Heart failure (HF) accounts for 6.5 million hospital days per year. It remains unknown if socioeconomic factors are associated with hospital length of stay (LOS). We analyzed predictors of longer hospital LOS [mean (days), 95% confidence interval (CI)] among participants with incident hospitalized HF (n = 1,300) in the Atherosclerosis Risk in Communities (ARIC) cohort from 1987 to 2005. In a statistical model adjusted for median household income, age, gender, race/study community, education level, hypertension, alcohol use, smoking, Medicaid status, and Charlson comorbidity index score, Medicaid recipients experienced a longer LOS (7.5, 6.3–8.9) compared to non-Medicaid recipients (6.2, 5.7–6.7), and patients with a higher burden of comorbidity had a longer LOS (7.5, 6.4–8.6) compared to patients with a lower burden (6.2, 5.7–6.9). Median household income and education were not associated with longer LOS in multivariable models. Medicaid recipients and patients with more comorbid disease may not have the resources for adequate, comprehensive, out-of-hospital management of HF symptoms, and may require a longer LOS due to the need for more care during the hospitalization because of more severe HF. Data on out-of-hospital management of chronic diseases as well as HF severity are needed to further elucidate the mechanisms leading to longer LOS among subgroups of HF patients.
Heart failure (HF) accounts for 6.5 million hospital days per year, and is the leading cause of hospitalization among patients aged 65 and older in the United States (U.S.; Goff, Pandey, Chan, Ortiz, & Nichaman, 2000; Nieminen et al., 2005). Recent studies indicate an overall decline in hospital length of stay (LOS) among HF patients (Bueno et al., 2010; Fonarow et al., 2007). However, a longer LOS is seen in patients with more severe disease and more co-occurring illness (Forman et al., 2004; Formiga et al., 2008; Krumholz, Chen, Bradford, & Cerese, 1999), and LOS remains the main determinant of the cost of hospital stays among HF patients (Hauptman, Swindle, Burroughs, & Schnitzler, 2008). Identifying determinants of LOS among HF patients may help guide efforts in reducing the burden of hospital stays on both patients and hospitals.
Recently published data from the Atherosclerosis Risk in Communities (ARIC) cohort study indicate that neighborhood-level socioeconomic status (SES) is associated with readmissions and mortality among HF patients (Foraker et al., 2011). A study of Rome residents aged 75 and older found that decreasing neighborhood-level SES was associated with an increase in all-cause readmission rate, hospital LOS and comorbidity, as defined by the Charlson index of comorbidity (Antonelli-Incalzi et al., 2007). In addition to measures of SES, demographic, clinical, and health-behavior–related factors may contribute to a longer hospital LOS.
The association of neighborhood-level and individual-level SES with LOS has not been previously investigated among HF patients. We hypothesized that patients living in areas of lower SES and those with a lower level of educational attainment would have a longer LOS compared to patients in higher socioeconomic strata. We evaluated the association of neighborhood and individual-level SES with LOS during an incident HF hospitalization, taking into account other patient-level factors, including Medicaid status, which is often used as a proxy for individual SES (Croft et al., 1999). We further hypothesized that the LOS distribution may differ between patients discharged alive compared to those who had in-hospital deaths, and we assessed predictors of a longer LOS separately for these two groups.
Study Design and Methods
The ARIC cohort was comprised of 15,792 black and white men and women aged 45–64 years at baseline from four U.S. communities: Washington County, Maryland; suburbs of Minneapolis, Minnesota; Jackson, Mississippi; and Forsyth County, North Carolina. Blacks were sampled exclusively in Jackson and oversampled in Forsyth County. Cohort participants from Washington County and Minneapolis were predominantly white.
We analyzed ARIC cohort data over the time period 1987–2005. We restricted the analysis to participants without prevalent HF at baseline, and characterized incident hospitalized HF in the ARIC cohort (n = 1,458) as the first hospitalization with an International Classification of Diseases, Version 9 (ICD-9) discharge code of 428. We defined LOS as the number of days elapsed from the admission date to the discharge date, and excluded 156 patients with a missing admission date. In addition, we excluded two observations with a LOS longer than 90 days, resulting in a final sample size of 1,300 patients. Of those, 1,143 were discharged alive, and 147 died in-hospital.
We selected the neighborhood-level SES measure, median household income (nINC), for study from the 1990 U.S. Census. We assigned participants the nINC of the census tract in which they resided at baseline (1987–1989), and calculated nINC tertiles (low, <$24,777; medium, $24,777–36,071; high, ≥$36,071) based upon values from all census tracts in the four ARIC study communities (Foraker et al., 2011). We classified the individual-level SES measure, education level, into basic (<11 years), intermediate (12–16 years), and advanced (17–21 years).
