Sociodemographic and Clinical Predictors of Mortality in Adults With Congenital Heart Disease.
Mar 25, 2026DOI: 10.1161/jaha.125.045411Journal ArticleAHA
Written by: Donnchadh O'SullivanSahar SamimiScott AnjewierdanChung-Il WiKeila N LopezAlexander EgbeDave WatsonC Charlie JainYoung J JuhnWilliam R MirandaHeidi M ConnollyMalini MadhavanLuke J Burchill
This is a single‐center retrospective cohort study designed to evaluate the association of HOUSES with ACHD mortality. This study was approved by the Mayo Clinic Institutional Review Board (20‐007695) with waiver of informed consent. This research was conducted in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology guidelines for observational cohort studies. Analytic methods, including R notebooks, will be available from the corresponding author upon reasonable request; data may be shared on reasonable request, subject to data sensitivity and institutional data governance review.Patients with ACHD were identified through the previously described MACHD (Mayo ACHD Registry), which includes patients receiving ambulatory or hospital care at the Mayo Clinic ≥18 years of age.19 ACHD inclusion criteria and methods for the creation the MACHD Registry are described in Data S1. This enterprise cohort spans 3 Mayo Clinic sites (Rochester, MN; Jacksonville, FL; Scottsdale, AZ), supporting geographic diversity across 3 states within 1 integrated system. To ensure reliable HOUSES index data retrieval and calculation we included patients whose home address was documented from a hospital or clinic visit between January 1, 2004 to December 31, 2023. Individuals were excluded if they had incomplete address data preventing calculation of the HOUSES index or if mortality status could not be ascertained. Follow‐up was tracked from the date of the first ACHD clinic visit until death or last known contact, with administrative censoring on December 31, 2024.Sex, race, and ethnicity were obtained from the electronic health record as documented during clinical registration. Race and ethnicity are social constructs rather than biological variables and were included to characterize the study population and examine potential disparities in health outcomes. The electronic health record recorded race categories including White or Caucasian, Black or African American, Asian, American Indian or Alaska Native, Native Hawaiian or Pacific Islander, multiracial, other race, and unknown. For statistical analyses, race was categorized as White, Non‐White, or Unknown. The Non‐White category included individuals recorded as Black or African American, Asian, American Indian or Alaska Native, Native Hawaiian or Pacific Islander, multiracial, or other race in the electronic health record. The Unknown category included records labeled as unknown, missing, or not reported. Ethnicity was categorized as Hispanic or Latino, Not Hispanic or Latino, or Other/Unknown. The Hispanic or Latino category included individuals recorded in the electronic health record as Mexican, Central American, South American, Cuban, Puerto Rican, other Spanish culture of origin regardless of race, Hispanic, or Latino. The Other/Unknown category included records labeled as unknown, choose not to disclose, unable to provide, missing, or other ethnicity in the electronic health record. Primary language spoken was based on self‐report in the electronic health record and summarized as English or non‐English reflecting real‐world interpreter needs and potential communication barriers. Insurance status was extracted from Mayo Data Explorer and available insurance status was categorized as public, private, or other. Residential addresses were used to define rurality according to Census derived Rural–Urban Commuting Area codes. Primary codes (1–10) delineate metropolitan, micropolitan, small town, and rural commuting areas based on the size and direction of the primary (largest) commuting flows. Rural–Urban Commuting Area primary codes 4 to 10 are rural.
HOUSES index is a validated individual‐level housing SES measure based on 4 housing features (number of bedrooms, number of bathrooms, square footage, and assessed building value) linked to patient's reported address. To formulate the HOUSES index, each housing feature was first transformed to a standardized Z score, and then the Z scores of the 4 housing features were summed to create a composite score. The HOUSES index is standardized within county and calendar year based on available real property data, such that values reflect relative socioeconomic position within each county‐year. The HOUSES index was calculated and analyzed based on residence at the first clinic.5 HOUSES was stratified into quartiles (Q1 to Q4), with Q1 representing the lowest SES status.12
All‐cause mortality, ascertained by review of medical records and the Accurint mortality database, served as the primary outcome. Mortality was ascertained using the MACHD registry and Accurint database, which includes >33 billion records from >8800 different data sources.20 We selected all‐cause mortality as the primary outcome because it is the most consistently and comprehensively ascertainable end point over the 20‐year period across sites.
