Multidomain risk profiles and documented polyvascular involvement across atherosclerotic vascular beds Authors: Kelvin Supriami, Alyssa M Flores, Yang Sui, So Mi Jemma Cho, Aniruddh P Patel, Yunfeng Ruan, Krishna G Aragam, Pradeep Natarajan, Patrick T Ellinor, Akl C Fahed Published: 2026 ABSTRACT AbstractBackgroundAtherosclerotic cardiovascular disease (ASCVD) manifests heterogeneously across coronary (CAD), cerebrovascular (CeVD), and peripheral artery (PAD) beds.ObjectivesTo evaluate sociodemographic, clinical, lifestyle, and genetic determinants of incident ASCVD, vascular bed localization, and subsequent documented polyvascular involvement.MethodsWe analyzed 468,266 UK Biobank participants free of ASCVD at baseline. Risk factor associations were examined using multivariable Cox models, machine learning-derived SHapley Additive Explanations (SHAP), multinomial logistic regression, and discrete-time multistate models of clinically documented ASCVD states.ResultsAmong participants (mean age 57 years, 56% female) followed for 13 years, 2.8% developed CAD, 1.9% CeVD, and 1.5% PAD. Risk domains showed distinct relative associations across vascular beds. PAD onset was most strongly associated with inflammation, metabolic dysfunction, and smoking, including hsCRP (HR per SD 1.22, 95% CI 1.19–1.26); CAD was more strongly associated with LDL-C (HR 1.24, 95% CI 1.21–1.26) and CAD polygenic risk score (PRS). CeVD showed overlapping features, with additional contribution from socioeconomic deprivation. Healthy sleep and physical activity were protective across outcomes. SHAP analyses yielded similar profiles. In multinomial models, inflammation, glycemic status, and smoking were enriched in non-CAD phenotypes, whereas LDL-C and CAD PRS were more specific to CAD. PAD was most frequently followed by subsequent documented polyvascular ASCVD and mortality. In multistate models, PRSs were primarily associated with early transitions from healthy to incident ASCVD or death, whereas later documented transitions and mortality after ASCVD were more strongly linked to clinical and lifestyle factors.ConclusionsASCVD shows distinct risk profiles across vascular beds and disease stages. Genetic factors were more strongly associated with early incident ASCVD, while clinical and lifestyle factors predominated in subsequent documented polyvascular involvement and later mortality.Keywords: Atherosclerosis, Coronary artery disease, Cerebrovascular disease, Peripheral artery disease, Disease progression, Risk factors Nonstandard Abbreviations and Acronyms atherosclerotic cardiovascular disease coronary artery disease cerebrovascular disease high-sensitivity C-reactive protein ischemic stroke lipoprotein(a) peripheral arterial disease polygenic risk score SHapley Additive Explanations Townsend Deprivation Index Introduction Atherosclerosis underlies most cardiovascular diseases but manifests heterogeneously across vascular territories [1]. Coronary artery disease (CAD), cerebrovascular disease (CeVD), and peripheral artery disease (PAD) share risk factors but differ in their biological pathways, clinical presentation, and prognosis [[2], [3], [4], [5]]. Some individuals develop isolated CAD, others manifest with PAD or CeVD [6], while a subset progresses to polyvascular involvement [[7], [8], [9]] over the life course. As precision prevention becomes central to cardiovascular medicine [10], understanding how diverse risk factors contribute to disease development across arterial beds is increasingly important. Contemporary cardiovascular risk models often assume uniform underlying atherosclerosis across vascular beds, overlooking that different arterial beds may be driven by unique biological and environmental factors [[11], [12], [13]]. Emerging evidence reveals meaningful variation in lipid, inflammatory, thrombotic, and vascular remodeling pathways between coronary, peripheral, and cerebral arteries [3,14,15], motivating closer examination of how established risk factors operate across vascular beds. While most studies focus on predicting incident events, far less is known about what governs subsequent clinically documented involvement of additional vascular territories or the accumulation of polyvascular disease burden over time. The clinical impact is substantial [16]: individuals with polyvascular involvement face markedly higher risks of cardiovascular events and mortality than those with single-territory disease [[17], [18], [19], [20]]. Yet, no framework currently guides prevention based on patterns of documented vascular involvement. Although individual risk factors for ASCVD are well-established, prior studies have typically examined them within single vascular beds or without accounting for disease stage. A unified, comparative framework that quantifies how multidomain risk factors differentially share vascular localization and subsequent accumulation of clinically observed disease burden is lacking. We hypothesized that while ASCVD arises from overlapping drivers, the relative influence of sociodemographic, clinical, lifestyle, and genetic factors varies across arterial beds and stages of disease. We therefore aimed to (1) compare the strength of multidomain risk factors for documented ASCVD onset across beds; (2) evaluate enrichment of risk factors by vascular localization; and (3) examine associations between baseline multidomain risk profiles and subsequent clinically documented involvement of additional vascular beds and mortality. This framework advances understanding of ASCVD heterogeneity and may guide future work aimed at tailoring prevention to vascular bed and disease stage. Methods Study population and design The UK Biobank is a prospective population-based cohort study that enrolled approximately 500,000 individuals aged 40 to 69 years old across the United Kingdom between 2006 and 2010 [21] At enrollment, participants underwent comprehensive baseline assessments, including anthropometric measurements, biospecimen collection, and detailed questionnaires. For this analysis, we excluded individuals with prevalent CAD, CeVD, or PAD at baseline to establish an incident cohort (Supplemental Fig. 1). Follow-up was conducted through linkage to National Health Service (NHS), with outcomes available through 2022. Participants contributed time at risk from enrollment until incident events, subsequent vascular transitions, death, or censoring at the end of follow-up. All participants provided informed consent. Analyses were performed under UKB application 7089 and approved by the Mass General Brigham Human Research Committee Institutional Review Board (2018P002292). UKB data are