Plasma proteomic signatures associated with abdominal aortic aneurysm presence, growth, and risk of mortality

Abstract

Abdominal aortic aneurysm is a vascular disorder, often diagnosed incidentally, that poses substantial morbidity and mortality. Current risk assessment relies heavily on imaging and conventional risk factors. This study aimed to evaluate associations between plasma proteins and abdominal aortic aneurysm presence, progression, and mortality. A nested case-control cohort of 641 men aged 65–74 years from the Viborg Vascular screening trial (VIVA) with available plasma samples underwent proteomic profiling using mass spectrometry-based proteomics; 466 were abdominal aortic aneurysm cases, 175 were age-matched controls. Associations between individual protein levels and (1) presence of abdominal aortic aneurysm, (2) annual aneurysm growth, and (3) time-to-death were evaluated using multivariable regression models adjusted for clinical risk factors with correction for multiple testing by controlling the false discovery rate. Plasma proteins significantly associated with outcomes were further evaluated for added predictive value. Twenty-seven plasma proteins, including coagulation factor XIIIa and complement factor H, were independently associated with the presence of abdominal aortic aneurysm and significantly improved prediction beyond common risk factors (ΔAUC + 0.11, P = 0.00024). An exploratory 22-protein panel combined with diabetes status, smoking, and baseline aortic diameter improved discrimination between slow- and fast-growing aneurysms compared with aortic diameter alone (ΔAUC + 0.09, P = 0.0020). Thirteen proteins were independently associated with time-to-death and improved mortality prediction beyond clinical factors (Δc-statistics + 0.13, P = 4.07E-6). Proteomic profiling identified plasma proteins that were independently associated with abdominal aortic aneurysms presence, progression, and mortality, provided significant discrimination between slow- and fast-growing aneurysms, and improved prediction of mortality beyond established risk factors. In all supporting the potential role of plasma proteins in complementing risk stratification of abdominal aortic aneurysms.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-52218-y.

Introduction

An abdominal aortic aneurysm is a potentially life-threatening disease caused by the degradation of the abdominal aortic wall leading to dilation of the aorta and risk of aortic rupture. Abdominal aortic aneurysms measuring 30 mm, or more are present in more than 3% of the Danish male population between 65 and 74 years of age^1,2. The average growth rate of an abdominal aortic aneurysm is 2.0–3.0 mm/year, but some grow quicker³. Rapidly growing aneurysms act more aggressive and provide greater rupture risk⁴. Nationwide screening for abdominal aortic aneurysm among men above the age of 65 years has been implemented in a few countries, including Sweden and the UK^5,6.. In countries without screening, most cases of abdominal aortic aneurysm are diagnosed as coincidental findings in ultrasonography or computer tomography (CT) scans. They are asymptomatic until rupture, a complication associated with a mortality of nearly 90%⁷. The common risk factors for abdominal aortic aneurysm, such as male sex, increasing age, smoking, and known family history, are well-described⁸. While the option of preventive surgical repair is available, it is only recommended for aneurysms exceeding 55 mm in aortic diameter. This one-size-fits-all threshold has previously been suggested to be a poor predictor of the risk of rupture as some aneurysms rupture before reaching this threshold, while the opposite situation of larger aneurysms progressing without a rupture ever occurring also holds true⁹. Thus, there exists a clear clinical need for biomarkers capable of predicting the presence and progression of aneurysms.

Proteins related to the pathophysiological process of aortic wall dilation and subsequent degradation of the extracellular matrix have been suggested as aneurysm markers, e.g., elastin fragments¹⁰, desmosin¹¹, and matrix metalloproteinases (MMPs)¹², other proteases or their activators and inhibitors such as tissue-plasminogen activator (tPA), plasmin, elastase, chymase, tryptase, and the cathepsin-inhibitor cystatin C^13–18. Moreover, plasma proteins such as C-reactive protein (CRP) and D-dimer, commonly changed in other conditions, have also been considered. For example, CRP levels have been shown to be significantly correlated with abdominal aortic aneurysm diameter^19,20. Likewise, D-dimer, a fibrin degradation product, has been extensively studied in relation to the detection and progression of abdominal aortic aneurysm, with many studies suggesting that high levels of D-dimer is associated with both the presence and increased growth of abdominal aortic aneurysms^21–23. However, none of them have been shown to be useful as screening markers for abdominal aortic aneurysm.

Only few proteomics studies about abdominal aortic aneurysm exist, most of them old with a limited number of patients included^24,]25, and we hypothesize that assessing plasma samples from a large, well-described cohort using mass spectrometry-based proteomics will identify individual proteins or groups of proteins that are associated with the presence and growth of abdominal aortic aneurysms and provide additional associative and predictive value beyond clinical covariates.

