Socio-demographic and -economic factors associated with 30-day readmission for conditions targeted by the hospital readmissions reduction program: a population-based study

Background Early hospital readmissions remain common in patients with conditions targeted by the CMS Hospital Readmission Reduction Program (HRRP). There is still no consensus on whether readmission measures should be adjusted based on social factors, and there are few population studies within the U.S. examining how social characteristics influence readmissions for HRRP-targeted conditions. The objective of this study was to determine if specific socio-demographic and -economic factors are associated with 30-day readmissions in HRRP-targeted conditions: acute exacerbation of chronic obstructive pulmonary disease, pneumonia, acute myocardial infarction, and heart failure. Methods The Nationwide Readmissions Database was used to identify patients admitted with HRRP-targeted conditions between January 1, 2010 and September 30, 2015. Stroke was included as a control condition because it is not included in the HRRP. Multivariate models were used to assess the relationship between three social and economic characteristics (gender, urban/rural hospital designation, and estimated median household income within the patient’s zip code) and 30-day readmission rates using a hierarchical two-level logistic model. Age-adjusted models were used to assess relationship differences between Medicare vs. non-Medicare populations. Results There were 19,253,997 weighted index hospital admissions for all diagnoses and 3,613,488 30-day readmissions between 2010 and 2015. Patients in the lowest income quartile (≤$37,999) had an increased odds of 30-day readmission across all conditions (P < 0.0001). Female gender and rural hospital designation were associated with a decreased odds of 30-day readmission for most targeted conditions (P < 0.05). Similar findings were also seen in patients ≥65 years old. Conclusions Socio-demographic and -economic factors are associated with 30-day readmission rates and should be incorporated into tools or interventions to improve discharge planning and mitigate against readmission.


Introduction
Early hospital readmissions are associated with increased healthcare costs and remain common, particularly within the Medicare population [1,2]. For example, one in five patients hospitalized with an acute exacerbation of chronic obstructive pulmonary disease (AECOPD) are rehospitalized within 30 days of discharge, of which 10-55% may be preventable [3][4][5]. High rates of preventable readmissions are not specific to AECOPD and also occur with other common comorbidities including heart failure (HF), acute myocardial infarction (AMI), and pneumonia (PNA) [2]. Given the potentially high number of preventable readmissions, the Centers for Medicare and Medicaid Services (CMS) implemented the Hospital Readmissions Reduction Program (HRRP) in 2012 [6]. This policy penalizes hospitals exceeding expected rates of readmission within 30 days of discharge. In 2013, 67% of hospitals were penalized due to excessive readmissions, a disproportionate number of which were safety net hospitals serving low-income patients [7]. The program was targeted to three conditions, AMI, HF, and PNA, and, in 2015, the policy was expanded to cover readmissions associated with AECOPD [6].
Social risk factors are important determinants of health outcomes and are disproportionately represented in high-needs populations [8,9]; for example, the lowest socioeconomic groups are up to 14-times more likely to have respiratory disease [10]. Certain social characteristics such as race, ethnicity, socioeconomic status, place of residence, and disability may predict readmission risk, particularly in complex conditions such as HF and AMI [11][12][13][14]. Overall, low socioeconomic status and social disadvantage have been shown to be associated with an increased risk of readmission; however, there are few large population-level studies examining the relationships between sociodemographic and socioeconomic status and readmissions. From a policy perspective, controversy also exists as to whether readmissions measures used to reimburse hospitals should adjust for socioeconomic factors [15][16][17][18][19][20][21]. On the one hand, socioeconomic adjustment would avoid penalizing hospitals caring for disadvantaged patients, but on the other it could inadvertently excuse the delivery of substandard care to disadvantaged populations, committing hospitals to different standards for the outcomes of patients based on their socioeconomic background [17,18,20]. Alternatively, effective interventions based on a patient's social determinants of health (SDoH) beyond the hospital will be necessary to ultimately prevent hospital readmissions [22]. Providers and hospitals recognize the association of social needs with patient outcomes, yet these groups may be reluctant to assume responsibility for a patient's social-related needs given the complexity addressing these needs coupled with increasing clinical demands [23,24]. Identifying and addressing a patient's SDoH through the development of accessible and evidence-based programs will be needed as a complimentary approach to improve health outcomes.
Medical care only accounts for 10-20% of the modifiable contributors to health outcomes with up to 80% contributed through a patient's SDoH [25][26][27]. Current readmission policy does not adequately account for patient sociodemographic factors, which may further drive health inequity [28]. An approach to address the current shortcomings may be a combination of policy changes and concurrently working towards implementation of community-level evidence-based interventions. Additional evidence is required showing differences in health care utilization based on socio-economic and -demographic characteristics to ultimately enrich risk adjustment models, advocate towards policy changes, or develop interventions so they reflect a patient's whole health rather than just their comorbidity burden [29]. Therefore, the objective of this study was to determine whether three social and economic characteristics (gender, median income associated with a patient's zip code, and urban/rural hospital designation) are associated with 30-day readmission in patients with HRRP-targeted conditions (AECOPD, AMI, HF, and PNA).

