model with poor discrimination suggests that key risk factors were not measured, hence controlled, in the analysis. The c-statistic is commonly used to examine model discrimination. A c-statistic equal to 1 indicates perfect discrimination whereas a value of 0.5 indicates the model was no better than random predictions of outcomes.

Search Strategy for Pertinent Publications

The MEDLINE® database was searched for publications about volume-outcome relationships for procedures to repair AAA that were published after 1990 when endovascular procedures for repair of AAA became available, at least to investigators. Key words included abdominal aortic aneurysm, volume, outcome, and process assessment health care. Reference lists were searched for additional reports (Figure 2 on page 28).

Study Inclusion and Exclusion Criteria

Articles that met the following criteria were reviewed to summarize study methods and characterize volume-outcome relationships.

• The report had to be an original analysis of data representing repair of unruptured AAA in the endovascular era beginning in 1990.

• The report had to represent practices in the United States.

• The sample had to represent variation between hospitals or surgeons in a community or larger geographic area, thereby excluding single site cases series.

• The report had to present sample statistics (e.g., percentages, odds ratios) representing the relationship between a measure of hospital or physician volume and any good or bad outcome associated with AAA repair.

• The analysis had to attempt to make adjustments for known risk factors in an effort to reduce bias.

Health care systems, physician training, clinical care processes, and procedures to repair AAA have evolved, thus older and foreign data might not represent current practices or outcomes in the United States. Since health care practices including patient selection criteria, referral patterns, and types of devices sold outside the United States could differ from the U.S. in important ways, this review was restricted to studies within the U.S. This restriction corresponds with the presumed interests of U.S. insurance companies that requested this evidence review. Furthermore, inclusion of foreign studies would not address the limitations of U.S. studies, would create concern about extrapolating foreign results to the United States, and would not alter the conclusions of this review. More recent recommendations in the United States to seek out higher volume hospitals could be altering the volume-outcome relationships. In addition, adoption of EVAR in recent years most likely has altered the characteristics of cases being surgically repaired, and perhaps the volume-outcome relationship. This review focused on available data that represented practices and outcomes in the United States in the 1990s and beyond.

Data Extraction

All published analyses regressed an outcome of AAA on retrospectively defined volume measures. A single unblinded reviewer abstracted information from each study to characterize (1) the study population including the time period and geographical region, (2) volume measures, (3) outcome measures, (4) regression model including covariates used for risk adjustment and handling of clustering of patients within hospitals or surgeons, and (5) adjusted and unadjusted estimates of volume effects.

Data Analysis

The preponderance of evidence for relationships between the volume of procedures performed by hospitals and physicians to repair AAA and outcomes has been obtained by secondary regression analysis of administrative databases. Methodological differences make it difficult to compare results from the different studies. For example, investigators defined hospital and physician volume in a variety of ways, often using arbitrary categorizations with varying reference groups. Use of variables to control for pre-procedure risk (case-mix), physician training, and health care system attributes varied. Analytical methods varied as well. Therefore, we did not attempt to do a meta-analysis to combine estimates from different studies. This review was limited to tabulation of individual study characteristics and results.

Question 3: How do the characteristics of the aneurysm (size/location/shape) and the patient (age/gender) affect the benefits and harms of endovascular and open-surgical repair?

It is not known whether patient, aneurysm, provider, or device characteristics differentially affect treatment success rates. As discussed above, patient factors including age, gender, and comorbidities may not only affect the prevalence and size of AAA but also the risk of AAA rupture and/or operative/EVAR related complications. As noted in Question 1, AAA morphology including size, vascular angulation, size of the aortic neck, and/or involvement of renal or iliac vessels may alter the feasibility of repair options and influence outcomes.

Literature Search Strategy and Eligibility Criteria

A literature search strategy was conducted similar to question 1 (Figures 1 and 3 on pages 27 and 29). Emphasis was placed on studies that provided comparative effectiveness of one treatment strategy versus another (especially RCT), conducted in the United States, used devices currently available in the United States, and provided information on clinical outcomes according to AAA size, location, shape, patient age, gender, or race. We updated a systematic review by NICE⁶³ with emphasis on results from RCTs or studies conducted in the United States.

Outcomes

Data regarding study, patient, and device characteristics, as well as outcomes, were extracted. Outcomes were described as effectiveness (initial clinical success ≤30 days or during initial hospitalization); short-term (30 days to 6 months), midterm (6 months to 5 years), and long-term (>5 years); and adverse effects based on recommendations for reporting of outcomes for open and EVAR by the Society for Vascular Surgery.⁸ The primary outcome was midterm all-cause mortality. Additional outcomes included initial all-cause mortality, AAA mortality, AAA rupture, quality of life, and technical measures of EVAR, and OSR success. Adverse events included treatment related mortality and morbidity including endoleaks, graft rupture, migration, kinking, and need for additional interventions, including OSR. Specific characteristics of AAA and patient included: 1) AAA size (small <5.5 cm; large ≥5.5 cm), shape (sacular vs. fusiform), extension (into iliacs); 2) patient age (<65 vs. ≥65), race (White, Black, Hispanic, other); gender; and 3) device manufacturer and type.

Quality of studies, reduction of bias, and strength of evidence. We used the methods of Schulz et al. to assess the quality of RCT.7 We attempted to assess whether the effectiveness of interventions varied according to patient, aneurysm, or device characteristics including age, race, gender, EVAR device manufacturer, aneurysm size (<5.5 cm vs. ≥5.5 cm), or aneurysm location. No study quality measures were used for manufacturer submitted data available on the FDA website. We described how patients, AAA, EVAR characteristics, and outcomes compare with those from RCT.

Question 4: What are the costs-benefits for each of the procedures?

From the perspective of cost, different audiences have different concerns. Traditional economic analysis focuses on the cost to the economy of delivering each element of service. However, policy makers may be more interested in the payments they must underwrite than in the economic cost of a procedure. These costs vary somewhat depending on the health care system. For example, in the United States, where Medicare pays for hospital care using a DRG payment approach, differences in the costs of the prosthesis/graft or even in lengths of stay are irrelevant for Medicare because they are all folded into the overall rate paid. However, those differences are very salient to hospitals that must bear them. Likewise in the United Kingdom, working under the National Health System, issues of hospital efficiency are quite important.

The analysis becomes even more complex if a differential survival rate exists across treatments. Unless the results can be expressed in terms like QALYs, which capture both quality of life and survival, the estimated benefit may be exaggerated by comparing quality of life only among survivors.

The classic approach would be to estimate the difference in outcomes across treatments and then divide that by the difference in costs. However, in the case of AAA repair, the outcomes seem to be generally equivalent over the long term. The real difference lies in the short term.

EVAR is considered less invasive and may result in shorter in-hospital and intensive care unit length of stay compared with open repair. However, there are additional costs associated with EVAR. These include the high costs the stent device (which has accounted for over 50 percent of the total in hospital EVAR costs in one analysis) diagnostic cost, primarily because of increased imaging costs with EVAR, and professional fees.⁶⁴ Pre and intraprocedural imaging are more extensive with EVAR than open repair. Because of the concern related to EVAR complications or failures, long term radiologic monitoring is required. The complexities of and recommendations for conducting cost-effectiveness analyses comparing treatment options for AAA have been described.^{22, 65}

Search Strategy Methods

We adopted the search strategy used by the AHRQ report on AAA screening.²² Our search was conducted within MEDLINE®. Key search terms included abdominal aortic aneurysm combined with economics, nursing economics, pharmaceutical economics, cost, pharmacoeconomics, cost analysis, cost allocation, cost-benefit analysis, cost control, cost savings, cost of illness, cost sharing, “deductibles and coinsurance,” medical savings accounts, health care costs, direct service costs, drug costs, employer health costs, hospital costs, health expenditures, capital expenditures, hospital economics, hospital charges, hospital costs, medical economics, or medical fees. We focused on studies reporting on the cost associated with either procedure, EVAR, OSR, or both. This strategy yielded 27 articles published between 2000 and 2005. Six of these studies were eliminated from the analyses as they contained less than 50 patients in either arm of the study (i.e., EVAR or OSR).⁶⁶–⁷¹ Of the remaining 21 articles, 12 reported on a comparative analysis of EVAR and OSR, two reported on EVAR and Conservative Management (CM) (one of which also reported on EVAR and OSR), three on EVAR only, two on OSR and CM, and three on OSR only (Figure 4 on page 30).

Baseline Characteristics (Appendix E ^*, Table E1)

Baseline characteristics and a description of the study protocol are shown in Appendix E, Table E1. ADAM randomized 1,136 patients fit for elective OSR with an AAA between 4.0 to 5.4 cm in diameter out of a total of 5,038 patients screened and referred for recruitment. The majority of patients excluded had AAA outside the eligible parameters, severe comorbid conditions, or were judged to be unlikely to adhere to the study protocol. Patients were randomized to immediate OSR (n=569) or to undergo surveillance (n=567) by ultrasonography or computed tomography (CT) every 6 months. Patients in the surveillance group were followed without repair until: 1) AAA reached 5.5 cm in diameter; 2) enlargement of AAA by at least 0.7 cm in six months or 1.0 cm in 1 year; or 3) the development of symptoms attributable to the AAA by the attending vascular surgeon.

Enrolled patients were ages 50 to 79 years with a mean age of 68 and a baseline mean AAA diameter by CT scan of 4.7 cm. Comorbid conditions included hypertension (56 percent), coronary artery disease (42 percent), and diabetes (10 percent). Past or current smoking status was greater than 90 percent. Patients were overwhelmingly male and white race.

UKSAT randomized 1,090 out of 1,276 eligible patients with non-tender infrarenal AAA between 4.0 to 5.5 cm. Patients fit for elective OSR were randomly assigned to early OSR (n=563) or ultrasonographic surveillance (n=527). OSR in the surveillance group was recommended if patients met the following criteria: (1) AAA exceeded 5.5 cm in diameter; (2) enlargement of AAA >1.0 cm per year; (3) AAA became tender; or (4) iliac or thoracic repair of an aneurysm was needed. Patients had a mean age of 69 years (range 60 to 76) and a mean AAA diameter of 4.6 cm. The majority of patients were male (83 percent) and 37 percent were current smokers. Hypertension was prevalent in 38 percent, ischemic heart disease in 40 percent. Over 90 percent of patients had a history of tobacco use with 37 percent still smoking.

