Association of Neighborhood and Environmental Factors with Clinical Phenotypes and Outcomes in Heart Failure with Preserved Ejection Fraction

Heart failure with preserved ejection fraction (HFpEF) is heterogeneous with multiple comorbidities and limited therapeutic options.¹ Multiple pathologies contribute to the development of distinct clinical HFpEF phenogroups.¹ Evidence suggests that social determinants of health (SDoH) are pivotal in the pathogenesis of cardiovascular disease (CVD).² Defined by the Centers for Disease Control & Prevention and the World Health Organization, SDoH refer to “The conditions in the environments where people are born, live, learn, work, play, worship, and age, influencing health outcomes and quality-of-life risks”. SDoH encompasses five domains: (i) Education Access & Quality, (ii) Economic Stability, (iii) Social & Community Context, (iv) Health Care Access & Quality, (v) Neighborhood & Built Environment.

Studies how SDoH and environmental factors (Figure A1) impact HFpEF phenotypes are limited. Conceptualizing Neighborhoods & Built Environments is essential to understand how SDoH contribute to CVD health inequities.³ To decipher how SDoH impact HFpEF, we determined population-level neighborhood environments (NEnv) of California from Census Data. Second, to determine Built Environments, we used the publicly available datasets of the California Office of Environmental Health Hazard Assessment, “Crimegrade”, and “Howloud”. Third, we conducted retrospective analyses of HFpEF patients at UC Davis to correlate NEnvs with their phenotypes.

Figure. Neighborhood and Environmental Stressors Are Correlated to Distinct HFpEF Phenotypes.

A1. HFpEF patients’ phenotype and outcome may be determined by 5 domains of Social Determinants of Health (SDoH), adapted from the World Health Organization (WHO) and the Centers for Disease Control & Prevention (CDC). A2. We clustered communities into 4 NEnvs using socio-economic and demographic variables. Clusters were computed using Python (with the following libraries: Pandas, Numpy, Scipy, Sklearn, StatModels, Seaborn, GeoPandas). A3. Each sociodemographic and economic variable on the heatmap can be assigned to the 5 domains of SDoH. We also considered race/ethnicity as demographic features next to SDoH. NEnv-1 is predominantly rural, mostly resided by White and Hispanic individuals, with lower income, health insurance coverage, college education attainment, and employment. NEnv-2 is also rural, predominantly White individuals, lower income and advanced age. NEnv-3 is urban, predominantly White & Asian individuals, higher income, (private) health insurance and education. NEnv-4 is urban, lowest portion of White individuals, highest portion of Hispanic and Black community, lowest income, low health insurance, lower education-level. Numbers in this table (and for A4) are the difference between Cluster-mean-score and Overall-mean-score (original units from Census data were converted to Z-scores), where red shading indicates positive z-scores and blue, negative z-scores (color-scale-bar normalized between −1 and 1). Difference between scores were evaluated using nonparametric Kruskal-Wallis test; asterisk (*) indicates statistical difference compared to NEnv-4 (P<0.05). A4. Air-pollution, lead-levels, noise-pollution, drinking-water-pollution, and crime-rates exhibited the highest levels in NEnv-4 and the lowest in NEnv-3, while pesticide-pollution showed the highest rate in NEnv-1 (color-scale bar normalized between minimum −0.6 to maximum 0.5). B. We mapped 6,850 HFpEF patients onto the 4 NEnv’s to analyze their phenotype and neighborhood-correlation. Median Age was higher in NEnv-1 and 3, and younger in NEnv-4. Sex was equally distributed across NEnv 1–4. Hypertension had the highest prevalence in NEnv-3 (70%) and was >50% in all groups. NEnv-4 displayed the highest prevalence for Diabetes, Obesity and Chronic kidney disease. NEnv-1 and 3 displayed highest prevalence for atrial fibrillation. Creatinine was higher in NEnv-4 vs. NEnv 1–3, while glomerular filtration rate (GFR) was lower in Env-4. Kruskal-Wallis and Chi-Square tests were used for continuous and categorical variables, respectively. C. B-type natriuretic peptide (BNP) levels were highest in NEnv-1 and 4 (Mean [95% CI], 5.7 pg/dL, [5.6–5.7] and 5.5 pg/dL [5.5–5.6], respectively). D. Left ventricular mass index to height (LVMI) was significantly higher in patients residing in NEnv-4 when compared to the rest of the groups. E. Cox regression survival analysis adjusted for age and sex showed that patients residing in NEnv-1 & 4 with lower income and more environmental stressors had higher mortality than patients residing in NEnv-3 with higher income and milder environmental stressors.

