- Email: [email protected]
Predictors of 1-Year Survival After Endovascular Aneurysm Repair
Eur J Vasc Endovasc Surg (2016)
-,
1e7
Predictors of 1-Year Survival After Endovascular Aneurysm Repair R.A. Fitridge a b
a,*
, M. Boult a, T. de Loryn a, P. Cowled a, M. Barnes
b
Discipline of Surgery, The University of Adelaide, Royal Adelaide Hospital and The Queen Elizabeth Hospital, Woodville, South Australia 5011, Australia CSIRO, Adelaide, South Australia 5000, Australia
WHAT THIS PAPER ADDS This article provides a comprehensive analysis of factors predicting 1-year all-cause survival after endovascular aneurysm repair (EVAR), something that has not been reported previously. We consider that 1 year is a reasonable and minimum amount of time that a patient undergoing elective surgery should expect to survive and therefore obtain benefit from the procedure. This study helps to identify which patients have a significantly elevated risk of dying within that first year after surgery. Surgeons could use this information to ensure that patients have understood and considered the risks and benefits of their elective EVAR prior to surgery.
Objective/background: The purpose of this study was to determine the preoperative variables that best predict 1-year survival following elective endovascular aneurysm repair (EVAR), a period of time that would suggest the patient had benefited from the procedure. Most EVAR survival studies focus on early and late survival; scant information is available for 1-year survival. Methods: Data from two Australian audits of EVAR (1999e2001 and 2009e13) were combined (n ¼ 1,647). Preoperative variables included routine demographic data, clinical health assessments, computed tomographyderived anatomical data, and all-cause mortality. Univariate and multivariate logistic regressions determined which variables best predicted 1-year survival. Results: One-year survival after EVAR was 93.7% (1,544/1,647) and 30-day survival was 98.4% (1,620/1,647). Univariate analyses found that nine preoperative variables were significantly associated with 1-year survival. Five variables were included in the final multivariate model: American Society of Anesthesiologists physical status, aneurysm diameter, creatinine, respiratory assessment, and severity of iliac artery calcification (receivere operator curve 0.717, R2 ¼ .117). Predicted 1-year survival ranged from 98.6% to 68.0%, based on differences in aneurysm size and patient comorbidities. Conclusion: Personalised 1-year survival risk enables surgeons and patients to consider seriously the risks and benefits of EVAR prior to surgery. Ó 2015 European Society for Vascular Surgery. Published by Elsevier Ltd. All rights reserved. Article history: Received 26 March 2015, Accepted 12 December 2015, Available online XXX Keywords: Abdominal aortic aneurysm, Mortality, Endovascular procedures
INTRODUCTION Survival after repair of an abdominal aortic aneurysm (AAA) is unarguably the most important end point for patients considering elective surgery and one they will need to balance against their likelihood of surviving without an operation. Estimates for risk of rupture per year based on the size of aneurysm are imprecise but are quoted as follows: <4 cm, <0.2%; 4e4.9 cm, 0.3e0.6% in men and 1.0e 3.0% in women; 5e5.9 cm, 3.0e15.0%; 6.0e6.9 cm, 10.0e 20.0%; 7.0e7.9 cm, 20.0e40.0%; 8 cm, 30.0e50.0%. * Corresponding author. E-mail address: robert.fi[email protected] (R.A. Fitridge). 1078-5884/Ó 2015 European Society for Vascular Surgery. Published by Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.ejvs.2015.12.019
However, risk can vary depending on pre-existing conditions and rates of aneurysm expansion.1,2 Undergoing elective endovascular aneurysm repair (EVAR) carries a risk. Two systematic reviews comparing endovascular and open repair found 30-day mortality to be lower after EVAR than after open repair (1.4% vs. 4.2%) but had no survival advantage thereafter and was associated with an increased rate of complications and reinterventions.3,4 The UK EVAR-1 and EVAR-2 trials reported 30-day mortality to be 1.8% for healthier patients (EVAR-1) and 7.3% for less fit patients (EVAR-2), showing a marked survival differential between healthier and sicker patients.5 It has previously been shown that the factors that most strongly influenced 3and 5-year survival are aneurysm diameter, American Society of Anesthesiologists (ASA) status, age, and creatinine.6,7 Three-year survival varied between 91% for younger patients with low ASA status, creatinine, and aneurysm size and 44% for older, less fit patients.8
Please cite this article in press as: Fitridge RA, et al., Predictors of 1-Year Survival After Endovascular Aneurysm Repair, European Journal of Vascular and Endovascular Surgery (2016), http://dx.doi.org/10.1016/j.ejvs.2015.12.019
2
Reliable figures for 1-year survival after EVAR have been less well documented than early and late survival, but published figures suggest ranges between 82% and 98% (details and references given in Table S1; see Supplementary Material). Hence, it is possible that a patient with a 6% risk of rupture could have an 18% risk of dying in the year after the procedure. In such a case the risk in that year outweighs the benefit, a factor that patients and clinicians may want to consider preoperatively. Highlighting the specific factors that most strongly influence 1-year survival after EVAR could help clinicians and patients decide whether, under certain circumstances, an intervention is an unnecessary burden. Not only is 1 year a reasonable period that a patient might expect to survive, but it is also more statistically reliable than early death (within 30 days of the procedure) as there are more deaths at 12 months. However, a note of caution about modelling is warranted: the ability to predict actual time of death for an individual is imprecise. The aim here is to provide a useful guide as to which aneurysm patients may have an elevated risk of dying within 1 year. METHODS Data and definitions This study was a retrospective review of two prospectively maintained datasets from two Australian audits of EVAR, which were combined for analysis. Patients underwent elective EVAR procedures between November 1999 and May 2001 (AUS-01)6 and between January 2009 and May 2013 (AUS-13). The majority (76%) of the grafts were Zenith grafts (Cook, Bloomington, IN, USA) and the aorto-bi-iliacbifurcated configuration was used in >90% of cases. Clinical data were self-reported by the surgeon or a staff member. Data from the AUS-13 dataset were also independently audited. All patients were followed for 1 year and date of death was supplied by the Australian Institute of Health and Welfare National Death Index (NDI), which captures all deaths in Australia. One-year survival was calculated from date of the primary EVAR procedure. Preoperative variables, their definitions and assumptions made when combining the data, are shown in Table S2 (see Supplementary Material). Information regarding iliac artery calcification and tortuosity was collected from preoperative computed tomography (CT) scans according to individual clinician opinion of severity (none, mild, moderate, severe). In the AUS-13 group, this was for both iliac arteries (common and external on both sides); in the AUS-01 group one global measurement was provided for left and right iliac arteries. For comparison purposes the iliac system with the most severe grading was selected. Ethics approval Ethics approval for AUS-01 was obtained from the Ethics Committee for the Royal Australasian College of Surgeons. The AUS-13 trial obtained state-wide and institutional ethics
R.A. Fitridge et al.