Followup and Treatment

The mean duration of followup was 4.9 years for ADAM and 4.6 years for the initial report of the UKSAT. Additional analysis of the UKSAT provided results at a mean followup duration of 8 years (range 6–10). In both trials approximately 60 percent of patients randomized to active surveillance underwent OSR sometime during the trial. The vast majority of delayed OSR were because AAA achieved predetermined criteria; typically AAA size >5.5 cm. In ADAM, AAA repair was performed in 93 percent of patients in the immediate OSR group, 72 percent within 6 weeks after randomization. In the active surveillance group, 62 percent of patients in the surveillance group had undergone AAA OSR by the end of the trial, and 9 percent were performed despite the fact that participants did not meet study criteria for OSR. Rate of repair in the surveillance group increased with the size of the aneurysm at baseline. Four years after randomization, 27 percent of AAA that had measured 4.0 to 4.4 cm at baseline had OSR as compared with 81 percent of those with baseline AAA 5.0 to 5.4 cm.

In UKSAT, 92 percent assigned to early repair had undergone OSR by the end of the trial and 87 percent within 5 months of randomization. Within the surveillance group, 62 percent had undergone repair, (82 percent of all delayed OSR were done according to protocol). The median time to surgery was 2.9 years. At the 8 year mean followup period, an additional 1 percent in the early repair group and 12 percent in the surveillance group had AAA repair. Treatment between the initial and longer followup report did not necessarily adhere to protocol. Approximately one of five patients in the surveillance group (105 of 527) died without undergoing AAA repair.

Mortality (Tables 2 and 3 on pages 42 and 43 and Appendix E ^*, Table E2)

All-cause mortality. Early/immediate OSR of AAA did not significantly reduce all-cause mortality compared with surveillance in either trial. After a mean followup of approximately 5 years, 27 percent of patients randomized to early OSR had died compared with 25 percent of the surveillance and delayed OSR group. Comparing immediate OSR to surveillance, the RR of death in the ADAM trial was 1.21 [95 percent CI 0.95 to 1.54] and the HR for the UKSAT was 0.94 [95 percent CI 0.75 to 1.17, p=0.56]. After a mean followup of 8 years, the adjusted hazard ratio was 0.83 [95 percent CI 0.69 to 1.00; p <0.05] with an estimated 7.1 and 8.3 deaths per 100 patient years for the early repair and surveillance groups, respectively.

In UKSAT (but not ADAM) mortality within the first 6 months of randomization was significantly higher in the early repair group compared with patients in the surveillance group (HR of 2.52 [95 percent CI 1.20 to 5.33] and absolute risk difference of 3 percent) (Appendix E ^*, Table E2). The 30-day mortality rate was approximately 6 percent for patients who underwent elective OSR and did not vary by time of repair. After 3 years, mortality was higher in the surveillance group in UKSAT but not ADAM. At the 8 year followup period, the mean duration of survival in the early OSR group was 6.7 years compared to 6.5 years for the surveillance group. Despite a mortality of 2.7 percent within 30 days of surgery or during hospitalization, the immediate OSR group of ADAM had lower cumulative survival compared to the surveillance and delayed OSR throughout the followup period. Independent predictors of death included higher serum creatinine level, lower weight, diagnosis of Chronic Obstructive Pulmonary Disease (COPD) or diabetes, larger AAA diameter, lower forced expiratory volume in one second, and nonuse of a beta-blocker.

Aneurysm-related mortality and rupture. (Table 2 on page 42) Aneurysm-related mortality was not reduced by early/immediate repair in the ADAM study (3.3 percent vs. 3.4 percent) but as defined differently in UKSAT was lower in the long-term followup report (7.5 percent vs. 11.4 percent.

In ADAM, 0.4 percent in the immediate OSR group and 1.9 percent in the surveillance group had AAA rupture. UKSAT assessed AAA rupture and death from OSR. The latter was defined as occurring within 14 days of OSR. The total rupture rate was 1.6 percent per year in the first 5 years of followup and 3.2 percent per year in the subsequent 3 years. Four percent of patients (8 percent of all deaths) in the surveillance group study died due to ruptured AAA compared to 1.8 percent (4 percent of all deaths) in the early OSR group after 8 years of followup.

Aneurysm related mortality in UKSAT, defined as deaths due to ruptured AAA, secondary AAA rupture, or AAA repair accounted for 19.3 percent vs. 15.3 percent of all deaths in the surveillance and early OSR groups respectively. Death associated with aneurysm-related disorders following OSR occurred in 19 patients, 15 in the surveillance group compared to four in the early repair group (p <0.001). Of these 19 deaths, three resulted from secondary AAA rupture, four from aortoduodenal fistula, and 12 from a ruptured thoracic aortic aneurysm. The risk of rupture was four times as high among women as among men. Additionally, fatal ruptures were more common in women than men; resulting in 14 percent of deaths for women versus 5 percent of deaths for men (p = 0.001). However, deaths for women from any cause were similar in the two treatment groups (8.4 percent versus 7.3 percent respectively; p = 0.99).

All-cause mortality according to diameter of aneurysm, age, gender, and smoking status. (Table 3 on page 43) Both trials reported no benefit from early OSR of AAA <5.5 cm in diameter or in any subgroup of patients defined by aneurysm diameter at entry. In UKSAT the reported test of interaction was not significant (p = 0.28) across AAA size or age (p = 0.18) or gender (p = 0.40). Calculated risk differences, reported relative risks and tests for interaction from ADAM were not significant and had point estimates favoring surveillance. In UKSAT, the overall death rate for patients who continued to smoke was 12.0 per 100 patient-years compared to 3.8 per 100 patient-years for patients who no longer smoked. However, these results were not reported according to randomized treatment assignment.

Operative and in-hospital mortality. The age and sex adjusted 30-day and in-hospital operative mortality did not differ in either the UKSAT or ADAM trial between patients receiving immediate OSR or those initially managed with surveillance but receiving delayed OSR. In UKSAT the 30 day operative mortality was 5.8 percent in the early repair group versus 7.1 percent in the group initially managed with surveillance (p = 0.30). In-hospital mortality was 5.8 percent and 7.2 percent respectively. Direct comparisons of these percents should be done cautiously because of differences between groups at the time of surgery in terms of age, aneurysm size, and number of tender or ruptured AAA at the time of OSR. In the ADAM study both the 30-day and in-hospital mortality were below 3 percent in the immediate OSR group and the surveillance group. Thus the finding in ADAM of no survival benefit with early OSR was found despite a low total operate mortality in the immediate repair group.

Complications of OSR (Appendix E ^*, Table E3)

AAA-related hospitalizations (besides those for the elective OSR) occurred more than twice as often for patients undergoing immediate OSR (44.8 percent) versus those randomized to surveillance (22.7 percent). Major complications of OSR with no operative death of unruptured AAA occurred in 4.5 percent of immediate repair patients and 7.6 percent of patients in surveillance who underwent delayed OSR. More than 50 percent of patients in either group who underwent OSR experienced “any complication.” Rehospitalizations for complications were slightly higher for early OSR versus surveillance (20.5 percent versus 16.5 percent). Late graft failure occurred in less than 0.5 percent of patients in either arm.

Rupture and Enlargement of Small Aneurysms in Patients Kept Under Ultrasound Surveillance

After nearly 5 years of followup, there were 25 and 14 total ruptures in UKSAT and ADAM, respectively. In ADAM, there were 12 ruptures in the surveillance group, resulting in eight deaths, yielding a rupture rate of 0.7 percent per year of followup of unrepaired aneurysms. The mean risk of rupture in the UKSAT study was 1.0 percent per year. The rate of aneurysm enlargement was 0.32 cm (interquartile range (IQR) 0.16 to 0.42) and 0.33 cm (IQR 0.20 to 0.53) per year for the ADAM and UKSAT surveillance groups, respectively. After 8 years of followup, UKSAT assessed total rupture rates (including AAA ≥5.5 cm and non-fatal ruptures) for two time periods. The total rupture up to the end of the trial was 1.6 percent per year and during the final 3 years of followup the rate was 3.2 percent per year (p = 0.008).

UKSAT investigators assessed risk factors associated with AAA rupture in 2,257 patients enrolled in the UKSAT or an associated study who were kept under ultrasound surveillance (data not shown in tables).⁷⁵ In addition to UKSAT participants, eligible patients analyzed had aortic diameter <4.0 cm (n=507), or ≥5.5 cm (n=100), refused randomization (n=122), were considered unfit for surgery (n=340), or other reasons (n=98). Ninety-eight percent of patients had initial aneurysm diameters between 3–6 cm. Three-quarters of the ruptures occurred in patients with AAA ≥5 cm in diameter. After 3 years of followup, the annual rate of AAA rupture was 2.2 percent (95 percent CI = 1.7 to 2.8). The risk of rupture was associated with female sex, larger initial aneurysm diameter, current smoking, and higher mean blood pressure. Women had a threefold higher risk of aneurysm rupture than men. Crude rupture rate in women was 4.6 per 100 person years vs. 2.0 per 1,000 person years in men.

Quality of Life (Appendix E ^*, Figures E1–E3 and Appendix E, Table E4)

Any differences in quality of life between treatment groups were small. ADAM utilized the Short Form-36 (SF-36) health status questionnaire in addition to measuring impotence and maximal activity level. The SF-36 scores for the physical and mental subscales are shown in Appendix E, Figure E1. There were no significant differences in any of the subscales between the immediate OSR and surveillance group with the exception of general health. All subscales showed a significant decrease throughout the followup period regardless of treatment group. The general health score was higher (p <0.001) at individual time points from 6 months to 2 years for the immediate OSR group. There were sporadic time point comparisons in other subscales that were significant, mental health at 6 months favoring immediate OSR and physical function at 5 years and role-physical at 6.5 years favoring the surveillance group. These should be interpreted cautiously due to the large number of comparisons. The baseline value for physical functioning was determined to be an independent predictor of mortality during the study duration.

The prevalence of impotence by treatment arm over the study duration is shown in Appendix E, Figure E2. Approximately 40 percent of all patients were impotent before OSR. Following randomization, impotence increased in the immediate OSR group between 18 months to 4 years (p <0.03), with the exception of 12 months. Maximum activity level did not significantly differ between the study groups. A significant interaction between treatment and followup (p <0.02) indicated a greater decline in maximum activity level over time in the immediate repair group.

Health-related quality of life status for UKSAT was assessed by the Medical Outcomes Study short-form 20-item questionnaire. There were no significant differences between groups in the mean change in scores from baseline at 12 months after randomization with the exception in current health perceptions subscale which favored the early repair group. The weighted mean difference in the mean change of scores from baseline between early repair and surveillance was 6.70 [95 percent CI 3.45 to 9.95] (Appendix E, Table E4, Appendix E, Figure E3). Within the surveillance group, mean score changes between baseline and 12 months after randomization for physical functioning, role functioning, social functioning, and bodily pain subscales decreased. In the early OSR group only physical functioning decreased significantly, but the health perceptions subscale improved significantly.

