General Review Risk-Adjusted Meta-analysis of 30-Day Mortality of Endovascular Versus Open Repair for Ruptured Abdominal Aortic Aneurysms Thomas Luebke, and Jan Brunkwall, Cologne, Germany
Background: In recent years, the relative benefits of endovascular repair (EVAR) in the treatment of ruptured abdominal aortic aneurysms (rAAAs) compared with those of open repair have been postulated. However, sufficient quantification and evidence-based validation of the role of EVAR in the care pathway for these patients is still lacking. The aim of the present meta-analysis was to investigate the impact of hemodynamic instability and other potential risk factors on 30-day mortality of EVAR versus open repair for rAAAs by performing a meta-regression analysis of previously published data. Methods: Studies comparing perioperative outcomes of endovascular and open repair of ruptured infrarenal or juxtarenal abdominal aortic aneurysm were considered for analysis. All types of comparative studies, including prospective or retrospective, observational studies, or randomized controlled trials (RCTs), were included. Meta-analysis was undertaken using the ManteleHaenszel method, with a standard continuity correction of 0.5. A random-effects model was used owing to the variability in baseline characteristics in each article. Furthermore, an odds ratio (OR) for 30-day mortality adjusted for patients’ hemodynamic condition at presentation in the hospital was calculated by performing a meta-regression analysis. Results: The entire meta-analysis population comprised 81,681 patients (63 studies), of whom 13,706 underwent EVAR and the remaining 67,975 had an open repair of their rAAA. Without correction for hemodynamic instability, patients undergoing EVAR had a significantly lower 30-day mortality rate than patients having open repair (OR, 0.512; 95% confidence interval [CI], 0.457e0.574; P < 0.01). Moderate heterogeneity among the studies was identified (I2 ¼ 53.303%), and the likelihood of publication bias was low (P ¼ 0.183). In the RCTs alone (3 studies), patients undergoing EVAR had no significantly lower 30-day mortality rate than patients with open repair (OR, 0.930; 95% CI, 0.691e1.253; P < 0.633). In all studies available, after adjustment for patients’ hemodynamic condition at presentation to the hospital, the OR for 30day mortality was 0.872 (95% CI, 0.598e1.270; P ¼ 0.474), as well, indicating no significant difference between the 2 therapeutic options. Conclusions: Because a hemodynamically unstable condition may result in poorer clinical outcome, we calculated the 30-day mortality OR adjusted for patients’ hemodynamic condition. After adjustment, there was no benefit in 30-day mortality for EVAR compared with that in open surgery.
Department of Vascular and Endovascular Surgery, University Hospital of Cologne, Cologne, Germany. Correspondence to: Thomas Luebke, MD, Department of Vascular and Endovascular Surgery, University Hospital of Cologne, Kerpener Street 62, 50937 Cologne, Germay; E-mail: [email protected]
Ann Vasc Surg 2015; 29: 845–863 http://dx.doi.org/10.1016/j.avsg.2014.12.014 Ó 2015 Elsevier Inc. All rights reserved. Manuscript received: September 25, 2014; manuscript accepted: December 22, 2014; published online: February 26, 2015.
845
846 Luebke and Brunkwall
Annals of Vascular Surgery
INTRODUCTION
Eligibility Criteria and Study Selection
In recent years, the relative benefits and advantages of endovascular repair (EVAR) in the treatment of ruptured abdominal aortic aneurysms (rAAAs) compared with those of open repair have been postulated1,2. As a result, several institutions have implemented the policy to treat patients with a rAAA with EVAR, provided the anatomy is suitable. However, despite the establishment of standardized protocols for the management of this critical condition in the emergency setting, sufficient quantification of the role of EVAR in the care pathway for these patients is still lacking. Although several cohort studies, randomized trials, and even metaanalyses have provided valid evidence indicating only a short-term advantage of elective endovascular treatment for abdominal aortic aneurysms (AAAs) over conventional surgical repair3e10, in case of rAAAs, several single-center retrospective reports with small number of patients, multicenter cohort, and population-based nonrandomized studies, and few meta-analyses and to date, only 3 randomized controlled trials (RCTs) provided data for EVAR compared with those for conventional open surgery1,2. These studies provided heterogeneous and mixed results regarding the survival benefit of emergency EVAR for rAAA1,2. Even today, there is limited understanding in managing hemodynamically unstable rAAA patients by EVAR and a lack of data on outcomes of hemodynamically stable versus unstable patients with rAAA undergoing EVAR. To enable comparison of the results of EVAR with results of open surgery in patients with a rAAA, it is important to systematically evaluate these published studies and to adjust for differences in inclusion criteria among the studies. The aim of the present meta-analysis was to investigate the impact of hemodynamic instability and other potential risk factors on 30-day mortality of EVAR versus open repair for rAAAs by performing a meta-analysis and meta-regression analysis of previously published data.
Studies comparing perioperative outcomes of endovascular and open repair of ruptured infrarenal or juxtarenal AAA were considered for analysis. All types of comparative studies, including prospective or retrospective, observational studies, or RCTs, were included. Studies reporting patients with acute or symptomatic aneurysm without clearly stating the presence of rupture were excluded (Table I). The primary outcome measure was defined as the 30-day mortality of any cause.
MATERIAL AND METHODS Standard reporting guidelines set by the Preferred Reporting Items for Systematic Reviews and Meta-Analyses Group11 were followed to identify RCTs and other high-level evidence reporting comparative outcomes of endovascular and open repair of rAAA.
