Neighborhood‐Level Socially Determined Vulnerabilities and Outcomes After Heart Failure Hospitalization

Abstract

Background

Outcomes are poor for older adults following a hospitalization for heart failure (HF). Neighborhood‐level socially determined vulnerabilities could contribute to this, but this is not well studied. We aimed to determine associations between neighborhood‐level socially determined vulnerabilities (density of clinical treatment facilities, physical activity facilities, and unhealthy food sources) and outcomes following HF hospitalization.

Methods

We included 366 Medicare beneficiaries in the REGARDS (Reasons for Geographic and Racial Differences in Stroke) cohort study aged ≥65 years and living in urban areas, who were discharged to home following HF hospitalization between 2003 and 2017. The main exposures were low versus high density (based on median) of clinical treatment facilities, physical activity facilities, and unhealthy food sources within 1 km of each participants' home. The primary outcomes were 90‐day and 6‐month composite of all‐cause mortality or readmission.

Results

Living in an area with high density of unhealthy food sources was associated with a 6‐month composite of all‐cause mortality and hospitalization (adjusted hazard ratio [aHR], 1.47 [95% CI, 1.09–1.98]). This was largely driven by readmission (aHR, 1.52 [95% CI, 1.12–2.07]). There were no associations between density of clinical treatment or physical activity facilities with outcomes at either time point.

Conclusions

Higher density of unhealthy food sources in a participant's neighborhood environment is associated with worse outcomes after HF hospitalization among older adults living in urban environments. Neighborhood‐level interventions, such as those that can mitigate exposure to unhealthy foods, may be important to improve posthospitalization outcomes in people living with HF.

Nonstandard Abbreviations and Acronyms

REGARDS

Reasons for Geographic and Racial Differences in Stroke

SDV

socially determined vulnerabilities

Clinical Perspective.

What Is New?

• High neighborhood density of unhealthy food sources is associated with worse posthospitalization outcomes.

What Are the Clinical Implications?

• Neighborhood‐level interventions to mitigate exposure to unhealthy foods could be an important modifiable target to improve outcomes.

Outcomes are poor for older adults following a hospitalization for heart failure (HF). Indeed, ≈20% will be readmitted ¹ and 11% will die ² , ³ within 30 days, >30% will be readmitted ¹ and 15% to 20% will die ³ within 90 days, and >40% will be readmitted and >20% will die within 6 months. ⁴ , ⁵ Although several interventions aimed at improving posthospitalization outcomes at the individual, hospital, and policy level have been studied and in some cases implemented, posthospitalization outcomes (namely readmission rates) remain largely unchanged over the past 10+ years. ¹ This supports the urgent need to identify novel modifiable factors that contribute to posthospitalization outcomes, which can be leveraged toward the development of effective interventions for this vulnerable subpopulation.

Socially determined vulnerabilities (SDV) have emerged as important influences on myriad health outcomes in HF including posthospitalization outcomes. ⁶ , ⁷ SDV exist at multiple levels, including the individual, neighborhood/community, health system, and sociopolitical levels. ⁸ , ⁹ The neighborhood level may be particularly relevant to facilitate health behaviors that are important following a HF hospitalization. For example, access to postdischarge medical follow‐up, green space to engage in physical exercise, and available foods without excess salt may be beneficial neighborhood‐level SDV. Yet, studies specifically examining the associations of access to these health‐related resources with post‐HF hospitalization outcomes are lacking. In the current study, we examined whether 3 neighborhood‐level SDV (density of clinical treatment facilities, density of exercise facilities, and density of unhealthy food sources near participant residence) were associated with posthospitalization outcomes for older adults hospitalized for HF.

METHODS

To abide by its obligations with National Institutes of Health/National Institute of Neurological Disorders and Stroke and the Institutional Review Board of the University of Alabama at Birmingham, REGARDS (Reasons for Geographic and Racial Differences in Stroke) facilitates data sharing through formal data use agreements. Any investigator is welcome to access the REGARDS data and documentation through this process. Requests for data access may be sent to the REGARDS study at regardsadmin@uab.edu. The study was approved by the institutional review boards of the participating institutions.

