Advertisement

Influence of Neighborhood Conditions on Recurrent Hospital Readmissions in Patients with Heart Failure: A Cohort Study

      Abstract

      Purpose

      This study examined how certain aspects of residential neighborhood conditions (ie, observed built environment, census-based area-level poverty, and perceived disorder) affect readmission in urban patients with heart failure.

      Methods

      A total of 400 patients with heart failure who were discharged alive from an urban-university teaching hospital were enrolled. Data were collected about readmissions during a 2-year follow-up. The impact of residential neighborhood conditions on readmissions was examined with adjustment for 7 blocks of covariates: 1) patient demographic characteristics; 2) comorbidities; 3) clinical characteristics; 4) depression; 5) perceived stress; 6) health behaviors; and 7) hospitalization characteristics.

      Results

      A total of 83.3% of participants were readmitted. Participants from high-poverty census tracts (≥20%) were at increased risk of readmission compared with those from census tracts with <10% poverty (hazard ratio [HR]: 1.53; 95% confidence interval: 1.03-2.27; P < .05) when adjusted for demographic characteristics. None of the built environmental or perceived neighborhood conditions were associated with the risk of readmission. The poverty-related risk of readmission was reduced to nonsignificance after including diabetes (HR: 1.33) and hypertension (HR: 1.35) in the models.

      Conclusions

      The effect of high poverty is partly explained by high rates of hypertension and diabetes in these areas. Improving diabetes and blood pressure control or structural aspects of impoverished areas may help reduce hospital readmissions.
      Clinical Significance
      • High area-poverty rate, but not built environment or perceived neighborhood conditions, was associated with increased risk for readmission in patients with heart failure.
      • Improving glucose and blood pressure control in patients with heart failure living in high-poverty areas may reduce the risk of readmission (eg, via postdischarge telemonitoring or home-visiting programs).

      Introduction

      Hospital readmissions are common and costly among patients with heart failure. More than 50% of patients are readmitted within 6 months after discharge.
      • Joynt KE
      • Jha AK.
      Who has higher readmission rates for heart failure, and why? Implications for efforts to improve care using financial incentives.
      Some readmissions may be driven by care processes that are under the control of hospitals or providers, whereas others are more likely to result from patient- and community-level factors.
      • Joynt KE
      • Jha AK.
      Who has higher readmission rates for heart failure, and why? Implications for efforts to improve care using financial incentives.
      Increasingly, social determinants of health, including the neighborhood in which patients reside, are being identified as playing an important role in heart failure care and outcomes.
      • Virani SS
      • Alonso A
      • Benjamin EJ
      • et al.
      Heart disease and stroke statistics-2020 update: a report from the American Heart Association.
      Patients with heart failure who live in impoverished neighborhoods may have a higher risk for recurrent all-cause and heart-failure-specific readmissions than patients who live in more affluent neighborhoods.
      • Kind AJH
      • Jencks S
      • Brock J
      • et al.
      Neighborhood socioeconomic disadvantage and 30-day rehospitalization.
      • Eapen ZJ
      • McCoy LA
      • Fonarow GC
      • et al.
      Utility of socioeconomic status in predicting 30-day outcomes after heart failure hospitalization.
      • Reid M
      • Kephart G
      • Andreou P
      • Robinson A.
      Potential of community-based risk estimates for improving hospital performance measures and discharge planning.
      • Conrad N
      • Judge A
      • Tran J
      • et al.
      Temporal trends and patterns in heart failure incidence: a population-based study of 4 million individuals.
      Disadvantaged neighborhoods may adversely affect heart failure readmissions through stressors in the physical environment, lower access to economic and medical resources, more severe comorbidities, or a noncohesive social environment.
      • Aneshensel CS
      • Wight RG
      • Miller-Martinez D
      • Botticello AL
      • Karlamangla AS
      • Seeman TE.
      Urban neighborhoods and depressive symptoms among older adults.
      Residents of disadvantaged neighborhoods may also report increased stress and incident or worsening depression,
      • Ross C.
      Neighborhood disadvantage and adult depression.
      which have been associated with risk of heart failure readmission. Recently, the built environment (ie, the human-created spaces and infrastructure with which people regularly interact) has gained attention in the measurement of street-level conditions because observed or perceived indicators of physical disorder (eg, presence of garbage, graffiti, or abandoned buildings; poor building conditions) have been associated with risk factors for readmission such as obesity, alcohol use, tobacco smoking, depression, and perceived stress.
      • Mayne SL
      • Jose A
      • Mo A
      • et al.
      Neighborhood disorder and obesity-related outcomes among women in Chicago.
      ,
      • O'Brien DT
      • Farrell C
      • Welsh BC.
      Broken (windows) theory: a meta-analysis of the evidence for the pathways from neighborhood disorder to resident health outcomes and behaviors.
      It is unclear if characteristics of the built environment also increase the risk of readmission in patients with heart failure.
      This study examined whether certain aspects of neighborhood conditions (ie, the built environment, area-level poverty, and perceived disorder) affect readmissions in urban patients with heart failure. Specifically, it determined 1) whether patients who live in neighborhoods with adverse conditions experience more readmissions than those who live in neighborhoods with better conditions; 2) the extent to which patient-level covariates (including patient demographic characteristics, comorbidities, clinical characteristics, depression, perceived stress, self-care, and hospitalization characteristics) account for any observed associations between neighborhood conditions and readmissions; and 3) the geographic levels (ie, census tract, zip code, street segment) at which any observed associations between neighborhood conditions and readmissions exist.

      Methods

      Study Participants

      Patients with a clinical diagnosis of heart failure were screened for study eligibility while hospitalized between July 1, 2014, and December 31, 2016, at [anonymized], an urban university teaching hospital in [anonymized city]. Eligibility required confirmation of current symptoms of heart failure according to the European Society of Cardiology criteria
      • Ponikowski P
      • Voors AA
      • Anker SD
      • et al.
      2016 ESC Guidelines for the diagnosis and treatment of acute and chronic heart failure: The Task Force for the diagnosis and treatment of acute and chronic heart failure of the European Society of Cardiology (ESC). Developed with the special contribution of the Heart Failure Association (HFA) of the ESC.
      and the ability to participate with interviews, questionnaires, and follow-up assessments among those who were discharged alive. Exclusion criteria were: 1) isolated right heart failure or reversible heart failure due to valve disease with impending surgical correction; 2) dementia; 3) medical comorbidities with a poor 1-year prognosis; 4) age younger than 18 years or older than 89 years; 5) active substance abuse or alcoholism; 6) bipolar disorder or schizophrenia; and 7) refusal by patient or attending physician. This 2-year prospective cohort study was approved by the Institutional Review Board of [anonymized]. Participants gave written informed consent.

