Vol. 00 No. 0 Month 2015

1

Comparative Effectiveness of Risk-Stratified Care Management in Reducing Readmissions in Medicaid Adults With Chronic Disease Sharon Hewner, Yow-Wu Bill Wu, Jessica Castner

Purpose Management of care transitions to avoid readmissions has become a priority for many acute care facilities because readmission rates are increasingly used as a measure of quality (Conway and Berwick, 2011; Kangovi and Grande, 2011). Efforts to prevent readmissions have focused on heart failure (HF), acute myocardial infarct, and pneumonia (Dharmarajan et al., 2013; Krumholz et al., 2013). However, recent studies suggest that avoiding readmissions is more complex than avoiding an exacerbation of a chronic disease and that care in other settings, such as primary care or emergency department (ED), has an impact on inpatient (IP) readmissions (Boutwell et al., 2011; Epstein et al., 2011). Furthermore, those with limited access to health services may be particularly vulnerable to readmission because they lack the social and economic resources needed to effectively self-manage their disease (Kangovi and Grande, 2011). Thus, adult Medicaid recipients with preexisting chronic illnesses may have a higher than expected risk of rehospitalization related to limited socioeconomic resources and access to care (Billings and Mijanovich, 2007; Trudnak et al., 2014). The purpose of this article was to examine the impact of chronic disease complexity and intensity of populationbased care management on readmissions in Medicaid adults with preexisting chronic disease. Looking at the 90 days after a hospitalization, we hypothesized that individuals enrolled in a managed

Abstract: Hospitalized adult Medicaid recipients with chronic disease are at risk for rehospitalization within 90 days of discharge, but most research has focused on the Medicare population. The purpose of this study is to examine the impact of population-based care management intensity on inpatient readmissions in Medicaid adults with pre-existing chronic disease. Retrospective analyses of 2,868 index hospital admissions from 2012 New York State Medicaid Data Warehouse claims compared 90-day post-discharge utilization in populations with and without transitional care management interventions. High intensity managed care organization interventions were associated with higher outpatient and lower emergency department post-discharge utilization than low intensity fee-for-service management. However, readmission rates were higher for the managed care cases. Shorter time to readmission was associated with managed care, diagnoses that include heart and kidney failure, shorter length of stay for index hospitalization, and male sex; with no relationship to age. This unexpected result flags the need to re-evaluate readmission as a quality indicator in the complex Medicaid population. Quality improvement efforts should focus on care continuity during transitions and consider population-specific factors that influence readmission. Optimum post-discharge utilization in the Medicaid population requires a balance between outpatient, emergency and inpatient services to improve access and continuity.

care organization (MCO) with intensive risk-adjusted care management programs would have fewer readmissions (IP) and ED treat and release visits, and would have higher rates of outpatient (OP) utilization. Based on previous analysis of 2009 utilization in Medicaid adults with chronic illness, we expected that individuals enrolled in MCO would have a lower likelihood of readmission than those in Medicaid Fee-for-service (FFS) as would

Keywords care/disease management Medicaid research – evaluation managed care disparities/equity of care Journal for Healthcare Quality Vol. 00, No. 0, pp. 1–14 © 2015 National Association for Healthcare Quality

2

Journal for Healthcare Quality

less complex cases than those with HF or chronic kidney failure (Hewner et al., 2014b).

Background Risk-stratified care management strategies used by the high-intensity MCO were initiated by telephonic outreach to patients within 72 hours of discharge to integrate services between the hospital and primary care office (Hewner, 2014; Hewner et al., 2014a). Outreach phone calls focused on assessment of unmet educational and health needs, medication reconciliation, arranged follow-up primary care, and offered up to four home visits by nurses to provide transitional support. Outreach phone calls were supported by a claimsbased virtual electronic health record at the MCO (Hewner and Seo, 2014) and population-specific care management teams developed a continuum of riskstratified interventions for Medicaid, Medicare, and privately insured populations (Hewner et al., 2014a). Finally, very unstable cases were referred to advanced practice nurses providing evaluation and monitoring in the home. The high-intensity MCO also provided disease management and care coordination specialists to assist primary care practices manage cases with high risk of readmission. The expectation is that highintensity interventions, such as those provided at the MCO since 2009, will result in fewer readmissions. The ability to proactively identify complex cases based on their health risk is critical to risk-stratified care management (Centers for Medicare and Medicaid, 2012). This is challenging because patient complexity is a “dynamic state in which the personal, social, and clinical aspects of the patient’s experience operate as complicating factors” (Shippee et al., 2012, p. 1041). However, medical complexity is often operationally defined as multiple interacting medical conditions as is found in patients with chronic diseases (Peek et al., 2009). A clinical algorithm that uses medical diagnoses and patterns of comorbidity during the past year would