publicly available (https://www.ukbiobank.ac.uk/); analysis codes are available upon reasonable request. Ascertainment of sociodemographic, clinical, and lifestyle risk factors Baseline sociodemographic variables included age, sex, race/ethnicity, and socioeconomic status measured using the Townsend Deprivation Index. Clinical characteristics included body mass index (BMI), systolic blood pressure (SBP), lipid profile, lipoprotein(a) (Lp(a)), glycated hemoglobin (HbA1c), high-sensitivity C-reactive protein (hsCRP), and estimated glomerular filtration rate (eGFR). Lifestyle exposures were derived from questionnaires and included smoking status, physical activity, diet quality, and sleep duration. Definitions, measurement protocols, and adjustment procedures are described in the Supplemental Methods. Construction of polygenic risk scores We constructed polygenic risk scores (PRSs) for CAD, Ischemic Stroke (IS), and PAD. The CAD PRS was computed from GPSmult (accession ID, PGS003725), a multi-ancestry PRS derived from CARDIoGRAMplusC4D 1000 Genomes-based genome-wide association studies (GWAS) with 1,296,172 variants [22]. PAD PRS was derived from recently validated genome-wide polygenic scores using multi-ancestry GWAS data for PAD and other 14 related traits from over 2 million individuals [23]. Lastly, IS PRS was calculated from publicly available weights for PRS_IS (accession ID, PGS000911) with over 530 thousand genetic variants from 446,696 individuals in 11 cohorts [24]. To minimize overfitting, we excluded 111,451 participants who were part of the GPSmult development set as well as 96,176 individuals whose data were used to train the PAD PRS. All polygenic risk scores were further adjusted for age at enrollment, sex, and the first 10 principal components (PCs) of genetic ancestry and normalized within the study population. Full details of UKB genotyping, quality control, and imputation are provided in the Supplemental Methods. Ascertainment of atherosclerotic outcomes To identify incident clinically observed atherosclerotic events, we utilized a combination of self-reported information and linked electronic health records (EHRs) available in the UK Biobank through March 2020. Specifically, we extracted data from the self-reported questionnaires completed by participants at baseline and the Hospital Episode Statistics (HESIN) master table, which comprehensively captures inpatient hospital encounters, including dates of admission and discharge, primary and secondary diagnoses, and procedural codes. We defined three major categories of ASCVD: CAD, PAD, and CeVD. To ensure a prospective assessment of risk, we limited our analyses to individuals free of clinical atherosclerotic disease at baseline. All participants with a history of CAD, PAD, or CeVD prior to their enrollment date were excluded from the incident cohort. Case definitions were based on a combination of self-reported diagnoses and procedures collected at baseline interviews, International Classification of Diseases 9th and 10th revisions (ICD-9 and ICD-10), and Classification of Interventions and Procedures v4 (OPCS-4) reflecting relevant vascular interventions. These coding systems were used in accordance with prior published algorithms and adapted to the UK Biobank coding framework. The specific codes, criteria, and operational definitions for each outcome are detailed in Supplemental Table 1. Events were classified as “incident” only if the earliest recorded diagnosis or procedure date occurred after the date of biobank enrollment. Polyvascular or multibed disease was defined as the presence of atherosclerotic in ≥two vascular beds. Statistical analysis Baseline characteristics were summarized using means and standard deviations (SD), medians with interquartile ranges (IQR), or proportions, as appropriate. Missing baseline covariates were addressed using multiple imputation with chained equations [25] generating five imputed datasets. The preservation of the distribution of imputed variables before and after imputation is shown in Supplemental Fig. 2. Genetic variables were excluded from the imputation procedure and analyzed as observed. Continuous variables were standardized within each imputed dataset prior to model fitting and applied consistently across outcome-specific analyses (Supplemental Table 2). All subsequent analyses were conducted across imputed datasets and final estimates were pooled using Rubin’s rules [26]. Covariates were selected a priori based on established ASCVD risk factors and clinical relevance [27]. Collinearity among continuous covariates was evaluated using pairwise correlation matrices and variance inflation factors, which did not indicate substantial multicollinearity (Supplemental Fig. 3; Supplemental Table 3). We first fit multivariable Cox proportional hazard models to examine associations of multidomain risk factors (sociodemographic, clinical, lifestyle, and genetic) with incident CAD, CeVD, and PAD, adjusting for five PCs of genetic ancestry. Hazard ratios (HRs) with 95% confidence intervals (CIs) were reported. For genetic susceptibility, primary models included the PRS matched to each outcome (e.g., CAD PRS for CAD) to preserve vascular bed-specific interpretability. Lifestyle variables were coded in the favorable direction such that HRs <1 reflect protective associations. Furthermore, we trained machine learning-derived gradient-boosted tree models (XGBoost) and computed SHapley Additive exPlanations (SHAP) values as a complementary, model-based ranking to assess concordance with Cox findings and capture potential non-linearities [28]. Effect modification was explored in secondary analyses by stratifying Cox models by sex and age at enrollment (40–49, 50–59, and 60–70 years). Age-specific risk trajectories were further assessed for the most influential risk factors identified by SHAP. To evaluate disease burden and localization, we applied multinomial logistic regression to compare risk factor enrichment for single-bed and multibed versus no ASCVD and for extracoronary (CeVD, PAD, or CVD+PAD) versus coronary (CAD) presentations. Odds ratios (ORs) with 95% CIs were reported. Lastly, to characterize disease-state transitions across the lifespan, we implemented a discrete-time multistate [29] framework with mutually exclusive states (healthy, single-bed disease, dual- and triple-bed ASCVD, and death). Transitions represent changes in clinically documented disease status captured through EHRs rather than systematic vascular screening and may reflect both underlying disease occurrence and patterns of clinical evaluation and detection. Transition-specific risks