Article Text

Original research
Association of household income and adverse outcomes in patients with atrial fibrillation
1. Anna Rose LaRosa1,
2. J'Neka Claxton2,
3. Wesley T O'Neal3,
4. Pamela L Lutsey4,
5. Lin Y Chen5,
6. Lindsay Bengtson6,
7. Alanna M Chamberlain7,
8. Alvaro Alonso8,
9. Jared W Magnani9
1. 1 Department of Internal Medicine, University of Pittsburgh Medical Center, Pittsburgh, Pennsylvania, USA
2. 2 Department of Epidemiology, Emory University, Atlanta, Georgia, USA
3. 3 Department of Medicine, Emory University School of Medicine, Atlanta, Georgia, USA
4. 4 Division of Epidemiology and Community Health, School of Public Health, University of Minnesota, Minneapolis, Minnesota, USA
5. 5 Cardiovascular Division, University of Minnesota School of Medicine, Minneapolis, Minnesota, USA
6. 6 Health Economics and Outcomes Research, Life Sciences, Optum, Eden Prairie, Minnesota, USA
7. 7 Division of Epidemiology, Department of Health Research Sciences, Mayo Clinic, Rochester, Minnesota, USA
8. 8 Division of Epidemiology and Community Health, School of Public Health, University of Minnesota, Minneapolis, Minnesotaa, USA
9. 9 Division of Cardiology, Department of Medicine, University of Pittsburgh, Pittsburgh, Pennsylvania, USA
1. Correspondence to Dr Jared W Magnani, University of Pittsburgh, Pittsburgh, PA 15260, USA; magnanij{at}pitt.edu

## Abstract

Background Social determinants of health are relevant to cardiovascular outcomes but have had limited examination in atrial fibrillation (AF).

Objectives The purpose of this study was to examine the association of annual household income and cardiovascular outcomes in individuals with AF.

Methods We analysed administrative claims for individuals with AF from 2009 to 2015 captured by a health claims database. We categorised estimates of annual household income as <$40 000;$40–$59 999;$60–$74 999;$75–$99 999; and ≥$100 000. Covariates included demographics, education, cardiovascular disease risk factors, comorbid conditions and anticoagulation. We examined event rates by income category and in multivariable-adjusted models in reference to the highest income category (≥$100 000). Results Our analysis included 336 736 individuals (age 72.7±11.9 years; 44.5% women; 82.6% white, 8.4% black, 7.0% Hispanic and 2.1% Asian) with AF followed for median (25th and 75th percentile) of 1.5 (95% CI 0.6 to 3.0) years. We observed an inverse association between income and heart failure and myocardial infarction (MI) with evidence of progressive risk across decreased income categories. Individuals with household income <$40 000 had the greatest risk for heart failure (HR 1.17; 95% CI 1.05 to 1.30) and MI (HR 1.18; 95% CI 0.98 to 1.41) compared with those with income ≥$100 000. Conclusions We identified an association between lower household income and adverse outcomes in a large cohort of individuals with AF. Our findings support consideration of income in the evaluation of cardiovascular risk in individuals with AF. • atrial fibrillation • quality and outcomes of care http://creativecommons.org/licenses/by-nc/4.0/ This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/. View Full Text ## Statistics from Altmetric.com ## Introduction Atrial fibrillation (AF) is a common, highly morbid cardiac arrhythmia that is associated with multiple cardiovascular and non-cardiovascular complications. AF results in significant social and medical burdens and accounts for$6 billion in annual US healthcare spending.1 Care for AF is complex and requires adequate social resources and health literacy to monitor for symptoms, navigate and adhere to complicated medication regimens (such as anticoagulation) and coordinate primary and specialty care services.2

Social determinants of health are relevant to AF given their potential to affect treatment and associated outcomes. To date, the foremost focus of social determinants in AF has been on race. Studies have underscored racial differences in treatment, awareness of the condition and risk for adverse outcomes.3 4 In a community-based study, income has likewise been related to increased risk of developing AF.5 Median neighbourhood income has also been positively associated with access to direct oral anticoagulants (as opposed to warfarin).6 Additional social determinants of health, such as neighborhood-level and community-level factors and healthcare access, likewise have relevance to health outcomes7 and may mediate associations between income and health outcomes in AF. Examination of the interaction of race and income in AF remains limited and may demonstrate how inter-related social factors contribute towards adverse outcomes and thereby direct treatment efforts.

