Primary Care–Mental Health Integration in the VA: Shifting Mental Health Services for Common Mental Illnesses to Primary Care

Aiming to improve access to mental health care, the U.S. Department of Veterans Affairs (VA) health care system underwent large-scale dissemination and implementation of team-based primary care models through primary care–mental health integration (PC-MHI) beginning in 2007 (1). This initiative mandated that all large VA primary care clinics needed to have a blended model that included colocated collaborative care and care management (2), components of which are supported by more than 79 randomized controlled trials, such as Improving Mood-Promoting Access to Collaborative Treatment (3–5). PC-MHI embeds nurse care managers and mental health specialists (predominantly psychologists, social workers, and licensed mental health counselors) to provide brief, time-limited, evidence-based treatment directly in primary care (6). In conjunction with the VA’s patient-centered medical home initiative (chiefly, the use of patient-aligned care teams [PACTs]) (7), PC-MHI strives to provide the majority of mental health care for primary care patients with mental health conditions of low to moderate complexity.

Over the past decade, several studies have examined the impact of PC-MHI on mental health care utilization in the VA. The association between PC-MHI and overall mental health services, however, has been mixed and has depended on how PC-MHI is characterized, that is, comparing clinics with or without PC-MHI programs (8) and comparing patients with or without a PC-MHI visit (9). More recently, we examined PC-MHI through a primary care clinic’s level of engagement in such programs by calculating the proportion of primary care patients who received PC-MHI services in a clinic site. Greater clinic PC-MHI engagement was associated with increased use of total VA mental health care and reduction in use of non–primary care based mental health specialty (MHS) services, suggesting substitution of PC-MHI for MHS visits (10).

Because PC-MHI is designated as a lower level service on the spectrum of intensity of VA mental health care, it remains unclear whether decreasing MHS visits by substituting PC-MHI visits reflects good mental health care for primary care patients. PC-MHI services have demonstrated effectiveness for common conditions amenable to primary care–based treatment, such as depression) (11), and are not meant to serve as a gatekeeper or to replace outpatient VA MHS services. In contrast to PC-MHI, MHS services are provided outside of primary care, emphasize chronic care for individuals with serious mental illness, and use a full range of mental health specialists. VA MHS services occur on a continuum as follows: general team-based mental health, specialty outpatient programs, and residential rehabilitation and treatment programs. PC-MHI, however, uses mental health staff in a collaborative model, often with psychiatrist consultation, to treat patients with mild to moderately severe conditions, such as depression, anxiety, and alcohol misuse. It is, therefore, important to rule out whether the reduction in MHS services is indiscriminate or inappropriate management of individuals with serious mental illness in primary care

Our study aims were to understand which types of MHS visits are reduced by increasing clinic PC-MHI engagement over time and which patient subgroups are affected by this reduction. First, we hypothesized that greater PC-MHI engagement among primary care clinics would be associated only with lower use of general MHS visits, without reducing more specialized MHS services (for example, services for individuals with serious mental illness). Second, we hypothesized that patients with depression, a PC-MHI target condition, would receive fewer general MHS services because a primary care clinic engages more fully in PC-MHI. Because patients with psychosis (patients with bipolar disorder and schizophrenia) may be best managed through MHS services, we hypothesized no change in the use of MHS services for these patients in relation to primary care clinic engagement in PC-MHI.

Methods

Study Design and Cohort

To identify mental health care utilization trends, we performed a retrospective longitudinal cohort study from fiscal year (FY) 2009 to FY 2013 (October 1, 2008–September 30, 2013). As detailed in previous analyses (10), eligible patients visited one of 29 primary care practices (four hospital-based, 25 community-based) in Southern California at least twice in the initial year and were assigned a home clinic on the basis of where the majority of their primary care visits were received. We included patients (N=66,638) with known mental health needs and with one or more of the following diagnoses on at least two separate encounters during the study period: alcohol use disorders, drug use disorders, depression, bipolar disorder, schizophrenia, posttraumatic stress disorder (PTSD), personality disorders, or anxiety and other psychiatric disorders.

Measures

Primary outcomes.