of entering subsequent observed states were estimated using complementary log-log link [30] models with risk sets defined for each allowed transition (Supplemental Table 4 and Supplemental Fig. 4). Results were reported as HRs with 95% CIs. To assess the robustness of findings, we performed several sensitivity analyses. First, we quantified the standalone discriminatory power of individual risk factors using C-statistics from univariate Cox models, and we repeated the fully adjusted Cox models within a mutually exclusive outcome framework (CAD-only, CeVD-only, PAD-only versus no ASCVD) to minimize overlap-related correlation. Next, we repeated our primary analysis using Cox models but now after excluding individuals on lipid-lowering or antihypertensive therapy at baseline to address potential treatment confounding. Third, to reduce potential misclassification from undiagnosed prevalent ASCVD at baseline, we performed a 1-year landmark sensitivity analysis excluding participants who developed CAD, CeVD, or PAD within 1 year after enrollment and follow-up began 1 year after the biobank enrollment among participants remaining free of ASCVD. Fourth, we performed cause-specific Cox analyses accounting for death before first vascular events as a competing risk. Finally, in multistate models, we collapsed low-frequency transitions (e.g., from PAD to PAD+CeVD) into a unified “polyvascular ASCVD” state to improve stability and recalculated transition-specific estimates. All statistical tests were two-sided, and statistical significance was denoted as P < 0.05. For the multistate analysis, we applied false discovery rate (FDR) correction to account for multiple testing across transitions and risk factors. All analyses were performed using R version 4.1.2 (R Foundation for Statistical Computing, Vienna, Austria). Additional details of analytic model specifications and imputation procedures are provided in the Supplemental Methods. Results Baseline characteristics We included 468,266 participants free of ASCVD at baseline (mean [SD] age, 56.8 [8.1] years; 259,959 [55.5%] female). Most participants were White (94.2%), with small proportions of Black (1.6%), South Asian (1.9%), Chinese (0.3%), and other racial/ethnic backgrounds. Median follow-up was 13.3 years (interquartile range [IQR], 11.7–15.1). At baseline, hypertension was present in 28.5%, hypercholesterolemia in 14.0%, and diabetes in 5.1%; 21.2%, 14.8%, and 3.6% were using antihypertensive, lipid-lowering therapy or diabetes medication, respectively. Median (IQR) SBP and diastolic blood pressure (DBP) were 142.9 (130.0–155.5) mmHg and 82.3 (75.0–89.0) mmHg. Ten percent were current smokers. Median (IQR) Lp(a) was 20.9 (9.5–61.2), and mean (SD) LDL-C was 3.59 (0.86) nmol/L. During follow-up, 13,285 participants developed CAD (2.8%), 8,950 CVD (1.9%), and 7,105 PAD (1.5%); 24,202 developed single-bed disease (5.2%) and 2,484 polyvascular disease (0.5%). Complete characteristics are summarized in Table 1. Baseline characteristics stratified by subsequent ASCVD location are provided in Supplemental Table 5. Incident ASCVD groups generally had higher baseline cardiometabolic risk than participants without incident ASCVD. Table 1Characteristics of study participants.CharacteristicsStudy population (N = 468,266)Age at enrollment, y, mean (SD)56.8 (8.10)Female, n (%)259,959 (55.5%)Self-reported race/ethnicity White, n (%)441,340 (94.2%) Black, n (%)7,464 (1.6%) Chinese, n (%)1,480 (0.3%) South Asian, n (%)8,879 (1.9%) Other, n (%)9,103 (1.9%)Follow-up period, y, mean (SD)13.28 (1.87)Townsend Deprivation Index, median (IQR)−2.17 (−3.66–0.48)Body mass index, kg/m2, mean (SD)27.4 (4.77)Hypertension, n (%)133,286 (28.5%)Systolic blood pressure, mmHg, mean (SD)142.87 (21.85)Diastolic blood pressure, mmHg, mean (SD)82.3 (10.66)Anti-hypertensive medication, n (%)99,491 (21.2%)Hypercholesterolemia, n (%)65,654 (14.0%)LDL-cholesterol, mmol/L, mean (SD)3.59 (0.86)HDL-cholesterol, mmol/L, mean (SD)1.46 (0.38)Triglycerides, mmol/L, median (IQR)1.48 (1.04–2.14)Lipoprotein(a), nmol/L, median (IQR)20.90 (9.51–61.31)Lipid-lowering medication, n (%)69,511 (14.8%)Diabetes mellitus, n (%)24,026 (5.1%)Diabetes medication, n (%)16,747 (3.6%)Hemoglobin A1c, mmol/mol, median (IQR)35.20 (32.70–37.80)Fasting glucose, mmol/L, median (IQR)4.92 (4.60–5.30)Chronic kidney disease, n (%)1,270 (0.3%)Creatinine, µmol/L, median (IQR)69.90 (61.30–80.40)eGFR, mL/min/1.73 m2, mean (SD)91.20 (13.25)hsCRP, mg/dL, median (IQR)1.32 (0.65–2.75)Current smoker, n (%)48,658 (10.4%)Good sleep behavior, n (%)317,336 (68.0%)Average sleep duration, h, mean (SD)7.15 (1.10)Good diet pattern, n (%)395,755 (84.5%)Favorable physical activity, n (%)367,235 (78.4%)Normalized CAD PRS, mean (SD)−0.00 (1.00)Normalized ischemic stroke PRS, mean (SD)−0.00 (1.00)Normalized PAD PRS, mean (SD)−0.01 (1.00)Incident CAD, n (%)13,285 (2.8%)Incident CeVD, n (%)8,950 (1.9%)Incident PAD, n (%)7,105 (1.5%)Incident single events, n (%)24,202 (5.2%)Incident multiple events, n (%)2,484 (0.5%)CAD, coronary artery disease; CeVD, cerebrovascular disease; eGFR, estimated glomerular filtration rate; HDL, high density lipoprotein; hsCRP, high sensitivity C-reactive protein; IQR, interquartile range; LDL, low density lipoprotein; PAD, peripheral artery disease; PRS, polygenic risk score; SD, standard deviation. Characteristics of study participants. CAD, coronary artery disease; CeVD, cerebrovascular disease; eGFR, estimated glomerular filtration rate; HDL, high density lipoprotein; hsCRP, high sensitivity C-reactive protein; IQR, interquartile range; LDL, low density lipoprotein; PAD, peripheral artery disease; PRS, polygenic risk score; SD, standard deviation. Comparative risk factor associations across vascular beds Across CAD, CeVD, and PAD, the same multidomain risk factors contributed to ASCVD, but the relative strength of associations differed substantially by vascular bed (Fig. 1; Supplemental Tables 6–7). The CAD PRS was the strongest genetic determinant (HR, 1.63 [95% CI, 1.59–1.66]), whereas IS and PAD PRSs were weaker. Among sociodemographic factors, age was the strongest risk factor for CeVD and PAD, whereas male sex had the greatest effect for CAD (HR, 2.67 [2.53–2.81]). Socioeconomic deprivation showed minimal influence for CAD but had greater explanatory value for CeVD. Within the clinical domain, SBP consistently ranked among the most influential factors (HRs, 1.21–1.29). LDL-C demonstrated stronger associations with CAD (HR, 1.24 [1.21–1.26]), whereas hsCRP and HbA1c showed relatively larger associations with PAD and CeVD with HRs consistently ≥1.10. Smoking demonstrated the largest