Each clinical comorbidity was defined as present or absent based on clinical diagnoses documented at the index visit. The CHD group classification encompassed a diverse range of congenital heart defects, each categorized based on the following anatomical and physiological characteristics: (1) right heart/conotruncal lesions include tetralogy of Fallot, Ebstein anomaly, valvular pulmonary stenosis, pulmonary atresia with intact ventricular septum, double‐chambered right ventricle, truncus arteriosus, and various transposition of the great arteries (TGA) variants, such as TGA‐arterial switch operation, TGA‐Rastelli, and TGA‐double‐outlet right ventricle post‐Rastelli repair; (2) left heart lesions primarily comprise coarctation of the aorta, mitral stenosis (MS), subaortic stenosis, aortic stenosis, and supravalvular aortic stenosis; (3) noncyanotic shunt lesions include partial anomalous pulmonary venous return, atrial septal defect, ventricular septal defect, and atrioventricular canal defects; (4) cyanotic heart disease is represented by Eisenmenger syndrome and unrepaired or palliated single‐ventricle physiology; (5) systemic right ventricle conditions encompass congenitally corrected TGA and TGA repaired with a Mustard/Senning procedures; (6) those with single ventricle physiology who had undergone Fontan palliation and (7) an other category that includes patent ductus arteriosus and cor triatriatum.
Clinical heart failure was defined as the simultaneous presence of New York Heart Association class II, III, or IV symptoms and concurrent prescription of loop diuretics at the index visit. ACHD severity was classified according to defect complexity (mild, moderate, complex) in keeping with Bethesda criteria.21
For descriptive analyses, HOUSES quartiles were compared to identify differences in demographics, comorbidities, and patterns of CHD.
Continuous variables were expressed as medians with interquartile ranges and compared across HOUSES quartiles using the Kruskal–Wallis test. Categorical variables were described as frequencies (percentages), and overall group differences were assessed using chi‐square or Fisher's exact tests as appropriate. Variables chosen for subsequent multivariate models included those that differed significantly by HOUSES quartile (P<0.05), were significant in univariate survival analyses (P<0.05), or were deemed clinically relevant, while also controlling for collinearity. To account for correlation among individuals residing in the same county and residual unmeasured county‐level heterogeneity (eg, geographic differences in care access/referral patterns), county of residence was additionally modeled as a shared frailty (county‐level random effect) term in Cox models. Cox proportional hazards models were then fitted to estimate hazard ratios (HR) with 95% CIs. In the multivariable model, age at first visit and year of birth were examined for collinearity to avoid redundancy. Age at first ACHD visit represents timing of entry ACHD clinic and the evolving care paradigms across the study span. Multivariable modeling used a complete‐case cohort for all covariates included in the model and no imputation was performed. Thus, variables such as race and ethnicity had no missingness within the modeled cohort. The model's discriminative ability was assessed by concordance statistic and was fitted using the Cox proportional hazards framework. Additional assessments of model performance included time‐dependent receiver operating characteristic analysis at 1, 3, and 5 years using time‐dependent receiver operating characteristic methods applied to the linear predictor from the single fitted Cox model. All statistical analyses were performed using R version 4.1.0 (R Foundation for Statistical Computing, Vienna, Austria). Throughout all analyses, a 2‐sided P<0.05 was considered statistically significant. Results are reported as point estimates (HR) with 95% CI.
RESULTS
A total of 7315 patients with ACHD were identified. There were 1181 patients with unavailable HOUSES that were excluded from the study resulting in a total cohort of 6134 patients with ACHD. The mean age at the first ACHD clinic visit was 38 (SD 16) years. Patients were stratified by the HOUSES index into 4 groups: Q1 (lowest SES: n=1032), Q2 (n=1290), Q3 (n=1613), and Q4 (highest SES: n=2199). Detailed demographic characteristics, comorbidity profiles, and CHD types for the entire cohort by HOUSES quartile were presented in Table 1. Although CHD complexity and anatomic subgroup distributions did not differ significantly across HOUSES quartiles (P=0.28 for CHD groups and P=0.67 for CHD severity), significant disparities in certain comorbid conditions were observed. For example, the prevalence of obesity was lower in Q4 (19.2%) compared with Q1 (23.7%, P<0.001), and frequencies of atrial arrhythmias and atrial fibrillation were lower in Q4 (21.1% and 13.6%, respectively) than in Q1 (25% and 15.8%, respectively; P=0.01 and P=0.002). As expected, there were more individuals with public insurance in Q1 compared with Q4 (Table 1). A geospatial ZIP‐level map demonstrates patient dispersion across the United States and highlights density surrounding Mayo enterprise sites, stratified by urban versus rural classification (Figure 1) and by insurance class (Figure S1).
