Setting Denver Health is an integrated healthcare system. During the time period that the trial was conducted, 2010 to 2012, Denver Health comprised eight federally qualified community health facilities serving primarily uninsured and Medicaid patients in Denver, Colorado, USA. The mainstay treatment for depression was antidepressant medication only, as access to specialty mental health counseling or psychiatric services was not readily available at the participating primary care facilities during the time frame of the trial.
Randomization and Masking We recruited seven primary care facilities housing 10 independent care teams (clusters). Three facilities had two teams each, and four had one team. Although staff and providers from different facilities and teams attend role-specific training together and may share managers, each team has unique support staff, providers, and patient panels. We used a blocked design to limit the possibility of patients crossing over from a team assigned to one randomization group to one in the other group (contamination). Because of a negative association between minority race/ ethnicity and depression diagnosis and treatment receipt, we stratified the teams by the race/ ethnicity of the respective patient populations at each facility (plurality Hispanic, non-Hispanic black, or non-Hispanic white). Prior to randomization, we blocked eight of 10 teams into four pairs within strata because they: (a) shared the same facility (three facilities with two teams each); or (b) were from closely located neighborhoods (two facilities with one team each).
Outcome assessors who obtained written consent, conducted the baseline depression assessment, and collected clinical outcomes were blinded to cluster allocation and were not involved in other study procedures. Patient participants were blinded to group assignment. Participating providers were not blinded.
Protocol Change Prior to recruiting patients into the intention-to-treat (ITT) groups, we changed the primary outcome at the trial registration site on 4/29/10 from the outcome specified in the original 2008 protocol, "adherence to antidepressant medication", to "remission from depression." However, detecting remission, defined as an event where depressive symptoms fall below a clinically significant level for a prolonged period, entails weekly or bi-weekly outcome assessments (Nierenberg, 2001; Rush et al., 2006). Limited study resources precluded closely spaced outcome assessments over 36 weeks. Therefore, to better understand trajectories of clinical outcome for the primary care patients assigned to teams staffed by providers trained to discuss depression with Motivational Interviewing versus those patients assigned to teams staffed by untrained providers, measures of both continuous depressive symptoms and a binary outcome measure (Patient Health Questionnaire-9 (PHQ-9) score \< 5) were calculated for each group at 6, 12, and 36 weeks.
Primary Care Providers In May 2009, the research team emailed 53 providers from 13 primary care teams (eligible clusters) in eight facilities with an invitation to participate in a study of "training for a new counseling method for treating depression." Inclusion criteria included working at least 1.5 days per week in outpatient care, previous experience treating depression, and availability for a one-day baseline MI training on 7/25/09. We obtained consent from providers prior to randomization.
Patients During the recruitment phase, the research team called sequential eligible patients 2 to 3 days prior to scheduled visits with participating providers. The team assessed for initial exclusion criteria and invited those not excluded to complete an audio-recorded partial waiver of consent and stage-I depression screen, the Patient Health Questionnaire-2 (PHQ-2). Prior to the visit, a recruiter met patients scoring 2 or higher on either item of the PHQ-2, described the study, obtained written informed consent, and invited consenting patients to complete the PHQ-9.
Patients were excluded if they were attending a one-time disability qualification examination, had no personal phone, were homeless, were treated for depression with antidepressant medication within the previous 3 months, were currently in therapy provided by a mental health specialist, had impaired memory, had a life-threatening physical disease, expressed severe suicidal ideation, were pregnant or breastfeeding, or exhibited alcohol/ drug dependence, bipolar disorder or current psychosis assessed with a diagnostic schedule.
The recruiter notified the provider and patient prior to the clinical encounter when patients scored 10 or higher on the PHQ-9, signifying 'probable major depression' . A board-certified psychiatrist, blinded to patient allocation, diagnosed depression and determined additional exclusion criteria. Patients diagnosed with a major depressive episode, not excluded for other reasons, and completing a subsequent baseline telephone interview within 8 days, were enrolled as the Intention-To-Treat (ITT) sample.
