Social Determinants of Health and Polygenic Susceptibility in Nonsyndromic Ascending Aortic Aneurysm

Abstract

Background

Nonsyndromic ascending aortic aneurysm (AsAA) may remain clinically silent until dissection or rupture. The contributions of multidomain social determinants of health (SDOH) and polygenic susceptibility to AsAA risk are not well defined.

Methods

In a cross‐sectional analysis of 124 180 participants in the National Institutes of Health All of Us Research Program, 14 SDOH measures were summarized as a standardized composite score, and an ascending aortic diameter polygenic risk score (PRS; polygenic score, 004927) was standardized within ancestry strata. Associations were tested using Firth‐penalized logistic regression adjusted for clinical risk factors; discrimination was assessed by area under the receiver operating characteristic curve.

Results

Among 124 180 participants, 346 had AsAA. Compared with controls, cases were older (mean age, 70.7±9.0 versus 57.1±16.4 years) and more often men (60.4% versus 37.1%). Each 1‐SD increase in the SDOH composite was associated with AsAA after clinical adjustment (odds ratio, 1.21 [95% CI, 1.04–1.40]) and in joint models including PRS (odds ratio, 1.28 [95% CI, 1.11–1.48]). The PRS was associated with AsAA after clinical adjustment (odds ratio per SD, 1.94 [95% CI, 1.74–2.16]); AsAA prevalence increased from 0.08% (lowest PRS decile) to 0.77% (highest). SDOH and PRS were minimally correlated (r=−0.002), with no evidence of interaction (interaction odds ratio, 1.03; P=0.62). Discrimination improved from basic covariates (area under the receiver operating characteristic curve, 0.774) to clinical covariates (area under the receiver operating characteristic curve, 0.841) and to the full model including SDOH and PRS (area under the receiver operating characteristic curve, 0.870 [95% CI, 0.851–0.889]).

Conclusions

Social adversity and higher polygenic susceptibility were associated with nonsyndromic AsAA, supporting further study of integrated social and genetic risk stratification.

Nonstandard Abbreviations and Acronyms

AoU

All of Us

AsAA

ascending aortic aneurysm

PC

principal component

PGS

polygenic score

PRS

polygenic risk score

SDOH

social determinants of health

WGS

whole‐genome sequencing

Research Perspective.

What Is New?

• In a large National Institutes of Health All of Us Research Program cohort with linked survey, genomic, and electronic health record data, multidomain social adversity, captured by a composite score spanning housing insecurity, food insecurity, loneliness, and other domains, was associated with nonsyndromic ascending aortic aneurysm after accounting for conventional clinical risk factors and a diameter‐derived polygenic risk score.

• A polygenic score originally developed for ascending aortic diameter (polygenic score, 004927) demonstrated a strong, monotonic gradient of risk for diagnosis‐ascertained, nonsyndromic ascending aortic aneurysm, extending prior imaging‐based work to a clinically recognized electronic health record phenotype.

What Question Should Be Addressed Next?

• Prospective studies with imaging‐confirmed aortic diameter measurements, multiancestry cohorts, and longitudinal follow‐up are needed to determine whether integrated social and genetic risk assessment may improve early detection and targeted surveillance of nonsyndromic ascending aortic aneurysm.

Ascending aortic aneurysm (AsAA) is frequently clinically silent until dissection or rupture, events that remain associated with a high mortality rate despite advances in surgical and critical care. ¹ , ² Current guideline‐driven screening and surveillance largely focus on patients with clear high‐risk triggers such as heritable aortopathy syndromes, bicuspid aortic valve, or a strong family history. ³ , ⁴ , ⁵ As a result, many individuals with nonsyndromic disease lack a clear pathway to early risk identification and targeted surveillance.

Social and economic adversity is consistently linked to cardiovascular morbidity and death, yet its role in thoracic aortic disease remains comparatively understudied. ⁶ , ⁷ , ⁸ , ⁹ Material hardship and social disconnection may plausibly shape aortic risk through higher cardiometabolic burden, adverse health behaviors, and reduced access to preventive care and longitudinal follow‐up. Whether adverse social context is associated with nonsyndromic AsAA risk independent of inherited susceptibility is unknown.

