Social Determinants of Health and Cardiologist Involvement in the Care of Adults Hospitalized for Heart Failure

Key Points Question Are social determinants of health associated with cardiologist involvement in the management of adults hospitalized for heart failure? Findings This cohort study of 1000 participants hospitalized for heart failure at 549 unique US hospitals found that low income was inversely associated with cardiologist involvement after adjusting for patient-based and hospital-based factors. Meaning These findings identify socioeconomic status as an important social determinant of health that may bias the care provided to patients hospitalized for heart failure.


Introduction
][8][9][10][11][12][13][14][15][16] Yet, cardiologist involvement is not universal when adults are hospitalized for HF.Some of this might relate to cardiologist availability, which is especially problematic in rural settings with shortages of cardiologists. 2,7wever, another potential contributor may be implicit bias.Indeed, in 2 studies 17,18 of adults hospitalized with HF, Black race was associated with decreased odds of cardiologist involvement in the intensive care unit (ICU) and on the floor.In another single-center study, 19 both Black and Latinx adults hospitalized for HF were less likely than White adults to be admitted to a cardiology service and subsequently experienced worse outcomes.1][22] In a multicenter study, Auerbach et al 17 found that adults admitted with severe HF who had an annual income less than $11 000 were less likely to receive care from a cardiologist.These observations raise concern that social determinants of health (SDOH), an increasingly recognized contributor to health care delivery, 23,24 lead to suboptimal care in the management of adults hospitalized for HF.

Methods
The protocol for this cohort study was reviewed and approved by Weill Cornell Medicine's and the University of Alabama at Birmingham's institutional review boards.All participants provided written informed consent.This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines for cohort studies.

Study Population
The study population included adults from REGARDS with an adjudicated HF hospitalization (as principal reason for hospitalization) in 2009 to 2017.In brief, the REGARDS cohort originally included 30 239 community-dwelling Black and White men and women aged 45 years or older recruited via mailing followed by telephone contact in 2003 to 2007 from all 48 contiguous states in the US and the District of Columbia to investigate racial and geographic disparities in stroke mortality. 33Black adults and residents of the so-called Stroke Belt (ie, Alabama, Arkansas, Georgia, Louisiana, Mississippi, North Carolina, South Carolina, and Tennessee), an area in the southeastern US with high stroke mortality, were oversampled by design; this includes a 153-county region in Georgia, North Carolina, and South Carolina known as the Stroke Buckle, which has even higher rates of stroke compared with the rest of the Stroke Belt.Participants were recruited using commercially available lists.They underwent an initial 45-minute computer-assisted telephone interview about medical history, health behaviors, and risk factors followed by an in-home visit for baseline vital signs, electrocardiograms, medication reconciliation, bloodwork, and urine samples.At 6-month intervals, participants were contacted by telephone and asked about their health status and study outcomes such as hospitalizations.When participants reported a hospitalization for a possible cardiovascular cause, their medical records were retrieved and underwent review and adjudication.HF adjudication was conducted by 2 expert clinicians (not coauthors of the present article) on the basis of medical record-level data on symptoms, physical examinations, laboratory values, imaging, and medical For this study, we excluded participants hospitalized at institutions that lack cardiology services according to linked American Hospital Association data. 34Of note, there were hospitals included in the American Hospital Association data set that lacked information on the availability of cardiology services.To mitigate risk of bias related to excluded patients hospitalized at these hospitals, we imputed values (for 176 participants) for the availability of cardiology services using a random forest method that incorporated most other American Hospital Association data (including, but not limited to, location, bed size, teaching status, geriatrics service, palliative care service, and licensed pharmacists).To do this, we used the R package missForest. 35,36We examined each participant's first adjudicated hospitalization during the study period.
Similar to prior work, [25][26][27][28][29][30][31][32] we examined (1) Black race, (2) social isolation (defined as having 0-1 visits from a family or friend in the past month), (3) social network and/or caregiver availability (defined as whether the participant reported having someone to care for them if ill), (4) low educational attainment (less than high school education), (5) low annual household income (<$35 000), 38 (6) living in rural areas (defined as living in an isolated or small rural area on the basis of Rural Urban Commuting Area Codes), (7) living in a zip code with high poverty (>25% of residence below the federal poverty level), ( 8) living in a Health Professional Shortage Area, and ( 9) living in a state with poor public health infrastructure (assessed using data from the America's Health Ranking, 39 which ranked states according to their contribution to lifestyle, access to care, and disability; states where REGARDS participants lived that fell into the bottom 20th percentile for their ranking for Ն8 of the 10 years spanning 2000 to 2010, concordant with REGARDS recruitment between 2003 to 2007, were considered to have poor public health infrastructure).

