Individual and Neighborhood-level Socioeconomic Status and Somatic Mutations Associated with Increased Risk of Cardiovascular Diseases and Mortality: A Cross-Sectional Analysis in the Women’s Health Initiative

Abstract

BACKGROUND:

Clonal hematopoiesis of indeterminate potential (CHIP), the expansion of leukemogenic mutations in white blood cells, has been associated with increased risk of atherosclerotic cardiovascular diseases, cancer, and mortality.

OBJECTIVE:

We examined the relationship between individual- and neighborhood-level socioeconomic status (NSES) and CHIP and evaluated effect modification by interpersonal and intrapersonal resources.

METHODS:

The study population included 10,799 postmenopausal women from the Women’s Health Initiative without hematologic malignancy or antineoplastic medication use. Individual- and neighborhood (Census tract)-level SES were assessed across several domains including education, income, and occupation, and an NSES summary z-score, which captures multiple dimensions of SES, was generated. Interpersonal and intrapersonal resources were self-reports. CHIP was ascertained based on a prespecified list of leukemogenic driver mutations. Weighted logistic regression models adjusted for covariates were used to estimate risk of CHIP as an odds ratio (OR) and 95% confidence interval (95% CI).

RESULTS:

The interval-scale NSES summary z-score was associated with a 3% increased risk of CHIP: OR (95% CI) = 1.03 (1.00, 1.05), P = 0.038. Optimism significantly modified that estimate, such that among women with low/medium and high levels of optimism, the corresponding ORs (95% CIs) were 1.03 (1.02, 1.04) and 0.95 (0.94, 0.96), P_Interaction < 0.001.

CONCLUSIONS:

Our findings suggest that reduced risk of somatic mutation may represent a biological pathway by which optimism protects contextually advantaged but at-risk women against age-related chronic disease and highlight potential benefits of long-term, positive psychological interventions.

1. INTRODUCTION

Clonal hematopoiesis of indeterminate potential (CHIP) is a common aging-related phenomenon in which hematopoietic stem cell mutations lead to the overexpansion of genetically distinct subpopulations of white blood cells (Calvillo-Arguelles et al., 2019; Jaiswal et al., 2017; Khetarpal et al., 2019; Libby et al., 2019; Steensma, 2018; Steensma et al., 2015). Although CHIP prevalence increases with age, affecting 10–20% of community-dwelling adults over 70 years old, it has been consistently associated with an increased risk of hematologic cancer (Steensma et al., 2015), coronary heart disease (Jaiswal et al., 2017), heart failure (Yu et al., 2021), early-onset myocardial infarction, stroke (Bhattacharya et al., 2022), chronic kidney disease (Vlasschaert et al., 2022), and mortality (Jaiswal, 2020). Inflammation and immunity, which play critical roles in the development of aging-related conditions such as atherosclerosis, also appear to be involved in enhancing the cardiovascular consequences of CHIP (Calvillo-Arguelles et al., 2019; Sano et al., 2018).

Recent studies have identified associations of cardiovascular risk factors like smoking (Dawoud et al., 2020; Ramanathan et al., 2021), obesity (Haring et al., 2021), unhealthy diet (Bhattacharya et al., 2021a, 2021b), and shorter leukocyte telomere length (LTL) (Aviv & Levy, 2019) with increased prevalence of CHIP, but to date, none have examined socioeconomic status (SES) as a risk factor for CHIP. The relationships between individual- and neighborhood-level socioeconomic disadvantage and CHIP therefore remain uncharacterized.

Individual- and neighborhood-level SES are distinct constructs with independent effects on health (Diez Roux & Mair, 2010; Ludwig et al., 2012; Ludwig et al., 2011). Lower individual-level SES (Chetty et al., 2016; Elo, 2009; Foraker et al., 2019; Freeman et al., 2016; Gebreab et al., 2015; Lewer et al., 2020; Stringhini et al., 2017) and lower neighborhood-level SES (independent of individual-level SES) (Arcaya et al., 2016; Diez Roux & Mair, 2010; Meijer et al., 2012; Pickett & Pearl, 2001; Robert, 1999; Truong & Ma, 2006) have been associated with increased risk of cardiovascular disease, depression, and premature mortality.

Individual and neighborhood socioeconomic disadvantage can lead to chronic psychological stress, which may result in high allostatic load (i.e., accumulating “wear and tear”) on the neuroendocrine and immune systems (Geronimus et al., 2006). Racial and ethnic minoritized individuals are more likely to be disadvantaged (Warnecke et al., 2008) and persistently exposed to systemic racism that is associated with high allostatic load and poor health outcomes (Guidi et al., 2021). Moreover, socioeconomically disadvantaged individuals are more likely to engage in unfavorable health behaviors such as physical inactivity and consumption of an unhealthy diet (Diez Roux & Mair, 2010; Rimmele et al., 2009; Volek et al., 2008).

The Reserve Capacity Model (RCM) is a theoretical framework proposed for understanding how interpersonal resources (e.g., social integration and social support) and intrapersonal resources (e.g., optimism) serve as pathways connecting SES to health outcomes like CHIP (Gallo et al., 2009; Gallo & Matthews, 2003). Under the RCM, interpersonal and intrapersonal resources can mediate or modify associations between SES, stress, and health. For example, modification could involve low SES individuals who maintain a reserve capacity or bank of adaptive interpersonal and intrapersonal resources having resistance to stressful events associated with their socioeconomic disadvantage (Gallo et al., 2009).