      Neighborhood Conditions

      Three aspects of neighborhood conditions were examined: 1) the built environment; 2) neighborhood poverty (ie, zip code and census tract); and 3) perceived neighborhood disorder.

      Built Environment

      The participant's home address was obtained at enrollment in the study. Residential addresses were geocoded with ArcGIS 9.3 (ESRI). Geocoding made it possible to identify the street segment of each patient's residence for auditing the built environment. A street segment was defined as the section of the road between 2 consecutive intersections. Segment sizes may differ across neighborhoods.
      A specially trained study team member used each participant's residential street address at baseline to complete the Active Neighborhood Checklist to measure neighborhood conditions using Google Street View (GSV) to assess the built environment. The checklist assesses land-use characteristics; sidewalks; shoulders and bike lanes; street characteristics; and quality of the environment for pedestrians (see Appendix, available online, for components of the built environment). We selected the date of the GSV that was closest to the date of a participant's baseline interview. The validity of historic and more recent GSV imagery were similar compared to in-person field audits.
      • Kelly CM
      • Wilson JS
      • Baker EA
      • Miller DK
      • Schootman M.
      Using Google Street View to audit the built environment: inter-rater reliability results.

      Neighborhood Poverty

      Population characteristics at the census-tract and zip-code-tabulated-area (ZCTA) level were obtained from the 2008-2012 American Community Survey and consisted of the percentage of the population that is below the federal poverty line. Poverty was selected because it is a robust indicator of socioeconomic status across levels of geography and time, has been associated with various health outcomes, and has relevance for policymakers. Area poverty level was classified as the percent living in poverty in the residents’ census tract/ZCTA and classified into 0%-9.9%, 10.0%-19.9%, or ≥20% to permit examination of nonlinear effects.

      Perceived Neighborhood Conditions

      The Ross-Mirowsky Perceived Neighborhood Disorder (PND) questionnaire was administered at baseline. The PND captures the extent to which residents perceive cues in their neighborhoods signaling the breakdown of social control.
      • Ross CE
      • Mirowsky J.
      Disorder and decay: the concept and measurement of perceived neighborhood disorder.
      It includes 15 four-point Likert-type items and consists of 2 subscales, Social Disorder (n = 9 items) and Physical Disorder/Decay (n = 6 items).

      Psychosocial Questionnaires

      Other questionnaires administered at baseline included the Patient Health Questionnaire (PHQ-9)
      • Kroenke K
      • Spitzer RL
      • Williams JB.
      The PHQ-9: validity of a brief depression severity measure.
      to assess depression, the Generalized Anxiety Disorder questionnaire (GAD-7)
      • Spitzer RL
      • Kroenke K
      • Williams JB
      • Löwe B.
      A brief measure for assessing generalized anxiety disorder: the GAD-7.
      to assess anxiety, and the Self-Care of Heart Failure Index (SCHFI v6.2)
      • Riegel B
      • Carlson B
      • Moser DK
      • Sebern M
      • Hicks FD
      • Roland V.
      Psychometric testing of the self-care of heart failure index.
      to assess heart failure self-care. These questionnaires were readministered at 3-month intervals for 24 months.

      Readmissions

      The patients’ electronic medical record was monitored to identify all readmissions to [anonymized hospital] following the date of discharge until the end of the study period (December 31, 2018), death, or loss to follow up. Patients were interviewed at 6-month intervals to identify admissions to other hospitals, and medical records were obtained to document these admissions.