Information Sources and Search Strategy Studies were identified by searching electronic bibliographic databases and scanning reference lists of articles. The search was applied to MEDLINE, EMBASE, and Health and Psychosocial Instruments databases, using Ovid Online (version: OvidSP_UI03.04.02.112) from the time of the first report of endovascular treatment of rAAA by Yusuf et al12 in 1994 to August 2014. No language restrictions or filters used to restrict study designs were applied. The following Expanded Medical Subject Headings and keywords were combined with Boolean operators: ‘‘aortic rupture’’ or ‘‘aortic aneurysm rupture’’ or ‘‘ruptured aortic aneurysm’’ and ‘‘EVAR’’ or ‘‘endovascular repair’’ or ‘‘open repair.’’ In addition, the reference lists were manually interrogated as part of a second-level search. A supplementary search of related articles suggested by the PubMed search engine was conducted. All types of comparative studies, including prospective or retrospective, observational studies, or RCTs, were included if: (1) patients who underwent EVAR were compared with patients who underwent open surgery, (2) each treatment group included at least 5 patients (exclusion of small case series or case reports with high risk of selection bias and heterogeneity), (3) information about patients’ hemodynamic condition at presentation to the hospital was reported, and (4) 30-day mortality was reported for each treatment group. If researchers in the studies reported results of current and historic control patients separately, we included the results of only current control patients. When multiple reports from 1 institution were retrieved, the most recent report was included to avoid double counting. Methods of Data Synthesis and Analysis Information extracted from each study can be grouped into (1) that related to the study design
Study period
Type of study
Wu/2014 Gupta/2014 Van Schaik/2011 Eefting/2013 Wallace/2013 Wanhainen/2008 Mani/2011 Chen/2013 Trenner/2013 Mukherjee/2014 R€ odel/2012 Meijenfeldt/2014 Antonopoulos/2014
2005e2012 2005e2010 2006e2008 1991e2012 2007e2012 1994e2005 2005e2009 1998e2009 1999e2010 2007e2011 2006e2010 2000e2013 2006e2012
IMPROVE/2014 Mohan/2013 Edwards/2013
2009e2013 2001e2010 2001e2008
Lo/2013 Mehta I/2006 Park/2013 Reimerink/2013 Nedeau/2012 Noorani/2012 Saqib/2012 Mayer/2012 Mandawat/2012 Ioannidis/2012 Ten Bosch/2011 Sarac/2011 Holt/2010 Starnes/2010
2003e2011 2002e2005 2005e2009 2004e2011 2000e2010 2006e2010 2001e2010 1998e2009 2003e2007 2003e2008 2002e2008 1990e2008 2003e2008 2007e2009
Bosch/2010
2002e2008
Chagpar/2010 Lyons/2010 Davenport/2010 Giles/2009
2003e2008 2006e2007 2005e2007 2000e2005
Hospital-based, retrospective study NSQIP database Hospital-based, retrospective study Hospital-based, retrospective study Hospital-based, retrospective study Swedvasc database VASCUNET database NHIRD database DGG database Hospital-based, retrospective study Hospital-based, retrospective study Hospital-based, retrospective study Multicenter hospital-based observational study, prospectively collected data Multicenter prospective RCT Population-based retrospective, NIS Population-based retrospective study, Medicare beneficiaries VSGNE database Hospital-based, retrospective study Population-based retrospective study RCT Hospital-based retrospective study Hospital-based retrospective study Hospital-based retrospective study Hospital-based retrospective study Population-based retrospective study Hospital-based retrospective study Hospital-based observational study Hospital-based retrospective study Population-based retrospective study Hospital-based prospective observational study Hospital-based prospective observational study Hospital-based retrospective study Hospital-based retrospective study Population-based retrospective study Population-based retrospective study
Total patients, N
35 1,447 56 195 100 3,516 7,040 537 4,859 45 105 221 418
EVAR
15 499 15 78 61 92 824 39 575 38 35 83 113
Open repair
20 9,487 41 117 39 3,424 6,216 498 4,284 7 70 138 305 297 33,986 9,872 (1,099)
NOS/Jadad score
5 5 6 5 5 5 5 6 5 5 7 6 6
623 42,126 10,998 (matched: 2,198) 429
316 8,140 1,126 (1,099)
3 6 6
91
338
16,557 116 74 102 148 361 271 43 129 160 4,414 51
3,796 57 19 52 37 198 64 20 25 32 335 27
12,761 59 55 50 111 163 207 23 104 128 4,079 24
6 7 6 3 7 7 6 7 5 5 7 7 6 7
58
25
33
6
167 37 427 28,429
32 18 99 2,323
135 19 328 26,106
6 6 6 6 (Continued)
Risk-adjusted meta-analysis of 30-day mortality 847
First author/year
Vol. 29, No. 4, May 2015
Table I. Study characteristics
Study period
Type of study
Vun/2009 Verhoeven/2009 Vogel/2009 Visser/2009
2004e2008 2002e2009 2001e2005 2004e2006
Giles/2009 Veith/2009 Wibmer/2008 Lee/2008 Moore/2007
2005e2007 1994e2008 2003e2006 2002e2006 2001e2006
Sharif/2007 Ockert/2007 Anain/2007 Acosta/2007 Mehta II/2013 Arya 2006
2001e2006 2000e2005 2001e2006 2000e2004 2002e2011 2002e2004
Franks/2006
1996e2003
Greco/2006 Coppi/2006 Hinchliffe/2006 Peppelenbosch/2006
2000e2003 1999e2006 2002e2004 2003e2004
Dalainas/2006 Larzon/2005 Kapma/2005 Castelli/2005 Alsac/2005 Vaddineni/2005 Brandt/2005 Lee/2004 Reichart/2003 Resch/2003 Yilmaz/2002
1998e2005 2001e2004 1998e2004 2001e2004 2001e2004 1999e2004 2003e2004 2002e2004 2000e2002 2001e2002 1999e2001
