Hospital inpatient readmissions and emergency department (ED) use are potentially avoidable and expensive health care events. Among nonelderly Medicaid members admitted to a hospital for a mental or substance use disorder, 12% are readmitted within 30 days (
1). In 2011, the mean cost of mental health–related admissions exceeded $10,000 and the mean cost of substance use disorder admissions exceeded $7,000 (author calculation from data at
http://hcupnet.ahrq.gov). The presence of a mental or substance use disorder increases hospitalization rates and per capita costs (
2).
Several factors are associated with behavioral health readmissions. Previous hospitalization is the most consistent predictor of future hospitalization (
3,
4). In addition, individuals with comorbid mental and substance use disorders, severe impairments in self-care, severe and persistent psychiatric symptoms, poor medication adherence, psychotic illness, and general medical comorbidities have a higher risk of readmission (
4–
8). The risk of treatment dropout, relapse, homelessness, and suicide in the weeks following hospital discharge are especially high for patients with mental disorders (
9), and these patients may be more likely to be readmitted. Among Medicaid patients with disabilities, the combination of schizophrenia and substance abuse was found to increase the likelihood of readmission by 73% (
10).
Research on the association between postdischarge service use and readmissions has produced mixed findings. Outpatient aftercare has been associated with increased readmissions (
11,
12), reduced readmissions (
13), and no effect on readmissions (
14,
15). Results have been mixed because many studies focus on aftercare services in general rather than specific types of aftercare, such as intermediate services. More detailed studies have shown that relationships between aftercare and readmissions are complex. Studies have shown associations between receiving care at a community mental health center and decreased readmissions (
4) and between increased use of community services and increased hospitalization (
3). One study found that some use of posthospitalization care was related to decreased readmissions but that higher than average use of outpatient services was related to increased readmissions (
16).
Few, if any, researchers have examined the extent to which intermediate services promote a successful transition from inpatient to outpatient care and help patients stabilize over time. In this study, intermediate services were defined as services that bridge acute treatment for behavioral health disorders and long-term outpatient care. Common behavioral health intermediate services include residential treatment, partial hospitalization, and intensive outpatient services.
Several studies have examined the relationship between residential treatment services and readmissions for substance use disorders, but there have been no studies on the relationship between intermediate services and readmissions for mental health conditions. Moos and colleagues (
17) examined the effect of treatment in community residential treatment facilities on readmissions for individuals with substance use disorders. They found that a longer stay in community residential treatment services was associated with fewer readmissions and that individuals who dropped out of such programs had particularly high rates of readmission. Groppenbacher and colleagues (
18) found that veterans who completed residential and outpatient substance use disorder programs were less likely to be readmitted than those who completed only residential treatment or those who did not complete either program.
The use of aftercare, particularly intermediate services, may be an effective strategy for reducing behavioral health hospital readmissions and ED use. However, research has not examined the effect of intermediate service use on readmissions and subsequent ED use. The objective of this study was to examine the impact of intermediate services on readmissions and subsequent ED use. Because the rate of readmission is particularly high among Medicaid enrollees with disabilities, we examined the impact of intermediate services separately for individuals eligible for Medicaid because of disability and those eligible through other means.
Methods
Data
We used data from the 50-state 2008 Centers for Medicare and Medicaid Services (CMS) Medicaid Analytic eXtract (MAX) data set and limited our analysis to 30 states for which we could identify intermediate services in the claims data (Arkansas, Arizona, Colorado, Connecticut, Florida, Hawaii, Iowa, Idaho, Illinois, Indiana, Massachusetts, Maryland, Minnesota, Missouri, Montana, North Carolina, North Dakota, Nebraska, New Hampshire, New Jersey, New Mexico, New York, Ohio, Rhode Island, Texas, Virginia, Vermont, Wisconsin, West Virginia, and Wyoming). The MAX data set is an administrative claims database derived from the Medicaid Statistical Information System.
