Clinical Characteristics of High-Altitude Patients Undergoing Endovascular Repair for Abdominal Aortic Aneurysm

Abstract

Background: Abdominal aortic aneurysm (AAA) poses a substantial risk of rupture and mortality, and endovascular aortic repair (EVAR) remains the predominant treatment. Individuals living at high altitude experience chronic hypoxia and unique physiological adaptations, yet the impact of altitude on AAA characteristics has not been systematically evaluated. No prior study has applied machine learning to identify features distinguishing high-altitude AAA patients.

Methods: This dual-center study included 197 patients who underwent EVAR between 2016 and 2023. Clinical characteristics, medical history, medication use, and preoperative laboratory data were collected. Univariable logistic regression was used to identify variables associated with high-altitude residence. Least absolute shrinkage and selection operator (LASSO) regression and seven machine-learning algorithms were applied to select predictive features and develop classification models. Model performance was assessed using receiver operating characteristic (ROC) curves in training and validation sets.

Results: Among 197 patients, 28 patients were from high-altitude regions. These patients were younger, predominantly female, and had markedly fewer traditional cardiovascular risk factors. They showed lower creatinine, uric acid, and albumin levels but higher hemoglobin, red blood cell count, and elevated RDW indices. LASSO identified 17 key variables, with RBC, RDW-SD, PLR, and HRR showing positive associations with high-altitude status. Machine-learning models, especially Random Forest, demonstrated strong discriminative performance. Even with age and sex alone, Logistic Regression, LDA, and Naïve Bayes maintained high predictive accuracy.

Conclusion: High-altitude AAA patients exhibit distinct demographic, and inflammatory profiles. Machine-learning models effectively identify altitude-associated features and may support precision risk stratification for AAA patients living at high altitude.

Graphical abstract

Keywords

Abdominal aortic aneurysm / High altitude / Endovascular aortic repair (EVAR) / Machine learning

Introduction

Abdominal aortic aneurysm (AAA), defined as a pathological dilation of the abdominal aorta exceeding 3.0 cm, is predominantly observed in elderly individuals and carries a high risk of rupture and mortality[1,2]. According to the 2024 EJVES guidelines, surgical intervention is recommended for men with an aneurysm diameter > 55 mm and women with a diameter > 50 mm[3]. Treatment options include open surgical repair (OSR) and endovascular aortic repair (EVAR). With advancements in endovascular techniques and guideline endorsement, approximately 80% of patients now undergo EVAR[4,5]. Although EVAR is associated with lower perioperative morbidity compared with OSR, it remains an intermediate-to-high risk cardiovascular procedure, underscoring the need for accurate preoperative risk assessment[4].

Residents of high-altitude regions are chronically exposed to hypoxia, low atmospheric pressure, and strong ultraviolet radiation. These environmental factors may contribute to distinct physiological and metabolic adaptations[6], yet their impact on the clinical characteristics of AAA has not been systematically evaluated. Meanwhile, machine-learning approaches have shown considerable potential in medical prediction tasks due to their ability to identify complex, nonlinear patterns. Previous studies have applied machine learning to AAA growth prediction[7,8], rupture risk estimation[9], optimal timing of intervention, and postoperative outcome prediction[10]. However, no existing research has specifically compared high-altitude and non–high-altitude AAA populations or utilized machine-learning models to identify key features associated with high-altitude AAA.

Therefore, this study utilized a dual-center clinical dataset and applied multiple machine-learning algorithms to identify the distinguishing characteristics of high-altitude AAA patients. Our goal was to explore the potential physiological, metabolic, and inflammatory differences associated with high-altitude exposure and to provide a data-driven foundation for improved risk assessment and clinical management of AAA patients living at high altitude.

Methods

Data collection

The study adheres to the ethical principles outlined in the Declaration of Helsinki. The Medical Ethics Committee of Beijing Hospital and the Xizang Autonomous Region People’s Hospital have approved the study (Approval Letter No. 2025BJYYEC-KY342-01 and No. ME-TBHP-25-073).

