Implementation of Collaborative Depression Management at Community-Based Primary Care Clinics: An Evaluation

A robust literature has emerged supporting the value of quality improvement initiatives, such as collaborative care models, that provide integrated care within primary care settings for common mental disorders such as depressive and anxiety disorders (1–6). The Improving Mood-Promoting Access to Collaborative Treatment (IMPACT) trial, one exemplar of collaborative care, demonstrated significant improvement in depression outcomes among elderly patients and cost savings in the long term (3). These promising results have engendered considerable national interest in the dissemination of integrated models and in the corresponding need to understand how to implement them in real-world settings (7,8). Presently, little is known about the care provided or the outcomes achieved when such models are adopted by community-based practices, particularly those in resource-poor settings.

To address this gap in knowledge, the Hogg Foundation for Mental Health sponsored a demonstration project to gain understanding of the process of implementing a collaborative care model at community health centers. In a separate report that was based on findings from the project, we found that patients who were older, more depressed, and more anxious were more likely to receive high-quality care and that patients with Spanish language preference and lower anxiety had the best outcomes (unpublished data, Bauer AM, Azzone V, Goldman HH, et al, 2011). In this study, we analyzed the role of site on specific process measures of depression treatment (early follow-up and appropriate pharmacotherapy) and on clinical outcomes (depression improvement and remission).

Methods

We conducted a quantitative evaluation of quality indicators and patient outcomes at six community health organizations. Our analysis was also informed by qualitative observations derived from the extensive contact of the investigators with the organizations, including loosely structured site visits. Although the latter component did not qualify as a rigorous mixed-methods approach, these observations informed the interpretation of our quantitative findings. The institutional review board at Harvard Medical School approved this evaluation.

Organizations

Six community health organizations in Texas treating predominantly uninsured patients were selected through a competitive review process in response to a request for proposals issued by the Hogg Foundation for Mental Health. Key selection criteria included the organizations' demonstrated grasp of the collaborative care model, their preparedness to implement the model, and their ability to demonstrate their organization's investment and commitment to participation and sustainability after the grant period. Because one organization had two participating clinics, data from a total of seven clinic sites were analyzed. [A table describing the organizations in detail is available in an online appendix to this report at ps.psychiatryonline.org.]

Patients and intervention

Patients aged 18 and older were identified by their primary care teams as having symptoms of depression and were enrolled between 2006 and 2009. Organizations adopted a collaborative care model that included the key elements of using a Web-based disease registry to track patients, care management to support treatment by primary care providers and provide proactive follow-up, and organized psychiatric consultation. The care manager conducted initial assessments of patients with standardized instruments, described below, provided patient education, tracked patients' treatment response longitudinally at the clinic and by telephone with standardized instruments, coordinated medication management with primary care providers, and assisted with referrals to psychotherapy or other specialty care as indicated.

Most sites had at least one care manager who provided services in Spanish and employed licensed care managers or separate licensed therapists to provide brief cognitive-behavioral therapy. The consulting psychiatrist provided weekly supervision to care managers either in person or by telephone, reviewed treatment plans, performed consultations for patients who had not improved, and was available by telephone to answer primary care providers' questions. The disease registry included prompts to assist care managers to identify patients needing follow-up and tools to facilitate identification of patients who had not responded to treatment (9). Primary care providers received training in evidence-based depression management, as described below, and were responsible for prescribing psychotropic medications and maintaining primary responsibility for patients' care.

Training and technical assistance

A technical assistance team from the University of Washington provided training for care managers, consulting psychiatrists, and primary care providers as well as access to the registry and logistical support for implementing the collaborative care program through annual in-person training and monthly conference calls. During a kickoff meeting, the University of Washington team provided the sites with an orientation to the collaborative care model. Following the meeting, staff members at the sites began making individual monthly calls to a University of Washington coach. The calls covered clinical and operational and program issues, such as engaging hard-to-reach patients. Annual in-person training included full-group training sessions on such program issues as triage of patients as well as separate sessions by site or by team role.