Our sample included inpatient admissions for adults ages 22 to 64 who received a primary diagnosis of a mental or substance use disorder between April 1 and September 30, 2008. These dates allowed for a 90-day prehospitalization period to assess health care utilization patterns and a 90-day posthospitalization period to assess readmissions. We excluded individuals enrolled in both Medicaid and Medicare as well as those in either comprehensive or behavioral health managed care plans because of the potential for incomplete claims data. To permit a clean period of exposure to intermediate care, we excluded admissions of individuals who had an additional hospitalization within seven days before or after the index admission. If an individual had more than one inpatient admission that met our criteria, we selected the first admission. These analyses were exempt from requirements for institutional review board approval because they do not meet the definition of human subjects research.
Variables
We examined two outcome variables: time to behavioral health readmission and time to behavioral health ED visit occurring within 90 days of discharge from the index hospitalization. We defined these behavioral health events as associated with a primary diagnosis of a mental or substance use disorder.
Our primary independent variables were receipt of the following intermediate services within seven days of the discharge date: residential services, partial hospitalization services, intensive outpatient treatment, other rehabilitative services, and any intermediate service (a composite variable indicating that an individual had received at least one intermediate service). Other rehabilitative services included rehabilitation CPT codes that did not clearly fall into residential treatment, partial hospitalization, or intensive outpatient treatment service categories.
From administrative claims, we collected data on age, sex, race-ethnicity, basis of eligibility, primary diagnosis, and comorbidities for the index hospitalization; health care use in the 90 days prior to the index hospitalization; and health care use in the seven days following discharge. We also examined prescription fills in the 90 days before hospitalization, including antipsychotic, antidepressant, and substance use disorder medications (
5).
Analysis
We generated descriptive statistics for the two cohorts as well as the relationships between the independent variables and the outcome measures for the two cohorts. [A table presenting these results is available in an online supplement to this article.] We performed a multivariate Cox proportional hazards model to estimate the relationship between the independent variables and time to readmission and ED use.
Because the relationship between use of intermediate services and outcomes was of specific interest and the number of beneficiaries who used intermediate services was low, we used propensity score matching to examine the impact of intermediate service use on readmissions and subsequent ED use. We performed two separate within-state, one-to-one propensity score matches for the final survival analysis for the use of intermediate services by individuals eligible for Medicaid because of disability and those eligible for other reasons. Within each cohort, we calculated the propensity score for each index admission as the predicted probability of receiving intermediate services within seven days after hospital discharge on the basis of the patients’ demographic characteristics, their preadmission service and medication use, and clinical characteristics from the index hospitalization, including primary diagnoses, comorbidities, and length of stay. We did not match on activities that occurred after the admission, because the decision to refer or not to refer to intermediate services was dependent on what occurred prior to or during the index admission.
Each individual who received intermediate services was matched to an individual with a similar propensity score who did not receive intermediate services by using the method of nearest neighbor with a caliper of 25% of the standard deviation of the propensity score. This resulted in two groups of equal size (each with N=223 for the nondisabled cohort and N=1,244 for the disabled cohort) that were equally likely to have received intermediate services given propensity matching that included adjustment for many variables. After matching, we examined the data to ensure appropriate results by comparing the standardized mean differences between the covariates of the matched samples. We used a Cox proportional hazards regression model to estimate the association between intermediate services and the risk of readmission or an ED visit for adults with and without disabilities in the matched samples.
Results
In our sample of 32,037 adult admissions, 27.6% were nondisabled adult admissions (N=8,854) and 72.4% were disabled adult admissions (N=23,183).
Table 1 presents data on the characteristics of our samples of nondisabled and disabled adult index admissions. Most of the nondisabled participants were female (55.2%) and between age 22 and 44 (76.0%). Fifty percent had a primary diagnosis of a substance use disorder. The average length of stay for the nondisabled sample was 6.3±6.7 days. In the 90 days prior to their index hospitalization, 16.9% had a behavioral health inpatient stay, 28.7% had a behavioral health outpatient visit, 26.3% had a behavioral health ED visit, 34.1% had at least one antidepressant prescription fill, and 19.6% had at least one antipsychotic prescription fill. After discharge, 45.5% received no services within seven days, 42.4% filled a prescription for a behavioral health disorder, and 17.2% had a behavioral health outpatient visit. Only 2.5% received intermediate services.
