Hospital Length of Stay and All-Cause 30-day Readmissions among High-Risk Medicaid Beneficiaries

Abstract

This study examined the association between index hospitalization characteristics and the risk of all-cause 30-day readmission among high-risk Medicaid beneficiaries using multi-level analyses. A retrospective cohort with a baseline and a follow-up period was used. The study population consisted of Medicaid beneficiaries (21-64 years) with selected chronic conditions, continuous fee-for-service enrollment through the observation period, and at least one inpatient encounter during the follow-up period (N=15,806). The outcome of 30-day readmission was measured using inpatient admissions within 30-days from the discharge date of the first observed hospitalization. Key independent variables included length of stay, reason for admission, and month of index hospitalization (seasonality). Multi-level logistic regression that accounted for beneficiaries nested within counties was used to examine this association, after controlling for patient-level and county-level characteristics. In this study population, 16.7% had all-cause 30-day readmissions. Adults with greater lengths of stay during the index hospitalization were more likely to have 30-day readmissions [AOR=1.03, 95% CI 1.02,1.04]. Adults who were hospitalized for cardiovascular conditions [AOR=1.20, 95% CI 1.08,1.33], diabetes [AOR=1.23, 95% CI 1.10,1.39], cancer [AOR=1.55, 95% CI 1.26,1.90], and mental health conditions [AOR=2.17, 95% CI 1.98,2.38] were more likely to have 30-day readmissions compared to those without these conditions.

INTRODUCTION

Hospital readmissions that occur within 30 days of discharge are an important measure for assessing performance of the healthcare system and the quality of patient care.^1,2 According to the Healthcare Cost and Utilization Project (HCUP) report, there were approximately 3.3 million adults with all-cause 30-day readmissions in the United States (U.S.) in 2011, incurring nearly $41.3 billion in hospital costs.³ Reducing 30-day readmissions has become a priority for payers, providers, and policymakers seeking to achieve improved quality of healthcare at lower cost.

The implementation of the Affordable Care Act (ACA) provided the Centers for Medicare & Medicaid Services (CMS) statutory authority under the “Hospital Readmissions Reduction Program” to reduce payments for certain hospital readmissions that it deemed avoidable.⁴ Although initial focus was on Medicare and readmissions related to heart failure, myocardial infarction, and pneumonia, CMS is now considering expanding the list beyond the three conditions covered by the program.^4,5 Therefore, it is important to understand major risk factors for readmissions in beneficiaries with chronic conditions.

Medicaid consists of the largest number of beneficiaries among all payers in the U.S., with approximately 62 million beneficiaries in 2013.⁵ The Medicaid population is further expected to increase with the coverage expansions under the ACA. In addition, the state Medicaid programs incur an estimated $374 billion in healthcare expenditures and provide healthcare services to the vulnerable, indigent, and disabled. It is estimated that 61% of adult Medicaid beneficiaries have chronic or disabling conditions which place them at an increased risk of hospitalization.⁶ A series of HCUP statistical briefs reported several findings. First, the Medicaid all-cause readmission rates were comparable with Medicare but double the rate of private insurance.⁷ Second, for readmissions following non-surgical hospitalizations, 30-day Medicaid readmission rates were higher than Medicare and private insurance for both acute and chronic conditions.¹ The effects of these costly utilization patterns, such as hospital readmissions, for this large and growing population necessitates heightened importance under healthcare reform.

The balance between hospital efficiency and quality of care is another crucial aspect for our healthcare system. However, length of stay (LOS), a marker of efficiency, may conflict with hospital readmission rates, an indicator of quality. Further, CMS plans to bundle 30-day readmission rates to reimbursement for the index hospitalization.⁸

The effect of LOS on readmission rates is complex and previous studies have provided conflicting data regarding the relationship between LOS and subsequent readmission risk. Some indicate that shorter LOS is associated with a higher risk of readmission,^8,9 while others suggest that extended LOS is associated with a higher risk of readmission.^10–12 However, most research on readmissions has focused on the Medicare beneficiaries.^11,13,14 The readmission patterns of the Medicaid population differ from those of the geriatric Medicare population, from both clinical and socio-economic perspective. Considering the importance of 30-day readmission for payers and policy makers, there is a need to understand the role of LOS and to develop treatment and management strategies.