A total of 197 patients who underwent endovascular abdominal aortic aneurysm repair (EVAR) between January 2016 and December 2023 were included. Clinical data were collected as follows:

(1) Clinical characteristics: age, sex, height, weight, body mass index (BMI), maximum AAA diameter, and the diameter-to-BMI ratio (DBR).

(2) Medical history: malignancy, major adverse cardiovascular events (MACE), smoking, hypertension, diabetes mellitus, coronary artery disease, myocardial infarction, coronary artery bypass grafting (CABG), percutaneous coronary intervention (PCI), stroke, heart failure, hyperlipidemia, chronic obstructive pulmonary disease (COPD), and chronic renal insufficiency.

(3) Medication history: aspirin, statins, antidiabetic medications, and others.

(4) American Society of Anesthesiologists (ASA) physical status classification.

(5) Preoperative laboratory tests: creatinine, uric acid, albumin, D-dimer, white blood cell count, platelet count, red blood cell count, hemoglobin, red blood cell distribution width–coefficient of variation (RDW-CV), red blood cell distribution width–standard deviation (RDW-SD), neutrophil count, lymphocyte count, monocyte count, platelet-to-lymphocyte ratio (PLR), neutrophil-to-lymphocyte ratio (NLR), systemic immune-inflammation index (SII), systemic inflammation response index (SIRI), and the hemoglobin-to-red cell distribution width ratio (HRR).

Diagnosis and treatment of AAA

The diagnostic and therapeutic procedures followed the protocols described in our previous work[10]. All patients underwent computed tomography angiography (CTA) prior to surgery to confirm the diagnosis. Indications for EVAR included: AAA diameter > 5.0 cm in men, AAA diameter > 4.5 cm in women, or rapid aneurysm growth exceeding 10 mm per year.

Inclusion and exclusion criteria

(1) Inclusion criteria: Patients undergoing EVAR for the first time. Signed informed consent with understanding of alternative treatment options. Complete perioperative and follow-up data available. Age ≥ 18 years.

(2) Exclusion criteria: Emergency admission, previous open AAA repair, or repeat EVAR. Concomitant thoracic aortic aneurysm, ruptured AAA, or endoleak after EVAR. Clinical and/or laboratory evidence of preoperative infection (elevated temperature or leukocytosis). Trauma or surgery within 2 months prior to enrollment. Autoimmune diseases such as Takayasu arteritis or Behçet’s disease.

Statistical analysis and machine-learning algorithms

All analyses were performed using R version 4.4.0 and Python 3.10.12 (Scikit-learn). Continuous variables with normal distribution were expressed as mean ± standard deviation, while categorical variables were summarized as counts and percentages. Depending on data characteristics, comparisons were performed using two-tailed t-tests, χ² tests, or Fisher’s exact tests. Univariable logistic regression was conducted to identify factors associated with high-altitude residence. A P value < 0.05 was considered statistically significant.

Feature selection was performed using the least absolute shrinkage and selection operator (LASSO) regression. Data were randomly divided into a training set and validation set at a 7:3 ratio. Receiver operating characteristic (ROC) curves and calibration plots were generated for both sets.

Multiple machine-learning models were constructed to develop prediction models, including: logistic regression, random forest, linear discriminant analysis (LDA), Naïve Bayes, k-nearest neighbors (KNN), support vector machine (SVM), decision tree. ROC curves of both the training and validation datasets were plotted to evaluate model performance.

Results

Baseline characteristics

The baseline characteristics of the study cohort are presented in Table 1. A total of 197 patients undergoing endovascular repair for AAA were included, comprising 169 individuals from non–high-altitude regions and 28 from high-altitude regions. Marked demographic and clinical differences were evident between the two populations.