facilitate proactive identification of the cohort with chronic disease. This approach differs from multiple clinical algorithms that focus on the most complex hospitalized patients to predict mortality (Charlson et al., 1987; Naessens et al., 1992) and hospital length of stay (LOS) (Young et al., 1994; Zhong et al., 2012). Use of claims data would support identification of the cohort without chronic disease or those with only hypertension or lipid disorders who are at risk of developing major chronic diseases. This type of approach stratifies the population and creates cohorts standardized for medical complexity, and this facilitates the development of interventions that target the needs of the cohort and the analysis of health outcomes. The rate of avoidable readmissions is generally considered a measure of the quality of care provided in an acute hospital (Kangovi and Grande, 2011; Press et al., 2013). Ineffective care transitions result in adverse events after discharge including avoidable readmissions within 30 days related to poor communication between providers, failures in medication reconciliation, unmet educational needs of patients and family, and healthcare system factors (Rennke et al., 2013). However, 30-day readmission rates are often not correlated with other markers of quality (Krumholz et al., 2013; Press et al., 2013). Comparison of privately insured MCO and FFS surgical hospitalizations demonstrated that unobserved hospital quality was associated with surgical complications but that this association disappeared when patient health status was controlled for (Maeng and Martsolf, 2011). In addition, the results of care transitions management often have a limited impact on rehospitalizations (Boutwell et al., 2011; Rennke et al., 2013). Factors that are commonly used to predict hospitalization such as disease complexity or comorbidity, utilization, age, and sex have limited predictive ability for readmissions (Kansagara et al., 2011). In their analysis of national Medicare claims from 2008, Epstein and colleagues (2011) found that regional hospitalization rates were the

Vol. 00 No. 0 Month 2015

strongest predictor of all-cause readmissions, accounting for 32–48% of the variance. Clearly, readmissions have complex causation and efforts to reduce readmissions need to consider safety and communication across settings to improve continuity of care. Since eligibility for Medicaid is based on income, we commonly use insurance status as a proxy for more in-depth information about poverty and social determinants of health. Readmissions in the non-obstetric adult population, aged 21– 64 years, showed that the 30-day readmission rate was higher for Medicaid (10.7%) than privately insured individuals (6.3%) (Jiang and Wier, 2010). In 2011, the top non-obstetric causes of readmission for Medicaid adults (18–64 years) were mood disorders, schizophrenia, diabetes, alcohol-related disorders, HF, septicemia, chronic lung disease (chronic obstructive pulmonary disease [COPD]), and substance-related disorders; of these diagnoses, only HF, septicemia, and COPD were also causes for Medicare (Hines et al., 2014). Review of the literature regarding Medicaid readmissions highlights the importance of mental illness and substance abuse as well as suggests that care coordination is an especially effective intervention in the Medicaid population (Bielaszka-DuVernay, 2011; Regenstein and Andres, 2014). Readmissions in the Medicaid population may be based on different factors than those in the Medicare population.

Study Design and Methodology This secondary analysis used existing 2012 electronic claims data from the New York State Medicaid Data Warehouse (MDW) to compare 90-day postdischarge utilization in the cohort with chronic disease in 2 counties served by the high-intensity MCO. Deidentified demographic and claims tables were extracted from the MDW. The demographic table included a research identification number, age, sex, months of enrollment, and insurer. This was joined to a claims table that included the research identification numbers, the

date of service (admission and discharge dates), type of service, and the first 5 International Classification of Diseases, 9th Revision (ICD-9) codes on the claim for IP, ED, and OP encounters. The University at Buffalo Social Behavioral Internal Review Board determined that the project did not meet the definition of human subject research since there is neither intervention nor interaction between the research team and the subjects to obtain the data and the data are not identifiable private information. The COMPLEXedex clinical algorithm was used to classify individuals into hierarchical disease categories, complexity segments, and chronic disease cohorts based on ICD-9 codes. Individuals were flagged for the presence of nine prevalent chronic diseases, and these were then ranked for comorbidity and complexity (Hewner and Seo, 2014). The COMPLEXedex is used to stratify the population into those with and without chronic disease and is not a risk-adjustment methodology. Data were cleaned to remove duplicate claims for encounters of the same type and date of service. Concurrent with classification into the hierarchy, total counts of IP, ED, and OP utilization were created. The summary table included the individual identifier, a flag for each of the nine chronic conditions, the count of IP, ED, and OP visits for the year, and the classification along with demographic information. The next step was to identify cases with one or more admissions. The first admission in 2012 was called the index admission. The analysis examines utilization in the 90 days after the index admission date of discharge, so that each individual with an admission was counted once. Only admissions between January 1, 2012, and September 30, 2012, were included to allow for the full 90-day postdischarge follow-up period. Figure 1 shows the data cleaning process that started with 357,228 cases and ended up with 2,868 index admissions and 556 readmissions. Cases with fewer than 10 months of enrollment, children younger than 18 and adults older than 65 years, individuals enrolled in other

3

4

Journal for Healthcare Quality

Figure 1. Exclusion criteria to identify index admissions and r.admissions within 90-days.