Large Aneurysm (≥5.5 cm) Randomized Controlled Trials

Three RCTs that enrolled patients with AAA ≥5.5 cm and considered medically fit for OSR have been completed.¹¹–¹³ None were conducted in the United States. EVAR devices used may not have been approved for, or are currently in use, in the U.S. Two of the trials, British Endovascular Aneurysm Repair-1 (EVAR-1) and DREAM, recently published midterm (2–4 year) results. EVAR-1 and DREAM report on procedures performed from 1999-2003. The Cuyper's trial objective was to compare cardiac response after EVAR and OSR. Cardiac complications in both groups were assessed up to 1 month postoperatively, and no long-term outcomes were reported. Two additional trials are undergoing recruitment, the Open Versus Endovascular Repair (OVER) in the U.S. and the AnEVARysme de l'aorte abdominal: Chirurgie versus Endoprothese (ACE) project in France. OVER has a recruitment goal of 900 and ACE is planning on enrolling 600 patients. The VA OVER trial will compare procedures done from 2002 to 2007. OVER is unique in that the specific device the patient will receive if randomized to EVAR will be recorded before randomization, thus allowing for subgroup analysis by EVAR compared with randomized controls for that graft. OVER will include a cost-effectiveness analysis of U.S. utilization and costs.

EVAR Versus OSR Patients Who Are Medically Fit for OSR

Study and patient characteristics. (Appendix E ^*, Table E5). Three trials comparing EVAR with OSR randomized a total of 1,489 patients¹¹–¹³ In the largest of the three trials, EVAR-1, 2,068 patients were anatomically eligible for EVAR, of which 52 percent gave consent and were randomized. Reasons for exclusion included: 1) refusal of further assessment following initial screen (n=327); 2) deemed unfit for OSR after local fitness assessment and offered EVAR-2 (n=399); and 3) refusal to participate (n=260).

Eligible patients had to be candidates for either OSR or EVAR based on aneurysm size (at least 5.0 cm) and anatomy as well as surgical risk. Over 90 percent of enrollees were male with an average age of approximately 70 years. Mean AAA diameter ranged from 5.4 cm in the small study by Cuypers to 6.5 cm in EVAR-1. More than 40 percent had a history of cardiac disease, 10–16 percent had diabetes, and the majority a history of tobacco use. All studies reported analyses comparing groups by intention-to-treat. EVAR-1 reported “per protocol” results.

Short-Term Mortality and Morbidity (Table 4 on page 44, Figure 5 on page 49, and Appendix E, Table E6)

EVAR resulted in a lower postoperative 30-day mortality compared to OSR. Pooled 30-day mortality was 1.6 percent for EVAR compared with 4.7 percent for OSR (Risk Difference [RD] = -3, 95 percent CI -5 to -1; RR = 0.34, 95 percent CI 0.17 to 0.65) (Table 4 and Figure 5). In-hospital mortality was similar, 1.9 percent for EVAR vs. 5.8 percent for OSR (RR = 0.32 [0.17 to 0.59]). Of the nine deaths within 30 days in the EVAR group reported in EVAR-1, one occurred after emergency OSR for AAA rupture and one after from rupture following elective OSR. In the OSR group, there was one death after AAA rupture and emergency repair within 30 days.

DREAM reported more severe or moderate systemic operative complications for the OSR group, 26.4 percent versus 11.7 percent for EVAR, an absolute risk reduction of 15 percent [95 percent CI -23 to -7]. A higher rate of pulmonary complications in the OSR group accounted for the majority of this difference. There were significantly more local vascular or implant related complications in the EVAR group, 16.4 percent compared with 8.6 percent for the OSR group (RR = 1.90, 95 percent CI 1.05 to 3.43). In addition, there were three conversions to OSR.

Endoleaks and Secondary Interventions (Table 5 on pages 45–46 and Appendix E ^*, Table E6)

Secondary intervention occurred in 9.8 percent of the EVAR group during primary admission in EVAR-1, including ten conversions to OSR and 18 endoleak corrections. The OSR group had 5.8 percent secondary interventions during initial admission, half for re-exploration of OSR.

Midterm Mortality and Morbidity (Table 4 on page 44, Figures 6 and 7 on pages 50 and 51, and Appendix E, Table E6)

In both EVAR-1 and DREAM, reduction in all-cause mortality seen with EVAR at 30 days had disappeared after 2 years (Table 4 and Figure 6). Pooled all-cause mortality was similar between EVAR and OSR groups, 12.7 percent versus 13.2 percent (ARR = -1. 95 percent CI -4 to 3), respectively (DREAM, EVAR-1). It is unclear whether this was just the endovascular group catching up on delayed deaths, or if late mortality will continue to increase with EVAR due to complications or failure to prevent rupture. Both studies reported some advantage for EVAR in aneurysm-related deaths at 2 years (Figure 7), but this may be because all early post-operative deaths (mostly open) are classified as aneurysm-related, whereas late aneurysm-related deaths (mostly EVAR) may be easily misclassified as something else (Table 4 and Figure 6). Overall, 2.5 percent of EVAR deaths were classified as aneurysm-related compared with 5.6 percent for the OSR control, risk reduction of 3 percent [95 percent CI -5 to -1]. The EVAR-1 trial presented results at 4 years after randomization. All-cause mortality was similar between EVAR and OSR, but aneurysm-related mortality continued to be significantly lower in the EVAR group, 3.5 percent versus 6.3 percent (ARR = -3, 95 percent CI -5 to 0). No significant interactions were reported for all-cause or AAA-related mortality with age, sex, AAA diameter, or creatinine concentration subgroups.

In DREAM the rate of survival free of moderate or severe complications was similar in the two groups at two years (65.6 percent [EVAR] versus 65.9 percent [OSR]). No details on specific adverse events, including graft complications were reported. No documented postoperative AAA ruptures were reported, although there was one possible but unconfirmed case.¹²

Midterm Endoleaks and Secondary Interventions (Table 5 on pages 45–46, and Appendix E, Table E6)

At 4 years after randomization, the overall rate of complications in EVAR-1 was five times as common with EVAR compared with OSR, 17.6 vs. 3.3 per 100 person years. However, a large percentage of these complications included radiographic detected endoleaks which are often asymptomatic and do not require a reintervention. Complications occurred in 35 percent of successfully completed EVAR procedures compared to 8.4 percent of OSRs completed [absolute risk difference of 27 percent; 95 percent CI 22 to 31] (Table 5). The most frequent complications were endoleak type 2 (42 percent), endoleak type 1 (15 percent), graft migration (6 percent) and graft thrombosis (6 percent). Reinterventions occurred three times as often in the EVAR group, exceeding 20 percent at 4 years in contrast to 6 percent for the OSR group. In DREAM the rate of reintervention after EVAR was almost three times the rate after OSR during the first nine months of followup. Thereafter, reintervention rates similar (Appendix E ^*, Table E6). Most reinterventions reported in EVAR-1 were endovascular, primarily repairing endoleaks. In the OSR group, reexploration of OSR was the most common re-intervention (44 percent of reinterventions). Overall, 14 EVAR patients required conversion to OSR, including eight completed after initial discharge from the hospital.

Health Related Quality of Life (Appendix E, Figures E4–E8)

The DREAM study reported on quality of life (Appendix E, Figures E4 and E5) and sexual function from the first 153 patients who underwent randomization. The preoperative scores were comparable to the general population of similar age. The physical function but not mental health scores of the SF-36 favored the EVAR group in the early postoperative period, but by six months, scores in the OSR group equaled or surpassed those in the EVAR group.^{80, 81} Similar findings were reported for sexual function.

In EVAR-1, EQ5D and SF-36 physical component summary scores were lower with OSR to 3 months, with no differences after or at any point in the SF-36 mental component summary scores (Appendix E, Figures E6–E8).

EVAR Versus No Intervention for AAA ≥5.5 cm in Patients Judged Medically Unfit for OSR (Appendix E, Table E7)

Study and patient characteristics. The EVAR-2 study is the only RCT that has evaluated EVAR versus no intervention for AAA ≥5.5 cm among individuals judged to be medically unfit for OSR (Appendix E, Table E7). This study was conducted at the same clinical centers and in parallel with EVAR-1. It enrolled 338 patients at 31 hospitals in the United Kingdom. Each center had to have performed and submitted to the RETA registry at least 20 EVAR procedures. Patients ≥60 years of age, with AAA diameter ≥5.5 cm were considered for participation in the EVAR-1 trial. If they had suitable vascular anatomy for EVAR and were judged medically unfit for major surgery as determined locally by surgeon, radiologist, anesthesiologist, and cardiologist they were offered randomization to EVAR or no intervention. In general, unacceptable surgical risk was due to severe cardiopulmonary conditions. Nearly 90 percent of enrollees were men. The mean age was 76.4 years and mean AAA diameter = 6.3 cm. Local centers could select the graft. Zenith grafts were inserted in 59 percent of patients, Talent in 21 percent, Excluder in 7 percent, and Aneurex in 6 percent. Bifurcated grafts were used in 87 percent of patients. Further description of the grafts was not provided. The mean duration of followup was 3.3 years. One subject randomized to receive EVAR declined and 47 patients randomized to no intervention received treatment with EVAR (35) or OSR (12). The stated reason (n) for repair included rupture (2); tender AAA (11), fast AAA growth (5), became fit for OSR (1), patient preference (14); unknown (14).

Overall and AAA Related Mortality (Tables 6 and 7 on pages 47 and 48 and Appendix E ^*, Tables E8 and E9)

All-cause mortality was higher in patients receiving EVAR compared to no intervention though not statistically so (44.6 percent vs. 39.5 percent; HR = 1.21; [95 percent CI 0.87 to 1.69; p = 0.25]) (Table 6). Overall mortality at 4 years for both groups combined was 64 percent according to Kaplan-Meier estimates (Table 7). AAA rupture occurred in 3.9 percent EVAR patients and 12.2 percent in the no intervention group (rupture rate in the no intervention group = 9.0 per 100 person years.) AAA specific mortality was similar between groups (12 vs. 13 percent) (HR = 1.01 [0.55 to 1.84]).

30-day mortality and treatment-related complications. Mortality within 30 days of the primary procedure occurred in 8.7 percent of patients assigned to EVAR and 2.1 percent among the 47 patients in the no intervention group who subsequently underwent either EVAR or OSR (Table 6, and Appendix E, Table E8).