Training sessions for care managers covered such topics as managing caseloads and communicating with primary care providers. Training sessions for psychiatrists and primary care providers covered such topics as evidence-based depression treatment and managing treatment-resistant depression. All sites had access to the same training and technical assistance resources, except that organization 6 started the grant program approximately a year later and missed the kickoff meeting and the first year of coaching calls.

Data collection

All data were obtained from the disease registry, which included data on patients' demographic characteristics, type and timing of follow-up contacts, and if applicable, type and dose of psychotropic medication prescribed. All data were collected in the routine course of delivering care. Sites were not asked to collect additional data for the evaluation.

Quality indicators.

To assess an organization's performance, care process measures can serve as indicators of health care quality (10). Two care process measures were selected to serve as indicators of health care quality—early follow-up, defined as follow-up with the care manager within three weeks of treatment initiation, and appropriate pharmacotherapy, defined as presence of an antidepressant medication at a therapeutic dose based on published guidelines in the treatment plan at the visit preceding outcome assessment (11–13).

Outcome measures.

At every contact with a care manager, patients completed the Patient Health Questionnaire-9 (PHQ-9), a widely used measure of depression severity and treatment response (14–18). On the basis of baseline PHQ-9 scores, depression severity was categorized as mild (0–9), moderate (10–14), or severe (15–27) (14). Patients who achieved a 50% reduction in PHQ-9 score at follow-up or scored ≤5 were considered to have improved; patients whose scores were ≤5 were classified as being in remission. Outcomes for the acute phase of depression treatment were obtained from follow-up contacts that occurred between six and 12 weeks after the initial assessment, given that the benefits of both antidepressant medications and psychotherapy necessitate at least four to eight weeks of treatment (11–13). For acute-phase analyses, patients with a PHQ-9 score during the acute-phase period were considered completers, whereas patients who did not have a PHQ-9 score were counted as dropouts.

Covariates.

Demographic information (age, gender, and preferred language) was obtained from the registry. The Overall Anxiety Severity and Impairment Scale (OASIS), a five-item self-report measure of anxiety symptoms that has been validated among primary care patients, was used to assess anxiety symptoms (6,19).

Data analysis

To examine the impact of site on the probability of receiving early follow-up and appropriate pharmacotherapy and of experiencing improvement of depression, logistic regression models were specified. To estimate the effect of site on the number of weeks elapsed before improvement was observed (time to improvement), Cox proportional hazards models were constructed. Because the patients served by each clinic differed in their demographic and clinical characteristics, factors that are associated with depression care and outcomes (20–23), all analyses accounted for potential confounding by adjustment for patients' age, gender, preferred language, and baseline scores for depression and anxiety.

Results

A total of 2,821 patients were identified as having symptoms of depression between 2006 and 2009 and were enrolled in treatment; they were predominantly female, ranged widely in age, spoke English or Spanish, and had moderate-to-severe depression and anxiety symptoms (Table 1). Patients (N=271) who met criteria for remission at baseline as well as patients (N=540) for whom some predictor variables were missing were excluded from analyses of outcomes, resulting in a valid sample for the regression models of 2,010 patients. Table 1 presents data for each clinic on the caseload, patient characteristics, and details of the treatment and outcomes.

Both quality indicators were strongly associated with site (Table 1), differences that were not attributable to characteristics of the patients served (Table 2). On the basis of multivariate logistic regression models, the adjusted probability that a patient received an early follow-up contact ranged from .34 to .88, whereas the adjusted probability for receipt of appropriate pharmacotherapy ranged from .27 to .69.

Depression outcomes at 12 weeks, four months, and six months varied substantially across clinics in unadjusted analyses (Table 1). The multinomial logistic regression model revealed that after adjustments for patient characteristics, the probability of discontinuing treatment differed significantly across clinics (range .29–.75), as did the probability of improvement (range .36–.84) among patients (N=1,063) who had valid PHQ-9 scores between six and 12 weeks (Table 2).