Data source
The Nationwide Readmissions Database (NRD) is one of the largest publicly-available all-payer databases in the United States. The NRD is derived from the Health Care Utilization Project's (HCUP) State Inpatient Databases sponsored by the Agency for Healthcare Research and Quality (AHRQ). The NRD includes admission information for community hospitals but not rehabilitation and long-term acute care hospitals. The unweighted sample compiles approximately 50% of all U.S. hospitalizations. With the NRD's complex survey design, sample weights are applied to raw data to generate national estimates. The University at Buffalo Institutional Review board exempted this study from review, as data were deidentified and publicly available through the AHRQ.

Study sample
Data from the NRD compiled approximately 19 million weighted index admissions of individuals with a diagnosis of AECOPD, AMI, HF, and PNA. Stroke was studied as a control condition because it was not included in the HRRP. International Classification of Disease, Ninth Revision, Clinical Modification (ICD-9-CM) codes were used to identify patients with each of the five conditions between January 2010 and September 2015. After September 2015, International Classification of Disease, Tenth Revision (ICD-10) codes were implemented, so these data were excluded to avoid code inconsistencies. Additionally, we limited our index hospitalization to January 1 to November 30, 2010-2014 in order to capture all 30 day readmissions. ICD-9-CM codes were selected based on the codes created for the CMS for the HRRP to assess all-cause readmissions for AECOPD, HF, AMI, PNA, and stroke [30][31][32][33]. Adults 18 years and older with the diagnoses of HF, AMI, PNA, and stroke were included. Only adults ≥40 years of age with a diagnosis of AECOPD were identified for our study sample to be consistent with previous COPD readmission studies [34][35][36]. Patients were excluded if they died during the index hospitalization, were discharged against medical advice, whose length of stay was missing, whose readmission was elective, or who were discharged to another acute care facility.

Primary exposures
The primary exposures were three socio-demographic and -economic factors: (i) patient gender (male vs. female); (ii) hospital's urban/rural designation; and (iii) median income associated with patient zip code. Urban/ rural designation was based on the county of the hospital as identified by the American Hospital Association and categorized as urban (large and small metropolitan) or rural (micropolitan and not metropolitan) [37]. This categorization is consistent with the Urban Influence Codes utilized by the NRD, which groups large and small metropolitan areas as "metropolitan" and micropolitan and non-metropolitan areas as "not metropolitan". The NRD categorizes income into quartiles based on the estimated median household income of residents in the patient's zip code ($1-37,999, $38,000-$47,999, $48,000-$63,999, and ≥ $64,000) derived from zip codedemographic data obtained from Claritas [37]. The estimated median household income data were reported in 2013 dollars.

Covariates
Covariates were included due to their potential to impact 30-day readmissions and were stratified based on subgroup. Patient and hospital characteristics included age, comorbidities, insurance type, and index length of stay. Age was stratified into 18-39, 40-64, 65-74, and 75-90 years. After reviewing the current literature, specific comorbidities for these populations were analyzed such as congestive heart failure, diabetes (uncomplicated), hypertension (uncomplicated and complicated), and renal failure. Insurance type was classified as Medicare, Medicaid, private, and self-pay/no charge. The Elixhauser Comorbidity Index was utilized to develop mortality and readmission scores based on the 29 comorbidity variables in the HCUP database [38]. Current NRD comorbidities were altered to match the Elixhauser Comorbidity Index scores utilized by the HCUP.

Outcomes
The primary outcome was unplanned all-cause 30-day readmission, consistent with definitions set by the CMS HRRP measures [39]. The 30-day readmission variable was defined as any patient with a readmission 1-30 days after discharge from an index hospitalization. These readmissions were not counted as a new index hospitalization. All additional hospitalizations during the 30-day period were not included as a 30-day readmission for the same condition. Hospitalizations occurring after this 30-day period were considered an index hospitalization. This consistent with HRRP policy for counting readmissions.