Indeed, social support has been identified as a critical buffer to the deleterious effects of sustained stress (Cohen & Wills, 1985). Social support is associated with a healthier and more resilient “biological profile” (Uchino, 2006), which may enable individuals to mount adaptive psychological and physiological responses to stressors (Epel et al., 2018). Constructs of social support have been associated with stress-related autonomic, neuroendocrine, and immune function (Cohen, 2004; Taylor et al., 2006). For example, companionship is associated with relatively salutary cortisol profiles, blunted inflammatory responses to acute stress (Cacioppo et al., 2015; Eisenberger et al., 2017; Hawkley & Cacioppo, 2010), and later mortality (Holt-Lunstad et al., 2015).

Dispositional optimism—a psychological trait characterized by the generalized expectation of a positive future—has been shown to be protective for health against the detrimental impacts of stress or adversity (Scheier & Carver, 1992; Smith & MacKenzie, 2006; Solberg Nes, 2016), including low socioeconomic status (SES; (Morozink et al., 2010; Zilioli et al., 2015). The mechanisms underlying the potential protective effect of optimism on health remain unclear. Physiological mechanisms (e.g., better immune function and lower levels of inflammation) and behavioral mechanisms (e.g., increased likelihood of engaging in healthy behaviors) may counteract the physiological effects of negative emotions (Fredrickson et al., 2000; Ong & Allaire, 2005), promote benefit-finding and adaptive coping skills (Bower et al., 2008; Folkman & Moskowitz, 2000), and thereby explain the protective effects of optimism on health (Sin, 2016).

The primary aim of this study was to examine the relationships between individual- and neighborhood-level SES and CHIP as a means of understanding potential mechanisms underlying relationships between individual- and neighborhood-level socioeconomic disadvantage and related health outcomes such as cardiovascular disease. To test the RCM hypothesis that the SES-CHIP associations are modified by reserve capacity, a secondary aim of this study was to determine if social integration, social support, and optimism modify the associations of individual- and neighborhood-level SES with CHIP. We hypothesized that lower individual- and neighborhood-level SES will be associated with higher risk of CHIP. We also hypothesized that social integration, social support, and optimism will attenuate the association of individual-level and neighborhood-level SES and CHIP.

2. METHODS

2.1. Study population

The Women’s Health Initiative (WHI) is a longitudinal study of risk factors for cardiovascular disease, cancer, osteoporotic fractures, and other causes of morbidity and mortality among postmenopausal women (Hays et al., 2003; TWHI, 1998). Between 1993 and 1998, its 40 centers throughout the U.S. enrolled post-menopausal women aged 50–79 years, either in its observational study (OS, n = 93,676) or one or more of its overlapping randomized, controlled clinical trials (CT, n = 68,132) of hormone therapy (HT), calcium / vitamin D supplementation, and dietary modification (Anderson et al., 2003). WHI participants completed a screening visit (SV) (1993–1998). CT participants also completed an annual visit at one (1995–2001), three (1997–2002), six (2000–2004), and nine years (2004) after randomization (AV1, AV3, AV6, AV9), and OS participants three years after enrollment (AV3). An additional visit among a subset of CT and OS participants occurred between 2012 and 2013 (ranging from 14 to 19 years after enrollment) as part of the WHI Long Life Study (WHI-LLS, 2021). Institutional review boards at participating institutions approved all study protocols, and all participants provided written informed consent.

2.2. Design

Within WHI, we conducted a case-control study of the 11,023 OS and CT participants with extracted DNA and database of Genotypes and Phenotypes (dbGAP) consent, who had been sampled for whole genome sequencing by the Trans-Omics for Precision Medicine (TOPMed) program (Taliun et al., 2021). The WHI TOPMed sample included all 5,874 previously ascertained cases of stroke or venous thromboembolism and an age-, race-, and HT-stratified, 3.8% random sample of 5,149 controls matched 1:1 to cases (TOPMed, 2021). We excluded 224 (2.0%) of the participants with a history of hematologic malignancy, antineoplastic medication use, or missing sampling weights, yielding a final sample of 10,799 participants with DNA from either the SV (1993–1998) (31%), AV1 (1995–2001) (27%), AV3 (1997–2002) (35%), AV6 (2000–2004) (5%), or AV9 (2004) (1%) exam.

2.3. Exposures

2.3.1. Individual-level Socioeconomic Status

We assessed individual-level SES in three distinct domains that have varied effects on health: education, income/wealth, and occupation (Elo, 2009). To approximate them, we obtained individual-level education (<high school, high school / trade school / GED, some college or associate degree, bachelor’s degree or higher), family income (<$20,000, $20,000–$34,999, $35,000–$49,999, $50,000–$74,999, ≥$75,000), and occupation (homemaker only, service / labor, technical / sales / administrative, managerial / professional) variables from the SV questionnaire. We then dichotomized education (< and ≥ college), family income (< and ≥ $50,000), and occupation (homemaker only / service / labor and technical / sales / administrative / managerial / professional).

2.3.2. Neighborhood-level Socioeconomic Status

We linked six U.S. Census of Population (year 2000) variables to the accurately geocoded addresses of WHI participants at the time of blood draw: (1) median household income, (2) median value of housing units, (3) percentage of households receiving interest, dividend, or net rental income, (4) percentage of adults ≥ 25 years of age who had completed high school, (5) percentage of adults ≥ 25 years of age who had completed college,^, and (6) percentage of employed persons ≥ 16 years of age in executive, managerial, or professional specialty occupations. The variables represent several dimensions of income/wealth, education, and occupation aggregated at the U.S. Census tract, the lowest geographic level historically associated with accurate and reliable assignment of Federal Information Processing System codes (Whitsel et al., 2006; Whitsel et al., 2004).

We log transformed variables 1–2 and then z-transformed variables 1–6, i.e., subtracted the population-specific mean from participant-specific values and then divided the differences by the population-specific standard deviation. The six resulting z-scores indicated the deviation of a given value from the corresponding, population-specific mean and summed to zero. For example, a z-score of +1.0 for log (median household income) indicated that the value was one standard deviation above the population-specific mean. By summing the z-scores of variables 1–6, we then constructed a neighborhood SES summary z-score, increases in which imply increasing neighborhood socioeconomic advantage (Diez Roux et al., 2001).