      Covariates

      Previous studies of neighborhood conditions in patients with heart failure
      • Kind AJH
      • Jencks S
      • Brock J
      • et al.
      Neighborhood socioeconomic disadvantage and 30-day rehospitalization.
      ,
      • Morris AA
      • McAllister P
      • Grant A
      • et al.
      Relation of living in a “food desert” to recurrent hospitalizations in patients with heart failure.
      and studies of predictors of heart failure readmission
      • Freedland KE
      • Carney RM
      • Rich MW
      • Steinmeyer BC
      • Skala JA
      • Dávila-Román VG
      Depression and multiple rehospitalizations in patients with heart failure.
      ,
      • Jhund PS
      • Ponikowski P
      • Docherty KF
      • et al.
      Dapagliflozin and recurrent heart failure hospitalizations in heart failure with reduced ejection fraction: an analysis of DAPA-HF.
      were used as starting points for identifying covariates for the current study. We examined 7 blocks of covariates that could explain a potential association between residential neighborhood conditions and readmission (Table 1), including 1) patient demographic characteristics; 2) comorbidities; 3) clinical characteristics; 4) depression; 5) perceived stress;
      • Cohen S
      • Kamarck T
      • Mermelstein R.
      A global measure of perceived stress.
      6) health behaviors; and 7) hospitalization characteristics. Serum markers such as blood urea nitrogen, hemoglobin, and prescribed beta-blockers, angiotensin-converting enzyme inhibitors or angiotensin receptor blocker, nitrates, and mineralocorticoid antagonist were not associated with readmission in our cohort and were not included in the current analysis.
      • Freedland KE
      • Carney RM
      • Rich MW
      • Steinmeyer BC
      • Skala JA
      • Dávila-Román VG
      Depression and multiple rehospitalizations in patients with heart failure.
      ,
      • Freedland KE
      • Skala JA
      • Steinmeyer BC
      • Carney RM
      • Rich MW.
      Effects of depression on heart failure self-care.
      Table 1Demographic, Medical, and Psychosocial Characteristics at Enrollment and Unadjusted Risk of Readmission (N = 400)
      VariableFrequency (%)Hazard ratio (95% CI)
      Demographic characteristics
       Age (years)58.4 ± 13.10.89 (0.82, 0.97)
      Per 10-unit increase
       Gender (female)202 (50.5)0.99 (0.77, 1.29)
       Race (white)196 (49.0)0.85 (0.66, 1.10)
       Education ≤ 12 years166 (41.5)1.22 (0.95, 1.58)
       Household income (<$30,000)206 (51.5)1.37 (1.02, 1.83)
        <10K42 (10.5)
        10K-19.9K59 (14.8)
        20K-29.9K45 (11.3)
        30K-39.9K35 (8.8)
        40K-49.9K27 (6.8)
        50K-59.9K29 (7.3)
        60K-69.9K14 (3.5)
        70K-79.9K10 (2.5)
        80K and higher27 (6.8)
        Refused33 (8.3)
        Unknown79 (19.8)
       Married or partnered159 (39.8)0.96 (0.74, 1.26)
      Comorbidities
       Chronic obstructive pulmonary disease114 (28.5)1.57 (1.18, 2.09)
       Sleep apnea162 (40.5)1.36 (1.05, 1.77)
       Diabetes202 (50.5)1.68 (1.30, 2.16)
       Hypertension356 (89.0)2.67 (1.74, 4.09)
       eGFR
        On dialysis26 (6.5)2.22 (1.55, 3.17)
        <60 mL/min/1.73m2186 (46.5)1.21 (0.92, 1.59)
        60+ mL/min/1.73m2188 (47.0)Reference
       History of acute coronary syndrome143 (35.8)1.36 (1.05, 1.76)
       Body mass index (kg/m2)34.3 ± 10.51.03 (0.99, 1.06)
      Per 3-unit increase
      Clinical characteristics
       New York Heart Association Class
       Ordinal scale2.5 ± 0.81.64 (1.42, 1.90)
       I-II187 (46.8)0.49 (0.38, 0.63)
       Left ventricular ejection fraction
        Interval scale (%)38.2 ± 20.00.99 (0.95, 1.03)
      Per 5-unit increase
        <45%256 (64.0)0.93 (0.71, 1.23)
      Medications
       Angiotensin receptor-neprilysin inhibitors60 (15.0)1.10 (0.75, 1.62)
       Hydralazine110 (27.5)1.32 (0.99, 1.75)
       Isosorbide115 (28.8)1.44 (1.09, 1.89)
      Depression measures
       PHQ-9 depression score9.3 ± 6.21.05 (0.99, 1.12)
      Per 3-unit increase
       Antidepressant at baseline100 (25.0)1.19 (0.90, 1.58)
      Perceived stress
       PSS perceived stress score15.4 ± 8.91.04 (0.99, 1.08)
      Per 3-unit increase
      Health behavior
       Never smoker187 (46.8)1.08 (0.83, 1.39)
       SCHFI Maintenance score60.4 ± 17.21.10 (1.01, 1.19)
      Per 10-unit increase
       SCHFI Management score54.2 ± 24.51.07 (1.01, 1.12)
      Per 10-unit increase
       SCHFI Confidence score58.5 ± 21.51.05 (0.99, 1.12)
      Per 10-unit increase
      Hospitalization characteristics
       Time since last known HF hospitalization (months)
      Estimates are presented as median (IQR) due to a left-skewed distribution
      24.1 (47.4)0.98 (0.97, 0.99)
       Discharge disposition (home)370 (92.5)1.05 (0.69, 1.60)
       Length of index stay (days)
      Estimates are presented as median (IQR) due to a left-skewed distribution
      5.0 (6.0)1.01 (0.98, 1.04)
      Per 3-unit increase
      CI = confidence interval; HF = heart failure; IQR = interquartile range; PSS = Perceived Stress Scale; SCHFI = Self-Care of Heart Failure Index; SD = standard deviation.
      Estimates are reported as mean ± SD for interval-scaled variables and number (%) for categorical variables.
      low asterisk Per 10-unit increase
      Per 3-unit increase
      Per 5-unit increase
      § Estimates are presented as median (IQR) due to a left-skewed distribution

      Statistical Analysis

      We used marginal proportional rates models to determine the association of adverse neighborhood conditions and each of the blocks of covariates on the number of all-cause readmissions over 2 years, controlling for demographic characteristics.
      • Liu D
      • Schaubel DE
      • Kalbfleisch JD.
      Computationally efficient marginal models for clustered recurrent event data.
      Deaths following study enrollment were incorporated into the models as terminating events in the readmission process. Cox frailty and random coefficients models were fitted to the longitudinal depression and terminal event processes, respectively. We assessed the extent of the clustering of participants within zip codes and census tracts to determine if a multilevel approach was needed. Because 79.7% of the 172 zip codes contained only 1 or 2 participants, a multilevel approach was not indicated.
      First, we calculated associations among the 3 types of neighborhood conditions (i.e., built environment, area-level poverty rate, and perceived disorder). Next, we determined whether neighborhood conditions predicted the risk of readmission separately by type of condition. For the neighborhood conditions that were associated with the risk of readmission, we then determined whether the covariates explained the association between adverse neighborhood conditions and readmission by adding all variables comprising a block of covariates to the marginal proportional rates model containing the adverse neighborhood conditions and sociodemographic characteristics. Examining each of the 7 blocks of covariates separately informed the extent to which each block of variables explained observed associations between neighborhood conditions and readmissions. We examined specific variables within a particular block of covariates when the block showed a significant reduction in odds ratio for the association of neighborhood conditions with readmissions. The assumptions and fit of each model were assessed via Schoenfeld and Martingale residuals. To test the robustness of our findings, we performed a sensitivity analysis by 1) using zip codes rather than census tracts for area-level poverty rate and 2) comparing the results for the imputed built environments and excluding participants with missing data.
      Statistical tests were 2-tailed with α = 0.05. SAS version 9.3 (SAS Institute) was used for imputation, to construct mean cumulative function (MCF) plots and to fit the proportional rates model. Twenty multiply imputed data sets with all independent variables and any auxiliary variables that were moderately correlated (r ≥ 0.30) with the outcome were generated for each model; outcome data were not imputed.

