Impact of Medical Homes on Expenditures and Utilization for Beneficiaries With Behavioral Health Conditions

In the United States, an estimated 43.6 million adults have a mental illness, and 14.5 million adults have a substance use disorder (1). Because primary care providers play a central role in the identification and treatment of many mental health conditions, a diverse literature supports integration of primary and behavioral health care services (2–10). Some clinicians and policy makers have looked to the patient-centered medical home (PCMH) as an avenue for this integration. The PCMH is an approach to transforming and organizing primary care to improve receipt of coordinated, comprehensive, accessible, and patient-centered health care, and an emerging literature has demonstrated that the PCMH model may be successful in improving access to and quality of health care and reducing the costs of care (11,12).

On the basis of this emerging evidence, in late 2011 and early 2012 the Centers for Medicare and Medicaid Services (CMS) began participating in eight ongoing state-led PCMH initiatives in Maine, Michigan, Minnesota, New York, North Carolina, Pennsylvania, Rhode Island, and Vermont (13). These eight initiatives included each state’s Medicaid agency and some of the commercial payers operating within these states. This unique arrangement was known as the Multi-Payer Advanced Primary Care Practice (MAPCP) demonstration. Unlike other PCMH initiatives, which have involved a single payer or have focused on implementing practice transformation within the context of a single health system (14–18), the MAPCP demonstration blended funding from both public (Medicare and Medicaid) and commercial payers to support practices in PCMH transformation. Participating primary care practices also received critical nonfinancial support through the demonstration, such as technical assistance and feedback on the cost and utilization impacts of their efforts.

Over the course of the MAPCP demonstration, behavioral health integration gained support as a core principle of the PCMH (5,19,20), and individuals with significant morbidity, including those with mental illnesses and substance use disorders, were expected to benefit from PCMH activities, such as improved care management. These activities in turn could lead to greater use of outpatient behavioral health services and decreases in the rates of hospitalizations and emergency department (ED) visits. If reductions in hospitalizations and ED visits are enough to offset the increased use of outpatient services, there could be an overall reduction in total expenditures. However, if practices are successful in linking these individuals to services for which there was pent up demand, then overall expenditures could increase.

This study explored the potential impacts on both non–behavioral health and behavioral health service expenditures and utilization. Specifically, we studied the association between participation in the MAPCP demonstration and changes in expenditures and utilization for Medicare and Medicaid beneficiaries with behavioral health conditions. We also explored how states’ initiatives evolved over the course of the MAPCP demonstration to improve care for these beneficiaries.

Methods

Qualitative and Quantitative Data Sources

We used qualitative and quantitative methods to develop an in-depth understanding of the transformative processes occurring within participating PCMH practices. Qualitative data were collected via review of each state’s MAPCP demonstration program documents (for example, quarterly reports to CMS and updates posted on the states’ Web sites) and site visits to each MAPCP state in the fall of 2012, 2013, and 2014. We conducted over 700 structured interviews with clinical and nonclinical staff, state officials, and payers across the eight states. Among those interviewed were individuals familiar with emerging behavioral health initiatives. The goal of the interviews was to learn more about the motivation for, successes of, and challenges with these initiatives. [Additional details on qualitative data collection, including interview topics, are presented in an online supplement to this article.]

Quantitative data included Medicare and Medicaid enrollment data, Medicare fee-for-service claims, and data on Medicaid managed care encounters and fee-for-service claims, covering a multiyear period (typically two to four years) before Medicare joined the state initiatives through December 2014 (that is, the end of the demonstration period) [see the online supplement for details on state-specific analytic time periods].

MAPCP and Comparison Group Identification

To identify the impact of PCMH transformation, we compared Medicare and Medicaid beneficiaries engaged with participating MAPCP practices with beneficiaries engaged with primary care practices that were not medical homes, defined as not having PCMH recognition by the National Committee on Quality Assurance (21). These comparison practices were chosen from within the state, if possible, or from geographically similar out-of-state areas when it was not possible to choose from within the state because the initiative was statewide. Medicare beneficiaries were attributed to the MAPCP or comparison group practice from which they received a plurality of their primary care services. Medicaid enrollees were similarly attributed, or they were attributed because the state’s Medicaid program assigned them a primary care provider who practiced at a designated MAPCP-participating practice or a comparison group practice. The number of total attributed MAPCP demonstration beneficiaries by state and payer are reported in Table 1. [Additional details on selection of the comparison group practices, eligibility criteria, and the attribution process are included in the online supplement.]