Covariates included age, gender, race/study community, hypertension (blood pressure ≥140/90 mm Hg or taking medication for hypertension), current alcohol use, current smoking, Medicaid status as indicated in the medical record, and Charlson comorbidity index score. We incorporated the Charlson index score previously adapted for use with administrative data (Deyo, Cherkin, & Ciol, 1992). Consistent with our previous work, we assigned a high comorbidity burden to persons with an index score of two or greater, and a low comorbidity burden to persons with an index score of less than two (Foraker et al., 2011).
We estimated LOS [mean (days), 95% confidence interval (CI)] using generalized linear models accounting for the geographic clustering of patients. We log-transformed LOS for the analyses, and then back-transformed LOS for interpretability of the results. We characterized predictors of LOS for the overall sample, as well as for those discharged alive, and for those who died in-hospital. We used SAS software (SAS Institute, Cary, NC) for all analyses.
Institutional Review Board Approval
This study was conducted with the approval of the corresponding study sites of the ARIC study.
Among the 1,300 participants with incident hospitalized HF, a greater proportion were living in low nINC neighborhoods at baseline and were more likely to be male, hypertensive, and have less than a high school education (Table 1). At the time of the hospitalization, the mean age of the participants was 67 years, and nearly 11% were Medicaid recipients.
Table 1. Characteristics of Atherosclerosis Risk in Communities (ARIC) Cohort Participants with an Incident Heart Failure (HF) Hospitalization (1987–2005), Overall and by Disposition on Discharge
Overall (n = 1,300)
Alive (n = 1,153)
Dead (n = 147)
SES, socioeconomic status.
Body mass index (kg/m2)
Educational attainment (years)
Overall, the mean LOS among incident HF patients was 8.3 days (median: 6.0; range: 0–81). The mean LOS among incident HF patients discharged alive was 7.9 days (median: 6.0; range: 0–75), while for those who died in-hospital, the mean LOS was 12.0 days (median: 7.0; range: 0–81). As seen in Table 1, a greater proportion of patients who died in-hospital were living in low nINC neighborhoods, female, hypertensive, current smokers, and Medicaid recipients compared to those discharged alive.
In univariate models, neither neighborhood-level nor individual-level SES was associated with a longer LOS, whether in the overall sample, among those discharged alive, or among those who died in-hospital (Model 1, Table 2). As seen in Table 2, there was also not a consistent pattern of decreasing LOS from low to high SES. In multivariable models, adjusted for nINC, age, gender, race/study community, education level, hypertension, alcohol use, smoking, Medicaid status, and Charlson comorbidity index score, nINC and education level remained unassociated with longer LOS, and inconsistent patterns from low to high SES persisted (Model 2, Table 2).
Table 2. Mean Length of Stay (in Days, and 95% Confidence Intervals) among Atherosclerosis Risk in Communities (ARIC) Cohort Participants with an Incident Heart Failure (HF) Hospitalization (1987–2005), Overall and by Disposition on Discharge
However, in the multivariable model, both Medicaid status and Charlson comorbidity index score were significant predictors (p < .05) of longer LOS (Figure 1) in the overall sample. Specifically, Medicaid recipients experienced a longer LOS (7.5, 6.3–8.9) compared to non-Medicaid recipients (6.2, 5.7–6.7). Patients with a higher burden of comorbidity, as indicated by an index score of two or more, had a longer LOS (7.5, 6.4–8.6) compared to patients with a score of less than two (6.2, 5.7–6.9).
Multivariable model results differed by disposition of the patient on discharge. Among those discharged alive (Figure 2), a higher burden of comorbidity predicted a longer LOS (7.0, 6.0–8.1) compared to a lower burden of comorbidity (6.0, 5.4–6.6). Receipt of Medicaid was not a statistically significant predictor of a longer LOS (p = .10) among those discharged alive. In contrast, there were no significant (p < .05) predictors of longer LOS among those who died in-hospital, although there was a trend toward a longer LOS among those receiving Medicaid (p = .08).
These data include men and women from a broad age range representing diverse communities across the U.S. An additional strength of this study is that the analyses incorporated both neighborhood and individual SES data. Limitations of these data include a lack of information regarding out-of-hospital management of both HF and co-occurring diseases. In addition, medical record data abstracted for study did not consist of data to distinguish between acute and chronic HF, nor did it include data that indicated severity of disease with regard to HF.
We also acknowledge that the generalizability of these results to Medicaid recipients across the U.S. is limited. Medicaid eligibility criteria, although generally including having certain medical conditions or low individual-level income (Ku, 2005; Rosenbaum S., 2002), vary by state. In addition, due to the longitudinal nature of these data, the programs for which Medicaid recipients were eligible (i.e., special HF programs) likely changed over the course of the study.
Directions for Future Research
Data on out-of-hospital management of chronic diseases as well as HF severity are needed to further elucidate the mechanisms leading to longer LOS among subgroups of HF patients.