Sensitivity analyses examined whether the quality indicators could account for the differences observed in outcomes across site. The inclusion of the quality indicators in the multinomial model did not attenuate the effects of site on outcome (data not shown). Patients with early follow-up were less likely to drop out (odds ratio [OR]=.50, p<.001) and more likely to improve (OR=1.64, p<.01), whereas patients who received appropriate pharmacotherapy were less likely to drop out (OR=.73, p<.05) but were no more likely to improve.

Survival analysis was used to estimate time to improvement across sites while accounting for censoring of patients who dropped out of care. The unadjusted survival curve, Figure 1, illustrates that the proportion of patients who experienced improvement across clinics diverged as early as four weeks and that these differences persisted over time. Due to censoring, these results cannot be compared directly to rates of improvement reported in Tables 1 and 2, although the findings are similar. Site differences in time to improvement remained significant after adjustment for patients' age, gender, preferred language, and baseline scores for depression and anxiety according to a Cox proportional hazards model (not shown).

Discussion

Overall, this demonstration project revealed that organizations with markedly differing characteristics were able to integrate mental health care into primary care for disadvantaged and underserved patients. In all sites, a plurality of patients experienced meaningful improvement in depression, and in some sites, the majority experienced rapid improvement following initiation of treatment.

Despite such broad successes, the organizations differed dramatically in indicators of quality and in outcomes, differences that were not attributable to the characteristics of the patients we measured. Moreover, the variability in quality measures and outcomes indicated that substantial room remained for improvement in quality of care.

Retention in depression treatment is poor for primary care patients, particularly among low-income members of minority groups (24,25). Outcome measures were collected exclusively in the course of clinical treatment and, therefore, were obtained at different times for different patients. Consequently, many patients who did not have a PHQ-9 score between six and 12 weeks were counted as dropouts for acute-phase outcomes, but only 18% were truly lost to follow-up, a rate that closely approximates retention in collaborative care research trials (1,3,4,26). Moreover, our survival models accounted for attrition in constructing estimates of outcomes over time.

In research on patients treated in a collaborative model, about 60% improved and 25% achieved remission during the acute phase, compared with rates for patients in usual care of about 40% and 10%, respectively (3,4,26,27). Depression outcomes are worse among patients whose conditions are complex, including the elderly, socioeconomically disadvantaged, anxious, or treatment resistant (3,4,6,26–31).

Our results refuted the notion that trial results could not be reproduced by organizations in economically deprived, minority communities serving heterogeneous patients. Several participating organizations were federally qualified health centers in poor urban border towns that have high concentrations of Spanish speakers and were safety-net providers for uninsured individuals, including undocumented immigrants who are ineligible for Medicaid. At six of seven sites, half of patients improved, and at five sites more than 40% of patients achieved remission, a more challenging clinical target.

Thus, the results provided robust support to demonstrate that collaborative care, when adapted by community health centers, can meet or surpass the outcomes achieved in research settings, even in poorly resourced settings. This finding is highly relevant in light of strong interest in integrating health care in the context of national health care reform.

Although all organizations received the same training, guidance, and support and all used the same Web-based registry, the care provided and the outcomes achieved differed markedly. Outcomes diverged across sites as early as four weeks, highlighting the critical importance of the early course of treatment. For example, loss to follow-up occurred for only 4% of patients at organization 2, compared with more than one-third of patients at organization 6. Organization 2 was successful in all metrics of care; 87% of patients received early follow-up, 66% had four or more follow-up contacts within 12 weeks (average=4.8), and 68% received appropriate pharmacotherapy. Patients treated at this clinic had excellent acute-phase outcomes—two-thirds of patients improved, and half achieved remission. In contrast, the site with the worst acute-phase outcomes (organization 1, clinic B) performed relatively poorly on process measures. As many as 19% of patients were lost to follow-up, only one-third of patients received early follow-up, just 5% of patients received four or more follow-up contacts within 12 weeks, and approximately half of patients received appropriate pharmacotherapy.

Having conducted site visits, we had the impression that some sites implemented the collaborative care model more effectively than others. Those sites that appeared to implement well demonstrated better performance on both quality indicators and patient outcomes. These differences were not attributable to patient characteristics. Although early follow-up was significantly associated with improved outcomes, we were not able to conclude that more patients improved because they received more intensive follow-up. It is possible that the indicators we measured tapped into broader constructs of the quality of services provided across organizations and were not responsible for improved outcomes but, nevertheless, were associated with such outcomes.