Methods

Data sources and design

This nested case-control study is derived from the large, population-based, multicenter randomized controlled VIVA screening trial including 50,156 men, 65 to 74 years of age in the Central Denmark region²⁶. The trial had no exclusion criteria during the enrollment period from 2008 to 2011. A total of 18,748 participants were screened at baseline in the VIVA trial. Baseline parameters were provided through clinical interviews capturing medical history and cardiovascular risk factors (e.g., smoking history and familial predispositions), blood tests, and ultrasound measurements of the abdominal aorta. Baseline clinical data were supplemented with individual-level nationwide register data from the Danish health and administrative registries using the unique person identification numbers from the Central Person Registry²⁷. Mortalities during the study time were identified using data from the Register of Causes of Death. Baseline data on prescription drug use were self-reported during interview. Diagnosis of abdominal aortic aneurysms was carried out by a team of specially trained nurses in mobile clinics in the trial area using portable ultrasound scanners. Abdominal aorta measurements were based on a right-angled, anterior-posterior measurement, and an inner-inner aortic diameter of at least 30 mm measured at peak systole was defined as aneurysmal. Annual follow-up scans were offered to monitor if any of the expansions would exceed 55 mm. Growth rates were calculated for each participant diagnosed with abdominal aortic aneurysm using linear regression based on all available measurements. Patients with abdominal aortic aneurysms had a median of 4 aortic measurements, ranging from 2 to 7, during the 5-year follow-up. The date of screening was defined as the index date, and participants were followed for up to five years and were censored at time of death or completion of 5 years from the index date. All 466 participants diagnosed with abdominal aortic aneurysm at baseline were included, and 175 age-matched controls without abdominal aortic aneurysm were randomly selected from the remaining study population. Participants with abdominal aortic aneurysms were referred for consultation regarding surgical repair if their dilation surpassed 50 mm; of the 466 cases, 199 underwent surgical repair during the five-year period.

To protect sensitive patient information, data from the VIVA screening trial and subsequent proteomic analyses are not publicly available but can be shared upon reasonable request in compliance with European GDPR regulations.

Proteomic screening design

In this sub-study, 641 individuals from the main trial had plasma samples available for proteomic analysis. This number is adequate for measuring the relative changes in protein abundance based on previous sample size calculations using common significance and statistical power cut-offs used in proteomic clinical studies (α = 0.05 and β = 0.8)²⁸. The proteomic screening was performed on plasma samples using highly sensitive mass spectrometry-based proteomics, which have previously been described in detail, and was performed using either an Orbitrap Eclipse mass spectrometer (Thermo Fisher Scientific, San Jose, CA, USA) or an Orbitrap Exploris 480 mass spectrometer (Thermo Fisher Scientific, Bremen, Germany) equipped with a nano-HPLC interface (Dionex UltiMate 3000 nano-HPLC, Thermo Scientific, Bremen, Germany)²⁹. Instrument-related batch effects were evaluated by principal component analysis (PCA) across all quantified proteins, and no evidence of sample clustering by instrument was observed. In addition, investigators performing the proteomic analyses were blinded to case/control status until the dataset was finalized. Consequently, samples were randomly allocated to TMT sets without regard to clinical group assignment. Please see Supplemental Methods S1 for additional details on the proteomic analysis.

Statistics

The primary outcome of this sub-study focused on identifying proteins able to predict the presence of abdominal aortic aneurysms, annual growth, and time-to-death in aneurysmal patients. A full overview of covariate definitions is available in Table S1. Differences between groups were evaluated using the Student’s t-test for continuous data and chi-squared test for categorical data. Continuous data were reported as means ± standard deviation and categorical data as counts with percentages.

Proteins were first evaluated through association analyses with abdominal aortic aneurysms presence, growth, and mortality. All proteins identified as significantly associated with these outcomes were subsequently included in prediction analyses to evaluate their incremental value compared to clinical variables. For abdominal aortic aneurysm presence, separate regression models were constructed for each individual protein, adjusted for established clinical risk factors for abdominal aortic aneurysms, including age, body surface area, smoking, diabetes, hypertension, family history of abdominal aortic aneurysms, and prior acute myocardial infarction³⁰. Discriminative abilities of proteins significantly associated with abdominal aortic aneurysm presence were assessed using the area under the curve (AUC) for the receiver operating characteristic curve, and differences in AUCs were tested using DeLong’s test. A sensitivity analysis using LASSO (Least Absolute Shrinkage and Selection Operator) regression was performed to further evaluate model robustness and potential redundant protein predictors (see Supplementary Methods S2).

For abdominal aortic aneurysm growth analyses, patients aneurysms were classified according to the yearly growth rate, defined as such: slow (< 2.5 mm/year), medium (2.5–5.0 mm/year), or fast (> 5.0 mm/year) growing. Because linear regression analyses did not identify significant associations, an exploratory approach was deemed appropriate with the use of machine learning-based variable selection for the classification analyses was performed instead (see Supplementary Methods S2). All available proteins and clinical variables were included as potential predictors, and the most robust markers for distinguishing between slow-, medium-, and fast-growing aneurysms were identified. Multinominal logistic regression combined with stability-based variable selection of clinical factors and proteins was used to classify aneurysm growth categories. Model performance was evaluated by comparing the AUCs for each growth category (slow, medium, fast) and overall performance was summarized as the average AUC (macro-AUC).