Hospital-based retrospective study Hospital-based retrospective study Population-based retrospective study Multicenter prospective observational study Population-based retrospective study Hospital-based retrospective study Hospital-based retrospective study Hospital-based retrospective study Hospital-based, prospective observational study Hospital-based observational study Hospital-based retrospective study Hospital-based observational study Hospital-based retrospective study Hospital-based retrospective study Hospital-based, retrospective study with prospectively collected data Historically controlled retrospective cohort study Population-based retrospective study Hospital-based observational study RCT Multicenter prospective observational study Hospital-based observational study Hospital-based retrospective study Hospital-based retrospective study Hospital-based retrospective study Hospital-based observational study Hospital-based retrospective study Hospital-based retrospective study Hospital-based retrospective study Hospital-based retrospective study Hospital-based observational study Hospital-based observational study
Total patients, N
EVAR
Open repair
NOS/Jadad score
45 159 700 201
7 45 82 58
38 114 618 143
5 6 5 8
567 57 47 37 126
121 45 16 17 56
446 12 31 20 70
6 6 7 6 8
126 58 40 162 283 51
52 29 30 56 120 17
74 29 10 106 163 34
6 6 6 5 5 7
43
21
22
7
5798 124 32 100
290 33 15 49
5508 91 17 51
6 7 2 7
28 41 197 46 37 24 24 17 19 37 46
20 15 25 25 17 9 11 13 6 14 17
8 26 172 21 20 15 13 4 13 23 29
6 7 6 7 6 6 7 6 6 7 6
The NewcastleeOttawa Scale was applied to evaluate the methodology quality of the studies. This scale was developed to assess the quality of studies using a ‘‘star system’’ (maximum 9 stars), in which a study is judged on 3 broad perspectives: (S) the selection of the study groups (maximum 4 stars), (C) the comparability of the groups (maximum 2 stars), and (O) the ascertainment of outcome of interest (maximum 3 stars). The Jadad scale was applied for the assessment of RCTs.
Annals of Vascular Surgery
First author/year
848 Luebke and Brunkwall
Table I. Continued
Vol. 29, No. 4, May 2015
and characteristics (first author, year of publication, patient recruitment period, type of study, and inclusion/exclusion criteria for patient enrollment); (2) data related to clinical and demographic characteristics of the study populations, including age, sex, presence of hypertension, coronary artery disease, diabetes mellitus, chronic renal insufficiency, and hemodynamic instability; and (3) outcome parameters, as outlined previously mentioned. The methodologic quality of the included observational studies was ascertained with the NewcastleeOttawa Scale (NOS),13 and the Jadad scale was applied for the assessment of RCTs.14 The NOS tool evaluates 3 main methodologic elements of caseecontrol studies: selection methods (adequate case definition, representativeness of the cases, appropriate selection, and definition of controls), comparability of cases and controls on the basis of the design or analysis, and assessment of exposure (ascertainment of exposure and nonresponse rate). The scale uses a star system, with a maximum of 9 stars; studies achieving 6 stars were considered to be of higher quality. Statistical Analysis Meta-analysis was undertaken using the Mantele Haenszel method, with a standard continuity correction of 0.5. A random-effects model was used owing to the variability in baseline characteristics in each article. An a level of 0.05 was used to determine statistical significance. The heterogeneity of treatment effect among trials was assessed using the I2 statistic. This describes the percentage of total variation across studies that is due to heterogeneity rather than chance or random error.15 A value greater than 50% reflects significant heterogeneity owing to real differences in study populations, protocols, interventions, and outcomes. Individual study odds ratios (ORs) and 95% confidence intervals (CIs) were calculated from event numbers extracted from each study before data pooling. In calculation of the OR, the total number of patients assigned in each group was used as the denominator. Summary estimates of ORs were obtained with a randomeffects model. The crude OR for 30-day mortality, with the 95% CI, was calculated for EVAR versus open surgery (i.e., an OR >1.0 favored EVAR). A prespecified sensitivity analysis of observational studies was undertaken by pooling the ORs of EVAR versus OR adjusted for at least age, sex, and hemodynamic instability. Meta-regression analyses were subsequently performed on operative mortality in an attempt to explain the observed
Risk-adjusted meta-analysis of 30-day mortality 849
heterogeneity between-study estimates. The effect of year of publication, gender, coronary artery disease, diabetes mellitus, chronic renal insufficiency, and hemodynamic instability were all included individually as covariates in the meta-regressions. For each meta-regression, the slope coefficient (standard error [s.e.]), the P value, and the Tau2 are reported along with a meta-regression plot of each covariate against operative mortality. The P value indicates the strength of association, the graph shows the direction (i.e., positive or negative), and the slope informs how much the outcome changes per unit increase in the covariate. The Tau2 (the betweenstudy variance) indicates how much residual heterogeneity exists which has not been explained by the covariate. A full meta-analysis random-effect approach to the regression had been used, where studies are weighted by a combination of their within-study variance and the degree of heterogeneity. Furthermore, an OR for 30-day mortality adjusted for patients’ hemodynamic condition at presentation in the hospital was calculated. For this analysis, we used the parameter hemodynamic instability to predict the log OR as the effect size. To test the regression coefficient of the metaregression for significance, we used a t test of the form. Although in meta-analysis, the coefficient for any covariate (B) and its standard error (SEB) are based on groups of studies rather than groups of subjects, in meta-regression, the test is based on the Zdistribution. Thus, we tested the significance of the slope of the meta-regression by applying the formula Z¼