Study Population

We examined participants from the REGARDS cohort study, which originally enrolled 30 239 self‐reported Black and White participants aged at least 45 years from the 48 contiguous states and Washington, DC, in 2003 to 2007 and has since had longitudinal follow‐up. As part of this cohort study, participants were asked by computer‐assisted telephone interview to provide baseline information on their demographics and medical history through a telephone interview. Approximately 1 month after their interview, they had in‐home visits to collect additional data such as medications, physical examination, and urine and blood samples. Over follow‐up, participants were contacted every 6 months, and medical records for potential cardiovascular hospitalizations were retrieved for adjudication by 2 clinicians to determine study outcomes, including HF hospitalization. Disagreements were resolved by an expert committee. All participants provided informed consent and the protocol for the study was approved by the participating institutions' ethics committees. Further details have previously been published. ⁶ , ⁷ , ¹⁰ , ¹¹ For each HF hospitalization, clinical variables were abstracted as previously described. ¹¹

For this study, we selected participants with an adjudicated HF hospitalization in 2003 to 2017 who were discharged to home, who were at least 65 years of age, and who had continuous Medicare parts A and B coverage during the period of follow‐up (90 days; and 6 months) (Figure 1). Because patterns are likely to be very different in rural compared with urban areas, ¹² , ¹³ , ¹⁴ we decided to focus specifically on participants living in urban areas, defined as areas where ≥75% of the US census tract population reside inside urban areas/clusters. We excluded participants who died during the index hospitalization, those discharged with hospice services, and those with missing neighborhood‐level SDV data. For participants with multiple hospitalizations, we examined the first hospitalization only.

Figure 1. Inclusion and exclusion cascade.

HF indicates heart failure; and RECVD, Retail Environment and Cardiovascular Disease study.

Main Exposures

The exposures of interest included 3 neighborhood‐level SDV: density of clinical treatment facilities, density of exercise facilities, and density of unhealthy food sources within 1 km of the participant's home. For each category, we calculated a median and defined high density as greater than the median, and low density as less than or equal to the median.

Data on neighborhood‐level SDV were obtained via the RECVD (Retail Environment and Cardiovascular Disease) ancillary study, which categorizes retail location data from the National Establishment Time Series data set into health‐related categories and links them to REGARDS participants. ¹⁵ Within this data set, medical facilities are individually identified by Standard Industrial Classification codes and then categorized into specific types of facilities. Word and name‐based searches employing keywords or names from chain name lists were also used for categorization. Clinical treatment facilities are defined as an “all clinical treatment facilities” category, which includes hospital‐based inpatient facilities, community‐based facilities (retail clinics, urgent care), physical therapy facilities, kidney centers, mental and behavioral health facilities, dental facilities, and ambulatory offices or clinics of health practitioners. Physical activity facilities included light/moderate physical activity facilities (based on a metabolic equivalent task value of 1.6–5.9 assigned to each Standard Industrial Classification code per the Compendium of Physical Activities ¹⁶ ), vigorous physical activity facilities (>5.9 metabolic equivalent tasks), and multiuse facilities (those that offer a wide range of activities or the specific activity could not be ascertained). Unhealthy food sources included small grocers/bodegas, bakery/candy/ice cream facilities, convenience stores, fast food restaurants, pizza restaurants, and coffee shops. Further details regarding categorization and a catalogue of variables with definitions can be found on the National Establishment Time Series website. ¹⁷ Densities were calculated by buffering 1 km around geocoded REGARDS participant residence locations (based on the address before and as close to the hospitalization date as possible), counting facilities using National Establishment Time Series data from the year of hospitalization (or from 2014 if hospitalized after 2014), then dividing by the area of the land in the buffer (excluding inland and coastal water bodies), as previously described. ¹⁸

Outcomes

Our primary outcomes were the composite of all‐cause mortality or readmission at 90 days and, separately, 6 months posthospitalization (which included all events during the period including those that occurred within 90 days). Secondary outcomes included individual components of the composite outcomes.

Covariates

Covariates for adjustment included sex, self‐reported race (Black versus White), and geographic region; social determinants of health including low household income (defined as < $35 000), low education (defined as less than high school), and living in a zip code with a high poverty level (defined as a zip code whereby >25% of residents live below federal poverty line); Get With The Guidelines score, which is a validated risk score for posthospitalization outcomes that includes age and other relevant clinical parameters ¹⁹ ; and discharge hospital bed size.

