The proportion of institutions performing at least 25% open abdominal aortic aneurysm repair (OAR) surgeries fell significantly over the study period (A). However, the proportion of deaths per year following both endovascular abdominal aortic aneurysm repair (EVAR) and OAR did not significantly change over time (B).
Aside from operative approach, intrinsic patient risk was the strongest independent predictor of in-hospital death following AAA repair. Hospital volume (Leapfrog criteria ≥50 AAA cases/year) and adequate institutional experience with open AAA repair techniques (≥25% open cases) also play a significant role. Relative impact represents β coefficient of each variable from the multivariable regression model.
aP < .05.
Caitlin W. Hicks, Joseph K. Canner, Isibor Arhuidese, Tammam Obeid, James H. Black, Mahmoud B. Malas. Comprehensive Assessment of Factors Associated With In-Hospital Mortality After Elective Abdominal Aortic Aneurysm Repair. JAMA Surg. 2016;151(9):838–845. doi:10.1001/jamasurg.2016.0782
Patient- and hospital-level factors affecting outcomes after open and endovascular abdominal aortic aneurysm (AAA) repair are each well described separately, but not together.
To describe the association of patient- and hospital-level factors with in-hospital mortality after elective AAA repair.
Design, Setting, and Participants
Retrospective review of the Nationwide Inpatient Sample database (January 2007-December 2011). The review included all patients undergoing elective open AAA repair (OAR) or endovascular AAA repair (EVAR) and was conducted between December 2014 and January 2015.
Main Outcomes and Measures
Factors associated with in-hospital mortality were analyzed for OAR and EVAR using multivariable analyses, adjusting for previously defined patient- and hospital-level risk factors.
Of the 166 443 surgeries (131 908 EVARs and 34 535 OARs) that were performed at 1207 hospitals, 133 407 patients (80.2%) were men, 123 522 patients (89.6%) were white, and the mean (SD) age was 73 (0.04) years. Overall in-hospital mortality was 0.7% for EVAR and 3.8% for OAR. Mortality after EVAR was significantly higher among hospitals with high general surgery mortality (mortality quartile ≥ 50%; odds ratio [OR], 1.37; 95% CI, 1.01-1.86; P = .04) and there was no difference in mortality among hospitals meeting the Leapfrog criteria for AAA repair (OR, 0.64; 95% CI, 0.38-1.09; P = .09). Mortality after OAR was significantly lower among hospitals performing at least 25% of AAA repairs using open techniques (OR, 0.68; 95% CI, 0.52-0.88; P = .004). Neither hospital bed size nor teaching status was significantly associated with mortality after either EVAR or OAR. Overall, OAR (OR, 6.07; 95% CI, 4.92-7.49) and intrinsic patient risk (Medicare score; OR, 4.81; 95% CI, 3.45-6.72) were most likely associated with in-hospital mortality after AAA repair, although hospitals with poor general surgery performance (OR, 1.31; 95% CI, 1.06-1.63) and those with at least a 25% proportion of open cases (OR, 1.39; 95% CI, 1.10-1.75) were also significantly associated with mortality (all P < .002). Notably, the proportion of institutions performing at least 25% open cases fell from 41% in 2007 to 18% in 2011 (P < .001).
Conclusions and Relevance
Patient-level factors were associated with in-hospital mortality outcomes after elective AAA repair. Hospital case volume and practice patterns were also associated. This demonstrates the importance of adequate institutional experience with OAR techniques, which appear to be critically declining. Based on these data, appropriate patient selection and medical optimization appear to be the most important means by which we can improve outcomes following elective AAA repair, although patient referral to high-volume aortic centers of excellence should be a secondary consideration.