In parallel, genome‐wide association studies have enabled polygenic scores (PGSs) that capture inherited liability for ascending aortic diameter, a proximal quantitative trait related to aneurysm development. ¹⁰ , ¹¹ , ¹² The availability of linked survey, genomic, and electronic health record (EHR) data in the All of Us (AoU) Research Program provides an opportunity to evaluate social and genetic risk within a single, clinically grounded framework. ¹³ , ¹⁴ In this study, we evaluated the associations of multidomain social determinants of health with EHR‐ascertained, nonsyndromic AsAA, assessed whether a diameter‐derived PGS stratifies risk for this diagnosis‐ascertained phenotype, and tested whether social and genetic measures were associated with AsAA separately, jointly, or through statistical interaction while quantifying their incremental value beyond conventional clinical covariates.

METHODS

Data Availability

The data will not be made directly available by the authors. This study used data from the All of Us (AoU) Research Program Controlled Tier, which is accessible to authorized researchers at https://www.researchallofus.org/. The analytic code used for this study will be made available upon reasonable request.

Study Design and Cohort

We conducted a cross‐sectional study using the AoU Research Program Researcher Workbench in the Controlled Tier, a secure cloud‐based research environment providing access to curated participant‐level survey, EHR, and whole‐genome sequencing (WGS) data. ¹³ Analyses used the Curated Data Repository version 8 (C2024Q3R5), consistent with AoU release‐specific data structures. EHR data were harmonized to the Observational Medical Outcomes Partnership Common Data Model, enabling standardized phenotype and covariate derivation. ¹⁵ The AoU Research Program protocol was approved by the National Institutes of Health Institutional Review Board. All participants provided written informed consent. This secondary analysis was conducted within the AoU Researcher Workbench and was determined exempt from additional institutional review board review under 45 CFR 46.104(d)(4).

Participants were eligible if they (1) completed the AoU Social Determinants of Health (SDOH) survey (index date defined as survey completion date); (2) had short‐read WGS data with available genetic ancestry predictions and principal components (PCs); and (3) had sufficient EHR data to ascertain outcomes and covariates. We assembled the analytic cohort by intersecting these data sources, excluding participants with missing polygenic risk score (PRS) or ancestry/PC information.

Furthermore, participants with congenital or syndromic aortopathy features, including bicuspid aortic valve and genetic connective‐tissue syndromes associated with thoracic aortopathy (Marfan, Loeys–Dietz, Ehlers–Danlos, and Turner syndromes), were excluded. Cohort construction and exclusions are summarized in the study flow diagram (Figure 1); phenotype definitions and concept sets are detailed in Data S1 through S5. The largest source of attrition arose from selective completion of the AoU SDOH survey, which was available only in a subset of participants; additional exclusions reflected incomplete linkage across genomic and EHR data sources and missing ancestry/PRS information (Figure 1).

Figure 1. Study flow diagram.

Participant selection from the All of Us Research Program. Cell counts <20 are not reportable due to National Institutes of Health All of Us Research Program data privacy policy. AsAA indicates ascending aortic aneurysm; EHR, electronic health record; PRS, polygenic risk score; SDOH, social determinants of health; and WGS, whole‐genome sequencing.

Outcome Ascertainment

Nonsyndromic AsAA case status was ascertained from Observational Medical Outcomes Partnership EHR tables using Systematized Nomenclature of Medicine Clinical Terms standard concepts and International Classification of Diseases, Tenth Revision, Clinical Modification (ICD‐10‐CM) source codes for ascending aortic aneurysm (Data S2). Participants with ≥1 qualifying diagnosis code were classified as cases; all remaining eligible participants served as controls.

Exposures and Covariates

Social Determinants of Health

SDOH exposures were derived from AoU surveys and organized according to Healthy People 2030 domains: economic stability, education access and quality, health care access and quality, neighborhood and built environment, and social and community context. ¹⁶ We operationalized 14 SDOH measures as continuous, binary, or ordinal variables derived from validated survey instruments including the Social Cohesion Scale, Physical Activity Neighborhood Environment Scale, modified Medical Outcomes Study Social Support Survey, UCLA Loneliness Scale, Everyday Discrimination Scale, Perceived Stress Scale, and Hunger Vital Sign (Table S1). Noninformative responses were treated as missing.