High-altitude patients were substantially younger than their non–high-altitude counterparts (58.07 ± 10.9 vs. 72.36 ± 8.4 years, P < 0.001) and exhibited a significantly lower proportion of males (25.0% vs. 88.2%, P < 0.001). Conventional cardiovascular risk factors—including smoking, coronary artery disease, previous myocardial infarction, and hyperlipidemia—were common in the non–high-altitude group yet entirely absent among high-altitude residents (all P < 0.001). The prevalence of hypertension did not differ significantly between groups (72.8% vs. 64.3%, P = 0.356).

Distinct laboratory features were also observed. High-altitude residents had lower creatinine, uric acid, and albumin levels (all P ≤ 0.002), alongside higher hemoglobin and red blood cell counts (P < 0.05). Red cell distribution indices (RDW-CV and RDW-SD) were markedly elevated in the high-altitude group (P = 0.004 and P < 0.001, respectively). Among inflammatory markers, PLR was significantly increased (P = 0.002), whereas NLR, SII, and SIRI showed no meaningful intergroup differences.

Univariable logistic regression

Univariable analyses identified multiple demographic and biochemical variables associated with high-altitude residence in AAA patients (Table 2).

Male sex showed a strong inverse association (OR 0.04, 95% CI 0.02–0.12, P < 0.001), while higher ASA grade predicted high-altitude status (OR 3.48, 95% CI 1.59–7.62, P = 0.002). Age (OR 0.84 per year, P < 0.001) and height (OR 0.91 per cm, P < 0.001) were negatively associated with high-altitude residence, whereas weight and BMI were not significant predictors. Neither aneurysm diameter nor DBR demonstrated significant associations.

Several laboratory variables emerged as strong predictors. Lower creatinine, uric acid, and albumin levels were each independently associated with high-altitude residence (all P ≤ 0.004). Hematologic characteristics—including increased RBC count, hemoglobin, RDW-CV, and RDW-SD—were positively associated with high-altitude status (all P ≤ 0.001). Among inflammatory indices, lower lymphocyte count (OR 0.35, P = 0.009) and higher PLR (OR 1.01, P = 0.003) were significant predictors, whereas NLR, SII, SIRI, HRR, platelet count, and D-dimer showed no association.

LASSO feature selection

To identify the most informative predictors of high-altitude residence, LASSO logistic regression with 10-fold cross-validation was performed. The optimal penalty parameter corresponding to minimum binomial deviance was λ_min = 0.0018 (Figure 1(a)–(b)).

At λ_min, a total of 17 variables retained non-zero coefficients, including: sex, smoking, hypertension, coronary artery disease, CABG, aspirin use, statin use, age, aneurysm diameter, DBR, albumin, WBC, RBC, hemoglobin, PLR, RDW-SD, and HRR.

RBC (coef = 2.57), RDW-SD (coef = 0.45), PLR (coef = 0.008), and HRR (coef = 0.16) were positively associated with high-altitude status, highlighting the contribution of erythropoietic and inflammatory activation. In contrast, sex (coef = −3.67), smoking (coef = −2.17), age (coef = −0.097), albumin (coef = −0.20), and WBC (coef = −0.18) showed negative associations. These predictors represent the most robust set of discriminative features identified by the penalized regression approach.

Machine-learning models: feature importance

To further evaluate variable contributions, additional machine-learning classifiers were constructed.

Random forest

Feature importance ranking revealed RDW-SD as the most influential predictor, followed by age, creatinine, D-dimer, RBC count, uric acid, and hemoglobin (Figure 2(a)). Additional contributors included albumin, statin therapy, sex, RDW-CV, height, PLR, WBC, and HRR. The random forest model demonstrated excellent performance, achieving perfect classification in the training set and correctly identifying 34/36 low-altitude and 4/6 high-altitude patients in the validation set (Figure 2(b)–(C)).

AdaBoost

AdaBoost produced a distinct yet complementary feature ranking (Figure 2(d)), with HRR, RDW-SD, PREMACE, D-dimer, and age emerging as leading predictors. Smoking, sex, creatinine, atrial fibrillation, NLR, monocytes, coronary artery disease, hypertension, statin therapy, and aspirin also contributed meaningfully.