approval from all 25 participating institutions. Ethics approval to link datasets with the NDI was provided by the institutions and independently by the NDI ethics committee. Statistical analysis Data were tabulated as means and ranges or percentages. Student t tests and proportions tests (“stats” prop.test function) were used to compare the two datasets.9 Each preoperative variable was plotted against 1-year survival rates for the overall dataset and the two datasets separately. For continuous variables (such as age) data were divided into deciles to assess model fit graphically. If the direction of the relationship for a variable was inconsistent across the two datasets it was not included in the multivariate model. The binomial outcome variable, 1-year survival (survive or not), was used to predict the survival rate for each level of the independent (preoperative) variables. Preoperative variables were truncated at the 2.5% and 97.5% percentiles to provide a reasonable portion at each end of the range and hence more statistically robust estimates. Truncation ranges were as follows: age 60e88 years; maximum aneurysm diameter 43e85 mm; creatinine 60e208 mmol/L; infrarenal neck diameter 18e31 mm; infrarenal neck length 10e57 mm; aortic neck angle <90 ; white cell count 4.5e 13.1 109/L; aneurysm angle <80 . For example, age <60 were treated as 60 years and ages >88 were treated as 88 years. Cardiac assessment groups 1 and 2 had similar 1-year survival rates, as did respiratory assessment groups 1 and 2 and calcification groups 1 and 2. Each of these indices were merged for the purposes of this study. These adjustments generally improved R2 and/or receivereoperator curves (ROC). For multivariate models only preoperative variables with >1,000 total patient records (61%) were included. For rigour, Cox proportional hazards survival models and binomial logistic regression were performed using Frank Harrell’s “lrm” and “cph” functions from the “rms” library. For Cox proportional hazards models survival was truncated at 1 year. The proportional hazards assumption was tested using “cox.zph” function from the “survival” library.10 ROC, odds ratios (OR) and hazards ratios (HR) were reported. Missing data were omitted from analysis rather than being imputed. Owing to missing preoperative variables, not all regressions included the same number of patients. Backwards stepwise regression using the “fastbw” function from rms library and the Akaike’s Information Criterion were used to select which variables should be included in multivariate 1-year survival model.9,11 The le Cessieevan HouwelingeneCopaseHosmer unweighted sum of squares test statistic was used to assess goodness of fit in logistic regression models.12,13 The test was applied using the “residuals.lrm” function.11 Final models were internally validated using 1,000 bootstrap samples with replacement.14 The bias corrected Somers’
Please cite this article in press as: Fitridge RA, et al., Predictors of 1-Year Survival After Endovascular Aneurysm Repair, European Journal of Vascular and Endovascular Surgery (2016), http://dx.doi.org/10.1016/j.ejvs.2015.12.019
Predictors of 1-Year Survival After EVAR
Dxy rank correlation, bias corrected R2 index, and the maximum absolute difference in predicted and calibrated probabilities (Emax) were reported using the “rms validate.lrm” function.11 All statistical analyses were performed using R version 3.0.0.9 RESULTS Patient characteristics In total, 1,647 patients were included in the analysis (AUS01, n ¼ 959; AUS-13, n ¼ 688). One-year survival was 93.7% (1,544/1,647). There was a significant difference in 1-year survival between the two groups (AUS-01, 92.9%; AUS-13, 95.1% [p ¼ .03]), suggesting survival improved over time. Overall 30-day survival was 1,620/1,647 (98.4%); survival in AUS-01 was 98.1% (941/959) and in AUS-13 it was 98.7% (679/688). Table 1 shows the means and ranges for preoperative variables in each dataset, for the combined data and between-group p-values. There were statistically significant differences between the following AUS-01 and AUS-13 variables: ASA status, creatinine, aortic neck angle, aneurysm angle, infrarenal neck length, calcification ranking, and cardiac fitness. However, these differences were not deemed clinically relevant. Univariate analysis Table 2 shows the magnitude of effect for “one at a time” analysis of the combined data from the two patient cohorts. Results are shown for both logistic regression OR and Cox proportional HR. Nine variables were significantly correlated
3
with survival. While aortic neck angle is significant (p ¼ .03), the result is based on a small sample (39 deaths from 710 patients). The inverse of HR is nearly identical to the OR, as are the associated p-values, showing concordance between the two statistical methods (Table 2). HR > 1 denotes that the risk of death increases with increases in the preoperative variable; as aneurysm size increases, so does risk of death. Low and high values given in Table 2 provide the contrasted range for the OR and HR. Multivariate analysis Variables not included in the final multivariate model owing to insufficient data (<1,000) were aortic neck angle, aneurysm angle, foot pulse, white cell count, statin use, and cancer. Infrarenal neck length and infrarenal neck diameter were excluded as the relationship with survival reversed between the AUS-01 and AUS-13. When added into multivariate analysis, age, tortuosity, cardiac assessment, and sex did not contribute significantly at the p ¼ .05 level. Table 3 shows the final multivariate model with five preoperative variables: ASA status, aneurysm diameter, creatinine, respiratory assessment, and iliac artery calcification, all of which were statistically significant covariates according to the Cox proportional hazard criteria and Akaike information criterion. The ROC for the logistic regression was 0.716 (R2 ¼ .117). The risk of death within 1 year for each variable’s contribution, based on HR, shows an increase of two ASA units results in 3.59 times higher hazard of death; an increase of aneurysm diameter by 11 mm increases hazard of
Table 1. Comparison of preoperative variables between datasets AUS-01 and AUS-13 and the combined dataset. Variable n 1,576
Combined datasets Range or % Mean n 30e120 57.8 929
Aneurysm diameter (mm) Age (y) 1,647 52e95 75.1 959 ASA II 476/1,576 30.2% 322/944 ASA III 998/1,576 63.3% 559/944 ASA IV 102/1,576 6.50% 63/944 Female sex 205/1,647 12.4% 133/959 Creatinine 1,534 40e800 107.3 906 Aortic neck angle 710 0e150 26.9 223 Aneurysm angle 530 0e112 28.1 54 Infrarenal neck diameter 1,478 12e50 23.6 875 Infrarenal neck length 1,375 3e120 26.7 868 Calcification 1 or 2 939/1,362 68.9% 655/867 Tortuosity 1 or 2 797/1,379 57.8% 535/897 Cardiac 842/1,482 56.8% 446/959 assessment 1 or 2 Respiratory 1,097/1,450 75.7% 698/959 assessment 1 or 2 Foot pulse left 394/450 87.5% 0 Foot pulse right 396/447 88.5% 0 Statin use 211/676 31.2% 0 White cell count 615 3.2e25.6 7.7 0 Cancer (nonskin) 110/958 11.5% 110/958 Note. ASA ¼ American Society of Anesthesiologists; NS ¼ not significant.
AUS-01 Range Mean n 30e110 57.5 647
AUS-13 Range 35e120
p between Mean datasets 58.3 NS
52e95 24.3% 69.5% 6.20% 10.5% 40e409 0e90 0e112 14e36 4e99 57.4% 54.4% 75.5%
75.0
52e94 75.1 34.1% 59.2% 6.70% 13.8% 41e800 115 10e150 46.4 5e90 45.6 12e50 23.6 3e120 25.7 75.5% 59.6% 46.5%
688 154/632 439/632 39/632 72/688 628 487 476 603 507 284/495 262/482 396/523
72.8%
399/491 81.3%
11.5%
394/450 396/447 211/676 615 0
87.5% 88.5% 31.2% 3.2e25.6
96.3 18.0 26.2 23.7 28.6
NS .002 <.001 NS .047 <.001 <.001 <.001 NS <.001 < .001 NS <.001 NS
7.7
Please cite this article in press as: Fitridge RA, et al., Predictors of 1-Year Survival After Endovascular Aneurysm Repair, European Journal of Vascular and Endovascular Surgery (2016), http://dx.doi.org/10.1016/j.ejvs.2015.12.019
4
R.A. Fitridge et al.
Table 2. Univariate analysis: logistic regression odds ratios (OR) and Cox proportional hazard ratios (HR). Variable p OR 1/OR Cox HR Cox p Deaths Maximum aneurysm diameter <.0001 0.59 1.69 1.67 <.0001 100 ASA <.0001 0.15 6.67 6.16 <.0001 102 Creatinine <.0001 0.59 1.69 1.66 <.0001 91 Cardiac assessment .001 0.58 1.72 1.70 .001 91 Age .004 0.65 1.54 1.52 .003 103 Respiratory assessment .004 0.49 2.04 2.01 .004 89 Cancer (nonskin) .004 0.63 1.59 1.55 .005 69 Calcification grade .015 0.45 2.22 2.13 .016 89 Aortic neck angle .031 0.65 1.54 1.54 .028 39 Infrarenal neck length .068 1.26 0.79 0.80 .066 86 Infrarenal neck diameter .141 0.79 1.27 1.26 .138 91 Tortuosity grade .242 0.87 1.15 1.15 .232 89 Foot pulse palpable (left foot) .250 1.67 0.60 0.60 .250 20 Foot pulse palpable (right foot) .307 1.59 0.63 0.63 .307 20 Aneurysm angle .480 0.81 1.23 1.23 .485 27 WCC .646 0.55 1.82 1.82 .641 24 Statin use .697 1.17 0.85 0.85 .683 32 Sex .799 1.08 0.93 0.93 .796 103 Note. Bold values denote statistically significant results (p<.05) ASA ¼ American Society of WCC ¼ white cell count. M ¼ male; F ¼ female. a Low and high values represent the 25th and 75th percentiles for the continuous variables.