Furthermore, cost-containment strategies will likely be put in place to offset pending reductions in reimbursement for hospitalizations such as HF (Patient Protection and Affordable Care Act of 2010, 2010). LOS may be a target for improving efficiency and reducing healthcare costs (Clarke & Rosen, 2001), however, our results suggest that sicker patients are more likely to have longer LOS. Although an extra day or more LOS may be warranted in terms of providing quality care for these patients, the effect of a shortened LOS on HF patients on healthcare quality and patient outcomes requires further investigation.
This study was the first to our knowledge to investigate the association of neighborhood-level SES and individual-level SES with LOS among HF patients in the presence of other patient-level factors. Contrary to our hypothesis, patients living in areas of lower SES and those with a lower level of educational attainment did not have a longer LOS compared to patients in higher socioeconomic strata. Thus, our results are not consistent with a study of older adults in Rome, in which patients in the lowest decile of neighborhood income had a longer LOS than patients in the highest decile (Antonelli-Incalzi et al., 2007). Instead, we found that overall, incident hospitalized HF patients receiving Medicaid and those with a higher comorbidity burden experienced a longer LOS.
In previous studies, receipt of Medicaid has been associated with adverse health outcomes independent of neighborhood-level SES (Foraker et al., 2008; Ross & Mirowsky, 2000). Although we used educational level as a marker of individual-level SES in this cohort, it should be noted that Medicaid enrollment is often used as a surrogate for low individual-level SES in studies of hospital claims data (Croft et al., 1999), as its receipt is determined by having certain diseases or disabilities or an income below the poverty line (Ku, 2005; Rosenbaum S., 2002). Thus, receipt of Medicaid may reflect greater comorbid illness and individual-level poverty. This is an important issue, because healthcare systems may target policies toward shortening LOS in order to increase efficiency (Clarke & Rosen, 2001). With coming reductions in reimbursements for hospitalized conditions such as HF (Patient Protection and Affordable Care Act of 2010, 2010), it is imperative to consider vulnerable populations of patients before instituting cost-saving policies that may have an adverse effect on the quality of care of HF patients.
When the analyses were repeated according to patient disposition on discharge, however, having a high burden of comorbidity predicted a longer LOS only among patients discharged alive. This finding is consistent with other studies indicating a longer LOS among HF patients with more severe disease and more co-occurring illness (Forman et al., 2004; Formiga et al., 2008; Krumholz et al., 1999). Among the HF patients who died in-hospital, there were no statistically significant predictors of a longer LOS, although there was a trend toward significance for receipt of Medicaid. However, the relatively small number of patients who died in-hospital that were also Medicaid recipients precludes a clear interpretation of this result.
Implications for Practice
Medicaid recipients and patients with more comorbid disease may not have the resources for adequate, comprehensive, out-of-hospital management of HF symptoms, and may require a longer LOS due to the need for more care during the hospitalization because of more severe HF. Receipt of Medicaid and more comorbidities, taken together, are likely signs of a greater burden of illness, which requires a longer LOS in order to be well enough to be discharged from the hospital.
The ARIC study is carried out as a collaborative study supported by NHLBI contracts N01-HC-55015, N01-HC-55016, N01-HC-55018, N01-HC-55019, N01-HC-55020, N01-HC-55021, and N01-HC-55022. We thank the staff and participants of the ARIC study for their important contributions.
There are no financial relationships with any organizations that might have an interest in the submitted work in the previous 3 years, and no other relationships or activities that could appear to have influenced the submitted work.
Dr. Randi E. Foraker is an assistant professor of epidemiology and cardiology at The Ohio State University (OSU). Her roles and responsibilities at the OSU include conducting research, teaching courses, and supervising students.
Dr. Kathryn M. Rose is a Consultant at SRA International in Durham, NC. Her roles and responsibilities at SRA include conducting research for clients, and communicating the results of epidemiologic studies. She also serves on dissertation committees at the University of North Carolina at Chapel Hill in the School of Public Health as Adjunct Associate Professor of Epidemiology.
Dr. Patricia P. Chang is an assistant professor and cardiologist at the University of North Carolina at Chapel Hill (UNC-CH) in the School of Medicine. Her primary responsibilities are that of clinician, but she also conducts research as an ARIC study coinvestigator with the other coauthors, and she serves on dissertation committees at UNC-CH in the School of Public Health.
Dr. Chirayath M. Suchindran is a professor of biostatistics at the University of North Carolina at Chapel Hill (UNC-CH). His roles and responsibilities at the UNC-CH include conducting research, teaching courses, and supervising students.
Dr. Ann M. McNeill is a senior epidemiologist at Merck. Her roles and responsibilities at Merck include conducting pharmacoepidemiologic research and communicating the results of studies. She also serves on dissertation committees at the University of North Carolina at Chapel Hill in the School of Public Health as Adjunct Assistant Professor of Epidemiology.
Dr. Wayne D. Rosamond is a professor of epidemiology at the University of North Carolina at Chapel Hill (UNC-CH) in the School of Public Health. He is an ARIC study co-investigator and oversees the collection and dissemination of ARIC community surveillance data. His roles and responsibilities at UNC-CH include conducting research, teaching courses, and supervising students.