For prediction of mortality among patients with abdominal aortic aneurysm, each individual protein was first evaluated in separate Cox proportional hazard models adjusted for the aforementioned established clinical risk factors to assess its association with time-to-death. Proteins identified as significantly associated were subsequently included in a combined prediction model with the clinical risk factors to evaluate their incremental prognostic value. Improvements of adding proteins were evaluated based on the concordance statistics (c-statistic) and statistical significance was evaluated using likelihood-ratio testing.

For each outcome, model performance was assessed using 10-fold cross-validation and bootstrap bias-corrected 95% confidence intervals were calculated. All statistical analyses were carried out in R (v. 4.2.2). Procedures for multiple testing correction and detailed descriptions of the statistical methods are provided in Supplemental Methods S2.

Ethics

All participants provided both written and verbal informed consent upon enrollment in the VIVA screening trial. Ethical approval was obtained from the Ethics Committee of Central Denmark Region (journal number: M-20080028), and at the Danish Data Protection Agency (journal No. 1–16-02-1-08)²⁶. The study was performed in accordance with the Declaration of Helsinki.

Results

The VIVA trial screened 18,748 participants aged 65 to 74 years. This study included all participants with an abdominal aortic aneurysm ≥ 30 mm identified at baseline (n = 466) and 175 age-matched controls from the VIVA trial that had available plasma samples for proteomic analysis. Baseline characteristics are shown in Table 1. The mean age was 69.8 years and all participants were men. Participants with abdominal aortic aneurysms generally had a higher weight and were more likely to be former or current smokers. In addition, statin therapy and use of platelet inhibitors were more prevalent among individuals with abdominal aortic aneurysms.

Table 1.

Clinical characteristics in the study population stratified by abdominal aortic aneurysm status.

Individuals, N	< 30 mm	≥ 30 mm	P
	175 (27.3%)	466 (73.7%)	-
Demographics	Mean (± SD)	Mean (± SD)
Age, years	69.6 (2.87)	69.9 (2.81)	0.219
Height, m	1.76 (0.0604)	1.76 (0.0612)	0.268
Weight, kg	81.3 (11.7)	85.2 (12.6)	< 0.001
Body surface area, m2	1.97 (0.150)	2.02 (0.155)	0.001
Systolic blood pressure, mmHg	147 (18.2)	156 (21.9)	< 0.001
Diastolic blood pressure, mmHg	81.2 (9.85)	87.9 (12.1)	< 0.001
Total cholesterol, mmol/L	4.82 (1.15)	4.89 (0.902)	0.428
Smoking status	N (%)	N (%)	< 0.001
- Never	56 (32.0%)	40 (8.6%)	-
- Former	86 (49.1%)	232 (49.8%)	-
- Current	33 (18.9%)	194 (41.6%)	-
Comorbidities
Hypertension	78 (44.6%)	252 (54.1%)	0.040
Diabetes	28 (16.0%)	55 (11.8%)	0.201
Chronic obstructive pulmonary disease	2 (1.1%)	19 (4.1%)	0.107
Acute myocardial infarction	3 (1.7%)	26 (5.6%)	0.060
Familial disposition to AAA	6 (3.4%)	33 (7.1%)	0.124
Medical treatment
Statin therapy	64 (36.6%)	238 (51.1%)	0.001
Anticoagulant therapy	14 (8.0%)	32 (6.9%)	0.805
Platelet inhibitors	45 (25.7%)	230 (49.8)	< 0.001

Data are presented as mean (± standard deviation (SD)) for continuous variables and n (%) for categorical variables, respectively. P-values for differences between cases with abdominal aortic aneurysm and non-aneurysmal controls is calculated using Student’s t-test for continuous variables and chi-squared test for categorical variables. P-values < 0.05 is considered statistically significant. Abdominal aortic aneurysm (AAA); Standard Deviation (SD).

In this study, regression models were used for two related, but distinct purposes. First, individual proteins were evaluated in models adjusted for established clinical risk factors to determine whether were independently associated with three separate outcomes: presence, growth, and time-to-death. Second, proteins identified as significantly associated were incorporated into multivariable regression models to assess whether proteins improved predictive performance beyond baseline models containing clinical risk factors alone. Predictive performance was quantified using changes in discrimination metrics (e.g. ΔAUC and Δc-statistic). In total, 3186 proteins were quantified in at least one TMT set. After filtering to only retain proteins quantified in at least 50% of patients, 242 proteins were included in the final analyses.

Identification of proteins associated with abdominal aortic aneurysm presence, annual growth, and time-to-death

Proteins associated with the presence of abdominal aortic aneurysms

Logistic regression analysis was applied to identify individual proteins significantly associated with the presence and annual growth of abdominal aortic aneurysms, and to time-to-death. Associations between individual proteins and outcomes were adjusted for common risk factors for abdominal aortic aneurysms: age, body surface area, diabetes, hypertension, familial history, smoking, and previous acute myocardial infarctions³⁰.

Twenty-seven proteins remained significantly associated with the presence of abdominal aortic aneurysm after adjustment for multiple testing (FDR < 0.05; Table 2). Here, CFH (complement factor H) had the strongest positive association with presence of an abdominal aortic aneurysm, while F13A1 (coagulation factor XIII A chain) had the strongest negative association, and the majority of the identified proteins were related to either blood coagulation or the complement system.