2.3.3. Interpersonal and Intrapersonal Resources

We estimated interpersonal resources (social integration, social support) and intrapersonal resources (dispositional optimism) using self-reported information on the SV questionnaire. We measured social integration (our proxy of social network size) as the sum (range: 0–3) of three dichotomous indicators (0=no, 1=yes) for marital status, religious attendance in the past month, and social club or group attendance in the past month. We coded marital status as “yes” if the participant indicated being presently married or in a marriage-like relationship, and “no” if widowed, divorced, separated, or never married. We categorized those with the largest networks as socially integrated, those with medium networks as moderately integrated, and those with the smallest networks as socially isolated (Kroenke et al., 2020). We based social support on a previously validated measure including nine questions, each with a 5-point Likert scale response option ranging from “none of the time” (1) to “all of the time” (5) describing how often participants had someone available to talk to in various circumstances, for example, when needing someone to listen or give good advice (Stewart et al., 1992). We summed the scores of 1–5 on each of the nine questions, yielding a summary score that ranged from 9–45. Higher summary scores represent greater social support (Stewart et al., 1992), the scale of which is internally consistent (standardized Cronbach α=0.94) (Kroenke et al., 2020; Moser et al., 2012) and commonly used in WHI (Freeborne et al., 2019; Golaszewski et al., 2022; Kroenke et al., 2020; Kroenke et al., 2013; Messina et al., 2004). We assessed optimism using the six-item Life Orientation Test-Revised (e.g., “I’m always hopeful about my future” and “In unclear times, I usually expect the best”), with a Cronbach’s α of 0.78, test-retest reliability of 0.68 and adequate predictive and discriminant validity (Scheier et al., 1994). We computed a summary score (range: 6–30) from the six components, each coded from 1=strongly disagree to 5=strongly agree. Higher scores indicate greater optimism (Scheier et al., 1994).

2.3.4. Clonal Hematopoiesis of Indeterminate Potential

Whole genome sequencing was performed as part of the NHLBI TOPMed project at approximately 30x coverage (The Broad Institute, Boston MA) using DNA extracted from blood drawn at the time of the WHI visit (i.e., SV (1993–1998), AV1 (1995–2001), AV3 (1997–2002), AV6 (2000–2004), and AV9 (2004)) and Illumina HiSeq X technology as previously described (Bick et al., 2020). We identified CHIP using the GATK MuTECT2 somatic variant caller and a pre-specified list of 74 leukemogenic driver mutations known to promote clonal expansion of hematopoietic stem cells (e.g., DNMT3A, TET2, ASKL1, JAK2) as previously described in Bick et al (Bick et al., 2020; Jaiswal et al., 2017; TOPMed, 2021). We applied a variant allele frequency > 0.02 to minimize the false discovery rate (Bick et al., 2020).

2.4. Covariates

Potential confounders of associations between individual- and neighborhood-level SES and CHIP included demographic characteristics (age, self-reported race/ethnicity as a proxy for experiencing systemic racism (Lett et al., 2022), and U.S. Census region), behavioral / clinical attributes (smoking status, alcohol use, physical activity, body mass index, diabetes, and depressive symptoms), interpersonal / intrapersonal resources (social integration, social support, and optimism, defined above), and ancestry principal components. We recorded participant data for these covariates at and between visits via survey and examination following standardized protocols. In analyses, we used self-reported race/ethnicity (American Indian/Alaskan Native, Asian/Pacific Islander, Black/African-American, Hispanic/Latino, white (not of Hispanic origin), and other) and a four-level US Census region (Northeast, South, Midwest, and West) from the screening visit. Otherwise, covariates originated at the at time of blood draw, including age (years); self-reported smoking status (current/former and never); self-reported alcohol use (current/former and never); total energy expenditure based on the self-reported type, duration, and frequency of recreational physical activity (metabolic equivalent of task [MET]-hours/week) (Miller et al., 2020); body mass index (kg/m²) based on measured weight and height; diabetes mellitus (1=yes, 0=no) based on self‐report of physician diagnosis or use of antidiabetic medication; and depressive symptoms assessed using the Burnam eight-item depression screening instrument (Burnam et al., 1988). The instrument consists of six items from the Center for Epidemiologic Studies Depression Scale about the frequency of depressive symptoms in the past week and two items from the National Institute of Mental Health’s Diagnostic Interview Schedule about the duration of symptoms. Higher scores (range: 0–0.99) indicate greater depressive symptomatology. In the WHI, sensitivity and specificity of the Burnam screening instrument for current major depression and dysthymia were 74% and 87%, respectively (Tuunainen et al., 2001). Ancestry principal components were estimated in TOPMed from genome-wide common genotypes as previously described (TOPMed, 2017).

2.5. Statistical Analyses

2.5.1. Multiple Imputation

We leveraged multiple imputation using chained equations (MICE) (Azur et al., 2011) to impute 10 datasets in which missing individual- and neighborhood-level SES and covariate values (range: 0.05%–9.3%) were replaced with values based on patterns of non-missing covariates within participants and the relationships between non-missing covariates among participants. We used the datasets in all analyses to account for the uncertainty of the imputation and ensure correct standard error estimation (Little & Rubin, 2002).

2.5.2. Correction for Sampling Design

We used inverse probability of stroke and venous thromboembolic event case sampling weights to correct all analyses for sampling design and thereby enable inference from the WHI TOPMed sample to the WHI sampling frame from which it was drawn (TOPMed, 2021).