With the emphasis on hospital quality and outcomes, interventions focused on improving perioperative mortality following surgical procedures are increasing. In 2003, the Leapfrog Group1 published safety standards designed to improve outcomes following 5 major high-risk surgical procedures including abdominal aortic aneurysm (AAA) repair. They recommended that hospitals perform a minimum of 50 elective AAA repairs per year to meet the minimum volume standard required to provide a relative risk reduction in mortality of 50%.1 These recommendations were made based on national data demonstrating that patients undergoing AAA repair at low-volume institutions had a risk of death of 5.1%, compared with 3.1% at high-volume institutions.2
Since the Leapfrog Safety Standards were published, there have been a number of studies published that analyze the effects of both hospital-level and patient-level factors on postoperative outcomes following AAA repair.3- 12 Hospital volume has been repeatedly validated as an important predictor of death following open AAA repair (OAR),10- 13 although the effect of case volume on mortality after endovascular AAA repair (EVAR) is less clear.14,15 Patient-level risk factors are also well reported, and multiple risk stratification scores for calculating patients’ risk of death following AAA repair are available as a result.3- 6,8,16,17
Despite this myriad of data, to our knowledge, no comprehensive study investigating the effect of both patient- and hospital-level factors on mortality after AAA repair exists. Given the push for aortic centers of excellence as a means to improve patient outcomes following elective AAA repair, it is important to understand precisely what variables are most likely to have the largest effect on the perioperative course to optimize patient selection and triaging. In this study, we aimed to quantify patient- and hospital-level factors affecting in-hospital mortality after OAR and EVAR.
Question How do patient- and hospital-level factors affect in-hospital mortality after elective abdominal aortic aneurysm repair?
Findings In this review of 166 443 patients from the Nationwide Inpatient Sample database, use of an open repair approach and intrinsic patient risk factors were associated with in-hospital mortality after abdominal aortic aneurysm repair, although hospitals with poor general surgery performance and those with at least a 25% proportion of open repairs were also significant.
Meaning Appropriate patient selection and medical optimization appear to be the most important means by which we can improve outcomes following elective abdominal aortic aneurysm repair, although patient referral to high-volume aortic centers of excellence should be a secondary consideration.
Patients undergoing elective (nonruptured) EVAR or OAR were identified from the Nationwide Inpatient Sample (NIS) between January 2007 and December 2011 using International Classification of Diseases, Ninth Revision codes (EVAR: 39.71, 39.77, and 39.78 and OAR: 38.34, 38.44, and 38.64). Hospital-level factors were determined secondarily, meaning that hospitals that performed no AAA repairs were not captured. Patients who underwent conversion of EVAR to OAR (n = 214), those who underwent nonelective repairs (n = 33 079), and those with missing data pertaining to mortality status, age, sex, hospital bed size, hospital teaching status, and severity of illness (all-patient refined diagnosis-related groups) (n = 1759) were excluded. The Johns Hopkins institutional review board approved this study prior to its initiation. No consent was obtained because the data was obtained from a publically available database.
Basic demographic information, including age, sex, race/ethnicity, and comorbid conditions, was collected for all patients. Each patient was then classified into low-, medium-, and high-risk groups using the scoring method for prediction of mortality after AAA repair developed by Giles et al8 within a Medicare population. In this model, each patient’s risk of mortality following AAA repair is calculated using a logistic regression equation that considers patient characteristics (sex and age) and comorbidities (renal insufficiency/failure, congestive heart failure, vascular disease, and cerebrovascular disease):
Risk Score = 4I (Female) + 1I (70 < Age ≤75) + 6I (75 < Age ≤80) + 11I (Age >80) + 9I (End-Stage Renal Disease) + 7I (Chronic Renal Insufficiency) + 6I (Congestive Heart Failure) + 3I (Vascular Disease)
In accordance with prior reports,8,18 Medicare risk scores greater than 11 were designated as high risk, scores ranging between 3 and 11 were designated as medium risk, and scores less than 3 were designated as low risk.