death 1.54 times; an increase in 39 mmol/L creatinine results in 1.48 times higher hazard of death; an increase of one respiratory assessment unit increases hazard of death by 1.68; and an increase in severity of iliac artery calcification by two units increases hazard of death by 2.11 (see Table 3). Table 4 shows 1-year survival as predicted by ASA status, aneurysm diameter, respiratory assessment, iliac artery calcification and creatinine. The best-case scenario for fitter patients with smaller aneurysms (50 mm) shows > 98% of patients could expect to survive 1 year. This contrasts with 68% survival of less fit patients with larger aneurysms (65 mm). Global goodness-of-fit was carried out to assess how well the final multivariate model fits the observations; a “good” fit is indicated by large p-values (>.05). The test pvalue was 0.254 for the final 1-year survival model, suggesting there was a “good” fit. Internal validation from 1,000 bootstrap samples showed a relatively high bias corrected Somers’ Dxy rank correlation (.421). The corrected R2 index was .109 and the calibration error for the multivariate 1-year survival model was reasonable (corrected Emax ¼ .036).
n Low valuea 1,576 51 1,576 2 1,534 81 1,482 2 1,647 70.3 1,450 2 958 N 1,362 2 710 10 1,375 20 1,478 21 1,379 2 450 N 447 N 530 10.0 615 3.2 676 Y 1,647 M Anesthesiologists; Y ¼
High valuea 62 4 120 3 80.0 4 Y 4 40 32 26 3 Y Y 43.8 13.1 N F yes; N ¼ no,
Survival analysis Fig. 1 (AeE) shows 1-year KaplaneMeier survival curves for the five variables, with the highest HR included in the multivariate model: ASA status, aneurysm diameter, creatinine, respiratory assessment, and iliac artery calcification. These illustrate the differences in survival for variable subgroups. DISCUSSION This study reports on the development of a predictive model derived from two datasets that were collected with a gap of 8 years between the datasets. On univariate analysis nine preoperative variables were significantly associated with 1-year survival. The best multivariate model included five variables: ASA status, aneurysm diameter, creatinine, respiratory fitness, and iliac artery calcification. While age and cardiac fitness were significant independent predictors, they were not significant on multivariate analysis and did not “improve” the model, suggesting that they are correlated with other predictor variables. A study by Mani et al. reported that older patients fit for surgery had better relative survival than an age-matched general population of
Table 3. Final multivariate model for 1-year survival. Variable
Logistic regression Cox proportional hazard Missing data p OR 1/HR p HR ASA .006 .28 .28 .0028 3.59 71 Aneurysm diameter <.001 .64 .65 <.001 1.54 71 Creatinine .001 .67 .68 <.001 1.48 113 Respiratory assessment .048 .59 .60 .034 1.68 197 Calcification grade .026 .46 .47 .022 2.11 285 Note. OR ¼ odds ratio; HR ¼ hazard ration; ASA ¼ ASA ¼ American Society of Anesthesiologists. a Unit change shown for each variable were contrasted in the Odds Ratios and Hazard Ratios.
Unit changea From To 2 4 51 62 81 120 2 4 2 4
Please cite this article in press as: Fitridge RA, et al., Predictors of 1-Year Survival After Endovascular Aneurysm Repair, European Journal of Vascular and Endovascular Surgery (2016), http://dx.doi.org/10.1016/j.ejvs.2015.12.019
Predictors of 1-Year Survival After EVAR Table 4. Prediction of 1-year survival for patients undergoing endovascular aneurysm repair. ASA Max. AAA Iliac Creatinine status diameter calcification 80 mmol/L 120 mmol/L (mm) Respiratory assessment 2 4 2 4 II 50 2 98.6 97.7 98.0 96.7 4 97.1 95.3 95.8 93.1 65 2 97.5 95.9 96.4 94.1 4 94.9 91.7 92.6 88.2 III 50 2 97.5 95.8 96.3 93.9 4 94.7 91.5 92.4 87.9 65 2 95.5 92.7 93.4 89.5 4 90.8 85.6 87.0 80.0 IV 50 2 95.3 92.4 93.2 89.2 4 90.6 85.2 86.6 79.4 65 2 91.9 87.1 88.4 81.9 4 84.1 76.0 78.1 68.0 Note. Respiratory assessments 2 (mild disease) and 4 (severe disease) are defined in Table S2 (see Supplementary Material). ASA ¼ American Society of Anesthesiologists; AAA ¼ abdominal aortic aneurysm.
younger patients undergoing EVAR.15 This suggests that older EVAR patients are selected for the procedure based on their fitness and therefore age alone would not be expected to predict 1-year survival. The reduction in the ASA status HR from 6.16 (univariate analysis) to 3.6 (multivariate model) also shows other factors simultaneously account for this hazard. The decision to pool the two datasets was based on the results of a large comparison study in which EVAR patients were compared across time and geographical locations.16 For the two Australian data collections, there were small differences for age, ASA status, aneurysm size, infrarenal neck length and diameter, and tortuosity. Larger differences were seen between the older Australian data and a dataset from the UK, yet in previous work it was shown that our endovascular aneurysm repair risk assessment (ERA) model was good at predicting outcomes for these UK data, despite the differences between datasets.7 Differences between the two Australian datasets may work in the model’s favour. Basing a predictive model only on one dataset may result in generalisability problems so the current model should therefore work on disparate datasets. Combining the two datasets also ensured that a reasonable number of deaths could be included in the final analysis. Table 4 clearly shows that certain patients have a much higher risk of dying within 1 year than their healthier counterparts. In addition, the risk of death for some elective EVAR patients is similar to, or exceeds, the risk of death due to aneurysm rupture with no treatment. This model contributes useful information for identifying patients at increased risk in the first year after EVAR surgery. Internal validation shows the model generally performed well in terms of bias-corrected predictive discrimination, particularly with respect to bias corrected Somers’ Dxy rank correlation, indicating that this is a relatively good multivariate model and compares well with a 3-year survival
5
model.6 That R2 is not higher shows the difficulties of predicting deaths within 1 year; however, calibration error was reasonable. Few other published studies show factors that can predict 1-year survival after EVAR. Beck et al. found congestive heart failure and aneurysm size 6.5 mm were independent predictors of 1-year survival after EVAR (n ¼ 639) on multivariate analysis.17 The authors developed a predictive model to identify patients with shorter life expectancies for whom “medical management would be preferred”. Lall et al. found a correlation between survival, age, and aneurysm size, but there were only three deaths and 32 patients in the EVAR arm following multivariate analysis.18 In our 2008 risk assessment model 30-day survival was mainly predicted by age and aneurysm size, whereas aneurysm size, age, ASA status, and creatinine all contributed significantly to 3 and 5-year survival.6e8 Respiratory fitness and calcification were not analysed in 2008, so it is interesting to find that they do contribute to the 1-year survival model. Variables that did not predict 1-year survival included statin use, white cell count, presence or absence of foot pulse, aneurysm angle, tortuosity, and sex. Sex was known for all patients, adding to the growing body of literature that shows no difference in survival for women after EVAR.19e21 Tortuosity data were also known for most patients but showed no survival effect. This fits with recent work that found no link between subjective assessments of tortuosity of the entire iliac artery and postoperative outcomes, whereas automated measurement of maximal tortuosity on a smaller 10-mm scale was a reliable predictor of early postoperative complications.22 Statin use and white cell count data were available for 676 and 615 patients, respectively. While limited in number, it is noteworthy that they do not appear to influence 1-year survival. The link between aneurysm size and survival is consistent but perplexing, possibly suggesting in some way it is a marker of systemic health. Conversely, the influence of ASA and respiratory status makes clinical sense, as these are indicators of patient fitness. The problem with ASA status is not so much that it is subjective but rather that most EVAR patients are designated ASA III, suggesting a method of subcategorising ASA III patients is required. It was recently reported that poor exercise capacity was correlated with all cause mortality in the ASA III patients in the AUS-13 cohort.23 Increasing serum creatinine is a measure of poor renal function and a good predictor of mortality. Iliac artery calcification also appears to contribute to 1-year survival, but whether this is due to operative complications arising from highly calcified vessels or as an overall measure of a patient’s fitness in the long-term requires further investigation. In agreement with this result, Allison et al. also reported that iliac artery calcification was associated with a significantly increased hazard for total mortality.24 This study evaluated all cause mortality rather than cardiovascular or aneurysm-related mortality for a number of reasons. Firstly, it is difficult to determine precisely the immediate cause of death in older people when they have
Please cite this article in press as: Fitridge RA, et al., Predictors of 1-Year Survival After Endovascular Aneurysm Repair, European Journal of Vascular and Endovascular Surgery (2016), http://dx.doi.org/10.1016/j.ejvs.2015.12.019
6
R.A. Fitridge et al.