2.5.3. Statistical Modeling

We approximated the risk of CHIP associated with individual- and neighborhood-level SES as an odds ratio (OR) and 95% confidence interval (95% CI) using multivariable logistic regression models adjusted for age, race/ethnicity, U.S. Census region, study participation (OS, CT), randomly assigned hormone therapy treatment arm, smoking status, alcohol use, physical activity, body mass index, diabetes, depressive symptoms, social support, social integration, optimism, and the top 10 ancestry principal components to account for residual confounding due to genetic admixture. We modeled CHIP (1=present, 0=absent) as a function of individual-level SES variables, neighborhood-level SES variables, or neighborhood SES summary z-score, all dichotomized as described above or at their medians. Low individual- and neighborhood-level socioeconomic advantage groups served as referents. We also modeled the linear associations between ordinal-scale, individual-level measures of SES, interval-scale neighborhood-level measures of SES, and CHIP.

We investigated effect modification of the association between SES (individual-level SES and neighborhood SES summary z-score) and CHIP by social integration, social support, and optimism by including interaction terms between them in separate models. For tests of interaction, we modeled social integration (socially isolated/moderately integrated, socially integrated), social support (low/medium, high), optimism (low/medium, high) as binary variables, and SES (individual-level SES and neighborhood SES summary z-score) as binary (dichotomized at the median) and interval-scale variables.

We used a significance level of alpha (α) < 0.05 for tests of association for individual- and neighborhood-level SES with CHIP. We used the Bonferroni method to correct for multiple comparisons (P < α = 0.05/3 = 0.017) (Dunn, 1961) for the three tests of interaction between interpersonal/intrapersonal resources and neighborhood-level SES on CHIP. We completed all analyses using SAS version 9.4 (SAS Institute).

2.5.4. Sensitivity Analyses

In planned sensitivity analyses accounting for selective attrition—a form of bias that can occur when losses to follow-up are influenced by an exposure and outcome or their determinants (Banack et al., 2018)—we upweighted the contribution of participants who remained in WHI over time and were similar to those who did not, thereby compensating for loss to drop out or death (Cole & Hernán, 2008; Hernan et al., 2004; Weuve et al., 2012). We did so because attrition can be concerning in studies of age-related outcomes such as CHIP, given its strong association with morbidity and mortality (Jaiswal et al., 2014; Weuve et al., 2012). In the latter case, participant weights were the inverse product of the estimated probabilities of being alive and participating in the study, conditional on being alive (Supplementary Material). We also addressed the possibility of residual confounding by urbanicity of residence via further adjustment for U.S. Census tract-level, rural-urban commuting area, classified as an urban or metropolitan area; large rural city/town or micropolitan area; small rural town; or isolated small rural town (Cromartie, 2020). Finally, we re-estimated the observed, interval-scale, neighborhood SES summary z score-CHIP associations in post-hoc sensitivity analyses stratified by age (< 70 and ≥ 70 years), race/ethnicity (Black/African-American and white, not of Hispanic origin), and smoking status (current/former and never), hypothesizing that stronger positive associations will be observed among older (≥70 years old), Black/African-American, and currently/formerly smoking women.

3. RESULTS

3.1. Descriptive Statistics

After exclusions, 907 (8.4%) of the 10,799 women in WHI-TOPMed had CHIP (Table 1). Women with CHIP were older (69.4 years vs. 66.1 years), more likely to be white (85.4% vs 82.0%), and more likely to be current or former smokers (52.9 % vs. 48.0%) than women without CHIP. Women with CHIP also were more likely to have a college education (68.0% vs 66.8%), family income ≥ $50,000 (30.6% vs. 35.5%), and technical / sales / administrative / managerial / professional occupation (73.8% vs. 71.2%) than women without CHIP (Table 2). Moreover, women with CHIP were uniformly more likely to live in socioeconomically advantaged neighborhoods than women without CHIP, regardless of how socioeconomic status was measured (Table 3). In contrast, women without CHIP were more likely to be socially integrated (27.3% vs. 22.7%), have high social support (35.2% vs. 34.2%), and have high optimism (34.3% vs. 33.1%) than women with CHIP (Table 1).

Table 1.

Characteristics of Study Participants by CHIP Status, WHI-TOPMed (N=10,799)

	Mean (SE) or N (%)a
Characteristic	CHIP Absent	CHIP Present	P-value
Total	9,892 (91.6)	907 (8.4)
Age (years)	66.1 (0.1)	69.4 (0.2)	<0.0001
Race/ethnicity			0.009
American Indian/Alaskan
Native	44 (0.9)	5 (1.3)
Asian/Pacific Islander	188 (3.5)	12 (3.1)
Black/African-American	1,318 (9.4)	89 (7.7)
Hispanic/Latino	286 (4.2)	17 (2.3)
White (not of Hispanic origin)	8,001 (82.0)	780 (85.4)
Other	56 (0.1)	4 (0.1)
Study / Treatment Arm			0.014
Observational Study	4,034 (40.8)	408 (45.0)
Clinical Trials	5,858 (59.2)	499 (55.0)
Hormone Therapy			0.242
Estrogen	765 (7.7)	71 (7.8)
Estrogen Control	759 (7.7)	79 (8.7)
Estrogen + Progestin	1,175 (11.9)	88 (9.7)
Estrogen + Progestin Control	1,038 (10.5)	88 (9.7)
Not Randomized	6,155 (62.2)	581 (64.1)
Stroke and/or venous thromboembolism	5,278 (53.4)	488 (53.8)	0.796
Smokerb	4,808 (48.0)	468 (52.9)	0.097
Physical activity (MET-hours/week)	12.4 (0.2)	12.1 (0.6)	0.590
Drinkerb	8,706 (89.0)	805 (89.9)	0.397
Body mass index (kg/m2)	28.8 (0.1)	29.0 (0.4)	0.940
Diabetes	846 (5.3)	62 (5.9)	0.082
Depressive symptoms score	0.03 (0.002)	0.02 (0.004)	0.058
Socially integratedc	2,595 (27.3)	237 (22.7)	0.945
High social supportd	3,302 (35.2)	294 (34.2)	0.399
High optimisme	3,276 (34.3)	302 (33.1)	0.882

Abbreviations: CHIP, Clonal hematopoiesis of indeterminate potential; kg/m², kilogram per square meter; MET, metabolic equivalent of task; SE, standard error; WHI-TOPMed, Women’s Health Initiative Trans-Omics for Precision Medicine. All estimates are averages from ten rounds of multiple imputation combined using Rubin’s rule (Little and Rubin, 2002).