effect size among lifestyle factors and was strongly associated with PAD (HR, 3.10 [2.91–3.31]). Good sleep, diet, and physical activity were protective across beds (HRs, 0.82–0.95). Patterns of multidomain risk factor associations with incident CAD, CeVD, and PAD. The heatmap displays adjusted association for 18 standardized risk factors across incident CAD, CeVD, and PAD, grouped by risk-factor domain. HRs are shown on a linear scale centered at 1.0, with shading intensity reflecting effect size magnitude. Values represent from the pooled HRs per SD increase in each determinant (or vs. reference for binary variables), derived from Cox proportional hazards models adjusted for five principal components of genetic ancestry, with Rubin’s rules applied across 5 multiple imputed datasets. PRS variables reflect disease-specific scores: CAD PRS for CAD, PAD PRS for PAD, and ischemic stroke PRS for CeVD. Asterisks (*) indicate significant associations defined as P < 0.05. Risk factors are color-coded by domain: sociodemographic (purple), clinical (red), lifestyle (green), and genetic (blue). Protective associations (HR < 1) appear shaded toward gray; risk-enhancing associations (HR > 1) are shaded toward gold. BMI, body mass index; CAD, coronary artery disease; CeVD, cerebrovascular disease; eGFR, estimated glomerular filtration rate; HbA1c, hemoglobin A1c; HDL, high density lipoprotein; hsCRP, high sensitivity C-reactive protein; HR, hazard ratio; LDL, low density lipoprotein; PAD, peripheral artery disease; PRS, polygenic risk score, SD, standard deviation. Overall, XGBoost-SHAP rankings (Supplemental Fig.s 5–6) were broadly concordant with Cox results, with risk factors ranked highest in Cox also demonstrating high SHAP values. Sensitivity analyses restricted to individuals with single-bed ASCVD presentations, excluding baseline lipid-lowering or antihypertensive therapy, applying a 1-year landmark analysis, and accounting for competing death showed largely unchanged risk-factor rankings across beds (Supplemental Tables 8–12). C-index analyses offered complementary discrimination patterns: age and SBP provided the greatest overall discrimination, the CAD PRS contributed most prominently for CAD, and HbA1c and hsCRP for CeVD and PAD (Supplemental Table 13). An observed inverse association between baseline LDL-C and PAD was explained by differential statin use, as detailed in the Supplemental Results. Differential effects of risk factors by age, sex, and vascular bed In exploratory analyses, age- and sex-stratified analyses suggested vascular bed-specific variation in risk factor associations across the life course (Supplemental Table 14; Supplemental Fig.s 7–9). In CAD, genetic and lipid-related factors were most prominent at younger ages, particularly among men; the CAD PRS was the highest in men aged 40–49 (HR, 1.95 [1.80–2.10]) and attenuated with age, and LDL-C similarly showed larger associations in men with age-related attenuation. In contrast, HDL-C and SBP appeared more influential among younger females, with convergence across sexes in older strata. In CeVD, smoking remained the dominant determinant across age groups and was consistently stronger in women, peaking in midlife (e.g., women 50–59 years: HR, 2.21 [1.74–2.58]); SBP also appeared more strongly associated in women in midlife, whereas the IS PRS showed modest associations, mainly in men. In PAD, smoking demonstrated the largest age- and sex-specific gradients of any risk factor, with HRs >3 across most strata and the highest estimate in older females (ages 60–70: HR, 3.75 [3.31–4.26]). The PAD PRS peaked in midlife men (HR, 1.32 [1.23–1.42]) and was relatively stable in women, while HbA1c associations attenuated and hsCRP associations increased with age. Across beds, lifestyle factors (sleep and physical activity) were generally protective, with larger relative effects in younger strata. Age-varying models across all three vascular beds demonstrated parallel patterns: genetic effects strongest earlier in life, smoking and metabolic/inflammatory factors peaking in midlife, and declining effects at older ages, while lifestyle behaviors remained protective throughout (Supplemental Results). Determinants of ASCVD burden and localization In multinomial models (Supplemental Table 15), male sex, older age, higher SBP, and current smoking were associated with higher odds of both single- and multibed ASCVD. For example, each SD higher SBP was associated with 23% higher odds of single-bed disease and 47% higher odds of multibed disease. Favorable lifestyle traits, such as healthy sleep, diet, and physical activity, were inversely associated with the odds of both single- and multibed disease (Supplemental Fig. 10). When comparing vascular localization relative to CAD, LDL-C, HDL-C, and Lp(a) were more strongly associated with coronary disease, whereas HbA1c, hsCRP, and eGFR were enriched in PAD and CeVD. Socioeconomic deprivation was more strongly associated with extracoronary presentations (e.g., TDI OR, 1.28 [1.17–1.39] for CeVD+PAD vs CAD). Current smoking was most strongly associated with PAD and CeVD+PAD, and less so with CAD (Fig. 2). These analyses identified relative enrichment of drivers in extracoronary versus coronary disease, rather than absolute protective or harmful effects. Relative enrichment of risk factors in extra-coronary presentations. Odds ratios (ORs) with 95% confidence intervals (CIs) for determinants of incident cerebrovascular disease (CeVD), peripheral artery disease (PAD), or concurrent CeVD+PAD, each compared with incident coronary artery disease (CAD). ORs >1 indicate higher odds of the indicated outcome category relative to CAD-only (reference group), whereas ORs <1 indicate lower odds relative to CAD-only. Estimates were derived from multinomial logistic regression models run across 5 multiple imputed datasets, with results pooled using Rubin’s rules. Models were adjusted for five principal components of genetic ancestry. Risk factors are categorized into domains: sociodemographic, clinical, lifestyle, and genetic. Distinct PRSs for CAD, PAD, and ischemic stroke were included simultaneously to capture vascular bed-specific genetic risk. BMI, body mass index; CAD, coronary artery disease; CeVD, cerebrovascular disease; eGFR, estimated glomerular filtration rate; HbA1c, hemoglobin A1c; HDL-C, high-density lipoprotein cholesterol; hsCRP, high sensitivity C-reactive protein; IS, ischemic stroke; LDL-C, low-density lipoprotein cholesterol; PAD, peripheral artery disease; PRS, polygenic risk score; SBP, systolic blood pressure; SD, standard deviation; TDI, Townsend Deprivation