Among patients having a successful EVAR (including the 12 randomized to no intervention) 32.6 percent had a study defined post-intervention complication and 18 percent required reintervention (Tables 6–7 and Appendix E, Table E8). Endoleaks occurred in 18 percent of patients, and there was one graft rupture. The most common reason for reintervention was for repair of endoleaks (mostly type 1). Four-year point estimates for complications were 43 percent vs. 18 percent; EVAR vs. no intervention; for reintervention: 26 percent and 4 percent. Complications following EVAR were comparable in EVAR-1 vs. EVAR-2 patients (43 percent vs. 41 percent). However, the reintervention rate was higher in unacceptable surgical risk EVAR-2 patients (11.5 per 100 person years) compared to EVAR-1 (6.9 per 100 person years).

Health related quality of life. Baseline quality of life scores were lower in EVAR-2 participants compared to EVAR-1 enrollees. There were no differences in quality of life or health status scores between patients assigned to EVAR versus no intervention as measured by the EuroQol 5-D and SF-36 questionnaires (physical and mental components) at any time point among patients who were still alive and completed forms. (Appendix E, Table E9). The authors state that because their results demonstrate that “compared to no intervention EVAR did not show a survival benefit, had little effect on Health Related Quality of Life, was more costly, and involves a continuing need for surveillance and reintervention, they did not see a need to pursue cost-effectiveness modeling at this time (i.e., no intervention was a dominant strategy). Continued monitoring of patients for an additional 6 years is planned.”

Patient preferences. There are few data to accurately determine patient treatment preferences according to age, race, AAA size, and comorbid conditions. Determining patient preference by rates or volume of a given procedure is subject to influences by provider or industry. We looked at registries of patients screened for participation in RCTs of OSR versus surveillance and delayed OSR for AAA <5.5 cm, EVAR versus OSR for patients judged medically fit for OSR who have AAA ≥5.5 cm, and EVAR versus no intervention among individuals judged medically unfit for OSR who have AAA ≥5.5 cm. Estimating preferences by these methods is difficult because preferences are not known among patients refusing to be considered for enrollment. Additionally, in the two countries conducting EVAR trials, EVAR outside the RCT setting was limited. Patient willingness to participate in and receive EVAR may have been influenced by the opportunity to be treated with otherwise unavailable technology.

Among patients with small AAA and eligible for UKSAT randomization to early OSR or ultrasonographic surveillance, 14 percent refused randomization (reasons not stated). Among those randomized to early OSR 2.3 percent refused surgery. Among those randomized to ultrasonographic screening and delayed OSR 7.2 percent had OSR against protocol. In the ADAM study 2 percent eligible patients refused randomization. In the surveillance group, 9 percent had procedures performed despite the fact that the AAA did not meet criteria for OSR

In DREAM, one percent declined to undergo AAA repair. One patient assigned to OSR crossed over to EVAR. In EVAR-1, 24 percent eligible and offered randomization refused randomization. Stated treatment preferences among those refusing included EVAR (32 percent), OSR (60 percent), no intervention (8 percent), and unknown (2 percent).

Patients who were eligible but refused entry into EVAR-2 were evenly divided between preference for EVAR and no intervention. Eight percent assigned to no intervention subsequently received intervention due to patient preference. Only one patient assigned to EVAR refused. How RCT results will influence preferences in the U.S. is not known.

Study and Patient Characteristics

The evidence of the effectiveness and adverse effects of EVAR from sources other than RCTs is based primarily on an update of a recently published systematic review by Drury (http://www.nice.org.uk/pdf/ip/AAA%20stent%20review%20body%20report%20for%20web.pdf.⁶ The authors included 47 studies published between 2000 and early 2004 (n=19,160, ranging from 46 to 4,613), including one RCT,¹¹ nine nonrandomized controlled trials (NRCTs) in comparison to OSR,^{54, 83}–⁹⁰ 16 comparative (two or more subgroups undergoing EVAR) observational studies,^{56, 91}–¹⁰⁵ and 21 case series.^{57, 106}–¹²⁵ Studies were excluded if they were case series with fewer than 50 patients, not primary studies, assessed ruptured AAA, had insufficient outcomes of interest, or more recent or relevant publications were available. We verified or corrected results from their included studies, summarized findings, updated with data from two recently published national registries, Lifeline (conducted in the U.S., n=2,664)¹²⁶ and RETA (n=1,000)¹²⁷ and information available on the FDA website from device-specific data. Additionally, we provide information from a post-hoc defined “high-risk subgroup” of patients enrolled in these five mutlicenter investigational device exemption clinical trials leading to FDA approval.¹⁴ Discussion of results from this report are provided under Key Question 3.

Studies were included if the intervention was EVAR, included adults with asymptomatic infrarenal AAA undergoing elective intervention, and assessed clinical outcomes related to efficacy/safety of EVAR. Patients with thoracic and thoraco-abdominal aneurysms and symptomatic or ruptured aneurysms were excluded. Main clinical outcomes included EVAR deployment success, primary and 30-day technical success, aneurysm rupture following successful EVAR, conversion rates (primary and secondary), and secondary interventions. Complications of EVAR were assessed by the frequency of technical problems (stent/graft complications and endoleaks) and major adverse events including 30-day mortality.

Mean ages of the study participants ranged from 65 to 76 years and mean followup times ranged from 7 to 36 months for the studies reporting these variables. AAA diameter ranged from 5 to 7 cm with a mean of 5.8 cm. The average AAA size would be classified as “large.” Over 70 percent of studies had a mean followup of at least 12 months, seven with 24 months. Mean followup was not reported in 12 studies. Nearly 90 percent of patients were men. Most had coexisting medical conditions including diabetes (14 percent), hypertension (54 percent), coronary artery disease (59 percent), tobacco use (67 percent), and chronic obstructive pulmonary disease (35 percent).

The methodological quality of included studies reported varied considerably. Reported limitations included lack of detail regarding inclusion/exclusion of study participants, description of dropouts or patients lost to followup, operator experience, and blinding of outcomes assessors. Furthermore, because these studies were not randomized trials, it is not possible to make definitive statements about the relative effectiveness or adverse effects of different endografts. Generally, statistical analyses of study outcomes were not adjusted for confounding factors. The majority of the included studies did not provide a source of funding. Funding was provided by device manufacturers (eight studies),^{87, 94, 95, 99, 104, 105, 108, 123} and the U.S. government in one.¹¹⁶

Mortality

30-day mortality. Fourteen case-control reports, including registries or nonrandomized but controlled trials, compared 30-day mortality of EVAR to OSR.^{51, 54, 84}–^{88, 90, 126, 128}–¹³² There was a higher 30-day mortality in the OSR group, 3.8 percent versus 1.5 percent for EVAR (absolute risk reduction of 2 percent [95 percent CI 2 to 3] favoring EVAR). Incidence and risk reduction in mortality for EVAR were similar to results reported from RCT (Appendix E ^*, Table E10). The pooled 30-day mortality rate for non-device specific EVAR as reported from both case series and comparative studies was slightly higher than either the RCT or the comparative studies (2.6 percent) and ranged from 0 to 8.4 percent (Appendix E, Table E11).

All-cause and AAA-related mortality. All-cause and AAA-related deaths from reports limited to registries and device-specific studies are shown in Appendix E, Table E12. Mean followup times varied widely or were not reported, limiting interpretation of the results. In the pooled Lifeline registry comprised of device trials conducted in the U.S., 23 percent of patients had died at a mean followup of 34 months, nearly 2 percent due to AAA-defined cause.¹²⁶ In data not shown, the overall incidence of AAA-related death was 3.0 percent (range 1.3 percent to 6.8 percent) in three studies (n=1,171) with followup periods ranging from 21 to 36 months. Non-AAA mortality was assessed in eight studies (n=1,228), including one nonRCT. In the nonRCT, using the Talent device, mortality not related to AAA following EVAR was 8.3 percent versus 4.8 percent for OSR at a mean followup of 13 months.⁸⁷

Early and Delayed Aneurysm Rupture (Appendix E, Table E13)

Nine studies (n=8,772) including the four clinical trials comprising Lifeline, reported incidence of early AAA rupture, defined as occurring within 30 days or less following EVAR. The rupture rate was 0.2 percent (16 events) and ranged from 0 percent to 1.6 percent. Delayed rupture, occurring after 30 days postoperatively, was reported in 18 studies (n=9,720). The percent of delayed rupture following EVAR was 0.5 percent during mean followup times ranging from 7 to 72 months. In the device-specific studies, early and late ruptures were reported only for AneuRx device, three (0.3 percent) and ten (0.84 percent), respectively.¹²⁴

Primary and Delayed Conversion to OSR (Appendix E, Table E14)

Primary conversion is defined as conversion to OSR immediately after a failed EVAR procedure. Primary conversion results were reported in 18 studies (n=10,832) including the four clinical trials comprising Lifeline. The overall percentage of patients converting to OSR was 2.2 percent and ranged from 0 to 10 percent. In the device-specific trials, primary conversion ranged from 0 percent for the Excluder device up to 9.7 percent for the EGS.^{89, 133} In a sub-study of the AneuRx clinical trial, Shames compared EVAR outcomes between men (n=203) and women (n=42) (data not shown).¹⁰⁴ Six women (14 percent) required conversion to OSR compared to one man. Two studies compared early to late experience in performing EVAR. Conversion to OSR for “early experience” was 27 percent and 8.9 percent compared to 4.5 percent and 0 percent for “late experience” in the Flora and Resch studies, respectively.^{93, 117}

A delayed conversion to OSR was considered to be a conversion after a successful EVAR procedure. The pooled percentage occurrence of delayed conversions from 18 studies (n=10,141) was 1.8 percent with mean followup times ranging from 7 to 72 months. In the device-specific trials, delayed conversion was reported for 38 patients receiving AneuRx (3.2 percent)¹²⁴ Only one study had occurrences of delayed conversion greater than 5 percent.⁹³

Secondary Intervention Following EVAR (Appendix E ^*, Table E15)

Secondary intervention was defined by Drury as any surgical or radiological procedure following EVAR that was conducted to maintain exclusion of the aneurysm sac from the circulation or to maintain graft patency. The number and percentage of secondary interventions for 22 studies (n=10,793) are shown in Appendix E, Table E15. Overall, 15.3 percent (range 3.8 to 55.5 percent) of analyzed patients required a secondary procedure within mean followup times ranging from 6 to 72 months. In the device-specific studies, secondary interventions were undertaken for 12 and 13 percent of patients receiving Powerlink and Excluder devices after approximately 2 years, respectively.^{133, 134} For patients receiving Ancure and EGS devices, 37 percent followed for 5 years required a secondary procedure.⁸⁹

Additional Outcomes

The Drury review assessed clinical outcomes not examined further in our report. These include: (1) deployment success; (2) primary and 30-day technical success; (3) change in aneurysm size during followup; (4) procedural blood loss; (5) length of intensive care unit (ICU) stay; and (6) length of hospital stay. The results are summarized: (http://www.nice.org.uk/pdf/ip/AAA%20stent%20review%20body%20report%20for%20web.pdf).