Figure 1. KaplaneMeier survival plots for (A) American Society of Anesthesiologists (ASA) grade, (B) aneurysm size, (C) creatinine, (D) respiratory grade, and (E) iliac artery calcification. Numbers at risk for each time point are provided in the associated tables.
multiple and interacting comorbidities, particularly as very few autopsies are performed on older patients in Australia. Secondly, while fact of death is accurate and up to date in the NDI, there were a number of patients for whom an attributed cause of death was not available. Published studies have also found errors in cause and/or manner of death certification to occur in approximately 33e41% of cases.25 There were also only three deaths resulting from AAA ruptures in the datasets.
It is the authors’ opinion that 1-year survival is a better and more statistically robust means of assessing risk than early death (too few events), having a high degree of relevancy to the patient. It should inform the shared decision-making process between patient and surgeon, enabling the surgeon to discuss the potential impact of EVAR surgery with patients who are at higher risk of dying within 1 year.
Please cite this article in press as: Fitridge RA, et al., Predictors of 1-Year Survival After Endovascular Aneurysm Repair, European Journal of Vascular and Endovascular Surgery (2016), http://dx.doi.org/10.1016/j.ejvs.2015.12.019
Predictors of 1-Year Survival After EVAR
ACKNOWLEDGEMENTS The authors thank the Australian Institute of Health and Welfare National Death Index, and all surgeons who contributed their time and data. APPENDIX A. SUPPLEMENTARY DATA Supplementary data related to this article can be found at http://dx.doi.org/10.1016/j.ejvs.2015.12.019. CONFLICT OF INTEREST None. FUNDING This study was funded by a project grant (number 565335) from the National Health and Medical Research Council of Australia. REFERENCES 1 RESCAN collaborators, Brown MJ, Sweeting MJ, Brown L, Powell JT, Thompson SG. Surveillance intervals for small abdominal aortic aneurysms: a meta-analysis. JAMA 2013;309: 806e13. 2 Moll FL, Powell JT, Fraedrich G, Verzini F, Haulon S, Maltham M, et al. Management of abdominal aortic aneurysms clinical practice guidelines of the European Society for vascular Surgery. Eur J Vasc Endovasc Surg 2011;41:S1e58. 3 Chambers D, Epstein D, Walker S, Fayter D, Paton F, Wright K, et al. Endovascular stents for abdominal aortic aneurysms: a systematic review and economic model. Health Technol Assess 2009;13(48). 4 Paravastu SCV, Jayarajasingam R, Cottam R, Palfreyman SJ, Michaels JA, Thomas SM. Endovascular repair of abdominal aortic aneurysm. Cochrane Database Syst Rev 2014;1: CD004178. 5 Brown LC, Powell JT, Thompson SG, Epstein DM, Sculpher MJ, Greenhalgh RM. The UK EndoVascular Aneurysm Repair (EVAR) trials: randomised trials of EVAR versus standard therapy. Health Technol Assess 2012;16:655e65. 6 Barnes M, Boult M, Maddern G, Fitridge R. A model to predict outcomes for endovascular aneurysm repair using preoperative variables. Eur J Vasc Endovasc Surg 2008;35:571e9. 7 Barnes M, Boult M, Thompson MM, Holt PJ, Fitridge RA. Personalised predictions of endovascular aneurysm repair success rates: validating the ERA model with UK Vascular Institute data. Eur J Vasc Endovasc Surg 2010;40:436e41. 8 Boult M, Maddern G, Barnes M, Fitridge R. Factors affecting survival after endovascular aneurysm repair: results from a population based audit. Eur J Vasc Endovasc Surg 2007;34: 156e62. 9 Core Team R. R: A Language and Environment for Statistical Computing. Vienna: R Foundation for Statistical Computing; 2013.
7 10 Therneau TM, Grambsch PM. Modeling Survival Data: Extending the Cox Model. New York: Springer; 2000. 11 Harrell Jr FE. Rms: Regression Modeling Strategies. R Package Version 4.2-1. 2014. 12 Harrell Jr FE. Regression Modeling Strategies with Applications to Linear Models, Logistic Regression, and Survival Analysis. New York: Springer; 2001. 13 Hosmer DW, Hosmer T, Le Cessie S, Lemeshow S. A comparison of goodness-of-fit tests for the logistic regression model. Stat Med 1997;16:965e80. 14 Steyerberg EW, Harrell Jr FE, Borsboom GJ, Eijkemans MJ, Vergouwe Y, Habbema JD. Internal validation of predictive models: efficiency of some procedures for logistic regression analysis. J Clin Epidemiol 2001;54:774e81. 15 Mani K, Bjorck M, Lundkvist J, Wanhainen A. Improved longterm survival after abdominal aortic aneurysm repair. Circulation 2009;120:201e11. 16 Fitridge RA, Boult M, Mackillop C, De Loryn T, Barnes M, Cowled P, et al. International trends in patient selection for elective endovascular aneurysm repair: sicker patients with safer anatomy leading to improved 1-year survival. Ann Vasc Surg 2015;29:197e205. 17 Beck AW, Goodney PP, Nolan BW, Likosky DS, EldrupJorgensen J, Cronenwett JL. Predicting 1-year mortality after elective abdominal aortic aneurysm repair. J Vasc Surg 2009;49:838e43. 18 Lall P, Gloviczki P, Agarwal G, Duncan AA, Kalra M, Hoskin T, et al. Comparison of EVAR and open repair in patients with small abdominal aortic aneurysms: can we predict results of the PIVOTAL trial? J Vasc Surg 2009;49:52e9. 19 Biebl M, Hakaim AG, Hugl B, Oldenburg WA, Paz-Fumagalli R, McKinney JM, et al. Endovascular aortic aneurysm repair with the Zenith AAA endovascular graft: does gender affect procedural success, postoperative morbidity, or early survival? Am Surg 2005;71:1001e8. 20 Dubois L, Novick TV, Harris JR, Derose G, Forbes TL. Outcomes after endovascular abdominal aortic aneurysm repair are equivalent between genders despite anatomic differences in women. J Vasc Surg 2013;57:382e9. 21 Sampaio SM, Panneton JM, Mozes GI, Andrews JC, Noel AA, Karla M, et al. Endovascular abdominal aortic aneurysm repair: does gender matter? Ann Vasc Surg 2004;18:653e60. 22 Dowson N, Boult M, Cowled P, De Loryn T, Fitridge R. Development of an automated measure of iliac artery tortuosity that successfully predicts early graft-related complications associated with endovascular aneurysm repair. Eur J Vasc Endovasc Surg 2014;48:153e60. 23 Boult M, Howell S, Cowled P, De Loryn T, Fitridge R. Self-reported fitness of American Society of Anesthesiologists class 3 patients undergoing endovascular aneurysm repair predicts patient survival. J Vasc Surg 2015;62:299e303. 24 Allison MA, Hsi S, Wassel CL, Morgan C, Ix JH, Wright CM, et al. Calcified atherosclerosis in different vascular beds and the risk of mortality. Arterioscler Thromb Vasc Biol 2012;32:140e6. 25 Brooks EG, Reed KD. Principles and pitfalls: a guide to death certification. Clin Med Res 2015;13:74e82.