After beneficiaries were attributed to primary care practices, we subset the sample to individuals with behavioral health conditions, defined as having at least one inpatient claim or two or more outpatient claims with a primary diagnosis of a mental or substance use disorder (identified by ICD-9-CM codes 291–316) during the 12-month period before their participation in the demonstration. The decision to rely on primary diagnoses was deliberate in order to understand how primary care transformation may have affected outcomes among those with the highest needs. For the Medicaid analysis, we also subset the study sample to nonelderly adults (ages 18–64). On the basis of these criteria, the prevalence of behavioral health conditions ranged from a low of 10% (N=17,029) among Medicare beneficiaries in the demonstration and comparison groups in North Carolina to a high of 18% among these groups in Minnesota (N=34,773) and 20% among these groups in Maine (N=20,495). Among Medicaid beneficiaries, the prevalence ranged from a low of 2% among the demonstration and comparison groups in Pennsylvania (N=641) to a high of 14% among these groups in Maine (N=6,614).

The prevalence of behavioral health conditions was lower in the Medicaid population compared with the Medicare population because we restricted the sample to only those with a primary diagnosis of a behavioral health condition and because Medicare-Medicaid dually eligible beneficiaries were excluded from the Medicaid sample (instead we included them in the Medicare sample). The prevalence of behavioral health conditions in Pennsylvania and North Carolina was particularly low and nonrepresentative of individuals with behavioral health conditions because Pennsylvania and North Carolina had comprehensive Medicaid behavioral health carve-outs, and they were unable to provide the additional behavioral health data for this analysis. As a result, we intentionally do not report Pennsylvania’s or North Carolina’s results for Medicaid service use and expenditures.

Dependent Variables: Health Service Use and Expenditures

The first set of dependent variables included measures of health service use: all-cause inpatient admissions, behavioral health inpatient admissions, ED visits, behavioral health ED visits, and behavioral health outpatient visits. The second set included two measures of expenditures: total expenditures and expenditures for services with a principal diagnosis of a behavioral health condition. Behavioral health services and expenditures were defined as services for which the primary diagnosis was for a behavioral health condition. Michigan and Minnesota did not report expenditures in the Medicaid managed care encounters, and because the majority of the study sample was enrolled in Medicaid managed care, we do not report on Medicaid expenditures for these states. Furthermore, North Carolina did not provide Medicaid claims for chemical dependency services pursuant to their interpretation of 42 CFR Part 2; however, the other states did so (22).

Statistical Analyses

We used covariate-adjusted, difference-in-difference regression modeling to examine changes in utilization and expenditures before and after participation in the MAPCP demonstration, comparing MAPCP beneficiaries to the comparison group (23). In each regression model, we controlled for age, race, urban place of residence, gender, dual Medicare-Medicaid enrollment (for Medicare), enrollment due to disability, enrollment due to end-stage renal disease (for Medicare), residence in an institutionalized setting, morbidity risk scores (Charlson Index and Hierarchical Conditions Category for Medicare and the Chronic Illness and Disability Payment System and presence of perinatal conditions for Medicaid), count of comorbid conditions, enrollment in fee-for-service or managed care (for Medicaid), and continuity of enrollment (for Medicaid). We estimated Medicare and Medicaid expenditures by using ordinary least squares and counts of visits and hospitalizations in Medicare by using negative binomial regression. Because persons in the nonelderly Medicaid population do not use services as often as those in the Medicare population, visits and hospitalizations in Medicaid were modeled as dichotomous (ever used a service) and estimated by using logistic regression. Finally, Medicaid expenditures exceeding the 99th percentile were top-coded at the 99th percentile to reduce the influence of outlier observations. Medicare expenditures had fewer outliers and were not top-coded. All regression analyses included clustered standard errors (enrollees clustered within primary care practices).