Collectively, the two machine-learning models consistently highlighted erythropoiesis-related markers (RBC, hemoglobin, RDW-SD, HRR), inflammation-related indices (PLR, NLR), and metabolic variables (creatinine, uric acid, albumin) as key determinants distinguishing high-altitude from non–high-altitude AAApatients.

Machine learning models using age and sex

To evaluate the predictive value of simple demographic variables, machine-learning classifiers were constructed using only age and sex. Performance varied substantially across the seven models tested (Figure 3(a)–(d)).

Logistic regression and LDA showed the most reliable discrimination, achieving cross-validated AUCs of approximately 0.90–0.93 with minimal overfitting. Random forest also performed well (train AUC = 0.97; test AUC = 0.89), whereas Naïve Bayes and K-nearest neighbors achieved modest-to-good test AUCs (about 0.93 and 0.86, respectively).

Support vector machine demonstrated poor generalizability (train AUC = 0.83; test AUC = 0.61). Decision Tree models displayed near-perfect training performance (AUC = 0.97) but maintained reasonably high test accuracy (AUC = 0.92), despite evidence of overfitting.

Overall, these findings indicate that even using only two demographic variables—age and sex—several machine-learning algorithms (logistic regression, LDA, Naïve Bayes, random forest) are capable of effectively discriminating high-altitude AAA patients, whereas SVM and KNN show reduced stability with minimal feature input (Figure 4).

Discussion

This study provides comparisons of AAA patients living in high-altitude vs. non–high-altitude environments and identifies a distinct clinical signature associated with altitude exposure. Using both traditional statistical analyses and machine-learning approaches, we demonstrate that high-altitude AAA patients exhibit unique inflammatory, and demographic features that differ substantially from the typical atherosclerotic-driven AAA phenotype observed at low altitude. These findings suggest that high-altitude AAA may represent a special clinical subtype influenced by chronic hypoxia and altitude-related physiological adaptation, rather than conventional cardiovascular risk factors.

In this cohort, high-altitude patients were markedly younger and included a significantly lower proportion of males compared with low-altitude patients. More importantly, none of the classical cardiovascular risk factors, smoking, coronary artery disease, previous myocardial infarction, or hyperlipidemia, were different from the high-altitude group. The high-altitude environment includes chronic hypoxia, extreme cold, low atmospheric pressure, and increased ultraviolet radiation [11]. These exposures induce profound physiological and genetic adaptations, especially among long-term residents of the Xizang Plateau, where variants in EPAS1 and EGLN1 have been implicated in hypoxia tolerance[12,13]. Such adaptations may influence vascular homeostasis, extracellular matrix remodeling, and inflammatory pathways, thereby creating a pathophysiological basis distinct from low-altitude AAA[14].

The most striking biochemical differences were the elevations in red blood cell (RBC) count, hemoglobin, RDW-CV, and RDW-SD among high-altitude patients[15]. These markers were consistently retained as top predictors in LASSO regression, random forest, and AdaBoost models, indicating that erythropoietic activation is the most robust classifier of high-altitude status. RDW elevation reflects heterogeneity in erythrocyte size and has been associated with hypoxia-driven stress erythropoiesis, chronic inflammation, and iron metabolism imbalance. Unlike low-altitude AAA, where RDW is typically interpreted as an inflammation-related prognostic marker, the pattern observed here likely represents altitude-induced hematologic adaptation. The convergence of biological plausibility and statistical importance highlights RBC-related indices as potential biomarkers of high-altitude AAA.

High-altitude patients demonstrated significantly elevated platelet-to-lymphocyte ratio (PLR), whereas NLR, SII, and SIRI values did not differ between groups. This contrasts with previous AAA literature where NLR and SII are strongly associated with aneurysm progression, inflammation, and postoperative complications[16,17]. The discrepancy suggests that high-altitude AAA may follow a different inflammatory trajectory, influenced more by hypoxia-mediated platelet activation than by neutrophil–lymphocyte dynamics.