We used nationally designated electronic encounter codes from the National Patient Care Database to subdivide outpatient VA mental health care as follows: PC-MHI, general MHS, and more specialized MHS care. [A brief description of VA electronic encounter codes used to identify MHS visits is available as an online supplement to this article.] Because our study focused on understanding the reduction of non-primary care–based MHS services, we did not examine PC-MHI services as an outcome. Our primary outcomes included the relative rates of patient MHS visits, subdivided into general MHS services (those provided by general mental–health team based care providers) and more specialized MHS services (specialty outpatient programs, residential rehabilitation, and treatment programs). We analyzed more specialized MHS visits pooled together and separately as follows: serious mental illness, homelessness, substance use disorder, PTSD, day hospitalization, rehabilitation, primary care in mental health, geriatric, home based, sexual trauma, and chaplain services.

Primary predictor.

As previously described (10), our main predictor was clinic PC-MHI engagement, which we defined as the number of PC-MHI service users divided by the number of primary care patients in each clinic in each year. Clinic PC-MHI engagement is conceptually like PC-MHI penetration rate, which is a national VA performance metric; however, our measure relates only to Southern California VA patients and is calculated irrespective of whether a clinic was required to have a significant level of PC-MHI services. We dichotomized clinics by whether they fell above or below the baseline (FY 2009) median PC-MHI engagement rate of 1.1% (range 0%–15.9%), which we labeled as “high PC-MHI” versus “low” in descriptive analyses, but we used clinic PC-MHI engagement as a continuous variable in regression analyses, hypothesizing that there might be a linear relationship.

Covariates.

We examined utilization-related patient characteristics, including age, gender, race-ethnicity, marital status, health insurance, income proxies (because patients may be eligible for VA care on the basis of a means test or service-connected disability), homelessness, and distance from home address to home clinic. Using ICD-9 diagnostic codes, we adjusted for each patient’s mental health conditions and calculated each patient’s Charlson comorbidity index by using the Deyo-Quan approach for each patient in each year (12,13). Finally, we created an indicator based on year of clinic participation in an evidence-based quality improvement (EBQI) intervention to enhance PACT adoption (EBQI-PACT), which was previously found to affect health care utilization (14,15).

Analysis

In descriptive analyses, we examined the relative proportions of psychiatric diagnoses among study patients in all clinics for each study year. Next, we examined the proportion of psychiatric diagnoses among patient users of different types of mental health care over the five-year study period. We used t tests to compare the mean numbers of mental health care visit types by each psychiatric diagnosis. We additionally stratified by low versus high PC-MHI engagement clinic category and used chi-square tests to compare the proportions of psychiatric diagnoses by low versus high PC-MHI engagement. Finally, we used unadjusted regression models with year fixed effects to estimate the relationship between clinic PC-MHI engagement and mental health care utilization outcomes across all study years.

In multivariable analyses, we estimated the effect of clinic PC-MHI engagement on mental health care utilization outcomes for all study patients after adjusting for year and clinic fixed effects, another clinic intervention (EBQI-PACT participation), and utilization-related patient characteristics. Then, we separately analyzed patients with depression (N=37,616) and patients with psychosis (bipolar disorder and schizophrenia; N=7,662). Additionally, we analyzed a pooled sample of both patient subgroups with variables for diagnostic type (depression or psychosis) to determine whether either diagnosis had an interactive effect with clinic PC-MHI engagement. We included year and clinic fixed effects to account for secular trends and invariant clinic characteristics and included patient random effects because of having multiple nonindependent observations per patient over the five study years. Additionally, we adjusted standard errors for clustering of patients within clinics (16). To account for overdispersion in the distributions of our health care utilization outcomes, we used multilevel negative binomial regression in unadjusted and adjusted models and reported incidence rate ratios.

In sensitivity analyses, we stratified patients by whether their clinics were required to have PC-MHI programs (mandated if clinics have 5,000 or more patients per year), excluded patients who were ages 65 and older and eligible for Medicare coverage (N=21,569), and excluded patients who had no visits during the final study year (because they left VA care) and patients who died during the study period (N=13,499). In additional analyses, we examined whether clinic PC-MHI engagement affected VA–directly provided mental health care costs, which we obtained from the Decision Support System files. We used log-transformed costs in a multilevel linear regression model to account for the skewed distribution of mental health care costs in each year. For all models, we determined significance by using a two-tailed alpha of .05 and analyzed data in Stata, version 14.0. The VA Greater Los Angeles Institutional Review Board approved this study (PCC-2013-101432).