MCOs, those without chronic conditions, and cases with LOS above the 99th percentile of 44 days were eliminated. Finally, index admissions were linked to the hospital identifiers in all but 198 cases. Of these 2,670 cases, 94% of the index admissions occurred in 10 hospitals, and all but three were grouped into 2 health systems resulting in 2,020 cases for the survival analysis. The index admission table included demographic information, complexity classification, admission and discharge dates, LOS for index admission, and for encounters in the 90 days after discharge, type of encounter, and interval from discharge date. These data were used to calculate weekly rates of IP, ED, and OP encounters. Bivariate analysis included Kaplan–Meier plots of estimated readmission probabilities to visually inspect

each categorical variable (sex, intervention intensity, hospital system, system failure) and days to readmission for proportional hazards assumption. In addition, comparison of LOS by network was completed using t test. Logistic regression and chi-square tests were used to analyze differences in readmission based on intensity of intervention (MCO or FFS). Cox regression survival analysis was used to evaluate the likelihood of readmission with intervention intensity, complexity segments, health system, LOS, age, and sex as covariates in the equation.

Results Prevalence of chronic diseases was somewhat higher for the population with multiple ($2) admissions in 2012. Table 1 lists the chronic conditions included in the

Vol. 00 No. 0 Month 2015

Table 1. Prevalence of Chronic Conditions in the COMPLEXedex Algorithm in Adult Medicaid Patients in Chronic Cohort With Multiple Hospitalizations in 2012

Population size Mental health diagnosis (includes depression)* Substance abuse diagnosis (excluding smoking)* Hypertension Diabetes Lipid disorder Asthma Smoker Chronic obstructive pulmonary disease Obesity Chronic kidney disease Coronary artery disease Heart failure *

FFS, Chronic

MCO, Chronic

FFS, 21 Admits

MCO, 21 Admits

11,213 47%

9,527 52%

1,248 51%

822 56%

17%

18%

43%

43%

29% 26% 18% 18% 13% 12% 8% 7% 6% 4%

32% 23% 23% 27% 19% 10% 12% 3% 5% 3%

29% 21% 12% 13% 17% 16% 6% 13% 8% 11%

35% 28% 23% 20% 21% 17% 10% 8% 12% 10%

ICD-9 codes for behavioral health diagnoses based on crosswalk of ICD-9 with DSM-III classification.

COMPLEXedex classification with the percentage of the population (N = 20,740 individuals with chronic disease) with claims for that condition for the chronic cohort for both FFS and MCO and for the subgroup with two or more admissions in 2012. Chronic conditions are listed in approximate order of their prevalence. Mental health (including depression) and substance abuse (excluding smoking) have the greatest prevalence in those with multiple admissions, whereas HF, coronary artery disease, and chronic kidney disease have the lowest prevalence. Although the prevalence of mental health is comparable with that in the cohort, the rate of substance abuse in the subset with two or more hospitalizations is more than twice than that of the chronic cohort overall (Castner et al., 2015). Table 2 compares population characteristics in FFS and MCO groups with an index admission (N = 2,868 index admissions). The MCO group included more women (69% compared with 59% in FFS), fewer cases with heart or chronic kidney failure (16% compared with 29% in FFS), and were younger on average (42 years)

than the FFS group (48 years). The MCO had higher percentage of the population with a readmission in the first 30 days after discharge (13% compared with 7% in FFS) and 31–60 days (6% compared with 2%). However in 61–90 days, they had a lower rate of readmission (4% compared with 7% in FFS). The MCO group had a shorter LOS for the index admission for the segment with chronic conditions (4.93 days compared with 6.02 days). Length of stay was comparable for those with system failure. The rate of readmissions with the same diagnosis in FFS was 27% compared with 31% in the MCO. Descriptive analysis of the observed (unadjusted) rate per 1,000 index admissions (N = 2,868 index admissions) for OP and ED utilization demonstrates higher OP utilization and lower ED utilization in the MCO with intensive risk-stratified care management (Fig. 2). Outpatient utilization in the MCO is higher throughout the 90 days after discharge, with 83% of cases having an OP visit in the first 30 days after discharge compared with 55% in FFS. Furthermore, FFS had higher ED utilization with one third having an ED visit in

5

6

Journal for Healthcare Quality

Table 2. Comparison of Characteristics of the Low- and HighIntensity Intervention Group With an Index Admission in 2012 (2,868 Index Admissions)