B SEB
The Z-test can be used to test the statistical significance of any single coefficient. Quantify the Magnitude of the Relationship The Z-test, like all tests of significance, speaks to the question of statistical, rather than substantive, significance. Therefore, in addition to reporting the test of significance, we reported the magnitude of the relationship. Here, the relationship of hemodynamic instability to effect (expressed as a log OR) is lnðORÞ ¼ Intercept SlopeðXÞ where X is the hemodynamic instability. In addition to that, we calculated the 95% CI for by using the following formulas:
850 Luebke and Brunkwall
LLB ¼ B 1:96 SEB ULB ¼ B þ 1:96 SEB As a result, we could calculate the real coefficient with its lower and upper limits. These limits were used to generate the CIs. Potential publication bias (i.e., bias resulting from the greater likelihood of publishing favorable results) was assessed with the Egger test and represented graphically with Begg funnel plots of the natural log of the OR versus its s.e.16,17. In the present study, we plotted the reciprocal of the s.e. of the 30-day mortality OR of each study as a function of the natural logarithm of the 30-day mortality OR. If no publication bias is present, the data points are distributed in a symmetric fashion. Sensitivity analyses were performed to assess the contribution of each study to the pooled estimate by excluding individual studies one at a time and recalculating the pooled OR estimates for the remaining studies. Separate meta-analyses of risk-adjusted observational studies for in-hospital (or 30-day) mortality and RCTs were undertaken. We explored potential heterogeneity in estimates with univariate meta-regression to identify trends in the treatment effects over time and by comparing summary results obtained from subsets of studies grouped by study design and quality. Furthermore, to test for heterogeneity in patients’ hemodynamic condition at presentation to the hospital across the studies, we used the chi-square test. A 2-sided P value of < 0.05 was regarded as significant for all analyses. Analyses were performed the Comprehensive Meta-Analysis 2.0 software (Biostat, Englewood, NJ).
RESULTS Literature search identified 5,124 relevant articles, of which 74 were reviewed in full text (Fig. 1). Five records had been excluded because they were series from the same institutions with duplicate clinical material or studies reported mixed outcome data of treatments in symptomatic and rAAA, and 4 were single case reports. Further 2 articles had been excluded because they were reviews or population-based studies from which accurate data could not be extracted. Another 5,050 records were excluded because the title or the abstract were not relevant. This left 64 studies (81,681 patients) for analysis (Table I)18e82,86,87,91e93. The entire meta-analysis population comprised 81,681 patients, of whom 13,706 underwent EVAR and the remaining 67,975 had an open repair of their rAAA. The
Annals of Vascular Surgery
studies selected for analysis were published between 2002 and 2014, whereas the patient recruitment period extended from 1990 to 2013. Three RCTs (Hinchliffe, AJAX, and IMPROVE ) were identified, with the remaining reports being single-center or multicenter prospective or retrospective observational studies or populationbased reports. The methodologic quality of the RCTs, represented in the Jadad score, was low. Similarly, a small proportion of the observational studies achieved a NOS score >6 (19 of 63 studies). Main demographic and clinical features of the study populations are outlined in Tables I and II.
OUTCOMES 30-Day Mortality All 64 studies reported data of the primary outcome parameter set for the meta-analysis. 30-Day mortality occurred in 3,644 of 13,706 (26.59%) patients treated with EVAR and in 28,480 of 67,975 (41.90%) of the patients who underwent open repair. Thus, unstratified patients undergoing EVAR had a significantly lower 30day mortality rate than patients having open repair (OR, 0.512; 95% CI, 0.457e0.574; P < 0.01). Moderate heterogeneity among the studies was identified (I2 ¼ 53.303%), and the likelihood of publication bias was low (P ¼ 0.183; Fig. 2). A funnel plot for 30-day mortality revealed no large degree of asymmetry (Fig. 3). Sensitivity Analyses and Meta-regression The combined outcome estimate of 30-day mortality was not substantially affected when the primary analysis was repeated with altered data sets after excluding each single study at a time (OR, 0.509; 95% CI, 0.455e0.569; P < 0.01). Furthermore, the effect on 30-day mortality was not different in hospital-based (OR, 0.458; 95% CI, 0.384e0.547; P < 0.01) compared with population-based studies (OR, 0.581; 95% CI, 0.494e0.684; P < 0.01; P ¼ 0.17 for interaction). In addition, sensitivity analysis was performed by analyzing the RCTs and the nonrandomized data separately. In the RCTs, patients undergoing EVAR had no significantly lower 30-day mortality rate compared with patients with open repair (OR, 0.930; 95% CI, 0.691e1.253; P < 0.633). No significant heterogeneity among the RCTs was identified (I2 ¼ 0%), and the likelihood of publication bias was low (P ¼ 0.29; Fig. 4). In contrast, in
Vol. 29, No. 4, May 2015
Risk-adjusted meta-analysis of 30-day mortality 851
Fig. 1. Flow chart showing the strategy used for the literature search.
