Impact of Behavioral Health Homes on Cost and Utilization Outcomes

Individuals with serious mental illness are more likely than persons in the general population to experience negative health outcomes (1–3). The annual economic, indirect cost of serious mental illness in the United States due to loss of productivity and earnings is estimated to be $193 billion (4). The integration of general medical health and wellness in behavioral health treatment settings has been shown to decrease health care disparities for individuals with serious mental illness (5–7); however, little is known about whether integrated care reduces cost to payers or health care utilization in this population.

A recent literature review identified 11 studies conducted between 2000 and 2018 assessing the impact of behavioral health homes (BHHs) (8). Four of these studies assessed cost-related outcomes and yielded mixed findings; however only two were peer reviewed (9, 10). Both of these studies provided important information, but their applicability and generalizability were limited. The randomized controlled trial by Druss et al. (9), an important seminal study, found no significant difference in total cost of care between those receiving integrated care in the offices of community mental health providers and a control group. The applicability of the findings to current policy is limited because the study was published nearly 20 years ago, was based on findings from two U.S. Department of Veteran’s Affairs (VA) clinics, and used VA cost accounting data with a small number of patients (N=120) over the course of a single year. Similarly, Krupski and colleagues (10) did not find a significant decrease in per member per month (PMPM) cost in their study of an approach implemented in mental health safety net clinics with a larger patient sample (N=762). This study included cost data only from two clinics and an affiliated medical center, and the matched comparison cohort had significantly lower behavioral health acuity. Several studies that evaluated health care utilization, including emergency department use, inpatient stays, and receipt of outpatient general medical care, resulted in heterogeneous findings (5, 6, 9–11).

Community Care Behavioral Health Organization (Community Care), a nonprofit behavioral managed care organization that is a subsidiary of UPMC, implemented two BHH approaches, provider-supported care and self-directed care, in behavioral health provider sites (“providers”) across Pennsylvania as part of a comparative effectiveness study funded by the Patient-Centered Outcomes Research Institute (PCORI contract #271). At the completion of the parent study, we conducted a subsequent analysis with the study population and a matched comparison cohort to examine the impact of the interventions on cost and health care utilization and to understand how our findings compare with current evidence.

Methods

Study Setting

Participating community mental health provider sites were within Community Care’s provider network and were chosen on the basis of their rural and suburban locations. Providers could not have previously participated in the implementation of a BHH approach and had to provide case management services. Eleven providers across Pennsylvania were randomly assigned to either provider-supported care (N=5) or self-directed care (N=6). Each provider implemented its assigned BHH approach over a 2-year period (6).

Study Interventions

Figure 1 provides an overview of the common and unique components of each BHH approach. In both approaches, provider staff, namely case managers and peer specialists, were trained in wellness coaching to support patients with identification of and progress toward general health and wellness goals. Interventions were implemented within the context of a culture of wellness in which sites were encouraged to support healthy lifestyles, disease prevention, health education and promotion, and coordination with primary care providers (6). Neither approach required on-site primary care service delivery.

Provider-supported sites employed a full-time on-site registered nurse (i.e., wellness nurse) who provided consultation to patients and wellness coaches. Nurses educated provider staff about common medical conditions and assisted wellness coaches in the development of patient wellness plans. In addition, wellness nurses supported patient access to and utilization of outpatient primary and specialty care (6).

Self-directed sites provided tools and resources to patients to support the adoption of healthy habits and management of chronic conditions. Tools included education about common chronic medical conditions and trackers to support positive sleep hygiene, smoking cessation, improved nutrition, and weight management. Patients could access information via a secure online portal or in a paper toolkit format (6).

Patient Sample

Eligible participants were Medicaid recipients receiving services at participating provider sites, age 21 or older, with a diagnosis of serious mental illness based on the Commonwealth of Pennsylvania’s definition (i.e., major depression, bipolar disorder, and schizophrenia or schizoaffective disorder) (12). The parent study enrolled 1,229 participants—713 across provider-supported sites and 516 across self-directed sites. We obtained informed consent from all individuals participating in the parent study. The study was reviewed and approved by the University of Pittsburgh Institutional Review Board (6).