Sex Male Female Complexity segment† Chronic System failure Readmissions, days 30 60 90 Age (mean), years Length of stay, days Chronic mean System failure mean Same diagnosis readmission (576 readmissions) Readmissions Same diagnosis * †

Low-Intensity FFS* (n = 1,693)

High-Intensity MCO* (n = 1,175)

689 (40.7%) 1,004 (59.3%)

367 (31.2%) 808 (68.8%)

1,211 (71.5%) 482 (28.5%)

991 (84.3%) 184 (15.7%)

121 (7.1%) 49 (2.9%) 121 (7.1%) 48

154 (13.1%) 73 (6.2%) 49 (4.2%) 42

6.02 6.23

4.93 6.36

298 79 (27%)

278 85 (31%)

FFS is Medicaid Fee-for-service and MCO is Medicaid managed care organization. The COMPLEXedex divides the population with chronic disease into those with heart or chronic kidney failure (in addition to other comorbidities) and those with chronic conditions without system failure. It also creates hierarchical disease categories.

first 4 weeks postdischarge compared with 28% in the MCO. These results supported our hypotheses. However, IP utilization was higher in the MCO, and this does not support our hypothesis of fewer readmissions with high-intensity interventions. The IP readmission graph shows the weekly rates in the MCO and FFS groups. Thirteen percent of MCO cases had a readmission in the first 30 days postdischarge compared with 7% in FFS. Table 3 shows the results of the unadjusted odds ratio (chi-square) and logistic regression. In the logistic regression model, sex, relative risk of hospitalization for hierarchical disease category, index admission LOS, and age were covariates to adjust for unexplained variance. Differences between the 2 groups were

statistically significant with a reversal of the pattern between 61 and 90 days postdischarge when the MCO had lower readmission rates. In the subsample with hospital identifier information (10 hospitals, n = 2,516), there was variation in the rate of readmissions, which ranged from 3% to 12% of index admissions. Of the 10 hospitals, 4 were part of an Accountable Care Organization (ACO), 3 were part of a non-ACO hospital health system, and 3 were not affiliated or provided primarily obstetric care. In the 7 hospitals associated with 2 health systems (n = 2,020 index admissions), the 30-day readmission rates were 10% in non-ACO system (n = 1,287) and 7% in the ACO (n = 733). The odds ratio for readmission in the non-ACO was 1.5

Vol. 00 No. 0 Month 2015

Figure 2. 90-day post-discharge utilization rates per 1000 index admission by Intervention Intensity, 1/1/2012 - 9/30/2012 (N = 2,868 index admissions).

7

8

Journal for Healthcare Quality

Table 3. Likelihood of Readmission (Odds Ratio) Within 1–30, 31–60, and 61–90 Days of Discharge in 2012 (2,868 Index Admissions) Readmission (Days Postdischarge) 1–30 days Yes No 31–60 days Yes No 61–90 days Yes No

c2 (p)

Odds Ratio (unadjusted)

Exp(B) (Adjusted*)

Logistic Regression (p)

154 1,021

,.01

0.51

0.46

,.01

49 1,644

73 1,102

,.01

0.45

0.40

,.01

121 1,572

49 1,126

,.01

1.78

1.69

,.01

FFS, Observed

MCO, Count

121 1,572

*In the logistic regression model, sex, relative risk of hospitalization for disease categories by complexity, index admission length of stay, and age were covariates to adjust for unexplained variance. FFS, Fee-for-service low-intensity interventions; MCO, managed care organization high-intensity intervention.

times that of the ACO. Furthermore, FFS cases were 1.8 times more likely to be cared for in the non-ACO health system. The mean LOS was 6.19 days for the non-ACO system compared with 4.34 days for the ACO, and this difference was statistically significant (t test, p # .001). The logistic regression analysis was completed separately on the 2,020 index admissions to the ACO and nonACO, and the remaining 496 nonaffiliated index admissions. Although the hospital system was a significant predictor (p = .018), there were differences between the networked and nonaffiliated cases. Time to readmission in the 2,020 index admissions in networked hospitals was tested using unadjusted Kaplan–Meier plot of readmission probabilities and adjusted analysis using Cox regression. Unadjusted Kaplan– Meier plots of readmission probabilities for each categorical independent variable and time to readmission are displayed in the Appendix (Supplemental Digital Content http://links.lww.com/JHQ/A6). Table 4 provides Kaplan–Meier estimates of readmission probabilities for 374 readmission events with 1,646 censored. Cox regression was used to analyze the time to readmission. Overall, the model was statistically significant (p , .001) and high intervention intensity MCO (p , .001), system failure complexity segment (p , .001), non-ACO