      Results

      Of 1440 patients with heart failure assessed for eligibility, a total of 1159 met the heart failure inclusion criteria. Of these, 1048 met 1 or more exclusion criteria or refused participation, resulting in 400 participants with heart failure completing the baseline interview (Figure). Study attrition was due in large part to mortality (n = 117; 29.3%); 277 patients (69.2%) completed the 2-year follow up with 6 of 400 (1.5%) participants being lost to follow-up or refusing participation. Data concerning the built environment could not be obtained for 67 of 400 patients (16.8%) because the GSV data were not collected by Google, the exact address was unknown because the patient entered a post office box number for the home address, or no houses were observed on the street segment; these data were imputed.
      Figure
      FigureCONSORT diagram of reasons for exclusion from the study.
      Table 1 shows that study participants were on average 58.4 years old (range: 22-88 years), 50.5% were females, and 49.0% were white. In all, 39.8% were married or partnered and 51.5% reported household incomes less than $30,000 per year. Comorbidities were prevalent, with more than half of participants having diabetes. In addition, 64.0% of participants had heart failure with reduced ejection fraction and 53.3% were in New York Heart Association Class functional class III or IV before study participation. Participants lived in 261 different census tracts and 172 different zip codes. Many characteristics were associated with the unadjusted risk of readmission, including age, chronic obstructive pulmonary disease, sleep apnea, diabetes, hypertension, estimated glomerular filtration rate, history of acute coronary syndrome, New York Heart Association Class, and heart failure self-care management. Depression at baseline, as measured by the PHQ-9, was not associated with risk of readmission, and the mean PHQ-9 score was just below the usual cutoff of 10 for clinically significant depression.
      Table 2 shows that less than one-quarter of participants lived on streets with graffiti, broken or boarded windows, litter, or broken glass or on streets with nonresidential destinations. Only 26.0% of streets had the highest sidewalk quality on all 9 characteristics of high-quality sidewalks. The date of the GSV imagery was on average 48.8 months (standard deviation = 8.0) after the baseline interview. In all, 50.0% of the participants lived in census tracts with ≥20% poverty rate. This was 56.7% based on zip code poverty rates. Most (72.5%) participants did not move during the study period. Moving during the study period was not related to the rate of readmission (hazard ratio [HR]: 0.94; 95% confidence interval [CI]: 0.71-1.23; P = .66).
      Table 2Frequency of Neighborhood Conditions and Unadjusted Association with Risk of Readmission
      Neighborhood conditionN (%)Number (%) readmittedHazard ratio (95% CI)
      Built environment (vs No)
       Mixed use93 (23.3)73 (18.3)1.05 (0.66, 1.66)
       Abandoned buildings/vacant lots86 (21.5)71 (17.8)1.08 (0.78, 1.50)
       Mixed housing55 (13.8)44 (11.0)0.86 (0.63, 1.17)
       Graffiti/litter/broken windows or glass55 (13.8)42 (10.5)1.12 (0.77, 1.64)
       Sidewalk quality104 (26.0)84 (21.0)0.83 (0.62, 1.12)
      Census tract poverty rate
       <10.0%78 (19.5)63 (15.8)1.00
       10.0%-19.9%122 (30.5)98 (24.5)1.24 (0.85, 1.81)
       20.0+%200 (50.0)172 (43.0)1.59 (1.13, 2.23)
      Zip code poverty rate
       <10.0%76 (19.0)60 (15.0)1.00
       10.0%-19.9%97 (24.3)80 (20.0)1.39 (0.93, 2.06)
       20+%227 (56.8)193 (48.3)1.65 (1.18, 2.31)
      Perceived neighborhood conditions (per 1 unit)*
       Physical disorder9.9 ± 3.91.02 (0.98, 1.06)
       Social disorder14.7 ± 6.01.01 (0.98, 1.03)
      CI = confidence interval; SD = standard deviation.
      Estimates are reported as mean ± SD for interval-scaled variables and number (%) for categorical variables.
      Overall, 83.3% (333/400) of participants were readmitted for a total of 1581 all-cause rehospitalizations during the 2-year study period.

      Neighborhood Poverty Rate

      Participants who lived in census tracts with high poverty rates (≥20%) were at increased risk of readmission compared to those who lived in census tracts with <10% poverty (HR: 1.59; 95% CI: 1.13-2.23) in unadjusted analysis (Table 2). Participants who lived in census tracts with 10.0%-19.9% poverty rate were equally likely to be readmitted as those who lived in census tracts with <10% poverty (HR: 1.24; 95 CI: 0.85-1.81). The risk of readmission was attenuated and became nonsignificant after adjusting for comorbidity (Table 3, Model 3). However, the risk remained significant when accounting for each of the other 5 blocks of factors (ie, clinical characteristics, depression, perceived stress, health behaviors, and hospitalization characteristics).
      Table 3Unadjusted and Adjusted Associations (Hazard Ratio and 95% Confidence Interval) of Neighborhood Poverty Rate and Risk of Rehospitalization
      Model 1Model 2Model 3Model 4Model 5Model 6Model 7Model 8
      Census tract poverty rate
       <10.0%1.001.001.001.001.001.001.001.00
       10.0%-19.9%1.24

      (0.85, 1.81)
      1.18

      (0.81, 1.73)
      1.04

      (0.73, 1.48)
      1.18

      (0.83, 1.69)
      1.14

      (0.79, 1.64)
      1.14

      (0.79, 1.66)
      1.18

      (0.81, 1.72)
      1.20

      (0.82, 1.74)
       20+%1.59

      (1.13, 2.23)
      1.53

      (1.03, 2.27)
      1.21 (ns)

      (0.83, 1.77)
      1.47

      (1.02, 2.12)
      1.49

      (1.02, 2.18)
      1.52

      (1.04, 2.24)
      1.55

      (1.05, 2.28)
      1.51

      (1.01, 2.24)
      Zip code poverty rate
       <10.0%1.001.001.001.001.001.001.001.00
       10.0%-19.9%1.39

      (0.93, 2.06)
      1.31

      (0.88, 1.96)
      1.18

      (0.81, 1.70)
      1.27

      (0.87, 1.85)
      1.31

      (0.88, 1.94)
      1.32

      (0.89, 1.96)
      1.32

      (0.89, 1.96)
      1.36

      (0.91, 2.01)
       20+%1.65

      (1.18, 2.31)
      1.58

      (1.09, 2.29)
      1.27 (ns)