For this analysis, we identified 859 individuals from our study sample (522 from provider-supported sites and 337 from self-directed sites) who were Medicaid eligible at some point in each of the 3 analysis years and who received case management services consistently from the same provider. Those not meeting these criteria dropped from Medicaid eligibility for 1 or more study years, switched the provider from which they were receiving case management services during the study period, or both. Those not included in the study for these reasons were slightly older and had modestly higher PMPM costs. A comparable pattern was found for members not in the sample in the larger population receiving services from other nonparticipating providers who lost Medicaid eligibility or switched providers. Of individuals receiving case management services, those who remained eligible for Medicaid and received services consistently from the same provider were slightly younger and had lower PMPM spending, compared with those who lost eligibility or switched providers over the 2-year study period.

Group composition of the provider-supported approach was as follows: total, 522; males, 213; females, 309, mean age, 45.1 (comparison cohort: total, 522; males, 213; females, 309; mean age, 43.2). Group composition of the self-directed approach was as follows: total, 337; males, 111; females, 226; mean age, 43.6 (comparison cohort: total, 337; males, 112; females, 225; mean age, 42.0).

We used historical claims data to identify comparison cohorts of individuals with characteristics similar to those of participants in the parent study, including serious mental illness diagnosis and receipt of case management services. The matched cohorts were identified from Community Care’s members in the same Medicaid program who were not receiving services at provider sites participating in the parent study. We matched with replacement using an iterative algorithm that considered serious mental illness diagnoses, total PMPM costs, continuity of Medicaid enrollment, age, gender, and PMPMs for the service categories included in this analysis. Although the Medicaid enrollment data included a race-ethnicity indicator, the completeness and accuracy of this information were of uncertain reliability; thus, we did not include it in our evaluation. Both cohorts were matched by using data only from the study’s baseline year (2013). Assessment of the comparison cohort’s health risk profiles relative to those of the two treatment groups with use of the Chronic Illness and Disability Payment System (13) determined that the comorbidity burden and acuity of the treatment and control groups were highly similar.

Data and Measures

We analyzed Medicaid total cost of care, including behavioral health, general medical, and pharmacy claims encounter data from calendar years 2013–2015 for the intervention and comparison cohorts. The claims cost to payers as the cost outcome measure is a significant advance over prior studies. Preimplementation occurred in calendar year 2013 (baseline), and 2014 (year 1) and 2015 (year 2) served as implementation measurement years.

Outcomes included service utilization rates (annual average of the monthly percentage of members utilizing services) and PMPM costs, with both measures calculated for behavioral health, general medical health, and combined spending; medical inpatient and behavioral health; emergency department use; prescription drugs; and total case management services. General medical claims did not contain payment amounts; thus, costs were calculated by applying a publicly available standardized pricing algorithm calibrated against Pennsylvania Medicaid unit costs. This algorithm was applied to general medical encounter data and behavioral health claims data for all the pre- and postimplementation periods. The National Quality Forum endorsed the algorithm’s integrity and validity in standardizing and measuring cost of care (14). Full payment information was available for a subset of the members whose claims costs allowed us to verify that there was no systematic bias introduced by the algorithm, which had the additional advantage of not being influenced by unrelated shifts in specific provider-contracted rates. Cost measures were not adjusted for inflation over the study years.

Statistical Methods

Three comparative analyses were conducted using Stata11 (15). First, in the primary analysis, changes in total claim spending were compared for each treatment arm (self-directed care and provider-supported care) to the comparison cohort matched to each approach’s subpopulation. Second, the impact of the approaches was confirmed and further illuminated by comparing each approach to its own baseline year. These two analyses estimated whether the BHH approaches had a statistically significant impact relative to an estimated counterfactual (i.e., comparison cohort and difference-in-difference for the first and second analyses, respectively). Third, to more clearly estimate and display the statistical significance of differences in impact between provider-supported and self-directed sites (i.e., which treatment arm worked “better”), the difference between these two arms was assessed directly for magnitude and statistical significance.

Analyses were conducted at the level of the individual service user. Dependent variables included annual average monthly utilization (i.e., percentage of members accessing service) and annual average PMPM expenditures, overall and for each service category analyzed. Average annual PMPM (i.e., total expenditure/total eligible member months) was calculated for service users in each of the 3 analysis years.