Table 2.

Predictors of abdominal aortic aneurysm presence.

Variables	Protein names	Association		Prediction
		Adjusted OR (95%CI)	P-value	Adjusted OR (95% CI)	P-value
Age				0.99 (0.92; 1.09)	0.95
Body surface area				7.33 (1.38; 40.95)	0.021
Smoking status
- Former				3.61 (1.92; 6.89)	< 0.001
- Current				10.14 (4.8; 22.19)	< 0.001
Diabetes				0.26 (0.12; 0.54)	< 0.001
Hypertension				1.78 (1.11; 2.87)	0.018
Family history of abdominal aortic aneurysm				2.21 (0.75; 8.02)	0.18
Previous event of acute myocardial infarction				4.06 (1.00; 22.44)	0.072
Proteins
ABCB9	ABC-type oligopeptide transporter ABCB9	0.07 (0.03; 0.20)	< 0.001	0.02 (0.00; 0.13)	< 0.001
ACTB	Actin, cytoplasmic 1	5.99 (2.02; 19.21)	0.023	4.14 (0.96; 19.7)	0.066
AGRN	Agrin	2.07 (1.27; 3.48)	0.043	1.50 (0.80; 2.93)	0.22
ORM1	Alpha-1-acid glycoprotein 1	2.70 (1.35; 5.50)	0.049	2.34 (0.81; 7.17)	0.12
A2M	Alpha-2-macroglobulin	4.34 (1.69; 11.85)	0.033	4.19 (1.21; 15.85)	0.029
APOD	Apolipoprotein D	0.15 (0.06; 0.38)	0.0021	0.53 (0.15; 1.92)	0.34
APOM	Apolipoprotein M	0.060 (0.01; 0.23)	0.0016	0.20 (0.03; 1.25)	0.088
C4BPA	C4b-binding protein alpha chain	7.68 (2.24; 27.40)	0.019	3.19 (0.41; 25.96)	0.27
F13A1	Coagulation factor XIIIa	0.04 (0.01; 0.11)	< 0.001	0.07 (0.02; 0.33)	< 0.001
C3	Complement C3	49.86 (7.89; 332.87)	0.0011	1.05 (0.06; 16.45)	0.97
C4A	Complement C4-A	1.95 (1.25; 3.10)	0.043	1.71 (0.98; 3.08)	0.066
C9	Complement component C9	5.58 (1.96; 16.29)	0.019	1.68 (0.25; 11.7)	0.60
CFH	Complement factor H	122.71 (16.89; 966.78)	< 0.001	256.78 (10.06; 7503.99)	< 0.001
CFI	Complement factor I	11.31 (2.37; 56.6)	0.030	0.66 (0.05; 9.39)	0.76
ECM1	Extracellular matrix protein 1	0.12 (0.04; 0.37)	0.0050	1.13 (0.19; 6.58)	0.89
FGA	Fibrinogen alpha chain	0.10 (0.04; 0.27)	< 0.001	8.21 (0.30; 292.57)	0.23
FGB	Fibrinogen beta chain	0.10 (0.03; 0.28)	< 0.001	0.89 (0.01; 78.12)	0.96
FGG	Fibrinogen gamma chain	0.09 (0.03; 0.24)	< 0.001	0.07 (0.02; 3.32)	0.19
FN1	Fibronectin	0.36 (0.23; 0.56)	< 0.001	1.40 (0.62; 3.2)	0.42
LBP	Lipopolysaccharide-binding protein	9.60 (3.65; 26.58)	< 0.001	1.99 (0.51; 8.14)	0.33
LYZ	Lysozyme C	3.84 (1.57; 9.96)	0.043	1.64 (0.48; 6.02)	0.44
SERPINF1	Pigment epithelium-derived factor	12.39 (3.06; 52.13)	0.0074	2.60 (0.44; 16.1)	0.30
PPBP	Platelet basic protein	2.06 (1.40; 3.09)	0.0060	1.92 (0.92; 4.05)	0.083
PF4	Platelet factor 4	3.06 (1.54; 6.40)	0.025	0.90 (0.26; 3.17)	0.86
S100A8	Protein S100-A8	2.91 (1.48; 6.44)	0.043	0.83 (0.3; 2.53)	0.73
S100A9	Protein S100-A9	4.20 (2.10; 9.08)	0.0025	3.45 (1.11; 11.41)	0.038
APCS	Serum amyloid P-component	6.96 (2.48; 20.31)	0.0050	1.63 (0.32; 8.41)	0.56

Adjusted odds ratios (ORs) and 95% confidence interval (CI) for abdominal aortic aneurysm. In the association analyses, each protein was evaluated in separate logistic regression model adjusted for common risk factors for abdominal aortic aneurysms: age, body surface area, smoking, diabetes, hypertension, familial history of abdominal aortic aneurysm, and previous acute myocardial infarction. In the multivariable prediction model, all proteins identified as significantly associated in the association analyses were utilized simultaneously in a single logistic regression model together with the same clinical risk factors. P-values for the association analyses were adjusted for multiple testing, and p-values < 0.05 were considered statistically significant.