Successful EVAR deployment, defined as EVAR placement in the correct position without surgical intervention or OSR conversion, was reported in 13 studies (n=5,480).^{56, 87, 89, 99, 102, 103, 106, 108}–^{110, 114, 115, 120, 121, 125} Deployment success rate was 97 percent. In two device-specific nonRCTs, success for Ancure/EGS and Talent was 93 percent and 99 percent, respectively.^{87, 89}

Primary technical success was reported in 11 studies (n=4,998).^{56, 87, 89, 92, 99, 102, 103, 108, 109, 112, 116, 120}–^{122, 125} The majority of studies defined this outcome as the successful placement of the endoluminal-stent with complete exclusion of the aneurysm from the circulation.^{87, 92, 99, 109, 125}

Three studies provided an alternative definition^{56, 103, 116} and four studies^{102, 108, 112, 120} provided no definition. Success was verified with either angiogram upon completion or pre-discharge angiograms. Primary technical success varied widely, from 61 to 91 percent, (80 percent overall). One nonRCT (Talent) reported a success of 70 percent.⁸⁷ Thirty-day technical success was defined as successful placement of the graft resulting in complete exclusion of the AAA, with or without prior secondary intervention. Overall success was 91 percent in eight studies (n=1,493) and ranged from 77 to 100 percent. Success for the Talent study was 86 percent.⁸⁷

Increase in AAA size was reported in 14 studies (n=2,509)^{83, 87, 95, 96, 108}–^{113, 121, 122, 124, 135} and ten (n=2,065) reported reduction in AAA^{83, 95, 96, 108, 110, 112, 113, 121, 124, 135} following EVAR. Eight studies reported significant change as an increase or decrease of greater than 5 mm.^{87, 95, 110}–^{112, 122, 124, 135} One study defined decrease as a change over 10 mm and increase as a change over 5 mm from preoperative size.⁸³ The range of increase in the included studies was 0 – 27 percent with an overall average of 7.7 percent. Approximately 44 percent of patients experienced a decrease in AAA (range 11 – 82.5 percent).

Overall procedural blood loss following EVAR was 397 ml, reported in 12 studies (n=2,594).^{56, 84, 85, 87}–^{89, 96, 106, 108, 110, 116, 121} Mean blood loss was less for EVAR (414 ml, range 96–783) compared to OSR (1,329 ml range 451–1,800). Mean length of ICU stay with EVAR was less than one-half of OSR (0.7 vs. 1.6 days).^{11, 84, 86}–⁸⁹ The mean length of hospital stay for EVAR from nine nonRCT was also less than one-half that of OSR (4.2 days vs. 9.9 days).

Endoleaks and Complications

Safety of EVAR was assessed according to the incidence of EVAR device-related adverse events, or technical complications. Endoleaks are included in this section although they are not classified as a complication. Endoleaks, detected and classified through radiographic imaging, are often asymptomatic and do not require a reintervention. Complications included stent migrations, fractures, graft-limb thrombosis, graft stenosis, and access artery injury.

Endoleaks (Appendix E ^*, Tables E16–E18)

Any endoleaks. The incidence of any EVAR endoleaks is shown in (Appendix E, Table E16). Data were extracted from two registries, RETA and Lifeline with additional data from the four clinical trials comprising Lifeline. Within 30 days postoperatively, any endoleaks occurred in nearly one-quarter in the U.S. Lifeline series¹²⁶ and 14.6 percent in the European RETA database.¹²⁷ For the individual devices, incidences ranged from 13.9 percent AneuRx¹²⁴ to 22.7 percent for Powerlink.¹³⁴ The incidence was two to three fold as high (42.2 percent) at discharge for combined Ancure and EGS bifurcated devices.⁸⁹

Type I endoleaks. Ten studies (n=2,617) reported occurrences of Type I endoleaks within 30 days following EVAR (Appendix E, Table E17). The overall incidence was 4.2 percent. In the device-specific studies this ranged from 0.9 percent (Powerlink and AneuRx) to 5.8 percent for Talent.^{134 87, 124} The overall incidence was 3.5 percent (range from 0–14 percent) after one year in 13 studies (n=2,544) reporting. The incidence of Type 1 endoleaks beyond 1 year was 6.7 percent (range 0 to 21.5 percent) for 18 studies reporting (n=7,848).

Type II endoleaks. Eleven studies (n=2,712) reported occurrences of Type II endoleaks within 30 days following EVAR (Appendix E, Table E18). The overall incidence was 10.5 percent with a wide range (1.4–31.2 percent). The percentage of events for the device-specific studies was 1.4 to 19.3 percent for 2 AneuRx studies ^{102, 124} 11.7 percent for Excluder, ¹³³ and 20 percent for Powerlink.¹³⁴ The highest number of events occurred in the combined Ancure and EGS study,⁸⁹ although this event was reported for bifurcated grafts only. At one year, 14.7 percent of individuals had a Type II endoleak. For the device-specific studies, the 1-year range was 5.4 percent (Talent) up to 21.8 percent (combined Ancure and EGS, bifurcated grafts only).^{87, 89} After 1 year the percent of events was still 10.2 percent (range 1.0–20.8 percent; 14 studies; n=7,066).

Type III endoleaks. Type III endoleaks were infrequent. Three studies (n=1,290) reported 30 day outcomes. All 15 events occurred in the RETA registry.^{127, 133, 134} Overall, 4.0 percent had a Type III endoleak after 1 year (nine studies, n=6,599). The majority were from Eurostar.¹¹⁹ Of the device-specific studies, only AneuRx reported events (8 of 383, 2.1 percent).¹²⁴

Technical complications. (Appendix E ^*, Table E19). The outcome “any technical complication” for the duration of postoperative hospitalization was reported by the RETA registry.¹²⁷ There were 55 events in 976 patients (5.6 percent). Stent migration was defined as caudal displacement greater than 10 mm. Stent migration within 30 days following EVAR or up to discharge occurred in less than 1 percent of patients in the two registries, Lifeline and Eurostar providing data.^{120, 126} The one year stent migration rate was less than 1 percent for three studies reporting (n=1,599). After 1 year, 4.4 percent had stent migration with a wide range in results (1.7–18.9 percent, eight studies, n=7,027). Stent migration occurred in over 6 percent of patients in the AneuRx device study.¹²⁴

No stent fractures were reported in the 47 included studies by Drury. Stent wire fractures can occur without a full stent fracture and are typically detected by followup x-rays or CT scans. Stent-wire fractures were reported in three studies, two device-specific.^{87, 108, 134} Overall, 17 events in 659 patients occurred (2.6 percent). The Powerlink study reported no stent wire fractures while the Talent reported an incidence of 4.6 percent.^{87, 134}

Graft-limb thrombosis was reported in five studies (n=659) within 30 days following EVAR (2.4 percent, range 0.7 to 6.3 percent). Incidences for the two device-specific studies were comparable. The overall percentage was similar at 1 year, 2.5, with a range of 0 to 11 percent (11 studies, n=1,657). After 1 year, the incidence increased to 3.8 percent (range 1.9 to 6.1 percent, eight studies, n=6,602). The 1-year incidence for device-specific Powerlink and combined Ancure and EGS reports were 2.1 and 5.4 percent, respectively.^{89, 134}

Graft stenosis was reported by four studies. Only Eurostar provided data within 30 days (n=2,862) and greater than 1 year (n=4,613) following EVAR (0.3 percent and 1.4 percent respectively). Up to 1 year, the overall incidence of graft stenosis was 2.7 percent in three studies reporting (n=365), including one device-specific study (Powerlink), which reported three events in 192 patients.¹³⁴ In data not shown, access artery injury was reported in seven studies (n=2,561) with an overall incidence of 4.8 percent (range 1.4–12.9 percent).^{89, 92, 102}–^{104, 109, 118, 121}

Question 2. What is the relationship of volume, both hospital and physician, to the benefits and harms of endovascular procedures to repair of AAA?

EVAR of AAA

The literature search did not find any qualifying studies of the association between the volume of endovascular AAA repairs done by hospitals or physicians and any beneficial or harmful outcome.

The only report from the U.S. was based on a statewide discharge database in New York that compared hospitals that did fewer than five EVAR to repair AAA to hospitals doing 30 or more.⁵⁴ The crude in-patient mortality rates of 1.9 percent versus 0.8 percent did not differ significantly in this small sample. Investigators were not able to make adjustments for potential differences in risk due to patient, physician, or hospital characteristics.

Analysis of data in a European registry did not find a relationship between 30-day mortality in the first 11 cases versus subsequent cases treated at a medical center.⁵⁷ Midterm outcomes were related to center experience. However, methodological issues discussed in the Background section, including use of withdrawn devices that are not used in the U.S. in earlier cases, make it difficult to determine whether these observations would apply to contemporary practice in the U.S.⁵⁸

OSR of AAA

Previous studies of the relationship between hospital and surgeon volume of OSRs and outcomes were reviewed to inform future studies of EVAR. Table 8 on pages 59–64 summarizes studies of relationships between the volume of OSRs of AAA and outcomes in the United States in the 1990s. Endovascular repair was primarily an investigational procedure during this time, and these findings should not be extrapolated to current practice where endovascular procedures are being used to repair a substantial select fraction of the AAA.

Five studies of OSR of AAA used a national sample of hospital discharges, discharges of elderly Medicare enrollees, or operations in VA medical centers.^{15, 16, 136}–¹³⁹ Three investigations analyzed hospital discharges in Maryland, Florida, or New York.¹⁴⁰–¹⁴² All studies were retrospective analyses of available databases. The VA study was the only one that had prospectively collected preoperative data including clinical variables.¹³⁹ Corresponding with the incidence of AAA, all samples were predominantly elderly males, and data for females is sparse.