      (0.89, 1.83)
      1.51

      (1.06, 2.15)
      1.56

      (1.08, 2.25)
      1.58

      (1.09, 2.29)
      1.61

      (1.11, 2.34)
      1.58

      (1.09, 2.29)
      Model 1: Unadjusted.
      Model 2: Adjusted for demographic characteristics.
      Model 3: Adjusted for demographic characteristics and comorbidity.
      Model 4: Adjusted for demographic characteristics and clinical characteristics.
      Model 5: Adjusted for demographic characteristics and depression.
      Model 6: Adjusted for demographic characteristics and perceived stress.
      Model 7: Adjusted for demographic characteristics and health behaviors.
      Model 8: Adjusted for demographic characteristics and hospitalization characteristics.
      Table 4 shows that when examining specific comorbidities, the risk of readmission was reduced to nonsignificance after adding diabetes (Model 3) or hypertension (Model 4) to Model 1. However, the risk changed little for participants in the high-poverty census tracts when the other comorbidities (chronic obstructive pulmonary disorder, estimated glomerular filtration rate, and acute coronary syndrome) were added to Model 1. Our findings were robust with respect to using zip codes rather than census tracts in sensitivity analyses of poverty rate.
      Table 4Unadjusted and Adjusted Associations (Hazard Ratio and 95% Confidence Intervals) of Neighborhood Poverty Rate and Risk of Rehospitalization Focusing on Comorbidity
      Model 1Model 2Model 3Model 4Model 5Model 6
      Census tract poverty rate
       <10.0%1.001.001.001.001.001.00
       10.0%-19.9%1.18 (0.81, 1.73)1.16 (0.80, 1.68)1.12 (0.77, 1.62)1.12 (0.77, 1.62)1.18 (0.81, 1.71)1.19 (0.82, 1.73)
       20+%1.53 (1.03, 2.27)1.47 (0.99, 2.17)1.33 (0.88, 2.01)1.35 (0.91, 1.98)1.55 (1.06, 2.26)1.51 (1.02, 2.23)
      Zip code poverty rate
       <10.0%1.001.001.001.001.001.00
       10.0%-19.9%1.31 (0.88, 1.96)1.29 (0.87, 1.91)1.22 (0.82, 1.81)1.33 (0.89, 1.97)1.28 (0.87, 1.89)1.26 (0.85, 1.88)
       20+%1.58 (1.09, 2.29)1.56 (1.08, 2.26)1.35 (0.92, 1.98)1.49 (1.03, 2.15)1.53 (1.05, 2.23)1.52 (1.05, 2.21)
      eGFR = estimated glomerular filtration rate.
      Model 1: Adjusted for demographic characteristics.
      Model 2: Adjusted for demographic characteristics and chronic obstructive pulmonary disease.
      Model 3: Adjusted for demographic characteristics and diabetes.
      Model 4: Adjusted for demographic characteristics and hypertension.
      Model 5: Adjusted for demographic characteristics and eGFR.
      Model 6: Adjusted for demographic characteristics and acute coronary syndrome.

      Built Environment and Perceived Neighborhood Conditions

      Table 2 describes readmission rates as a function of the built environment and perceived neighborhood conditions. None of these factors were associated with the risk of readmission. In 2 sensitivity analyses, the findings were quite robust.

      Discussion

      The main finding of this study was that high area poverty rate, but not built environment or perceived neighborhood disorder, was associated with increased risk for readmission in patients with heart failure. Participants who lived in census tracts or zip codes with high poverty rates (≥20%) were at increased risk of readmission compared to those who lived in areas with lower poverty rates, but this was largely explained by concomitant medical comorbidities (diabetes and hypertension).
      Many patients with heart failure have multiple comorbidities, including diabetes and hypertension.
      • Conrad N
      • Judge A
      • Tran J
      • et al.
      Temporal trends and patterns in heart failure incidence: a population-based study of 4 million individuals.
      Type 2 diabetes is present in 40%-45% of patients with heart failure regardless of ejection fraction.
      • Groenewegen A
      • Rutten FH
      • Mosterd A
      • Hoes AW.
      Epidemiology of heart failure.
      Reducing HbA1c lowers the risk of hospitalization among those with heart failure and diabetes; for every 1% absolute decrease in HbA1c, the risk of hospitalization is reduced by 23%.
      • Cintra RM
      • Nogueira AC
      • Bonilha I
      • et al.
      Glucose-lowering drugs and hospitalization for heart failure: a systematic review and additive-effects network meta-analysis with more than 500 000 patient-years.
      High systolic, diastolic, and pulse pressure levels are associated with higher rates of adverse events in patients with heart failure, which supports the importance of optimized blood pressure control.
      • Lip GY
      • Skjøth F
      • Overvad K
      • Rasmussen LH
      • Larsen TB.
      Blood pressure and prognosis in patients with incident heart failure: the Diet, Cancer and Health (DCH) cohort study.
      These observations suggest focusing on improving glucose and blood pressure control in patients with heart failure living in high-poverty areas (eg, via postdischarge telemonitoring or home-visiting programs).
      • Dawson NL
      • Hull BP
      • Vijapura P
      • et al.
      Home telemonitoring to reduce readmission of high-risk patients: a modified intention-to-treat randomized clinical trial.
      In addition to targeting patients, focusing on structural aspects of neighborhoods with high poverty rates, such as improving access to nutritional foods and exercise facilities, could also reduce readmission risk in patients with heart failure. This macro approach may also be worthwhile since none of the built environment or perceived neighborhood conditions were associated with the risk of readmission.
      Strengths of our study include its racially diverse sample, longitudinal design, low loss to follow-up, and evaluation of multiple aspects of neighborhood conditions (ie, zip code, census tract, street segment, and perceived conditions). We used GSV, which has been found to yield valid and reliable information about the built environment. Advantages include efficiency, researcher safety, low cost, unobtrusive data collection, and access to historical images of street segments. Also, previous studies did not control for patients’ income, which may have resulted in overestimating the effect of living in disadvantaged neighborhoods because patients who live in disadvantaged neighborhoods are themselves more likely to have lower incomes.
      However, the study also has some limitations. First, we only audited the street segment of each study participant's residence. Although the audited geographic area did not capture the entire neighborhood of study participants, our findings likely extend beyond the single street segment because adjacent segments tend to have similar characteristics.
      • Mooney SJ
      • Bader MDM
      • Lovasi GS
      • Neckerman KM
      • Teitler JO
      • Rundle AG.
      Validity of an ecometric neighborhood physical disorder measure constructed by virtual street audit.
      Second, this is a single-center study and generalizability may be limited to urban patients with heart failure.

      Conclusion

      In conclusion, high area poverty rate, but not built environment or perceived neighborhood conditions, is associated with increased readmissions in patients with heart failure. The effect of high poverty is explained, in part, by high rates of hypertension and diabetes in these areas. Improving diabetes and blood pressure control or structural aspects of impoverished areas may help reduce hospital readmissions among patients with heart failure who reside in high-poverty areas.

      Acknowledgment

      We thank Jeffrey Schootman for data collection about the built environment using Google StreetView.

      Appendix

      • 1.
        Active Neighborhood Checklist
      • 2.
        Perceived neighborhood disorder
      • 3.
        Table – sensitivity analysis. Frequency of neighborhood conditions and unadjusted association with risk of readmission, comparing the main analysis with the sensitivity analysis by excluding those with imputed built environmental variables.