Index. Observed ASCVD states transitions across the life course Sankey diagrams (Supplemental Fig. 11–12) illustrate observed transitions between clinically documented ASCVD states within successive 10-year age intervals, based on each individual’s first and last observed state during that period. Most individuals remained free of documented ASCVD within each age interval, although this proportion declined from >99% at age 40–50 to ∼95% at age 80–90. Incident clinically observed single-bed ASCVD in midlife was dominated by CAD, followed by CeVD and PAD. Stability decreased with advancing age for all single-bed states, with PAD showing the most pronounced age-related increase in observed transitions to polyvascular ASCVD and death, especially in older adults. Mortality rates per 1,000 person-years also increased progressively with age across all vascular states, with the highest rates observed among individuals with PAD (Fig. 3; Supplemental Tables 16–18). Age-specific progression and mortality from single-bed ASCVD. (A) Observed progression from isolated CAD, CeVD, or PAD across successive 10-year age intervals, defined as transition from the starting single-bed ASCVD state to a different atherosclerotic state, including polyvascular ASCVD or death, within the same age interval. (B) Mortality rates by starting vascular state across the same age intervals, expressed as deaths per 1,000 person-years. Transition estimates were derived from each individual’s first and last documented ASCVD state within each age interval. Numbers of individuals contributing to each age interval vary because of staggered cohort entry and left truncation in the UK Biobank cohort. The 40–50-year age interval for single-bed ASCVD states was omitted from the Fig. because of sparse event counts and unstable estimates in younger affected individuals; corresponding data are provided in Supplementary Tables 12–13. Denominators and transition counts are provided in Supplementary Tables 12–13, and detailed transition flow diagrams are shown in Supplementary Fig. 11. CAD indicates coronary artery disease; CeVD, cerebrovascular disease; PAD, peripheral artery disease; PY, person-years. Risk factor patterns for ASCVD onset, subsequent involvement, and mortality Across the multistate framework (Supplemental Table 19; Supplemental Fig. 13), transitions from healthy to incident CAD were most strongly associated with male sex, SBP, lipid markers, smoking, and the CAD PRS (all FDR < 0.05). Incident PAD was dominated by smoking (HR, 4.29 [4.04–4.55]) with contributions from male sex, HbA1c, hsCRP, and PAD PRS, while CeVD onset was associated primarily with smoking, SBP, and socioeconomic deprivation (HR, 1.12 [95% CI, 1.09–1.14]). Sleep and physical activity were protective across all incident outcomes, while genetic risk scores showed outcome-specific relevance. Transitions from single bed to polyvascular disease (Supplemental Table 19; Supplemental Fig.s 14–17) was most strongly associated with metabolic and inflammatory factors. For instance, observed transitions from CAD to CAD+PAD were associated with excess hazard from HbA1c (HR, 1.33 [95% CI, 1.25–1.43]), hsCRP (HR, 1.29 [95% CI, 1.16–1.43]), and current smoking (HR, 2.35 [95% CI, 1.79–3.08]). Subsequent involvement from PAD or CeVD to more advanced states showed fewer consistent determinants, though TDI was a recurrent factor in cerebrovascular-related polyvascular involvement (e.g., from CAD to CAD+CeVD and from CeVD to CeVD+PAD), with HbA1c and hsCRP also contributing. Determinants of mortality were distinct and dominated by clinical and behavioral factors (Supplemental Table 19; Supplemental Figs. 13–17). Inflammation, glycemic dysregulation, higher BMI, and smoking increased mortality risk; eGFR, sleep, and physical activity were protective. Genetic associations with mortality were largely absent once ASCVD had developed, although CAD PRS was significantly associated with death following PAD. By contrast, PRSs showed positive associations with transitions from healthy to death. Sensitivity analysis collapsing outcomes by polyvascular burden yielded qualitatively similar results, with more stable associations due to larger event counts (Supplemental Table 20; Supplemental Figs. 18–21). We also noted paradoxical inverse associations for LDL-C and Lp(a) in certain later-stage transitions. For LDL-C, landmark analyses addressing potential reverse causation, and for Lp(a), inspection of event timing and cumulative incidence curves, clarified these patterns (Supplemental Results). Discussion In this large, prospective cohort, we provide an integrated characterization of multidomain risk profiles across vascular beds and across the disease course. Rather than identifying novel individual risk factors, our analysis quantifies how shared sociodemographic, clinical, lifestyle, and genetic domains are differentially weighted across vascular territories and from initial onset through subsequent documented polyvascular involvement and mortality (Central Illustration). We demonstrate that the relative influence of these domains shifts meaningfully over time: genetic susceptibility is most prominent before clinical ASCVD and in premature events, whereas metabolic, inflammatory, behavioral, and socioeconomic factors show stronger associations with subsequent polyvascular disease involvement and late mortality. This comparative framework clarifies the heterogeneity of ASCVD expression within a single harmonized population and provides context for how risk should be interpreted differently according to vascular bed and clinically observed disease stage. Vascular bed differences in the expression of shared ASCVD determinants Consistent with prior studies [31], traditional risk factors such as age, hypertension, diabetes, and smoking contributed to ASCVD; however, our comparative analyses show that the same multidomain risk factors carry different weights across vascular beds. Continued smoking, inflammation, and glycemic dysregulation had disproportionately stronger effects on PAD, reflecting the heightened inflammatory-metabolic milieu characteristic of peripheral atherosclerosis [[32], [33], [34]], whereas LDL-C and the CAD PRS were more influential for coronary events. These contrasts, observed consistently across analytic frameworks, highlight that ASCVD is not biologically uniform, but a heterogenous vascular syndrome in which shared exposures demonstrate varying relative associations across vascular beds [35]. Understanding these differences may sharpen risk interpretation: patients with similar profiles may diverge toward CAD, CeVD, or PAD because the same