Innovations in Care Delivery for Patients With Serious Mental Illness Among Accountable Care Organizations

We conducted a cross-sectional analysis of responses to the 2017–2018 National Survey of Accountable Care Organizations (NSACO). We fielded the 2017–2018 NSACO from July 2017 to March 2018. The survey included questions about ACO contract and organizational characteristics and also specific mental health care delivery capabilities (25). We invited participants by using both Web and paper invitations to improve response rate; among all respondents, most completed the survey online (80%). The Dartmouth College Committee for the Protection of Human Subjects approved the survey protocol and communication with potential respondents, including informed consent.

Survey participants included leaders of organizations participating in ACO contracts, such as the organization’s chief executive or population health administrator. We defined ACO contracts as contracts that hold organizations responsible for the total cost of care and quality of care for a designated patient population, consistent with previous survey waves and the Centers for Medicare and Medicaid Services (CMS) rules (26, 27). We constructed the sampling frame by using publicly available lists of Medicare ACO participants combined with ACO registries curated by consultant groups specialized in tracking ACO activity (Leavitt Partners and Salesforce). We included all ACOs participating in 2017–2018 contracts, excluding dialysis treatment organizations, organizations without contact information, organizations not meeting our definition of an ACO (ineligible), and pilot organizations surveyed during survey development. Of an estimated 862 eligible organizations, 55% returned a survey and 49% completed the survey (a figure in an online supplement to this article provides further details).

Our primary outcome was self-reported use of strategies to integrate primary care and treatment for serious mental illness endorsed in response to this survey question: “Do any providers in your ACO use the following strategies to integrate primary care and treatment for serious mental illness, like schizophrenia or bipolar disorder?” Response options included: 1) “A care manager to ‘address general medical health treatment coordination’ OR to ‘address non-medical needs (e.g., job support, housing),’” 2) “Patient registries to track general medical health conditions (e.g., hypertension, glucose control),” and 3) “A co-located primary care clinician seeing patients in specialty mental health setting” (see online supplement for additional survey details). Because the term ACO could refer to contracts, an organization as a whole, or an entity that is part of a larger organization, we asked respondents to “respond on behalf of the organization most commonly participating in ACO contracts.”

Self-reported contract and organizational characteristics asked in the 2017–2018 NSACO were used as covariates (see online supplement). Contract characteristics included whether the respondent had an ACO contract with Medicare, Medicaid, or a commercial payer; whether the respondent’s ACO shared financial risk (i.e., took on downside risk) in its Medicare, its largest Medicaid, or its largest commercial contract; and whether the respondent’s ACO had prior experience with risk-based contracting. We also measured contract features specific to mental health services in respondents’ non-Medicare contracts; unlike Medicare contracts, commercial and Medicaid ACO contracts may exclude mental health services. We asked whether a respondent’s largest Medicaid contract—or, separately, largest commercial ACO contract—included mental health services in the total cost-of-care calculation (the spending benchmark), any behavioral health quality measures, quality measures specific to serious mental illness (e.g., adherence to antipsychotics for schizophrenia), or a “carve-out” of mental health services by contracting out care to a managed behavioral health care organization. Organizational characteristics included leadership of a respondent’s largest ACO contract (physician led versus not), size (quartile of number of physician full-time equivalents in largest contract), and types of providers included in the largest ACO contract (academic medical centers, public hospitals, federally qualified health centers, and specialty behavioral health providers). Finally, we asked respondents whether any providers in the ACO used any of three collaborative care strategies to integrate primary care and treatment for depression or anxiety (care management, consulting mental health provider, or patient registry; coded as a count of 0, 1–2, or 3 collaborative care strategies).

Each integrated care delivery strategy was measured as a 0,1 binary outcome, where 1 indicated that the respondent reported using this strategy at least once in at least one location to care for patients with serious mental illness.

