Abstract
Introduction
The Centers for Medicare & Medicaid Services (CMS) define a readmission as an admission to a subsection hospital within 30 days of a discharge from the same or another subsection (CMS, 2012). Hospital readmissions are at the forefront of health care debate and have been the focus of significant scrutiny surrounding high-value health care. Despite research and focused interventions, hospital readmissions remain prevalent, expensive, and potentially preventable (Boccuti & Casillas, 2017). One landmark study indicated that nearly 20% of Medicare beneficiaries were rehospitalized within 30 days after discharge at an annual expense of US$17 billion (CMS, 2012; Jencks, Williams, & Coleman, 2009). Preventable hospital readmissions are both a clinical and a financial problem (Hackbarth, 2009). For patients, readmission increases risk for complications, infections, and functional impairment (Cuffel, Bloom, Wallace, Hausman, & Hu, 2002; Forster et al., 2004; Krumholz, 2013). A study comparing readmission rates across procedures showed that many of the most common procedures in the United States also have the highest readmission rates. For example, treatment of femur/hip fracture, which is associated with morbidity and mortality and may disproportionately affect older adults, was the 17th most common procedure and had the 28th highest readmission rate (Weiss, Elixhauser, & Steiner, 2013).
Preventable readmissions represent a critical failure of the health care system and are the single largest driver of excess health care costs (Jencks et al., 2009). While not all readmissions are preventable, research suggests that unplanned readmissions could be prevented by addressing a series of risk factors patients face prior to, during, and after admission and discharge (Benbassat & Taragin, 2000). The Affordable Care Act aimed to reduce hospital readmissions by enacting the Hospital Readmission Reductions Program (HRRP), which imposed a 3% financial penalty on hospitals with 30-day readmission rates that exceeded the national average. The HRRP was initiated in 2012 and was originally applied to three clinical conditions (acute myocardial infarction, heart failure and pneumonia) and has currently expanded to additional conditions including chronic obstructive pulmonary disease (COPD), elective hip and knee replacement, and coronary artery bypass graft (Boccuti & Casillas, 2017). Despite concerns that hospitals achieved lower readmission rates by increasing their use of observation units, there is evidence that this practice did not account for lower rates of readmission (Zuckerman, Sheingold, Orav, Ruhter, & Epstein, 2016).
To continue to effectively reduce hospital readmissions, it is critical to understand both patient- and hospital-level factors that increase risk. Individual risk factors such as acute illness, chronic morbidity, comorbidity, socioeconomic status, and quality of patient care have received considerable research attention and demonstrated impact on readmission rates (Arbaje et al., 2008; Calvillo-King et al., 2013; Philbin, Dec, Jenkins, & DiSalvo, 2001; Robinson, Howie-Esquivel, & Vlahov, 2012). Community-level factors including community income level, primary care practitioner supply, and amount and quality of nursing homes are also shown to affect readmissions (Herrin et al., 2015). However, the understanding of how organizational characteristics impact readmissions is less well understood. The objective of this brief report is to examine both patient- and hospital-level factors associated with increased risk for 30-day hospital readmission using large nationally representative data.
Method
Sample
The data for this study were extracted from the 2013 Nationwide Readmission Database (NRD). The NRD is part of a family of databases and software tools developed for the Agency for Healthcare Research and Quality (AHRQ)-sponsored Healthcare Cost and Utilization Project (HCUP). It is a unique and powerful database designed to support various types of analyses of national readmission rates for all payers and the uninsured. The NRD sample includes discharges for patients treated at community hospitals in the United States. The NRD excludes patients who are less than 1 year, patients who died in the initial hospital admission (i.e., index event), and discharges from rehabilitation and long-term acute care facilities. Same-day stays and transfers, defined as index event discharges and readmissions with the same date, were also excluded. The database addresses a large gap in health care data, namely, the lack of nationally representative information on hospital readmissions for all ages. Unweighted, the NRD contains data from approximately 17 million discharges annually. Weighted, it estimates roughly 36 million discharges. The NRD is drawn from the HCUP National Inpatient Database (NID) and is constructed from 21 states with reliable, verified patient linkage numbers in the NID that are used to track patients across hospitals. The NRD includes data from states with large populations of older adults (e.g., New York, California, and Florida) and includes states that are geographically dispersed, accounting for 49.3% of the total U.S. resident population and 49.1% of all U.S. hospitalizations.