Our study examined the association between index hospitalization characteristics (LOS and reason for admission) and all-cause 30-day readmission risk in fee-for-service high-risk Medicaid beneficiaries (i.e. those with selected chronic conditions). The study is limited to patients who are high-risk and examines the differentiating factors within this high-risk population. For the purpose our study, variables were selected based on a priori knowledge and Andersen's Behavioral Model of health service utilization framework. This model suggests that potential health service use is determined by interactions among predisposing (demographics, index hospitalization characteristics), enabling (county-level, e.g., socio-economic status), and need (health status) characteristics of individuals and also the healthcare systems in the communities where they reside.¹⁵

METHODS

Study design

Retrospective cohort approach was used with baseline and follow-up periods. Baseline period was defined as the admission date of index hospitalization (first observed hospitalization) between January 1, 2007 and December 31, 2007. Patients were followed for 180 days after discharge date of the associated index hospitalization.

Data source

Medicaid administrative claims files from California, Illinois, New York, and Texas, between 2006 and 2008, were used. Personal Summary file included information on demographics, Medicaid enrollment, and eligibility status. Outpatient and Inpatient files included claims for services provided in ambulatory and inpatient settings and contained International Classification of Diseases 9^th edition Clinical Modification (ICD-9-CM) codes. Information on county-level characteristics was obtained from 2009 Area Health Resource File (AHRF), which was linked to Medicaid administrative claims files using state and county codes in which the beneficiary resided.

Study Population

Study population consisted of non-elderly (21-64 years) fee-for-service Medicaid-only beneficiaries with selected chronic conditions and continuous enrollment during baseline and follow-up period (Figure 1). Analyses were restricted to those who had at least one inpatient admission in 2007 and were conducted at person-level.

Figure 1.

Schematic Presentation of Selection Criteria

*Selected chronic conditions: asthma, arthritis, cardiac arrhythmias, coronary artery disease, cancer, congestive heart failure, chronic kidney disease, chronic obstructive pulmonary disease, dementia, diabetes, hypertension, hyperlipidemia, hepatitis, acquired immunodeficiency syndrome, osteoporosis, stroke, depression, schizophrenia, and substance use disorders.

For the purpose of this study, Medicaid beneficiaries with 19 chronic conditions were selected—asthma, arthritis, cardiac arrhythmias, coronary artery disease, cancer, congestive heart failure, chronic kidney disease, chronic obstructive pulmonary disease, dementia, diabetes, hypertension, hyperlipidemia, hepatitis, acquired immunodeficiency syndrome, osteoporosis, stroke, depression, schizophrenia, and substance use disorders. These conditions were identified based on the strategic framework developed and adopted by the Department of Health and Human Services for research, policy, program, and practice.¹⁶

Dependent Variable

Individuals were categorized into two groups, those with and without all-cause 30-day readmission. All-cause 30-day readmission was identified as subsequent hospitalization within 30 days of discharge date of the index hospitalization.

Key Independent Variables

These included index hospitalization characteristics, where LOS was the primary independent variable, reasons for admission was the secondary independent variable, and month of index hospitalization (included to control for potential seasonal effect).

Other independent variables

Patient-level characteristics included demographics (age, gender, and race/ethnicity) and Medicaid eligibility status (cash and medical need). Primary care access included continuity of care, measured using previously published continuity index (Modified, Modified continuity index) and coordination of care, measured as primary care visit within 14 days of discharge date. Healthcare utilization was measured as emergency room visit within six months prior to index hospitalization.

Variables accounting for county socio-economic conditions included educational attainment, per-capita income, employment rate, poverty level, and metropolitan statistical area. Variables related to availability of providers and healthcare facilities were AHRF designations for primary/mental healthcare shortage areas, presence of federally qualified health centers, rural health centers, and community mental health centers. Hospital and primary care provider density was defined as total number of hospitals or primary care providers per 100,000 individuals, respectively.