      Active Neighborhood Checklist

      We examined 5 characteristics of the built environment of the street segment on which a patient resided. Mixed housing was present when a combination of single-family homes, multiunit homes, or apartments was observed. Otherwise, the participant did not live on a street segment with mixed housing. We also assessed whether an abandoned building, home, or vacant lot was present. A mixed-use variable was created to indicate whether a nonresidential destination was present on the street segment; nonresidential destinations included grocery and convenience stores, supermarkets, food establishments, entertainment venues, libraries, post offices, banks, dry cleaners, indoor fitness facilities, parks, off-road walking trails, sports and playing fields, basketball, tennis, and volleyball courts, playgrounds, outdoor pools, parks with exercise or playground facilities, and designated green spaces. We also observed whether graffiti, broken or boarded windows, litter, or broken glass were present. Sidewalk quality was based on the presence of 9 characteristics of high-quality sidewalks, including the presence of a buffer between the curb and sidewalk, sidewalk continuity within street segment, sidewalk continuity between street segments, width of sidewalks >3 feet, no missing ramps or curb cuts, no major bumps, cracks, or holes, no permanent obstructions, tree shade, and flat or gentle slope. Having no sidewalk was coded as 0. A dichotomous variable was created, with 0 indicating poor quality or no sidewalk and 1 indicating high-quality sidewalks on one or more of the nine characteristics.
      Start of Block: Auditor and Segment Details
      Q1.1 Auditor and Segment Details:
      • Today's Date: (1)
      • Auditor ID:: (2)
      • Segment ID: (3)
      • Street Name: (4)
      • Date Image Captured: (5)
      • Start Time: (6)
      End of Block: Auditor and Segment Details
      Start of Block: A. Land Uses
      Q2.1 Section A. What land uses are present?
      Q2.2 1. Are residential and non-residential land uses present?
      • All residential (1)
      • Both residential and non-residential (2)
      • All non-residential (3)
      Q2.3 2. What is the predominant land use? Check one or two that apply.
      • Residential buildings/yards (1)
      • Commercial, institutional, office or industrial building(s) (2)
      • School/school yards (elementary, middle, high school) (3)
      • Parking lots or garages (4)
      • Park with exercise/sport facilities or playground equipment (5)
      • Abandoned building or vacant lot (6)
      • Undeveloped land (7)
      • Designated green space (includes park with no exercise/play facilities) (8)
      • Other non-residential, specify: (9)
      Display This Question:
      If 1. Are residential and non-residential land uses present? = All residential
      Or 1. Are residential and non-residential land uses present? = Both residential and non-residential
      Q2.4 3. What types of residential uses are present? Select all that apply.
      • None (1)
      • Abandoned homes (2)
      • Single family homes (3)
      • Multi-unit homes (2-4 units) (4)
      • Apartments or condominiums (>4 units, 1-4 stories) (5)
      • Apartments or condominiums (>4 stories) (6)
      • Apartment over retail (7)
      • Other (retirement home, dorms) (8)
      Q2.5 4. What functioning parking facilities are present? Select all that apply.
      • None (no parking allowed on street most or all of the time) (1)
      • On-street, including angled parking (2)
      • Small lot or garage ( (3)
      • Medium to large lot (4)
      • Garage (5)
      Q2.6 5. What public recreational facilities and equipment are present (including in the schoolyard if publicly accessible)? Select all that apply.
      • None (1)
      • Park with exercise/sport facilities or playground equipment (2)
      • Off-road walking/biking trail (3)
      • Sports/playing field (4)
      • Basketball/tennis/volleyball court (5)
      • Playground (6)
      • Outdoor pool (7)
      • Other: (8)
      Display This Question:
      If 1. Are residential and non-residential land uses present? = Both residential and non-residential
      Or 1. Are residential and non-residential land uses present? = All non-residential
      Q2.7 6. (OPTIONAL) What types of non-residential uses are present? Select all that apply.
      • None (1)
      • Abandoned building or vacant lot (2)
      • Small grocery, convenience store (including in gas station), or pharmacy (3)
      • Supermarket (4)
      • Food establishment (restaurant, bakery, cafe, coffee shop, bar) (5)
      • Entertainment (e.g., movie theater, arcade) (6)
      • Library or post office (7)
      • Bank (8)
      • Laundry/dry cleaner (9)
      • Indoor fitness facility (10)
      • School (elementary, middle, high school) (11)
      • College, technical school, or university (12)
      • High-rise building (>5 stories) (13)
      • Big box store (e.g., Wal-Mart, Best Buy) (14)
      • Mall (15)
      • Strip mall (16)
      • Commercial walking street (17)
      • Large office building, warehouse, factory, or industrial building (18)
      Q2.8 Land use notes:
      _________________________________
      _________________________________
      _________________________________
      _________________________________
      _________________________________
      End of Block: A. Land Uses
      Start of Block: B. Public Transportation
      Q3.1 Section B. Is public transportation available?
      Q3.2 1. Any transit stop (bus, subway, or other)?
      • No (1)
      • Yes, one side (2)
      • Yes, both sides (3)
      Skip To: Q3.4 If 1. Any transit stop (bus, subway, or other)? = No
      Q3.3 1a. Bench or covered shelter at transit stop?
      • No (1)
      • Yes, one side (2)
      • Yes, both sides (3)
      Q3.4 Transit stop notes:
      _________________________________
      _________________________________
      _________________________________
      _________________________________
      _________________________________
      End of Block: B. Public Transportation
      Start of Block: C. Street Characteristics
      Q4.1 Section C. What street characteristics are visible?
      • 1. Enter posted speed limit (99 if none): (1) ______________________________________
      • 2. Enter special speed zone (99 if none): (2) ______________________________________
      • 3. Enter total # of lanes on street: (3) _______________________________________
      Q4.2 What street characteristics are visible?
      Tabled 1
      No (1)Yes (2)
      4. Marked lanes? (1)
      5. Median or pedestrian island? (2)
      6. Turn lane? (3)
      7. Stop sign or light for crossing this segment? If YES, go to C8. (4)
      7a. Any stoplight(s) without a walk signal? (5)
      8. Crosswalk for crossing this segment? (6)
      9. Traffic calming device (roundabout, speed bump, brick road, other)? If yes, specify type(s). (7)
      10. Cul-de-sac (dead-end street)? If NO, go to D1. (8)
      10a. Sidewalk cut-through in cul-de-sac? (9)
      11. Footbridge? (10)
      12. Pedestrian underpass? (11)
      Q4.3 Street characteristic notes:
      _________________________________
      _________________________________
      _________________________________
      _________________________________
      _________________________________
      End of Block: C. Street Characteristics
      Start of Block: D. Quality of the Environment
      Q5.1 D. What is the quality of the environment?
      Tabled 1
      No (1)Yes (2)
      1. Any commercial buildings adjacent to the sidewalk? (1)
      2a. Bench (excluding at transit stop)? (2)
      2b. Pedestrian-scale lighting? (3)
      2c. Other, specify: (4)
      3. Public art (e.g., statues, sculptures)? (5)
      4. Graffiti, or evidence of graffiti that has been painted over, on buildings, signs, or walls? (6)
      5. Boarded-up or abandoned buildings in the block face? (7)
      6. Empty beer or liquor bottles visible in streets, yards, or alleys? (8)
      7. Abandoned cars? (9)
      8. Burned-out buildings in the block face? (10)
      9. Any buildings have windows with bars? (11)
      10. Vacant or undeveloped land? (12)
      Q5.2 11. Litter or broken glass?
      • None or a little (1)
      • Some (2)
      • A lot (3)
      Q5.3 12. Condition of most of the buildings on the block face?
      • Poor/badly deteriorated condition (1)
      • Fair condition (2)
      • Moderately well kept condition (3)
      • Very well kept/good condition (4)
      Q5.4 13. Tree shade on the walking area?
      • None or a little (1)
      • Some (2)
      • A lot (3)
      Q5.5 14. Steepest slope along the walking area?
      • Flat/gentle (1)
      • Moderate (2)
      • Steep (3)
      Q5.6 Pedestrian environment notes:
      _________________________________
      _________________________________
      _________________________________
      _________________________________
      _________________________________
      End of Block: D. Quality of the Environment
      Start of Block: E. Place to Walk/Bicycle
      Q6.1 E. Do you have a place to walk or bicycle?
      SIDEWALKS
      Tabled 1
      1. Sidewalk present? If NO, go to E10. (1)▼ No (1) ... Yes, both sides (3)
      2. Any grassy or other buffer between curb and sidewalk along most of the segment? If NO, go to E3. (2)▼ No (1) ... Yes, both sides (3)
      2a. Tree(s) in buffer? (3)▼ No (1) ... Yes, both sides (3)
      3. Sidewalk continuous within segment? (4)▼ No (1) ... Yes, both sides (3)
      4. Sidewalk continuous between segments at both ends? (5)▼ No (1) ... Yes, both sides (3)
      5. Width >3 ft (1 m) for most of the sidewalk? (6)▼ No (1) ... Yes, both sides (3)
      6. Width (7)▼ No (1) ... Yes, both sides (3)
      7. Any missing curb cuts or ramps at intersections or driveways? (8)▼ No (1) ... Yes, both sides (3)
      8. Any major bumps, cracks, holes, or weeds in the sidewalk? (9)▼ No (1) ... Yes, both sides (3)
      9. Any permanent obstructions (trees, signs, tables) blocking the 3 ft (1 m) walk area? (10)▼ No (1) ... Yes, both sides (3)
      Skip To: Q6.2 If E. Do you have a place to walk or bicycle? SIDEWALKS = 1. Sidewalk present? If NO, go to E10.
      Q6.2 10. If a sidewalk is not present on any part of the segment, do you have another safe place to walk, including:
      Tabled 1
      10a. Street or shoulder (if safe)? (1)▼ No (1) ... Yes, both sides (3)
      10b. Unpaved pathway? (2)▼ No (1) ... Yes, both sides (3)
      10c. Other? Specify: (3)▼ No (1) ... Yes, both sides (3)
      Q6.3 Sidewalk notes:
      _________________________________
      _________________________________
      _________________________________
      _________________________________
      _________________________________
      Q6.4 SHOULDERS (OPTIONAL)
      Tabled 1
      11. Designated bike route sign or marking? (1)▼ No (1) ... Yes, both sides (3)
      12. On-street, paved, and marked shoulder? If NO, go to E16. (2)▼ No (1) ... Yes, both sides (3)
      13. Width of marked shoulder ≥4 ft (1.2 m)? (3)▼ No (1) ... Yes, both sides (3)
      14. Shoulder continuous between segments at both ends? (4)▼ No (1) ... Yes, both sides (3)
      15. Any permanent obstructions in the shoulder (including drainage grates, parked cars)? (5)▼ No (1) ... Yes, both sides (3)
      Q6.5 16. If a paved, marked shoulder is not present on any part of the segment, do you have another safe place to bicycle, including:
      Tabled 1
      16a. Street? (1)▼ No (1) ... Yes, both sides (3)
      16b. Wide outside lane (∼15ft (4.5 m))? (2)▼ No (1) ... Yes, both sides (3)
      16c. Other? Specify: (3)▼ No (1) ... Yes, both sides (3)
      Q6.6 Shoulder notes:
      _________________________________
      _________________________________
      _________________________________
      _________________________________
      _________________________________
      Q6.7 Stop Time:
      _________________________________
      End of Block: E. Place to Walk/Bicycle