In contrast, the disabled sample was predominantly male (51.0%), and a larger proportion of individuals were between age 45 and 64 (48.9%). The average length of stay was 8.0±7.5 days. The largest proportion of disabled adults was admitted primarily with a diagnosis of schizophrenia and psychotic disorders (35.3%), and only 20.8% were admitted for a substance use disorder. This sample had generally higher rates of medical comorbidities than the nondisabled sample. In the 90 days prior to the index hospitalization, the disabled sample had similar rates of behavioral health inpatient stays (15.0%) but higher rates of behavioral health outpatient visits (36.9%), behavioral health prescription visits (29.4%), behavioral health ED use (32.5%), antidepressant prescription fills (46.2%), and antipsychotic prescription fills (51.4%). Within seven days of discharge, 29.7% did not receive any services, 59.5% received a behavioral health prescription, 17.3% received a behavioral health outpatient visit, 11.0% received a behavioral health prescription visit, and 6.9% received assertive community treatment. In addition, 5.4% received intermediate services, with 1.3% receiving services in a partial hospital setting and 3.1% receiving other rehabilitative services.
For nondisabled adults, 25.6% (N=2,268) and 19.3% (N=1,710) of index admissions were followed by a behavioral health readmission or ED visit, respectively, within 90 days. Individuals who received no services within seven days of discharge had less time to a readmission (hazard ratio [HR]=1.32, p<.001) and less time to an ED visit (HR=1.11, p=.027) (
Table 2). Furthermore, those who received medication-assisted treatment had less time to a readmission or ED visit (HR=1.08, p=.022, and HR=1.07, p=.025, respectively). Individuals who received any intermediate services had more time until a readmission (HR=.71, p=.041) or an ED visit (HR=.68, p=.002). As expected, some demographic and clinical characteristics as well as health care use prior to the index admission were associated with time to readmission and ED visit (
Table 2).
For adults with disabilities, 28.3% (N=6,570) and 25.9% (N=6,005) of index admissions were followed by a behavioral health readmission or ED visit, respectively, within 90 days. Individuals who received no services within seven days of discharge had less time to readmission (HR=1.14, p<.001) (
Table 3). Individuals who received assertive community treatment services within seven days had more time to readmission (HR=.96, p<.001), and those who received a behavioral health prescription visit or medication-assisted treatment within seven days of discharge had more time to an ED visit (HR=.95, p=.020, and HR=.96, p=.004, respectively). The receipt of intermediate services within seven days had no impact on time to readmission or ED use. As for the nondisabled adults, several demographic and clinical characteristics as well as health care use prior to index admission were associated with time to readmission and ED use (
Table 3).
Propensity Score Matching Results
The one-to-one propensity score matching for any intermediate service use by basis of eligibility was successful. The mean standardized difference between covariates was below 10% for all covariates. The results of the propensity score–matched survival analysis were not statistically significant for adults with and without disabilities (
Table 4) but suggested that the impact of intermediate services may depend on a beneficiary’s basis of eligibility.
Discussion
Less than 5% of adult Medicaid enrollees received intermediate services within seven days of an inpatient hospitalization. This low rate of intermediate service use after a behavioral health admission may reflect limited coverage and availability under 2008 Medicaid fee-for-service systems. In many states, fee-for-service Medicaid may not cover all intermediate services, and coverage through other state funding streams, such as block grants, would not be picked up in this analysis. Even if intermediate services are covered via Medicaid, geographic constraints may prevent access to them by individuals in rural areas (
19,
20). Also, outpatient treatment facility waitlists may prevent individuals from immediately engaging in intermediate services postdischarge. Previous research has demonstrated that wait times serve as a major barrier to engagement in substance use disorder treatment (
21,
22) and mental health treatment (
23). These factors may have contributed to the limited significance of the survival analyses in this study.
Among Medicaid enrollees who received intermediate services within seven days of an inpatient hospitalization, receipt of those services was more likely to be associated with lower readmissions and ED use for Medicaid enrollees without disabilities than for those with disabilities, but the difference was not statistically significant in the propensity score–matched analysis. One possible explanation for these differential effects is the differences in the sample characteristics in terms of demographic characteristics, clinical factors, and prior service utilization. The nondisabled population was younger, predominantly female, and dominated by index admissions for substance use disorder, with significantly fewer index admissions for severe mental illness than the disabled sample. Past research has suggested that diagnoses and illness severity (
4–
7,
10,
24), substance use disorder comorbidity (
4,
25), and adherence to medications (
7,
8,
26–
30) have a stronger effect on readmissions and ED visits than use of intermediate services. Furthermore, past research on intermediate services has shown residential services to be effective for individuals with substance use disorders (
17,
18).