Hospital-level factors, including hospital bed size, teaching status, Medicaid quartile, performance status, and AAA repair volume and practice patterns, were also collected for each patient. Hospital bed size was defined as a categorical variable (small, medium, or large) based on the institution’s number of acute care beds, geographical location, and teaching status, as described by the Agency for Healthcare Research and Quality.19 Hospital teaching status was defined as a binary variable (teaching vs nonteaching) based on data reported by the American Hospital Association Annual Survey of Hospitals.19 Safety-net hospitals were defined as those institutions that were at or above the 75th percentile for proportion of Medicaid patients treated among all hospitals in the NIS, as previously described.20 Hospital performance status was based on mortality quartiles for all general surgery cases reported within NIS, which were identified on the basis of the “major therapeutic” procedure classification within the database. Those institutions that fell within the top 2 quartiles for mortality (ie, mortality quartile ≥50%) for major therapeutic procedures were described as low-performance hospitals. Hospitals meeting Leapfrog criteria for AAA repair were defined as those hospitals performing more than 50 elective AAA repair cases per year according to the published Leapfrog Group volume standards.2 Hospital practice patterns were described by reporting the ratio of OARs to EVARs by an institution.
The primary outcome of our study was in-hospital mortality following AAA repair. Descriptive statistics were reported as weighted values with mean (standard error of the mean [SEM]) or count with percentages, as appropriate. Univariable statistics including t tests (continuous variables), Pearson χ2 tests (categorical variables), and logistic regression models were used to assess for associations between individual variables and mortality. Multivariable logistic regression was used to identify independent predictors of mortality following AAA repair.
Covariates in the multivariable model included both patient- and hospital-level factors. Common patient-level factors from prior AAA risk mortality prediction studies (including age, sex, and pertinent patient comorbidities),3- 9 were screened as possible covariates in the model. Ultimately, use of an OAR, adjusted Medicare risk score category, hypertension, and chronic obstructive pulmonary disease were selected for inclusion based on univariable analyses. Diabetes was excluded based on the observation that, after adjusting for chronic obstructive pulmonary disease and hypertension, it was not significantly associated with mortality (data not shown).
Similarly, previously defined hospital-level factors (including hospital size, volume, teaching status, performance, and catchment)1,14,15,21- 26 were screened as possible covariates in the multivariable model. Ultimately, hospital bed size, teaching status, classification as a low-performance or safety-net hospital, and AAA repair volume (Leapfrog criteria and OAR experience) were selected for inclusion as hospital-level factors based on univariable analyses (data not shown). Of note, the method used for weighting the analysis to the US population (Stata svyset command and Stata logistic command with svy: prefix) takes into account the clustering of patients within hospitals. This method is comparable with a multilevel mixed-effects model with random effects for hospital, but has the advantage of adjusting the variance calculations to account for the NIS sampling strata and to account for the sampling strata and hospitals that may not contain any patients undergoing AAA repair. All statistics were performed using Stata software, version 14.0 (StataCorp) with statistical significance defined as P ≤ .05.
Of the 166 443 operations (131 908 EVARs and 34 535 OARs) that were performed at 1207 hospitals, 133 407 patients (80.2%) were men, 123 522 patients (89.6%) were white, and the mean (SD) age was 73.0 (0.04) years. Based on the Medicare risk score,8 53 761 patients (32.3%) were classified as low risk for mortality following AAA repair, 64 899 (39.0%) were classified as medium risk, and 47 771 (28.7%) were classified as high risk. The breakdown of patient comorbidities and Medicare risk classification by surgical approach (ie, EVAR vs OAR) are summarized in Table 1.