Death was the primary outcome in all analyses. In-hospital death was analyzed in five of the eight reports^{16, 138, 140}–¹⁴² Others included deaths within 30-days to avoid bias that can be introduced by differences in hospital length of stay.¹⁴³ Hospital volume was analyzed in all studies. Two studies of Medicare data attempted to estimate each hospital's total volume, not just operations covered by Medicare.^{15, 136, 137} Volume measures reflected periods of time that varied from 1 to 6 years. Some were analyzed as average annual volume, while others reported total volume during the period of study. Half of the studies included a measure of surgeon volume. Only one examined the surgeon's specialty.¹³⁸ All studies ignored procedures that were done prior to the period of study as well as the surgeon's experience doing other types of vascular operations.

All studies employed logistic regression models with patients as the unit of analysis. Four studies did not report adjustments for clustering of patients within hospitals (or surgeons).^{16, 140}–¹⁴² Two studies utilized hierarchical regression models with hospitals and or surgeons as random effects.^{137, 139} Age and sex were used as covariates in all analyses except the VA study that was nearly 100 percent male. Race (white versus non-white) was included in six studies.^{15, 16, 136}–^{138, 141, 142} Two studies did not attempt to include co-morbidities because of concerns about completeness of diagnoses listed on discharge abstracts and the inability to differentiate postoperative complications from preexisting conditions.^{141, 142} Admission acuity was a covariate in all studies except one that excluded urgent and emergency admissions. Most excluded hospital transfers. Other covariates used by some, but not all, included year to control for time trends, estimated patient income or type of insurance, service intensity of DRGs, hospital length of stay, and characteristics such as bed size, ownership, being a teaching facility for physicians, and urban location. Information about aneurysm characteristics was not available for risk adjustment in any of the eight investigations. Another analysis of the data in Nationwide Inpatient Sample from 1998 to 2001 found that use of low volume hospitals (<10 AAA repairs/year) was associated with race/ethnicity, geographic region, rural setting, Medicaid, and lack of insurance and non-elective admissions.¹⁴⁴ If these variables are also associated with outcomes, they need to be controlled when estimating the effects of volume. Only two studies reported how well their regression model fit the data and three reported how well their model discriminated deaths. Reported c-statistics were modest, ranging from 0.68 to 0.75.

Overall mortality in the studies ranged from 3.5 percent to 5.7 percent. Seven of eight studies reported that significantly lower adjusted mortality was associated with higher volume hospitals. Two studies suggested that the volume effect was relatively consistent in patient subgroups with different preoperative risks of death.^{15, 16} Cutpoints for hospital volume that were associated with lower adjusted mortality ranged from 17 to 100 AAA OSR per year. These cutpoints were arbitrary and studies did not report analyses specifically to determine a volume threshold(s). The highest volume in the VA study that did not find an association between hospital volume and mortality was 32 procedures per year. This was the only study that prospectively collected and incorporated numerous preoperative clinical variables for risk adjustment. A summary plot of mortality versus volume from all studies was not possible given varying measures of mortality and volume and broad volume categorizations. Some studies only reported risk-adjusted results for volume analyzed as a logarithm-transformed continuous variable.

All three studies that analyzed surgeon volume reported that surgeon volume had an effect that was independent of hospital volume.^{137, 138, 140} One study that reported how surgeon volume altered the effect of hospital volume indicated that the addition of surgeon volume to the regression model greatly reduced, but did not eliminate, the effect of hospital volume.¹³⁷ Surgeon volume was inversely related to mortality when surgeon specialty was taken into account in the two studies.^{138, 142} These studies did not identify a volume cutpoint that could be used to identify surgeons who perform enough AAA repairs to maintain optimal outcomes.

Summary of Results

Adequate studies of the relationship between the volume of EVAR procedures done by hospitals and physicians in the United States have not been reported to date. Most previously published studies of OSRs of AAA done by hospitals and surgeons suggest that there is an inverse relationship between volume and short-term mortality. Whether or not a similar inverse relationship exists between the volume of EVAR procedures done by hospitals or physicians and mortality or any other outcome has not been established. The poorly defined relationship between the volume of OSR and short-term mortality should not be extrapolated to EVAR. Studies specific to EVAR that address important limitations of previous volume-outcome studies discussed in the Background section are needed to guide policy.

Question 3: How do the characteristics of the aneurysm (size/location/shape) and the patient (age/gender) affect the benefits and harms of endovascular and open-surgical repair?

Aneurysm Size (Small AAA)

Natural History: AAA initial or attained diameter is the biggest known predictor of rupture risk. AAA typically have been classified as small (<5.5 cm) or large. These classifications have been used for inclusion criteria in RCTs and clinical decision making.

The annual rupture rate for AAA <5.5 cm was less than 1 percent in patients enrolled in the active surveillance arm of ADAM and UKSAT. A cohort of 2,257 adults was comprised of patients enrolled in the UKSAT trial and the associated study for patients' ineligible or refusing randomization. After 3 years the annual rate of AAA rupture was 2.2 percent. Additional analyses assessed the risk of rupture and last known or estimated AAA diameter categorized as ≤3.9 cm, 4.0 to 4.9 cm, 5.0 to 5.9 cm, and ≥6.0 cm. The number of ruptures per 100 patient years increased from 0.3 for AAA <3.9 cm to 1.5 and 6.5 for patients with AAAs 4.0 to 4.9 cm and 5.0 to 5.9 cm respectively. A Mayo Clinic study suggested that a 1 cm larger initial diameter was associated with an approximately 50 percent increase in the adjusted rupture risk.

OSR. As noted previously (Question 1) both ADAM and UKSAT analyzed mortality according to treatment arm (early OSR versus surveillance) and age or aneurysm diameter at baseline, based on tertile groups. The adjusted hazard ratio at 8 years followup in UKSAT tended toward a greater benefit of early OSR among younger patients, men, and those with larger aneurysms. However, neither study demonstrated a significant interaction between treatment group and age baseline, AAA diameter, or gender. The reported test of interaction was not significant across AAA size (p = 0.28) or age (p = 0.18). Calculated risk differences, reported relative risks, and tests for interaction from ADAM were not significant and had point estimates favoring surveillance. In UKSAT, the overall death rate for patients who continued to smoke was 12.0 per 100 patient years compared to 3.8 per 100 patient years for patients who no longer smoked. Results were not reported according to randomized treatment assignment. Results from ADAM demonstrated that independent predictors of death among all enrollees included higher serum creatinine level, lower weight, diagnosis of COPD or diabetes, larger AAA diameter, lower forced expiratory volume in one second, and nonuse of a beta-blocker. There were few women enrolled in either trial. The test for interaction with gender was not significant in UKSAT (p = 0.40). However, fatal ruptures were nearly three times more common in women than men and the risk of AAA rupture was four times higher among women. Greater than 90 percent of enrollees were of white race, and no outcomes according to race were provided.

EVAR. Despite an RCT that reported no benefit in survival or quality of life with immediate OSR versus active surveillance in patients with AAA <5.5 cm there is a possibility that some of these patients might benefit from EVAR, especially if EVAR morbidity is very low. One study recently began recruitment to evaluate EVAR versus surveillance for small AAA. The estimated rupture rate for the surveillance group used in sample size calculations for this trial are higher than reported in the two trials of OSR versus surveillance.

Large AAA

No intervention. The rupture rate for large AAA in patients judged fit for OSR is difficult to determine because most undergo early AAA repair. Lederle reported that the 1-year incidence of probable rupture by initial AAA diameter in those refusing or judged medically unfit for OSR. Rupture rates were 9.4 percent for AAA of 5.5 to 5.9 cm, 10.2 percent for AAA of 6.0 to 6.9 cm, and 32.5 percent for AAA ≥7.0 cm. Among patients who attained an AAA diameter exceeding 8.0 cm, 25.7 percent ruptured within 6 months.³⁶ The rupture rate among the no intervention group in EVAR-2 was lower than that reported by Lederle. In EVAR-2, (mean AAA = 6.3 cm) AAA rupture occurred in 12.2 percent of individuals in the no intervention group (rupture rate in the no intervention group = 9.0 per 100 person years).

EVAR versus OSR or no intervention. Outcomes of EVAR from randomized controlled trials for large AAA were assessed relative to OSR or no intervention based on whether patients were considered medically fit for OSR. Enrolled patients were all considered candidates for EVAR. However, the proportion of patients with large AAA (i.e., ≥5.5 cm) who might be eligible for EVAR in clinical settings was estimated from the EVAR-1 and 2 trial registries as 54 percent.¹³ This is consistent with other reports from EVAR utilization patterns in the U.S. where an estimated 40–80 percent of AAA could be amenable to EVAR based on aneurysm size, morphology, and patient surgical risk characteristics.^{39, 40} Data from the EVAR-2 trial suggests that approximately one-fifth of patients with large AAA who are EVAR candidates are judged medically unfit for OSR.¹⁴⁵

Patient characteristics. In trials of EVAR vs. OSR over 90 percent of enrollees were male with an average age of 70 years. Almost all were white. The mean AAA diameter ranged from approximately 5.4 cm in the small study by Cuypers to 6.5 cm in EVAR-1. More than 40 percent of patients had a history of cardiac disease, 10–16 percent diabetes, and the majority were current or past smokers. No outcomes by race or these selected comorbidities were reported.

Detailed findings from RCTs of EVAR versus OSR or no intervention are described in Question 1. In the DREAM trial the perioperative survival advantage with EVAR as compared with OSR was not sustained after the first postoperative year. In the early postoperative period there was a small quality of life advantage for EVAR. At 6 months and beyond QOL was better after OSR. In the larger EVAR-1 trial, compared with OSR, EVAR offered no advantage in all-cause mortality and health related quality of life, was more expensive, and was associated with a greater number of complications and reinterventions. There was a 3 percent better AAA-related survival with EVAR, though it is possible that cause of death ascertainment was biased against later AAA related deaths in the EVAR group. There were no significant interactions for all-cause or AAA-related mortality with age, sex, aneurysm diameter, or creatinine concentration. Followup time period suggested that hazard ratios for AAA-related mortality, total mortality, complications, and need for intervention began to favor OSR after 6 months.

Patients in EVAR-2 were considered medically unfit for OSR and had large AAA (mean AAA diameter = 6.4 cm; range = 6.0–7.4).¹⁴⁵ Patients were older (76 vs. 70 years) and had worse pulmonary function based on spirometry (FEV₁: 1.6 vs. 2.1L) than patients enrolled in EVAR-1. Baseline blood pressure, serum creatinine, and cholesterol levels as well as the percent males or individuals with diabetes, smoking, or cardiac history appeared similar. The severity or duration of these comorbidities was not described in detail, but 2 year mortality in the no intervention group was greater than 60 percent. In the EVAR-2 trial, EVAR resulted in a 30-day operative mortality of 9 percent in patients unfit for OSR. EVAR did not improve overall or AAA related mortality or quality of life over no intervention and was associated with a need for continued surveillance and reinterventions at increased cost.