Statistical Analysis

Baseline characteristics were stratified by density (low versus high) of each exposure variable. Continuous variables were represented as medians and interquartile ranges (IQR), and categorical variables were represented as counts and percentages. The 90‐day and 6‐month composite outcomes (total number and percentages) were determined in each group. Unadjusted survival curves were subsequently estimated using the Kaplan–Meier method. Adjusted Cox regression models with complete case analysis were used to estimate the association of outcomes with the density of clinical treatment facilities, exercise facilities, and unhealthy food sources in a 1‐km buffer of the participant's address. The proportional hazards assumption was evaluated using the estat phtest function in Stata 18, which is a test based on fitting a generalized linear regression of scaled Schoenfeld's residuals +regression coefficient (β) on time; it tests for a null hypothesis of slope=0; a significant nonzero slope indicates violation of proportional hazards assumption.

We conducted a sensitivity analysis where multiple imputation with chained equations and Rubin's rules to summarize estimation results across 10 imputations was used for missing covariates. We also conducted sensitivity analyses in which we repeated all analyses whereby the exposure variables were defined based on density within a 5‐km buffer of participant address.

Because 3 different exposures for each outcome were examined, a Bonferroni correction was used whereby a P value <0.016 was considered statistically significant for each outcome. All analyses were conducted in R 4.2.1. ²⁰

Dr Goyal had full access to all the data in the study and takes responsibility for their integrity and the data analysis. Because of the sensitive nature of the data collected for this study, requests to access REGARDS study data from qualified researchers trained in human subject confidentiality protocols may be sent to the REGARDS study at regardsadmin@uab.edu. The data use agreement with the Centers for Medicare and Medicaid Services does not allow the authors to share participant‐level data on Medicare claims. Data management and statistical code is available upon request to the corresponding author.

RESULTS

The study sample included 366 participants (Figure 1). Baseline characteristics for each study group are listed in Table 1. The median age at time of HF hospitalization was 77 years (IQR, 73–83) and 45.9% of participants were women. Within the cohort, 36.6% were Black, 52.5% had low annual household income, 18.6% had less than a high school education, and 19.8% lived in a zip code with a high poverty level. In addition, there was a significant burden of social isolation (73.2%), hypertension (79.2%), and coronary heart disease (68.3%).

Table 1.