The overall mean (SEM) number of AAA repair operations performed per hospital per year was 109 (4) (EVAR = 85.4 [2.9] and OAR = 23.7 [1.00]). Sixty percent (n = 729) of hospitals met the Leapfrog criteria of performing at least 50 cases/year,1 but nearly all surgeries (n = 153 915, 92.5%) were performed at Leapfrog-compliant hospitals. Most surgeries were performed at large hospitals (n = 119 848, 72.0%) with an academic affiliation (n = 96 076, 57.7%). Sixty-seven percent (n = 11 467) were performed at hospitals with a predominantly (>75%) endovascular AAA practice. Less than half of surgeries (n = 119 848, 41.6%) were performed at low-performance institutions, and only 18.3% (n = 30 540) were performed at safety-net hospitals. A complete summary of hospital characteristics broken down by EVAR and OAR approaches is provided in Table 1.
Overall in-hospital mortality following AAA repair was 0.9% (n = 1548), including 0.7% (n = 923) for EVAR and 3.8% (n = 625) for OAR.
Among patient-level factors, the increasing Medicare risk category was associated with a significantly increased risk of death for both EVAR (OR, 4.17; 95% CI, 2.31-7.51 for medium-risk categories and OR, 10.3; 95% CI, 2.31-7.51 for high-risk categories) and OAR (OR, 2.49; 95% CI, 1.69-3.65 for medium-risk categories and OR, 4.86; 95% CI, 3.29-7.16 for high-risk categories) (P < .001; Table 2). In contrast, hypertension was protective for both approaches (OR, 0.24; 95% CI, 0.18-0.32 for EVAR and OR, 0.30; 95% CI, 0.06-0.08 for OAR; P < .001; Table 2). Chronic obstructive pulmonary disease was associated with increased risk of death following EVAR (OR, 1.46; 95% CI, 1.09-1.97; P = .01) but not OAR (OR, 1.08; 95% CI, 0.83-1.41; P = 56).
Among hospital-level factors, mortality following EVAR was significantly higher for hospitals with high general surgery mortality (mortality quartile ≥ 50%; OR, 1.37; 95% CI, 1.01-1.86; P = .04) and somewhat lower among hospitals meeting the Leapfrog criteria for AAA repair (OR, 0.64; 95% CI, 0.38-1.09; P = .09). Outcomes for OAR were significantly better among hospitals performing at least 25% of AAA repairs using open techniques (OR, 0.68; 95% CI, 0.52-0.88; P = .004). Neither hospital bed size nor teaching status significantly affected outcomes after either EVAR or OAR (all P = .17; Table 2). There was also no significant difference in mortality for EVAR or OAR cases performed at institutions with high numbers of Medicaid patients (ie, safety-net hospitals) (both P = .10; Table 2).
Overall, OAR (OR, 6.07; 95% CI, 4.92-7.49; P < .001) and patient risk factors (OR, 2.73; 95% CI, 1.98-3.45 and OR, 4.81; 95% CI, 3.45-6.72 for medium- and high-risk Medicare risk categories, respectively; P < .001) were associated with in-hospital mortality after AAA repair (Table 3). Hospitals with poor general surgery performance were also significantly associated with an increased risk of death (OR, 1.31; 95% CI, 1.06-1.63; P = .01). Hypertension (OR, 0.33; 95% CI, 0.27-0.41; P < .001) and hospitals with at least 25% proportion of open cases (OR, 0.72; 95% CI, 0.57-0.91; P = .005) were associated with lower rates of mortality. Notably, the proportion of institutions performing at least 25% open surgeries fell from 41% in 2007 to 18% in 2011 (P < .001; Figure 1A). However, the proportion of deaths per year following both EVAR and OAR did not significantly change over the study period (P = .76; Figure 1B).
There was no difference in mortality among hospitals meeting Leapfrog criteria for AAA repairs (OR, 0.71; 95% CI, 0.50-1.02; P = .06) or among mortality among hospitals with safety-net status (OR, 1.27; 95% CI, 0.97-1.66; P = .08). Hospital bed size and teaching status was not significantly associated with outcomes following AAA repair (P = .27; Table 3). The relative effect of each variable on patient risk of death following AAA repair ordered sequentially from greatest (open approach) to least (teaching status) based on multivariable model β coefficients is shown in Figure 2.
