Abstract
Background: Disparities in depression care exist among the poor. Community Partners in Care (CPIC) compared a community coalition model with technical assistance to improve depression services in under-resourced communities. We examine effects on health, social, and utilization outcomes among the poor and, non-poor depressed, and poor subgroups.
Methods: This study analyzed clients living above (n = 268) and below (n = 750) the federal-poverty level and, among the poor, 3 nonoverlapping subgroups: justice-involved (n = 158), homeless and not justice-involved (n = 298), and other poor (n = 294). Matched programs (n = 93) from health and community sectors were randomly assigned to community engagement and planning (CEP) or resources for services (RS). Primary outcomes were poor mental health–related quality of life and 8-item Patient Health Questionnaire scores, whereas community-prioritized and utilization outcomes were secondary. Effects were scrutinized using false discovery rate–adjusted P values to account for multiple comparisons.
Results: In the impoverished group, CEP and RS clients of participating study programs did not differ in primary outcomes, but CEP more than RS improved mental wellness among the depressed poor (unadjusted P = .004) while providing suggestive evidence for other secondary outcomes. Within the poor subgroups, evidence favoring CEP was only suggestive but was strongest among justice-involved clients.
Conclusions: A coalition approach to improving outcomes for low-income clients with depression, particularly those involved in the justice system, may offer additional benefits over standard technical assistance programs.
Depression is associated with increased morbidity and physical health comorbidity.1,2 While prevalence of depressive symptoms is similar across cultural groups,3 ethnic minorities and persons of lower socioeconomic status have worse access to evidence-based care and worse outcomes than white populations.4,5 Quality improvement (QI) programs for depression in primary care, such as the collaborative care model,6 are effective in minority and low-income communities.7⇓⇓⇓⇓⇓⇓⇓⇓–16 Miranda et al17 demonstrated that depression treatments are effective for socioeconomically vulnerable, minority women, whereas Partners in Care found greater benefits under collaborative care compared with usual care for African Americans and Latinos relative to non-Hispanic whites across 5 years of follow-up.16,18 Such interventions are not widely available in safety-net practices, despite calls for implementation in policy sectors.19 National studies demonstrate persistent racial and ethnic disparities in access to and quality of depression care.20
Participatory coleadership with patient and community stakeholders is recommended for engaging under-resourced communities in improving their health.21,22 Community Partners in Care (CPIC) found that a Community Engagement and Planning (CEP) intervention, cultivating multiagency networks as a community coalition to implement QI interventions for depression across health and community sectors, was more effective at improving mental health quality of life (MHQOL) and reducing hospitalizations over 6 to 12 months than was Resources for Services (RS), which made expert assistance available to individual agencies.14,23 This article examines the extent to which the effects of CEP over RS for the whole sample also applied to persons living in poverty and to nonpoor participants, and it explores intervention effects for specific subgroups of the poor of clinical and policy interest.
CPIC Intervention
Community-partnered participatory research was used to design and implement the CPIC study and guide development of the CEP community coalition intervention to enhance depression services. Clients were recruited from programs delivering health (mental health, primary care, public health, substance abuse) and/or social and community services (homeless-services, prisoner reentry, family preservation and faith-based programs, community centers, hair salons, and exercise clubs).24 Two implementation interventions were developed by academic and community partners. Program administrators, providers, and community and academic study leaders in CEP used an intervention work group to conduct biweekly meetings for 5 months to develop and implement a written plan for engaging groups of providers in training based on existing QI toolkits (eg, clinical assessments, medication management, case management, cognitive-behavioral therapy manuals, and patient education materials), with efforts to build a coalition for depression care.14 By contrast, RS provided technical assistance through written and online resources and a series of 12 “train-the-trainer” webinars, plus primary care site visits to support toolkit implementation. Relative to RS, CEP increased the participation of eligible staff in depression training.25 Over 6 months, relative to RS, CEP reduced the proportion of individuals who had poor MHQOL. For secondary outcomes, CEP (relative to RS) increased physical exercise, reduced risks for homelessness, and reduced behavioral health–related hospitalizations,14 while shifting outpatient visits for depression away from specialty mental health medication management toward depression services provided at faith-based centers and park-based programs.23 Over 12 months, there were modest continuing benefits of CEP over RS in terms of a reduced proportion of individuals with poor MHQOL and fewer hospital stays for behavioral health, but some findings were sensitive to modeling assumptions in the data analysis.23 There were no significant comparative intervention effects on depressive symptoms or use of health care depression treatments over 6 to 12 months. CPIC is noted in a Cochrane review as the only study of the added value of a community coalition intervention to improve the health of minority communities over a noncoalition comparator.26
This article examines the extent to which the effects of CEP relative to RS on primary and secondary outcomes are confirmed among poor or nonpoor clients; it also explores effects of CEP over RS within 3 mutually exclusive subgroups of poor: (1) justice-involved, (2) clients at risk for homelessness but not justice-involved, and (3) other poor (ie, neither homeless nor justice-involved). Community leaders requested the inclusion of these subgroups during the study design process. For these analyses, health policy partners recommended examining mutually exclusive subgroups, focusing first on those who are justice-involved as understudied and relevant to policy, homeless who are not justice-involved as a distinct group for similar reasons, and other poor to provide findings across all impoverished participants. Subgroup analyses were considered exploratory and were done to inform future research and potentially clinical practice. Secondary outcomes explore related social risk factors, including stressful life events or difficulties, the reduction of which may improve mental health outcomes.27 Given the dearth of literature on socially disenfranchised groups, these secondary analyses, albeit exploratory, can provide guidance to the field.