We investigated the association of household income and cardiovascular outcomes in individuals with AF. We used a database of deidentified, aggregated commercial and Medicare Advantage health claims, thereby enabling us to conduct our analysis in a large, socially diverse cohort with enhanced generalisability. Our primary hypothesis was that lower income would be associated with increased risks of heart failure, myocardial infarction and stroke in this large cohort of individuals with AF. Our secondary hypothesis was that lower income would be a stronger risk factor for adverse outcomes in racial and ethnic minorities compared with white referents.

## Methods

### Cohort selection

We obtained data from Optum Clinformatics Data Mart (Eden Prairie, Minnesota, www.optum.com), a large US database composed of inpatient, outpatient, emergency department, pharmacy and laboratory health claims. Data are deidentified, and informed consent is thereby waived. Medical claims include International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) diagnosis and procedure codes; Current Procedural Terminology, Version 4 procedure codes; Healthcare Common Procedure Coding System procedure codes; and site of service codes. The database includes commercial and Medicare Advantage enrollees and is geographically diverse across the USA and similar to the insured population.8 Patients or the public were not involved in the design, or conduct, or reporting, or dissemination of our research.

We selected individuals having a diagnosis of AF during the period of 1 January 2009–30 September 2015. AF was defined by the presence of at least one inpatient or two outpatient claims for AF (ICD-9-CM 427.31 or 427.32 in any diagnostic position). We required two outpatient claims in order to minimise the impact of rule-out diagnoses and improve the specificity of our definition of AF; this restriction is not applied to inpatient claims, which have greater specificity for the diagnosis of AF. This approach has been validated elsewhere.9 Outpatient AF diagnostic codes were temporally separated by a minimum of 1 week to a maximum of 1 year. Date of diagnosis (first in-patient claim or second out-patient claim) was defined as the date of study entry for time-dependent analyses. We used the second outpatient claim of AF to preclude the immortal time bias inherent to initiating analysis at the first date of AF diagnosis. We identified 727 935 individuals with a diagnosis of AF.

### Household income

Household income was derived by AmeriLINK Consumer Marketing Database, which provides estimates of annual household income. Income data are collected by monthly survey from a representative cross section of the US population of >30 000 households and are informed by 130 variables that encompass ZIP +4 (a highly specific geographic locator), Internal Revenue Service data, address-level home value, aggregated credit and short-term loans. Derived estimates of household income are validated by comparison to self-reported income collected by household surveys.10 The claims database divides household income into six categories: <$40 000,$40–$49 999,$50–$59 999,$60–$74 999,$75–$99 999 and ≥$100 000. For this analysis, we categorised income as <$40 000;$40–$59 999;$60–$74 999;$75–$99 999; and ≥$100 000. We combined the $40–$49 999 and $50–$59 999 categories due to the limited numbers cohort participants in these two categories of estimated annual household income.

### Outcome ascertainment

We identified incident cardiovascular events that occurred within the enrolment period after the date of AF diagnosis. The outcomes of interest were obtained from inpatient claims and included heart failure,11 myocardial infarction12 13 and ischaemic stroke.14 Each outcome was defined using the primary discharge diagnosis in an inpatient claim as follows: heart failure was defined by the presence of ICD-9-CM codes 402.x1, 404.x1, 404.x3 and 428; myocardial infarction was defined by the presence of ICD-9-CM discharge diagnosis code of 410.xx; ischaemic stroke was defined by the presence of ICD-9-CM codes 434.xx and 436.xx. online supplementary table 1 provides a comprehensive list of ICD codes used to identify the outcomes of heart failure, myocardial infarction and stroke.

### Covariates

Age, sex and race are included in the claims data. The database collects race and ethnicity from public records (eg, driver’s licence) and by imputation with commercially available software (E-Tech, Ethnic Technologies, South Hackensack, New Jersey) that employs validated algorithms incorporating racial and ethnic neighbourhood composition as ascertained by the US Census, residential zip code and first and last name.15 Race and ethnicity were subsequently categorised as white, black, Asian or Hispanic. Education level was derived from Census data at the ZIP +4 level and categorised as less than high school diploma, high school diploma, less than bachelor degree, bachelor degree or higher or unknown. For this analysis, we categorised education as less than high school diploma and high school diploma; less than bachelor degree; or bachelor degree or higher. Additional clinical covariates were selected from prior analyses of AF with recognised contributions to the outcomes studied here and relevance to AF.16 These covariates included hypertension, diabetes, prior coronary heart disease, prior heart failure, prior stroke, chronic obstructive pulmonary disease and chronic kidney disease. All covariates were defined by ICD-9-CM codes in inpatient and outpatient claims prior to or at the time of AF diagnosis, as listed in online supplementary table 1.