There were no significant interactions for the effect of EVAR with age, sex, aneurysm diameter, or creatinine concentration. The 30-day EVAR mortality was higher in the sicker EVAR-2 patients than those receiving endovascular repair who were judged fit for surgery in EVAR-1 (9 percent and versus 1.7 percent; p <0.0001). Additionally, compared to healthy EVAR-1 patients, there was a greater need for internal iliac artery embolization, blood products, renal dialysis, length of hospital stay (and a trend for overall reintervention rate).

A wide range of EVAR devices was used in these trials. Over 90 percent of EVAR procedures used commercially available aortobiiliac device systems while aorto-aortic (tube) grafts were used in the majority of OSRs. Selection for both EVAR and OSR was based on the discretion of the surgical team. The most commonly used EVAR devices were Zenith (33–59 percent of devices) and Talent (27 to 33 percent). Additional description of the devices was not provided. Prior to participating, investigators were required to submit data related to at least 20 completed EVAR procedures.

High Surgical Risk Patients

No randomized trials of EVAR versus OSR have specifically limited recruitment to patients judged to be at high surgical risk due to age, medical comorbidities, or abdominal anatomy (e.g., horseshoe kidney, abdominal adhesions, etc.). As noted previously, patients enrolled in the EVAR and OSR trials were elderly and had numerous comorbid conditions that would result in a limited life expectancy and a relatively high surgical risk. One report enrolled 100 patients believed to be at high risk for complications from OSR and subsequently treated them with the Zenith AAA EVAR.¹³¹ Pathophysiologic conditions used to determine high risk status included age >80 years, creatinine >2.0 mg/dL, disabling COPD, ejection fraction <25 percent, stroke with residual deficit, medically intractable hypertension, previous renal bypass surgery, or myocardial infarction within the past 6 months. Mean baseline AAA and patient characteristics were not reported and it is not clear how many, if any, of these patients were considered “medically unfit” for OSR. Based on Kaplan-Meir plot estimates, overall survival at 1 year was 92 percent and at 2 years was 78 percent (number at risk not provided). AAA reported survival was estimated at 94 percent for both 1 and 2 year. An additional report evaluated outcomes of EVAR and OSR in a “high surgical risk” subgroup of patients entered into any of five nonrandomized multicenter IDE studies leading to FDA approval of EVAR devices.¹⁴ Patients with AAA ≥5.5 cm were retrospectively categorized as “high surgical risk” based on age >60 years and having at least one cardiac, pulmonary, or renal comorbidity. Inclusion, criteria for the IDE studies required that patients were candidates for OSR, though in one study patients were prospectively defined as being at high risk for OSR complications as noted above. Three-quarters of “high-risk” patients had only one comorbid category and <1 percent had all three categories. There were 14.2 percent endoleaks reported at 30 days, 17.5 percent at 1 year, and 18.9 percent at 4 years. Endograft migration occurred in 2.7 percent at 4 years. Major complications were not reported. After 4 years, deaths categorized as due to AAA were similar between EVAR and OSR patients (4.2 percent and 5.1 percent respectively (p = 0.58). Overall-survival in EVAR treated patients was 10 percent lower compared to OSR (56 percent vs. 66 percent), though this difference was not statistically significant (p = 0.23).

Device-Specific Results from FDA Reports and Lifeline Registry (Tables 9 and 10 on pages 70–72)

Few studies provided an evaluation according to device or patient characteristics. We previously described results from nonrandomized studies of EVAR with description of device-specific results (key question 1 results from nonrandomized trials). A summary of some of these findings is also provided here with attention to reports from the United States (Tables 9 and 10). Pooled results from EVAR devices (n = 3,016) included in the published Lifeline report (Ancure, AneuRx, Excluder, and Powerlink) as well as FDA data related to the Cook-Zenith device and corresponding OSR controls (n = 414) are provided. Information from manufacture/device specific FDA reports is summarized. None represent direct comparisons from randomized trials. Patient or aneurysm factors could influence outcomes. Therefore, it is hazardous to make definitive statements regarding relative safety or effectiveness.

Compared to RCT enrollees, patients receiving EVAR included in these reports had smaller AAA (mean = 5.6 cm). Compared to the listed OSR controls, EVAR patients had slightly smaller AAA (5.6 vs. 5.8 cm), were older (73 vs. 70 years), more frequently male (89 vs. 81 percent), and have a history of coronary artery disease (79 versus 53 percent). Across devices there were no large differences in mean age, AAA diameter, gender, or comorbid conditions.

The number of EVAR devices submitted for FDA review ranged from 121 for Ancure to 416 for AneuRx. The pooled results as well as device-specific data obtained separately from FDA websites indicates that 30-day mortality rates with different devices ranged from 1 percent with Gore Excluder to 4.2 percent with Guidant Ancure. EVAR and OSR 30-day mortality rates were 1.6 and 1.4 percent respectively. Major complications or adverse events within 30 days were not provided in the Lifeline report but ranged from 13.6 percent in FDA reports with Gore Excluder to 35.6 percent with Guidant Ancure.

Device-Specific Results from Published Nonrandomized Studies

The mean AAA diameter was ≥5 cm in patients included in the nonrandomized-EVAR reports, and available outcomes would be considered related to large AAA. However, three reports describe outcomes of EVAR for small versus large AAA.^{94, 100, 146} Arko reported no significant difference in endoleaks between small, medium, or large AAA (p = 0.41). Ouriel reported that in a cohort of 700 patients treated with EVAR, overall survival, AAA related death, conversion, and Type 1 endoleaks were worse in large (≥5.5 cm) versus small AAA. Similar results from the Eurostar registry were noted by Peppelenbosch.

In the device-specific published studies (which often included additional patients and outcomes data compared to the FDA reports), early and late ruptures were reported only for AneuRx device and were 0.3 and 0.8 percent respectively.¹²⁴ Primary conversion ranged from 0 percent for the Excluder device up to 9.7 percent for the EGS. In a sub study of the AneuRx clinical trial, Shames compared EVAR outcomes between men (n = 203) and women (n = 42) (data not shown).¹⁰⁴ Six women (14 percent) required conversion to OSR compared to one man. The incidence of delayed conversions for 18 studies (n = 10,141) was 1.8 percent with mean followup times ranging from 7 to 72 months.

In device-specific trials delayed conversion was reported for 3.2 percent of patients receiving AneuRx. Secondary interventions were undertaken for 12 percent, and 13 percent of patients receiving Powerlink and Excluder devices after approximately 2 years. For patients receiving Ancure and EGS devices, 37 percent followed for 5 years required a secondary procedure.

The incidence of any EVAR endoleaks is shown in Table 9. Data were extracted from two registries, RETA and Lifeline, with additional data from four clinical trials comprising Lifeline. Within 30 days postoperatively, the percentage of any endoleaks for the two registries was 24.7 percent for Lifeline and 14.6 percent for RETA. For the individual devices, incidences ranged from 13.9 percent AneuRx to 22.7 percent for Powerlink. Incidence was 42.2 percent at discharge for combined Ancure and EGS devices. This included only bifurcated grafts.

The overall incidence of Type I endoleaks within 30 days following EVAR was 4.2 percent and ranged from 0.9 percent to 11 percent. The percentage of events in the device-specific studies ranged from 0.9 percent (Powerlink and AneuRx) to 5.8 percent for Talent. Overall incidence of Type III endoleaks within 30 days following EVAR was 1.2 percent in three reports, including Powerlink and Excluder. Powerlink and Excluder reported no Type III endoleaks at any followup period.

There was a wide range in the percentage of Type II endoleaks within 30 days following EVAR (1.4 percent to 19.3 percent in two AneuRx studies, 11.7 percent for Excluder and 20 percent for Powerlink). The highest number of events occurred in the combined Ancure and EGS study, although this event was reported for bifurcated grafts only. In 12 studies (n = 2,598) 14.7 percent had events at 1 year. The range was 5.4 percent (Talent) to 21.8 percent (combined Ancure and EGS, bifurcated grafts only). The Powerlink study reported no stent fractures.

Graft-limb thrombosis was reported in five studies (n = 659) within 30 days following EVAR with an overall incidence of 2.4 percent (range 0.7 percent to 6.3 percent). Incidences for the two device-specific studies were comparable. The 1 year incidence for device-specific Powerlink and combined Ancure and EGS NRCTs were 2.1 percent and 5.4 percent, respectively. Up to 1 year, the overall incidence of graft stenosis was 2.7 percent in three studies reporting, including one device-specific study Powerlink which reported events in 1.6 percent of patients.

Question 4: What are the costs-benefits for each of the procedures?

Aside from studies developing Markov models to estimate lifetime costs and the cost effectiveness of EVAR and OSR, the majority of studies reviewed for the economic analysis report either on the costs associated with the initial elective procedure (EVAR, OSR, or both) or on costs of followup care for EVAR. If studies reporting on cost also include measures of effectiveness, we report those findings as well as conclusions regarding cost effectiveness.

The results of the two EVAR RCTs for AAA ≥5.5 cm (EVAR vs. OSR and EVAR vs. no intervention) were used as the basis for a cost-effectiveness modeling. With those two exceptions, most of the data on costs or effectiveness (rarely both) come from observational studies, some of which are simply case series with no comparative analysis.

All of the retrospective reviews and prospective case series included in our analyses have taken an institutional perspective and report only the costs incurred by the hospital(s) and/or clinics. None take the societal perspective of including the cost or quality of life associated with the patient's or caregiver's time lost from work or other activities while hospitalized, traveling, or attending followup visits.

Table 11 on page 79 presents an overview of the included studies. These studies vary not only by type (i.e., RCTs, systematic reviews, Markov models, retrospective reviews, and prospective case series), but also along several substantive dimensions including country, population, time horizon, type of grafts, patient selection criteria, and economic perspective (societal or institutional). A few studies report on hospital incentives supporting the use of these two procedures within the context of their profitability relative to Medicare reimbursement rates for DRGs 110 and 111 (AAA repair with and without complications, respectively).

The costs reported within the studies reviewed also vary by whether they considered direct, indirect, variable, and/or fixed costs. As individual hospital accounting systems do not necessarily follow a standardized approach to allocating costs across departments, the costs reported within this systematic review reflect a variety of costing methodologies. Despite these qualifications, we found a general consensus in the relative costs of EVAR and OSR within each of these studies regardless of the time frame considered.