Chronic hypoxia is known to enhance platelet reactivity, endothelial dysfunction, and HIF-dependent inflammatory signaling[18]. Thus, the selective elevation in PLR may reflect a unique hypoxia-driven inflammatory phenotype rather than traditional atherosclerotic inflammation[19]. High-altitude AAA patients showed significantly lower creatinine, uric acid, and albumin levels. These findings likely reflect the combined effects of reduced muscle mass among high-altitude residents, altered hepatic and renal metabolism under chronic hypoxia, nutritional differences in plateau populations, altitude-induced shifts in oxidative metabolism[20]. These metabolic patterns support the concept that high-altitude AAA develops under physiological constraints distinct from those in low-altitude populations. And the aortic function may differ across different altitudes[21,22].

Interestingly, machine-learning models built solely on age and sex achieved AUCs as high as 0.90–0.93 in cross-validation, indicating that altitude-associated demographic patterns are highly stereotyped and reliably distinguishable. This degree of classification accuracy with only two variables is rarely seen in cardiovascular studies and suggests a potential genetic or ethnic component in high-altitude AAA not present in low-altitude populations.

Across all models, LASSO, random forest, and AdaBoost, three major biological clusters repeatedly emerged: (1) Hypoxia-related erythropoietic activation: RBC, hemoglobin, RDW-SD, and HRR were uniformly strong predictors. (2) Hypoxia-modified inflammatory response: PLR was consistently elevated, while NLR was not discriminative. (3) Altered metabolic and nutritional markers: Lower creatinine, uric acid, and albumin contributed to group separation. The consistency of these features across multiple algorithms reinforces the notion that machine learning can uncover altitude-specific biological signatures even in modest sample sizes.

This study has several limitations. First, the number of high-altitude patients was relatively small (n = 28), although machine learning can partially mitigate small-sample effects by revealing stable patterns. Second, this was a dual-center study and may not fully represent all high-altitude populations. Third, only preoperative laboratory values were analyzed; longitudinal data could better characterize dynamic altitude-related physiological changes. Finally, mechanistic studies—including genomic, transcriptomic, and proteomic analyses—are necessary to determine whether high-altitude AAA constitutes a distinct etiologic entity.

Conclusion

High-altitude AAA patients display a unique phenotype characterized by younger age, lower male predominance, absence of traditional cardiovascular risk factors, marked erythropoietic activation, distinct inflammatory signatures, and altered metabolic profiles. These differences strongly suggest that high-altitude AAA may represent a non-atherosclerotic, hypoxia-associated subtype influenced by environmental adaptation. Machine-learning models effectively captured these patterns and may provide a framework for future investigations into altitude-specific vascular disease mechanisms.

References

[1] Wu ZY, Trenner M, Boon RA, Spin JM, Maegdefessel L. Long noncoding RNAs in key cellular processes involved in aortic aneurysms. Atherosclerosis. 2020;292:112-118.

[2] Chen ZG, Tan SP, Diao YP, Wu ZY, Miao YQ, Li YJ. The long-term outcomes of open and endovascular repair for abdominal aortic aneurysm: a meta-analysis. Asian J Surg. 2019;42(10):899-906.

[3] Zettervall SL, Schanzer A. ESVS 2024 clinical practice guidelines on the management of abdominal aorto-iliac artery aneurysms: a north american perspective. Eur J Vasc Endovasc Surg. 2024;67(2):187-189.

[4] Suckow BD, Goodney PP, Columbo JA, et al. National trends in open surgical, endovascular, and branched-fenestrated endovascular aortic aneurysm repair in medicare patients. J Vasc Surg. 2018;67(6):1690-1697.e1.

[5] Yang Y, Li C, Wu ZY, Chen ZG, Diao YP, Li YJ. Effect of antiplatelet agents on abdominal aortic aneurysm process: a systematic review and meta-analysis. Rev Cardiovasc Med. 2023;24(12):357.