Please cite this article in press as: Fitridge RA, et al., Predictors of 1-Year Survival After Endovascular Aneurysm Repair, European Journal of Vascular and Endovascular Surgery (2016), http://dx.doi.org/10.1016/j.ejvs.2015.12.019
-,
1e7
Predictors of 1-Year Survival After Endovascular Aneurysm Repair R.A. Fitridge a b
a,*
, M. Boult a, T. de Loryn a, P. Cowled a, M. Barnes
b
Discipline of Surgery, The University of Adelaide, Royal Adelaide Hospital and The Queen Elizabeth Hospital, Woodville, South Australia 5011, Australia CSIRO, Adelaide, South Australia 5000, Australia
WHAT THIS PAPER ADDS This article provides a comprehensive analysis of factors predicting 1-year all-cause survival after endovascular aneurysm repair (EVAR), something that has not been reported previously. We consider that 1 year is a reasonable and minimum amount of time that a patient undergoing elective surgery should expect to survive and therefore obtain benefit from the procedure. This study helps to identify which patients have a significantly elevated risk of dying within that first year after surgery. Surgeons could use this information to ensure that patients have understood and considered the risks and benefits of their elective EVAR prior to surgery.
Objective/background: The purpose of this study was to determine the preoperative variables that best predict 1-year survival following elective endovascular aneurysm repair (EVAR), a period of time that would suggest the patient had benefited from the procedure. Most EVAR survival studies focus on early and late survival; scant information is available for 1-year survival. Methods: Data from two Australian audits of EVAR (1999e2001 and 2009e13) were combined (n ¼ 1,647). Preoperative variables included routine demographic data, clinical health assessments, computed tomographyderived anatomical data, and all-cause mortality. Univariate and multivariate logistic regressions determined which variables best predicted 1-year survival. Results: One-year survival after EVAR was 93.7% (1,544/1,647) and 30-day survival was 98.4% (1,620/1,647). Univariate analyses found that nine preoperative variables were significantly associated with 1-year survival. Five variables were included in the final multivariate model: American Society of Anesthesiologists physical status, aneurysm diameter, creatinine, respiratory assessment, and severity of iliac artery calcification (receivere operator curve 0.717, R2 ¼ .117). Predicted 1-year survival ranged from 98.6% to 68.0%, based on differences in aneurysm size and patient comorbidities. Conclusion: Personalised 1-year survival risk enables surgeons and patients to consider seriously the risks and benefits of EVAR prior to surgery. Ó 2015 European Society for Vascular Surgery. Published by Elsevier Ltd. All rights reserved. Article history: Received 26 March 2015, Accepted 12 December 2015, Available online XXX Keywords: Abdominal aortic aneurysm, Mortality, Endovascular procedures
INTRODUCTION Survival after repair of an abdominal aortic aneurysm (AAA) is unarguably the most important end point for patients considering elective surgery and one they will need to balance against their likelihood of surviving without an operation. Estimates for risk of rupture per year based on the size of aneurysm are imprecise but are quoted as follows: <4 cm, <0.2%; 4e4.9 cm, 0.3e0.6% in men and 1.0e 3.0% in women; 5e5.9 cm, 3.0e15.0%; 6.0e6.9 cm, 10.0e 20.0%; 7.0e7.9 cm, 20.0e40.0%; 8 cm, 30.0e50.0%. * Corresponding author. E-mail address: robert.fi[email protected] (R.A. Fitridge). 1078-5884/Ó 2015 European Society for Vascular Surgery. Published by Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.ejvs.2015.12.019
However, risk can vary depending on pre-existing conditions and rates of aneurysm expansion.1,2 Undergoing elective endovascular aneurysm repair (EVAR) carries a risk. Two systematic reviews comparing endovascular and open repair found 30-day mortality to be lower after EVAR than after open repair (1.4% vs. 4.2%) but had no survival advantage thereafter and was associated with an increased rate of complications and reinterventions.3,4 The UK EVAR-1 and EVAR-2 trials reported 30-day mortality to be 1.8% for healthier patients (EVAR-1) and 7.3% for less fit patients (EVAR-2), showing a marked survival differential between healthier and sicker patients.5 It has previously been shown that the factors that most strongly influenced 3and 5-year survival are aneurysm diameter, American Society of Anesthesiologists (ASA) status, age, and creatinine.6,7 Three-year survival varied between 91% for younger patients with low ASA status, creatinine, and aneurysm size and 44% for older, less fit patients.8
Please cite this article in press as: Fitridge RA, et al., Predictors of 1-Year Survival After Endovascular Aneurysm Repair, European Journal of Vascular and Endovascular Surgery (2016), http://dx.doi.org/10.1016/j.ejvs.2015.12.019
2
Reliable figures for 1-year survival after EVAR have been less well documented than early and late survival, but published figures suggest ranges between 82% and 98% (details and references given in Table S1; see Supplementary Material). Hence, it is possible that a patient with a 6% risk of rupture could have an 18% risk of dying in the year after the procedure. In such a case the risk in that year outweighs the benefit, a factor that patients and clinicians may want to consider preoperatively. Highlighting the specific factors that most strongly influence 1-year survival after EVAR could help clinicians and patients decide whether, under certain circumstances, an intervention is an unnecessary burden. Not only is 1 year a reasonable period that a patient might expect to survive, but it is also more statistically reliable than early death (within 30 days of the procedure) as there are more deaths at 12 months. However, a note of caution about modelling is warranted: the ability to predict actual time of death for an individual is imprecise. The aim here is to provide a useful guide as to which aneurysm patients may have an elevated risk of dying within 1 year. METHODS Data and definitions This study was a retrospective review of two prospectively maintained datasets from two Australian audits of EVAR, which were combined for analysis. Patients underwent elective EVAR procedures between November 1999 and May 2001 (AUS-01)6 and between January 2009 and May 2013 (AUS-13). The majority (76%) of the grafts were Zenith grafts (Cook, Bloomington, IN, USA) and the aorto-bi-iliacbifurcated configuration was used in >90% of cases. Clinical data were self-reported by the surgeon or a staff member. Data from the AUS-13 dataset were also independently audited. All patients were followed for 1 year and date of death was supplied by the Australian Institute of Health and Welfare National Death Index (NDI), which captures all deaths in Australia. One-year survival was calculated from date of the primary EVAR procedure. Preoperative variables, their definitions and assumptions made when combining the data, are shown in Table S2 (see Supplementary Material). Information regarding iliac artery calcification and tortuosity was collected from preoperative computed tomography (CT) scans according to individual clinician opinion of severity (none, mild, moderate, severe). In the AUS-13 group, this was for both iliac arteries (common and external on both sides); in the AUS-01 group one global measurement was provided for left and right iliac arteries. For comparison purposes the iliac system with the most severe grading was selected. Ethics approval Ethics approval for AUS-01 was obtained from the Ethics Committee for the Royal Australasian College of Surgeons. The AUS-13 trial obtained state-wide and institutional ethics
R.A. Fitridge et al.