exposures exert differing influence in different beds [35]. In this context, we use “weights” to refer to the strength of association (HR per standardized unit) within this cohort, rather than disease burden. Accordingly, this perspective also supports more precise evaluation based on a patient’s dominant risk-factor pattern, helping clinicians anticipate which vascular territory may be most vulnerable. Although beyond the scope of this analysis, the marked heterogeneity we observed may provide a conceptual foundation for future research on whether vascular-bed differences should be incorporated into risk-prediction or trial design, particularly in PAD, where inflammatory and metabolic pathways appear especially prominent. PAD as a marker of systemic risk Individuals presenting with clinically diagnosed PAD exhibited the highest observed rates of subsequent polyvascular involvement and death compared with those presenting with CAD or CeVD. This pattern likely reflects a combination of advanced systemic atherosclerotic burden at the time of PAD recognition and the clinical context in which PAD is typically diagnosed. Patients with PAD often present with clustered risk factors, including chronic kidney disease, diabetes, and smoking [36,37], and may remain underrecognized until advanced stages of disease [38,39]. In routine practice, evaluation for PAD often prompts more comprehensive cardiovascular assessment, and coexisting CAD or CeVD may be detected during preoperative or vascular workup [[40], [41], [42], [43], [44]]. Accordingly, these transition estimates should be interpreted as reflecting observed clinical trajectories within routine care rather than definitive evidence of intrinsic pathophysiologic spread between vascular territories. Within this context, PAD may be understood as a clinical marker of systemic atherosclerotic risk rather than a limb-limited condition. A diagnosis of PAD should prompt careful assessment for atherosclerotic involvement in other territories and early, aggressive intensification of modifiable factors management, including smoking cessation, glycemic and blood pressure control, and reduction of inflammatory burden [45]. These findings reinforce current guideline recommendations and underscore persistent missed opportunities for timely recognition and comprehensive management of high-risk PAD populations. Genetic risk: early impact, diminishing later Genetic susceptibility, captured by PRSs, was most consequential before ASCVD onset, especially for CAD, consistent with prior evidence linking higher genetic risk to greater coronary plaque burden and earlier coronary events [46,47]. What becomes clear in our transition-based framework is how this influence declined markedly once clinical ASCVD had been documented. Across vascular beds, PRSs showed minimal association with subsequent polyvascular involvement or mortality, with the exception that CAD PRS remained associated with mortality in individuals with PAD. In contrast, PRSs continued to associate with observed transitions from healthy to death, a pattern that may reflect contributions from genetically mediated premature fatal events such as sudden cardiac death, fatal myocardial infarction, or early-onset stroke [[48], [49], [50]]. Clinically, these observations may delineate a potential window of utility for inherited risk: genetic information may be most informative earlier in life, before competing risks and accumulated exposures dominate, and less discriminating once overt disease develops in older adulthood [51,52]. This life-course gradient supports further evaluation of genetic information for earlier risk stratification, while later documented disease stages appear to be more strongly shaped by modifiable risk factors. Clinical, lifestyle, and social determinants of polyvascular burden and mortality Across vascular beds, our multistate analyses suggest that the risk-factor patterns associated with incident ASCVD differ from those associated with later polyvascular burden and mortality, offering clearer insight into where clinical action is most impactful. Once disease was documented, advancement to polyvascular involvement or death was associated predominantly with a compounded burden of metabolic dysfunction, inflammation, renal impairment, and adverse lifestyle factors [53]—domains that outweighed genetic susceptibility at later stages. A particularly important and clinically actionable finding was the consistent protective association of favorable lifestyle behaviors, including physical activity and healthy sleep, across vascular beds and throughout the disease course. These associations persisted beyond primary prevention and remained evident in advanced disease states and mortality, underscoring that modifiable behaviors retain prognostic relevance even after ASCVD onset [54]. This durability contrasts with genetic susceptibility, which was most strongly associated with early observed disease transitions and attenuated after. Socioeconomic disadvantage further amplified risks of subsequent polyvascular burden and mortality, particularly in cerebrovascular-related patterns, reinforcing the importance of considering social context when assessing patient vulnerability [55]. Taken together, these findings suggest that post-onset risk stratification should place greater weight on modifiable clinical and behavioral exposures, as well as socioeconomic context, to help identify which patients may benefit from closer surveillance and intensification of preventive therapy to reduce polyvascular progression. Strengths and limitations This study has several strengths. It leverages a large, contemporary, population-based cohort with extended follow-up and multidomain phenotyping, enabling comprehensive risk profiling across clinical, lifestyle, and genetic domains. The use of multiple complementary analytic frameworks, including Cox models, SHAP-based machine learning, age-varying hazard models, and multistate transition modeling, allowed us to assess both the relative importance and temporal dynamics of risk factors across disease stages and vascular beds. By integrating polygenic risk scores alongside clinical and behavioral data, we evaluated their relative contribution to early disease susceptibility and subsequent documented polyvascular burden. However, several limitations warrant consideration. ASCVD outcomes were derived from linked hospital and mortality records, which may lead to misclassification or under-ascertainment of outpatient or subclinical disease. Accordingly, multistate transitions should be interpreted as clinically documented patterns of ASCVD involvement