Baseline Characteristics^*

	Overall (N=366)	Density of clinical treatment facilities		Density of physical activity facilities		Density of unhealthy food sources
		Low (N=183)	High (N=183)	Low (N=183)	High (N=183)	Low (N=184)	High (N=182)
Demographics
Age at admission, y, median (IQR)	77 (73–83)	77 (72–82)	78 (73–84)	78 (73–83)	77 (72–82)	78 (74–83)	77 (72–82)
Female sex	168 (45.9%)	80 (43.7%)	88 (48.1%)	78 (42.6%)	90 (49.2%)	89 (48.4%)	79 (43.4%)
Black race	134 (36.6%)	79 (43.2%)	55 (30.1%)	70 (38.3%)	64 (35.0%)	57 (31.0%)	77 (42.3%)
Income < $35 000	192 (52.5%)	103 (56.3%)	89 (48.6%)	95 (51.9%)	97 (53.0%)	89 (48.4%)	103 (56.6%)
Unknown	46 (12.6%)	17 (9.3%)	29 (15.8%)	21 (11.5%)	25 (13.7%)	27 (14.7%)	19 (10.4%)
Less than high school education	68 (18.6%)	44 (24.0%)	24 (13.1%)	43 (23.5%)	25 (13.7%)	32 (17.4%)	36 (19.8%)
Unknown	3 (0.8%)	0 (0%)	3 (1.6%)	1 (0.5%)	2 (1.1%)	3 (1.6%)	0 (0%)
Southeast region residency	190 (51.9%)	109 (59.6%)	81 (44.3%)	114 (62.3%)	76 (41.5%)	110 (59.8%)	80 (44.0%)
Zip code level poverty	72 (19.8%)	35 (19.1%)	37 (20.6%)	37 (20.3%)	35 (19.3%)	25 (13.8%)	47 (25.8%)
Lack of public health infrastructure	123 (33.6%)	66 (36.1%)	57 (31.1%)	68 (37.2%)	55 (30.1%)	67 (36.4%)	56 (30.8%)
Social isolation	268 (73.2%)	133 (72.7%)	135 (73.8%)	135 (73.8%)	133 (72.7%)	127 (69.0%)	141 (77.5%)
Hypertension	289 (79.2%)	139 (76.0%)	150 (82.4%)	136 (74.7%)	153 (83.6%)	137 (74.9%)	152 (83.5%)
Diabetes	137 (38.8%)	77 (43.5%)	60 (34.1%)	70 (39.5%)	67 (38.1%)	69 (38.5%)	68 (39.1%)
Heart failure subtype
Preserved EF (≥ 50%)	134 (36.6%)	67 (36.6%)	67 (36.6%)	65 (35.5%)	69 (37.7%)	73 (39.7%)	61 (33.5%)
Midrange EF (41–49%)	30 (8.2%)	9 (4.9%)	21 (11.5%)	16 (8.7%)	14 (7.7%)	11 (6.0%)	19 (10.4%)
Reduced EF (≤ 40%)	144 (39.9%)	76 (41.5%)	68 (37.2%)	67 (36.6%)	77 (42.1%)	72 (39.1%)	72 (39.6%)
Unknown	58 (15.8%)	31 (16.9%)	27 (14.8%)	35 (19.1%)	23 (12.6%)	28 (15.2%)	30 (16.5%)
Coronary heart disease	250 (68.3%)	127 (69.4%)	123 (67.2%)	119 (65.0%)	131 (71.6%)	119 (64.7%)	131 (72.0%)
Atrial fibrillation	72 (20.4%)	36 (20.7%)	36 (20.1%)	37 (21.3%)	35 (19.6%)	40 (22.9%)	32 (18.0%)
GWTG score, median (IQR)	39 (34–45)	38 (33–43)	41 (35–46)	39 (35–44)	39 (34–45)	39 (35–45)	39 (34–44)
GWTG score category
0–33	77 (21.6%)	48 (26.7%)	29 (16.5%)	40 (22.3%)	37 (20.9%)	38 (21.1%)	39 (22.2%)
34–50	249 (69.9%)	122 (67.8%)	127 (72.2%)	123 (68.7%)	126 (71.2%)	122 (67.8%)	127 (72.2%)
51–57	28 (7.9%)	10 (5.6%)	18 (10.2%)	15 (8.4%)	13 (7.3%)	19 (10.6%)	9 (5.1%)
58–61	1 (0.3%)	0 (0.0%)	1 (0.6%)	0 (0.0%)	1 (0.6%)	0 (0.0%)	1 (0.6%)
62–65	1 (0.3%)	0 (0.0%)	1 (0.6%)	1 (0.6%)	0 (0.0%)	1 (0.6%)	0 (0.0%)
Unknown	10	3	7	4	6	4	6
Hospital bed size, median (IQR)	368 (227–565)	362 (235–545)	372 (220–569)	341 (225–536)	394 (228–587)	343 (220–544)	384 (263–581)

EF indicates ejection fraction; GWTG, Get With The Guidelines; and IQR, interquartile range.

^*Categorical variables displayed as n (%) and continuous variables displayed as median (IQR).

Within a 1‐km buffer, there were a median of 1.6 (IQR, 0.3–4.7) clinical treatment facilities, 0.3 (IQR, 0–1.0) physical activity facilities, and 1.9 (IQR, 0.6–3.5) unhealthy food stores. Participants who lived in an area with low density of clinical treatment facilities were more likely to be Black, to have less than a high school education level, and to be admitted to a hospital with a lower Get With The Guidelines score (Table 1). Participants who lived in an area with low density of physical activity facilities were more likely to have less than a high school education. Participants who lived in an area with high density of unhealthy food sources were more likely to be Black and live in a zip code with high poverty.

Regarding outcomes, 40% of participants experienced a composite 90‐day outcome (11% died and 37% were readmitted), and 55% experienced a composite 6‐month outcome (16% died and 52% were readmitted [Table S1]). Figure 2 shows a Kaplan–Meier curve of 90‐day outcomes for each exposure, and Figure 3 shows a Kaplan–Meier curve of 6‐month outcomes for each exposure.