Prior research on the impact of intermediate services for individuals with mental disorders has focused largely on how these services compare with inpatient services. This study went further by examining the impact of intermediate services on outcomes, such as readmissions or ED visits, when intermediate services were used as a bridge to outpatient services. There are several possible explanations for the failure of intermediate services to affect readmissions or ED visits in the disabled population in this analysis. First, the type of intermediate service may have differential effects on readmissions and ED visits. Although this analysis did not have sufficient numbers to study individual services, the unadjusted relationships suggest potential differences in the impact of different services (supplemental material). Second, regardless of eligibility status, the presence of any intermediate service use may be less predictive than the receipt of some threshold number of services. The relationship may resemble a J shape, with a sufficient but limited amount of aftercare related to improved outcomes and a large amount acting as a marker for decompensation or severity and being associated with poor outcomes (
3,
16).
Third, the use of intermediate services may serve as a proxy for clinical severity that was not sufficiently addressed by the propensity score matching. Finally, other types of outpatient care after discharge also affect readmissions and ED use. Previous research has demonstrated an association between receipt of assertive community treatment (
31–
33), general and behavioral health outpatient visits (
13), or prescriptions for mental or substance use disorder medications (
34) and reduced readmissions. Similarly, the multivariate analyses in this study found significant associations between lack of services and decreased time to readmissions as well as significant associations between assertive community treatment and increased time to readmission and antidepressant or antipsychotic medication use and increased time to readmission and ED use in the disabled population.
This study had several limitations. First, we restricted the analysis to patients who did not have a readmission within seven days of their hospitalization to create a meaningful opportunity for exposure to intermediate services. This excluded readmissions within seven days that might have been avoided with intermediate services. Second, the percentage of individuals receiving intermediate services within seven days was low. As a result, we restricted the analysis to states where intermediate services were identified. This approach limited the generalizability of the analysis, as did the use of only fee-for-service Medicaid records. Managed care programs have a greater incentive than fee-for-service programs to use intermediate care as a cost-saving measure, and they may make it more readily available as a result.
Third, the low number of intermediate service users limited the power of the analysis, as did the low number of individuals who could be matched using propensity scoring; this may have limited the power to detect significant differences. However, the direction and magnitude of the HRs in the propensity score models were consistent with those in the regression models. That the analysis was underpowered also may indicate a failure to detect all intermediate services. In addition, we used both national and state-specific codes in states where we could identify their use, but we may have miscategorized or overlooked codes that indicate intermediate services. States may use nonspecific codes to reimburse for intermediate services. Fourth, propensity scoring can balance observations only on measured covariates. Although our models included covariates that have been significantly correlated with the outcomes of interest, there may have been unobserved patient and state characteristics that also accounted for some observed relationships in the subset of those who were matched and those who were unmatched. In particular, there may be a need for more detailed clinical variables than those available in claims data to effectively match patients. Examining more extended patterns of past service use may capture additional characteristics.
Conclusions
This analysis showed low use of intermediate care services in fee-for-service Medicaid programs after an inpatient behavioral health hospitalization. We did not find consistent statistically significant relationships between intermediate service use and behavioral health readmissions or ED visits within 90 days of discharge. Given the high rates of readmission and ED use in the Medicaid population, identifying levers to reduce these rates and improve mental health outcomes is important. Additional research with sufficient numbers of recipients of intermediate services is needed to develop an understanding of the impact of the type and number of intermediate services on outcomes for both the disabled and the nondisabled population.
Acknowledgments
The authors thank Tim Bullock, Ph.D., and Degang Wong, Ph.D., for programming assistance, Molly Bailey, B.S., for research assistance, Paige Jackson, M.S., and Linda Lee, Ph.D., for editorial assistance, and Mitchell Berger, M.P.H., for reviewing the manuscript.