hospital network (p , .001), and shorter LOS (p , .001) all predict a greater likelihood of readmission within 90 days of discharge. In contrast, neither age nor sex was a significant predictor of readmission. Table 5 shows the results of the variables in the equation with their hazard ratio (Exp (B)). Figure 3 compares likelihood of readmission after index admission for the high-intensity postdischarge intervention in the study MCO compared with FFS with minimal postdischarge care transitions management. Based on these results, we accept the hypothesis that intensive risk-standardized care management resulted in more OP utilization in the 90 days after discharge and fewer ED visits. Our hypotheses that high-intensity interventions result in fewer readmissions or a lower likelihood of readmission must be rejected. However, the hypothesis that those with more complex diseases are more likely to be readmitted was supported by the results. Other factors increasing the likelihood of readmission were non-ACO hospital network and shorter LOS.

Directions for Future Research A research implication is that the Medicaid population may differ in complex ways

Vol. 00 No. 0 Month 2015

9

Table 4. Kaplan–Meier Estimates of Days to Readmission (374 Readmission Events, 1,646 Censored)

Variable Sex Intervention Intensity Health System Complexity Segment

95% Confidence Interval

Value

Total, N

Censored, %

Mean Estimate

Lower

Upper

Female Male FFS MCO ACO Non-ACO Chronic Sys. Fail.

1,286 734 1,270 750 773 1,287 1,466 554 2,020

82.7 79.3 84.1 77.1 84.6 79.7 83.9 81.5 81.5

82.324 80.200 84.025 77.365 83.226 80.559 82.726 78.446 81.552

81.09 78.42 82.93 75.39 81.69 79.27 81.60 76.28 80.54

83.56 81.98 85.12 79.34 84.76 81.32 83.86 80.61 82.57

Overall

ACO, Accountable Care Organization; FFS, Fee-for-service low-intensity interventions; MCO, managed care organization high-intensity intervention; Non-ACO, health system not an accountable care organization; Sys. Fail., system failure complexity segment.

from the older Medicare population (Jencks et al., 2009) and the more affluent population of privately insured individuals (Maeng and Martsolf, 2011) in their disease complexity and in the importance of social determinants of health in postdischarge utilization. Although the rates of admission for our population are comparable to those reported in the literature for Medicaid (Bielaszka-DuVernay, 2011; Jiang et al., 2003; Jiang et al., 2005; Raven

et al., 2011; Regenstein and Andres, 2014), few studies have explored readmissions in the Medicaid population (Gill et al., 2003; Jiang and Wier, 2010). In an analysis of non-obstetric adult 30-day readmissions comparing Medicaid and private insurance populations (Jiang and Wier, 2010), readmission rates with comorbid mental disorders, substance abuse, diabetes, or chronic lung disease were consistently higher in Medicaid (11.9–12.6%) than in

Table 5. Cox Proportional Hazard Analysis for Readmission Survival Analysis Results (374 Readmission Events, 1,646 Censored) 95% Confidence Level Variable MCO SEG NET LOS Sex Age

B

SE

Wald

df

Sig.

Exp(B)

Lower

Upper

0.623 0.571 0.281 0.0.29 0.163 -0.004

0.109 0.114 0.115 0.007 0.106 0.005

32.871 25.217 19.238 13.186 2.340 0.810

1 1 1 1 1 1

,0.001 ,0.001 0.015 ,0.001 0.126 0.368

1.865 1.770 1.325 1.030 1.177 0.996

1.507 1.416 1.057 1.016 0.995 0.987

2.308 2.211 1.659 1.043 1.450 1.005

LOS, length of stay; MCO, managed care organization high-intensity intervention; NET, non-ACO hospital network; SEG, system failure complexity segment.

10

Journal for Healthcare Quality

Figure 3. Likelihood of readmission comparing high intensity (MCO) with low intensity interventions (FSS), 1/1/2012 - 9/30/2012 (N = 2,868 index admissions, 567 readmissions).

privately insured (7.3–8.1) (p.10). Future research should continue to address this disparity.

Limitations Our analysis did not differentiate between avoidable and unavoidable readmissions. We were able to identify a difference in readmission with the same or closely related diagnosis between MCO (31%) and FFS (27%); however, these readmissions may not have been avoidable and are unlikely to be all of the avoidable readmissions (which might include an infection or medication error). A second issue is that there was a higher than expected prevalence of substance abuse and mental health conditions. This suggests the need for further research examining the impact of a broader range of behavioral health

diagnoses on both admissions and readmissions and to differentiate psychiatric hospitalizations from general medical admissions (Billings and Mijanovich, 2007; Huff, 2000; Prince et al., 2009). Because this is an observational study, we were not able to control for characteristics and differences between the MCO and FFS groups with index admissions that may confound the results of the study. The initial data extraction did not include the hospital provider, and efforts to match index admissions with a hospital or hospital system resulted in a loss of 848 index admission. As the market changes and health systems incorporate risk-stratified care management strategies in efforts to reduce readmissions, it will be necessary to identify providers and insurers. Moving forward, comparison of insurance-based interventions may be contaminated by

Vol. 00 No. 0 Month 2015

conflicting interventions in the marketplace. Observational studies will need to identify ways to determine the dose of care management in a given environment.