^aMeans, standard errors and percentages are weighted for inverse sampling probabilities.

^bCurrent or former.

^cScore ≥ 1.

^dScore ≥ 41.

^eScore ≥ 25.

Table 2.

Individual-level Socioeconomic Status by CHIP Status, WHI-TOPMed (N=10,799)

	N (%)a
Individual Characteristic	CHIP Absent	CHIP Present	P-value
Education ≥ college	6,283 (66.8)	611 (68.0)	0.020
Annual family income ≥ $50,000b	2,865 (35.5)	237 (30.6)	0.040
Technical/sales/administrative/managerial/professional occupation	6,797 (71.2)	637 (73.8)	0.399

Abbreviations: CHIP, clonal hematopoiesis of indeterminate potential; WHI-TOPMed, Women’s Health Initiative Trans-Omics for Precision Medicine. All estimates are averages from ten rounds of multiple imputation combined using Rubin’s rule (Little & Rubin, 2002).

^aPercentages are weighted for inverse sampling probabilities.

^bMedian = $20,000–$34,999.

Table 3.

Neighborhood-level Socioeconomic Status by CHIP Status, WHI-TOPMed (N=10,799)

	N (%) > Mediana
Neighborhood Characteristicb	CHIP Absent	CHIP Present	P-value
% adults aged ≥ 25 years with high school degreec	4,671 (49.7)	476 (53.4)	0.001
% adults aged ≥ 25 years with college degreed	4,733 (49.8)	466 (52.0)	0.042
Median household incomee	4,808 (49.9)	469 (51.8)	0.029
Median home valuef	4,784 (49.6)	469 (54.8)	0.036
% households receiving interest, dividend, or net rental incomeg	4,656 (49.8)	466 (52.8)	0.014
% civilians ≥ 16 years w/ professional/managerial/executive occupationh	4,750 (49.5)	490 (56.5)	<0.001
Neighborhood socioeconomic status summary z-scorei	4,767 (49.7)	475 (54.0)	0.009

Abbreviations: CHIP, Clonal hematopoiesis of indeterminate potential; WHI-TOPMed, Women’s Health initiative Trans-Omics for Precision Medicine. All estimates were averages from ten rounds of multiple imputation combined using Rubin’s rule (Little & Rubin, 2002).

^aPercentages and medians are weighted for inverse sampling probabilities.

^bUS Census tract-level.

Weighted medians: ^c 0.883; ^d 0.373; ^e $49,158; ^f $143,005; ^g 0.446; ^h 0.379; ⁱ −0.384.

3.2. Socioeconomic Status, Social and Personal Resources, and CHIP

In both adjusted and unadjusted models, an individual-level college education and technical / sales / administrative / managerial / professional occupation were positively, but not significantly, associated with CHIP (Table 4). Much the same can be said about the association between neighborhood-level socioeconomic advantage and CHIP (Table 5). Indeed, after adjustment for covariates, including individual-level SES, the estimated risk of CHIP overall was 10% higher among women with a neighborhood SES summary z-score > versus ≤ median: (OR (95% CI), P-value) = (1.10 (0.87, 1.39), P = 0.433). However, optimism significantly modified that estimate (P_Interaction = 0.002), such that among women with low/medium and high levels of optimism, the corresponding ORs (95% CIs) were quite distinct: 1.45 (1.11, 1.89) and 0.65 (0.43, 0.99) (Table 6).

Table 4.

Associations between Individual-level Socioeconomic Status and CHIP: WHI-TOPMed (N=10,799)

	Unadjusted		Adjusteda
Dichotomized-Individual Characteristic	OR (95% CI)	P-value	OR (95% CI)	P-value
Education ≥ college	1.07 (0.84,1.35)	0.596	1.08 (0.82,1.41)b	0.590
Annual family income ≥ $50,000	0.97 (0.73,1.28)	0.814	0.96 (0.72,1.28)c	0.776
Technical/sales/administrative/managerial/professional occupation	1.12 (0.87,1.43)	0.392	1.21 (0.94,1.57)d	0.144

Abbreviations: CHIP, Clonal hematopoiesis of indeterminate potential; OR (95% CI), odds ratio (95% confidence interval); WHI-TOPMed, Women’s Health Initiative Trans-Omics for Precision Medicine.

^aModels adjusted for age, race/ethnicity, U.S. Census region, study participation, randomly assigned treatment arm, smoking, alcohol use, physical activity, body mass index, diabetes, depressive symptoms, social support, social integration, optimism, and ancestry principal components.

Also adjusted for ^b annual family income and occupation; ^c education and occupation; and ^d education and annual family income. All estimates were averages from ten rounds of multiple imputation combined using Rubin’s rule (Little and Rubin, 2002).

Table 5.