Methods
CPIC, a group-level randomized comparative effectiveness trial, assessed the effectiveness of CEP compared with RS14 using a community-partnered participatory research approach28 that includes community and academic partner coleaders with equal authority in all aspects of research. The institutional review boards of RAND Corp. and participating agencies approved procedures. CPIC was funded (2007) before the inclusion of health services interventions in required clinical trial registration and was not considered a clinical trial by the National Institutes of Health since no experimental treatments were included. The design of the study is described elsewhere.14,23,24,29
Sampling
Agencies that serviced at least 15 clients/week, had ≥1 staff member, and were not focused exclusively on psychotic disorders or home services were sampled from 2 Los Angeles (LA) communities (South LA, Hollywood Metro) with high minority populations and poverty rates. Comprehensive lists of services agencies and community partner recommendations within each community were used to identify mental health, primary care, substance abuse, social services, homeless, and other community agencies. Sixty eligible agencies were offered the opportunity to participate; 133 of 194 programs within those agencies were potentially eligible. Within each community, programs or clusters were paired based on geographic location, service sector, size, population served, services provided, and funding. One of each pair was randomized to CEP and the other to RS. Following site visits to finalize enrollment, 93 programs from 50 consenting agencies were enrolled.
Within programs, clients were screened in waiting rooms or at events from March to November 2010. Staff who were blinded to agency intervention status approached 4645 adults (age ≥18 years) on 2 or 3 days/program; 4440 (95.6%) agreed to be screened for depression. Eligibility for the study included the ability to provide contact information and being depressed (eg, score ≥10 on the 8-item Patient Health Questionnaire).30 Of the 4440 adults screened, 1322 (29.8%) were eligible and 1246 (94.3%) were enrolled. Of the enrolled clients, 981 completed a baseline telephone survey; 759 and 733 completed the 6- and 12-month follow-up surveys, respectively. The CONSORT flowchart detailing inclusion and exclusion is presented elsewhere.14 The most common reason for exclusion at different waves of follow-up was inability to contact.
Our analytic sample includes 1018 participants (81.9% of those who consented minus those who died) with baseline or follow-up data, of whom 268 lived above and 750 below the federal poverty level (FPL). Of the 750 poor participants, we created mutually exclusive subgroups in order of decreasing social disenfranchisement: (1) clients arrested or on probation in the 6-months before baseline (n = 158) (“justice-involved”); (2) clients who were homeless or had ≥2 risk factors for homelessness, but who were not arrested or on probation in the 6-months before baseline (n = 298); and (3) the remaining clients living below the FPL (n = 294) (“other poor”).
Baseline Measures
Baseline measures include age, sex, education, race/ethnicity, current marital and work statuses, health insurance coverage, number of chronic medical conditions, and intervention assignment. Responses from the Mini-International Neuropsychiatric Interview31 were used to create an indicator for 12-month major depressive disorder. Alcohol abuse or use of illicit drugs were assessed by the 3-item AUDIT32 and the 10-item Drug Abuse Screening Test33 at follow-up.
