PDF
Citation
Print
INTRODUCTION
Elective open repair surgery (ORS) and endovascular aortic repair (EVAR) are management options for asymptomatic but huge abdominal aortic aneurysms (AAAs). Asymptomatic AAA incidence is increasing due to screening for high-risk patients and worldwide aging [1]. Therefore, it is essential to determine treatment by predicting rupture risk, evaluating independent surgery risk factors, and objectively assessing life expectancy while performing surveillance.
Risk prediction tools, such as the Glasgow Aneurysm Score (GAS) in which the primary focus is ruptured aneurysms, have been introduced [2]. Subsequent advancements led to the development of more accurate risk prediction models such as the V-POSSUM (Vascular-Physiological and Operative Severity Score for the enUmeration of Mortality and Morbidity), the Medicare model, and Vascular Governance North West (VGNW) [3]. However, a notable drawback of these models is their reliance on complicated calculations that require dedicated websites to facilitate the computation process. In addition, there is a limitation of external validation. Eslami et al. [4] created the Vascular Study Group of New England (VSGNE) cohort-based risk predictive model, a more accessible, simpler, and parsimonious model for predicting the mortality of elective operable AAA patients. Because the VSGNE reflects treatment options, anatomical features, and the underlying risk factors of patients, the model is useful for predicting short-term risk in actual clinical practice. In addition, internal validation showed that the VSGNE can be used to predict in-hospital mortality more accurately than other scoring models (GAS, Medicare, VGNW). The VSGNE was introduced in the Society for Vascular Surgery guidelines in 2018, recently was recognized as an external validation for the Vascular Quality Initiative (VQI) database [5], and currently is used for risk stratification [16].
Recently, the prevalence of AAAs has notably increased on a national scale in Korea. Furthermore, a significant increase in the percentage of unruptured elective AAA repair cases has been observed. In addition, the average count of total AAA cases has been high in major metropolitan areas, reflecting the influence of accessibility to treatment [7].
It is important to identify high-risk patients and to focus on this in experienced high-volume centers in metropolitan areas. This approach can provide insight into national trends. The primary objective of this study is to apply the VSGNE to actual patients to assess its ability to predict short-term risk. In addition to identifying and comparing risk factors incorporated in the scoring schema, we aim to evaluate factors that impact short-term mortality.
METHODS
Ethics Statement
This study was approved by the Institutional Review Board of Samsung Medical Center in Seoul, Korea (No. 2023-11-076). The need for informed consent was waived due to the retrospective nature of the study.
Study design and patient selection
A retrospective analysis was conducted on patients who underwent elective repair for AAA (EVAR or ORS) from January 2005 to December 2021 at Samsung Medical Center in Seoul, Korea. To ensure precision of the scoring system, comparable with the VSGNE prediction model, only degenerative AAAs were considered. Aneurysm locations were limited to the abdominal region, including suprarenal, juxtarenal, and infrarenal areas, and the isolated iliac artery.
Patients were excluded from the analysis if they underwent nonelective AAA repair due to a ruptured aneurysm or other etiology or previous aortic surgery or if they lacked essential information regarding age, sex, or type of procedure. Furthermore, those who had undergone supraceliac or thoracoabdominal AAA repairs were excluded.
Patient data collection and study endpoint
We collected comprehensive demographic and clinical data for each patient, including age, sex, body mass index (BMI), medical comorbidities, baseline serum creatinine level, family history of degenerative aortic aneurysm, and the maximal anterior-posterior (AP) diameter of the AAA sac measured in centimeters. Chronic obstructive pulmonary disease (COPD) was defined as a moderate obstructive pattern observed on preoperative pulmonary function tests or ongoing nebulizer treatment for COPD. We also included patients with diabetes mellitus and myocardial and cerebrovascular diseases. The presence of congestive heart failure was determined based on medical records and preoperative zechocardiography.
The risk score was calculated using the criteria proposed by Eslami et al. [6]. Points were assigned as follows: treatment: EVAR (0 points), ORS (infrarenal, 2 points), ORS (suprarenal, 4 points); aneurysm size ≥65 mm (1 point); age ≥75 years (1 point); sex: male (0 points), female (1 point); comorbidities: myocardial disease (1 point), cerebrovascular disease (1 point), COPD (2 points); creatinine ≥1.5 mg/dL (1 point).
The endpoints of this study were 30-day and 1-year mortality. The VSGNE model discrimination and calibration analysis were employed, and 1-year mortality was used for risk factor analysis of short-term mortality. The follow-up period for mortality analysis extended until March 31, 2023.