Basic Characteristics of Study Cohort by HOUSES Index
Demographic, socioeconomic, and clinical characteristics of the study cohort by HOUSES Index quartiles (Q1: lowest SES to Q4: highest SES). Continuous variables are presented as median (interquartile range) and compared using the Kruskal–Wallis test, and categorical variables are reported as n (%) and analyzed using the χ 2 or Fisher's exact test. Column denominators reflect HOUSES group distribution, sample sizes for individual variables may vary due to missing data. # <0.05 was the p‐value cut off.
CHD indicates congenital heart disease; HOUSES, housing‐based socioeconomic status index; SES, socioeconomic status; and VT, ventricular tachycardia.
The Other category included records labeled as unknown, choose not to disclose, unable to provide, missing, or other ethnicity or race in the electronic health record.
The Non‐White category included individuals recorded as Black or African American, Asian, American Indian or Alaska Native, Native Hawaiian or Pacific Islander, multiracial, or other race in the electronic health record.
RUCA indicates Rural–Urban Commuting Area.
In univariate analysis (Table 2), patients in the lowest HOUSES quartile (Q1) did not exhibit increased mortality compared with those in the highest quartile (Q4) (HR, 1.12 [95% CI, 0.90–1.40], P = 0.31). As in prior analyses, older age at first ACHD clinic visit remained linked to worse outcomes (HR, 1.04 per year [95% CI, 1.04–1.05], P<0.001), whereas a later year of birth was protective (HR, 0.96 per year [95% CI, 0.96–0.96], P<0.001). The most powerful predictors of death were unrepaired cyanotic congenital heart disease with a 5.7× higher risk of mortality, and Fontan palliation with a 2.6× higher risk of mortality. Additional factors associated with higher mortality included complex CHD severity (HR, 1.87 [95% CI, 1.61–2.63], P<0.001), systemic right ventricle (HR, 1.42 [95% CI, 1.08–1.86], P=0.012), male sex (HR, 1.20 [95% CI, 1.05–1.37], P=0.007), “other” ethnicity (HR, 1.29 [95% CI, 1.20–1.75], P<0.001), public insurance (HR, 2.43 [95% CI, 2.09–2.82], P<0.001), and non‐English language (HR, 1.48 [95% CI, 1.28–1.71], P<0.001). Cardiac rhythm disturbances also conferred excess risk, notably atrial arrhythmias (HR, 2.51 [95% CI, 2.19–2.88], P<0.001), atrial fibrillation (HR, 2.89 [95% CI, 2.50–3.33], P<0.001), and sustained ventricular tachycardia (HR, 1.91 [95% CI, 1.33–2.74], P<0.001). Rural residence showed a borderline association with mortality (HR, 1.16 [95% CI, 0.99–1.34], P=0.054).
Univariate Cox Regression Analysis
Univariate Cox regression analysis. Hazard ratios with 95% CIs and P values are reported. # <0.05 was the p‐value cut off.
CHD indicates congenital heart disease; and RH, right heart.
The Other category included records labeled as unknown, choose not to disclose, unable to provide, missing, or other ethnicity or race in the electronic health record.
The Non‐White category included individuals recorded as Black or African American, Asian, American Indian or Alaska Native, Native Hawaiian or Pacific Islander, multiracial, or other race in the electronic health record.
The multivariate model included 5909 patients in total after accounting for missing observations, with 711 mortality events (Table 3). After adjustment for CHD complexity, sociodemographic factors, and key comorbidities, HOUSES quartile was not independently associated with mortality, and no dose–response gradient was observed across quartiles. Age at first visit remained incrementally associated with mortality (HR, 1.05 per year [95% CI, 1.04–1.05], P<0.001). Cyanotic congenital heart disease (HR, 5.70 [95% CI, 2.90–11.23], P<0.001) and Fontan palliation (HR, 4.98 [95% CI, 2.56–9.70], P<0.001) remained the strongest independent predictors of mortality. Among sociodemographic variables, rural residence (HR, 1.25 [95% CI, 1.06–1.47], P=0.007), public insurance (HR, 1.60 [95% CI, 1.35–1.90], P<0.001), non‐English language (HR, 1.75 [95% CI, 1.38–2.23], P<0.001), and male sex (HR, 1.19 [95% CI, 1.03–1.39], P=0.02) were associated with increased mortality risk, whereas “other” insurance types were associated with lower risk (HR, 0.53 [95% CI, 0.36–0.76], P<0.001). Atrial fibrillation (HR, 1.39 [95% CI, 1.09–1.78], P=0.007) and sustained ventricular tachycardia (HR, 1.71 [95% CI, 1.16–2.52], P=0.007) also conferred elevated mortality risk. After adjustment, CHD severity category, systemic right ventricle, race, and ethnicity no longer reached statistical significance.