Outcome
The primary outcome was cardiologist involvement, which was defined as involvement of a cardiologist as the primary responsible clinician or as a consultant.This variable was collected through a rigorous medical record review and abstraction process previously described. 40Medical record review and abstraction involved review of admission notes, progress notes, consultation notes, discharge summaries, medication reconciliations, laboratory values, and imaging studies to collect relevant data related to the HF hospitalization including cardiologist involvement.

Covariates
Covariates were selected on the basis of the Andersen Behavioral Model, which outlines predisposing, enabling, and need factors that contribute to health care utilization. 41

Statistical Analysis
Data analysis was performed from November 2022 to January 2023.We calculated the frequency and distribution of each SDOH and examined collinearity using φ coefficients (eTable 1 in Supplement 1).We then examined associations between each SDOH and cardiologist involvement using relative risks (RRs).Similar to prior REGARDS studies, [25][26][27][28][29][30][31][32] candidate SDOH with statistically significant associations (P < .10)were retained for further multivariable analysis.All SDOH were collected at the time of initial REGARDS baseline surveys.
To describe patient demographic and clinical characteristics and hospital characteristics, descriptive statistics such as frequencies, percentages, medians, and IQRs were calculated.χ 2 tests and Mann-Whitney U tests were conducted for categorical and continuous variables, respectively, to identify significant differences (P < .05)across categories of the retained SDOH.
We then examined the association between retained SDOH and cardiologist involvement by conducting a modified Poisson regression with robust SE, adjusting for the previously described covariates.We chose a modified Poisson regression to estimate RR given the high prevalence of cardiologist involvement. 44To examine for a potential interaction by age 65 years or older, we added cross-product terms to the models and performed a Wald test.
We used 2-sided hypothesis testing with P < .05for all analyses performed.Multiple imputation by chained equations was used to impute missing SDOH and covariates to minimize bias under the assumption that data were missing at random.

Results
We In a fully adjusted model, low income remained inversely associated with cardiologist involvement during hospitalization (adjusted RR, 0.89; 95% CI, 0.82-0.97;P < .001)(Figure 2).We did not find a significant interaction with age 65 years or older.