Discussion There are a number of possible contributing factors that may affect health outcomes in the non-obstetric adult Medicaid population. Maeng and Martsolf (2011) explore possible causes for worse outcomes in MCO compared with FFS populations and conclude that certain chronic conditions are more prevalent in MCO. We also found differences in disease prevalence with higher prevalence of diabetes, COPD, chronic kidney disease in FFS and higher prevalence of mental health diagnoses, hypertension, lipid disorders, asthma, smoking, and obesity in MCO. Another possibility is that telephone outreach that can result in up to four nurse visits in the month after discharge, such as that provided by the study MCO, might result in earlier identification of complications of care resulting in readmission (Epstein et al., 2011). In this case, readmission might be a sign of improved case identification and improved quality of care. Furthermore, the MCO group had a shorter LOS which may be related to the availability of nursing follow-up in home and more efficient care. The FFS cases may have remained in the hospital longer to complete teaching and arrange discharge services. More FFS were in the non-ACO hospital with limited system-wide care management interventions. Intensive postdischarge interventions do result in additional OP follow-up and fewer ED visits, both indicative of improved access to care. This improved access before and after hospitalization may also lead to better case finding of behavioral health issues including depression which is higher in the study MCO than FFS. The second issue is the question of behavioral health comorbidity as a predictor of readmission. Many tools use the number of comorbid conditions as a proxy for disease complexity (Bowles et al., 2012, 2014; Holland and Harris, 2007), and the impact

of behavioral health comorbidity on hospitalization and readmission has been examined (Billings et al., 2006; Huff, 2000; Jiang and Wier, 2010; Prince et al., 2009). However, most algorithms predicting risk of hospitalization do not include behavioral health issues (Kansagara et al., 2011). The COMPLEXedex clinical algorithm is currently being adapted to include substance abuse and a broader range of behavioral health diagnoses to explore the impact of comorbidity on admission and especially readmission rates. Finally, social complexity is rarely included as part of the discussion of readmission; however, in the Medicaid population, social determinants of health may play a significant role in readmissions (Kangovi and Grande, 2011; Kansagara et al., 2011). Knowledge about social complexity may be outside what can be answered using existing electronic claims; however, it is possible to use health information exchange to facilitate outreach phone calls by a care coordinator to evaluate the potential impact of social determinants and to intervene proactively, preventing readmission. Improved care coordination for Medicaid based in primary care settings resulted in a reduction of readmission rates from 30 to 7% and shorter LOS in the hospital (BielaszkaDuVernay, 2011). Technology to support this type of intervention exists, and simple screening tools can be added to support decision making in the early postdischarge period. This type of intervention must be embedded in a continuum of services to proactively prevent hospitalization and reactively identify avoidable readmissions.

Implications for Practice We need to consider the relative merits of avoiding readmissions compared with avoiding preventable admissions. Current approaches focus on improving the discharge process for high-risk medical patients (Kangovi and Grande, 2011), ignoring the impact of access in both decreasing and paradoxically increasing readmissions. But often readmissions are unavoidable because disease progression

11

12

Journal for Healthcare Quality

and acute exacerbations of chronic illness necessitate acute care follow-up. Avoidable admissions are often caused by medical error and unsafe care in the hospital setting (Rennke et al., 2013), and vulnerable individuals with preexisting chronic disease are more likely to succumb to a readmission that could have been avoided by improved safety, stronger medication reconciliation, and better communication across settings (Burke and Coleman, 2013). Use of readmission as a quality measure and to determine fines for specific chronic disease encourages a disease-specific response from acute care providers. Instead of improving handoffs and communication (Conway and Berwick, 2011), the focus has been on improved teaching for specific conditions such as HF. Further examination of the complex reasons for readmission in the socially vulnerable Medicaid population may help us to understand how to prevent avoidable admissions in the broader population.

Acknowledgments The authors thank Navinder Mehrok, MS, and Harish Mangalampalli, graduate student programmers for data management, and Jin Young Seo, MS, Doctoral student for assistance in manuscript preparation. The authors are grateful to the New York State Department of Health for allowing access to the Medicaid Data Warehouse to complete this analysis.