Multivariate Cox Regression Analysis
Multivariate Cox regression analysis. Hazard ratios with 95% CIs and P values are reported. # <0.05 was the p‐value cut off.
CHD indicates congenital heart disease; and HR, hazard ratio.
The Other category included records labeled as unknown, choose not to disclose, unable to provide, missing, or other ethnicity or race in the electronic health record.
The Non‐White category included individuals recorded as Black or African American, Asian, American Indian or Alaska Native, Native Hawaiian or Pacific Islander, multiracial, or other race in the electronic health record.
The final multivariate Cox regression model demonstrated robust discriminative ability with a concordance statistic of 0.77. The forest plot (Figure 2) summarizes the relative contributions of the covariates in the final model. Time‐dependent receiver operating characteristic curve analyses showed areas under the curve of 0.82 at 1 year, 0.78 at 3 years, and 0.79 at 5 years (Figure 3), indicating strong performance in distinguishing patients at higher versus lower risk of mortality.
Points indicate adjusted hazard ratios and horizontal bars indicate 95% CIs (x axis on log scale). Significant predictors included cyanotic heart disease, Fontan palliation, non‐English language, sustained ventricular tachycardia, public insurance, atrial fibrillation, rural residence, and male sex. Significance levels are denoted as follows: P < 0.05; P < 0.01; ***P* < 0.001; CHD indicates congenital heart disease; HOUSES, housing‐based socioeconomic status; NS, not significant; and RUCA, Rural–Urban Commuting Area.
AUC indicates area under the curve; and ROC, receiver operating characteristics.
DISCUSSION
This research advances our understanding of ACHD health outcomes by integrating clinical and SDOH into a unified model of mortality risk. We demonstrate for the first time that ACHD mortality is strongly influenced by clinical and SDOH indicators, with functional barriers to care including language, insurance, and geography, being especially important for mortality risk, independent of SES. In the adjusted analyses, public insurance, rural residence, non‐English primary language, male sex, age at first ACHD clinic visit, ACHD anatomic/clinical severity, atrial fibrillation, and sustained ventricular tachycardia were independently associated with mortality, whereas HOUSES did not show an independent association after adjustment. The findings highlight a critical opportunity to improve survival in patients with ACHD by integrating SDOH into risk stratification models and implementing targeted support to help at‐risk individuals overcome barriers to care.
Our findings align with the World Health Organization's Conceptual Framework for Action on the Social Determinants of Health, which identifies upstream structural determinants (such as SES) as operating through downstream determinants to shape health outcomes. Downstream determinants include social support, biological, behavioral, and functional barriers to health care access, all of which directly shape an individual's ability to navigate and benefit from health systems. In our study, functional barriers including non‐English language, public insurance, and rural residence, were strongly associated with ACHD mortality whereas individual‐level SES, as measured by the HOUSES index, was not. Although SES remains important as an upstream structural determinant of health outcomes, the findings of this research suggest downstream functional barriers to health care access may be exerting a more immediate and measurable impact on survival in this population.22
Prior research has demonstrated a range of SDOH indicators as being important for ACHD health outcomes. Insurance status, education level, income, race or ethnicity, and geographic access to specialty care have each been associated with disparities in ACHD health outcomes including hospital readmission,23 quality of life,24 higher resource use,25 loss to follow‐up26 and surgical complications.27 In this study, patients with ACHD with lower SES were more likely to have a higher prevalence of comorbidities, particularly arrhythmias and obesity. These findings mirror earlier studies that identify socioeconomic factors as predictors of increased disease burden in ACHD.28, 29, 30
The association between public insurance and mortality may reflect a composite of administrative/network barriers to subspecialty access, travel/transportation constraints, disability status, and reduced continuity, aligning with the access‐barrier framing. Although our study cannot directly quantify intermediate mechanisms (eg, education level, health‐system navigation), the non‐English signal plausibly captures communication and care‐coordination barriers.