approval from all 25 participating institutions. Ethics approval to link datasets with the NDI was provided by the institutions and independently by the NDI ethics committee. Statistical analysis Data were tabulated as means and ranges or percentages. Student t tests and proportions tests (“stats” prop.test function) were used to compare the two datasets.9 Each preoperative variable was plotted against 1-year survival rates for the overall dataset and the two datasets separately. For continuous variables (such as age) data were divided into deciles to assess model fit graphically. If the direction of the relationship for a variable was inconsistent across the two datasets it was not included in the multivariate model. The binomial outcome variable, 1-year survival (survive or not), was used to predict the survival rate for each level of the independent (preoperative) variables. Preoperative variables were truncated at the 2.5% and 97.5% percentiles to provide a reasonable portion at each end of the range and hence more statistically robust estimates. Truncation ranges were as follows: age 60e88 years; maximum aneurysm diameter 43e85 mm; creatinine 60e208 mmol/L; infrarenal neck diameter 18e31 mm; infrarenal neck length 10e57 mm; aortic neck angle <90 ; white cell count 4.5e 13.1 109/L; aneurysm angle <80 . For example, age <60 were treated as 60 years and ages >88 were treated as 88 years. Cardiac assessment groups 1 and 2 had similar 1-year survival rates, as did respiratory assessment groups 1 and 2 and calcification groups 1 and 2. Each of these indices were merged for the purposes of this study. These adjustments generally improved R2 and/or receivereoperator curves (ROC). For multivariate models only preoperative variables with >1,000 total patient records (61%) were included. For rigour, Cox proportional hazards survival models and binomial logistic regression were performed using Frank Harrell’s “lrm” and “cph” functions from the “rms” library. For Cox proportional hazards models survival was truncated at 1 year. The proportional hazards assumption was tested using “cox.zph” function from the “survival” library.10 ROC, odds ratios (OR) and hazards ratios (HR) were reported. Missing data were omitted from analysis rather than being imputed. Owing to missing preoperative variables, not all regressions included the same number of patients. Backwards stepwise regression using the “fastbw” function from rms library and the Akaike’s Information Criterion were used to select which variables should be included in multivariate 1-year survival model.9,11 The le Cessieevan HouwelingeneCopaseHosmer unweighted sum of squares test statistic was used to assess goodness of fit in logistic regression models.12,13 The test was applied using the “residuals.lrm” function.11 Final models were internally validated using 1,000 bootstrap samples with replacement.14 The bias corrected Somers’
Please cite this article in press as: Fitridge RA, et al., Predictors of 1-Year Survival After Endovascular Aneurysm Repair, European Journal of Vascular and Endovascular Surgery (2016), http://dx.doi.org/10.1016/j.ejvs.2015.12.019
Predictors of 1-Year Survival After EVAR
Dxy rank correlation, bias corrected R2 index, and the maximum absolute difference in predicted and calibrated probabilities (Emax) were reported using the “rms validate.lrm” function.11 All statistical analyses were performed using R version 3.0.0.9 RESULTS Patient characteristics In total, 1,647 patients were included in the analysis (AUS01, n ¼ 959; AUS-13, n ¼ 688). One-year survival was 93.7% (1,544/1,647). There was a significant difference in 1-year survival between the two groups (AUS-01, 92.9%; AUS-13, 95.1% [p ¼ .03]), suggesting survival improved over time. Overall 30-day survival was 1,620/1,647 (98.4%); survival in AUS-01 was 98.1% (941/959) and in AUS-13 it was 98.7% (679/688). Table 1 shows the means and ranges for preoperative variables in each dataset, for the combined data and between-group p-values. There were statistically significant differences between the following AUS-01 and AUS-13 variables: ASA status, creatinine, aortic neck angle, aneurysm angle, infrarenal neck length, calcification ranking, and cardiac fitness. However, these differences were not deemed clinically relevant. Univariate analysis Table 2 shows the magnitude of effect for “one at a time” analysis of the combined data from the two patient cohorts. Results are shown for both logistic regression OR and Cox proportional HR. Nine variables were significantly correlated
3
with survival. While aortic neck angle is significant (p ¼ .03), the result is based on a small sample (39 deaths from 710 patients). The inverse of HR is nearly identical to the OR, as are the associated p-values, showing concordance between the two statistical methods (Table 2). HR > 1 denotes that the risk of death increases with increases in the preoperative variable; as aneurysm size increases, so does risk of death. Low and high values given in Table 2 provide the contrasted range for the OR and HR. Multivariate analysis Variables not included in the final multivariate model owing to insufficient data (<1,000) were aortic neck angle, aneurysm angle, foot pulse, white cell count, statin use, and cancer. Infrarenal neck length and infrarenal neck diameter were excluded as the relationship with survival reversed between the AUS-01 and AUS-13. When added into multivariate analysis, age, tortuosity, cardiac assessment, and sex did not contribute significantly at the p ¼ .05 level. Table 3 shows the final multivariate model with five preoperative variables: ASA status, aneurysm diameter, creatinine, respiratory assessment, and iliac artery calcification, all of which were statistically significant covariates according to the Cox proportional hazard criteria and Akaike information criterion. The ROC for the logistic regression was 0.716 (R2 ¼ .117). The risk of death within 1 year for each variable’s contribution, based on HR, shows an increase of two ASA units results in 3.59 times higher hazard of death; an increase of aneurysm diameter by 11 mm increases hazard of
Table 1. Comparison of preoperative variables between datasets AUS-01 and AUS-13 and the combined dataset. Variable n 1,576
Combined datasets Range or % Mean n 30e120 57.8 929
Aneurysm diameter (mm) Age (y) 1,647 52e95 75.1 959 ASA II 476/1,576 30.2% 322/944 ASA III 998/1,576 63.3% 559/944 ASA IV 102/1,576 6.50% 63/944 Female sex 205/1,647 12.4% 133/959 Creatinine 1,534 40e800 107.3 906 Aortic neck angle 710 0e150 26.9 223 Aneurysm angle 530 0e112 28.1 54 Infrarenal neck diameter 1,478 12e50 23.6 875 Infrarenal neck length 1,375 3e120 26.7 868 Calcification 1 or 2 939/1,362 68.9% 655/867 Tortuosity 1 or 2 797/1,379 57.8% 535/897 Cardiac 842/1,482 56.8% 446/959 assessment 1 or 2 Respiratory 1,097/1,450 75.7% 698/959 assessment 1 or 2 Foot pulse left 394/450 87.5% 0 Foot pulse right 396/447 88.5% 0 Statin use 211/676 31.2% 0 White cell count 615 3.2e25.6 7.7 0 Cancer (nonskin) 110/958 11.5% 110/958 Note. ASA ¼ American Society of Anesthesiologists; NS ¼ not significant.
AUS-01 Range Mean n 30e110 57.5 647
AUS-13 Range 35e120
p between Mean datasets 58.3 NS
52e95 24.3% 69.5% 6.20% 10.5% 40e409 0e90 0e112 14e36 4e99 57.4% 54.4% 75.5%
75.0
52e94 75.1 34.1% 59.2% 6.70% 13.8% 41e800 115 10e150 46.4 5e90 45.6 12e50 23.6 3e120 25.7 75.5% 59.6% 46.5%
688 154/632 439/632 39/632 72/688 628 487 476 603 507 284/495 262/482 396/523
72.8%
399/491 81.3%
11.5%
394/450 396/447 211/676 615 0
87.5% 88.5% 31.2% 3.2e25.6
96.3 18.0 26.2 23.7 28.6
NS .002 <.001 NS .047 <.001 <.001 <.001 NS <.001 < .001 NS <.001 NS
7.7
Please cite this article in press as: Fitridge RA, et al., Predictors of 1-Year Survival After Endovascular Aneurysm Repair, European Journal of Vascular and Endovascular Surgery (2016), http://dx.doi.org/10.1016/j.ejvs.2015.12.019
4
R.A. Fitridge et al.
Table 2. Univariate analysis: logistic regression odds ratios (OR) and Cox proportional hazard ratios (HR). Variable p OR 1/OR Cox HR Cox p Deaths Maximum aneurysm diameter <.0001 0.59 1.69 1.67 <.0001 100 ASA <.0001 0.15 6.67 6.16 <.0001 102 Creatinine <.0001 0.59 1.69 1.66 <.0001 91 Cardiac assessment .001 0.58 1.72 1.70 .001 91 Age .004 0.65 1.54 1.52 .003 103 Respiratory assessment .004 0.49 2.04 2.01 .004 89 Cancer (nonskin) .004 0.63 1.59 1.55 .005 69 Calcification grade .015 0.45 2.22 2.13 .016 89 Aortic neck angle .031 0.65 1.54 1.54 .028 39 Infrarenal neck length .068 1.26 0.79 0.80 .066 86 Infrarenal neck diameter .141 0.79 1.27 1.26 .138 91 Tortuosity grade .242 0.87 1.15 1.15 .232 89 Foot pulse palpable (left foot) .250 1.67 0.60 0.60 .250 20 Foot pulse palpable (right foot) .307 1.59 0.63 0.63 .307 20 Aneurysm angle .480 0.81 1.23 1.23 .485 27 WCC .646 0.55 1.82 1.82 .641 24 Statin use .697 1.17 0.85 0.85 .683 32 Sex .799 1.08 0.93 0.93 .796 103 Note. Bold values denote statistically significant results (p<.05) ASA ¼ American Society of WCC ¼ white cell count. M ¼ male; F ¼ female. a Low and high values represent the 25th and 75th percentiles for the continuous variables.