nonrandomized data, patients undergoing EVAR had a significantly lower 30-day mortality rate than patients having open repair (OR, 0.490; 95% CI, 0.438e0.549; P < 0.01). Moderate heterogeneity among the nonrandomized studies was identified (I2 ¼ 50.46%), and the likelihood of publication bias was low (P ¼ 0.076). In
nonrandomized data excluding low-quality studies (
Background: In recent years, the relative benefits of endovascular repair (EVAR) in the treatment of ruptured abdominal aortic aneurysms (rAAAs) compared with those of open repair have been postulated. However, sufficient quantification and evidence-based validation of the role of EVAR in the care pathway for these patients is still lacking. The aim of the present meta-analysis was to investigate the impact of hemodynamic instability and other potential risk factors on 30-day mortality of EVAR versus open repair for rAAAs by performing a meta-regression analysis of previously published data. Methods: Studies comparing perioperative outcomes of endovascular and open repair of ruptured infrarenal or juxtarenal abdominal aortic aneurysm were considered for analysis. All types of comparative studies, including prospective or retrospective, observational studies, or randomized controlled trials (RCTs), were included. Meta-analysis was undertaken using the ManteleHaenszel method, with a standard continuity correction of 0.5. A random-effects model was used owing to the variability in baseline characteristics in each article. Furthermore, an odds ratio (OR) for 30-day mortality adjusted for patients’ hemodynamic condition at presentation in the hospital was calculated by performing a meta-regression analysis. Results: The entire meta-analysis population comprised 81,681 patients (63 studies), of whom 13,706 underwent EVAR and the remaining 67,975 had an open repair of their rAAA. Without correction for hemodynamic instability, patients undergoing EVAR had a significantly lower 30-day mortality rate than patients having open repair (OR, 0.512; 95% confidence interval [CI], 0.457e0.574; P < 0.01). Moderate heterogeneity among the studies was identified (I2 ¼ 53.303%), and the likelihood of publication bias was low (P ¼ 0.183). In the RCTs alone (3 studies), patients undergoing EVAR had no significantly lower 30-day mortality rate than patients with open repair (OR, 0.930; 95% CI, 0.691e1.253; P < 0.633). In all studies available, after adjustment for patients’ hemodynamic condition at presentation to the hospital, the OR for 30day mortality was 0.872 (95% CI, 0.598e1.270; P ¼ 0.474), as well, indicating no significant difference between the 2 therapeutic options. Conclusions: Because a hemodynamically unstable condition may result in poorer clinical outcome, we calculated the 30-day mortality OR adjusted for patients’ hemodynamic condition. After adjustment, there was no benefit in 30-day mortality for EVAR compared with that in open surgery.
Department of Vascular and Endovascular Surgery, University Hospital of Cologne, Cologne, Germany. Correspondence to: Thomas Luebke, MD, Department of Vascular and Endovascular Surgery, University Hospital of Cologne, Kerpener Street 62, 50937 Cologne, Germay; E-mail: [email protected]
Ann Vasc Surg 2015; 29: 845–863 http://dx.doi.org/10.1016/j.avsg.2014.12.014 Ó 2015 Elsevier Inc. All rights reserved. Manuscript received: September 25, 2014; manuscript accepted: December 22, 2014; published online: February 26, 2015.
845
846 Luebke and Brunkwall
Annals of Vascular Surgery
INTRODUCTION
Eligibility Criteria and Study Selection
In recent years, the relative benefits and advantages of endovascular repair (EVAR) in the treatment of ruptured abdominal aortic aneurysms (rAAAs) compared with those of open repair have been postulated1,2. As a result, several institutions have implemented the policy to treat patients with a rAAA with EVAR, provided the anatomy is suitable. However, despite the establishment of standardized protocols for the management of this critical condition in the emergency setting, sufficient quantification of the role of EVAR in the care pathway for these patients is still lacking. Although several cohort studies, randomized trials, and even metaanalyses have provided valid evidence indicating only a short-term advantage of elective endovascular treatment for abdominal aortic aneurysms (AAAs) over conventional surgical repair3e10, in case of rAAAs, several single-center retrospective reports with small number of patients, multicenter cohort, and population-based nonrandomized studies, and few meta-analyses and to date, only 3 randomized controlled trials (RCTs) provided data for EVAR compared with those for conventional open surgery1,2. These studies provided heterogeneous and mixed results regarding the survival benefit of emergency EVAR for rAAA1,2. Even today, there is limited understanding in managing hemodynamically unstable rAAA patients by EVAR and a lack of data on outcomes of hemodynamically stable versus unstable patients with rAAA undergoing EVAR. To enable comparison of the results of EVAR with results of open surgery in patients with a rAAA, it is important to systematically evaluate these published studies and to adjust for differences in inclusion criteria among the studies. The aim of the present meta-analysis was to investigate the impact of hemodynamic instability and other potential risk factors on 30-day mortality of EVAR versus open repair for rAAAs by performing a meta-analysis and meta-regression analysis of previously published data.
Studies comparing perioperative outcomes of endovascular and open repair of ruptured infrarenal or juxtarenal AAA were considered for analysis. All types of comparative studies, including prospective or retrospective, observational studies, or RCTs, were included. Studies reporting patients with acute or symptomatic aneurysm without clearly stating the presence of rupture were excluded (Table I). The primary outcome measure was defined as the 30-day mortality of any cause.
MATERIAL AND METHODS Standard reporting guidelines set by the Preferred Reporting Items for Systematic Reviews and Meta-Analyses Group11 were followed to identify RCTs and other high-level evidence reporting comparative outcomes of endovascular and open repair of rAAA.