Our community clustering algorithm^4,5 used 24 sociodemographic and socioeconomic variables previously described,⁵ across 5 main domains of SDoH, at ZIP Code Tabulation Areas (ZCTA) level (from the American Community Survey, https://data.census.gov/cedsci/ [accessed 2022-03-07]). Prominent variables were Education level, Access to healthcare, Household income, and Employment. We also included Race/ethnicity as demographic features. An unsupervised machine learning algorithm stratified the variables into community clusters, unveiling 4 main NEnvs (1–4), each connected to ZCTAs (Figure A2 & A3). NEnv-1 (602 ZCTAs) is predominantly rural, mostly resided by White and Hispanic individuals, with lower income, health insurance coverage, college education attainment, and employment. NEnv-2 (239 ZCTAs) is rural, predominantly older White individuals with low employment. The other two NEnvs are urban. NEnv-3 (481 ZCTAs) is mainly resided by White and Asian individuals, with fewer Hispanic individuals and a higher median age, higher rates of employment, income, education, and access to health-insurance. NEnv-4 (418 ZCTAs) has the highest numbers of underserved multiethnic populations with lower income and moderate employment rate. NEnv-4 displays the highest rates for lead, air-pollution, crime (neighborhood & property), ozone-gas, drinking-water-pollution, and chemical-waste, especially compared to NEnv-3. Pesticide exposure was highest in NEnv-1 (Figure A4).

We mapped 6,850 HFpEF patients at UC Davis to their ZCTAs to analyze their phenotype-neighborhood-correlation (Figure B). We found that 1,188 patients resided in NEnv-1, 418 in NEnv-2, 2,704 in NEnv-3, and 2,539 in NEnv-4. Statistical analysis was conducted with SAS 9.4. As expected from our clustering of NEnv that includes race/ethnicity as variables, while 53% of 6,850 were White patients, NEnv-1 exhibited the highest percentage of White patients (71.3%) followed by NEnv-3 (60.7%). NEnv-4 had the lowest percentage of White (36.2%) patients and the highest percentages of Black (21%) and Hispanic (16.7%) patients compared to NEnvs1 to 3 (~10% or lower, p=3.0×10^−160). Notably, we observed remarkable phenotype differences across NEnv’s. Patients from NEnv 2 and 3 were older (median [IQR] age 70.0 [59.0–80.0] and 73.0 [63.0–81.0] years, respectively) vs. NEnv-2 and 4 (69.1 and 66.6 years, respectively). Second, NEnv-4 exhibited the highest percentage of diabetes (34.6%) and obesity (44.7%). NEnv-3 displayed the highest hyperlipidemia prevalence (50.8%). Prevalence of atrial fibrillation was higher in NEnv-1 and 3 (23% and 22.8% respectively) vs. NEnv-2 and 4 (16.7% and 12.2% respectively) likely due to the more advanced age. Third, while most laboratory values were not-statistically-significant (NS) across the NEnvs, median [IQR] creatinine 1.1 [0.8–1.8] mg/dL was higher and GFR of 52.0 [31.0–59.0] mL/min was lower in NEnv 4 vs. NEnv 1–3, indicating a higher risk for kidney dysfunction. Median [IQR] B-type natriuretic peptide levels were significantly higher in NEnv-1 and 4 which could indicate different HF severity and/or phenotypes (Figure C). Fourth, echocardiographic parameters including tricuspid annular pulmonary systolic excursion (TAPSE) and left ventricular ejection fraction (LVEF) were NS across the 4 NEnvs (mean LVEF of 60%), however, left ventricular mass index (LVMI) was significantly higher in HFpEF patients in NEnv-4 (p=0.004) (Figure D). Finally, Survival analysis (adjusted for age) showed that HFpEF patients from NEnv-3 (urban, higher-income) had the highest survival compared to NEnv-1 (rural, lower-income, high Pesticide) and NEnv-4 (urban, lower-income, excessive environmental stressors) over 8 years (p<0.001) (Figure E). It is critical that our study is interpreted with caution. Although NEnvs were partly determined by race/ethnicity, area-level race/ethnicity do not equate to SDoH and are not modifiable factors to improve health outcomes. SDoH are upstream modifiable factors known to explain differences in CVD among race/ethnic groups.

Summary:

HFpEF patients exhibit phenotypic differences in survival associated with NEnvs. Hence, Neighborhoods and Built Environments in SDoH, can stratify health-risk for HFpEF patients, which may be reduced by targeted community-driven interventions.