References Bielaszka-DuVernay, C. Improving the coordination of care for Medicaid beneficiaries in Pennsylvania. Health Aff (Millwood) 2011;30: 426–430. Billings, J., & Mijanovich, T. Improving the management of care for high-cost Medicaid patients. Health Aff (Millwood) 2007;26: 1643–1654. Billings, J., Dixon, J., Mijanovich, T., & Wennberg, D. Case finding for patients at risk of readmission to hospital: development of algorithm to identify high risk patients. BMJ 2006;333:327–332. Boutwell, A.E., Johnson, M.B., & Rutherford, P., et al. An early look at a four-state initiative to reduce avoidable hospital readmissions. Health Aff (Millwood) 2011;30:1272–1280.

Bowles, K.H., Holland, D.E., & Potashnik, S.L. Implementation and testing of interdisciplinary decision support tools to standardize discharge planning. Nurs Inform 2012;2012:41–45. Bowles, K.H., Hanlon, A., & Holland, D., et al. Impact of discharge planning decision support on time to readmission among older adult medical patients. Prof Case Manag 2014;19:29–38. Burke, R.E., & Coleman, E.A. Interventions to decrease hospital readmissions: keys for cost-effectiveness. JAMA Intern Med 2013; 173:695–698. Castner, J., Wu, Y.B., & Paul, S., et al. Frequent emergency department utilization and behavioral health comorbidities. Nursing Res. 64(1): 3–12. DOI: 0.1097/NNR.0000000000000065. Centers for Medicare and Medicaid. Fact sheet: comprehensive primary care initiative. 2012. Available at: http://innovation.cms.gov/ Files/fact-sheet/CPCI-Fact-Sheet.pdf. Accessed August 14, 2013. Charlson, M.E., Pompei, P., Ales, K.L., & MacKenzie, C.R. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis 1987;40:373–383. Conway, P.H., & Berwick, D.M. Improving the rules for hospital participation in Medicare and Medicaid. JAMA 2011;306:2256–2257. Dharmarajan, K., Hsieh, A., & Lin, Z., et al. Diagnoses and timing of 30-day readmissions after hospitalization for heart failure, acute myocardial infarction, or pneumonia. JAMA 2013;309:355–363. Epstein, A.M., Jha, A.K., & Orav, E.J. The relationship between hospital admission rates and rehospitalizations. N Engl J Med 2011; 365:2287–2295. Gill, J., Mainous, A.G., 3rd, & Nsereko, M. Does having an outpatient visit after hospital discharge reduce the likelihood of readmission? Del Med J 2003;75:291–298. Hewner, S. A population-based care transitions management model for chronically ill elders. Nurs Econ 2014;32:109–166; 141. Hewner, S., & Seo, J.Y. Health informatics’ role in bridging population and person-centered perspectives on complexity to improve care transitions in chronic disease. Online J Nurs Inform 2014;18(2). Hewner, S., Seo, J.Y., Gothard, S., & Johnson, B.J. Aligning population-specific care management with chronic disease complexity. Nurs Outlook 2014a;62:250–258. Available at: http:// dx.doi.org/10.1016/j.outlook.2014.03.003. Hewner, S., Mehrok, N., Wu, Y.W., & Doloresco, F. Impact of intervention intensity on health outcomes in Medicaid adults with chronic disease: comparative effectiveness research using existing clinical electronic data. 2014b. Manuscript Submitted Publication.

Vol. 00 No. 0 Month 2015

Hines, A.L., Barrett, M.L., Jiang, H.J., & Steiner, C.A. Conditions With the Largest Number of Adult Hospital Readmissions by Payer, 2011. HCUP Statistical Brief #172. Rockville, MD: Agency for Healthcare Research and Quality; 2014. Available at: www.hcup-us. ahrq.gov/reports/statbriefs/sb172-ConditionsReadmissions-Payer.pdf. Accessed February 4, 2015. Holland, D.E., & Harris, M.R. Discharge planning, transitional care, coordination of care, and continuity of care: clarifying concepts and terms from the hospital perspective. Home Health Care Serv Q 2007; 26:3–19. Huff, E.D. Outpatient utilization patterns and quality outcomes after first acute episode of mental health hospitalization is some better than none, and is more service associated with better outcomes? Eval Health Prof 2000; 23:441–456. Jencks, S.F., Williams, M.V., & Coleman, E.A. Rehospitalizations among patients in the Medicare fee-for-service program. N Engl J Med 2009;360:1418–1428. Jiang, H.J., & Wier, L.M. All-Cause Hospital Readmissions Among Non-Elderly Medicaid Patients, 2007. HCUP Statistical Brief #89. Rockville, MD: Agency for Healthcare Research and Quality; 2010. Available at: www.hcup-us. ahrq.gov/reports/statbriefs/sb89.pdf. Accessed February 4, 2015. Jiang, H.J., Stryer, D., Friedman, B., & Andrews, R. Multiple hospitalizations for patients with diabetes. Diabetes Care 2003;26: 1421–1426. Jiang, H.J., Andrews, R., Stryer, D., & Friedman, B. Racial/ethnic disparities in potentially preventable readmissions: the case of diabetes. Am J Public Health 2005;95: 1561–1567. Kangovi, S., & Grande, D. Hospital readmissions– not just a measure of quality. JAMA 2011;306: 1796–1797. Kansagara, D., Englander, H., & Salanitro, A., et al. Risk prediction models for hospital readmission. JAMA 2011;306:1688–1698. Krumholz, H.M., Lin, Z.Q., & Keenan, P.S., et al. Relationship between hospital readmission and mortality rates for patients hospitalized with acute myocardial infarction, heart failure, or pneumonia. JAMA 2013;309:587–593. Maeng, D.D., & Martsolf, G.R. Comparing patient outcomes across payer types: implications for using hospital discharge records to assess quality. Health Serv Res 2011; 46:1720–1740. Available at: http://dx.doi. org/10.1111/j.1475-6773.2011.01285.x. Naessens, J.M., Leibson, C.L., Krishan, I., & Ballard, D.J. Contribution of a measure of disease complexity (COMPLEX) to prediction of outcome and charges among