Results

Unadjusted Analyses

Table 1 depicts the proportion of psychiatric diagnoses among study patients in all clinics during the baseline study year; the pattern remained similar over five years. Consistently, the most common psychiatric diagnoses given to veterans were depression (35%), anxiety (36%), and PTSD (22%). Our patients, on average, had 1.2 psychiatric diagnoses, indicating that comorbid illnesses were common. We found significant differences between patients in clinics with high versus low PC-MHI engagement. There were more substance use disorders (χ²=159, df=1, p<.001), anxiety disorders (χ²=74.9, df=1, p<.001), and schizophrenia (χ²=8.1, df=1, p<.01) diagnosed among patients receiving treatment from high PC-MHI engagement clinics and more depression (χ²=62.3, df=1, p<.001), PTSD (χ²=27.1, df=1, p<.001), and alcohol use disorders (χ²=28.5, df=1, p<.001) diagnosed among patients receiving treatment from low PC-MHI clinics.

When we examined the proportion of psychiatric diagnoses among patients using different types of mental health care over five years, we found that PC-MHI target conditions were more often associated with PC-MHI service users than with general MHS service users (Table 2). PC-MHI service users more often had depression (30.6% vs. 26.7%; p<.001), anxiety (26.6% vs. 23.3%; p<.001), and PTSD (21.4% vs. 20.9%; p<.001) diagnoses, whereas general MHS service users more often had bipolar disorder (5.7% vs. 4.1%; p<.001) and schizophrenia (4.7% vs. 2.5%; p<.001) diagnoses. These differences remained when we stratified results by high versus low clinic PC-MHI engagement.

Despite a decrease in all mental health visits over five years (−19.8%), there was a large increase in PC-MHI visits in descriptive analyses (fivefold). Notably, general MHS visits decreased in high PC-MHI engagement clinics (−6.1%) and increased in low PC-MHI clinics (37.3%). In unadjusted analyses, an increase of 1 percentage point in a clinic’s PC-MHI engagement rate was associated with a 1.5% lower rate for general MHS visits (95% confidence interval [CI]=−1.6 to −1.4, p≤.001) and a 2.2% lower rate for more specialized MHS visits (CI=−2.5 to −2.0, p≤.001) per year for patients treated at that clinic.

Adjusted Analyses

We found a reduction in effect size of PC-MHI on outcomes in adjusted analyses compared with unadjusted analyses, likely because of significant differences we previously identified in high versus low PC-MHI clinics and their patients (10). Thus, after controlling for patient- and clinic-level factors, we found that an increase of 1 percentage point in a clinic’s PC-MHI engagement rate was associated with a 1.2% lower general MHS visit rate (CI=−2.0 to −.4, p<.001) per year. However, there was no evidence of a PC-MHI effect on more specialized MHS visit outcomes (difference=.3%; CI=−.8 to 1.5, p=.58), pooled together or analyzed separately (Figure 1).

Additionally, we found a decreasing trend in general MHS visits over time (data not shown). Several patient characteristics were associated with lower general MHS utilization: older age, male gender, black race, Hispanic ethnicity, and longer distance to clinic. Others were associated with higher general MHS utilization: single, having a service-connected disability, being homeless, having multiple chronic comorbid conditions, and having any psychiatric diagnosis. Covariate results were similar for more specialized MHS visits, except that male gender, black race, and Hispanic ethnicity positively predicted use, and clinic participation in EBQI-PACT negatively predicted use.

When we stratified patients by psychiatric diagnoses, we found differences between patients with depression and patients with psychosis (bipolar disorder and schizophrenia). We observed an association between clinic PC-MHI engagement and general MHS utilization when we restricted our sample to patients with depression, which, in most cases, is a condition ideally treated in PC-MHI (difference=−1.1%; CI=−1.8 to –.4, p=.01). This association was not present when we restricted our sample to patients with psychosis, which are conditions that generally require management through MHS services (difference=−.4%; CI=−1.1 to .3, p=.22; Figure 1). The difference between patient subgroups in use of general MHS services, however, was not statistically significant when we included an interactive effect with clinic PC-MHI engagement.

Sensitivity and Additional Analyses

We saw some indications of a difference between patients who belonged to clinics that were required or not required to have PC-MHI, whereby the association between PC-MHI engagement and general MHS visits existed for PC-MHI nonmandated clinics (difference=−1.5%; CI=−2.6 to −.3, p=.01) and not for PC-MHI mandated clinics. However, this difference between clinic type was not significant when we included an interactive effect with clinic PC-MHI engagement.