death 1.54 times; an increase in 39 mmol/L creatinine results in 1.48 times higher hazard of death; an increase of one respiratory assessment unit increases hazard of death by 1.68; and an increase in severity of iliac artery calcification by two units increases hazard of death by 2.11 (see Table 3). Table 4 shows 1-year survival as predicted by ASA status, aneurysm diameter, respiratory assessment, iliac artery calcification and creatinine. The best-case scenario for fitter patients with smaller aneurysms (50 mm) shows > 98% of patients could expect to survive 1 year. This contrasts with 68% survival of less fit patients with larger aneurysms (65 mm). Global goodness-of-fit was carried out to assess how well the final multivariate model fits the observations; a “good” fit is indicated by large p-values (>.05). The test pvalue was 0.254 for the final 1-year survival model, suggesting there was a “good” fit. Internal validation from 1,000 bootstrap samples showed a relatively high bias corrected Somers’ Dxy rank correlation (.421). The corrected R2 index was .109 and the calibration error for the multivariate 1-year survival model was reasonable (corrected Emax ¼ .036).
n Low valuea 1,576 51 1,576 2 1,534 81 1,482 2 1,647 70.3 1,450 2 958 N 1,362 2 710 10 1,375 20 1,478 21 1,379 2 450 N 447 N 530 10.0 615 3.2 676 Y 1,647 M Anesthesiologists; Y ¼
High valuea 62 4 120 3 80.0 4 Y 4 40 32 26 3 Y Y 43.8 13.1 N F yes; N ¼ no,
Survival analysis Fig. 1 (AeE) shows 1-year KaplaneMeier survival curves for the five variables, with the highest HR included in the multivariate model: ASA status, aneurysm diameter, creatinine, respiratory assessment, and iliac artery calcification. These illustrate the differences in survival for variable subgroups. DISCUSSION This study reports on the development of a predictive model derived from two datasets that were collected with a gap of 8 years between the datasets. On univariate analysis nine preoperative variables were significantly associated with 1-year survival. The best multivariate model included five variables: ASA status, aneurysm diameter, creatinine, respiratory fitness, and iliac artery calcification. While age and cardiac fitness were significant independent predictors, they were not significant on multivariate analysis and did not “improve” the model, suggesting that they are correlated with other predictor variables. A study by Mani et al. reported that older patients fit for surgery had better relative survival than an age-matched general population of
Table 3. Final multivariate model for 1-year survival. Variable
Logistic regression Cox proportional hazard Missing data p OR 1/HR p HR ASA .006 .28 .28 .0028 3.59 71 Aneurysm diameter <.001 .64 .65 <.001 1.54 71 Creatinine .001 .67 .68 <.001 1.48 113 Respiratory assessment .048 .59 .60 .034 1.68 197 Calcification grade .026 .46 .47 .022 2.11 285 Note. OR ¼ odds ratio; HR ¼ hazard ration; ASA ¼ ASA ¼ American Society of Anesthesiologists. a Unit change shown for each variable were contrasted in the Odds Ratios and Hazard Ratios.
Unit changea From To 2 4 51 62 81 120 2 4 2 4
Please cite this article in press as: Fitridge RA, et al., Predictors of 1-Year Survival After Endovascular Aneurysm Repair, European Journal of Vascular and Endovascular Surgery (2016), http://dx.doi.org/10.1016/j.ejvs.2015.12.019
Predictors of 1-Year Survival After EVAR Table 4. Prediction of 1-year survival for patients undergoing endovascular aneurysm repair. ASA Max. AAA Iliac Creatinine status diameter calcification 80 mmol/L 120 mmol/L (mm) Respiratory assessment 2 4 2 4 II 50 2 98.6 97.7 98.0 96.7 4 97.1 95.3 95.8 93.1 65 2 97.5 95.9 96.4 94.1 4 94.9 91.7 92.6 88.2 III 50 2 97.5 95.8 96.3 93.9 4 94.7 91.5 92.4 87.9 65 2 95.5 92.7 93.4 89.5 4 90.8 85.6 87.0 80.0 IV 50 2 95.3 92.4 93.2 89.2 4 90.6 85.2 86.6 79.4 65 2 91.9 87.1 88.4 81.9 4 84.1 76.0 78.1 68.0 Note. Respiratory assessments 2 (mild disease) and 4 (severe disease) are defined in Table S2 (see Supplementary Material). ASA ¼ American Society of Anesthesiologists; AAA ¼ abdominal aortic aneurysm.
younger patients undergoing EVAR.15 This suggests that older EVAR patients are selected for the procedure based on their fitness and therefore age alone would not be expected to predict 1-year survival. The reduction in the ASA status HR from 6.16 (univariate analysis) to 3.6 (multivariate model) also shows other factors simultaneously account for this hazard. The decision to pool the two datasets was based on the results of a large comparison study in which EVAR patients were compared across time and geographical locations.16 For the two Australian data collections, there were small differences for age, ASA status, aneurysm size, infrarenal neck length and diameter, and tortuosity. Larger differences were seen between the older Australian data and a dataset from the UK, yet in previous work it was shown that our endovascular aneurysm repair risk assessment (ERA) model was good at predicting outcomes for these UK data, despite the differences between datasets.7 Differences between the two Australian datasets may work in the model’s favour. Basing a predictive model only on one dataset may result in generalisability problems so the current model should therefore work on disparate datasets. Combining the two datasets also ensured that a reasonable number of deaths could be included in the final analysis. Table 4 clearly shows that certain patients have a much higher risk of dying within 1 year than their healthier counterparts. In addition, the risk of death for some elective EVAR patients is similar to, or exceeds, the risk of death due to aneurysm rupture with no treatment. This model contributes useful information for identifying patients at increased risk in the first year after EVAR surgery. Internal validation shows the model generally performed well in terms of bias-corrected predictive discrimination, particularly with respect to bias corrected Somers’ Dxy rank correlation, indicating that this is a relatively good multivariate model and compares well with a 3-year survival
5
model.6 That R2 is not higher shows the difficulties of predicting deaths within 1 year; however, calibration error was reasonable. Few other published studies show factors that can predict 1-year survival after EVAR. Beck et al. found congestive heart failure and aneurysm size 6.5 mm were independent predictors of 1-year survival after EVAR (n ¼ 639) on multivariate analysis.17 The authors developed a predictive model to identify patients with shorter life expectancies for whom “medical management would be preferred”. Lall et al. found a correlation between survival, age, and aneurysm size, but there were only three deaths and 32 patients in the EVAR arm following multivariate analysis.18 In our 2008 risk assessment model 30-day survival was mainly predicted by age and aneurysm size, whereas aneurysm size, age, ASA status, and creatinine all contributed significantly to 3 and 5-year survival.6e8 Respiratory fitness and calcification were not analysed in 2008, so it is interesting to find that they do contribute to the 1-year survival model. Variables that did not predict 1-year survival included statin use, white cell count, presence or absence of foot pulse, aneurysm angle, tortuosity, and sex. Sex was known for all patients, adding to the growing body of literature that shows no difference in survival for women after EVAR.19e21 Tortuosity data were also known for most patients but showed no survival effect. This fits with recent work that found no link between subjective assessments of tortuosity of the entire iliac artery and postoperative outcomes, whereas automated measurement of maximal tortuosity on a smaller 10-mm scale was a reliable predictor of early postoperative complications.22 Statin use and white cell count data were available for 676 and 615 patients, respectively. While limited in number, it is noteworthy that they do not appear to influence 1-year survival. The link between aneurysm size and survival is consistent but perplexing, possibly suggesting in some way it is a marker of systemic health. Conversely, the influence of ASA and respiratory status makes clinical sense, as these are indicators of patient fitness. The problem with ASA status is not so much that it is subjective but rather that most EVAR patients are designated ASA III, suggesting a method of subcategorising ASA III patients is required. It was recently reported that poor exercise capacity was correlated with all cause mortality in the ASA III patients in the AUS-13 cohort.23 Increasing serum creatinine is a measure of poor renal function and a good predictor of mortality. Iliac artery calcification also appears to contribute to 1-year survival, but whether this is due to operative complications arising from highly calcified vessels or as an overall measure of a patient’s fitness in the long-term requires further investigation. In agreement with this result, Allison et al. also reported that iliac artery calcification was associated with a significantly increased hazard for total mortality.24 This study evaluated all cause mortality rather than cardiovascular or aneurysm-related mortality for a number of reasons. Firstly, it is difficult to determine precisely the immediate cause of death in older people when they have
Please cite this article in press as: Fitridge RA, et al., Predictors of 1-Year Survival After Endovascular Aneurysm Repair, European Journal of Vascular and Endovascular Surgery (2016), http://dx.doi.org/10.1016/j.ejvs.2015.12.019
6
R.A. Fitridge et al.