Information Sources and Search Strategy Studies were identified by searching electronic bibliographic databases and scanning reference lists of articles. The search was applied to MEDLINE, EMBASE, and Health and Psychosocial Instruments databases, using Ovid Online (version: OvidSP_UI03.04.02.112) from the time of the first report of endovascular treatment of rAAA by Yusuf et al12 in 1994 to August 2014. No language restrictions or filters used to restrict study designs were applied. The following Expanded Medical Subject Headings and keywords were combined with Boolean operators: ‘‘aortic rupture’’ or ‘‘aortic aneurysm rupture’’ or ‘‘ruptured aortic aneurysm’’ and ‘‘EVAR’’ or ‘‘endovascular repair’’ or ‘‘open repair.’’ In addition, the reference lists were manually interrogated as part of a second-level search. A supplementary search of related articles suggested by the PubMed search engine was conducted. All types of comparative studies, including prospective or retrospective, observational studies, or RCTs, were included if: (1) patients who underwent EVAR were compared with patients who underwent open surgery, (2) each treatment group included at least 5 patients (exclusion of small case series or case reports with high risk of selection bias and heterogeneity), (3) information about patients’ hemodynamic condition at presentation to the hospital was reported, and (4) 30-day mortality was reported for each treatment group. If researchers in the studies reported results of current and historic control patients separately, we included the results of only current control patients. When multiple reports from 1 institution were retrieved, the most recent report was included to avoid double counting. Methods of Data Synthesis and Analysis Information extracted from each study can be grouped into (1) that related to the study design
Study period
Type of study
Wu/2014 Gupta/2014 Van Schaik/2011 Eefting/2013 Wallace/2013 Wanhainen/2008 Mani/2011 Chen/2013 Trenner/2013 Mukherjee/2014 R€ odel/2012 Meijenfeldt/2014 Antonopoulos/2014
2005e2012 2005e2010 2006e2008 1991e2012 2007e2012 1994e2005 2005e2009 1998e2009 1999e2010 2007e2011 2006e2010 2000e2013 2006e2012
IMPROVE/2014 Mohan/2013 Edwards/2013
2009e2013 2001e2010 2001e2008
Lo/2013 Mehta I/2006 Park/2013 Reimerink/2013 Nedeau/2012 Noorani/2012 Saqib/2012 Mayer/2012 Mandawat/2012 Ioannidis/2012 Ten Bosch/2011 Sarac/2011 Holt/2010 Starnes/2010
2003e2011 2002e2005 2005e2009 2004e2011 2000e2010 2006e2010 2001e2010 1998e2009 2003e2007 2003e2008 2002e2008 1990e2008 2003e2008 2007e2009
Bosch/2010
2002e2008
Chagpar/2010 Lyons/2010 Davenport/2010 Giles/2009
2003e2008 2006e2007 2005e2007 2000e2005
Hospital-based, retrospective study NSQIP database Hospital-based, retrospective study Hospital-based, retrospective study Hospital-based, retrospective study Swedvasc database VASCUNET database NHIRD database DGG database Hospital-based, retrospective study Hospital-based, retrospective study Hospital-based, retrospective study Multicenter hospital-based observational study, prospectively collected data Multicenter prospective RCT Population-based retrospective, NIS Population-based retrospective study, Medicare beneficiaries VSGNE database Hospital-based, retrospective study Population-based retrospective study RCT Hospital-based retrospective study Hospital-based retrospective study Hospital-based retrospective study Hospital-based retrospective study Population-based retrospective study Hospital-based retrospective study Hospital-based observational study Hospital-based retrospective study Population-based retrospective study Hospital-based prospective observational study Hospital-based prospective observational study Hospital-based retrospective study Hospital-based retrospective study Population-based retrospective study Population-based retrospective study
Total patients, N
35 1,447 56 195 100 3,516 7,040 537 4,859 45 105 221 418
EVAR
15 499 15 78 61 92 824 39 575 38 35 83 113
Open repair
20 9,487 41 117 39 3,424 6,216 498 4,284 7 70 138 305 297 33,986 9,872 (1,099)
NOS/Jadad score
5 5 6 5 5 5 5 6 5 5 7 6 6
623 42,126 10,998 (matched: 2,198) 429
316 8,140 1,126 (1,099)
3 6 6
91
338
16,557 116 74 102 148 361 271 43 129 160 4,414 51
3,796 57 19 52 37 198 64 20 25 32 335 27
12,761 59 55 50 111 163 207 23 104 128 4,079 24
6 7 6 3 7 7 6 7 5 5 7 7 6 7
58
25
33
6
167 37 427 28,429
32 18 99 2,323
135 19 328 26,106
6 6 6 6 (Continued)
Risk-adjusted meta-analysis of 30-day mortality 847
First author/year
Vol. 29, No. 4, May 2015
Table I. Study characteristics
Study period
Type of study
Vun/2009 Verhoeven/2009 Vogel/2009 Visser/2009
2004e2008 2002e2009 2001e2005 2004e2006
Giles/2009 Veith/2009 Wibmer/2008 Lee/2008 Moore/2007
2005e2007 1994e2008 2003e2006 2002e2006 2001e2006
Sharif/2007 Ockert/2007 Anain/2007 Acosta/2007 Mehta II/2013 Arya 2006
2001e2006 2000e2005 2001e2006 2000e2004 2002e2011 2002e2004
Franks/2006
1996e2003
Greco/2006 Coppi/2006 Hinchliffe/2006 Peppelenbosch/2006
2000e2003 1999e2006 2002e2004 2003e2004
Dalainas/2006 Larzon/2005 Kapma/2005 Castelli/2005 Alsac/2005 Vaddineni/2005 Brandt/2005 Lee/2004 Reichart/2003 Resch/2003 Yilmaz/2002
1998e2005 2001e2004 1998e2004 2001e2004 2001e2004 1999e2004 2003e2004 2002e2004 2000e2002 2001e2002 1999e2001