In multivariate analyses adjusting for quality indicators, site remained strongly associated with outcomes, suggesting that the effect of being treated in a particular clinic is at least as salient as receiving (or not receiving) early follow-up or appropriate pharmacotherapy. On the surface that may appear surprising; however, it was consistent with our impression that the quality measures we quantified were indicators of broader processes occurring within the clinics. Similar patterns may have emerged if we had quantified other quality indicators. Clinics with better implementation may have been more likely to provide treatment intensification for patients not responding to initial treatments, effects that were not captured by our analyses. Similarly, clinics that achieved superior outcomes may have been more effective in tailoring treatment to individuals' needs, for example, by accounting for patient preferences, availability of services, and adaptation of treatment to individual outcomes.

Because a small number of organizations participated in this demonstration project, we were not able to quantitatively evaluate how characteristics of the organizations were associated with outcomes. However, because of the great interest in replicating collaborative care models in community settings, we speculate on some potential contributors.

Several organizations located in impoverished communities with no preexisting mental health services, little or no experience with integrated care, and few resources were largely successful in implementing collaborative care. In contrast, the organization that had the least success in program implementation was a large organization with a significant preexisting structure of mental health providers both within primary care and at specialty sites. Having a system in place that separated mental health and primary care appeared to impede implementation. Despite receiving training and additional resources, the organization appeared to make few changes to its practice, and care remained so fragmented among a wide array of providers that many patients “fell through the cracks.” Thus, building an integrated delivery model de novo may be more straightforward than reengineering a system of existing services.

Organization 4 provided an example of a success story. Although the leadership at this organization was unsupportive and at times antagonistic to the initiative, the organization had strong support from capable primary care providers and care managers who embraced the model. In this case, it appears that the clinicians were successful in enacting substantial practice change despite a lack of strong support from the administration. They engaged patients early on and provided appropriate pharmacotherapy to most of them. These examples suggest that paying attention to the preexisting system of care may be an important element for organizations to consider as they implement an integrated model and that getting buy-in from clinical providers on the ground may be particularly crucial.

In considering the implications of these findings, several limitations are important. First, unmeasured differences in the patient populations treated across sites may have accounted for some of the variability observed. Information was not available on patients' race-ethnicity, education, or income, factors that may be associated with depression outcomes. Because the organizations predominantly treated uninsured patients in impoverished communities, these results may not represent organizations treating patients under more advantaged circumstances.

Without having data on treatment patterns and outcomes at these organizations before the start of the initiative, we were unable to indicate to what extent the treatment and outcomes observed were the result of the introduction of this model. The patients treated represented a heterogeneous mixture of clinical presentations, and data on diagnoses were not available for further description. Analyses were also limited by a lack of detail about patients' use of psychotherapy and psychiatric consultation. Finally, medication data were obtained from the registry and may not have reflected patients' actual medication use.

Conclusions

Several lessons from this demonstration project are instructive as health centers throughout the country grapple with the process of providing integrated health services. Successful implementation of integrated care occurred in a variety of settings and achieved admirable outcomes among a heterogeneous population of disadvantaged primary care patients. Moreover, organizations without preexisting mental health personnel were particularly successful in creating well-functioning integrated mental health care teams.

Organizations interested in implementing an integrated model may wish to pay particular attention to a few of our observations about the challenges these organizations faced. First, in settings with extensive existing mental health services, having a preexisting structure may have impeded practice change and integration. Second, insufficient follow-up may result if staffing is inadequate. Third, the effectiveness of the model may be limited if there is inadequate participation by a psychiatrist for supervision of care managers and for consultation. Successful sites engaged patients early in treatment and had multiple contacts with patients during the first 12 weeks. Thus organizations striving to provide collaborative care should focus their efforts on developing a system that allows for early and intensive follow-up, thereby sending patients a strong message of the organizations' commitment to engaging them in care.

Figures and Tables