Intervention - MI with Standard Management of Depression The MI training approach included interactive learning for the core MI skills. An 8-hour classroom training on 7/25/09 consisted of a brief overview of MI, videos and discussion of core MI skills and "MI Spirit," as well as skill-building practice. Providers learned how to use "OARS" and summaries to elicit change talk, how to implement the elicit-provide-elicit technique, and how to craft action plans with patients. Providers practiced using 0-10 Ruler questions to assess and increase patient importance and confidence in changing.
The training was designed according to Arkowitz and Burkes' (2008) recommendation for a three-component framework when using MI to discuss depression. In this framework, the provider first helps explore the patient's negative emotions through reflective listening. Second, the provider uncovers contributors to the negative emotions. Third, the patient's own ideas about how to handle their negative emotions are elicited.
To optimize treatment integrity, 4-hour refresher sessions were offered after 4 and 12 months. Over the first 14 months, feedback was provided via email and face-to face regarding audio-taped encounters (total two to four feedbacks per provider).
Control - Standard Management of Depression Alone All providers randomized to either intervention or to control received a 1-hour slideshow and the "American Psychiatric Association Practice Guideline for the Treatment of Major Depressive Disorder" (APA depression guideline) (2010). The resources recommended antidepressant medications and psychotherapy as evidence-based treatments for depression.
Measurements Baseline Descriptors. In addition to demographics, provider information included training (M.D. versus mid-level) and certification (Family Practitioner, General Internist, Physician's Assistant, or Nurse Practitioner), and years of clinical experience. Patient descriptors included generalized anxiety disorder (past 6 months), federal poverty level, and previous suicidal behavior.
Treatment Variables. Time spent discussing depression at the index visit was assessed from audio-recorded encounters. Two measures of follow-up over 36 weeks, total primary care visits dichotomized at the median and at least three or more visits with documented discussion of mental health were abstracted from the electronic medical record. Receipt of antidepressant medication and specialty mental health counseling was also determined.
Statistical Analysis Missing Data. Total missingness and whether missingness was differential by group or cluster was assessed. Little's likelihood ratio test was used to test the hypothesis that data is Missing Completely at Random (MCAR). To determine plausibility of Missing at Random (MAR), correlations between baseline or auxiliary variables and missingness by group at each time point were calculated.
Prior to performing the ITT analyses, Multiple Imputation through Chained Equations (MICE) was conducted to deal with missing data. The Fully Conditional Specification algorithm in the SAS MI procedure was used to generate 30 imputed datasets. The effects of the intervention on depressive symptoms and on remission were determined on each imputed data set. MI repetition results were combined to generate final estimates of parameters and confidence intervals for depressive symptoms. Rubin's rules were used to combine the parameter estimates and errors into a single set of results.
Treatment Outcomes. Analyses of treatment outcomes were run by ITT. Generalized linear mixed effects models, adjusted for clustering of patients within care team, were used to examine differences in outcomes by randomization group. Whereas randomization was stratified by class (race/ ethnicity), the interaction between class and treatment group as well as class itself were included in all multivariate analyses.
Main Effects. Main outcome analyses were run by ITT utilizing imputed datasets. General linear mixed effects models, or generalized models for a binary outcome, were used to examine whether the group x time interactions for outcomes depressive symptoms and remission were significant. The general linear mixed model assumed unstructured variance-covariance matrices, maximum likelihood (method = ML) estimation, and adjusted for clustering of patients within care teams and repeated measures within patients. Whereas the random effect of clustering patients within care teams on remission in the adjusted model was zero, patients were not clustered within care teams in the generalized model for the binary outcome.
Moderator hypothesis: A hypothesis concerning moderator variables was included in the Colorado Multiple Institutional Review Board (COMIRB) protocol for the trial. In the protocol approved with a COMIRB stamp on 2/3/10, specific aim #2 was "ii. Specific Aim #2: Aim 2 is to explore for moderators and mediators of the effect of AMI-TAD on adherence and outcome. We hypothesize that higher baseline patient self-efficacy will differentially increase the effect of AMI-TAD with SCD." AMI-TAD is an earlier abbreviated name for the the MI training intervention.