Hospital-based retrospective study Hospital-based retrospective study Population-based retrospective study Multicenter prospective observational study Population-based retrospective study Hospital-based retrospective study Hospital-based retrospective study Hospital-based retrospective study Hospital-based, prospective observational study Hospital-based observational study Hospital-based retrospective study Hospital-based observational study Hospital-based retrospective study Hospital-based retrospective study Hospital-based, retrospective study with prospectively collected data Historically controlled retrospective cohort study Population-based retrospective study Hospital-based observational study RCT Multicenter prospective observational study Hospital-based observational study Hospital-based retrospective study Hospital-based retrospective study Hospital-based retrospective study Hospital-based observational study Hospital-based retrospective study Hospital-based retrospective study Hospital-based retrospective study Hospital-based retrospective study Hospital-based observational study Hospital-based observational study
Total patients, N
EVAR
Open repair
NOS/Jadad score
45 159 700 201
7 45 82 58
38 114 618 143
5 6 5 8
567 57 47 37 126
121 45 16 17 56
446 12 31 20 70
6 6 7 6 8
126 58 40 162 283 51
52 29 30 56 120 17
74 29 10 106 163 34
6 6 6 5 5 7
43
21
22
7
5798 124 32 100
290 33 15 49
5508 91 17 51
6 7 2 7
28 41 197 46 37 24 24 17 19 37 46
20 15 25 25 17 9 11 13 6 14 17
8 26 172 21 20 15 13 4 13 23 29
6 7 6 7 6 6 7 6 6 7 6
The NewcastleeOttawa Scale was applied to evaluate the methodology quality of the studies. This scale was developed to assess the quality of studies using a ‘‘star system’’ (maximum 9 stars), in which a study is judged on 3 broad perspectives: (S) the selection of the study groups (maximum 4 stars), (C) the comparability of the groups (maximum 2 stars), and (O) the ascertainment of outcome of interest (maximum 3 stars). The Jadad scale was applied for the assessment of RCTs.
Annals of Vascular Surgery
First author/year
848 Luebke and Brunkwall
Table I. Continued
Vol. 29, No. 4, May 2015
and characteristics (first author, year of publication, patient recruitment period, type of study, and inclusion/exclusion criteria for patient enrollment); (2) data related to clinical and demographic characteristics of the study populations, including age, sex, presence of hypertension, coronary artery disease, diabetes mellitus, chronic renal insufficiency, and hemodynamic instability; and (3) outcome parameters, as outlined previously mentioned. The methodologic quality of the included observational studies was ascertained with the NewcastleeOttawa Scale (NOS),13 and the Jadad scale was applied for the assessment of RCTs.14 The NOS tool evaluates 3 main methodologic elements of caseecontrol studies: selection methods (adequate case definition, representativeness of the cases, appropriate selection, and definition of controls), comparability of cases and controls on the basis of the design or analysis, and assessment of exposure (ascertainment of exposure and nonresponse rate). The scale uses a star system, with a maximum of 9 stars; studies achieving 6 stars were considered to be of higher quality. Statistical Analysis Meta-analysis was undertaken using the Mantele Haenszel method, with a standard continuity correction of 0.5. A random-effects model was used owing to the variability in baseline characteristics in each article. An a level of 0.05 was used to determine statistical significance. The heterogeneity of treatment effect among trials was assessed using the I2 statistic. This describes the percentage of total variation across studies that is due to heterogeneity rather than chance or random error.15 A value greater than 50% reflects significant heterogeneity owing to real differences in study populations, protocols, interventions, and outcomes. Individual study odds ratios (ORs) and 95% confidence intervals (CIs) were calculated from event numbers extracted from each study before data pooling. In calculation of the OR, the total number of patients assigned in each group was used as the denominator. Summary estimates of ORs were obtained with a randomeffects model. The crude OR for 30-day mortality, with the 95% CI, was calculated for EVAR versus open surgery (i.e., an OR >1.0 favored EVAR). A prespecified sensitivity analysis of observational studies was undertaken by pooling the ORs of EVAR versus OR adjusted for at least age, sex, and hemodynamic instability. Meta-regression analyses were subsequently performed on operative mortality in an attempt to explain the observed
Risk-adjusted meta-analysis of 30-day mortality 849
heterogeneity between-study estimates. The effect of year of publication, gender, coronary artery disease, diabetes mellitus, chronic renal insufficiency, and hemodynamic instability were all included individually as covariates in the meta-regressions. For each meta-regression, the slope coefficient (standard error [s.e.]), the P value, and the Tau2 are reported along with a meta-regression plot of each covariate against operative mortality. The P value indicates the strength of association, the graph shows the direction (i.e., positive or negative), and the slope informs how much the outcome changes per unit increase in the covariate. The Tau2 (the betweenstudy variance) indicates how much residual heterogeneity exists which has not been explained by the covariate. A full meta-analysis random-effect approach to the regression had been used, where studies are weighted by a combination of their within-study variance and the degree of heterogeneity. Furthermore, an OR for 30-day mortality adjusted for patients’ hemodynamic condition at presentation in the hospital was calculated. For this analysis, we used the parameter hemodynamic instability to predict the log OR as the effect size. To test the regression coefficient of the metaregression for significance, we used a t test of the form. Although in meta-analysis, the coefficient for any covariate (B) and its standard error (SEB) are based on groups of studies rather than groups of subjects, in meta-regression, the test is based on the Zdistribution. Thus, we tested the significance of the slope of the meta-regression by applying the formula Z¼
