Aneurysm is the second most frequent disease of the aorta after atherosclerosis [1]. Aortic aneurysm (AA) accounts for over 10,000 deaths in America annually [2]. AA is subdivided into thoracic aortic aneurysm (TAA) and abdominal aortic aneurysm (AAA). Despite diagnostic advances, in-time diagnosis of AAs in an early stage is difficult due to lack of classic symptoms. AAAs are found in up to 8% of men aged 65 years, yet usually remain asymptomatic until they rupture [3]. Early TAAs remain asymptomatic as well, until aortic dissection or rupture occurs, which is related to a high mortality of approximately 50% [4]. With the aging of population, the incidence of aortic diseases will increase further [5].

Nowadays, surgery is still the main treatment of AA, of which open surgery repair (OSR) and endovascular aortic repair (EVAR) are the most common two. In the recent years, the latter is used more often because of its low mortality and probability of comorbidities in perioperative phase, since EVAR avoids ischemia of organ and large trauma, compared with OSR [1]. However, OSR is still an important choice of AA patients due to its efficacy within a wide range. First of all, OSR shows long-term advantages, compared with EVAR. The Dutch Randomized Endovascular Aneurysm Management (DREAM) trial shows that although EVAR is better than OSR in aspect of 30-day mortality, comorbidity and length of stay (LOS) in hospital, there is no significant difference in 6-year follow-up after surgery, and reintervention rate of EVAR is even higher than that of OSR [6]. What’s more, EVAR1, a clinical trial reports that patients who have received EVAR have a lower survival rate than patients that have received OSR in 15-year follow-up, mainly owing to an increase of secondary aneurysmal sac rupture in EVAR group [6]. Therefore, OSR is still a treatment worthy of consideration. Moreover, OSR is still the standard when dealing with some types of AA. In cases of Marfan disease and other connective tissue diseases, when dealing with TAA in the descending part of the aorta, OSR should be preferred over thoracic endovascular aortic repair (TEVAR), since there is no evidence supporting any use of TEVAR in patients with connective tissue disease, except in emergency situations in order to get initial stabilization as a bridge to definitive surgical therapy [1]. As for AAA, in patients with complex aortic anatomy, including those with aneurysms in close proximity to or involving the renal arteries, EVAR is unsuitable, and OSR remains the standard [1]. In addition, OSR is a remedy when EVAR fails or causes comorbidities. For instance, OSR is a selection of prompt treatment to secondary endoleak following EVAR and TEVAR [7]. Since OSR has high efficacy but still with problems in safety simultaneously, the prognosis of it is still worthy of enough consideration. However, it seems that the existing predictive models for AA mortality have several shortcomings, and are not useful and practical for clinical decision making [8]. Thus, new predictors of AA prognosis should be discovered. What’s more, elevated anion gap (AG) is a risk factor for mortality of critically ill patients. Kim et al indicates that corrected AG at ICU admission may be used to predict mortality in children, regardless of underlying etiology [9]. Since there had not been a research that study the relationship between AG and mortality of AA, in our previous study, we conducted a research aiming to confirm AG as a prognosis factor for prognosis of AA patients in ICU, using the patient information obtained from Medical Information Mart for Intensive Care III (MIMIC-III), and found admission serum anion gap might serve as a strong predictor of ICU mortality for AA patients [10]. However, in that study we did not include patients with thoracoabdominal aneurysm, and we did not focus on OSR specially, which has higher perioperative mortality. Since OSR has certain risk of adverse prognosis, especially in perioperative phase, it’s necessary to build and find reliable prognosis factors of OSR. By conducting the current research, we aimed to find out that whether AG had a better prognosis effect when dealing with the subgroup receiving OSR. Moreover, we wanted to use more kinds of statistical tools to find out the association between AG and clinical outcome and estimate the clinical net benefit of the predictive models.

As we grouped patients by their admission level of AG, we found that death and LOS between those groups were all different. It proved that the subgroups divided by AG really indicated the different risk of death and hospital or ICU stay. Moreover, as we compared the risk discrimination ability of AG model and SAPSII, we found that there was no significant difference between the performance of these two models, except in hospital mortality and 1 year-mortality (AG model performed better). The AUC of AG for discrimination of survivors and non-survivor reached 0.882 (95% CI 0.818–0.946), which was higher than our previous research that studied all kinds of AA patients in ICU (AUC: 0.8513, 95% CI 0.7698–0.9328). Also, a report showed that EVAR caused smaller change of acid–base status than OSR [18]. These indicated that the prognostic prediction ability of AG might be better when dealing with patients who received OSR.