### Oral anticoagulation

We identified oral anticoagulants including warfarin, dabigatran, rivaroxaban and apixaban prescribed within 3 months prior to 6 months after the date of AF diagnosis. Data were obtained from outpatient pharmaceutical claims, which provide the National Drug Code, the prescription fill date and the number of days supplied.

We summarised the distributions of continuous and categorical variables. Our primary analysis was the association of income with incident heart failure, myocardial infarction and ischaemic stroke in individuals with AF. Date of AF diagnosis was defined as time 0, and time of study entry for each individual was included in our cohort. We calculated the rates of incident events during follow-up through database disenrolment or 30 September 2015, whichever came first. We then examined associations of income with myocardial infarction, heart failure and ischaemic stroke in multivariable-adjusted Cox proportional hazards models that compared risk by income category using income ≥$100 000 as the referent. For each outcome assessed, we excluded the respective prevalent disease. We then examined for interactions by race and income as well as for income and education. All analyses were adjusted initially for age, sex and race (model 1); then for age, sex, race, education, hypertension, diabetes, ischaemic stroke, coronary heart disease and heart failure (model 2); and then adjusted for all covariates including age, sex, race, education, hypertension, diabetes, ischaemic stroke, coronary heart disease, heart failure, chronic obstructive pulmonary disease and chronic kidney disease as well as oral anticoagulant use (model 3). We stratified individuals by race and ethnicity to assess for differences in risk by income level to investigate differences by race and ethnicity strata. We verified the assumption of proportional hazards with Schoenfeld residuals. All statistical analyses were performed using SAS V.9.4. Given the potential for residual confounding in our analysis, we conducted a bias analysis using the methods articulated by VanderWeele and Ding.17 Specifically, we calculated the E-value for the HRs obtained comparing the lowest to the highest income category for those endpoints in which we found associations. The E-value can be interpreted as the minimum strength of association that an unmeasured confounder would need to have with both the exposure and the outcome, after adjusting for the measured covariates, to explain away an observed association.17 ## Results After excluding individuals with <180 days of enrolment before a diagnosis of AF (n=3 12 958), and those missing income (n=66 822), race or ethnicity (n=11 130), or education (n=289), there were 336 736 individuals included in the analysis (mean age 72.7±11.9 years; 44.5% women), as summarised in table 1. Table 1 Characteristics of AF cohort, 2009–2015 The majority were white race (82.6%) with 28 286 (8.4%) black, 23 510 (7.0%) Hispanic and 6976 (2.1%) Asian. A greater proportion of black (54%) and Hispanic (45%) individuals belonged to the lowest income category compared with whites (37%) or Asians (28%). Education was likewise distributed by income, with a greater proportion of individuals with lower education belonging to lower income categories. In contrast, 58% of individuals with a bachelor’s degree or higher were in the highest income category. During a median (25th and 75th percentile) follow-up of 1.5 (95% CI 0.6 to 3.0) years, there were 4736 cases of heart failure; 1444 cases of myocardial infarction; and 3435 cases of stroke. table 2 summarises the incidence rates for each of the outcomes by income category. With increasing income category, event rates decreased with respect to myocardial infarction, heart failure and stroke. Table 2 Age-standardised, sex-standardised and race-standardised incidence rates and 95% CIs (per 1000 person-years) of cardiovascular disease (heart failure, myocardial infarction and ischaemic stroke) by income categories Table 3 summarises the HRs with p values for the trend of the associations of income with each of the cardiovascular outcomes, while figure 1 presents them graphically. Table 3 HRs and 95% CIs of cardiovascular outcomes by income category and adjusted for covariates in cohort with AF Figure 1 A–C: HRs with 95% CIs of incident heart failure, myocardial infarction and ischaemic stroke by income category and adjusted for covariates in cohort with AF. Figure parts A–C are forest plots of multivariable-adjusted HRs with 95% CIs of incident cardiovascular outcomes heart failure, myocardial infarction and ischaemic stroke by income category. Incident event rates were calculated and related to income in multivariable-adjusted Cox proportional hazard models that compared risk by income category with income >$100 000 as referent. Model 1: adjusted for age, sex and race; model 2: adjusted for age, sex, race, education, hypertension, diabetes, prevalent coronary heart disease and prevalent heart failure; model 3: adjusted for age, sex, race, education, hypertension, diabetes, prevalent coronary heart disease, prevalent heart failure, chronic obstructive coronary disease, chronic kidney disease and oral anticoagulant use. (A) Heart failure by income category; (B) myocardial infarction by income category; (C) ischaemic stroke by income category adjusted for the three models. AF, atrial fibrillation.

