Depression causes substantial suffering, health care expenditure, additional economic loss, and death through suicide and inferior self-care of other chronic diseases (
1,
2). Most depression treatment is administered in primary care settings, where diagnosis, medication selection and dosing, duration of treatment, and response to treatment failure are often suboptimal (
3,
4).
Designed to improve depression outcomes for primary care patients, collaborative care involves expanding primary care teams with the addition of two new roles. A psychiatrist conducts regular case reviews and advises on diagnosis and treatment. A care manager educates patients about depression, coordinates referrals, promotes behavior changes that decrease depression symptoms, supports adherence to treatment regimens, administers serial depression symptom questionnaires, and notifies primary care providers when responses to those questionnaires indicate inadequate improvement and a possible need to revise the treatment plan. On the basis of meta-analyses, researchers have concluded that collaborative care substantially improves depression outcomes (
5,
6). A randomized controlled trial involving primary care patients ages 60 years and older found that collaborative care generated substantial health care cost savings (
7).
From 2008 to 2012, the Institute for Clinical Systems Improvement (ICSI) assisted 87 primary care clinics in implementing collaborative care through the Depression Improvement Across Minnesota, Offering a New Direction (DIAMOND) initiative (
8). The clinics received similar training and support but varied in how they staffed and configured care manager positions. Some care managers were registered nurses (RNs), whereas others were certified medical assistants (CMAs) and licensed practical nurses (LPNs). Some were dedicated to depression care, but others had additional responsibilities. Some served patients at one clinic, whereas others served patients at multiple clinics in the same medical group. To inform future implementation of collaborative care, we analyzed associations between two key measures of program effectiveness—patient enrollment and six-month remission rates—with care manager background and clinic characteristics.
Methods
Between March 2008 and March 2010, a group of clinics began DIAMOND implementation every six months, for a total of five groups. A total of 87 clinics operated by 21 medical groups participated. This study focused on the 63 clinics, operated by 12 medical groups, that remained active through early 2012.
Staff of the participating clinics aimed to enroll all patients with
ICD-9 diagnoses of 296.2X, 296.3X, or 300.4 and a score of ≥10 on the Patient Health Questionnaire–9 (PHQ-9), suggesting probable major depression or significant dysthymia (
9). Patients were discharged from DIAMOND if they met criteria for remission, defined as having PHQ-9 scores of <5 on two consecutive occasions over a period of greater than two months.
Care managers were expected to collect PHQ-9 data from patients at every contact. They contacted patients weekly to monthly on the basis of the severity of depression symptoms. Monthly, clinic staff submitted deidentified data on service delivery and PHQ-9 scores for each enrolled patient. From these data, the ICSI calculated total enrollment and remission rates.
Throughout implementation, the ICSI tracked clinic progress and provided technical assistance. In early 2012, the ICSI surveyed project leaders at each active medical group about care manager and clinic characteristics. The data indicated whether the care manager was an RN, a CMA, or an LPN; whether the care manager was dedicated solely to depression care or had other duties; and whether the care manager provided depression services at one or multiple clinics. Of the 63 clinics, seven clinics indicated modifying their staffing models within two months of program launch, and the data reflect the final staffing model. Clinics located in or near the Twin Cities (in Hennepin or Ramsey County or adjacent counties) were classified as metropolitan, and the other clinics were classified as outstate. Complete data were obtained for all 63 active clinics.
We analyzed the relationship of care manager and clinic characteristics to rate of enrollment of eligible patients and rate of remission six months (±30 days) after the enrollment date. Generalized linear mixed models (GLMMs) were used to account for multiple sources of clustering within the data (
10). We conducted two separate GLMMs, one using enrollment rates and one using remission rates as the binomial outcome variable. For each model, the independent variables were the three care manager variables and clinic location. A random-effects term accounted for medical group effect. A sequence effect term denoted the five different start dates. A term for number of enrolled patients controlled for clinic volume. We conducted the analysis with SAS, version 9.2 (
11).