The 14 SDOH measures were selected to correspond to the validated psychometric constructs included in the AoU SDOH Survey. Annual household income, educational attainment, and employment status are collected separately in the AoU Basics survey and were therefore not incorporated into the primary composite score, which was intended to summarize SDOH survey–based experiential and perception‐based constructs. To assess whether the SDOH composite provided information beyond demographic socioeconomic position, we performed a sensitivity analysis additionally adjusting primary SDOH and PRS association models for annual household income, educational attainment, and employment status from the Basics survey (Table S5).

Demographic and Clinical Covariates

Demographics included age at index date and sex assigned at birth. Conventional cardiovascular risk factors derived from EHR data included hypertension, diabetes, body mass index, smoking status, and hyperlipidemia. Medication exposures included statin and antihypertensive therapy, defined using class‐based Observational Medical Outcomes Partnership concept sets.

Genetic Ancestry and Population Structure

Genetic ancestry group predictions (European, African, Admixed American, East Asian, Middle Eastern, and South Asian) and 16 precomputed genetic PCs were obtained from AoU‐provided ancestry prediction files. ¹⁴ We used the first 10 PCs for population structure adjustment in all genetic models.

Polygenic Risk Score

We applied a published PRS for ascending aortic diameter directly from the PGS Catalog (PGS004927; GRCh38 build; 1 111 632 variants), derived from the AORTA Gene Score reported by Pirruccello et al, without rederivation, retuning, or recalibration within the present data set. ¹⁰ PRS calculation was performed using the AoUPRS package version 0.2.3, which implements Hail‐based scoring on the AoU WGS Variant Dataset with chunked processing for computational efficiency. Per‐participant PRS values were computed as the sum of weighted effect–allele dosages across variants after harmonizing to reference weights and excluding relatedness‐flagged samples. The only postcomputation processing was ancestry‐stratified standardization (Z scoring) before association modeling to account for ancestry‐related distributional differences (Data S5).

Missing Data and Cross‐Fitting Framework

Missingness in SDOH measures and body mass index was addressed using random‐forest imputation with predictive mean matching. To minimize information leakage, imputation, composite score estimation, and prediction were performed using a 5‐fold cross‐fitting framework. ¹⁷ Within each fold, imputation models and score weights were fit using training data only; held‐out participants received out‐of‐sample imputed values, composite scores, and predicted probabilities generated without using their outcome information (Data S6).

Composite SDOH Score Analyses

To summarize multidomain social adversity, a composite AsAA–SDOH score was constructed as a weighted linear combination of the 14 SDOH measures. Weights were estimated using elastic‐net penalized logistic regression (α=0.5) to accommodate intermeasure correlation and select the most informative contributors, with λ tuned via 10‐fold cross‐validation within each training fold. The resulting composite score was standardized per 1 SD for interpretation and served as the primary prespecified SDOH exposure. Weights contributing to the composite score, including fold selection frequency and mean coefficients, are reported in Tables S1.

As secondary descriptive analyses, individual SDOH measures were also evaluated by fitting separate Firth‐penalized logistic regression models for each measure, adjusting for age, sex, genetic ancestry indicators, 10 PCs, and PRS. Because the primary inferential target was the composite score, no formal multiple‐comparison correction was applied to these individual‐domain analyses; however, the strongest associations were also examined for robustness to Bonferroni correction. Effects are reported as odds ratios (ORs) with 95% CIs per 1‐SD increase for continuous measures or relative to reference categories for binary measures (Figure 2A).

Figure 2. Social and polygenic risk for nonsyndromic AsAA.

A, Forest plot showing odds ratios with 95% CIs for individual SDOH measures associated with AsAA. Odds ratios are per 1‐SD increase for continuous measures and vs reference category for binary measures. Three models shown: basic (age, sex, genetic principal components), basic plus PRS, and basic plus PRS plus clinical risk factors. B, Receiver operating characteristic curves for nested prediction models demonstrating incremental discrimination. AUC values with 95% CIs shown for each model. C, Violin plots showing distribution of composite SDOH score by AsAA status. Box plots indicate median and interquartile range. D, Violin plots showing distribution of ancestry‐standardized PRS by AsAA status. Box plots indicate median and interquartile range. AsAA indicates ascending aortic aneurysm; AUC, area under the curve; PRS, polygenic risk score; and SDOH, social determinants of health.