Discussion
In this cohort study of adults hospitalized for HF, we found that low annual income (<$35 000) was associated with lower likelihood of cardiologist involvement during hospitalization.Prior work has shown the potential influence of SDOH on cardiologist involvement among adults with HF, but  studies have been limited regarding the number and breadth of SDOH included.For example, Breathett et al 18 previously reported that Black race was inversely associated with receiving care from a cardiologist among adults admitted to the ICU with HF.Their study included race but did not include socioeconomic status.Auerbach et al 17 showed that low income was inversely associated with having a cardiologist as the primary responsible clinician (primary attending physician), but their study only included those with advanced HF with reduced EF and did not adjust for other potential SDOH, such as social isolation, social network, or area of residence.Our study included 9 different SDOH (including race and income) that span multiple domains and examined adults from more than 500 hospitals across the US across the entire spectrum of left ventricular EF and severity.Moreover, our study uniquely examined involvement of cardiologists as either the primary responsible clinician  or a consultant; prior work to date has primarily examined factors associated with having a cardiologist as the primary responsible clinician.Accordingly, our study now provides one of the most comprehensive examinations of SDOH and cardiologist involvement to date.
Race is a well-established critical factor associated with bias and disparities in health outcomes, including for cardiac therapeutic options and outcomes. 45Interestingly, race did not emerge as a significant factor in our study examining the involvement of a cardiologist in the care of adults hospitalized for HF.Although this should not alter our conceptual understanding of how race affects health outcomes, our findings should serve as a reminder that there are other SDOH that are important and play critical roles in clinician behavior, care processes, and health outcomes.Our findings here further highlight the value of examining a broad range of SDOH to optimally understand the impact that these factors have within health care. 46The field would also benefit from a deeper examination of mediating factors such as discrimination, allostatic load, housing instability, and childhood exposures, to name a few.
Our observations remained even after controlling for length of stay, severity of disease, HF subtype, and hospital characteristics.This raises concern that our observation could relate to implicit or even explicit bias.Implicit bias is the subconscious stereotype or attitude that consequentially affects behavior and actions, which can contribute to health care disparities. 47,48The concept of SDOH contributing to implicit bias with a subsequent effect on care provision is not new. 16,49,50wever, this is one of the first studies to suggest that low income can contribute to implicit bias. 50cause income status may not be readily apparent to clinicians, further investigation is needed to better understand whether specific patient attributes or behaviors contribute to the suspected implicit bias observed among adults with low income.There is also a possibility of explicit bias, whereby clinician behavior may be directly impacted by awareness of the patient's income status and related notions about treatment adherence and/or affordability of services for adults with low income.Developing treatment strategies that incorporate social factors is important, but systematically withholding therapies can lead to worsening of disparities.Our observations could also be explained by patients themselves requesting a cardiology consultation; whether patients with specific SDOH such as low income are less likely to request involvement of a specialist is not known and merits additional investigation.An important factor that may mediate these findings is insurance status.Although most participants in this cohort had insurance, we were unable to differentiate patients with different types of insurance.Future work should examine whether insurance type and related concepts such as network participation and cost-sharing are contributors to our findings.It is also not known whether lower rates of cardiologist involvement mediate the well-described associations between low socioeconomic status and hospital readmission or mortality among adults with HF. [51][52][53][54] Perhaps more important than cardiology involvement are care processes such as optimizing guideline-directed medical therapy, providing counseling on self-care, and ensuring appropriate postdischarge follow-up.Rather than efforts to increase cardiology involvement (especially given shortages in select areas), it may be more fruitful to develop strategies that can facilitate these key care processes to ensure the highest quality of care to all patients regardless of SDOH, such as low income.
Our findings raise some questions related to a recent call for elicitation and documentation of SDOH experienced by adults with chronic diseases, including HF. 47 Given the ubiquitous role of the electronic medical record in patient care, some have advocated for incorporating SDOH fields into the electronic medical record as a strategy to improve the care of vulnerable populations. 47,55On the one hand, identifying adults with SDOH, such as low income, is important to identify a subpopulation at particularly high risk for poor outcomes.On the other hand, identification of SDOH could lead to implicit bias that negatively affects care provision.Consequently, it is critical that efforts intended to increase elicitation and documentation of SDOH be paired with strategies to address implicit bias from the health care system.[56][57][58][59] Clearly, effort is needed to routinely incorporate these strategies as a means to address the multitiered effects of SDOH.

Figure 1 .
Figure 1.Prevalence of Individual Social Determinants of Health 60 Social Determinants of Health and Cardiologist Involvement in the Care of Adults With HF 43sk Score on admission (a score that estimates in-hospital all-cause mortality with a range of 0 to Ն79),43length of stay, ICU stay, and in-hospital cardiac arrest.Comorbid conditions included coronary artery disease, history of arrythmia (atrial and/or ventricular), diabetes, history of stroke, and the number of noncardiac comorbid conditions (>40) spanning multiple organ systems.Hospital characteristics included each hospital's total number of beds, presence or absence of cardiac ICU, and academic teaching status.
Because the primary exposures (SDOH) are considered enabling factors, covariates included predisposing factors and need factors.Predisposing factors included age (at time of admission) and baseline selfreported race, sex, and an indicator for Stroke Belt residence.Need factors included the following 3 subcategories: HF characteristics, comorbid conditions, and hospital characteristics.HF characteristics and comorbid conditions were collected according to medical record abstraction, and hospital characteristics were collected according to American Hospital Association survey data, which have been linked to the REGARDS cohort.HF characteristics included New York Heart Association Class (class I-IV), HF subtype (HF with preserved ejection fraction [EF] if EF Ն50% or qualitative description of normal systolic function), 42 Get With The Guidelines-Heart Failure (GWTG-HF)

Table 2 .
Age-Adjusted RR for the Association of Candidate SDOH With Cardiologist Involvement a Retained for further analysis.