To form outcomes for analysis, we created a vector of care integration strategies (care manager, patient registry, and colocated primary care clinician) “stacked” by strategy at the ACO level. For example, for respondents reporting using a care manager only, this vector would equal 1,0,0. We regressed this stacked outcome on all ACO characteristics and strategy type (28). We also included interaction terms between each ACO characteristic and strategy type indicator to test whether the association between ACO characteristics and use of a care integration strategy differed by strategy type. To estimate the relationship of interest, the association between the marginal (i.e., unconditional) probability of each care integration strategy and the ACO characteristics, we fit the model by using generalized estimating equations, specifying a logit link function, an unstructured correlation matrix, and robust standard errors (29, 30). This approach—an extension of standard logistic regression—permitted regression coefficients on ACO characteristics to vary by care integration strategy, an estimated correlation matrix between care strategies, and statistical inferences accounting for any statistical dependencies between the outcomes for the same ACO. (Additional details explaining our rationale and full model results are available in the online supplement.)

To more easily interpret effects, we transformed log-odds into the expected average change in probability of implementing each integrated care strategy studied, holding other variables at their observed values (average marginal effects). We considered p<0.05 statistically significant.

We used multiple imputation to account for missing covariate data from 48 respondents (12%) (estimates below use imputed data; for complete case results, see online supplement). We included all covariates from our analytic model in the imputation model and specified 10 data sets across which to pool final estimates.

Of ACO respondents, 399 answered the survey questions about the three strategies used to integrate primary care and treatment for serious mental illness. Most respondents contracted with Medicare (84%) and commercial payers (72%), while just a quarter had an ACO arrangement with Medicaid (24%) (Table 1). Most respondents reported having prior experience with risk-based contracts (64%), but a minority (37%) reported sharing financial risk in at least one contract (Medicare contract, the largest Medicaid contract, or the largest commercial ACO contract). Few respondents (21%) reported that the largest Medicaid or largest commercial ACO contract included a financial carve-out of mental health services to managed behavioral health care organizations. Less than half of respondents (43%)—including respondents that did and did not report having a financial carve-out—reported that the largest Medicaid or largest commercial ACO contract included mental health services in the total cost of care calculation (the financial benchmark). A third of respondents (34%) reported that the largest Medicaid or largest commercial ACO contract held them accountable for any behavioral health care quality measures, and just 18% were subject to performance measures specific to serious mental illness, such as follow-up after mental health hospitalization or antipsychotic adherence in these non-Medicare contracts.

When asked whether any providers in the ACO integrated primary care and treatment for serious mental illness by using a care manager, patient registry, or colocated primary care provider, respondents most commonly reported using a care manager to address physical or social needs (N=281, 70%; Figure 1). Approximately one-third of respondents reported using a patient registry to track general medical conditions (N=146, 37%) or having a colocated primary care clinician (N=118, 30%; Figure 1). Approximately one-quarter of respondents (N=107, 27%) relied exclusively on care managers to integrate services, another quarter reported using a combination of a care manager and either a patient registry or colocated primary care clinician (N=109, 27%), and 16% (N=65) of respondents reported using all three strategies. Nearly a quarter of respondents (N=96, 24%) reported using no strategies.

ACO respondents subject to quality measures relevant for populations with serious mental illness in their largest commercial or largest Medicaid contract were 12 percentage points, or 20%, more likely to use patient registries to track physical health conditions (95% confidence interval [CI]=−3, 27; Figure 2). The association between behavioral health quality measures and care integration strategy use did not differ significantly for care managers or colocated primary care clinicians relative to patient registry (Figure 2).

Among organizational characteristics that we measured, only one—use of collaborative care for patients with depression or anxiety—had a consistently positive statistically significant association with the likelihood of use of each of the three integrated care strategies for patients with serious mental illness. After the analysis adjusted for ACO contract and organizational characteristics, respondents implementing one or two collaborative care strategies were 22 percentage points (95% CI=12, 31), or 80%, more likely to report using patient registries, compared with respondents using no collaborative care strategies, with similar associations for care managers or primary care colocation. Respondents implementing all three collaborative care strategies were 73 percentage points (95% CI=61, 84) more likely to report using any integrated care delivery strategy for their patients with serious mental illness, suggesting that respondents fully implementing collaborative care were well positioned to integrate care for serious mental illness as well. Finally, physician-led ACOs were 19 percentage points (95% CI=7, 31) more likely to report having primary care colocation. However, ACOs that were physician led were no more likely than other ACOs to adopt care managers or patient registries.