Outcome Measures
Health Outcomes
For CPIC, the 2 preplanned primary mental and physical health outcomes were (1) poor MHQOL, defined as 1 standard deviation below the population mean on the 12-item Short Form Health Survey Mental Health Composite,34 and (2) reduction of depressive symptoms on the 9-item Patient Health Questionnaire (PHQ-9).35 Following stakeholder input, community-prioritized outcomes were developed for mental wellness and good physical health.10
Social Outcomes
Community-prioritized social outcomes included current homelessness or living in a shelter, or having multiple risk factors for future homelessness (eg, no place to stay for ≥2 nights in the past 6 months or eviction from a primary residence, financial crisis or food insecurity in the past 6 months). Expanded social factors included report of worry about cost keeping one from getting help for emotional problems and total number of life difficulties experienced of a total of 15 (eg, no place to stay, eviction, witness of violence, lost custody of a child, death of a loved one, worry that food would not last, arrest or on probation, fired or laid off, reduced work hours, loss of health insurance, financial problems, gave up trying to find work, arguments with others at work, arguments with others at home, and arguments with others not living with you).
Service Utilization
Use of health services over a 6-month period was evaluated before the 6- and 12-month follow-ups based on client self-report. Outcomes include utilization of the formal health care sector: nights spent in a behavioral health hospital; any mental health specialty (MHS) outpatient visits during which advice was given about medication for emotional or mental health problems such as depression; number of visits where counseling was received; any use of outpatient substance abuse services or stay in a residential treatment program for substance abuse problems; and any use of outpatient primary care or public health clinics (primary care provider). The number of visits in the informal sector (social services, religious institutions/programs, parks, hotlines, other, but not including mental health self-help) where clients report receiving help for depression or emotional problems was also included. Finally, the total number of outpatient contacts for depression in all sectors (formal and informal) was included.
Statistical Methods
Our analytic sample included 1018 individuals who completed ≥1 survey at baseline, 6 months, or 12 months. We used item-level imputation for missing data and wave-level imputation for missing surveys to adjust findings to the observed analytic sample. We used weights to account for nonenrollment among eligible clients and attrition.14 All analyses, which were conducted using SUDAAN version 11.0.1 (RTI International, Research Triangle Park, NC: http://www.rti.org/sudaan/), account for clustering (clients within programs),36 weighting, and multiple imputations.37
We estimated 2 models using linear regression for continuous variables, logistic regression for binary variables, or Poisson regression for count outcome variables. Model 1 used the total sample (N = 1018) and estimated a poverty-by-intervention (CEP relative to RS) interaction model adjusted for age, education, race/ethnicity, 12-month depressive disorder, community, and the baseline status of each outcome. Model 2 was fit to the subset of those in poverty (n = 750) and included indicators for justice-involved, homeless but not justice-involved, and other poor, each of which interacted with intervention status using the same set of predictors. Comparisons focused on the effects of interventions within subgroups (above or below the FPL and the 3 nonoverlapping subgroups among those below the FPL).
Viewing our analyses as encompassing aspects of what Benjamini and Yekutieli38 described as a problem in subgroup analysis regarding the comparison of 2 conditions and a problem involving multiple end points, we built on the false discovery rate (FDR) framework of Benjamini and Hochberg39 as extended by Yekutieli and Benjamini40 by using FDR-adjusted P values in interpreting results across a large number of regression analyses.38⇓–40 FDR assesses whether the number of results with a significant unadjusted P value (ie, without adjustment for multiple comparisons) is disproportionate to what would be expected as a result of chance variation; the FDR-adjusted P value (PFDR) can be interpreted as the threshold FDR value below which the particular null hypothesis in question would be rejected. Results with PFDR < .05, which adjusts for multiple comparisons, are considered to be convincing evidence of a difference; consistent with rationales for adapting the interpretation of significance findings to contextual factors such as the nature of the intervention and the size of the sample,41 higher alternative PFDR thresholds are considered as signaling suggestive evidence within a sensitivity analysis framework.
To broaden our understanding of patterns, we report Cohen's effect sizes (ESs).42,43 We reflect ESs through standardized predictions by intervention and subgroups generated from fitted regression models.44 We also provide between-group differences, odds ratios, and incidence rate ratios with 95% confidence intervals (CIs). For continuously scaled variables, ESs are estimated as the difference between 2 adjusted intervention group means divided by a pooled standard deviation from a random effects model accounting for the cluster-randomized design.45 For dichotomous outcomes, we provide Cohen's effect size index h, defined as the difference between 2 arcsine-transformed proportions,43 where h values of 0.20, 0.50, and 0.80 are used to reflect small, medium, and large ESs, respectively.