Associations between Neighborhood-level Socioeconomic Status and CHIP: WHI-TOPMed (N=10,799)

Median-Dichotomized Neighborhood Characteristic	Unadjusted		Adjusteda
	OR (95% CI)	P-value	OR (95% CI)	P-value
% adults aged ≥ 25 years with high school degreeb	1.16 (0.93,1.45)	0.185	1.07 (0.84,1.36)	0.569
% adults aged ≥ 25 years with college degreec	1.09 (0.88,1.36)	0.431	0.98 (0.77,1.24)	0.869
Median household incomed	1.08 (0.87,1.34)	0.496	1.01 (0.80,1.28)	0.959
Median home valuee	1.23 (0.99,1.53)	0.062	1.11 (0.87,1.40)	0.402
% households receiving interest, dividend, or net rental incomef	1.13 (0.91,1.41)	0.267	1.01 (0.79,1.28)	0.995
% civilians ≥ 16 years with professional/managerial/executive occupationg	1.32 (1.06,1.65)	0.013	1.21 (0.96,1.54)	0.112
Neighborhood socioeconomic status summary z-scoreh	1.19 (0.96,1.48)	0.120	1.10 (0.87,1.39)	0.433

Abbreviations: CHIP, Clonal hematopoiesis of indeterminate potential; GED, general education development; OR (95% CI), odds ratio (95% confidence interval); WHI-TOPMed, Women’s Health Initiative Trans-Omics for Precision Medicine.

^aAll models adjusted for age, race/ethnicity, U.S. Census region, study participation, randomly assigned treatment arm, smoking status, alcohol use, physical activity, body mass index, diabetes, depressive symptoms, social support, social integration, optimism, and ancestry principal components. Models for each neighborhood-level SES measure included all individual-level SES variables (i.e., education level, annual family income, and occupation) and adjustment covariates.

Weighted median: ^b 0.883; ^c 0.373; ^d $49,158; ^e $143,005; ^f 0.446; ^g 0.379; ^h −0.384. All estimates were averages from ten rounds of multiple imputation combined using Rubin’s rule (Little and Rubin, 2002).

Table 6.

Association between the Median-Dichotomized, Neighborhood Socioeconomic Status Summary Z Score and CHIP, by Social and Personal Resource Level: WHI-TOPMed (N=10,799)

Interpersonal and Intrapersonal Resourcea	Adjusted OR (95% CI)b	P for Interactionf
Social integrationc
Socially isolated/moderately integrated	1.07 (0.83,1.39)	0.654
Socially integrated	1.21 (0.76,1.93)
Social supportd
Low/Medium	1.12 (0.84,1.49)	0.847
High	1.07 (0.73,1.58)
Optimisme
Low/Medium	1.45 (1.11,1.89)	0.002
High	0.65 (0.43,0.99)

Abbreviations: CHIP, Clonal hematopoiesis of indeterminate potential; OR (95% CI), odds ratio, (95% confidence interval).

^aSocial integration, sum of three dichotomous indicators for marital status, religious attendance in the past month, and social club or group attendance in the past month; Social support: low, ≤33; medium, 34–40; high, ≥41; Optimism: low, 6–22; medium, 23–24; high, 25–34.

^bAll models adjusted for age, race/ethnicity, U.S. Census region, study participation, randomly assigned treatment arm, smoking status, alcohol use, physical activity, body mass index, diabetes, depressive symptoms, education level, annual family income, occupation, and ancestry principal components.

^cModels for social integration included social support, optimism, and adjustment covariates.

^dModels for social support included social integration, optimism, and adjustment covariates.

^eModels for optimism included social integration, social support, and adjustment covariates.

^fP for interaction is P>Chi-Square for the interaction term between social/individual resources and continuous measure of neighborhood socioeconomic status. All estimates were averages from ten rounds of multiple imputation combined using Rubin’s rule (Little and Rubin, 2002).

The interval-scale neighborhood SES summary z-score also was associated with CHIP. After adjustment, a one-unit (i.e., standard deviation) increase in z-score was significantly associated with a 3% increase in the estimated risk of CHIP: (OR (95% CI), P) = (1.03 (1.00, 1.05), P = 0.038) (Table S1). Among women with low/medium and high levels of optimism, the corresponding ORs (95% CIs) were 1.03 (1.02, 1.04) and 0.95 (0.94, 0.96), (P < 0.001). The interval-scale neighborhood SES summary z-score-CHIP association also varied among women who were socially isolated/moderately integrated versus those who were social integrated and those with low/medium versus high social support: OR (95% CI) = 1.02 (1.00, 1.03) versus 1.00 (0.99, 1.01), (P = 0.017) and 1.01 (1.00, 1.02) versus 0.99 (0.98, 1.00), (P < 0.001), respectively (Table S2).

3.3. Sensitivity Analyses

Associations were robust in planned sensitivity analyses to additional adjustment for selective attrition (Tables S3–S7) and urbanicity of residence (Tables S8–S12). In post-hoc sensitivity analyses, the interval-scale neighborhood SES summary z-score-CHIP associations also remained positive among older (≥70 years old), white, and never smoking women: (OR (95% CI), P-value) = (1.04 (1.00, 1.08), P=0.035), (1.02 (0.99,1.05),P=0.219), and (1.05 (1.01, 1.09), P=0.006), respectively (Tables S13–S21). In addition, associations were robust in analyses substituting ordinal-scale for dichotomized individual-level measures of SES (Tables S22–S28).

4. DISCUSSION

4.1. Global Summary of Study Findings

In this large study of postmenopausal women, across multiple dimensions of individual- and neighborhood-level SES, we observed a higher risk of CHIP among those of high individual- and neighborhood-level SES compared to those of low individual- and neighborhood-level SES. Although not statistically significant, finding a higher risk of CHIP among women of high SES is opposite of what was hypothesized. Moreover, when we captured neighborhood advantage as a multidimensional, interval-scale metric, a one standard deviation increase in the z-score was significantly associated with a 3% increased risk of CHIP, even after adjustment for individual-level socioeconomic status. Although the positive socioeconomic advantage-CHIP associations that we observed were at odds with our a priori hypotheses, we found that they were even stronger among older, white, and never smoking women. However, dispositional optimism protected women residing in socioeconomically advantaged neighborhoods who were otherwise at greatest risk of CHIP, an emerging risk factor for cancer, cardiovascular disease, and mortality.