We observed a graded, inverse association between income and the risks for heart failure and myocardial infarction. Individuals in the lowest income category had the greatest risk for heart failure with a HR of 1.17 (95% CI 1.05 to 1.30) and myocardial infarction with an HR of 1.18 (95% CI 0.98 to 1.41) relative to the highest income category. The association between income and heart failure remained significant in the lowest income category after multivariable adjustment with the clinical covariates, oral anticoagulant use and education. We did not observe an association between income and ischaemic stroke.

Interactions for race/ethnicity and income and for income and education were not statistically significant in multivariable-adjusted models. The results of analyses stratified by race and ethnicity are presented in online supplementary table 2A and 2D. Our bias assessment identified an E-value of 2.17 (CI, lower limit, 1.88) for the association between extreme income categories and heart failure in model 1, and an E-value of 2.04 (CI, lower limit, 1.54) for the association between extreme income categories and myocardial infarction in Model 1. We concluded from these calculations that potential unmeasured confounders, such as neighbourhood SES or racial segregation, have weaker associations with the endpoints of interest and therefore are unlikely to be responsible for the observed associations.18 19

## Discussion

### Rationale for the association of income and health outcomes

Multiple potential pathways have been proposed to relate income and health. The socioecological theory asserts that health is shaped by multiple factors: social, family and community networks; living and working conditions; and broad social, economic, cultural and environmental conditions and policies.25 Low income as determined by socioeconomic status has been associated in general with diminished access to preventative care,26 specialty care27 and poorer clinical outcomes.28 Low socioeconomic status yields decreased access to prescription medications29 and may contribute towards competing priorities for using health services, medications and diverse other needs. Low income may complicate adherence to medications or appointments, as individuals struggle to choose between material necessities and medical care. Furthermore, limited health literacy is more prevalent in lower income individuals and likewise may add to the heterogenous contributions of social determinants on complicated chronic disease such as AF.2 This study demonstrates an important association between income and cardiovascular disease in AF and we recognise that further research is needed to understand the complex pathways by which patients with low income and AF develop adverse outcomes.

### Relevance of social determinants to clinical care and practice

Recognising that social determinants are associated with health outcomes, there has been increased focus on their incorporation into routine clinical care. The National Academy of Medicine has recommended inclusion of social and behavioural determinants in electronic health records due to the ‘substantial empirical evidence of the contribution of social and behavioural factors to function status and the onset, progression, and effective treatment of disease’.30 Our analysis indicates the substantive contribution of income towards adversity and its prominent role as a social determinant of health. Our study is unable to distinguish the multiple mechanisms and pathways that may relate income and adverse cardiovascular outcomes; however, our findings underscore the importance of further studies to examine the practical implications of using income in routine clinical assessment with the goal of identifying and effectively intervening on high-risk patient populations. Collecting and using income data as one component of a risk stratification tool that incorporates a suite of social determinants of health in the clinical setting may be of interest to clinicians seeking to identify and intervene on vulnerable patients, health systems attempting to improve the outcomes of populations and health plans seeking to provide high value, cost conscious care.

### Strengths and limitations

Our analysis had several strengths, most particularly the availability of nationwide health claims data from over 300 000 individuals diagnosed with AF. The generalisability of our analysis to insured persons with AF is strengthened by the utilisation of a database that is geographically diverse and racially representative as well as inclusive of enrollees of both private and Medicare advantage plans.