Because Medicare and Medicaid beneficiaries were not randomly assigned to MAPCP and comparison practices, we used entropy-balanced weighting to correct for potential bias introduced by observed sociodemographic and geographic characteristics and features of the assigned primary care practice that differed by MAPCP and comparison groups (24). Entropy-balanced weighting statistically adjusts the study sample so that there are no significant differences between groups on observed characteristics. After balancing, the Medicare and Medicaid study samples within each state more closely resembled each other.

Results

Throughout the MAPCP demonstration, site visit interviewees in all states reported significant unmet need for behavioral health care among their patients, particularly their Medicaid patients. Interviewees often noted that improvements in the physical health of and utilization of general medical care by patients with behavioral health conditions were hindered because these patients were not receiving adequate behavioral health treatment. In an effort to bridge gaps in care, several states embarked on unique initiatives to improve access to care. Vermont implemented a model, known as hub and spoke, for integrating medication-assisted treatment services into all primary care practices, including demonstration practices, for Medicaid enrollees with co-occurring mental and substance use disorders (25). Maine’s demonstration practices coordinated with behavioral health organizations to deliver behavioral health care to Medicaid enrollees (26). Rhode Island’s demonstration practices were required to develop “compacts” (that is, a framework for communication and care transitions) with behavioral health providers, and the state initiated a work group to lead the transformation of integrated primary and behavioral health care. Minnesota’s participating practices received enhanced MAPCP demonstration payments from Medicare, Medicaid, and the commercial payers to coordinate care for individuals with severe and persistent mental illness. Maine, Michigan, New York, North Carolina, Rhode Island, and Vermont also leveraged support teams, which were developed to assist practices with patient panel management and the delivery of additional clinical and nonclinical services, to connect patients with behavioral health conditions to treatment services. State officials in Michigan, New York, North Carolina, and Pennsylvania did not report implementing initiatives aimed specifically at enhancing access to behavioral health services or coordinating care for this population.

Regardless of a state’s decision to direct specific resources under the demonstration, providers in all states reported engaging in numerous activities to provide accessible, comprehensive, and coordinated care for this population. Examples of activities included screening more patients for behavioral health disorders, training providers to improve management of behavioral health conditions, introducing on-site care managers to connect patients to community-based behavioral health providers, referring patients with behavioral health needs to support teams for intensive care management, colocating behavioral health specialists within primary care, and using telehealth for behavioral health provider consultations. Notably, at both the state and practice levels, it took time to turn attention to this population, with many reported activities beginning in years 2 and 3 of the demonstration.

Table 2 summarizes sociodemographic, area-level, and practice-level characteristics of the MAPCP demonstration group, by state. The comparison group in each state was weighted (as discussed above) to resemble the intervention group on these characteristics.

In Tables 3 and 4, we present state-specific estimates of the impacts of the MAPCP demonstration on behavioral and non–behavioral health care expenditures and utilization among Medicare and Medicaid beneficiaries relative to those for the comparison group. As shown in Table 3, there were few associations between MAPCP participation and changing patterns of care among Medicare beneficiaries. Most results were either statistically nonsignificant or significant and contrary to expectations (for example, an increase in ED visits). In North Carolina, however, we observed greater decreases over time for the MAPCP group in behavioral health inpatient admissions and expenditures for services with a principal diagnosis of a behavioral health condition, relative to the comparison group, resulting in significant difference-in-difference estimates.

As shown in Table 4, in the Medicaid population, there was little change for the MAPCP group in spending on either overall medical care or on services with a principal diagnosis of a behavioral health condition, relative to the comparison group. Total Medicaid expenditures increased over time for MAPCP and comparison group beneficiaries in New York and Rhode Island, and expenditures grew significantly faster for MAPCP beneficiaries than for the comparison group, resulting in significant difference-in-difference estimates (New York, $96.17, p<.05; Rhode Island, $54.71, p<.05). Similarly, in Vermont, spending on behavioral health care increased over time for the MAPCP and comparison group beneficiaries, and expenditures grew faster for MAPCP beneficiaries than for those in the comparison group ($62.47, p<.05). We also found that the MAPCP demonstration had little impact on service utilization, with a few exceptions. In Vermont, the likelihood of having a behavioral health outpatient visit increased over time for both groups, but the growth was not as fast for MAPCP beneficiaries as it was for those in the comparison group, resulting in a significant difference-in-difference estimate (−7.92, p<.05).