Informed consent was not obtained from patients, nor was institutional review board approval sought, because the project was a quality improvement initiative, and no protected health information was shared with the researchers.
Results
Of 55,594 eligible patients, 9,179 patients enrolled in DIAMOND as of early 2012, for a 17% enrollment rate across all clinics. Enrollment rates varied from 1% to 55% across the participating clinics. The model in
Table 1 indicates the association between enrollment and remission rates and clinic characteristics after the analyses accounted for medical group–level effects. Clinics with a dedicated care manager had higher enrollment rates than clinics in which the care manager had multiple roles (odds ratio [OR]=2.65, p=.050). There were no significant differences in enrollment on the basis of care manager licensure (RN versus CMA or LPN), number of clinics served, or clinic location.
Across the clinic sites, 7,438 enrolled patients were eligible for six-month PHQ-9 follow up. Of these patients, 2,323 attained remission at six months, yielding an overall remission rate of 31%. Six-month remission rates varied from 3% to 46% across the clinics. As shown in Table 1, the independent variables had no significant effects on remission rates. There was a trend for sites with a dedicated care manager to have higher remission rates (OR=1.96, p=.105).
Clinics that were among the first three groups to implement DIAMOND had higher enrollment and remission rates. Enrollment rates were significantly lower at larger versus smaller clinics, but remission rates were significantly higher at larger clinics.
Discussion
The U.S. Preventive Services Task Force (USPSTF) recommends depression screening in primary care settings only “when staff-assisted depression care supports are in place to assure accurate diagnosis, effective treatment, and follow-up” (
12). Collaborative care clearly comprises the most effective supports.
Under the DIAMOND initiative, as the clinics implemented collaborative care, they varied in how they staffed the care manager role. Our analysis found no differences in outcomes obtained by RNs, CMAs, and LPNs, who had completed the same training. Our analysis also found that staffing by a dedicated care manager was associated with significantly greater enrollment and, possibly, superior treatment outcomes compared with staffing by a care manager with multiple duties, even when dedicated staff were shared across multiple clinics.
A possible explanation for the superior performance of the clinics in the first three groups to implement DIAMOND may be related to selection effects. Earlier-participating clinics may have been most ready and motivated to initiate the program. The association between volume of patients served and better outcomes may indicate that managers with dedicated time are more ready and motivated to develop their skills.
One limitation of the study was its focus on clinics that were still implementing collaborative care two to four years later. The data did not identify the factors that may have been associated with program discontinuation. Additional limitations were the retrospective nature of the analyses, which were possibly confounded by differences in implementation that were not measured and by the effects of idiosyncratic practices at the clinics studied. Furthermore, the study design allowed conclusions about association, not causation, given that decisions about staffing of care manager positions could have influenced service delivery or vice versa.
The findings of this study support the “teamlet” model of primary care, in which medical assistants are trained to execute health coaching functions (
13). A larger prospective study would provide more definitive guidance on how best to configure and staff collaborative care programs.
Even if clinical data suggest how to optimize staffing, decisions about staffing must take into account the regulatory environment and financing opportunities. Regulations governing which individuals may deliver collaborative care may vary across states. In Minnesota, for example, paraprofessionals are permitted to deliver collaborative care, and major commercial payers honor a billing code for paraprofessional-administered collaborative care (
14). This type of flexibility and financing is lacking in other states. Currently, Medicare reimburses for depression care only when depression care supports are in place, consistent with USPSTF recommendations. However, it does not reimburse for ongoing collaborative care. Thus most primary care settings cannot finance collaborative care through fee-for-service reimbursement.
Conclusions
Increasingly, health care providers are rewarded for delivering value rather than services, for example, as part of accountable care organizations. Research has documented a positive return on investment for collaborative care, which should help drive dissemination (
7). As collaborative care spreads, patients and health care purchasers will benefit, especially if staffing decisions and relevant regulations are based on research findings rather than traditional role concepts.
Acknowledgments
Dr. Brown is the owner and CEO of Wellsys, LLC, which helps health care settings and workplaces implement behavioral screening and intervention. The other authors report no competing interests.