Primary Association Analyses

Prevalent AsAA was modeled using Firth penalized‐likelihood logistic regression to reduce small‐sample bias and separation. Prespecified nested adjustment blocks were used to quantify the incremental contribution of exposures: (1) a basic model including age, sex, and genetic PCs/ancestry; (2) basic plus the composite SDOH score; (3) basic plus PRS; (4) basic plus PRS and SDOH; (5) a clinical model including conventional risk factors and medications; and (6) a full model including clinical factors plus PRS and SDOH. Effect modification was assessed by including a PRS×SDOH interaction term in models 4 and 6. The interaction term was not included in models 1, 2, 3, or 5, which did not jointly include both PRS and the SDOH composite.

Model Performance and Score Distribution

Model performance was assessed using discrimination, calibration, and decision‐analytic metrics. Discrimination was evaluated using receiver operating characteristic curves and area under the receiver operating characteristic curve (AUC) with DeLong 95% CIs (Figure 2B). Predicted probabilities for each prespecified model were generated out‐of‐sample within the 5‐fold cross‐fitting framework and pooled to estimate unbiased discrimination metrics. Supplementary assessments included Brier scores, grouped‐decile calibration plots, and decision curve analysis (Data S9; Table S2; Figures S10–S11). Because Firth penalization can affect absolute predicted probabilities in rare‐event settings, Firth logistic regression with intercept correction was applied to the fitted Firth models before computing probability‐based performance metrics. The distribution of the composite SDOH score and ancestry‐stratified PRS by AsAA status was visualized using violin plots (Figure 2C and 2D).

Survey Nonresponse Bias Assessment

Because SDOH survey completion may be selective, we evaluated potential nonresponse bias using 3 approaches. First, we modeled the probability of SDOH survey response among all WGS participants using demographic characteristics (age, sex, race, ethnicity, education, income) via generalized boosted models and constructed stabilized inverse‐probability‐of‐response weights (Figures S1 and S2). Covariate balance before and after weighting was assessed using standardized mean differences. Second, we compared response‐weighted distributions across the 14 SDOH measures (Figure S3). Third, we compared association estimates between the full eligible cohort and the SDOH‐respondent subset for sociodemographic variables (Figure S4) and clinical risk factors (Figure S5), summarizing selection effects as relative ORs.

Risk Stratification and Secondary Analyses

We categorized standardized PRS and composite SDOH scores into quintiles and examined AsAA prevalence across the 5×5 joint strata with binomial 95% CIs (Figure S7). AsAA prevalence by PRS decile in the nonsyndromic cohort was also examined (Figure S8). Marginal predicted probabilities of AsAA across age, stratified by PRS quintile, were estimated using the full model while holding other covariates at reference values (Figure S9).

To contextualize pathways linking social adversity to aortic disease risk, we evaluated secondary associations between the composite SDOH score and cardiometabolic traits and related EHR‐derived conditions (eg, smoking, diabetes, hypertension, hyperlipidemia, peripheral artery disease, coronary artery disease) using regression models with consistent covariate adjustment (Figure S6). As a prespecified sensitivity analysis, we also performed a conventional 70/30 train/test split in which imputation models and elastic‐net SDOH weights were derived exclusively in the training set and association estimates were evaluated in the independent held‐out test set (Data S10; Table S4).

Statistical Analysis

Statistical analyses were performed using R version 4.2.2 in the AoU Researcher Workbench. Key packages included bigrquery for data extraction, glmnet for elastic‐net regression, logistf for Firth regression, pROC for discrimination metrics, missRanger for imputation, and twang for propensity score analysis.

PRS computation was performed using Python version 3.10 with Hail version 0.2.134 and AoUPRS version 0.2.3 (3).