Discussion

Even though primary care practices made concerted efforts to improve access to care for their patients with behavioral health conditions, few significant associations were found between the MAPCP demonstration and Medicare and Medicaid behavioral health utilization and expenditure outcomes. No consistent patterns of utilization and expenditures emerged across states. When we saw significant changes in expenditures, it was most often indicative of increased expenditures for the MAPCP demonstration groups relative to the comparison group, and the few significant utilization measures did not suggest any clear trend toward increased or decreased utilization.

As a subanalysis, we examined the same outcomes, comparing the MAPCP sample to a comparison group of beneficiaries who were receiving care in PCMHs that were not participating in the demonstration, and we found results similar to those presented here.

There are relatively few studies with which to compare our results. One survey found little evidence that PCMHs had an impact on use of recommended preventive care for adults with mental illness (27); however, other studies have found modest shifts in care for Medicaid enrollees in the PCMH with severe behavioral health conditions or general medical and psychiatric comorbidity (28–31). The clinical-condition criteria used to select our Medicaid sample were less restrictive than those in studies that found some positive effects, and thus there may be differential impacts on persons with more severe conditions that are masked when effects within a population of mixed medical complexity are examined.

The lack of quantitative results should be considered in context. First, the needs of this population did not go unnoticed by primary care providers. In all states, providers discussed efforts to change practice patterns to meet the needs of patients with behavioral health conditions, even though doing so was not an easy task. Providers in New York, North Carolina, Rhode Island, and Vermont noted that behavioral health provider shortages posed a significant barrier to coordination. The MAPCP demonstration in New York and North Carolina was implemented in rural areas, and providers noted that coordination was made even more challenging when compounded by the lack of available providers, regardless of medical specialty, in rural communities.

Furthermore, defining the roles, responsibilities, and work flows to successfully integrate behavioral health care with primary care also proved to be challenging for practices in the MAPCP demonstration. Providers in Maine and Rhode Island frequently reported that creating an effective, efficient system of coordinating with external resources, such as behavioral health organizations, takes significant time.

In addition, states that implemented specific behavioral health initiatives implemented them over years 2 and 3 of the MAPCP demonstration, and individual practices reported on their integration efforts more frequently in years 2 and 3 as well. Even though Medicare joined initiatives that were ongoing, site visit interviewees frequently reported that year 1 of the MAPCP demonstration was the time for the foundational work of learning to become a PCMH—for example, meeting state’s PCMH requirements; installing and learning to use health information technology systems, such as electronic health records; reporting on program metrics required by the state initiative; and learning how to identify and target select groups for intensive care management and follow-up. This left a relatively short period (one to two years at best) to focus on behavioral health services, and making significant changes in patterns of care and expenditures over that time may not be a realistic expectation.

In addition, as mentioned above, we looked only at the subset of beneficiaries who had primary diagnoses of behavioral health conditions. These beneficiaries may tend to use more health care services than beneficiaries with secondary or tertiary diagnoses of these conditions. Accordingly, the mixed and null results may reflect a population whose cost and utilization outcomes are difficult to change.

Finally, it is not clear whether comparison practices made any substantive changes to their practices during the study period. For example, some comparison practices may have partnered with behavioral health providers in a way similar to that of practices who participated in the MAPCP demonstration. This type of contamination could attenuate findings.

Conclusions

State- and federal-led initiatives to support PCMH transformation continue at a rapid pace, and many of these initiatives are paying considerable attention to integrating behavioral health care into the PCMH. For example, Medicaid programs can receive technical assistance on behavioral health integration through the Medicaid Innovation Accelerator Program, and some states participating in the CMS-funded State Innovations Models demonstration are focused on behavioral health integration in primary care. However, the challenges to integration are many, and as states and health systems experiment with promising approaches to integrating systems of health care and community-based resources to meet the needs of individuals with behavioral health conditions, policy makers will need to consider what supports, both financial and nonfinancial, may be most effective in assisting providers of general medical care, behavioral health care, and social services in these efforts.