rather than direct evidence of biological spread across vascular territories. Risk factors were assessed only at baseline, and we could not fully account for time-updated changes in clinical status, medication use, or risk factor control over time. Some later-stage transitions in the multistate models were based on few events, yielding wide confidence intervals and unstable estimates. These results should be considered exploratory and interpreted with caution. In addition, several cardiometabolic and inflammatory variables included in multivariable models might be correlated, which may influence adjusted effect size estimates and the stability of relative rankings across predictors. Similarly, comparisons across PRSs should be interpreted cautiously because the CAD, IS, and PAD PRSs were derived from different GWASs with different phenotype definitions, discovery sample sizes, and predictive performance; thus, differences in observed associations may reflect PRS robustness as well as vascular bed-specific genetic architecture. Although the IS PRS was the closest available validated cerebrovascular genetic risk score, it may not fully represent the genetic architecture of the broader CeVD phenotype which includes carotid stenosis. Therefore, small differences in adjusted effect estimates or relative rankings should be interpreted cautiously. Furthermore, lifestyle measures were self-reported at baseline and may be subject to measurement error and healthy user/residual confounding. The study cohort from the UK Biobank consisted largely of White people of European ancestry and had a relatively healthy volunteer profile, which limits generalizability to more diverse or higher-risk populations. Finally, as with all observational studies, residual confounding, including effects related to treatment use, competing risks, reverse causation, or unmeasured social and behavioral factors, cannot be fully excluded. Conclusions ASCVD arises from shared risk domains, but the magnitude of these domains differs meaningfully across vascular beds and documented disease stages. Genetic factors contribute most to early disease susceptibility, whereas clinical and lifestyle factors dominate later polyvascular involvement and mortality. PAD is frequently accompanied by polyvascular involvement, underscoring the importance of broad evaluation and rigorous risk-factor control. These insights offer a framework for interpreting ASCVD risk throughout the disease course and highlight opportunities for vascular bed-specific risk stratification and prevention. Central Illustration: Among 468,266 UK Biobank participants free of ASCVD at baseline, CAD, CeVD, and PAD arose from shared sociodemographic, clinical, lifestyle, and genetic domains, but with distinct emphases across vascular beds (e.g., lipid/genetic factors for CAD; socioeconomic factors for CeVD; inflammation, metabolic dysfunction, and smoking for PAD). PAD was most frequently followed by documented polyvascular involvement (18.8%) and death (33.3 deaths/1,000 PY), functioning as a sentinel marker of systemic atherosclerotic risk. Genetic factors were most strongly associated with early disease susceptibility and premature death, whereas clinical and lifestyle factors were more prominent in later polyvascular involvement and mortality. These patterns underscore that the clinical relevance of shared risk factors differs by vascular bed and documented disease stage, informing more precise ASCVD risk interpretation. ASCVD, atherosclerotic cardiovascular disease; CAD, coronary artery disease; CeVD, cerebrovascular disease; HbA1c, glycated hemoglobin; hsCRP, high-sensitivity C-reactive protein; LDL-C, low-density lipoprotein cholesterol; PAD, peripheral artery disease; PRS, polygenic risk score; PY, person-years; TDI, Townsend Deprivation Index. Disclosure of interest Y.S. reports serving as a consultant from Arboretum Lifesciences. P.N. reports research grants from Allelica, Amgen, Apple, Boston Scientific, Cleerly, Genentech / Roche, Ionis, Novartis, and Silence Therapeutics, personal fees from AIRNA, Allelica, Apple, AstraZeneca, Bain Capital, Blackstone Life Sciences, Bristol Myers Squibb, Creative Education Concepts, CRISPR Therapeutics, Eli Lilly & Co, Esperion Therapeutics, Foresite Capital, Foresite Labs, Genentech / Roche, GV, HeartFlow, Incyte, Magnet Biomedicine, Merck, Novartis, Novo Nordisk, TenSixteen Bio, and Tourmaline Bio, equity in Bolt, Candela, Mercury, MyOme, Parameter Health, Preciseli, and TenSixteen Bio, royalties from Recora for intensive cardiac rehabilitation, and spousal employment at Vertex Pharmaceuticals, all unrelated to the present work. P.T.E. has received sponsored research support from Bayer AG and IBM Health, and he has served on advisory boards or consulted for Bayer AG, MyoKardia and Novartis. A.C.F. reports being co-founder of Goodpath and Avigena, serving as scientific advisor to MyOme, Aditum Bio, HeartFlow, and Arboretum Health. The remaining authors do not report any disclosures. No other relationships or activities that could appear to have influenced the submitted work. Ethical approval The UK Biobank study protocol was approved by the North West Multi-centre Research Ethics Committee (11/NW/0382). The secondary analysis of UKB for this present study was approved by the Mass General Brigham human research committee institutional review board (2018P002292). Declaration of generative AI use During the preparation of this work, OpenAI ChatGPT's image-generation capabilities were used to assist in creating and refining the central illustration. The authors reviewed and edited the output as needed and take full responsibility for the content of the published article. Data availability All data and materials are publicly available at the UKB and are accessible to researchers by application at https://www.ukbiobank.ac.uk/. The codes used in the analyses will be shared on reasonable request to the corresponding author. Funding Y.S. is supported by the TOPMed fellowship from the National Heart Lung and Blood Institute. S.J.C. is supported by a grant from the National Heart Lung and Blood Institute (K099HL177340). P.N. is supported by a grant from the National Heart Lung and Blood Institute (R01HL127564). P.T.E. is supported by grants from the National Institutes of Health (1R01HL092577, 1R01HL157635, 5R01HL139731), from the American Heart Association Strategically Focused Research Networks (18SFRN34110082), and from the European Union (MAESTRIA 965,286). A.C.F. is supported by grants from the National Heart Lung and Blood Institute (K08HL161448 and R01HL164629). The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication. CRediT