EVAR Prosthesis

Since the cohort year also represents the year that the grafts were used, it represents the stage of the technology and to some extent the learning curve associated with the procedure. The majority of the EVAR grafts reported on in these studies received FDA approval in 1999.⁶⁷ The cost of some of these grafts may have been discounted or paid for based on production costs at the time the studies were underway. Once grafts became commercially available, they became more expensive. As the driving force behind the majority of the cost of EVAR is the cost of the graft, associated supplies/equipment, and the degree of complications associated with followup, the cohort year relative to the year the grafts became commercially available is important.

Cost estimates for the EVAR graft range from a low of $7,000 reported for a 1997-1999 patient cohort¹⁵⁴ to $13,000 reported for a 2000 patient cohort.¹⁵⁶ The graft accounts for a range of 34 percent to 78 percent of the total hospitalization costs reported in these two studies respectively, with the differences in percentages largely a reflection of how comprehensively the authors itemized costs. Bosch et al. relied on graft costs reported in the literature near the time of commercialization and estimated a relatively low price for the graft relative to the cost of the hospitalization, thus contributing to underestimating the total hospitalization costs for EVAR, underestimating the difference in the cost of EVAR relative to OSR, and underestimating the percent of total hospitalization costs attributable to the graft.¹⁵⁴ In contrast, Dryjski et al. underestimated the cost of the hospitalization relative to the graft by excluding a number of categories of necessary hospital services from the cost estimate.¹⁵⁶

Reimbursement Issues

The adequacy of Medicare reimbursement for either procedure varies by type of hospital, geographic region, and type of grafts used primarily for EVAR. Several studies report that from a hospital's perspective, reimbursement may not be sufficient to cover the cost of EVAR hospitalization, primarily because the DRG rates were set before EVAR procedure was introduced, thus not taking into account the cost of the prosthesis (graft). Studies published in the late 1990s rely on production costs associated with grafts. Once grafts obtained FDA approval and became commercially available, grafts essentially doubled in price, making it more difficult for hospitals to break even based on Medicare reimbursement rates for AAA repair. As Medicare reimbursement rates for AAA repair do not differentiate between EVAR and OSR, they do not reflect the high cost of the EVAR grafts or differences in length of stay. Hospitals may thus be trading off savings from reduced length of stay against the cost of the EVAR prosthesis.

Measures of Effectiveness (Table 14 on pages 88–89)

Measures of effectiveness include the LOS associated with the initial hospitalization, ICU time, mortality (in-hospital, 30 day) and morbidity, complication, and reintervention rates. EVAR has a shorter LOS associated with the initial hospitalization, range 2.0 days,¹⁵¹ to 10 days¹³ includes preoperative assessments) compared to OSR, range 7.3 days¹⁵¹ to 15.7 days.¹³ The longer length of hospital and ICU length of stay are based on the randomized trials that were conducted in Europe and thus may not be comparable to U.S. settings. Less ICU time is associated with EVAR's initial hospitalization, range 0.6 days¹⁵² to 1.4 days,¹⁵⁶ compared to OSR, range 2 days⁶⁴ to 3.5 days.¹⁵¹ Thus, EVAR is associated with less resource use in terms of a patient's time spent in the hospital as well as use of the ICU. Because hospital beds are expensive and essentially define the hospital's capacity/ability to serve patients, from the hospital's perspective, EVAR may represent a more efficient use of hospital resources than OSR.

Short-term benefits associated with EVAR include reduced LOS, reduced use of ICU, reduced operative, and 30-day mortality and morbidity. However, results of two RCT comparing EVAR with OSR have provided additional midterm outcome data after 2–4 years of followup. The results from these studies (which were not conducted in the U.S.) indicate that compared to OSR, EVAR reduces AAA related mortality by 3 percent. However, after 2 years there was no difference in overall survival. Differences in health related quality of life were small, may not be clinically noticeable, and disappeared after 3 months. EVAR costs were higher than OSR and were associated with more complications, need for reinterventions, and continued monitoring.

Cost Effectiveness Analyses Using Markov Methodologies (Table 15 on pages 90–92)

Table 15 summarizes three studies that utilized Markov methodologies to analyze differences in the cost and effectiveness of EVAR and OSR over a patient's lifetime^{17, 149, 150} and two studies that compared OSR to conservative treatment. All three of the former found that EVAR was more expensive. Two of the studies were conducted in the U.S. and report that a higher quality of life was associated with EVAR and that the ICERs were $9,905 per QALY and $22,826 per QALY respectively.^{149, 150} While Bosh, Patel, and Michaels use differing baseline operative mortality rates and QOL measures, they all use 70 year olds with aneurysms between 5 and 6 cm as their baseline reference case. The Patel study used assumptions of effectiveness based on case series data. They conclude that the benefits were worth the cost with the qualification that their results were highly dependent on their assumptions regarding mortality and morbidity. Patel et al. reported that EVAR may be more cost effective than OSR if operative mortality rates were <1.2 percent and the surgical mortality rates were >1.7 percent.¹⁵⁰ However, if the combined mortality and long term morbidity rate of open surgery decreased from 9.1 percent to 4.7 percent, the authors concluded that EVAR may not be cost effective. Bosch et al. also used case series data as the basis for their model; they report the sensitivity of their conclusions to EVAR graft performance in terms of long term failure and rupture rates.¹⁴⁹ They fail, however, to explicitly account for the cost of the EVAR prosthesis by using Medicare reimbursement rates for DRG codes 110 and 111 (AAA repair with and without complications, respectively) to estimate the cost of the initial AAA repair. These DRG codes do not distinguish differences in the cost of AAA repair by procedure (i.e., by EVAR and OSR), and do not include the cost of the EVR prosthesis. Thus, Bosch et al. underestimate the cost of the EVAR initial hospitalization costs relative to OSR and report a conservative estimate of the costs in the cost effectiveness ratio. On the other hand, Bosch includes patients' time lost from work in terms of lost wages in the costs associated with each procedure. As they also estimate the impact of the procedures on the patients' quality of life and include this in the denominator of the cost effectiveness ratio, they have essentially double counted the impact of AAA repair from the patient's perspective.

Although Bosch and Patel varied their analytical assumptions and conducted sensitivity analyses, they did not vary their assumptions simultaneously by employing probabilistic sensitivity analyses to fully test the robustness of their findings as recommended by the U.S. Panel on Cost Effectiveness in Health and Medicine.¹⁶⁰–¹⁶²

In a more recent, rigorous approach to Markov modeling, Michaels et al.¹⁷ used data from the EVAR trials and concluded that EVAR was NOT cost effective, in that the extra cost was not worth the gain in QALYs.¹⁷ Their ICER was £110,000 per QALY, which is well above accepted norms for cost effective procedures.^{65, 160} Michaels et al.¹⁷ found that the cost effectiveness of EVAR and OSR is sensitive to assumptions regarding lower early morbidity for EVAR, higher operative mortality rate for OSR, higher need for followup care for EVAR, higher reintervention/ complication rates for EVAR, e.g., 25–30 percent for EVAR versus 0 percent for OSR, and higher costs associated with care provided to EVAR patients, regardless of time frame.

Michaels et al. point out that there is a time dependent effect.¹⁷ Studies conducted within a relatively short timeframe fail to adequately address the long-term benefits, harm, or costs associated with EVAR. Decreased operative mortality rates associated with EVAR may be offset by higher complication rates later in life. If followup costs associated with complications and reinterventions for EVAR are ignored, then EVAR's low operative mortality rates favor EVAR and may lead to the premature conclusion that EVAR is a cost effective alternative to OSR. This assumption is particularly relevant given the midterm results from RCTs demonstrating that compared with OSR, EVAR offered no advantage in all-cause mortality or health related quality of life and led to a greater number of complications and reinterventions.

In terms of the cost effectiveness of the two procedures, Michaels et al. found that OSR dominates if operative mortality rates are less than 3 percent.¹⁷ OSR is preferred if EVAR's cost/QALY is greater than $30,000 and if operative mortality rates for OSR are between 3 percent and 11 percent. EVAR dominates if operative mortality rates are greater than 40 percent. If EVAR costs are similar to OSR, and/or if EVAR reintervention rates are cut by 50 percent, then EVAR is preferred over OSR. If operative mortality rates are between 11 percent and 40 percent, Michaels et al. report that the outcome of whether EVAR is a cost-effective alternative to OSR is uncertain.¹⁷ Again, this cost effectiveness analysis did not include the midterm results from RCT.

Although Michaels takes the most rigorous approach employing probabilistic sensitivity analyses, they do not explicitly state their assumptions regarding what cost categories they considered for the initial procedure, nor for the cost of the EVAR prosthesis. The authors state that this new technology comes at a considerable price, yet no explicit monetary value was reported, nor was any sensitivity analyses on the cost of the graft and its effect on the cost effectiveness of EVAR relative to OSR conducted.

Michaels also reported that EVAR was cost effective compared to no intervention in individuals with large AAA who were judged medically unfit for OSR.¹⁷ However, they did not use results from EVAR-2 that directly compared EVAR with no intervention.¹⁴⁵ The findings from this RCT demonstrated that EVAR had a considerable 30-day operative mortality in patients already medically unfit for OSR. EVAR did not improve survival compared to no intervention, had little effect on health-related quality of life, and was associated with a need for continued surveillance and reinterventions, at substantially increased cost. Therefore, the authors of EVAR-2 concluded that they saw no reason to pursue cost-effectiveness modeling.

As shown in Table 15, each of these cost-effectiveness models used different levels of specificity when describing how the costs of treatment were calculated. The Michaels study provided almost no guidance about just what went into their calculations.

None of these Markov studies analyzing the cost effectiveness of AAA repair included the less favorable midterm results regarding quality of life, survival, complications, or need for reintervention described above from recently reported RCTs of EVAR versus OSR. Furthermore, as no RCT studies have been completed within the U.S., the relevance of the cost data incorporated within these Markov studies from other countries is questionable.

Although none of the studies give confidence intervals around the cost effectiveness ratios, the use of probabilistic sensitivity analyses (PSA) and the percent of simulations falling below the reference threshold addresses this issue. However, for those studies not using PSA, because of the large observed variability in operative mortality, EVAR complication rates and the need for reintervention, confidence intervals would help clarify the significance of the point estimate, and the comparability of the reported cost effectiveness ratios. For all of the reasons mentioned above, the cost effectiveness of EVAR versus OSR remains unclear, particularly for specific patient cohorts defined by age, aneurysm size, and comorbidity.