Figure 2. Ninety‐day composite outcome and density of clinical treatment facilities (A), physical activity facilities (B), and unhealthy food sources (C) within 1 km.

Figure 3. Six‐month composite outcome and density of clinical treatment facilities (A), physical activity facilities (B), and unhealthy food sources (C) within 1 km.

Participants living in areas with a high density of unhealthy food sources had increased risk for the 90‐day composite outcome compared with those living in areas with a low density of unhealthy food sources (unadjusted hazard ratio [HR], 1.46 [95% CI, 1.05–2.03]). In an adjusted model, this risk remained elevated (adjusted HR, 1.50 [95% CI, 1.06–2.12], P=0.021). This finding was largely driven by increased rates of readmission (adjusted HR, 1.54 [95% CI, 1.08–2.20], P=0.018) (Table 2). Of note, with Bonferroni correction, the composite outcome findings did not formally meet criteria for statistical significance.

Table 2.

Associations of Neighborhood‐Level Social Determinants of Health (Within 1 km) With Outcomes at 90 Days and 6 Months

	Composite outcome			Readmission			Mortality
	Unadjusted HR (95% CI)	Adjusted HR (95% CI)	P value	Unadjusted HR (95% CI)	Adjusted HR (95% CI)	P value	Unadjusted HR (95% CI)	Adjusted HR (95% CI)	P value
90‐d outcomes (N=351)
High density of clinical treatment facilities	1.17 (0.84, 1.62)	1.15 (0.81, 1.62)	0.4	1.28 (0.92, 1.80)	1.25 (0.87, 1.79)	0.2	0.95 (0.50, 1.77)	0.90 (0.45, 1.78)	0.8
High density of physical activity facilities	1.18 (0.85, 1.63)	1.21 (0.86, 1.71)	0.3	1.25 (0.89, 1.75)	1.27 (0.89, 1.81)	0.2	0.95 (0.51, 1.79)	1.04 (0.53, 2.03)	0.9
High density of unhealthy food sources	1.46 (1.05, 2.03)	1.50 (1.06, 2.12)	0.021	1.5 (1.07, 2.11)	1.54 (1.08, 2.20)	0.018	0.87 (0.47, 1.64)	0.88 (0.45, 1.70)	0.7
6‐mo outcomes (N=345)
High density of clinical treatment facilities	1.15 (0.87, 1.52)	1.12 (0.83, 1.52)	0.4	1.26 (0.95, 1.69)	1.25 (0.91, 1.70)	0.2	1.15 (0.70, 1.92)	1.10 (0.64, 1.90)	0.7
High density of physical activity facilities	1.06 (0.80, 1.40)	1.06 (0.79, 1.43)	0.7	1.12 (0.84, 1.50)	1.12 (0.82, 1.52)	0.5	1.07 (0.64, 1.77)	1.14 (0.67, 1.97)	0.6
High density of unhealthy food sources	1.45 (1.10, 1.92)	1.47 (1.09, 1.98)	0.011	1.51 (1.13, 2.01)	1.52 (1.12, 2.07)	0.008	0.80 (0.48, 1.33)	0.78 (0.46, 1.33)	0.4

Adjusted Cox proportional hazards regression models include covariates of sex, self‐reported race, geographic region, income, education, zip code level poverty, Get With The Guidelines score, and discharge hospital bed size. Complete case analysis was used. HR indicates hazard ratio.

Participants living in areas with a high density of unhealthy food sources also had increased risk for the 6‐month composite outcome compared with those living in areas with a low density of unhealthy food sources (unadjusted HR, 1.45 [95% CI, 1.10–1.92]). In an adjusted model, this association remained (HR, 1.47 [95% CI, 1.09–1.98], P=0.011). This finding was largely driven by increased rates of readmission (adjusted HR, 1.52 [95% CI, 1.12–2.07], P=0.008) (Table 2). There were no significant associations observed for either density of clinical treatment facilities or physical activity facilities for either the 90‐day or 6‐month composite outcomes (Table 2).

Of note, the proportional hazards assumption was not violated in any of our models. Table S2 shows the HR of each outcome for each covariate included in the model. In a sensitivity analysis where we conducted multiple imputation for missing covariates (only 4% were missing), findings were similar (Table S3). In a sensitivity analysis where we defined density based on a 5‐km buffer of participant residence, we did not find any associations between any of the neighborhood‐level SDV (density of clinical treatment facilities, physical activity facilities, or unhealthy food sources) and posthospitalization outcomes at either 90 days or 6 months (Table S4).