Proteins associated with abdominal aortic aneurysm growth

It is relevant to distinguish between slow-, medium-, and fast-growing abdominal aortic aneurysms because the growth rate independently predicts rupture risk and guides management. Therefore, the annual abdominal aortic aneurysm growth was divided into slow-growing (< 2.5 mm/year), medium-growing (2.5–5.0 mm/year) and fast-growing (> 5.0 mm/year) aneurysms³⁷(Table S2, Figure S1). Multinomial logistic regression analysis was initially applied to identify proteins associated with slow, medium or fast-growing abdominal aortic aneurysms. Unfortunately, none of the analyzed proteins were significantly associated with the growth of slow, medium or fast-growing abdominal aortic aneurysm after adjusting for multiple testing (Table S3).

Proteins associated with time-to-death

For the association of plasma proteins with time-to-death for individuals with an abdominal aortic aneurysm, 13 proteins were identified to be significantly associated, including CP (ceruloplasmin) with the strongest positive association while the protein F2 (prothrombin) demonstrated the strongest negative association after adjusting for multiple testing (Table 3).

Table 3.

Predictors of time-to-death of individuals with abdominal aortic aneurysms.

Variables	Protein names	Association		Prediction
		Adjusted HR (95% CI)	P-value	Adjusted HR (95% CI)	P-value
Age				1.06 (0.98; 1.15)	0.12
Body surface area				0.20 (0.04; 0.89)	0.035
Smoking status
- Former				0.90 (0.40; 2.05)	0.81
- Current				0.53 (0.23; 1.23)	0.14
Diabetes				1.83 (1.01; 3.30)	0.047
Hypertension				0.84 (0.53; 1.32)	0.44
Family history of abdominal aortic aneurysm				0.43 (0.13; 1.39)	0.16
Previous event of acute myocardial infarction				0.07 (0.01; 0.63)	0.017
Proteins
ALB	Albumin	0.030 (0.0029; 0.28)	0.042	0.76 (0.04; 12.96)	0.85
B2M	Beta-2-microglobulin	1.95 (1.35; 2.80)	0.016	1.07 (0.54; 2.12)	0.84
CP	Ceruloplasmin	27.40 (5.85; 128.46)	0.0032	4.04 (0.76; 21.54)	0.10
C9	Complement component C9	5.87 (2.14; 16.08)	0.020	2.07 (0.60; 7.11)	0.25
IGLV1-47	Immunoglobulin lambda variable 1–47	2.07 (1.45; 2.95)	0.0049	1.82 (1.26; 2.62)	0.0013
IGFBP2	Insulin-like growth factor-binding protein 2	2.74 (1.47; 5.12)	0.041	1.85 (0.85; 4.05)	0.12
ITIH4	Inter-alpha-trypsin inhibitor heavy chain H4	6.63 (2.01; 21.90)	0.042	3.05 (0.73; 12.86)	0.13
LYZ	Lysozyme C	2.46 (1.45; 4.16)	0.024	1.63 (0.77; 3.42)	0.20
PIGR	Polymeric immunoglobulin receptor	3.64 (2.11; 6.29)	< 0.001	2.79 (1.44; 5.39)	0.0023
PZP	Pregnancy zone protein	1.56 (1.23; 1.98)	0.014	1.34 (1.01; 1.78)	0.043
S100A8	Protein S100-A8	1.15 (1.06; 1.26)	0.042	1.13 (1.01; 1.25)	0.026
F2	Prothrombin	0.08 (0.02; 0.33)	0.020	0.09 (0.02; 0.39)	0.0011
PTPN1	Tyrosine-protein phosphatase non-receptor type 1	1.21 (1.07; 1.36)	0.042	1.28 (1.08; 1.53)	0.0043

Adjusted hazard ratios (HRs) and 95% confidence interval (CI) for time-to-death for individuals with abdominal aortic aneurysm. In the association analyses, each protein was evaluated in separate Cox proportional regression models adjusted for common risk factors for abdominal aortic aneurysms: age, body surface area, smoking, diabetes, hypertension, familial history of abdominal aortic aneurysm, and previous acute myocardial infarction. In the multivariable Cox proportional prediction model, all proteins identified as significantly associated in the association analyses were utilized simultaneously in a single Cox proportional hazards model together with the same clinical risk factors. P-values for the association analyses were adjusted for multiple testing, and p-values < 0.05 were considered statistically significant.

Prediction of the presence and growth of abdominal aortic aneurysms, and of time-to-death

Building on the association analyses, proteins identified as significantly associated with the clinical outcomes were subsequently incorporated into multivariable models to determine whether the proteins provided incremental predictive value beyond established clinical risk factors.