In this study, the analytic sample was restricted to include older adults who were 65 years or older at the index event and live discharges between January 1, 2013, and November 30, 2013. Of the 5,038,940 episodes for patients discharged from community hospitals during this time, 1,449,175 were for patients younger than 65 years and were excluded from analysis. The final sample for this study includes 3,589,765 hospital episodes for 2,622,961 individuals aged 65+.
Measures
Patient-level factors
Age, sex, zip income, and insurance type
Age in years was coded from 0 to 90 years, with patients aged 90 and older coded as 90. Sex was dichotomously coded male or female. The following 2013 national quartiles of the median household income for patients’ zip code was used to estimate patients’ income: (1st) US$1-US$37,999, (2nd) US$38,000-US$47,999, (3rd) US$48,000-US$63,999, and (4th) US$64,000 or more. Insurance types, defined as patients’ primary payer, included Medicare, Medicaid, private insurance, self-pay, no charge, and other. Because all patients were 65+ and therefore Medicare eligible, having Medicaid indicates dual-eligibility.
Mortality risk, illness severity, number of chronic conditions, elective admission, and weekend admission
Mortality risk and illness severity ratings were assigned in the NID using software developed by 3M Health Information Systems. Both measures include the base All Patient Refined Diagnosis Related Groups (APR-DRGs), the severity of illness subclass, and the risk of mortality subclass within each base APR-DRG. Mortality risk, defined as the likelihood of the patient dying, was categorized as minor, moderate, major, and extreme. Illness severity, based on impact on functioning, was coded as (a) minor loss of function, including cases where no comorbidity or complications were present; (b) moderate loss of function; (c) major loss of function; and (d) extreme loss of function. For analytic purposes, number of chronic conditions was categorized as less than two conditions, 2, 3, 4, 5, 6, 7, 8, and 9, or more. Both elective admission (elective/nonelective) and weekend admission (weekday/weekend) were analyzed dichotomously.
Hospital-level factors
Hospital size, urban-rural location, ownership type, and teaching status
Hospital size (small, medium, large) was based on number of beds. Hospital bed size categories are defined using region of the United States, the urban-rural designation of the hospital, in addition to the teaching status. Hospital urban-rural locations included large metropolitan areas (with at least 1 million residents), small metropolitan areas (with less than 1 million residents), micropolitan areas, and rural areas. Hospital ownership types included nonfederal public, nonprofit, and for-profit. The hospital’s teaching status was measured dichotomously (yes/no).
Outcome
30-day readmission
The main outcome variable for our investigation was 30-day all-cause unplanned hospital readmissions as defined by CMS. For the current study, it was defined as a hospitalization between January and December within 30 days of live discharge from an initial admission of a known duration between January and November. Readmissions across the same or different hospitals and those that resulted in death were included.
Data Analysis
Descriptive statistics were used to compare patients who were rehospitalized within 30 days of discharge from a prior admission. Because the NRD was compiled to reflect the population of all U.S. inpatient discharges treated at community hospitals, sampling and discharge weights were used to represent national estimates in the descriptive analyses. Multivariable logistic regression models using PROC GENMOD within SAS version 9.4 were utilized to examine whether and to what extent patient factors and hospital admission and discharge factors were associated with risk for readmission. All analyses were conducted at the individual hospital discharge level. Because episodes were clustered within facilities and because a significant number of patients who were readmitted had episodes at more than one hospital, generalized estimating equations (GEEs) with an exchangeable correlation matrix were used to account for the clustering of observations within hospitals and only the first rehospitalization was utilized for each patient so that patients were fully nested within hospitals. Sampling weights were used for the descriptive statistics (see Table 1), but not for the GEE analysis.
Comparison of Patient and Hospital Characteristics of the Index Admission by 30-Day Readmission Status.
Note. With the exception of teaching hospital status (p = .676), all other comparisons were statistically significant at the .001 level.