For most measures with service-specific categories, a majority of recipients reported no spending; thus, a two-part model for cost outcome variables was used to assess total costs and for each service use category. In the two-part model, the first stage predicted whether any cost would be incurred (and so included observations with both zero and nonzero cost), and the second stage predicted the level of cost given that cost had been incurred (and so included only observations with nonzero cost) (16). Logistic regression was used in the first model, and the second used generalized linear models with a gamma distribution and log link, which was a more suitable fit for handling skewness. The analysis controlled for age, gender, and baseline outcome values to guard against chance of imbalance between treatment arms. Difference-in-differences estimation by linear regression was used for comparisons to the baseline period within treatment arms.

Results were calculated and are presented in percentage change to clearly demonstrate the relative sizes of the various effects analyzed. Percentage change was calculated from the parameters of the statistical models. Using total cost in year 2 for provider-supported sites compared with their own historical baseline as an outcome example, the calculation for the percentage change was (year 2 PMPM–baseline year PMPM)/(baseline year PMPM).

Results

The narrative below provides an overview of each analysis, describing results significant at least at the level of p<0.05, and in most cases p<0.01. Table 1 presents summary results for each of the three analyses, including year-2 impacts, the arm-to-arm comparison, significance levels, and sample characteristics. (More detailed tables displaying absolute values and calculations of percentage changes for each of the three analyses are available in an online supplement.)

Each BHH Approach Versus Comparison Cohort

Provider-supported care.

The first and primary analysis compared each approach with its own comparison cohort over the same 2-year period. Relative to the comparison cohort, the provider-supported cohort incurred a 15% decrease in overall PMPM cost and a 28% reduction in behavioral health PMPM cost at the end of year 2. In addition, it incurred a 36% reduction in use of inpatient services and a 23% reduction in total inpatient cost, which was likely driven by the 41% reduction in use of inpatient behavioral health services and 49% reduction in associated behavioral health spending. Provider-supported care was also correlated with a reduced percentage of prescription drug use (30%) and lower total case management spending (17%). At the same time, significant increases were noted in access to general medical and behavioral health services (43% and 14%, respectively), with no significant change in general medical spending and a reduction in behavioral health spending of 28%.

Self-directed care.

The self-directed cohort experienced a significant 26% decrease in total PMPM costs and a 22% reduction in general medical cost at the end of year 2, while increasing use of general medical services by 29% and behavioral health services by 15%. Self-directed care was associated with significant reductions in overall inpatient service use (34%) and cost (44%) as well as reductions in prescription drug use (26%) and cost (23%). Although downward trends were observed for inpatient behavioral health costs, this finding was not statistically significant.

Each BHH Approach Versus Baseline

Given the large increases in utilization of general medical services and the large reductions in cost in the primary results discussed above, it is helpful to analyze how each measure shifted relative to the year prior to BHH. The findings indicate that by year 2 many cost measures showed a large absolute drop relative to baseline. In the treatment arms, costs for many categories fell below baseline levels by year 2, whereas in the comparison cohorts discussed in the previous section, costs grew marginally at year 2. Therefore, the cost difference between the treatment arms and the comparison cohort was larger in absolute value than the cost difference between the treatment arms and baseline, which, by definition, had zero growth.

Provider-supported care.

Participants in the provider-supported approach had a statistically significant 12% reduction in overall PMPM spending postimplementation, compared with baseline, primarily driven by a significant reduction in behavioral health spending of 25%. We observed a 16% increase in overall use of general medical services. At the same time, a significant 32% reduction was seen in total inpatient utilization (behavioral and general medical), as well as medical inpatient utilization. Percentage of members with prescription drug use decreased significantly by 13%, but the PMPM spending on prescription drugs increased by 27%, suggesting a shift in the mix of medications. A significant downward trend was shown in total case management use (19%) and cost (25%).

Self-directed care.