authorship contribution statement Kelvin Supriami: Writing – review & editing, Writing – original draft, Visualization, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Alyssa M. Flores: Writing – review & editing, Data curation. Yang Sui: Writing – review & editing, Methodology. So Mi Jemma Cho: Writing – review & editing, Methodology. Aniruddh P. Patel: Writing – review & editing, Data curation. Yunfeng Ruan: Writing – review & editing, Data curation. Krishna G. Aragam: Writing – review & editing, Data curation. Pradeep Natarajan: Writing – review & editing. Patrick T. Ellinor: Writing – review & editing. Akl C. Fahed: Writing – review & editing, Supervision, Methodology, Conceptualization. Declaration of competing interest The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Yang Sui, PhD reports a relationship with Arboretum Lifesciences that includes: consulting or advisory. Pradeep Natarajan, MD, MMSc reports a relationship with Allelica that includes: consulting or advisory and funding grants. Pradeep Natarajan, MD, MMSc reports a relationship with Amgen that includes: funding grants. Pradeep Natarajan, MD, MMSc reports a relationship with Apple that includes: consulting or advisory and funding grants. Pradeep Natarajan, MD, MMSc reports a relationship with Boston Scientific that includes: funding grants. Pradeep Natarajan, MD, MMSc reports a relationship with Cleerly that includes: funding grants. Pradeep Natarajan, MD, MMSc reports a relationship with Genentech that includes: consulting or advisory and funding grants. Pradeep Natarajan, MD, MMSc reports a relationship with Ionis that includes: funding grants. Pradeep Natarajan, MD, MMSc reports a relationship with Novartis that includes: consulting or advisory and funding grants. Pradeep Natarajan, MD, MMSc reports a relationship with Silence Therapeutics that includes: funding grants. Pradeep Natarajan, MD, MMSc reports a relationship with AIRNA that includes: consulting or advisory. Pradeep Natarajan, MD, MMSc reports a relationship with AstraZeneca that includes: consulting or advisory. Pradeep Natarajan, MD, MMSc reports a relationship with Bain Capital that includes: consulting or advisory. Pradeep Natarajan, MD, MMSc reports a relationship with Blackstone Life Sciences that includes: consulting or advisory. Pradeep Natarajan, MD, MMSc reports a relationship with Bristol Myers Squibb that includes: consulting or advisory. Pradeep Natarajan, MD, MMSc reports a relationship with Creative Education Concepts that includes: consulting or advisory. Pradeep Natarajan, MD, MMSc reports a relationship with CRISPR Therapeutics that includes: consulting or advisory. Pradeep Natarajan, MD, MMSc reports a relationship with Eli Lilly & Co that includes: consulting or advisory. Pradeep Natarajan, MD, MMSc reports a relationship with Esperion Therapeutics that includes: consulting or advisory. Pradeep Natarajan, MD, MMSc reports a relationship with Foresite Capital that includes: consulting or advisory. Pradeep Natarajan, MD, MMSc reports a relationship with Foresite Labs that includes: consulting or advisory. Pradeep Natarajan, MD, MMSc reports a relationship with GV that includes: consulting or advisory. Pradeep Natarajan, MD, MMSc reports a relationship with HeartFlow that includes: consulting or advisory. Pradeep Natarajan, MD, MMSc reports a relationship with Incyte that includes: consulting or advisory. Pradeep Natarajan, MD, MMSc reports a relationship with Magnet Biomedicine that includes: consulting or advisory. Pradeep Natarajan, MD, MMSc reports a relationship with Merck that includes: consulting or advisory. Pradeep Natarajan, MD, MMSc reports a relationship with Novo Nordisk that includes: consulting or advisory. Pradeep Natarajan, MD, MMSc reports a relationship with TenSixteen Bio that includes: consulting or advisory. Pradeep Natarajan, MD, MMSc reports a relationship with Tourmaline Bio that includes: consulting or advisory. Pradeep Natarajan, MD, MMSc reports a relationship with Bolt that includes: equity or stocks. Pradeep Natarajan, MD, MMSc reports a relationship with Candela that includes: equity or stocks. Pradeep Natarajan, MD, MMSc reports a relationship with Mercury that includes: equity or stocks. Pradeep Natarajan, MD, MMSc reports a relationship with MyOme that includes: equity or stocks. Pradeep Natarajan, MD, MMSc reports a relationship with Parameter Health that includes: equity or stocks. Pradeep Natarajan, MD, MMSc reports a relationship with Preciseli that includes: equity or stocks. Pradeep Natarajan, MD, MMSc reports a relationship with TenSixteen Bio that includes: equity or stocks. Pradeep Natarajan, MD, MMSc reports a relationship with Recora for intensive cardiac rehabilitation that includes:. Pradeep Natarajan, MD, MMSc reports a relationship with Vertex Pharmaceuticals that includes:. Patrick T. Ellinor, MD, PhD reports a relationship with Bayer AG that includes: board membership, consulting or advisory, and funding grants. Patrick T. Ellinor, MD, PhD reports a relationship with IBM Health that includes: funding grants. Patrick T. Ellinor, MD, PhD reports a relationship with MyoKardia that includes: consulting or advisory. Patrick T. Ellinor, MD, PhD reports a relationship with Novartis that includes: consulting or advisory. Akl C. Fahed, MD, MPH reports a relationship with Goodpath that includes: equity or stocks. Akl C. Fahed, MD, MPH reports a relationship with Avigena that includes: equity or stocks. Akl C. Fahed, MD, MPH reports a relationship with MyOme that includes: consulting or advisory. Akl C. Fahed, MD, MPH reports a relationship with Aditum Bio that includes: consulting or advisory. Akl C. Fahed, MD, MPH reports a relationship with HeartFlow that includes: consulting or advisory. Akl C. Fahed, MD, MPH reports a relationship with Arboretum Health that includes: consulting or advisory. If there are other authors, they declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Acknowledgements We are grateful to the participants and investigators of the UK Biobank for their contributions to this study. The authors also thank Dr. Brian L. Claggett for his assistance with statistical analyses and consideration. Footnotes Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.ajpc.2026.101710. Supplementary materials References Associated Data Supplementary Materials Data Availability Statement All data and materials are publicly available at the UKB and are accessible to researchers by application at https://www.ukbiobank.ac.uk/. The codes used in the analyses will be shared on reasonable request to the corresponding author.