RESULTS

Of 633 547 AoU participants, 143 825 had the required overlap of SDOH survey, sequencing, and linked EHR data (Figure 1); Individually, 259 189 had SDOH survey data, 447 281 had genomic data, and 393 601 had EHR data, with incomplete SDOH survey completion representing the largest source of exclusion. SDOH survey respondents differed systematically from nonrespondents among WGS participants. Before weighting, respondents were more likely to be White individuals, have higher educational attainment, be married, be of non‐Hispanic ethnicity, have higher annual household income, and be older and were less likely to be Black individuals, Hispanic individuals, or in the lowest income and education categories (Figure S1). After excluding 16 733 with missing PGS or ancestry data and 2912 with syndromic features (bicuspid aortic valve, 2254; Ehlers–Danlos syndrome, 591; Marfan syndrome, 74; Turner syndrome, 23; Loeys–Dietz syndrome, <20), 124 180 remained for analysis: 346 cases (0.28%) and 123 834 controls.

Cases were older (70.7±9.0 versus 57.1±16.4 years), more often men (60.4% versus 37.1%), and predominantly White individuals (90.2% versus 72.6%) (Table 1). Hypertension (87.0% versus 43.2%), atherosclerotic cardiovascular disease (68.2% versus 18.6%), and hyperlipidemia (84.1% versus 45.0%) were more prevalent among cases.

Table 1.

Baseline Characteristics

Variable	Cases (n=346)	Controls (n=123 834)
Demographics
Age, y, mean±SD	70.7±9.0	57.1±16.4
Female sex, n (%)	137 (39.6)	77 854 (62.9)
Hispanic ethnicity, n (%)	<20	14 613 (11.8)
Race, n (%)
White	312 (90.2)	89 873 (72.6)
Black	<20	10 703 (8.6)
Asian	<20	3466 (2.8)
American Indian/Alaska Native	<20	829 (0.7)
Native Hawaiian/Pacific Islander	<20	64 (0.1)
Multiple	<20	5107 (4.1)
Other*	<20	1729 (1.4)
Missing	<20	12 063 (9.7)
Clinical risk factors
Body mass index, kg/m2, mean±SD	29.8±6.4	29.5±7.3
Hypertension, n (%)	301 (87.0)	53 484 (43.2)
Diabetes, n (%)	88 (25.4)	22 435 (18.1)
Atherosclerosis/coronary artery disease, n (%)	236 (68.2)	23 012 (18.6)
Hyperlipidemia, n (%)	291 (84.1)	55 780 (45.0)
Smoking (ever), n (%)	50 (14.5)	16 653 (13.4)
Medications
Statin use, n (%)	68 (19.7)	8986 (7.3)
Antihypertensive use, n (%)	87 (25.1)	11 281 (9.1)
Genetic ancestry distribution
European	324 (93.6)	93 506 (75.5)
Admixed American	<20	13 350 (10.8)
African	<20	12 456 (10.1)
East Asian	<20	2666 (2.2)
South Asian	<20	1462 (1.2)
Middle Eastern	<20	394 (0.3)

Values are mean±SD for continuous variables and n (%) for categorical variables. Cell counts <20 are not reportable due to National Institutes of Health All of Us Research Program data privacy policy.

^*Race categories shown represent all categories available in the source data. “Other” was recorded as a single source category and cannot be further disaggregated. Hispanic ethnicity was collected separately and is reported in the table. Case counts fewer than 20 cannot be reported under the NIH All of Us Research Program privacy policy. Therefore, no additional race or ethnicity categories can be specified.

Elastic net regression identified housing insecurity (mean coefficient −0.385), food insecurity (−0.292), loneliness (−0.223), housing instability (−0.188), and perceived stress (−0.124) as primary contributors to the SDOH composite; housing insecurity and food insecurity were selected across all 5 cross‐validation folds (Table S1). Individual measures showed fully adjusted per‐SD ORs of 1.43 (95% CI, 1.19–1.71; P<0.001) for housing insecurity, 1.29 (95% CI, 1.08–1.55; P=0.005) for food insecurity, 1.24 (95% CI, 1.07–1.43; P=0.004) for loneliness, 1.21 (95% CI, 1.03–1.42; P=0.019) for housing instability, and 1.14 (95% CI, 1.00–1.30; P=0.049) for stress (Figure 2A). These individual‐domain findings should be interpreted as secondary descriptive analyses. Housing insecurity, food insecurity, and loneliness remained the most prominent signals and would survive Bonferroni correction for 14 comparisons (threshold P<0.0036).