The self-directed cohort also experienced a significant decrease in overall PMPM spending (8%), overall inpatient utilization (28%), medical inpatient utilization (32%), and inpatient behavioral health utilization (40%). We also observed significant reductions in prescription drug use (8%) and higher general medical services utilization (9%). In contrast to provider-supported care, self-directed care corresponded with reduced emergency department use (14%) and an increase (15%) in total case management cost.

Provider-Supported Versus Self-Directed Approaches

In the first two analyses, each arm was compared with its own comparator (comparison cohort in the first analysis and baseline year in the second analysis). In the third analysis, we observed modest differences in the impact of the provider-supported and self-directed approaches. Because the inclusion of a wellness nurse increased the cost of implementing the provider-supported approach, we conducted a direct comparison of the trial arms to provide an explicit test of the statistical significance of any differences in impact between the two arms observed in the analyses described above.

Relative to self-directed care, provider-supported care had regression parameters suggesting a more pronounced impact for most measures. Most measures were not statistically significant, with a few exceptions, including lower overall inpatient utilization (23%) and use of medical inpatient services (28%) and lower incurred cost for medical inpatient utilization (26%) and total case management (33%).

Again, because each intervention displayed significant changes in cost and utilization outcomes relative to both its own baseline period and to its matched comparison cohort (i.e., both interventions shifted utilization and costs in a similar manner), findings for our comparison of provider-supported to self-directed approaches tended to be less statistically significant.

Discussion

Existing literature underscores the value of health homes in supporting health care systems as they continue to work toward achieving the “triple aim” of improved care quality, cost, and patient outcomes (9, 17). To date, the potential of BHHs to reduce costs has not been conclusively supported (8).

Both provider-supported care and self-directed care resulted in a higher proportion of members utilizing general medical services, compared with those receiving services in nonintervention sites. Higher rates observed for the provider-supported approach may be driven by the contributions of the on-site wellness nurse, who focused on direct collaboration and communication with general medical providers around management of patients’ chronic diseases. This may have led to timely and consistent use of general medical care by patients. Nonetheless, the observed increase in use of general medical services for those receiving services at self-directed sites suggests that in cases where resources may not be available for on-site nursing staff, providers can still play a role in enhancing important care linkages. Provider-supported and self-directed approaches provide two BHH options of variable intensity to help facilitate coordinated, holistic care.

The implementation of patient-centered medical homes has resulted in successes in enhancing the spectrum and coordination of services available to members in primary care settings and improving patient outcomes (18–20). In the context of mixed findings in the literature (5, 9–11, 21, 22), our study provides further evidence that medical homes may be advantageous for individuals with complex conditions who are receiving care at community mental health provider sites. In addition to highlighting the need for expansion of the medical home approaches in behavioral health treatment settings, our findings highlight the importance of implementing and evaluating similar efforts to address outcomes for individuals with other chronic illnesses who have primary clinical connections to a specialty setting rather than with their primary care physician.

This study adds to the literature by showing that a BHH approach with modest changes in staffing that does not require the addition of on-site primary care provision can make efficient use of resources by reducing the total cost of care to payers. These improvements were large and statistically significant. Furthermore, outcomes were measured over a 2-year period, allowing the program to take root and develop its potential for impact, which is an additional contribution to current knowledge regarding this important public policy issue.

This study had some notable limitations. Because randomization occurred at the provider level, there may have been differences in providers or in implementation regions for which we could not control; however, safeguards related to statistical methods and control variables should have reduced any such effects. In addition, the available population for the comparison cohort came from more urban geographic areas, in contrast to the more rural and suburban locations of the study sites. One result of this limitation in available comparison data was that we were not able to match emergency department utilization for the intervention versus comparison cohorts in the first analysis. However, the similarity in magnitude and significance of the results of the matched cohort comparison analysis with the approach-versus-baseline results suggests that this limitation did not have a large effect.

Conclusions

Our analysis of two BHH approaches revealed a significant reduction in total cost of care to payers, a reduction in use of general inpatient services, and an increase in use of outpatient general medical services. Our prior research assessing the implementation of provider-supported and self-directed approaches demonstrated improved patient outcomes (6). This study showed that implementation of health homes in community mental health provider settings may also produce health care savings and more efficient health care utilization. Similar approaches may be advantageous for individuals with other complex chronic conditions.