hospitalized patients. Mayo Clinic Proc 1992; 67(12):1140–9. Peek, C.J., Baird, M.A., & Coleman, E. Primary care for patient complexity, not only disease. Fam Syst Health 2009;27:287–302. Press, M.J., Scanlon, D.P., & Ryan, A.M., et al. Limits of readmission rates in measuring hospital quality suggest the need for added metrics. Health Aff (Millwood) 2013;32: 1083–1091. Prince, J.D., Akincigil, A., & Hoover, D.R., et al. Substance abuse and hospitalization for mood disorder among medicaid beneficiaries. Am J Public Health 2009;99:160–167. Raven, M.C., Doran, K.M., & Kostrowski, S., et al. An intervention to improve care and reduce costs for high-risk patients with frequent hospital admissions: a pilot study. BMC Health Serv Res 2011;11:270. Regenstein, M., & Andres, E. Reducing hospital readmissions among medicaid patients: a review of the literature. Qual Manag Health Care 2014;23:20–42. Rennke, S., Nguyen, O.K., & Shoeb, M.H., et al. Hospital-initiated transitional care interventions as a patient safety strategy: a systematic review. Ann Intern Med 2013;158:433–440. Shippee, N.D., Shah, N.D., & May, C.R., et al. Cumulative complexity: a functional, patient-centered model of patient complexity can improve research and practice. J Clin Epidemiol 2012;65:1041–1051. Trudnak, T., Kelley, D., & Zerzan, J., et al. Medicaid admissions and readmissions: understanding the prevalence, payment, and most common diagnoses. Health Aff (Millwood) 2014;33:1337–1344. Young, W., Kohler, S., & Kowalski, J. PMC patient severity scale: derivation and validation. Health Serv Res 1994;29:367–390. Zhong, W., Chow, R., & He, J. Clinical charge profiles prediction for patients diagnosed with chronic diseases using Multi-level Support Vector Machine. Expert Syst Appl 2012;39:1474–1483. Available at: http://dx. doi.org/10.1016/j.eswa.2011.08.036.

Authors’ Biographies Sharon Hewner, PhD, RN, is a nurse anthropologist with clinical experience across the healthcare continuum. Her research interests include using existing clinical records (including claims data) to understand population health and comparative effectiveness research on health outcomes such as hospital and emergency utilization. Dr. Hewner developed the COMPLEXedex to stratify the population based on disease comorbidity and complexity. Yow-Wu Bill Wu, PhD. Dr. Wu’s expertise is in research methodology and applied statistics. He has served as

13

14

Journal for Healthcare Quality

a statistical consultant in previously funded research projects in the past two decades. His teaching experience includes advanced statistics techniques and questionnaire construction and psychometric property evaluation can help this research team to build a strong research proposal. Jessica Castner, PhD, RN, CEN, is a board certified emergency nurse. Her research interests include emergency department utilization, novel technology to reduce asthma exacerbations, and health effects of environmental air quality. She contributed to the conceptual planning, survival analysis, behavioral health measures, drafting, and editing of this paper.

For more information on this article, contact Sharon Hewner at [email protected]. Supported by the Patricia H. Garman Behavioral Health Nursing Endowment Fund Award, University at Buffalo, State University of New York. Supplemental digital content is available for this article. Direct URL citations appear in the printed text and in the HTML and PDF versions of the article at www.jhqonline.com. The authors declare no conflicts of interest.