Statistical Techniques

Chi-square tests of independence were used for categorical variables and t-tests for continuous variables to determine group differences in patient-level and county-level characteristics and all-cause 30-day readmission. Multi-level logistic regression models, which accounted for beneficiaries nested within counties, were used to examine association between all-cause 30-day readmission and index hospitalization characteristics. The reference group for the dependent variable was “no 30-day readmission”. Model 1 controlled for only patient-level characteristics. Model 2 controlled for both patient-level and county-level characteristics. In these two models, county was specified as a random intercept using GLIMMIX procedure. All analyses were conducted using Statistical Analysis Software version 9.3 (SAS Inc., Cary, North Carolina USA).

RESULTS

After the exclusion criteria, there were 15,806 Medicaid beneficiaries with selected chronic conditions and at least one inpatient encounter in 2007. Overall, 16.7% experienced all-cause 30-day readmissions. A description of the study population and unadjusted associations between independent variables and all-cause 30-day readmission are presented in Table 1.

Table 1.

Description of Study Population by all-cause 30-day readmissions Medicaid Fee-for-Service Beneficiaries with Selected Chronic Conditions Multi-state Medicaid 2006–2008

Variables	30-day readmission	No 30-day readmission
	N (%)	N (%)	Sig.
	2633 (16.7%)	13,173 (83.3%)
Demographic and Medicaid Eligibility Characteristics
Gender			***
Female	1715 (65.1%)	9274 (70.4%)
Male	918 (34.9%)	3899 (29.6%)
Age group			***
21–24 years	301 (11.4%)	1675 (12.7%)
25–34 years	567 (21.5%)	3578 (27.2%)
35–44 years	517 (19.6%)	2498 (19.0%)
45–54 years	673 (25.6%)	2971 (22.6%)
55–64 years	575 (21.8%)	2451 (18.6%)
Race/Ethnicity			***
Caucasian	847 (32.2%)	3831 (29.1%)
African American	988 (37.5%)	4270(32.4%)
Hispanic	608 (23.1%)	4245 (32.2%)
Asian/AI/PI	39 (1.5%)	169 (1.3%)
Other	151 (5.7%)	658 (5.0%)
Cash Eligibility	1529 (58.1%)	6666 (50.6%)	***
Medical need Eligibility	876 (33.3%)	3769 (28.6%)	***
Index Hospitalization Characteristics
Length of stay, Mean (SD)	6.62 (9.09)	4.29 (6.35)	***
Chronic conditions at admission
Arthritis/osteoporosis	99 (3.8%)	464 (3.5%)
Cancer	134 (5.1%)	429 (3.3%)	***
Cardiovascular conditions	995 (37.8%)	3733 (28.3%)	***
COPD/Asthma	541 (20.5%)	2197 (16.7%)	***
Diabetes	575 (21.8%)	2103 (16.0%)	***
HIV/Hepatitis	305 (11.6%)	1185 (9.0%)	***
Mental health conditions	1491 (56.6%)	4352 (33.0%)	***
Season of readmission			***
Spring	730 (27.7%)	3944 (29.9%)
Summer	401 (15.2%)	2332 (17.7%)
Fall	211 (8.0%)	1605 (12.2%)
Winter	1291 (49.0%)	5292 (40.2%)
Primary Care Access
Coordination of primary care	326 (12.4%)	1747 (13.3%)
Continuity of primary care
Complete care continuity	349 (13.3%)	1764 (13.4%)
Some care continuity	634 (24.1%)	2960 (22.5%)
No care continuity	1650 (62.7%)	8449 (64.1%)
Healthcare Utilization
Emergency room (ER) visit	893 (33.9%)	4449 (33.8%)
County-level Characteristics
Metropolitan Status
Non-metro	267 (10.1%)	1285 (9.8%)
Metro	2366 (89.9%)	11888 (90.2%)
Primary Care Shortage Area			*
Whole county	2034 (77.3%)	10147 (77.0%)
Part county	429 (16.3%)	2312 (17.6%)
No shortage	170 (6.5%)	714 (5.4%)
Mental Healthcare Shortage Area			**
Whole county	2015 (76.5%)	9925 (75.3%)
Part county	388 (14.7%)	2242 (17.0%)
No shortage	230 (8.7%)	1006 (7.6%)
	Mean (SD)	Mean (SD)
CMHC	0.81 (1.23)	0.94 (1.24)	***
Rural Health Center	0.62 (3.03)	1.06 (4.41)	***
FQHC	37.69 (44.31)	37.78 (42.98)
Education rate - 4+ yrs	25.39 (10.98)	23.77 (10.51)	***
Unemployment rate	4.57 (0.71)	4.67 (0.90)	***
% Below poverty level	15.11 (3.73)	15.06 (3.80)
Per-capita income (US Dollars)	58,761.96 (33,697.42)	54,029.16 (31,265.86)	***
Non-federal PCP density	307.10 (192.29)	279.97 (179.22)	***
Hospital density	1.74 (1.37)	1.65 (1.14)	***