Multivariable analyses yielded similar estimates when we excluded patients ages 65 and older, patients who left VA care, and patients who died before the end of the study period (data not shown). We observed a small but nonsignificant reduction in mental health costs per year associated with increasing clinic PC-MHI engagement (β=−.20, standard error=.23, p=.387).

Discussion

In our study, we observed increasing clinic engagement in PC-MHI services over time, accompanied by a reduction in general MHS visits but no change in more specialized MHS services. There appears to be a shift in location where veterans receive mental health care (i.e., from general MHS to primary care), which may indicate achievement of PC-MHI goals (e.g., collaborative care) or may be the product of reduced MHS availability (e.g., relative loss or reassignment of MHS staff in the VA). In either scenario, PC-MHI appears to be filling a need for mental health expertise in the primary care population. An earlier VA study similarly concluded that usage of PC-MHI programs has expanded in the VA and that these services were reaching patients new to the VA or with a gap in health care utilization (17). Primary care is often the de facto provider of mental health care (18,19), and other health care delivery systems have similarly attempted to integrate primary care and mental health (20,21). As such, these observed changes in mental health care utilization may also apply beyond the VA.

PC-MHI targets the most common mental illnesses (e.g., depression) faced by VA primary care patients. There is ample evidence to support integrated mental health services for mild to moderate depression and anxiety in primary care (3). Team-based primary care models, such as PC-MHI and PACT, aim to provide the majority of the mental health care needed by primary care patients with mental health conditions of low to moderate complexity (6). Our findings suggest that those patients are, indeed, receiving PC-MHI services, especially in clinics that more highly engage in PC-MHI. Future research should examine what is a suitable role for PC-MHI in the care of individuals with serious mental illness.

Our study adds to the literature on dissemination of integrated care, for which widespread implementation has been slow despite a strong evidence base (20,22). Furthermore, it remains novel to analyze service utilization in relation to clinics’ PC-MHI engagement (10). Although administrative data sources do not necessarily contain information on fidelity to PC-MHI implementation, this metric may still contribute to the development of meaningful and valid quality measures to monitor practice-level variation in integrated care implementation (23). Because the Centers for Medicare and Medicaid Services now provide financial support for integrated care (i.e., same day billing for mental health and primary care services), it may become more commonplace for all health systems to use existing administrative data to determine clinic engagement in such care models (24).

Our study had several potential limitations. First, our cohort study capitalized on the longitudinal variation in clinic PC-MHI engagement to understand panel utilization and costs trends, but it is limited by not accounting for patients who dropped out or switched clinics. The study population did not include additional primary care enrollees of these clinics who were not part of the longitudinal cohort. Second, our analyses did not control for several factors related to PC-MHI (e.g., mental health care staffing patterns, PC-MHI implementation fidelity, primary care physician provided mental health care) that may affect health care utilization and cost. Incorporating newly available data on mental health care staffing or surveying clinics about PC-MHI implementation fidelity may be areas for future research. Third, administrative data can have problems with coding inaccuracies, which was observed in VA data for mental health telephone visits. Finally, our study’s generalizability is limited to patient health care utilization patterns in Southern California’s VA health care systems. More research is needed to understand PC-MHI effects on clinical quality of mental health care and whether such findings can be broadly replicated in other health care systems.

Conclusions

Although patients with comorbid mental and general medical illnesses are frequent users of VA health care, their mental health needs may be treated in primary care through PC-MHI services rather than in specialty care. Our study found that increasing clinic PC-MHI engagement was associated with significant decreases in general MHS visits but not in more specialized MHS visits. PC-MHI programs appear to reduce reliance on general MHS clinics and thus may be effective in engaging those reluctant to seek mental health care. PC-MHI services in the VA appear to be focused on more “bread and butter” mental health conditions and are used more heavily by patients with depression than with psychosis. Implemented appropriately, these programs can help ensure that primary care is providing evidence-based stepped care and that patients with more intensive mental health concerns (e.g., psychosis) are appropriately referred to specialty services. Future research should examine outcomes involving clinical quality indicators for mental health care, expand on quality indicators for PC-MHI, and clarify the role of PC-MHI services for individuals with serious mental illness.