[6] Jiang Y, Ping J, Lu H, et al. Associations between high-altitude adaptation and risk of cardiovascular diseases: a bidirectional mendelian randomization study. Mol Genet Genomics. 2023;298(5):1007-1021.

[7] Hirata K, Nakaura T, Nakagawa M, et al. Machine learning to predict the rapid growth of small abdominal aortic aneurysm. J Comput Assist Tomogr. 2020;44(1):37-42.

[8] Lee R, Jarchi D, Perera R, et al. Applied machine learning for the prediction of growth of abdominal aortic aneurysm in humans. EJVES Short Rep. 2018;39:24-28.

[9] Ristl R, Klopf J, Scheuba A, et al. Growth prediction model for abdominal aortic aneurysms. Br J Surg. 2022;109(2):211-219.

[10] Turton EP, Scott DJ, Delbridge M, Snowden S, Kester RC. Ruptured abdominal aortic aneurysm: a novel method of outcome prediction using neural network technology. Eur J Vasc Endovasc Surg. 2000;19(2):184-189.

[11] Teklu AA, Heckenbach I, Petr MA, Bakula D, Keijzers G, Scheibye-Knudsen M. Deep learning reveals diverging effects of altitude on aging. Geroscience. 2025;47(3):3873-3889.

[12] Lin Z, Lu Y, Yu G, et al. Genome-wide DNA methylation landscape of four Chinese populations and epigenetic variation linked to tibetan high-altitude adaptation. Sci China Life Sci. 2023;66(10):2354-2369.

[13] Wen J, Liu J, Feng Q, et al. Ancestral origins and post-admixture adaptive evolution of highland tajiks. Natl Sci Rev. 2024;11(9):nwae284.

[14] Xu C, Xu J, Zou C, et al. Chronic intermittent hypoxia regulates CaMKII-dependent MAPK signaling to promote the initiation of abdominal aortic aneurysm. Oxid Med Cell Longev. 2021;2021:2502324.

[15] Teng B, Xie C, Zhao Y, et al. Identification of MEDAG and SERPINE1 related to hypoxia in abdominal aortic aneurysm based on weighted gene coexpression network analysis. Front Physiol. 2022;13:926508.

[16] Hafezi S, Arif B, Ruhel R, et al. C-C chemokine receptor 2 is a tissue biomarker for abdominal aortic aneurysmal and occlusive disease. Ann Surg. 2025.

[17] Zhao WX, Wu ZY, Zhao N, Diao YP, Lan Y, Li YJ. Novel systemic inflammatory markers predict all-cause mortality in patients undergoing endovascular abdominal aortic aneurysm repair. Rev Cardiovasc Med. 2024;25(6):202.

[18] Luo Z, Tian M, Yang G, et al. Hypoxia signaling in human health and diseases: implications and prospects for therapeutics. Signal Transduct Target Ther. 2022;7(1):218.

[19] Dihlmann S, Kaduk C, Passek KH, Spieler A, Böckler D, Peters AS. Exploring circulating cell-free DNA as a biomarker and as an inducer of AIM2-inflammasome-mediated inflammation in patients with abdominal aortic aneurysm. Sci Rep. 2025;15(1):20196.

[20] Qi Y, Jiang H, Lun Y, et al. Protein drug targets for abdominal aortic aneurysm and proteomic associations between modifiable risk factors and abdominal aortic aneurysm. J Am Heart Assoc. 2025;14(5):e037802.

[21] Wang YX, Wu ZY, Li YJ. Beyond diameter: functional endpoints may be considered in abdominal aortic aneurysm trials evaluating metformin. Eur J Vasc Endovasc Surg. 2025;S1078-5884(25)00476-9.

[22] Yong-Jun L. Evaluating the value of biological function in the diagnosis and treatment of aortic diseases. Vascular Research. 2025;1(1):1-2.

RIGHTS & PERMISSIONS

The Author(s) 2026. This article is published by Higher Education Press at journal.hep.com.cn.