      Perceived Neighborhood Disorder (Ross & Mirowsky, 1999)

      The following questions are about the neighborhood in which you live. Please take a moment to think about your neighborhood [pause]. Now I will ask you to rate your level of agreement with several statements. Please respond to each statement using the following choices: strongly agree, agree, disagree or strongly disagree. [Interviewers: Repeat the choices for the first few items and continue to repeat, if necessary.]
      Tabled 1
      Please readDo not read
      Strongly agreeAgreeDisagreeStrongly disagreeDon't know/

      Not sure
      Refused
      G1. There is a lot of graffiti in my neighborhood.123479
      G2. My neighborhood is noisy.123479
      G3. Vandalism is common in my neighborhood.123479
      G4. There are lots of abandoned buildings in my neighborhood.123479
      G5. My neighborhood is clean.123479
      G6. People in my neighborhood take good care of their houses and apartments.123479
      G7. There are too many people hanging around on the streets near my home.123479
      G8. There is too much drug use in my neighborhood.123479
      G9. There is too much alcohol use in my neighborhood.123479
      G10. I'm always having trouble with my neighbors.123479
      G11. There is a lot of crime in my neighborhood.123479
      G12. In my neighborhood, people watch out for each other.123479
      G13. The police protection in my neighborhood is adequate.123479
      G14. My neighborhood is safe.123479
      G15. I can trust most people in my neighborhood.123479
      TableSensitivity Analysis. Frequency of neighborhood conditions and unadjusted association with risk of readmission, comparing the main analysis with the sensitivity analysis by excluding those with imputed built environmental variables.
      Neighborhood Condition
      Built environment (versus No)N (%)Number (%) readmittedHazard ratio (95% CI)
      Mixed use

      93 (23.3)73 (18.3)1.05 (0.66, 1.66)

      1.05 (0.65, 1.70) Sensitivity analysis
      Abandoned buildings/vacant lots

      86 (21.5)71 (17.8)1.08 (0.78, 1.50)

      1.10 (0.79, 1.53) Sensitivity analysis
      Mixed housing

      55 (13.8)44 (11.0)0.86 (0.63, 1.17)

      0.86 (0.63, 1.17) Sensitivity analysis
      Graffiti/litter/broken windows or glass

      55 (13.8)42 (10.5)1.12 (0.77, 1.64)

      1.12 (0.76, 1.66) Sensitivity analysis
      Sidewalk quality

      104 (26.0)84 (21.0)0.83 (0.62, 1.12)

      0.82 (0.60, 1.11) Sensitivity analysis
      CI = confidence interval.