Patient-level and hospital-level factors affecting outcomes after AAA repair are each well described separately. However, to our knowledge, there are no data quantifying the impact of both aspects on risk of death following either OAR or EVAR approaches. Using NIS data made up of more than 165 000 surgeries, we reported the effects of patient comorbidities, operative approach, and hospital volume on in-hospital mortality following elective AAA repair. We found that, aside from operative approach, intrinsic patient factors were most strongly associated with in-hospital death following operative AAA repair. Overall hospital general surgery performance and AAA experience also played a role, although the relative effects of these factors was less than one-third that of individual patient risk (Figure 2).
Multiple patient-based risk scores for predicting mortality after AAA repair have been developed.3- 6,8,16,17 One of these, the Medicare risk score, is a risk algorithm derived using Medicare data that can be used to stratify patients as low, medium, or high risk for death following AAA repair based on their age, sex, comorbidities, and proposed surgical approach (OAR or EVAR).8 Although the accuracy of the model is good (C statistic = 0.73), to our knowledge, it has never been validated in a broad (non-Medicare) population.18 The results presented here demonstrate that a Medicare risk classification of “high risk” was the single strongest predictor of death aside from operative approach in our study. This was true for both EVAR and OAR, regardless of hospital-level factors. Interestingly, the baseline characteristics of patients with AAA undergoing AAA repair in the NIS database appear to be quite similar in most domains (age, sex, race/ethnicity, and comorbidities) to those included in the Medicare risk score study.8 As such, we both provide a validation of the Medicare risk score using national data and demonstrate that patient-level risk factors play an integral role in determining patient outcomes following this high-risk operation.
Interestingly, although patient-level risk factors were highly predictive of death following AAA repair, hospital-level factors had a much lower effect. Not surprisingly, hospitals falling within the higher quartiles for mortality following general surgery had worse outcomes. However, hospital bed size and teaching status had no significant effects. Previous studies examining the effects of hospital type and volume on mortality following AAA repair have reported a protective effect of teaching hospitals,21 although this finding can be difficult to differentiate from the differences in hospital case volume that frequently occur between teaching and nonteaching institutions.27 Using data from the American College of Surgeons National Surgical Quality Improvement Program, we previously demonstrated a significant mortality benefit for both OAR and EVAR performed at academic institutions, regardless of hospital size.13 However, the data from that study were institutional-level data that could not be paired with matching patient-level data and therefore were not risk-adjusted for hospital case mix. Both high-volume hospitals and teaching hospitals are more likely to have higher case mix indices and a different practice distribution (ie, open vs endovascular approaches) than smaller community hospitals, which could explain the unadjusted mortality differences previously reported between hospital types.13- 15 As we demonstrate here, intrinsic patient risk factors have more than a 3-fold higher effect on patient mortality following AAA repair than any measurable hospital-level factor, making the relative effect of institutional practice patterns relatively minor in comparison.
Despite the strong association between intrinsic patient risk and in-hospital mortality after AAA repair, it should be noted that hospital-level factors are also important. In keeping with the Leapfrog Group standards, hospital volume did have a moderate effect on patient mortality in our study, even after adjusting for patient-level differences between institutions. Our data suggest that if a similar Medicare-risk patient planned to undergo OAR, going to a hospital with an adequate (≥25%) open experience would significantly lower his/her predicted risk of death by an OR of 0.72 (95% CI, 0.57-0.91). Of note, we observed a significant decrease in the number of institutions with a robust OAR practice over the study period. This finding likely reflects the push toward the development of aortic centers of excellence, whereby more complex cases and those patients whose anatomy or baseline health status are not amenable to an endovascular approach are referred to more experienced centers.28 Failure-to-rescue rates, which account for a large proportion of perioperative deaths following AAA repair,24 are significantly lower in high-volume hospitals for both OARs and EVARs.29 Consistent with this notion, our data suggest that cases performed at institutions that maintain a reasonable OAR practice and those that meet the Leapfrog minimum volume standards for this operation have a lower risk of death. Although this risk reduction is markedly less than the risk associated with baseline patient factors, the implementation of minimum case volume standards can be effective. Among California hospitals, in-hospital mortality dropped by 61% following implementation of the Leapfrog evidence-based standards for AAA repair.30 As more and more cases are referred to high-volume centers with robust AAA repair experience, we expect to see the dichotomy in outcomes between expert and nonexpert institutions grow, especially as methods for preoperative medical optimization continue to evolve.