DISCUSSION

In this analysis of older adults living in urban areas who were hospitalized with HF, we observed that a high density of unhealthy food sources within 1 km of participant residence was associated with a 52% higher risk of experiencing a readmission at 6 months. This has important implications on providing care to older adults following a HF hospitalization, and if confirmed, could inform the development of strategies to prevent posthospitalization adverse events.

Prior studies have demonstrated that availability of unhealthy food outlets in the neighborhood environment is associated with the incidence of cardiovascular disease, including heart failure and coronary artery disease. ²¹ , ²² , ²³ , ²⁴ One study to date has specifically examined the impact of food deserts on long‐term outcomes in HF, demonstrating an increased risk of HF hospitalization among participants in 2 Atlanta‐based cohorts. ²⁵ Our study now builds on these findings by demonstrating a potential negative impact of local unhealthy food sources on outcomes following a HF hospitalization, a particularly vulnerable period in which diet (such as avoidance of excessive sodium) may be especially important. ²⁶ , ²⁷ , ²⁸

These findings could have implications for efforts to improve posthospitalization outcomes in HF. Although physicians may provide education to patients about the importance of healthy eating upon discharge following a HF hospitalization, this education may be less beneficial if the environment is not conducive to such behavior. This is particularly relevant for older adults, who frequently contend with frailty and mobility challenges, especially after a HF hospitalization. ²⁹ , ³⁰ , ³¹ Indeed, older adults with HF may not be able to venture out to ascertain healthier food sources, instead opting for food sources that are closest to them. Consistent with this notion, data indicate that older adults living in urban areas tend to spend time (and thus potentially preferentially use resources such as food sources) in the immediate vicinity of their home. ³² This may explain why we observed an association for unhealthy foods within a 1‐km buffer but not within a 5‐km buffer. Unhealthy food sources are relevant for multiple reasons. First, high‐sodium foods can contribute to fluid retention. Although there is ongoing debate about strict sodium restriction, ³³ there is less debate about the negative effects of high sodium. ³⁴ , ³⁵ Unhealthy food sources can also contribute to malnutrition, a condition observed to be highly prevalent and associated with adverse outcomes in older adults with HF across the left ventricular ejection fraction spectrum. ³⁶ , ³⁷ In addition, unhealthy food sources can contribute to food insecurity, thereby exacerbating malnutrition and contributing to cachexia. ³⁸ Taken together, a high density of unhealthy foods is an important neighborhood‐level SDV that merits additional attention in the HF community.

If our findings are causally related, they would have some important public health and policy implications. Simply providing education and counseling about healthy eating is unlikely to be sufficient to improve posthospitalization outcomes in HF. Instead, it may be important to consider neighborhood/community level interventions to facilitate improved outcomes following a HF hospitalization, as influencing neighborhood‐level SDVs will likely require interventions above and beyond physician‐ and hospital‐level approaches. Healthy food supplements and reimbursements, as well as coupons for healthy eating, could serve as potential strategies to promote healthy food consumption. ³⁷ , ³⁹ Another potential strategy for addressing this challenge could be home‐based delivery of foods. The ongoing GOURMET‐HF (Geriatric Out‐of‐Hospital Randomized Meal Trial in Heart Failure) study aims to evaluate this approach in patients with HF, with promising pilot results on quality of life and readmission. ⁴⁰

Our study also showed that the density of clinical treatment and physical activity facilities may be less relevant for posthospitalization outcomes among older adults with HF. Prior work has shown the benefits of early posthospitalization follow‐up, ⁴¹ , ⁴² but posthospitalization follow‐up may not be directly linked to density of clinical treatment facilities—in fact, posthospitalization follow‐up for older adults with HF may be a challenge even when density is high. In a study conducted at a large quaternary academic center in a city with high density of clinical facilities, only 52% of older adults aged at least 65 years had a scheduled follow‐up postdischarge appointment at the time of hospital discharge. ⁴³ Prior work has also shown the potential benefits of posthospitalization physical activity for older adults with HF, ³¹ but facilities for physical activity may not be sufficient to promote engagement of older adults with HF. This is because older adults often face barriers to in‐person interventions such as transportation limitations, distance, costs, and schedule conflicts; these issues may be further exacerbated by additional factors like traffic and inclement weather. ⁴⁴ , ⁴⁵ , ⁴⁶ In addition, these facilities are unlikely to be tailored to the particular needs of older adults recovering from a HF hospitalization. Taken together, our findings suggest that facilities for clinical encounters and physical activity may not be sufficient to improve posthospitalization outcomes. Whether telehealth and mobile physical rehabilitation can improve posthospitalization outcomes is an area of great interest currently under study.