Despite the clear importance of SDOH for ACHD outcomes, the relationship between SDOH and ACHD mortality has not been explored. Most studies of SDOH in ACHD have not included mortality as a primary outcome or have been limited by data sources that typically include only 1 or 2 SDOH indicators, most commonly insurance status and income.31 Meanwhile, studies of ACHD mortality have largely ignored the influence of SDOH when developing risk scores and models for predicting ACHD outcomes. These approaches have led to a critical gap in our understanding of how multilevel structural and social factors contribute to mortality risk in populations with ACHD and limit the ability of current models to accurately predict outcomes or guide equitable interventions for at‐risk patients.
Predictors of outcomes in CHD are complex and multifactorial, involving interplay between congenital lesion type, comorbidities, and social determinants that can accumulate over time. Cyanotic heart disease and Fontan circulation were potent predictors of mortality in this study, aligning with prior reports linking ACHD complexity to adverse outcomes.32 The persistence of strong associations between clinical factors, such as cyanotic heart disease and Fontan circulation, and mortality, even after adjustment for sociodemographic variables, underscores the continued need for specialized, lifelong surveillance and tailored care strategies in high‐risk populations with ACHD32 Importantly, however, several nonanatomic variables, including rural residence, public insurance, and non‐English language also emerged as independent predictors of mortality. Patients affected by these factors were at significantly elevated risk, echoing prior literature that highlights the detrimental impact of limited health care access, language barriers, and financial constraints on long‐term outcomes.33, 34 Notably, the combined impact of these modifiable nonanatomic factors was comparable to that of complex congenital anatomy, suggesting that SDOH may be as consequential as structural heart disease for shaping prognosis. This highlights a critical opportunity to enhance clinical care by integrating SDOH into standard risk stratification frameworks, supporting multidisciplinary, integrated approaches that address both biological complexity and the structural inequities that persist in the care of patients with ACHD.34
Several limitations merit consideration. Our single‐center design may reflect referral patterns favoring more severe cases or higher resource availability, potentially limiting generalizability. Although the ZIP‐level map demonstrates broad geographic representation, referral and ascertainment bias inherent to a tertiary‐care system may influence observed associations; external validation in multicenter cohorts beyond the Mayo system will be important to confirm generalizability. Although HOUSES is a validated measure of individual‐level SES, it captures only a snapshot of housing‐related factors and does not account for changes over time or broader community elements. Insurance type was categorized as private, public (eg, Medicare/Medicaid and other government programs), and other. The “other” category mostly included records in which a definitive private/public label was not recoverable during curation from Mayo Data Explorer. Validation on a random sample demonstrated a heterogeneous mix of private and public coverage in this group; therefore, “other” insurance group should be interpreted as a data‐labeling category, not a coherent clinical subgroup. Variables were modeled at time of entry to clinic; we lacked standardized longitudinal capture of time‐varying comorbidities and use across eras. The study cohort was mostly White (78.7%). The lack of association between race and mortality may reflect limited racial diversity in the cohort but could also suggest that other measures of structural disadvantage, such as language, insurance status, and rurality, may be more accurate indicators of barriers to care influencing survival in populations with ACHD. Causal mechanisms (eg, education, alternative therapy use, system capacity for interpreter services) were not directly measurable and are proposed for future study. Minnesota Medicaid expansion and institutional charity care could attenuate SES gradients; however, insurance type, language, and rurality remained independently associated with mortality after adjustment. The performance of our final multivariate model reinforces the synergistic importance of clinical and social determinants in predicting mortality. Although our internal validation metrics are promising, future work should focus on external validation to confirm the model's generalizability and clinical utility.
Conclusions
By evaluating multiple, distinct SDOH dimensions—including individual SES, language, insurance status, and rurality—alongside clinical complexity, this research highlights clinical risk and SDOH as not being separate dimensions but rather highly interconnected for shaping mortality risk in patients with ACHD. We present an integrated clinical‐SDOH risk model that allows for clearer insights into which clinical and social factors are driving mortality risk in ACHD. We provide evidence for functional access barriers (language, insurance, geography) being more strongly associated with mortality than SES alone. This suggests that modifiable system‐level barriers, rather than SES status, may be key intervention points a critical insight for health equity efforts in ACHD. Future research should include longitudinal assessments of SES/SDOH and interventions directed toward improving access to specialized care for the most vulnerable subgroups with ACHD, ultimately leading to improved survival in this growing patient population.
Sources of Funding
The Mayo Clinic Adult Congenital Heart Disease Registry is supported by the Al‐Bahar Research Grant (Mayo Clinic, Rochester, MN).
Disclosures
None.
Supporting information
Data S1
Figure S1
SS
Dr. Sahar Samimi
Internal Medicine · General Internal MedicineBaylor College of Medicinesahar