Figure 1. KaplaneMeier survival plots for (A) American Society of Anesthesiologists (ASA) grade, (B) aneurysm size, (C) creatinine, (D) respiratory grade, and (E) iliac artery calcification. Numbers at risk for each time point are provided in the associated tables.
multiple and interacting comorbidities, particularly as very few autopsies are performed on older patients in Australia. Secondly, while fact of death is accurate and up to date in the NDI, there were a number of patients for whom an attributed cause of death was not available. Published studies have also found errors in cause and/or manner of death certification to occur in approximately 33e41% of cases.25 There were also only three deaths resulting from AAA ruptures in the datasets.
It is the authors’ opinion that 1-year survival is a better and more statistically robust means of assessing risk than early death (too few events), having a high degree of relevancy to the patient. It should inform the shared decision-making process between patient and surgeon, enabling the surgeon to discuss the potential impact of EVAR surgery with patients who are at higher risk of dying within 1 year.
Please cite this article in press as: Fitridge RA, et al., Predictors of 1-Year Survival After Endovascular Aneurysm Repair, European Journal of Vascular and Endovascular Surgery (2016), http://dx.doi.org/10.1016/j.ejvs.2015.12.019
Predictors of 1-Year Survival After EVAR
ACKNOWLEDGEMENTS The authors thank the Australian Institute of Health and Welfare National Death Index, and all surgeons who contributed their time and data. APPENDIX A. SUPPLEMENTARY DATA Supplementary data related to this article can be found at http://dx.doi.org/10.1016/j.ejvs.2015.12.019. CONFLICT OF INTEREST None. FUNDING This study was funded by a project grant (number 565335) from the National Health and Medical Research Council of Australia. REFERENCES 1 RESCAN collaborators, Brown MJ, Sweeting MJ, Brown L, Powell JT, Thompson SG. Surveillance intervals for small abdominal aortic aneurysms: a meta-analysis. JAMA 2013;309: 806e13. 2 Moll FL, Powell JT, Fraedrich G, Verzini F, Haulon S, Maltham M, et al. Management of abdominal aortic aneurysms clinical practice guidelines of the European Society for vascular Surgery. Eur J Vasc Endovasc Surg 2011;41:S1e58. 3 Chambers D, Epstein D, Walker S, Fayter D, Paton F, Wright K, et al. Endovascular stents for abdominal aortic aneurysms: a systematic review and economic model. Health Technol Assess 2009;13(48). 4 Paravastu SCV, Jayarajasingam R, Cottam R, Palfreyman SJ, Michaels JA, Thomas SM. Endovascular repair of abdominal aortic aneurysm. Cochrane Database Syst Rev 2014;1: CD004178. 5 Brown LC, Powell JT, Thompson SG, Epstein DM, Sculpher MJ, Greenhalgh RM. The UK EndoVascular Aneurysm Repair (EVAR) trials: randomised trials of EVAR versus standard therapy. Health Technol Assess 2012;16:655e65. 6 Barnes M, Boult M, Maddern G, Fitridge R. A model to predict outcomes for endovascular aneurysm repair using preoperative variables. Eur J Vasc Endovasc Surg 2008;35:571e9. 7 Barnes M, Boult M, Thompson MM, Holt PJ, Fitridge RA. Personalised predictions of endovascular aneurysm repair success rates: validating the ERA model with UK Vascular Institute data. Eur J Vasc Endovasc Surg 2010;40:436e41. 8 Boult M, Maddern G, Barnes M, Fitridge R. Factors affecting survival after endovascular aneurysm repair: results from a population based audit. Eur J Vasc Endovasc Surg 2007;34: 156e62. 9 Core Team R. R: A Language and Environment for Statistical Computing. Vienna: R Foundation for Statistical Computing; 2013.
7 10 Therneau TM, Grambsch PM. Modeling Survival Data: Extending the Cox Model. New York: Springer; 2000. 11 Harrell Jr FE. Rms: Regression Modeling Strategies. R Package Version 4.2-1. 2014. 12 Harrell Jr FE. Regression Modeling Strategies with Applications to Linear Models, Logistic Regression, and Survival Analysis. New York: Springer; 2001. 13 Hosmer DW, Hosmer T, Le Cessie S, Lemeshow S. A comparison of goodness-of-fit tests for the logistic regression model. Stat Med 1997;16:965e80. 14 Steyerberg EW, Harrell Jr FE, Borsboom GJ, Eijkemans MJ, Vergouwe Y, Habbema JD. Internal validation of predictive models: efficiency of some procedures for logistic regression analysis. J Clin Epidemiol 2001;54:774e81. 15 Mani K, Bjorck M, Lundkvist J, Wanhainen A. Improved longterm survival after abdominal aortic aneurysm repair. Circulation 2009;120:201e11. 16 Fitridge RA, Boult M, Mackillop C, De Loryn T, Barnes M, Cowled P, et al. International trends in patient selection for elective endovascular aneurysm repair: sicker patients with safer anatomy leading to improved 1-year survival. Ann Vasc Surg 2015;29:197e205. 17 Beck AW, Goodney PP, Nolan BW, Likosky DS, EldrupJorgensen J, Cronenwett JL. Predicting 1-year mortality after elective abdominal aortic aneurysm repair. J Vasc Surg 2009;49:838e43. 18 Lall P, Gloviczki P, Agarwal G, Duncan AA, Kalra M, Hoskin T, et al. Comparison of EVAR and open repair in patients with small abdominal aortic aneurysms: can we predict results of the PIVOTAL trial? J Vasc Surg 2009;49:52e9. 19 Biebl M, Hakaim AG, Hugl B, Oldenburg WA, Paz-Fumagalli R, McKinney JM, et al. Endovascular aortic aneurysm repair with the Zenith AAA endovascular graft: does gender affect procedural success, postoperative morbidity, or early survival? Am Surg 2005;71:1001e8. 20 Dubois L, Novick TV, Harris JR, Derose G, Forbes TL. Outcomes after endovascular abdominal aortic aneurysm repair are equivalent between genders despite anatomic differences in women. J Vasc Surg 2013;57:382e9. 21 Sampaio SM, Panneton JM, Mozes GI, Andrews JC, Noel AA, Karla M, et al. Endovascular abdominal aortic aneurysm repair: does gender matter? Ann Vasc Surg 2004;18:653e60. 22 Dowson N, Boult M, Cowled P, De Loryn T, Fitridge R. Development of an automated measure of iliac artery tortuosity that successfully predicts early graft-related complications associated with endovascular aneurysm repair. Eur J Vasc Endovasc Surg 2014;48:153e60. 23 Boult M, Howell S, Cowled P, De Loryn T, Fitridge R. Self-reported fitness of American Society of Anesthesiologists class 3 patients undergoing endovascular aneurysm repair predicts patient survival. J Vasc Surg 2015;62:299e303. 24 Allison MA, Hsi S, Wassel CL, Morgan C, Ix JH, Wright CM, et al. Calcified atherosclerosis in different vascular beds and the risk of mortality. Arterioscler Thromb Vasc Biol 2012;32:140e6. 25 Brooks EG, Reed KD. Principles and pitfalls: a guide to death certification. Clin Med Res 2015;13:74e82.
Please cite this article in press as: Fitridge RA, et al., Predictors of 1-Year Survival After Endovascular Aneurysm Repair, European Journal of Vascular and Endovascular Surgery (2016), http://dx.doi.org/10.1016/j.ejvs.2015.12.019