      References

        • Joynt KE
        • Jha AK.
        Who has higher readmission rates for heart failure, and why? Implications for efforts to improve care using financial incentives.
        Circ Cardiovasc Qual Outcomes. 2011; 4: 53-59
        • Virani SS
        • Alonso A
        • Benjamin EJ
        • et al.
        Heart disease and stroke statistics-2020 update: a report from the American Heart Association.
        Circulation. 2020; 141: e139-e596
        • Kind AJH
        • Jencks S
        • Brock J
        • et al.
        Neighborhood socioeconomic disadvantage and 30-day rehospitalization.
        Ann Intern Med. 2014; 161: 765-774
        • Eapen ZJ
        • McCoy LA
        • Fonarow GC
        • et al.
        Utility of socioeconomic status in predicting 30-day outcomes after heart failure hospitalization.
        Circ Heart Fail. 2015; 8: 473-480
        • Reid M
        • Kephart G
        • Andreou P
        • Robinson A.
        Potential of community-based risk estimates for improving hospital performance measures and discharge planning.
        BMJ Open Quality. 2021; 10e001230
        • Conrad N
        • Judge A
        • Tran J
        • et al.
        Temporal trends and patterns in heart failure incidence: a population-based study of 4 million individuals.
        Lancet. 2018; 391: 572-580
        • Aneshensel CS
        • Wight RG
        • Miller-Martinez D
        • Botticello AL
        • Karlamangla AS
        • Seeman TE.
        Urban neighborhoods and depressive symptoms among older adults.
        J Gerontol B: Psychol Sci Soc Sci. 2007; 62: S52-S59
        • Ross C.
        Neighborhood disadvantage and adult depression.
        J Health Soc Behav. 2000; 41: 177-187
        • Mayne SL
        • Jose A
        • Mo A
        • et al.
        Neighborhood disorder and obesity-related outcomes among women in Chicago.
        Int J Environ Res Public Health. 2018; 15: 1395
        • O'Brien DT
        • Farrell C
        • Welsh BC.
        Broken (windows) theory: a meta-analysis of the evidence for the pathways from neighborhood disorder to resident health outcomes and behaviors.
        Soc Sci Med. 2019; 228: 272-292
        • Ponikowski P
        • Voors AA
        • Anker SD
        • et al.
        2016 ESC Guidelines for the diagnosis and treatment of acute and chronic heart failure: The Task Force for the diagnosis and treatment of acute and chronic heart failure of the European Society of Cardiology (ESC). Developed with the special contribution of the Heart Failure Association (HFA) of the ESC.
        Eur J Heart Fail. 2016; 18: 891-975
        • Kelly CM
        • Wilson JS
        • Baker EA
        • Miller DK
        • Schootman M.
        Using Google Street View to audit the built environment: inter-rater reliability results.
        Ann Behav Med. 2013; 45: S108-S112
        • Ross CE
        • Mirowsky J.
        Disorder and decay: the concept and measurement of perceived neighborhood disorder.
        Urban Affairs Rev. 1999; 34: 412-432
        • Kroenke K
        • Spitzer RL
        • Williams JB.
        The PHQ-9: validity of a brief depression severity measure.
        J Gen Intern Med. 2001; 16: 606-613
        • Spitzer RL
        • Kroenke K
        • Williams JB
        • Löwe B.
        A brief measure for assessing generalized anxiety disorder: the GAD-7.
        Arch Intern Med. 2006; 166: 1092-1097
        • Riegel B
        • Carlson B
        • Moser DK
        • Sebern M
        • Hicks FD
        • Roland V.
        Psychometric testing of the self-care of heart failure index.
        J Card Fail. 2004; 10: 350-360
        • Morris AA
        • McAllister P
        • Grant A
        • et al.
        Relation of living in a “food desert” to recurrent hospitalizations in patients with heart failure.
        Am J Cardiol. 2019; 123: 291-296
        • Freedland KE
        • Carney RM
        • Rich MW
        • Steinmeyer BC
        • Skala JA
        • Dávila-Román VG
        Depression and multiple rehospitalizations in patients with heart failure.
        Clini Cardio. 2016; 39: 257-262
        • Jhund PS
        • Ponikowski P
        • Docherty KF
        • et al.
        Dapagliflozin and recurrent heart failure hospitalizations in heart failure with reduced ejection fraction: an analysis of DAPA-HF.
        Circulation. 2021; 143: 1962-1972
        • Cohen S
        • Kamarck T
        • Mermelstein R.
        A global measure of perceived stress.
        J Health Soc Behav. 1983; 24: 385-396
        • Freedland KE
        • Skala JA
        • Steinmeyer BC
        • Carney RM
        • Rich MW.
        Effects of depression on heart failure self-care.
        J Card Fail. 2021; 27: 522-532
        • Liu D
        • Schaubel DE
        • Kalbfleisch JD.
        Computationally efficient marginal models for clustered recurrent event data.
        Biometrics. 2012; 68: 637-647
        • Groenewegen A
        • Rutten FH
        • Mosterd A
        • Hoes AW.
        Epidemiology of heart failure.
        Eur J Heart Fail. 2020; 22: 1342-1356
        • Cintra RM
        • Nogueira AC
        • Bonilha I
        • et al.
        Glucose-lowering drugs and hospitalization for heart failure: a systematic review and additive-effects network meta-analysis with more than 500 000 patient-years.
        J Clin Endocrinol Metab. 2021; 106: 3060-3067
        • Lip GY
        • Skjøth F
        • Overvad K
        • Rasmussen LH
        • Larsen TB.
        Blood pressure and prognosis in patients with incident heart failure: the Diet, Cancer and Health (DCH) cohort study.
        Clin Res Cardiol. 2015; 104: 1088-1096
        • Dawson NL
        • Hull BP
        • Vijapura P
        • et al.
        Home telemonitoring to reduce readmission of high-risk patients: a modified intention-to-treat randomized clinical trial.
        J Gen Intern Med. 2021; 36: 3395-3401
        • Mooney SJ
        • Bader MDM
        • Lovasi GS
        • Neckerman KM
        • Teitler JO
        • Rundle AG.
        Validity of an ecometric neighborhood physical disorder measure constructed by virtual street audit.
        Am J Epidemiol. 2014; 180: 626-635