B SEB
The Z-test can be used to test the statistical significance of any single coefficient. Quantify the Magnitude of the Relationship The Z-test, like all tests of significance, speaks to the question of statistical, rather than substantive, significance. Therefore, in addition to reporting the test of significance, we reported the magnitude of the relationship. Here, the relationship of hemodynamic instability to effect (expressed as a log OR) is lnðORÞ ¼ Intercept SlopeðXÞ where X is the hemodynamic instability. In addition to that, we calculated the 95% CI for by using the following formulas:
850 Luebke and Brunkwall
LLB ¼ B 1:96 SEB ULB ¼ B þ 1:96 SEB As a result, we could calculate the real coefficient with its lower and upper limits. These limits were used to generate the CIs. Potential publication bias (i.e., bias resulting from the greater likelihood of publishing favorable results) was assessed with the Egger test and represented graphically with Begg funnel plots of the natural log of the OR versus its s.e.16,17. In the present study, we plotted the reciprocal of the s.e. of the 30-day mortality OR of each study as a function of the natural logarithm of the 30-day mortality OR. If no publication bias is present, the data points are distributed in a symmetric fashion. Sensitivity analyses were performed to assess the contribution of each study to the pooled estimate by excluding individual studies one at a time and recalculating the pooled OR estimates for the remaining studies. Separate meta-analyses of risk-adjusted observational studies for in-hospital (or 30-day) mortality and RCTs were undertaken. We explored potential heterogeneity in estimates with univariate meta-regression to identify trends in the treatment effects over time and by comparing summary results obtained from subsets of studies grouped by study design and quality. Furthermore, to test for heterogeneity in patients’ hemodynamic condition at presentation to the hospital across the studies, we used the chi-square test. A 2-sided P value of < 0.05 was regarded as significant for all analyses. Analyses were performed the Comprehensive Meta-Analysis 2.0 software (Biostat, Englewood, NJ).
RESULTS Literature search identified 5,124 relevant articles, of which 74 were reviewed in full text (Fig. 1). Five records had been excluded because they were series from the same institutions with duplicate clinical material or studies reported mixed outcome data of treatments in symptomatic and rAAA, and 4 were single case reports. Further 2 articles had been excluded because they were reviews or population-based studies from which accurate data could not be extracted. Another 5,050 records were excluded because the title or the abstract were not relevant. This left 64 studies (81,681 patients) for analysis (Table I)18e82,86,87,91e93. The entire meta-analysis population comprised 81,681 patients, of whom 13,706 underwent EVAR and the remaining 67,975 had an open repair of their rAAA. The
Annals of Vascular Surgery
studies selected for analysis were published between 2002 and 2014, whereas the patient recruitment period extended from 1990 to 2013. Three RCTs (Hinchliffe, AJAX, and IMPROVE ) were identified, with the remaining reports being single-center or multicenter prospective or retrospective observational studies or populationbased reports. The methodologic quality of the RCTs, represented in the Jadad score, was low. Similarly, a small proportion of the observational studies achieved a NOS score >6 (19 of 63 studies). Main demographic and clinical features of the study populations are outlined in Tables I and II.
OUTCOMES 30-Day Mortality All 64 studies reported data of the primary outcome parameter set for the meta-analysis. 30-Day mortality occurred in 3,644 of 13,706 (26.59%) patients treated with EVAR and in 28,480 of 67,975 (41.90%) of the patients who underwent open repair. Thus, unstratified patients undergoing EVAR had a significantly lower 30day mortality rate than patients having open repair (OR, 0.512; 95% CI, 0.457e0.574; P < 0.01). Moderate heterogeneity among the studies was identified (I2 ¼ 53.303%), and the likelihood of publication bias was low (P ¼ 0.183; Fig. 2). A funnel plot for 30-day mortality revealed no large degree of asymmetry (Fig. 3). Sensitivity Analyses and Meta-regression The combined outcome estimate of 30-day mortality was not substantially affected when the primary analysis was repeated with altered data sets after excluding each single study at a time (OR, 0.509; 95% CI, 0.455e0.569; P < 0.01). Furthermore, the effect on 30-day mortality was not different in hospital-based (OR, 0.458; 95% CI, 0.384e0.547; P < 0.01) compared with population-based studies (OR, 0.581; 95% CI, 0.494e0.684; P < 0.01; P ¼ 0.17 for interaction). In addition, sensitivity analysis was performed by analyzing the RCTs and the nonrandomized data separately. In the RCTs, patients undergoing EVAR had no significantly lower 30-day mortality rate compared with patients with open repair (OR, 0.930; 95% CI, 0.691e1.253; P < 0.633). No significant heterogeneity among the RCTs was identified (I2 ¼ 0%), and the likelihood of publication bias was low (P ¼ 0.29; Fig. 4). In contrast, in
Vol. 29, No. 4, May 2015
Risk-adjusted meta-analysis of 30-day mortality 851
Fig. 1. Flow chart showing the strategy used for the literature search.
nonrandomized data, patients undergoing EVAR had a significantly lower 30-day mortality rate than patients having open repair (OR, 0.490; 95% CI, 0.438e0.549; P < 0.01). Moderate heterogeneity among the nonrandomized studies was identified (I2 ¼ 50.46%), and the likelihood of publication bias was low (P ¼ 0.076). In
nonrandomized data excluding low-quality studies (