Statistical analysis
To account for the complex survey design of the NRD, we used a survey-specific methodology with hospital as clusters, NRD stratum as strata, and discharge-level weights as weights to obtain study population data and weighted nationwide overall and annual 30-day readmission rates for each targeted condition. Baseline demographic and hospital characteristics were summarized into two groups based on whether the subject was readmitted within 30 days. Categorical variables were reported as percentages within each targeted condition. Hierarchical two-level logistic models with hospital ID as a random effect were used to evaluate the relationships between the targeted conditions and 30-day readmission. All models were adjusted for demographic variables, hospital characteristics, and Elixhauser comorbidity and mortality risk scores. Five separate multivariable models were run for each condition. An ageadjusted model was created for the social characteristics with the < 65 and ≥ 65 year age groups to analyze readmission rates in both the Medicare population (the policy target group) and the non-Medicare population. All statistical tests were two-tailed with a P < 0.05 level of significance and analyses were conducted using SAS version 9.4 (SAS Institute, Cary, NC).

Sample demographics
There were 19,253,997 weighted index hospital admissions for all diagnoses and 3,613,488 30-day readmissions between 2010 and 2015. Table 1 presents demographic data stratified by condition and readmission event. Readmissions were approximately equal for males and females across conditions except for AECOPD, where women accounted for 56% of readmissions, and for AMI, where men accounted for 56% of readmissions. Hospital designation was consistent across There was a significant trend towards higher readmission rates based on income quartile (Fig. 2D). Median household income was associated with an increased odds of readmission for income quartile ≤$37,999 (aOR 1.066; 95% CI 1.059-1.073, p < 0.0001) and $38,000-$47,999 (aOR 1.020; 95% CI 1.013-1.027, P < 0.0001).   Fig. 4C). All median household income groups showed a significantly increased odds of 30-day readmission compared to those in the highest income quartile (P < 0.001) (Fig. 4D).

Age-adjusted analyses between Medicare and non-Medicare populations
Males with a diagnosis of COPD and < 65 years old had an increased odds of early readmission (aOR 1.17; 95% CI 1.16-1.18, P < 0.0001) compared to the other four target conditions (Table 3). Hospitals with urban designations, a diagnosis of COPD, and < 65 years old were associated with an increased odds of early readmission (aOR 1.13; 95% CI 1.11-1.15, P < 0.0001). With respect to median household income, those with the greatest odds of early readmission were patients diagnosed with HF aged ≥65 years and with ≤$37,999 total income. The adjusted odds ratios between the defined age groups were also notable: females diagnosed with a stroke and < 65 years old had an increased odds of readmission (aOR 1.042; 95% CI 1.017-1.068, p < 0.0001) compared to those ≥65 years old (aOR .899; 95% CI 0.885-0.913, P < 0.0001). Patients diagnosed with pneumonia and < 65 years old also had a decreased odds of readmission in all median household income categories, and those ≥65 years old all had an increased odds of early readmission.