Each SD increase in the SDOH composite was associated with AsAA in basic models (OR, 1.32 [95% CI, 1.15–1.52]; P<0.001) and after clinical adjustment (OR, 1.21 [95% CI, 1.04–1.40]; P=0.012; Table 2). The PGS showed clear separation between cases (0.7±1.0) and controls (0.0±1.0) (Figure 2D). Per‐SD ORs were 1.97 (95% CI, 1.77–2.19; P<0.001) in basic models and 1.94 (95% CI, 1.74–2.16; P<0.001) after clinical adjustment. Prevalence rose from 0.08% in the lowest polygenic decile to 0.77% in the highest (Figure S8).

Table 2.

Results From Regression Models

Predictor	Covariates	OR per SD (95% CI)	P value
Models including SDOH
SDOH	Basic	1.32 (1.15–1.52)	<0.001
SDOH	+Clinical RF	1.21 (1.04–1.40)	0.012
Models including PRS
PRS	Basic	1.97 (1.77–2.19)	<0.001
PRS	+Clinical RF	1.94 (1.74–2.16)	<0.001
Models including SDOH and PRS with interaction
SDOH	Basic	1.28 (1.11–1.48)	<0.001
PRS		1.95 (1.75–2.17)	<0.001
SDOH×PRS		1.03 (0.91–1.16)	0.624
SDOH	+Clinical RF	1.18 (1.02–1.37)	0.028
PRS		1.92 (1.72–2.14)	<0.001
SDOH×PRS		1.02 (0.90–1.15)	0.752

Correlation between SDOH and PRS: r=−0.002 (95% CI, −0.008 to 0.003). Basic model adjusted for age, sex, and 10 genetic principal components. Clinical RF model additionally adjusted for hypertension, diabetes, body mass index, smoking status, statin use, and antihypertensive use. SDOH and PRS were standardized (mean=0, SD=1). Firth logistic regression was used due to rare outcome. OR indicates odds ratio; PRS, polygenic risk score; RF, risk factors; and SDOH, social determinants of health.

The SDOH composite and PGS were uncorrelated (r=−0.002 [95% CI, −0.008 to 0.003]). In joint models including both exposures, the SDOH composite and PRS were each associated with AsAA: SDOH composite OR, 1.28 (95% CI, 1.11–1.48; P<0.001), PGS OR, 1.95 (95% CI, 1.75–2.17; P<0.001). There was no statistical evidence of PRS×SDOH interaction (OR, 1.03 [95% CI, 0.91–1.16]; P=0.62; Table 2). Cross‐stratification by quintile showed the highest observed prevalence among individuals in the top stratum of both exposures (Figure S7). In a held‐out test‐set sensitivity analysis, effect estimates for both the SDOH composite and PRS remained directionally consistent and of similar magnitude, supporting the internal robustness of the primary findings (Table S4). In a sensitivity analysis adjusting for educational attainment, annual household income, and employment status, the SDOH composite remained associated with AsAA across model specifications, including the joint clinical/full model (socioeconomic status–adjusted OR, 1.19 [95% CI, 1.01–1.40]; P=0.040), and PRS estimates were materially unchanged (Table S5).

Model discrimination improved incrementally (Figure 2B). The basic model (age, sex, genetic principal components) achieved AUC 0.774 (95% CI 0.750–0.799). Addition of the SDOH composite yielded AUC 0.795 (ΔAUC +0.020), PGS AUC 0.816 (ΔAUC +0.042), and clinical risk factors AUC 0.841 (ΔAUC +0.067). The full model achieved AUC 0.870 (95% CI, 0.851–0.889; ΔAUC +0.096). After Firth logistic regression with intercept correction, mean predicted risk was aligned with the observed event proportion across nested models (Table S2). Brier scores changed only minimally across models, while grouped‐decile calibration plots showed residual imperfect absolute‐risk calibration, particularly in the full model, despite improved discrimination (Table S2; Figure S10). Decision curve analysis showed the greatest net benefit for the full model across the evaluated threshold‐probability range (Figure S11).