We would also like to summarise the important limitations of our study. First, inclusion in this cohort required that individuals have health insurance, thereby allowing health claims data to be captured. Consequently, the exclusion of a highly vulnerable patient population may underestimate the effect of income on adverse cardiovascular outcomes in individuals with AF and may not be generalisable to the uninsured population. Second, using claims data selects for individuals with AF who are more often in need of medical care, thereby limiting the generalisability to individuals with AF who do not seek care and do not generate health claims. Third, requiring two outpatient claims for AF was intended to increase specificity of the diagnostic algorithm but excluding individuals with only one outpatient claim may have selected for individuals with more symptomatic AF, again limiting the generalisability of our results to patients with asymptomatic paroxysmal AF. Fourth, we relied exclusively on health claims for ascertainment of AF, covariates and the cardiovascular outcomes; we are not able to correlate administrative claims with clinical data by individual review of health records. Fifth, several variables including income, race and education were derived using varied algorithms, thereby subject to misrepresentation. Sixth, we expect that there is residual confounding by omission of informative, socially relevant variables (neighbourhood-level determinants, treatment adherence and health literacy) that may be part of a causal pathway to relate income and adverse clinical outcomes associated with AF. However, our assessment for uncontrolled confounding suggested that only a strong unmeasured confounder (E-value >2) would be able to explain the observed associations. Seventh, our follow-up time for this study was limited to a median of 1.5 years. We expect that a longer study interval would have yielded more events to include in the analysis. The limited follow-up time also allows for the potential of reverse causation. Eighth, our analyses do not account for the effect of AF treatment during the observation period. However, we included prescription of oral anticoagulation for stroke prevention as a fundamental metric of routine clinical care for AF. Finally, we recognise our categorisation of income as limited. The measurement of income is complex and may be adjusted for family size, the addition of non-cash benefits such as food stamps or Medicare or broadened to include wealth and assets that are separate from income.25

## Conclusion

In conclusion, in a retrospective analysis of a large US healthcare utilisation database, we observed a significant association between income and risk of cardiovascular outcomes relevant to AF, specifically heart failure and myocardial infarction. Our results are consistent with prior evidence that underscores the relevance of income to increased risk for cardiovascular and health outcomes. Further study must now address how to incorporate data regarding income into the treatment of patients with AF to improve cardiovascular outcomes and mitigate adverse outcomes in vulnerable patient populations.

### Key messages

#### What is already known on this subject?

• Income is strongly associated with cardiovascular disease risk and mortality. There has been limited knowledge of the association between income and adverse cardiovascular outcomes in individuals with atrial fibrillation.

#### What might this study add?

• This study identified significant associations between household income and risks for myocardial infarction and heart failure in individuals with prevalent atrial fibrillation. Individuals in the lowest income quartile had 1.3-fold to 1.4-fold greater risk for myocardial infarction and heart failure relative to the highest income quartile. These results support prior evidence demonstrating the relevance of social resources to health outcomes.

#### How might this impact on clinical practice?

• Household income is a social determinant of health that mediates access to care and health outcomes. Incorporating social factors in health systems and the provision of care provide opportunities to promote equity, address disparities and improve outcomes in vulnerable patient populations.

View Abstract

## Footnotes

• Contributors AA and JWM conceived and designed the primary aim and method of this study. JC, LB, WTO, PLL, LYC and AMC designed and implemented data collection. AA, JWM, JC and ARL analysed the data. JWM and ARL drafted the manuscript. JWM and AA supervised the project. AA and PLL obtained funding. All authors critically reviewed and developed the final manuscript.

• Funding Research reported in this publication was supported by the National Institutes of Health's National Heart, Lung and Blood Institute award numbers R01HL122200 and R01HL143010. This work was supported by Grant 2015084 from the Doris Duke Charitable Foundation, New York, New York, and by American Heart Association grant 16EIA26410001.

• Disclaimer The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

• Competing interests Disclosures: LB is an employee of Optum. There are no other author disclosures.

• Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.

• Patient consent for publication Not required.

• Ethics approval Analysis was conducted at Emory University with the approval of the Institutional Review Board.

• Provenance and peer review Not commissioned; externally peer reviewed.

• Data availability statement The data employed in this analysis are not publicly available as they belong to Optum Cliniformatics. The data may be obtained by appropriate request and cost.

## Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.