Discussion
This study utilized the National Readmissions Database to evaluate the impact of social characteristics on 30-day readmission rates for targeted conditions. For the targeted conditions, lowest income quartile, male gender, and urban hospital designation were associated with an increased odds of 30-day readmission. Within the ageadjusted analysis, there were significant differences in the risk factors for early readmission for different targeted conditions. Social and economic disparities will continue to play an important role in patient care and health outcomes. Current evidence suggests that up to 80% of a patient's health outcomes are a result of social, behavioral, and economic factors rather than their medical care [25][26][27]. We have identified social and economic factors that continue to support the potential differences in health outcomes based on these social determinants of health. A major criticism of the HRRP is that hospitals in low socioeconomic areas incur more penalties for treating more complex patients. This is especially problematic for safety net hospitals. These hospitals serve medically and socially vulnerable patients and consequently have higher 30-day readmission rates [40]; safety net hospitals could therefore be disproportionately fined by the HRRP due to the populations they serve [41]. Maddox and colleagues argued that social risk is outside a hospital's Table 3 Age-adjusted associations between 30-day readmission and social risk factors in chronic obstructive pulmonary disease, pneumonia, acute myocardial infarction, heart failure, and stroke a,b control and hospitals should therefore not be penalized for treating patients at higher risk [42]. Social factors may be associated with higher readmission rates due to post-discharge healthcare access issues. Individuals with disproportionate social disadvantages may be unable to afford prescriptions, lack adequate transportation for follow-up appointments [43][44][45], have poor health literacy [45,46], or be unable to follow self-care regimens [45,47]. Attention to the association of social needs with medical outcomes is widespread, but dissemination of care delivery innovations in hospitals and medical practice is deficient [23,48,49]. Focusing post-discharge interventions on patient-related social needs will be just as important as any policy adjustment related to the CMS hospital readmission program [22]. This increasing recognition of the importance of social and health inequity justifies the development for accessible program that incorporate whole-person healthcare. Patients in the lowest income quartile were more likely to be readmitted, consistent with previous literature [50][51][52]. Low socioeconomic status has been associated with low health literacy, poor social support, and a higher prevalence of comorbidities such as hypertension, diabetes, and obesity [53]. The higher prevalence of these conditions contributes to comorbid burden and results in complex patients who may be more difficult to treat. Low income has also been linked to disparities in healthcare access; patients with low incomes may have trouble affording medications and obtaining transportation to appointments [53], factors independent of the quality of care received at a hospital. Knowing that low income patients are at risk of early readmission could become an important consideration in transition of care programs. Socially vulnerable patients enrolled in community support-based transition of care programs have been shown to be less likely to use acute care and experience multiple readmissions and more likely to receive timely post-acute care and received higher quality discharge information [45,54].
Female gender was associated with lower 30-day readmission rates for most targeted conditions, which may be due to greater utilization of preventative healthcare and healthcare in general. Females use significantly more preventative care including blood pressure checks, influenza immunizations, and cholesterol checks [53,55]. Women are thought to establish better routine care and be more active in their own healthcare since they are exposed to routine healthcare at a younger age than men, receive prenatal care, and are screened more often than men [56]. Women are also more likely to be responsible for their children's healthcare [55]. Despite receiving better care overall, the quality of care received by women in acute settings may be sub-optimal as compared to men [57]. Therefore, even though women have lower 30-day readmission rates, this is not the result of quality care in the acute setting but rather the result of better healthcare practices in the outpatient setting. Hospitals that underperform in providing care to women would therefore not be penalized under the HRRP. Similarly, hospitals that provide high-quality acute care to men could potentially be penalized if those male patients do not pursue post-acute care. Since post-acute care is heavily linked to income and social support, socially vulnerable patients are more likely to forgo post-acute care and are more likely to be readmitted.
Patients admitted to rural hospitals were less likely to be readmitted within 30 days of discharge from an acute care facility. There are conflicting findings on the relationship between hospital location and readmission rates. The targeted conditions studied here typically require follow-up with a specialist, and these specialists are often located only in urban centers [58]. Further, these early physician follow-ups are associated with lower 30-day readmission rates [59], so urban residents are likely to have better access to post-discharge specialized care and would therefore be less likely to be readmitted. Urban hospitals also tend to be larger and have additional infrastructure and established clinical protocols, suggesting that the quality of acute care may be better in these hospitals [60]. Conversely, it has also been reported that rural patients requiring additional resources or needing surgery are more likely to be treated at urban hospitals [61]. Therefore, clinical variables that measure the need for specialized care can serve as predictors of patient crossover from rural to urban hospitals. Urban hospitals may be associated with a higher odds of 30-day readmission because these hospitals see complex patients from both rural and urban areas who require more resources.
The findings of our study should be interpreted in the context of important limitations. First, we relied on CMS algorithms using ICD-9-CM codes to classify hospitalizations for our targeted conditions. The selections for the ICD-9-CM codes-based algorithms may have led to an underestimation of the number of hospitalizations for the targeted conditions. Since this methodology is used by the CMS to identify hospital admissions, we felt it was prudent to apply it to the entire study to provide national readmission estimates across age groups. Due to a limitation in the NRD, we excluded patients who were residents of different states. Persons are identified and tracked in the NRD with state-specific linkage numbers; therefore, a person readmitted between two different states cannot be tracked between states. This study was also limited by the inability to fully characterize social-related factors that may affect readmission such as access to transportation, social support, race and ethnicity, and health literacy. Extensive information on social risk factors are typically not collected as discrete data within inpatient databases, therefore limiting our analysis to available NRD data elements. Further research is needed to evaluate the confluence of social barriers and its impact on hospital readmission.

Conclusion
In conclusion, our population study of HRRP-targeted conditions indicates differences in readmission rates based on certain social and economic conditions. Exploring policy changes, innovative care delivery models, and tools that address social and economic needs at hospital discharge should be explored and tested to mitigate against hospital readmissions. Ultimately, a complimentary approach between policy and implementing accessible, evidence based SDoH programs will be necessary to improve patient care and minimize health inequity.