Statistical analysis
Baseline analysis
We compared the clinical characteristics of patients undergoing either EVAR or ORS and clamping site (infrarenal clamping vs. suprarenal clamping). Continuous variables were compared using the Student t-test and categorical variables with the chi-square test.
Discrimination and calibration analysis for the VSGNE risk prediction model
After the score distribution was identified, we compared it with that of the VSGNE cohort sample and other external validation studies. Using 30-day and 1-year all-cause mortality as endpoints, we constructed the receiver operating characteristics curve (ROC) to obtain the area under the curve (AUC) for the VSGNE score with a 95% confidence interval. Utilizing Youden’s index, we calculated the cutoff value for the VSGNE score associated with both 30-day and 1-year mortality. Furthermore, we corrected the reference value of the score risk group (low, medium, and high risk) based on data distribution.
To assess the performance of the model within the dataset, we stratified patients into 5 risk groups based on the VSGNE scoring system. Subsequently, we created a plot comparing predicted mortality with observed mortality, where perfect prediction aligned with the X = Y line.
Risk factor analysis
Multiple logistic regression analysis was utilized to assess the association between factors included in the VSGNE risk prediction model and short-term mortality. The significance of the regression model was determined using the Hosmer-Lemeshow goodness-of-fit test and a backward elimination procedure. The estimated coefficients from the logistic regression were exponentiated to obtain the odds ratios (ORs) for the predictor variables. Covariables such as diabetes mellitus, smoking history, congestive heart failure, and BMI were included as well as factors in the scoring model. The patients were stratified into low-, medium-, and high-risk score groups. The relevance of these risk stratifications was validated using univariate analysis, which minimized confounding by other variables included in the model. A multiple Cox regression model was used to further investigate the effects of these factors on mortality over time. Survival distributions were compared using the log-rank test. Survival rates across the risk-stratified groups were compared using Kaplan-Meier analysis, examining both 30-day and 1-year mortality.
R statistical software ver. 3.6.1 (The R Foundation) was used to perform all analyses, and significance was defined as a P-value of <0.05.
RESULTS
Baseline demographics
A total of 1,227 elective patients were treated for AAA over 17 years, 616 with EVAR and 611 with ORS. The mean age was 71.21 ± 7.78 years, 383 were >75 years of age (31.2%), and males were predominant (n = 1,061, 86.5%). Approximately 20% of patients had COPD; only 11 patients (0.9%) had congestive heart failure based on medical records and preoperative echocardiography. However, the proportion of patients with any angina event and coronary angiography or percutaneous coronary intervention history was 33.3%. Among subjects, 12.8% had cerebrovascular disease and 12.6% had baseline creatinine level >1.5 mg/dL. Regarding anatomical aneurysm features, the mean maximal sac size was 5.56 ± 1.08 cm; huge aneurysm (>6.5 cm) accounted for 17.2%.
Patients in the EVAR group were older, but a significant difference was not found in the group aged >75 years. In addition, more patients in the EVAR group were male and used statins preoperatively. Furthermore, basal creatinine level >1.5 mg/dL was more frequent in the ORS group than in the EVAR group (15.2% vs. 9.9%, P = 0.005), and the ORS group had more prevalent huge aneurysms (≥6.5 cm; 26.4% vs. 8.4%, P < 0.001). Significant differences were not observed in other underlying diseases or short-term mortality between EVAR and ORS groups (Table 1).
In total, the 30-day mortality was 0.7% (9 patients), the 1-year mortality was 3.7%, and other all-cause mortality was 33.5% in the follow-up period. The ORS group had a higher rate of all-cause mortality during the follow-up period (the mortality category in Table 1). However, the rate of basal creatinine >1.5 mg/dL was high in the group with suprarenal artery clamping, and 30-day mortality was higher in the suprarenal artery clamping group (3.4%); there was no significant difference in 1-year mortality based on clamping site (Table 2).
Discrimination and calibration of the VSGNE prediction model
The mean VSGNE model score among all patients was 2.81 (standard deviation, 2.08), lower than that of the VSGNE sample population of 3.18 (Fig. 1A). Notably, the distribution pattern was similar to that of the VSGNE sample population. Most patients (n = 980, 79.9%) scored within the range of 1–4, and only 3 patients (0.2%) were classified into the prohibitory high-risk group (Fig. 1B).
