Community-Partnered Cluster-Randomized Comparative Effectiveness Trial of Community Engagement and Planning or Resources for Services to Address Depression Disparities

The CPIC Council modified depression QI toolkits4,27 supporting clinician assessment and medication management (manual and pocket cards), case management (manual), patient and education (brochure, video), and CBT (individual and group manuals) for use by diverse agencies, adding a lay health worker manual and team support tools28 from a post-disaster initiative blending QI toolkits8,12,29 (Table 1). The toolkits were available to programs online and through flash drives and hardcopy. The CPIC Council developed two interventions: 1) Resources for Services (RS) offered programs in each community technical assistance in toolkit use under a “train-the-trainer” paradigm through webinars plus site visits to primary care for each community. The trainers included a nurse care manager, licensed psychologist CBT trainer, three expert board-certified psychiatrists for medication management and QI, support staff and community service administrator to support participation and cultural competence. 2) Community Engagement and Planning (CEP) invited administrators to bi-weekly meetings for 5 months to build training capacity for toolkits and networks for services. Planning was co-led by community and academic Council members and followed principles of CPPR,25,26 such as equal authority for decisions and two-way knowledge exchange. CEP councils were provided with a workbook for developing written implementation plans tailored to the community and monitored plan implementation with course corrections as needed.

Within enrolled programs, providers could attend trainings and providers and clients could use intervention resources, even if not personally enrolled. Enrolled client lists were provided to CEP but not RS administrators, except one agency with a shared waiting room where both were given lists. Programs were not asked to limit toolkit use to enrolled clients.

For RS, between December 2009 and July 2010 ten webinars were provided per community plus one make-up in one community; one primary care site accepted a visit. Khodyakov et al.49 describe CEP planning including development of trust and partnered co-leadership and completion of a written plan. Common features of plans included multiple training conferences, follow-up for individual programs and telephone and webinar supervision for therapists and case-managers, between December 2009 and July 2011. Los Angeles County’s Service Planning Area 6 (South LA) Council developed train-the-trainer capacities while the Service Planning Area 4 (Hollywood/Metro LA) Council used group case review to identify training modification needs. CEP offered more training and increased staff participation and total training hours relative to RS.

The two communities, Los Angeles County Service Planning Areas 6 (South Los Angeles) (1.5 million people) and 4 (Hollywood-Metro Los Angeles) (500K people), have high rates of unemployment, homelessness, avoidable hospitalizations, and low rates of insurance.50 We hosted community meetings to determine service sectors viewed as important to depressed clients,24 and to identify vulnerable populations for oversampling. Across communities, sectors were mental health, primary care and public health, substance abuse, social services, and faith-based programs, parks, hair salons, and exercise clubs. South Los Angeles partners nominated substance abuse clients and African Americans and Hollywood-Metro homeless clients and seniors for over-sampling.

With no standard approach to recruit diverse sectors, we used county lists supplemented by community nominations to identify agency names. We used an iterative process of contacting programs and assessing eligibility and interest to offer consent to 60 potentially eligible agencies having 194 programs, from which 133 potentially eligible programs (serving ≥ 15 clients per week, one or more staff, not focused on psychotic disorders or home services) were randomized (65 RS, 68 CEP) (Fig. 1). Within each community, programs were paired into exchangeable units or clusters of smaller programs, based on geographic location, service sector, size, population served, services provided, and funding streams; two larger agencies were their own stratum. One unit was randomized to CEP and the other to RS.51 Randomization was conducted by a statistician uninvolved with recruitment. Council members produced seed numbers for the randomization program.52 Site visits were conducted to finalize enrollment using recruiters blinded to assignment; 20 programs were ineligible, 18 refused and 95 programs from 50 consenting agencies enrolled (46 RS, 49 CEP; Fig. 1, Table 2). Administrators were informed of intervention status by letter. From census track data, participating and nonparticipating programs were comparable in age, sex, race, population density, and income at the zip code level (each p > 0.10).

After changes in response to community input, the study was designed to achieve a follow-up sample of 780 depressed clients, for a detectable effect size of 0.20–0.22 and percentage point difference between groups of 9.98–10.91 %, assuming power of 80 % with alpha = 0.05 (two-sided), and intraclass correlation coefficient (ICC) = 0 .00–.02.51,53,54 To achieve this, we planned to enroll 557–600 clients per condition assuming 65–70 % retention.

