Readmission Patterns and Effectiveness of Transitional Care Among Medicaid Patients With Schizophrenia and Medical Comorbidity

Setting

Seventy percent of North Carolina Medicaid recipients were enrolled in CCNC during the study period. CCNC is an enhanced PCCM managed care model that links recipients to a primary care medical home in 1 of 14 nonprofit, physician-led regional networks that coordinate care management and quality improvement activities [10, 11]. North Carolina Medicaid supports network activities through care management fees ranging from $3.72 to $13.72 per member per month. Both medical and behavioral health services are reimbursed on a fee-for-service basis, with the exception of a capitated behavioral health pilot program operating in 5 counties.

During the study period, Medicaid behavioral health services were provided by a wide range of largely nonprofit local providers with the oversight of a contracted utilization review company and public local management entities (LMEs), which were responsible for developing the network, conducting quality improvement activities, overseeing providers in the community, and identifying gaps in service. Each CCNC network had a behavioral health coordinator and a consulting psychiatrist who served as resources to care managers, supported primary care practices in managing less severe behavioral health conditions, and facilitated the referral of unstable patients to specialty behavioral health providers with oversight by the LMEs.

Data Sources Used in the Analysis

Demographic data (age, sex, race, ethnicity, language, county of residence, and CCNC enrollment status) came from the North Carolina Medicaid eligibility and enrollment files. Hospitalization service dates, diagnostic related groups (DRGs), discharge disposition, substance abuse diagnoses, and assertive community treatment (ACT) services were identified from paid Medicaid claims. Care manager assessments and interventions were documented in CCNC's care management information system in structured data fields according to standardized documentation requirements.

Patients were categorized by disease burden with the categorical and hierarchical Clinical Risk Group (CRG) methodology developed by 3M Health Information Services [12]. Based on diagnoses and procedure information in paid claims from July 1, 2010 through June 30, 2011, each CCNC beneficiary was assigned to 1 of 1,075 CRGs with others of similar disease burden and severity. Each CCNC beneficiary was then assigned a CRG risk score that reflected average total costs of care within that CRG relative to the CCNC population as a whole. CRGs can be combined into 44 Aggregated Clinical Risk Groups, which classify patients according to number and severity of acute and chronic conditions.

Cohort Assignment

Patients were retrospectively assigned to the transitional care cohort if a CCNC care manager completed any screening, assessment, or intervention for that patient between the date of hospital admission and 30 days post-discharge. Individuals readmitted within 30 days of discharge were considered transitional care patients if any care management activity occurred prior to readmission. Following an intent-to-treat design, all patients who were screened by a care manager were included in the transitional care cohort, even if the care manager decided to defer or was unable to contact the patient, or if the patient refused services. The usual care cohort consisted of patients who did not receive any screening, assessment, or intervention by a CCNC care manager from the time they were admitted until the end of their Medicaid eligibility or the end of the study period. Patients were excluded from the analysis if they did not receive a care management activity within 30 days of discharge but did receive care management during the subsequent 12 months.

During the time frame of this study, far more patients were discharged every day than the transitional care program could handle. Networks aimed to assess as many hospitalized enrollees as possible, with priority given to those in the aged, blind, or disabled eligibility category. Care managers screened patients through review of Medicaid claims history and hospital records, discussion with the hospital care team, and interviews with the patient or caregivers, as available. CCNC care managers were expected to use professional judgment to determine the frequency and duration of interventions, according to individual patient need, so higher-risk individuals were likely to receive longer and more intensive support following discharge. Further details about patient selection and the delivered intervention are published elsewhere [8, 9].

Statistical Analysis

The primary study outcome was time to readmission, defined as the first admission date following initial discharge, excluding transfers or same-day readmissions. We followed patients for up to 12 months post-discharge, observing readmission claims through September 30, 2011 and paid as of January 1, 2012. All patients remained in the analysis until readmission or until censored due to either death or a gap in Medicaid eligibility greater than 2 months. We calculated 12-month readmission survival estimates for transitional care and usual care patients, and we tested the statistical significance of group differences with the Wilcoxon-Gehan statistic. These analyses were conducted separately for psychiatric and nonpsychiatric index hospitalizations.

ACT is an intensive and highly integrated model of outpatient treatment for patients with serious mental illness who are at particularly high risk of psychiatric hospitalization [13]. As patients eligible for ACT services were expected to have the highest risk of hospitalization, and evidence-based ACT services can reduce readmissions [14], we conducted a subanalysis limited to patients in both the transitional care and usual care groups receiving ACT services at the time of the index hospitalization to evaluate the incremental impact of transitional care services for these high-risk patients already receiving intensive outpatient treatment.

We then applied a Cox proportional hazards [15] regression model to identify demographic, clinical, and hospital characteristics associated with a decreased risk of readmission, and we estimated the effect of transitional care while controlling for these characteristics. Because we only had Medicaid claims data for patients in the usual care group, predictors in the model were limited to variables reliably available through claims. Model variables included age, sex, race (white, African American, American Indian, Asian, or other), ethnicity (Hispanic or other), language (English, Spanish, or other), CRG risk score (relative clinical burden), and presence of a substance abuse or dependence diagnosis on any claim within the study period. As an additional indicator of severity of psychiatric care needs and readmission risk, we included a dichotomous variable indicating receipt of ACT services at any time during 3 years spanning the study period (July 1, 2009 through June 30, 2012). Characteristics of the index hospitalization were also considered; these factors included whether it was a psychiatric hospitalization (as indicated by an inpatient claim DRG of 876–887; coded as yes/no) and hospital size (less than 100 beds, 100–499 beds, or 500 or more beds). Age and CRG risk score were both entered as continuous variables; all other variables were categorical. All analyses were conducted in SPSS version 12.0.