Results
Patient characteristics and hospital characteristics associated with the hospital readmissions are shown in Table 2. The data are presented for episodes which resulted in readmission within 30 days versus episodes which did not result in readmission within 30 days. The group differences for each variable were tested by chi-square analysis. Our overall adjusted 30-day readmission rate was 11.9% (95% confidence interval [CI] = [11.7, 12.0]), which is relatively low compared with prior population-based research which has reported finding a 20% 30-day readmission rate among Medicare beneficiaries (Jencks et al., 2009). The main factors that predicted increased risk of 30-day readmission were mortality risk, illness severity, number of chronic conditions, and hospital ownership. Not surprisingly, patients with the highest illness severity and patients with the highest mortality risk were 74% (OR = 1.74, 95% CI = [1.69, 1.79], p < .001) and 43% (OR = 1.43, 95% CI = [1.39, 1.47], p < .001) more likely, respectively, to have a 30-day readmission than patients with the lowest illness severity and mortality risk. More than 90% of the analytic sample had more than two chronic conditions, and risk of 30-day hospital readmission increased consistently with more chronic conditions. Twenty-one percent of patients had nine or more chronic conditions and were 62% more likely to have a 30-day hospital readmission than patients with less than two (OR = 1.62, 95% CI = [1.57, 1.67], p < .001). Interestingly, patients admitted to a for-profit hospital were 11% more likely to have a 30-day hospital readmission (OR = 1.11, 95% CI = [1.08, 1.14], p < .001) as compared with patients admitted to a nonprofit hospital. Notably, dual-eligible patients did not have an increased risk of 30-day readmission, but were 3% more likely to have any readmission during the year as compared with patients with other forms of insurance (OR = 1.03, 95% CI = [1.00, 1.06], p = .040).
Patient and Hospital Factors Associated With Risk of Readmission.
Note. Logistic regression conducted at the patient-level of analysis. CI = confidence interval; age = age in years at initial hospital admission; REF = reference category for comparing categorical independent variables.
Discussion
Individual factors commonly observed to be related to hospital readmissions, such as mortality risk, illness severity, and number of chronic conditions, predicted higher rates of 30-day hospital readmissions in the current study. Insurance type was also associated with 30-day readmissions. Other studies have reported preliminary evidence that Medicaid payer status may be related to increased risk in hospital readmissions (Allaudeen, Vidyarthi, Maselli, & Auerbach, 2011; Kansagara et al., 2011). Interestingly, hospital profit status (ownership) was associated with greater risk for 30-day hospital readmission in this sample. While some research has found no significant difference in readmission risk related to hospital ownership (Kind, Bartels, Mell, Mullahy, & Smith, 2010), our finding appears to be consistent with more recent research (Birmingham & Oglesby, 2018). It may be that our large, nationally representative data set captured something not evident in smaller local data sets. This finding is critical and controversial and may have implications for underlying policies or procedures that impact quality of care received before and after hospital discharge.
Contrary to our expectation, older age was not a significant predictor of increased risk for 30-day hospital readmission. This finding is contrary to previous research suggesting that readmissions are particularly prevalent among older adults and patients aged 65 (Schwarz, 2000) and older adults account for 60% of all potentially preventable hospital readmissions (Hamilton et al., 2015). However, Berry and colleagues (2018) used a similar data set and found a decrease in the risk-adjusted odds of hospital readmission from age 64 to 65 years and relatively constant odds through age 90. Given these inconsistent findings, this is an area of research with deserves additional inquiry. It should be noted that our analysis potentially included patients who died within 30 days of live discharge from the index event. We expect that these patients account for no more than a small percentage of the “not readmitted” group but recognize this as a limitation of the current project. Future research should control for 30-day mortality when examining hospital readmission rates.
Conclusion
The findings of this brief report suggest that patients with severe physical illness and multiple chronic conditions should be the primary targets for hospital transitional care interventions, which may help to reduce the rate of potentially preventable readmissions among the populations at the highest risk. It also suggests “risk adjustment” in the Medicare HRRP program. We suggest that if a hospital admits a large proportion of adults with Medicaid coverage, who are low-income, and who have multiple chronic conditions, the expected readmission rate should be adjusted so that the HRRP penalty is reduced for those hospitals.
Footnotes
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by an internal award from the College of Behavior and Community Sciences.