DCA and multivariate analysis both showed the good clinical usefulness and prediction ability of AG.

AG is a factor that indicates the acid–base status of patients. Traditionally, metabolic acidosis is categorized to the presence or absence of unmeasured anions, inferred by calculating AG [19]. Since it is reported that metabolic acidosis is a powerful marker of poor prognosis of critically ill patients, AG might be a prognostic factor for adverse clinical outcome in ICU patients [20].

In ruptured AAA, preoperative unmeasured AG is a prognosis factor for mortality, and elevated unmeasured AG is associated with lactic acid, ketoacids, uremia, and intoxications with nonchloride-containing acids as well as in some forms of metabolic alkalosis [21]. In our study, an evidence of this hypothesis was that different levels of AG were significantly associated with different types of AA (P value < 0.001), as is shown in Table 2.

Besides, several studies have indicated that aortic cross-clamp is associated with metabolic acidosis, which probably causes negative cardiovascular effects without proper intervention [18, 22]. Therefore, we inferred that an underlying preoperative acidosis, indicated by the elevated AG level on admission, might escalate when aortic cross-clamp is conducted, and eventually lead to a worse prognosis and a higher mortality when compared with the situation without preoperative acidosis. [18]

What’s more, preoperative acidosis may be a surrogate for occult systemic malperfusion ostensibly, since lactate generated from hypoperfusion generates acidosis, and the resolution of lactic acidosis contributes to survival significantly [23, 24]. In patients with acute type A aortic dissection, severe preoperative acidosis is associated with malperfusion or shock, and contributes to mortality [25, 26]. A previous report demonstrated that acidosis included by lactate, pyruvate or HCl will increase whole blood viscosity and hematocrit; therefore, although systemic flow is normal or supranormal, increase of viscosity and hematocrit may partly contribute to regional hypoperfusion [24]. Lactic acidosis as a surrogate for systemic hypoperfusion or malperfusion is becoming an increasingly recognized and useful independent indicator of disease severity and a predictor of survival in the intensive care and trauma settings [23].

Also, preoperative acidosis may have associations with coagulopathy. It is indicated that acidosis is a probable reason for coagulopathy, and bleeding caused by coagulopathy contributes to intraoperative death of ruptured AAA patients [27]. Another study reported that in disseminated intravascular coagulation patients in the cardiothoracic surgical ICU, there is a strong relationship between acid–base derangement and mortality, since the elevated blood lactate concentration and base deficit is a reflection of severe tissue hypoxia and plays a key role in survival in critically ill patients [28]. In our study, as is shown in Table 2, different levels of AG are significantly associated with different incidences of coagulopathy (P value < 0.001), which might be a supporting evidence of this hypothesis.

In addition to acid–base disturbance, it is reported that in geriatrics, the elevated AG level is a prognosis factor for mortality, since it is associated with hypertension, low cardiorespiratory fitness and decreased renal function [29]. Reports show that smoking, existence of other cardiovascular diseases, hypertension and dyslipidemia are risk factors of AAA, and untreated hypertension is also a risk factor for AAA rupture [1]. Besides, although smoking seems to have no significant relationship with TAA, hypertension is a risk factor for TAA, and hyperlipidemia is one of the determining factors of expansion rate of TAA, probably causing rupture [30‐32]. These factors might cause negative effect on the prognosis of AA patients, and contribute to the elevated AG with the elevated mortality. The elevated incidence of cardiovascular events of AA patients is not associated to AA directly, but associated to the common risk factors and comorbidities. In the current study, although chronic pulmonary diseases, hypertension and renal failure didn’t show differences respectively between subgroups stratified by level of AG or those stratified by survival conditions, they might have a combined effect on AG and mortality, which probably explains the association between AG and mortality.

There are several limitations in this study. As this study is an observational study conducted by using an open database, we could just give a rough possible explanation of the mechanism that caused the association between AG level and mortality of AA patients in ICU after open surgery. Moreover, we did only obtain data from one database, for which we just conducted an internal validation in this study. An external validation is essential for proving the ability of our models in future studies. In addition, the net benefit of prediction models assessed by DCA was only showed through graphs, and there was not an exact numeric value that reflects the net benefit more precisely.

The level of serum AG is an important prognosis factor for AA mortality in ICU after open surgery, which might promote the refinement of the existing prediction models and the establishment of new models.