4.2. Interpretation

Collectively, these observational epidemiologic findings are consistent with the Reserve Capacity Model, which postulates that the detrimental effects of low SES (individual-level SES and neighborhood-level SES) are buffered by high levels of optimism (Gallo, 2009). However, under that model, neighborhood socioeconomic advantage also would have been expected to benefit health, not confer risk of acquiring a CHIP-defining somatic mutation. Furthermore, while the findings also suggest that reduced risk of somatic mutation is a biological pathway by which dispositional optimism protects contextually advantaged but at-risk women against age-related chronic disease, this possibility has not been confirmed by any long-term studies of positive psychological interventions. Unfortunately, extant randomized controlled trials of optimism training tend to be small and have brief interventions, short follow-up periods, and no DNA collections (Celano et al., 2019; Malouff & Schutte, 2017; Mohammadi et al., 2018; Nikrahan et al., 2021; Nikrahan et al., 2019), all of which preclude ideally designed, experimental assessment of optimism-driven change in CHIP.

In lieu of that assessment, alternative explanations for the positive socioeconomic advantage-CHIP associations and their greater strength among older, white, and never smoking women observed herein must be considered. Although the positive associations may be residually confounded, we minimized confounding (e.g., of the individual-level, education-CHIP association) by co-adjusting for each of the two remaining measures of individual-level socioeconomic status (annual family income and occupation) (Blakely et al., 2004; Chi et al., 2016; Kaufman et al., 1997) in addition to demographic characteristics, behavioral / clinical attributes, social / personal resources, and ancestry principal components. We also minimized confounding of the neighborhood-level socioeconomic advantage-CHIP associations by simultaneously adjusting for all three individual-level measures of education, annual family income, and occupation (Kaufman et al., 1997). Moreover, all of the aforementioned associations were equally robust to further adjustment for rural-urban commuting area (Powell-Wiley et al., 2017; Shahar et al., 2019; Xie et al., 2020). Such concerted efforts to minimize confounding reduce the potential magnitude of bias in the adjusted associations.

Another plausible methodological explanation for the positive socioeconomic advantage-CHIP associations observed includes reverse causation related to age. Indeed, age is the strongest known risk factor for CHIP (Heuser et al., 2016) and white, never-smoking, socioeconomically advantaged women live longer than disadvantaged women, thereby increasing their CHIP risk. However, we minimized the potential for reverse causation by controlling for age. It is also possible that attrition bias due to death or drop out after study enrollment distorted our results. Indeed, smoking, diabetes, arthritis, asthma, cancer, cardiovascular disease, hypertension, osteoporosis, and thyroid diseases were all associated with reduced participation in follow-ups, as were social isolation / moderate integration, low / medium social support, and low / medium optimism. Women with lower individual SES (as measured by education and income) and lower neighborhood SES (as measured by neighborhood SES summary z score) also were less likely to participate in follow-ups (Tables S29–S30). We nonetheless accounted for such differences in attrition by including these variables in the logistic regression model used to create inverse probability of attrition weights (IPAW). Then we minimized the potential impact of attrition bias on our study results by using IPAW to weight the logistic regression models of SES-CHIP associations and their modifications by interpersonal and intrapersonal resources. The resulting associations were robust to age adjustment and inverse probability of attrition weighting (Tables S4–S7, S22–S28).

In the absence of methodological explanations for the counterintuitively positive SES-CHIP associations observed herein, it is also tempting to consider putative biological mechanisms. After all, such predisposition may have been driven not by high SES per se, but by the molecular costs incurred by upward mobility among those women previously transitioning from low to high SES. The trade-off by which economic success of women is achieved over the life course, albeit at the expense of their physical health in adulthood, is a phenomenon called skin-deep resilience (Chen et al., 2022). Central to it is the experience of uncontrollable stress (e.g., that related to lack of belonging in high SES environments) and compensatory adoption of unhealthy behaviors (e.g., overeating, physical inactivity, and inadequate sleep) as coping mechanisms (Chen et al., 2022). Although stressors associated with upward mobility dysregulate stress response systems and cause cellular changes such as epigenetic aging (Austin et al., 2018; Chen et al., 2022; Chen et al., 2016), and may thereby increase risk of CHIP, distal SES measures were not uniformly available in this context, leaving unanswered questions regarding the extent to which the observed direction of the neighborhood SES-CHIP association is related to the burdens of upward mobility.

4.3. Limitations and Strengths

This study has several limitations, including its exclusive focus on post-menopausal women, small racial/ethnic subpopulations, and cross-sectional design. However, its inverse-probability of sampling-weighted association estimates enable inference to the much larger geographically and racially / ethnically diverse WHI sampling frame of women from which the study population was drawn. Moreover, these association estimates plausibly reflect the SES-related risk of incident CHIP under a modest rare outcome assumption, in lieu of residual center-or neighborhood-level confounding, in the absence of reverse causation by age, and even in its presence, when simultaneously age-adjusted and attrition-weighted, as described above. Furthermore, such well-adjusted associations of both individual- and neighborhood-level socioeconomic advantage in multiple domains with an emerging risk factor for cancer, cardiovascular disease, and mortality, and their modification by dispositional optimism, clearly distinguish the contributions of this study to the literature. Larger, more diverse populations will nonetheless be poised to enable more powerful confirmatory analyses of such associations as availability of repeated SES, CHIP, and optimism measures evolves over time (Reiner & Whitsel, 2019).