Prediction of presence of abdominal aortic aneurysms

For abdominal aortic aneurysm a baseline multivariable logistic regression model based on the aforementioned clinical risk factors (Model 1, Fig. 1) was compared to a multivariable logistic regression model based on the 27 proteins (Table 2) significantly associated with the presence of abdominal aortic aneurysms (Model 2, Fig. 1) and a model including both proteins and clinical risk factors (Model 3, Fig. 1). Using cross-validation, the inclusion of proteins into the baseline model (i.e., Model 3) substantially improved the AUC compared to the baseline model (i.e., Model 1) (ΔAUC: +0.11, P = 0.00024), and in comparison, only modestly improved the AUC compared to the proteins by themselves (i.e., Model 2) (ΔAUC: +0.03, P = 0.022) (Fig. 1). To assess the robustness of these findings, a sensitivity analysis was performed using LASSO regression applied to the same set of variables. The resulting LASSO-derived model showed similar performance (AUC = 0.82), with no significant difference compared with the combined model (ΔAUC=−0.01, P = 0.67), suggesting that the findings were stable and not dependent on overfitting or redundant predictors.

Fig. 1.

Receiver operating characteristic curves for prediction of the presence of abdominal aortic aneurysm. ROC curves for the prediction of abdominal aortic aneurysm presence. Three models were evaluated: Model 1 includes clinical risk factors (age, body surface area, diabetes, hypertension, smoking status, family history of abdominal aortic aneurysm, and prior acute myocardial infarction); Model 2 includes the 27 significantly associated proteins alone; and Model 3 combines both clinical variables and proteins. The area under the curve (AUC) are reported for each model comparison to assess incremental improvements in predictive accuracy. All statistics were calculated using data not used for model training.

Prediction of growth of abdominal aortic aneurysms

As no proteins were significantly associated with annual growth rate after multiple testing, an alternative approach was applied in an exploratory analysis, to generate new hypotheses that can later be validated. First, a machine learning-based selection procedure was used to identify the most informative predictors for distinguishing between slow-, medium-, and fast-growing abdominal aortic aneurysms. These variables included the baseline aortic diameter, smoking status, diabetes, and 22 proteins (Table S4). Model performance was then evaluated for each growth category separately using a one-vs-rest approach, e.g. fast-growing aneurysms were compared with the combined group of slow- and medium-growing aneurysms to assess the model’s ability to distinguish that specific group. Model performance compared across three expanded models, beginning with baseline aortic diameter alone (Model 1), then adding smoking status and diabetes (Model 2), and finally, the 22 selected proteins were incorporated into the full model (Model 3). Examination of the AUCs for each growth category (slow, medium, fast) showed that the 22 proteins consistently improved predictions across slow (Model 1 - AUC: 0.68 vs. Model 2 - AUC: 0.70 vs. Model 3 - AUC: 0.73), medium (0.58 vs. 0.57 vs. 0.70), and fast (0.70 vs. 0.74 vs. 0.79) growth categories, when combined with baseline aortic diameter, diabetes, and smoking status (Fig. 2). To summarize overall model performance, a macro-AUC was calculated as the average of the three category-specific AUCs, which reflects the performance across all growth categories equally. The full model (Model 3) showed a substantial improvement in the overall ability to distinguish between the three growth categories compared to baseline aortic diameter by itself (Model 1, + 0.09, P = 0.0020) and aortic diameter, smoking status and diabetes combined (Model 2, + 0.07, P = 0.0040) (Table 4).

Fig. 2.

Receiver operating characteristic curves for abdominal aortic aneurysm growth prediction. ROC curves for the discrimination patients with abdominal aortic aneurysms as either slow- (A), medium- (B) or fast-growing (C). Model 1 includes baseline aortic diameter alone; Model 2 adds machine learning selected clinical variables (smoking status and diabetes); and Model 3 is the full combined model including aortic diameter, clinical variables, and 22 machine learning selected proteins. The per-class area under the curve (AUC) are reported for each model comparison to assess incremental improvements in predictive accuracy. All statistics were calculated using data not used for model training.

Table 4.

Comparison of models for classifying aneurysm growth.

Models	Overall performance (Macro-AUC)	P-value
Model 1: Aortic diameter	0.65	reference
Model 2: Aortic diameter, diabetes, smoking	0.67	0.46
Model 3: Aortic diameter, diabetes, smoking + 22 proteins	0.74	0.0020

P-values were derived from bootstrap resampling (1,000 repetitions) to evaluate whether differences in the overall model performance were consistent. Overall performance is summarized as the macro-AUC, defined as the average ability of the model to distinguish between the three growth groups: slow, medium, and fast.

Prediction of time-to-death

For the prediction of time-to-death among participants with an abdominal aortic aneurysm (mean follow up: 5.03 years; 98 deaths), the 13 proteins identified as significantly associated with mortality were included in a baseline Cox proportional regression model containing established clinical risk factors (Table 3). Inclusion of the proteins improved the overall model fit, as demonstrated by a likelihood-ratio test, compared with the baseline clinical model (likelihood-ratio χ² = 79.92, p < 1e-11). The ability to distinguish between patients with shorter and longer survival improved substantially with addition of proteins compared to the baseline model, as demonstrated by the c-statistic increasing to c = 0.78 with the addition of proteins (c: +0.13, P = 4.07E-6).

