Behavioral Health Integration in Community Health Centers and Emergency Department Use

Integration of behavioral health services with primary care is a core component of the Medicaid Transformation Program (MTP) in Washington State, a 5-year agreement funded through a Section 1115 waiver that provides up to $1.5 billion of federal investment for regional health system transformation projects. Beginning in 2017, Washington embarked on an ambitious scope of reforms, including merging financing for general medical and behavioral health services under comprehensive Medicaid managed care organizations and promoting delivery system change through independent regional accountable communities of health (ACHs). Among other responsibilities, ACHs were tasked with collecting data from practices and reporting progress toward benchmarks on a variety of measures related to integrated care and patient outcomes. A priority outcome incentivized by the state was a reduction in ED utilization. To accomplish this goal, ACHs partnered with community health centers to provide technical support and distribute demonstration funds to build capacity for integrated care under the state demonstration.

To better understand the potential to improve ED use and other outcomes for the Medicaid population, we linked clinic-level survey data on behavioral health integration with Medicaid claims for a group of community health centers in King County, Washington, to examine the characteristics of highly integrated clinics. We also examined whether greater clinic capacity for behavioral health integration was associated with changes in ED use among enrollees who received their care at these clinics.

The Maine Health Access Foundation (MeHAF) Site Self-Assessment Plus was selected by the Washington ACH to track implementation of behavioral health integration in the evaluation of its MTP. We used clinic responses to the MeHAF for a 6-month performance period (January 2019–June 2019). The MeHAF organizes items into two domains related to services for integrated, patient-centered care and organizational supports for integration, which are rated on a scale of 1–10 for the level of integration achieved (with 1 indicating lower levels and 10 indicating higher levels).

No previous validation of the MeHAF has been published. During preliminary work, our team conducted an exploratory factor analysis and identified two alternative domains that redistributed subsets of questions to characterize behavioral health integration (see Table A1 in an online supplement to this column). These two alternative domains included infrastructure for integrated care (seven questions) and patient and family engagement activities (four questions). The two domains were found to have high internal consistency. Average scores across the items in each domain were calculated, and clinics were divided into terciles based on the scores, reflecting low, average, and high levels of integration for that domain.

Data from adult (18–64 years) Medicaid enrollees who could be attributed to health centers participating in the integrated care demonstration were included in this study. Further details of our cohort derivation and our study design are provided in the online supplement (Figures A1 and A2, respectively). We attributed enrollees to clinics where they received a plurality of their primary care during a 12-month period (July 2018 to June 2019). This period included visits during the MeHAF performance period, as well as an additional 6-month look-back period to ensure adequate capture of health care utilization. An office visit was considered a primary care encounter if it had both a primary care provider taxonomy code and a primary care–related ICD-10 diagnosis or procedure code (5). Next, we counted visits to clinics by using a combination of National Provider Identifier and address on the insurance claim to identify brick-and-mortar sites in a multisite organization.

We identified the proportion of attributed patients who had any ED visit if they had a claim with a revenue code of 045xx, Current Procedural Terminology code 99281–99285, or place of service listed as an ED. Logistic generalized estimating equations (GEEs) were used to model the association between tercile of behavioral health integration and ED use, by adjusting for age, gender, race and ethnicity (grouped into mutually exclusive, self-identified categories including Native American, Asian, Black, Hawaiian/Pacific Islander, White, Latinx/Hispanic, multiple races, and unknown), Gagne comorbidity scores, behavioral health service need, and clinic size. Behavioral health service need was determined through criteria from Washington State, which relies on a combination of diagnostic codes, mental health and substance use disorder service codes, and psychotropic medication prescriptions. GEE models account for clustering within clinics but may produce biased estimates when the number of clusters is small. To account for this possible bias, we applied a small cluster correction to our standard errors. Each domain of integrated care was modeled separately, comparing clinics with average versus low and high versus low levels of integration for all attributed enrollees. We also performed a subgroup analysis examining ED visits among the subset of enrollees with a need for behavioral health treatment.

Our sample included 17,587 Medicaid enrollees attributed to 22 clinics participating in the behavioral health integration demonstration in King County, Washington. Nearly all (91%, N=20) of the clinics were federally qualified health centers serving a mean±SD of 6,594±2,118 enrollees across 23,685±8,966 office visits per year. Clinics scoring in the low, average, and high terciles across both integrated care domains (team-based care and patient and family engagement) were largely similar in the number of patients served, annual office visits, provider staffing, and payer mix (see Table A2 in the online supplement). In general, clinics scoring higher in one domain tended to also report relatively higher scores in the other domain.

We did observe significant differences in case mix among clinics with lower versus higher levels of integrated care (see Table A3 in the online supplement). Notably, enrollees attributed to clinics that scored in the lowest tercile of infrastructure for team-based care were disproportionately more likely to be Black or Native American–Alaska Native and to have a higher burden of comorbid conditions, including behavioral health conditions, than those scoring in the highest tercile. Similar patterns were noted for the patient and family engagement domain, with clinics scoring in the lowest tercile disproportionately serving racial-ethnic minority patients across most groups and those with a higher burden of behavioral health diagnoses. The exception to this were individuals of Asian descent, who were more likely to be seen in clinics reporting higher levels of integration in both domains.