The ROC curve showed that the VSGNE model was effective at predicting 30-day mortality, with an AUC of 0.822. However, the performance of the model was lower for predicting 1-year mortality, with an AUC of 0.596. The cutoff value for predicting short-term mortality was 3.5 for all patients, while the EVAR and ORS groups had different cutoffs of 2.5 and 3.5, respectively, for 1-year mortality. Using the determined cutoff value and considering score distribution, the risk stages based on the score suitable for the study group were categorized as low (1–3 points), moderate (4–6 points), and high (≥7; Fig. 2).
Based on the VSGNE score for all patients, including the EVAR and ORS groups, the mean predicted mortality was overestimated compared with the actual 30-day mortality (1.1% vs. 0.7%) (Fig. 3A). There was a notable discrepancy in the ORS group; observed mortality was 1.10% while predicted mortality was 1.9%. Conversely, observed mortality in the EVAR group showed good agreement with the predicted value. In addition, when stratified based on the American Society of Anesthesiologists physical status classification, the predicted mortality was consistently overestimated compared with the actual mortality for all groups (Fig. 3B).
When total patient scores were stratified into quintiles for model calibration, observed mortality closely corresponded to predicted 30-day mortality. However, in cases of extremely high scores (≥9 points), observed mortality was lower than predicted mortality. In contrast, the actual 1-year mortality generally exceeded the predicted value, except for the group with the highest score, where observed mortality was lower than predicted (Fig. 4).
Risk factor analysis
In the logistic regression analysis, age >75 years, cerebrovascular disease, and creatinine level ≥1.5 mg/dL were significant predictors of short-term mortality. The association between myocardial disease and mortality was not significant, with an observed OR of 0.54 (P = 0.26). Preoperative statin use was protective, reducing the risk of 1-year mortality (OR, 0.45, P = 0.017). In contrast, a BMI <25 kg/m2 was associated with an increased risk of mortality (OR, 2.7, P = 0.015). These patterns persisted in the Cox regression model (Fig. 5A).
In the EVAR group, advanced age was not a risk factor, while sac size >6.5 cm had an OR for risk of 3.7 (P = 0.02). In the ORS group, subjects aged >75 years had markedly increased mortality risk (OR, 5.78, p < 0.01); however, aneurysm size was not a significant factor in this group (Fig. 5B).
The univariable analysis confirmed the VSGNE risk score and categorization into risk stages as significant in 1-year mortality. The risk of death increased 1.4-fold with each additional VSGNE point. The high-risk group had a mortality risk 4-fold higher than the low-risk group (Fig. 5C).
Kaplan-Meier analysis revealed a significant difference in mortality between the low-risk and high-risk groups over time, both at 30 days and 1 year. However, no significant difference was found in mortality between the low- and medium-risk groups. In 1-year mortality, a significant difference was observed in the EVAR group between the medium-risk group (18.1%) and the low-risk group (3.1%). Similarly, the ORS group exhibited notably higher mortality in the high-risk group (10.1%) compared with the low-risk group (2.1%) (Fig. 6).
DISCUSSION
In this analysis, the VSGNE risk model was applied to a single-center cohort to predict short-term mortality. The examination of risk score distribution compared with actual mortality confirmed the importance of the risk factors in the scoring system.
The distribution pattern of scores within the dataset exhibited similarities to the VSGNE sample cohort group. However, compared to the VSGNE cohort assessed by Eslami et al. [6], our mean score was slightly lower, measuring 2.81 in contrast to their 3.18. Notably, our mean score was similar to that of the VQI sample, in which a mean score of 2.6 was reported [5]. Furthermore, in comparison to the model’s prediction for the total population (mean 8.33), our patient scores were underestimated.
Our study demonstrated a high accuracy in predicting short-term mortality, with an AUC of 0.822. In terms of accuracy, the model tended to overestimate observed mortality. This finding contradicts the assessment made by de Guerre et al. [8], who reported this risk prediction model tends to underestimate mortality in their analysis of large datasets such as National Inpatient Sample, VQI, and National Surgical Quality Improvement Program. When stratifying risk scores into quintiles, our analysis revealed notably lower observed mortality than predicted within the cohort exhibiting extremely high scores.
Notably, advanced age (>75 years), cardiovascular disease, and elevated creatinine level (components included in the risk score) were significant factors influencing short-term mortality. Application of the model demonstrated an approximately 4-fold higher mortality in the high-risk group compared with the low-risk group.