Note: Based on 15,806 non-elderly (21-64 years) fee-for-service Medicaid beneficiaries residing in California, Illinois, New York and Texas with selected chronic conditions, who were alive and had continuous fee-for-service enrollment through the observation period, were not enrolled in Medicare, and had at least one inpatient encounter in the follow-up period. Significant group differences in all-cause 30-day readmissions were tested with chi-square and t-tests. Asterisks represent significant group differences between the “30-day readmission” and “no 30-day readmission” groups.

^***P < 0.001;

^**0.001 ≤ P < 0.01;

^*0.01 ≤ p < 0.05

Column percentages are reported for categorical variables.

Abbreviations - AI/PI: American Indian or Pacific Islander; CMHC: Community Mental Health Center; COPD: Chronic obstructive pulmonary disease; FQHC: Federally Qualified Health Center; HIV: Human immunodeficiency virus; PCP: Primary care practitioners; SD: Standard deviation; US: United States

Multi-level logistic regressions of all-cause 30-day readmissions are summarized in Table 2. Beneficiaries with longer LOS had significantly higher odds of 30-day readmission. In addition, presence of cancer, cardiovascular conditions, diabetes, and mental health conditions at index hospitalization significantly increased the odds of readmission. In addition, beneficiaries with cash or medical need eligibility had significantly higher odds of 30-day readmission.

Table 2.

Adjusted Odds Ratios (AOR) and 95% Confidence Intervals (CI) from Multi-level Logistic Regressions of all-cause 30-day readmissions Medicaid Fee-for-Service Beneficiaries with Selected Chronic Conditions Multi-state Medicaid, 2006–2008

	AOR	95% CI	Sig.
Length of stay	1.03	[1.03,1.04]	***
Chronic conditions at admission
Arthritis/osteoporosis	0.90	[0.72,1.13]
Cancer	1.55	[1.26,1.90]	***
Cardiovascular conditions	1.20	[1.08,1.33]	***
COPD/Asthma	1.01	[0.90,1.12]
Diabetes	1.23	[1.10,1.39]	***
HIV/Hepatitis	0.98	[0.85,1.12]
Mental health conditions	2.17	[1.98,2.38]	***
Season of readmission
Spring	0.79	[0.71,0.88]	***
Summer	0.77	[0.68,0.88]	***
Fall	0.58	[0.49,0.68]	***
Winter	Ref.
Cash eligibility	1.14	[1.01,1.27]	*
Medical need eligibility	1.21	[1.08,1.36]	**

Note: Based on 15,806 non-elderly (21–64 years) fee-for-service Medicaid beneficiaries residing in California, Illinois, New York and Texas with selected chronic conditions, who were alive and had continuous fee-for-service enrollment through the observation period, were not enrolled in Medicare, and had at least one inpatient encounter in the follow-up period.