Discussion

This study provides an important proof-of-concept for the use of proteomics in predicting essential outcomes related to abdominal aortic aneurysms, including the presence, progression, and risk of mortality. Using mass spectrometry-based proteomics, we identified multiple circulating proteins that were significantly associated with abdominal aortic aneurysm-related endpoints after adjusting for traditional clinical risk factors. In predictive models, these proteins improved the discrimination for abdominal aortic aneurysm presence and mortality and, importantly, showed potential clinical utility for predicting aneurysm growth when used in combination with conventional clinical variables. These results suggest that identification of individuals at high risk of adverse abdominal aortic aneurysm-related outcomes might be possible through the complementary use of proteomic profiling.

Many individual proteins were discovered to have strong associations with abdominal aortic aneurysm presence. These proteins were related to biological processes previously suggested to be involved in the formation and progression of abdominal aortic aneurysms, namely blood coagulation, complement activation and modulation of endothelial cell proliferation^31–33. Inclusion of proteins into a baseline model consisting of traditional risk factors markedly improved the prediction (AUC = 0.83), strengthening the rationale for including proteomic analyses in future population-screening studies. Our results align with recent proteomic studies that similarly showed improvements in the AUC using proteomics data for the prediction of abdominal aortic aneurysm presence^34–36.

Beyond presence, prediction of growth is of particular clinical relevance. Although no individual proteins showed significant associations with annual growth after correction for multiple testing, a machine learning approach based on penalized multinomial logistic regression was applied to all clinical variables and proteins. This exploratory analysis identified baseline aortic diameter, diabetes, smoking, and 22 proteins as the most informative variables, which could distinguish between slow-, medium- and fast-growing abdominal aortic aneurysms. Dividing patients with abdominal aortic aneurysm into three distinct categories offered key advantages over continuous predictions that suffered from high bias (Figure S2) in our study. By focusing on distinct growth groups, the models could better capture the biological signature of abdominal aortic aneurysm growth and provide more reliable predictions for potential clinical decision-making. This categorical approach is directly translatable to practice since management decisions are already guided primarily by growth thresholds and surveillance intervals. The selected cut-offs of < 2.5 mm and > 5.0 mm for categorizing aneurysms as either slow- or fast-growing, respectively, were adopted from a previous study³⁷. In this study, Chandrashekar and colleagues demonstrated that categorical prediction of growth was a viable alternative to continuous predictions, achieving an AUC of 0.80 using geometric features obtained from CT scans of abdominal aortic aneurysms. Several of the identified proteins for growth prediction are biologically plausible and may provide a more diverse insight into the progression of abdominal aortic aneurysms. Notably, multiple key components of the complement system were identified, including CFH (complement factor H) and C4A (complement C4A), aligning with previous evidence of the role of complement system dysregulation in aneurysmal development³⁸. Additionally, LPA (apolipoprotein (a)), the primary component of the well-established cardiovascular disease marker lipoprotein (a), was also significantly associated with abdominal aortic growth. PCYOX1 (prenylcysteine oxidase 1), a pro-oxidant enzyme involved in hydrogen peroxide production, and F5 (coagulation factor V), a key regulator of hemostasis, further underscore the multifactorial biology and supports the notion that inflammatory, oxidative, and thrombotic processes promote aneurysmal growth. The added predictive value of proteins with baseline aortic diameter was notable, demonstrating the feasibility of improving risk assessment of individuals with abdominal aortic aneurysm at the time of initial imaging. Although aneurysm growth is defined by serial imaging and imaging remains the clinical gold standard for measuring aneurysm size, baseline diameter alone does not fully capture the future growth behavior of an aneurysm. The results from the combined model suggest that circulating proteins may provide complementary biological information that can help to guide surveillance frequency by distinguishing slow- from fast-growing aneurysms.

Most proteins identified as relevant across the different abdominal aortic aneurysm outcomes were unique to the specific outcome (Figure S3A). However, gene ontology enrichment analyses revealed that the proteins share largely similar biological processes across presence, growth, and mortality (Figure S3B-D, respectively). This suggests that while distinct proteins are identified for each endpoint, the underlying pathways are shared, particularly those related to inflammation, and to a lesser degree complement activation, and coagulation.

In the context of mortality, the enriched biological pathways related to inflammation and homeostatic processes support the interpretation that proteins such as CP, PIGR, ALB, B2M, LYZ, and S100A8 likely capture systemic processes known to contribute to cardiovascular disease progression and overall frailty in general, rather than specific biological processes distinct for abdominal aortic aneurysms. This interpretation is supported by a large number of studies reporting that these proteins are also associated with mortality prediction in other cardiovascular diseases and a diverse range of other clinical conditions^39–48. As such, the use of all-cause mortality as the endpoint prevents the possibility of distinguishing AAA-related cause of death from other competing causes. Importantly, while the mortality-associated proteins likely represent a broader systemic risk, the inclusion of proteins into the multivariable Cox model substantially improved the c-statistic compared to established risk factors, suggesting that proteomic data may assist with identifying high-risk patients.