There were several strengths to this study, including use of a biracial nationwide cohort with robust data on HF admission characteristics. There were also some important limitations of this study. Some baseline and follow‐up data were collected through self‐reported telephone surveys, which may introduce reporting bias. In addition, given that the REGARDS cohort included only Black and White participants, the results may not be generalizable to other racial groups. The sample size was relatively small, which limited power to detect more subtle differences. The ability to detect more subtle differences was also limited by the fact that we tested multiple hypotheses and applied the Bonferroni correction; though even without applying the Bonferroni correction, the only other finding with statistical significance (P < 0.05) would have been the association between density of unhealthy food sources and outcomes at the 90‐day time point. On the other hand, this study had a high degree of geographic diversity (with hospitalizations occurring across the United States), which provides high generalizability; it is also notable that we found an important association despite the small sample size and the need for adjusting for multiple comparisons, reflecting the potential for high internal validity. In addition, our study examined Medicare beneficiaries in urban regions—future studies would benefit from examining whether these findings apply to younger populations and those in rural areas. We were unable to assess participant use of the neighborhood‐level facilities—for example, although we found that the density of clinical treatment facilities was not associated with posthospitalization outcomes, our findings may have been different if we specifically examined those who used the clinical treatment facilities (ie, completed follow‐up appointments). In addition, geographic buffers are only a proxy for resource availability, as facilities at the same radius may have differential time to travel from residential address. Finally, although we adjusted for several participant‐level and hospital‐level covariates, we could not exclude other residual unmeasured confounders like differences in public and subsidized transportation initiatives/availability and community‐based programs.

CONCLUSIONS

In conclusion, higher availability of unhealthy food sources in a participant's local neighborhood was associated with readmission following HF hospitalization among adults aged ≥65 years old living in urban environments. This finding supports the importance of developing strategies to mitigate the negative effects of unhealthy food sources for older adults with HF, with a specific focus on the neighborhood/community level.

Sources of Funding

This research project, via REGARDS, is supported by cooperative agreement U01 NS041588 co‐funded by the National Institute of Neurological Disorders and Stroke and the National Institute on Aging, National Institutes of Health, Department of Health and Human Services. It was also supported by National Institute on Aging grants (R01AG049970, R01AG049970‐S1, R01AG072634, R56AG049970), as well as by the National Heart, Blood, and Lung Institute (R01HL14843), National Institute on Alcohol Abuse and Alcoholism (R01AA028552), Commonwealth Universal Research Enhancement program funded by the Pennsylvania Department of Health 2015 Formula award (SAP #4100072543), the Urban Health Collaborative at Drexel University, and the Built Environment and Health Research Group at Columbia University. Additional funding was provided by National Heart, Blood, and Lung Institute grants R01HL080477 and R01HL165452, and by National Institute on Aging grant R03AG056446. The content is solely the responsibility of the authors and does not necessarily represent the official views of any of the aforementioned institutions; neither did these institutions have any role in the design and conduct of the study, the collection, management, analysis, and interpretation of the data, or the preparation or approval of the article.

Disclosures

Dr Safford is the founder of MedExplain, Inc., a company seeking to disseminate the Patient Activated Learning System approach to patient education. Dr Levitan is supported by research funding (to institution) from Amgen, Inc. and receives personal fees from University of Pittsburgh for serving on a Data and Safety Monitoring Board. Dr Goyal has received consulting fees from Agepha Pharma, Akros Pharma, Axon Therapies, and Sensorum Health; and has received personal fees for medicolegal consulting and expert testimony related to heart failure. The remaining authors have no disclosures to report.

Supporting information

Tables S1–S4

Data S1