Model controlled for both state effect, patient-level and county-level variables. Patient-level variables were demographic (gender, age, race/ ethnicity), Medicaid eligibility characteristics (cash and medical need), primary care access (continuity and coordination of care), and healthcare utilization (emergency room visits). County-level variables were metropolitan statistical area, primary care shortage areas, mental healthcare shortage areas, community mental health centers, rural health centers, federally qualified health centers, college education rate, unemployment rate, poverty level, per-capita income, density of primary care providers, and hospital density.

Asterisks represent significant group differences in 30-day readmission compared to the reference group. The logistic regressions also included intercept terms. The regressions accounted for clustering of individuals within counties.

^***P < 0.0001;

^**0.001 ≤ P < 0.01;

^*0.01 ≤ p < 0.05

Abbreviations: COPD: Chronic pulmonary obstructive disease; HIV: Human immunodeficiency virus

DISCUSSION

To the best of our knowledge, this is the first study examining patient-level and county-level characteristics associated with all-cause 30-day readmission in Medicaid beneficiaries with chronic conditions. In addition, our findings add to the nascent literature on readmissions among Medicaid beneficiaries, with valuable findings discussed as follows.

Length of stay has been reported as a risk factor for readmission both in elderly and non-elderly populations.¹¹ Our findings indicate that longer LOS is associated with increased odds of 30-day readmission, which could be attributed to severity of illness at index hospitalization.¹⁰ This finding could be related to unmeasured clinical severity (our models account for some comorbidities) and socioeconomic issues (as noted in the introduction), although we have tried to control for these as best we can in our model. This may have implications for discharge planning efforts, focusing on chronic disease management, which has previously shown to be effective in reducing readmissions.¹⁷ Our findings suggest 30-day readmissions can be predicted using variables that are readily available, few in number and simple to incorporate in discharge planning. An effective discharge planning taking into account chronic conditions and index hospitalization characteristics (LOS) may help organize post-discharge services, including coordination of care with physicians, medication reconciliation, follow-up care, and appropriate self-management for chronic conditions.

Our findings also indicate increased risk of 30-day hospital readmissions as well as longer LOS among Medicaid beneficiaries with cancer, cardiovascular conditions, diabetes, and mental health conditions at index hospitalization. This suggests that patient complexity/poor health status increases the risk of readmission. A more focused approach in treatment of these diseases can help reduce readmissions. Integrated care management interventions post-hospital discharge have been shown to reduce readmissions among those with heart disease; a coordinated care team including cardiologists, specialized nurses, and primary care physicians, and provision of integrated care following hospitalizations can reduce readmissions.^18,19 Emerging models of delivery such as accountable care organizations and patient-centered medical homes, which offer comprehensive, well-coordinated primary care services, may be needed to reduce readmission among Medicaid beneficiaries with chronic health conditions. In this respect, three of the four states represented (California, New York, and Texas), are CMS Innovation Model partner states and are presently awardees of Medicaid Incentives for the Prevention of Chronic Disease state grants.²⁰ It remains to be seen whether such programs can reduce the high prevalence of readmissions in Medicaid population.

Although our findings may have implications in reducing readmission risk, these results need to be interpreted in the light of study limitations. Our study was based on beneficiaries from only four states and cannot be generalized to the entire Medicaid population in the U.S. We also excluded individuals who were not enrolled in Medicaid health maintenance organizations. Given that less than one-third of the population receives fee-for-service care in Medicaid, our study may have selection bias. Our study design utilized retrospective cohort approach and cannot be used to establish causal relationships. Further, our study did not include adjustment for variables related to discharge planning and care coordination, which might influence the readmission risk of patients with complexities. Our study utilized data from 2006–2008 Medicaid administrative claims file.

Overall, our analyses revealed that patient-complexities increased the risk of all-cause 30-day readmission for high-risk Medicaid beneficiaries with chronic conditions, thus warranting the need for comprehensive care for those with chronic conditions. Programs designed to reduce the risk of 30-day readmissions may need to focus on appropriate disease management and post-hospitalization coordinated care.