The SDOH composite was associated with smoking (OR, 1.39 [95% CI, 1.35–1.43]), diabetes (OR, 1.30 [95% CI, 1.25–1.35]), hypertension (OR, 1.18 [95% CI, 1.14–1.22]), and peripheral artery disease (OR, 1.08 [95% CI, 1.05–1.11]), consistent with overlap between social adversity and cardiometabolic risk burden (Figure S6). Inverse probability of response weighting reduced measured covariate imbalance among observed sociodemographic variables (Figures S1–S3), and clinical risk factor associations were directionally consistent between the analytic cohort and broader eligible population (Figures S4 and S5). Inverse probability–weighted baseline characteristics (Table S3) showed only modest shifts in demographic composition relative to the unweighted analytic sample.

DISCUSSION

In this analysis of 124 180 National Institutes of Health AoU participants, we report 3 principal findings. First, multidomain social adversity was associated with EHR‐documented, nonsyndromic AsAA, including after adjustment for conventional clinical risk factors (composite OR, 1.21 after clinical adjustment; OR, 1.28 in joint models including the PGS). Second, a diameter‐derived PGS demonstrated a gradient of risk for diagnosis‐ascertained aneurysm and remained associated in fully adjusted models (OR, 1.95 per SD in joint models). Third, the SDOH composite and PRS were minimally correlated and no statistically significant PRS×SDOH interaction was observed (interaction OR, 1.03 [95% CI, 0.91–1.16]; P=0.62); however, given the modest case count, modest synergistic effects cannot be excluded. Integrating both domains improved discrimination beyond conventional clinical covariates (full‐model AUC, 0.870 [95% CI, 0.851–0.889]).

The SDOH signal in this study was concentrated in domains of material hardship and social connection, with housing insecurity, food insecurity, and loneliness among the strongest contributors to the composite and among the measures most strongly associated with AsAA. These associations are clinically plausible and may reflect overlapping pathways relevant to thoracic aortic disease, including greater cardiometabolic risk burden, adverse health behaviors, chronic stress physiology, and barriers to preventive care and longitudinal follow‐up. ⁶ , ⁷ In addition, adverse social context may shape disease progression and the timing of diagnosis through differences in blood pressure control, medication adherence, and consistency of surveillance. ¹⁸ , ¹⁹ , ²⁰ Consistent with this framework, the composite score was associated with smoking and cardiometabolic comorbidities in secondary analyses, supporting the interpretation that social adversity may affect the downstream risk profile in which aneurysm is recognized and managed. Although the per‐SD effect size of the clinically adjusted composite was modest (OR, 1.21), such effect sizes are common for multidomain social adversity measures in cardiovascular epidemiology and may become clinically meaningful across the full distribution of exposure burden and when considered jointly with genetic susceptibility (Figure S7). From a prevention standpoint, these findings motivate attention to unmet social needs and social isolation as risk context that may complement conventional risk factor modification, particularly in nonsyndromic patients who lack guideline‐based triggers for early imaging or structured surveillance. ³ , ²¹

More broadly, these findings may also be viewed through an “environmentalomics” or exposome framework, in which social, behavioral, environmental, and biological exposures accumulate over time and may correlate with vascular risk. ²² In the present study, SDOH measures should be interpreted as high‐level markers of exposure context rather than direct measures of biological embedding or causal pathways. Future cohorts integrating longitudinal SDOH, geospatial, imaging, and molecular data will be needed to test whether cumulative exposures are associated with aortic remodeling trajectories and whether such associations vary by inherited susceptibility.