The limitations of our study deserve discussion. The NIS is a retrospective administrative database and is thus susceptible to all of the limitations associated with potential coding errors and incomplete data. Our sample size of 166 443 patients should mitigate any statistically significant errors related to this bias, but the possibility remains that some patient-level data may be inaccurate. In addition, some of the OARs may have been performed in patients who were not candidates for EVAR. It would be interesting to see whether our findings change when corrected for differences in patient anatomy; we would expect that more difficult aneurysms are preferentially referred to tertiary hospitals, which may partly explain why hospital experience had a relatively lower effect on mortality than patient factors. Finally, our study outcomes are limited to in-hospital mortality, owing to the intrinsic limitations of the NIS database. As a result, our findings are restricted in-hospital death following AAA repair and should not be assumed to apply to longer-term outcomes. Additional research is critical for quantifying how patient- and hospital-level factors affect long-term outcomes after OAR and EVAR. Cost analyses would also be helpful for determining the potential economic effects of the findings that we present.
Patient- and hospital-level factors affecting outcomes after OAR and EVAR are well described separately, but to our knowledge, no comprehensive study exists describing the quantitative effects of both aspects together. Herein, we used data from more than 165 000 surgeries to characterize the effects of intrinsic patient risk and institutional experience and practice patterns on elective AAA repair. We demonstrate that, aside from operative approach (open vs endovascular), intrinsic patient risk factors were most strongly associated with risk of in-hospital death following elective AAA repair. Adequate institutional experience with OARs is the predominant institutional variable in determining patient outcomes, albeit to a much lesser degree. Notably, relative use of OAR techniques appear to be rapidly declining nationwide, which may be reflective of a shift in paradigm toward aortic centers of excellence vs a general change in practice patterns, but could negatively influence outcomes in the future. Based on these data, appropriate patient selection and medical optimization appear to be the most important means by which we can improve outcomes following elective AAA repair, although patient referral to high-volume aortic centers of excellence should be a secondary consideration. As such, we urge surgeons in the vascular community who have less experience with AAA repair techniques to use referral centers in their area, especially for the management of complex patients who may be at higher risk for poor outcomes.
Corresponding Author: Mahmoud B. Malas, MD, MHS, Department of Vascular and Endovascular Surgery, Johns Hopkins Bayview Medical Center, 4940 Eastern Ave, Bldg A/5, Ste 547, Baltimore, MD 21224 (firstname.lastname@example.org).
Accepted for Publication: March 11, 2016.
Published Online: May 18, 2016. doi:10.1001/jamasurg.2016.0782
Author Contributions: Drs Hicks and Malas had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Hicks, Arhuidese, Malas.
Acquisition, analysis, or interpretation of data: Hicks, Canner, Obeid, Black, Malas.
Drafting of the manuscript: Hicks, Arhuidese.
Critical revision of the manuscript for important intellectual content: Hicks, Canner, Obeid, Black, Malas.
Statistical analysis: Hicks, Canner, Arhuidese, Obeid.
Administrative, technical, or material support: Hicks, Arhuidese, Black.
Study supervision: Malas.
Conflict of Interest Disclosures: None reported.
Previous Presentations: The data herein were presented at the Eastern Vascular Society 29th Annual Meeting; September 26, 2015; Baltimore, Maryland.