5. CONCLUSIONS

This cross-sectional study examined the potential role of individual- and neighborhood-level socioeconomic status (SES) in clonal hematopoiesis of indeterminate potential (CHIP) and its modification by interpersonal and intrapersonal resources among a diverse group of post-menopausal women. It found that neighborhood socioeconomic advantage was positively associated with increased risk of CHIP, a counterintuitive association that may be related to the molecular costs of upward mobility and attenuated by dispositional optimism. We therefore posit that reduced risk of CHIP-defining somatic mutation could represent a biological pathway by which optimism protects contextually advantaged but at-risk women against age-related chronic disease. Large, longitudinal studies of SES, optimism, and incident somatic mutation in diverse cohorts and longer-term, randomized controlled trials of optimism training with repeated DNA collections could help assess this possibility.

6. IMPLICATIONS FOR PRACTICE AND/OR POLICY

This study found that high levels of optimism protect women from socioeconomically advantaged neighborhoods against CHIP, an outcome for which they are at increased risk. Therefore, long-term, positive psychological interventions maybe effective in lowering the risk of age-related chronic diseases among women at risk for CHIP.

Biographies

Shelly-Ann Love, PhD, MS, is an Epidemiologist at Social & Scientific Systems, Inc., a DLH Holdings Company and a Research Collaborator at the University of North Carolina at Chapel Hill. Her research focuses on environmental and genetic determinants of ageing outcomes.

Jason Collins, MPH, is a Doctoral Candidate and Research Assistant in the Department of Epidemiology at the University of North Carolina at Chapel Hill. His research focuses on environmental risk factors of cardiovascular diseases.

Kurtis M. Anthony, MPH, is a Doctoral Student and Research Assistant in the Department of Epidemiology at the University of North Carolina at Chapel Hill. His research focuses on behavioral and environmental risk factors of cardiovascular disease.

Sophie Buchheit is an Undergraduate Student at Brown University a Research Assistant in the Department of Epidemiology at the University of North Carolina at Chapel Hill. Her research focuses on environmental risk factors of cardiovascular disease.

Ebonee N. Butler, PhD, MPH, is an Assistant Professor in the Department of Epidemiology at the University of North Carolina at Chapel Hill. Her research focuses on cancer epidemiology and etiological studies.

Ganga S. Bey, PhD, MPH, is a Postdoctoral Fellow in the Department of Epidemiology at the University of North Carolina at Chapel Hill. Her research focuses on social determinants of cardiovascular disease.

Rahul Gondalia, PhD, MPH, is a Senior Epidemiologist at IQVIA and a Research Collaborator in the Department of Epidemiology at the University of North Carolina at Chapel Hill. His research focuses on injury epidemiology and environmental risks factors of cardiovascular disease.

Kathleen Hayden, PhD, MA, is a Professor in Social Sciences and Health Policy at Wake Forest University. Her research focuses on predictors of neurocognitive outcomes.

Anthony Zannas, MD, PhD, MSc, is an Assistant Professor in the Department of Psychiatry at the University of North Carolina at Chapel Hill. His research focuses on how epigenetic changes contribute to stress-related somatic and behavioral phenotypes.

Alexander G. Bick, MD, PhD, is an Assistant Professor in the Department of Medicine at Vanderbilt University Medical Center. His research focuses on understanding how the interplay between inherited germline genetic factors and acquired somatic mutations contributes to disease.

JoAnn E. Manson, MD, DrPH, MPH, is Professor of Epidemiology at Harvard University and Professor of Medicine at Brigham and Women’s Hospital. Her research focuses on women’s health, chronic disease epidemiology, type 2 diabetes mellitus, and lifestyle and endocrinologic determinants of breast cancer.

Pinkal M. Desai, MD, MPH, is an Assistant Professor of Medicine at Weill Cornell Medical College and Assistant Attending Physician at the New York-Presbyterian Hospital. Her research focuses on leukemia, myelodysplastic syndrome, myeloproliferative neoplasms, and clonal hematopoiesis.

Pradeep Natarajan, MD, MMSC, is an Associate Professor of Medicine at Harvard University and Director of Preventive Cardiology at Massachusetts General Hospital. His research focuses on using human genetic variation, biomedical informatics, integrative genomics, and genotype-based phenotyping to investigate cardiometabolic traits.

Romit Bhattacharya, MD, is a Cardiologist at the Massachusetts General Hospital and researcher at the Broad Institute of Harvard and MIT. His research focuses on comprehensive cardiometabolic risk profiling using a combination of genomic risk and lifestyle risk factors.

Siddhartha Jaiswal, MD, PhD, is an Assistant Professor of Pathology at Stanford University. His research focuses on understanding the biology of the aging hematopoietic system.

Ana Barac, MD, PhD, is an Associate Professor of Medicine and Oncology at Georgetown University and the Director of MedStar Heart and Vascular Institute’s Cardio-Oncology program. Her research focuses on the mechanisms, diagnosis and treatment of cardiac complications of cancer therapies.

Alex Reiner, MD, MSc, is a Research Professor of Epidemiology at the University of Washington. His research focuses on the genetic epidemiology of atherosclerotic, thrombotic and cardiovascular disease; population biology; and genomics.

Charles Kooperberg, PhD, is Professor and Program Head of the Biostatistics Program Public Health Sciences Division at Fred Hutch. His research specializes in statistical genetics analysis of complex data to better understand the genes and proteins that underlie human disease.

James D. Stewart, MA, is a Senior Spatial Analyst in the Department of Epidemiology at the University of North Carolina at Chapel Hill. Mr. Stewart work utilizes spatial analysis to generate pollution exposure estimates and related variables for genetic epidemiology studies.

Eric A. Whitsel, MD, MPH is a Professor of Epidemiology and Medicine at the University of North Carolina at Chapel Hill. His research focuses on electrocardiography, CVD surveillance, environmental epidemiology, genetics and epigenetics.