The PRS used in this study (PGS004927) derives from the AORTA Gene Score reported by Pirruccello et al, ²³ which was developed using ascending aortic diameter, explained ≈13% of variance in that quantitative trait and was externally validated across multiple cohorts including AoU. In the present analysis, this PGS demonstrated a strong and internally consistent gradient of risk for diagnosis‐ascertained, nonsyndromic AsAA. It clearly separated cases from controls (0.7±1.0 versus 0.0±1.0), was associated with AsAA in both basic and clinically adjusted models (per‐SD OR, 1.97 and 1.94, respectively) and showed a monotonic increase in disease prevalence from 0.08% in the lowest PRS decile to 0.77% in the highest. These findings extend the prior imaging‐based AORTA Gene Score work by demonstrating that this diameter‐derived score also stratifies risk for a clinically recognized, EHR‐ascertained aneurysm phenotype, while also evaluating its relationship to multidomain SDOH. At the same time, because the PRS was originally derived in predominantly European‐ancestry discovery cohorts and PRS performance can vary across ancestries, rigorous multiancestry validation and improved cross‐ancestry modeling remain important priorities, even with ancestry‐standardized scoring in this study. ²⁴

Current aortic guidelines emphasize family history and imaging‐based assessment to guide surveillance and prophylactic intervention, but this approach can miss many individuals with nonsyndromic disease who lack a clear trigger for early evaluation. ³ In this context, our analyses suggest that social and genetic information may provide nonredundant risk stratification beyond conventional clinical covariates: Compared with a basic model (AUC, 0.774), adding the SDOH composite improved AUC to 0.795; the PGS to 0.816; clinical risk factors to 0.841; and the full model incorporating social, genetic, and clinical domains achieved AUC 0.870. Future work should focus on validation in additional cohorts, evaluation for incident aneurysm and dissection outcomes, imaging‐confirmed aortic dimensions, improved multiancestry modeling to assess transportability, and further calibration assessment before clinical absolute risk prediction.

Limitations

Several limitations warrant consideration. First, the cross‐sectional design limits causal interpretation, and SDOH measures assessed at survey completion may not reflect the timing or duration of exposures during aneurysm development. Second, inclusion required overlap of survey, genomic, and linked EHR data, with SDOH survey completion representing the largest source of attrition; although inverse‐probability‐of‐response weighting reduced measurable imbalance (Figures S1–S5; Table S3), residual selection bias and differential health care ascertainment remain possible. Third, the analytic sample was predominantly White individuals, which may limit generalizability to populations with different social risk distributions or aneurysm epidemiology. Fourth, the PRS was derived from prior discovery efforts enriched for European ancestry, and performance may differ across ancestry groups despite ancestry‐stratified standardization. Fifth, our outcome relied on EHR‐based phenotyping, which may underascertain asymptomatic disease and preferentially capture individuals with greater health care contact; we also lacked imaging‐derived aortic diameters, limiting assessment of subclinical dilation. Sixth, although the main‐effect estimates were reasonably precise, the modest number of AsAA cases (n=346) limited power for subgroup and especially interaction analyses, so modest synergistic effects between social and genetic risk cannot be excluded. Seventh, although Firth logistic regression with intercept correction correction aligned mean predicted probabilities with the observed event proportion, calibration slope remained imperfect.

Finally, despite cross‐fitting and supportive split‐sample sensitivity analyses for the composite SDOH score (Table S4), validation in additional cohorts will be important to establish the broader transportability of the learned weights and prediction performance. AoU is currently the only large‐scale program linking individual‐level SDOH surveys, whole‐genome sequencing, and longitudinal EHR data, a unique infrastructure that enabled this first integrated analysis of social and genomic risk for nonsyndromic AsAA.

CONCLUSIONS

In a large National Institutes of Health AoU cohort with linked survey, genomic, and EHR data, multidomain social adversity and a diameter‐derived PGS were associated with nonsyndromic ascending aortic aneurysm and together improved risk discrimination beyond conventional clinical covariates. These results highlight the relevance of both social and inherited susceptibility to thoracic aortic disease risk and support further validation of integrated social–genetic risk assessment in prospective, imaging‐based, and multiancestry settings. More broadly, this work supports future studies using exposome‐oriented frameworks to evaluate how social, environmental, and inherited risk markers jointly relate to vascular disease.

Sources of Funding

None.

Disclosures

None.

Supporting information

Tables S1–S5

Figure S1–S11

References 24–33

AHA Journals Racial and Ethnic Disparities Reporting Guidelines

STROBE Checklist