Taken together, these findings resonate with the broader trends in precision medicine where proteomic-generated protein panels can achieve more meaningful improvement in predictive performance than traditional solitary biomarkers, e.g., NT-proBNP or CRP, in relation to cardiovascular diseases⁴⁹. The observed improvements in our study compare well with gains reported in other large-scale proteomic studies investigating abdominal aortic aneurysm^34,36. In particular, the demonstrated ability to improve growth prediction represents a potentially clinically useful advance, since the growth rate is directly related to the timing of surgical intervention. As such, proteomics could advance the risk assessment beyond static diameter thresholds toward more dynamic progression estimates. As proteomic technologies become more accessible and scalable, investment into multi-biomarker risk assessment tools may become clinically relevant.

The strengths of our study include that proteins were investigated for their association and predictive ability with relevant outcomes related to abdominal aortic aneurysms. Another strength of this study is that the main trial, from which our study cohort is derived, is based on a population-based screening design covering diverse geographic regions across Denmark using standardized measurement protocols. Unlike the majority of prior studies, our study does not rely on ICD-10 codes for diagnosis of abdominal aortic aneurysms, as all aneurysm patients were diagnosed by ultrasound scanning, and is therefore not subject to the same misclassification bias.

However, the study has several limitations that should be considered. First, the included participants were mainly white men aged 65–74 from Denmark, as such generalizability to other populations remains untested. Second, the aortic measurements were conducted by multiple trained personnel, thereby introducing inter-observer variation of the aortic anterior-posterior diameter. However, the study protocol of the VIVA trial only allowed for independent testing when a team’s inter-observer variation was less than 2 mm²⁶. Additionally, by incorporating every available measurement of the aortic diameter from each visit, we minimized the variation through the principle of regression toward the mean. Lastly, the proteins identified in the present study consist primarily of abundant plasma proteins, including complement factors, apolipoproteins, transport proteins, and coagulation factors. This reflects the characteristic nature of mass spectrometry-based proteomics, which reliably quantifies proteins present across vast concentrations in plasma. In contrast, very low-abundant proteins, like cytokines, are often not detected and fragments of these proteins cannot be identified unless extensive additional laboratory steps are performed to remove high-abundant proteins or selectively enrich low-abundant proteins⁵⁰. However, a key strength of mass spectrometry is the high specificity of the identified proteins, in contrast to affinity binding-dependent methods, such as SomaScan and Olink^51,52.

Conclusion

This study provides compelling evidence that plasma proteomics may improve the detection and risk stratifications of abdominal aortic aneurysms in older men. Several circulating proteins were independently associated with outcomes related to abdominal aortic aneurysms, highlighting known biologically plausible pathways related to complement activation, oxidative stress, and blood coagulation. Importantly, while no individual proteins were significantly associated with aneurysm growth after adjustment for multiple testing, a subset of proteins selected by machine learning improved the discrimination between fast- and slow-growing aneurysms beyond clinical variables. This ability to discriminate between growth categories could prove potentially clinically useful, as growth rates are directly related to timing surgical interventions. Proteins also demonstrated the improved ability to identify patients at higher risk of mortality compared with model based solely on clinical risk factors. Collectively, these results suggest that circulating protein biomarkers may complement established risk factors for the diagnosis, growth assessment, and management of abdominal aortic aneurysm. However, as all models were developed and evaluated within the same cohort, further validation of the proteins in independent cohorts and across populations that are more diverse is needed to confirm and assess their clinical utility.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

LPA

Apolipoprotein (a)

AUC

Area under the curve

CP

Ceruloplasmin

F5

Coagulation factor V

F13A1

Coagulation factor XIII A chain

C4A

Complement C4A

CFH

Complement factor H

CT

Computer tomography

CRP

C-reactive protein

FDR

False discovery rate

LASSO

Least Absolute Shrinkage and Selection Operator

MMPs

Matrix metalloproteinases

PIGR

Polymeric immunoglobulin receptor

PCYOX1

Prenylcysteine oxidase 1

F2

Prothrombin

tPa

Tissue-plasminogen activator

VIVA

Viborg Vascular trial

Author contributions

The authors acknowledge the following contributions: concept and design: H.C.B., L.M.R., J.S.L.; data acquisition: N.B.P., H.C.B., M.D., J.S.L.; data analysis or interpretation: A.H., M.S., N.B.P., J.S.L.; drafting of the manuscript: N.B.P., H.C.B.;A.J.C.; critical revision of the manuscript for important intellectual content: all authors; statistical analyses: A.H., M.S., N.B.P.; funding: N.B.P., H.C.B., L.M.R., J.S.L.; and supervision: H.C.B., L.M.R., J.S.L.

Funding

This study was supported by the 7th European Framework Programme, Central Denmark Region, Viborg Hospital, and the Danish Council for Independent Research. N.B.P. was partly supported by the Danish Cardiovascular Academy, which is funded by the Novo Nordisk Foundation, grant number NNF20SA0067242 and The Danish Heart Foundation. The funders had no role in study design, interpretation, or publication, and were not informed before submission.

Data availability

Data from the VIVA screening trial and subsequent proteomic analyses are not publicly available but can be shared upon reasonable request in compliance with European GDPR regulations.

Declarations

Competing interests

The authors declare no competing interests.