However, underlying myocardial disease or COPD did not show significant effects on short-term mortality in our analysis. Although these factors were expected to have a clinically significant effect, the presence of major cardiopulmonary complications did not correlate with short-term mortality in our dataset. The major complications leading to mortality may have been prevented by more frequent follow-up and close observation of underlying diseases. In addition, the beneficial effects of collaborative treatment in high-volume centers could contribute to the reduction of mortality. Myocardial disease significantly increased all-cause death. These findings indicate that long-term mortality outcomes may differ.
Although EVAR exhibited superior short-term outcomes regarding mortality compared with ORS in another study, this study found no significant difference in short-term mortality between the 2 groups in elective cases. In our study, the observed mortality (1.1%) in the ORS group was markedly lower than the predicted mortality (1.9%). These results indicated that, even in the short term, the risk of death associated with elective ORS was not higher compared with that of EVAR in our dataset. The low mortality associated with ORS observed in the data may have been influenced by hospital volume [9] and surgeon expertise [10]. Furthermore, corroborating findings from several randomized controlled trials and large cohort studies (EVAR-1, DREAM, OVER) indicate that EVAR reduced long-term survival and increased the need for secondary interventions [11]. Hwang et al. [12] emphasized the tendency to prefer EVAR due to patient vulnerability. They emphasized the importance of individual determination when considering delayed sac expansion, confirming the importance of the early survival advantage of EVAR when making decisions.
Our study identified a significant risk factor for short-term mortality in the EVAR group. In addressing issues associated with Instructions for Use compliance, it is crucial to consider sac size as well as neck length, angle, and anatomical severity as relevant factors. As these factors can affect mid-term mortality, including that due to endoleak, a comprehensive assessment of individual anatomical factors for risk stratification is warranted. Conversely, Min et al. [13] emphasized the importance of the ultrasound-based inner-to-inner maximum AP diameter, as recommended by the National Institute for Health and Care Excellence guidelines. However, they noted the potential for overtreatment due to the tendency of surgeons to measure the longest aortic diameter using CT [13], which could have led to the unnecessary inclusion of treated patients with small AAA (size < 5.5 cm), who do not meet the indications, especially because our dataset also relied on CT-based diameter measurements for screening.
In a study by McNally et al. [14], the use of statins before surgery was associated with a significant reduction in both short-term and long-term mortality in patients undergoing AAA repair. It appears that patient adherence and compliance to prescribed medications or treatment plans reduced mortality and the pharmacological preventive effects of statins. In addition, preoperative BMI was a significant risk factor, possibly indicating preoperative malnutrition, in conjunction with established factors such as hypoalbuminemia [15] and psoas muscle volume [16].
The limitations of our study are as follows. Our study is limited by its retrospective analysis within a single medical center, potentially reducing the strength of causal relationships with risk factors. Due to a focus on patients who had already received treatment, a possibility of selection bias existed because notably high-risk patients may have been excluded. In addition, the single-center data may not be generalized nationally.
Data collection relied only on medical records even for intraoperative details, including suprarenal and infrarenal artery clamping, and the low rate of congestive heart failure was determined based on medical records or preoperative echocardiography, introducing limitations in data collection accuracy.
We did not quantify the calibration between actual mortality and expected mortality according to risk score, and comparisons with other model systems were not feasible. Therefore, the superiority of the VSGNE risk prediction model compared with other models could not be verified within the study dataset. In terms of the small number of deaths in our study, 1-year mortality was evaluated instead of the more common in-hospital mortality. This difference may have affected the accuracy of our analysis. However, in elective AAA repair, traditional risk assessment methods have transitioned to the use of artificial intelligence and machine learning [1718]. This shift prompts a reevaluation of existing guidelines. In our study, we applied the modern VSGNE score, considered one of the best tools available, to assess patient risk in a real clinical setting. We aimed to determine whether the VSGNE score could effectively predict risks and identify factors affecting mortality.
In conclusion, our study evaluated the efficacy of the VSGNE risk prediction model in guiding preoperative risk assessment and treatment for patients with elective AAA. The application of this model facilitates the provision of objective risk information to patients, enhancing informed clinical decision-making.
Notably, our findings indicated lower than anticipated actual risk scores of patients who underwent treatment. Therefore, it is necessary to develop another risk prediction model suitable for the clinical characteristics of the Korean population. Further research should include multicenter data, ensuring a more comprehensive and representative analysis. Such research would be a significant step forward in the individual care of patients with AAA.
XML Download