In adjusted analyses (see Table A4 in the online supplement), we did not find an association between the level of behavioral health integration and ED use among enrollees overall. Across clinics with low, average, or high levels of infrastructure for team-based care, 40% (N=1,210 of 2,997), 31.2% (N=3,228 of 10,347), and 33% (N=1,409 of 4,243) of attributed enrollees had an ED visit during the 6-month performance period, respectively, corresponding to an insignificant difference in ED use of −9.2 percentage points (95% confidence interval [CI]=−21.0 to 2.6) for clinics in the average versus low terciles and −7.2 percentage points (95% CI=−19.2 to 4.8) for clinics in the high versus low terciles. Although these differences were statistically significant after adjustments for case mix and clustering across clinics, they became insignificant after we applied the small cluster correction to the standard errors of the estimated difference in ED visits. Across clinics with low, average, or high levels of patient and family engagement, 34.3% (N=3,170 of 9,224), 35% (N=1,155 of 3,263), and 30.6% (N=1,554 of 5,080) of attributed enrollees had an ED visit during the 6-month performance period, respectively. This pattern of visits corresponded to an insignificant difference in ED use of 1.1 percentage points (95% CI=−7.0 to 9.2) for clinics in the average versus low terciles and −3.7 percentage points (95% CI=−11.4 to 4.0) for clinics in the high versus low terciles. We found similar results when we looked at the subset of individuals with a behavioral health service need (see Table A5 in the online supplement).

Although these analyses of behavioral health integration in community health centers early in Washington’s MTP did not reveal significant associations between the studied integrated care domains and ED use, our examination revealed potentially important differences in case mix across clinics that had implications for access to high-quality and coordinated care. Specifically, we noted that clinics reporting lower levels of team-based care and those reporting fewer patient and family engagement activities were more likely to serve patients of color with a higher burden of comorbid conditions, including behavioral health conditions. This finding suggests that certain communities of color, including Medicaid enrollees who are Native American–Alaskan Native or Black, have less access to integrated care in Washington State relative to White enrollees, despite potentially having a higher need for those services.

Although previous work has shown that racial-ethnic minority patients are less likely to receive timely and appropriate behavioral health treatment compared with White patients (6), to date there is a paucity of studies examining racial disparities in access to integrated care. One of the mechanisms cited for disparities in access to behavioral health services is that individuals from racial-ethnic minority groups are more likely to be publicly insured compared with White individuals (7). Medicaid has particularly low rates of acceptance among behavioral health providers (8), and care disparities may therefore arise as a function of available provider networks.

Yet we observed similar disparities within a population of Medicaid enrollees, in which patients from racial-ethnic minority groups are more likely to receive care from community health centers with lower levels of integration. In this regard, our results are consistent with findings of other studies that have shown that racial-ethnic minority groups are more likely to be concentrated around primary care practices that have historically been underresourced and are of lower quality than those serving predominantly White patients (9, 10). This is a critical finding because recent evidence suggests that people from racial-ethnic minority groups experience disproportionate improvements in health outcomes from collaborative care (10), but these benefits would extend only to individuals who have access to these services. It is also important to highlight that the clinics in our sample that had the lowest integration scores also cared for a higher proportion of patients with general medical and behavioral health comorbid conditions who could benefit most from integrated care. Thus, it is very likely that these clinics would also realize greater improvements in care coordination over time if they were offered state and other regional support to build behavioral health integration capacity.

This pilot study had several limitations for deriving conclusions about behavioral health integration in community health centers. First, the measures of behavioral health integration relied on self-reports by clinic staff. Although the MeHAF is intended to be completed by all members of the care team in a consensus-based manner, it is unknown to what extent this occurred. Second, although the MeHAF was derived from validated chronic disease management tools, its utility as a measure of behavioral health integration for outcomes assessment is unknown. Third, we had limited power to detect associations of clinic-level factors (e.g., integration) with outcomes because of the small number of clinics in our sample and despite having a relatively robust sample of attributed enrollees. Fourth, this study was a cross-sectional analysis of the association between clinic-level integration and outcomes and could not assess causal effects. Fifth, enrollees with serious mental illnesses may have been underrepresented in our sample because they may have been less likely to show up for primary care visits, the visit type we used to attribute beneficiaries to a clinic. Finally, we could not identify the subset of attributed patients who received integrated services, including screening, referral, and case management from claims data, even though this would be a group where we would have expected to see improvement in outcomes.

Although we found limited evidence that behavioral health integration was associated with changes in ED use, we did observe evidence for racial-ethnic disparities in access to clinics offering high levels of integrated services. We also found that Medicaid enrollees with the greatest need for integrated services may be disproportionately cared for at clinics with less capacity to offer such services. These findings suggest opportunities for better targeting resources to improve population health that should be considered by state Medicaid policy makers.