Within programs, clients were screened in waiting rooms (consecutive) or events (random) from March 2010 to November 2010. Staff blinded to intervention status approached 4,649 adults (age ≥ 18) on 2–3 days per program and 4,440 (95.5 %) agreed to screening. Eligibility was limited to clients providing contact information and having at least mild depression (score ≥ 10 on the 8-item Patient Health Questionnaire (PHQ-8), with the same scoring characteristics and cut-point as the PHQ-9.55 Based on community input we counted a response on either of two versions of one PHQ-8 item with and without the word “depression”; Pearson correlation coefficient of items was 0.99. Of 4,440 screened, 1,322 (29.8 %) were eligible and of these 1,246 (94.3 %) consented, a high rate for QI studies.10,12,29 Between April 2010 and January 2011, 981 (78.7 % of consented) completed baseline telephone surveys (interviewers blinded to assignment) and between November 2010 and August 2011, 759 completed 6-month follow-up telephone surveys (61.1 % of consented minus three deaths), a lower rate than some QI studies,27,29 but similar to studies of safety-net clients8,31; 1,018 (81.9 % of consented minus deaths) with baseline or follow-up data (Fig. 1) comprise the analytic sample. One RS and two CEP programs had no clients completing surveys, yielding 90 programs with clients in the analytic sample. The study, funded in 2007, was not considered a clinical trial by the National Institutes of Health and no data safety monitoring board was required; post data collection the study was registered (ClinicalTrials.gov NCT01699789). No major design changes were made after recruitment began. Patients reporting recent suicidal ideation on surveys were contacted by telephone by a study clinician following a protocol. Procedures were approved by Human Subjects Protection Review Committees of RAND and participating agencies.

Baseline measures include program intervention assignment and sector and client data from the screener and baseline survey on demographics (age, sex), presence of three or more chronic conditions from 18 conditions, education level, race/ethnicity; from the 12-item Short Form health survey (SF-12)56 physical and mental health composite scores (PCS12, MCS12) and meeting federal family poverty criteria. Using the Mini-International Neuropsychiatric Interview (MINI),57 we assessed using Diagnostic and Statistical Manual of Psychiatric Disorders, Fourth Edition (DSM-IV) criteria 12-month major depressive or dysthymic disorder, current manic episode, recent anxiety disorder (1-month panic or post-traumatic stress or 6-month generalized anxiety disorder), and alcohol abuse or use of illicit drugs in 12 months. Study outcomes are based on client self-report at 6-month telephone follow-up. Pre-planned primary outcomes are poor mental HRQL (MCS12 ≤ 40, one standard deviation below population mean)56 and exceeding the cut-point for mild/moderate depression on the PHQ-955 (score ≥ 10). In response to community priorities, we developed indicators of mental wellness (three items of having in the prior 4 weeks some feeling of being calm or peaceful, having energy or being happy56); responding somewhat or definitely true to “my life is organized”58 as a resiliency measure; being at least fairly physically active from a single item; working for pay; missing work in the past 30 days; and either currently homeless/living in a shelter or having multiple risk factors for future homelessness,59,60 including no place to stay for at least two nights in the past 6 months or eviction from primary residence, financial crisis or food insecurity in the past 6 months from life events measures.61

We assessed services use in the past 6 months for overnight hospital stays for mental health or substance abuse, overnight substance abuse rehabilitation, emergency room (ER) visits, outpatient mental health or self/family groups visits, hotline calls, and use of outpatient primary care or public health clinics, substance abuse or social services programs, parks and community centers, and faith-based and other community locations, counting as depression-related visits for which the client reported receiving information, referral, counseling, or medication management for depression or emotional problems. We developed indicators for any use and being above the baseline median visits and counts of contacts. Since a single overnight stay could reflect emergency room use, we included for sensitivity analyses having ≥ four hospital nights. To account for potential bias in self-report, we asked participants to provide names and addresses for up to four providers per sector, and for high utilizers and “other” locations, confirmed sector assignment and feasibility of counts through internet searches and program telephone calls.

We conducted intent-to-treat, comparative-effectiveness analyses with intervention status as the independent variable, using logistic regression for dichotomous measures and log-linear models for counts. Consistent with group-randomized trial recommendations,51 we adjusted for baseline status of dependent variables and covariates (age, sex, ≥ three chronic conditions, education, race/ethnicity, family poverty, 12-month alcohol abuse or use of illicit drugs, 12-month depressive disorder, and community). We weighted data for 1,018 clients to characteristics of the eligible sample, with item-level imputation for missing data62,63 and wave-level64,65 imputation for missing surveys. Weights account for non-enrollment among eligible clients and attrition (see Supporting Online Material). All analyses were conducted using SUDAAN Version 10.0 (http://​www.​rti.​org/​sudaan/​) and accounted for clustering (clients within programs),66 weighting, and multiple imputations.64 Significance of comparisons by intervention status was based on regression coefficients. Results of logistic regression models are presented as odds ratios (OR) and log-linear models as rate ratios (RR) with 95 % confidence intervals. We illustrate results for intervention groups adjusted for covariates using standardized predictions generated from fitted regression models.27,67 For outcomes, we provide adjusted differences in predictions with standard errors.68 We present Cohen’s effect size index h, defined for dichotomous variables,53 where h = 0.20 is small, h = 0.50 is medium and h = 0.80 is large. We supplement adjusted models with unadjusted raw data to assess robustness. We applied a Bonferroni adjustment69 considering two primary outcomes and compared the proportion of results across outcomes favoring one intervention to an assumed null-hypothesis of 0.5.