Results
Baseline and demographic characteristics for the CEP and RS groups are similar for those above and below the FPL. Those in poverty relative to the nonpoor had a higher percentage of unmarried clients, had less education above high school, worked for less pay, and had less health insurance coverage (Table 1). Within each poor, socially disenfranchised subgroup, baseline demographic and clinical characteristics were similar for the CEP and RS arms. The justice-involved were more likely than the other subgroups to be male, not be working for pay, not to have health insurance, to have more total life events, and to have misused any drugs in the past 6 months. Other differences between groups are shown in Table 2.
Baseline Characteristics of Depressed Clients by Family Income Status
Baseline Characteristics of Depressed Clients Whose Family Income Is Below the Federal Poverty Level (n = 750), by Vulnerable Subgroups
Model 1
There are no significant intervention differences in primary outcomes (percentage of clients with poor MHQOL or PHQ-8 scores) among the nonpoor or poor (Table 3). Regarding community-prioritized outcomes among the poor, CEP over RS improved mental wellness at 6 months (P = .004; PFDR = .027; ES = 0.27), with suggestive evidence of better physical health at 12 months (P = .019; PFDR = 0.115; ES = 0.20) (Table 3). Regarding other secondary outcomes, there was suggestive evidence for reduced worry about cost of mental health services at 12 months among those in CEP relative to those in RS (P = .033; PFDR = .115; ES = 0.19). The percentage of clients hospitalized for behavioral health or with any mental health outpatient visits did not differ significantly by intervention status for clients in poverty, nor did the percentage attending any outpatient or residential substance abuse services (Table 4). There was suggestive evidence for a reduced number of MHS medication visits at 6 months among the poor in the CEP compared with RS (P = .006; PFDR = .059; ES = 0.30). At 12 months, primary care provider visits by the poor increased in the CEP group relative to the RS group (P = .004; PFDR = .049; ES 0.23). There is suggestive evidence that CEP increased mean depression-specific visits in community-based sectors at 6 and 12 months compared with RS (6 months: P = .032; PFDR = .159; ES = 0.15; 12 months: P = .033; PFDR = .163; ES = 0.14). No significant unadjusted P value differences by intervention were found within the nonpoor sample, and none of the interactions between poverty and intervention status emerged as significant in model 1.
Health and Social Outcomes by Poverty Status and Intervention Assignment (Model 1)
Service Utilization By Poverty Status and Intervention Assignment (Model 1)
Model 2
Among subgroups of poor clients, no significant intervention differences were seen in poor MHQOL or in PHQ-9 scores (Table 5). Regarding community-prioritized health outcomes, no significant intervention effects were found among justice-involved clients. CEP relative to RS showed suggestive evidence of improved physical health at 12 months for the homeless who were not justice-involved (P = .042; PFDR = .196; ES = 0.27) and improved mental wellness at 6 months for other poor (P = .009; PFDR = .062; ES = 0.33). For the justice-involved group, CEP relative to RS showed suggestive evidence of reducing the percentage who were currently homeless or who had multiple homelessness risk factors at 6 months (P = .036; PFDR = .126; ES = 0.46) and evidence of reducing total life difficulties (P = .017; PFDR = .120, ES = 0.44). An exploratory analysis showed evidence of an increase in having no place to stay for ≥2 nights in a row in the past 6 months for 26 of 88 clients in the RS group (95% CI, 15.0–36.4) versus 13 of 69 clients in CEP (95% CI, 5.6–20.6). Similarly, on the single item asking about arrests or probation in the last 6 months, there was a substantial difference from an average of 35 of 88 clients in the RS group (95% CI, 20.1–49.1) to 19 of 69 clients in the CEP group (95% CI, 8.7–28.9). There were no significant findings for social outcomes among the homeless but not justice-involved or other poor subgroups. Regarding utilization outcomes, the percentages of behavioral health hospitalizations, any mental health outpatient visits, and any outpatient or inpatient substance abuse services did not differ significantly by intervention status for subgroups (Table 6). For the justice-involved group, CEP showed suggestive evidence of reducing the number of both MHS medication visits (P = .038; PFDR = .192; ES = 0.31) and mental health counseling visits (P = .007; PFDR = .066; ES = 0.48) at 6 months. There are no significant utilization outcomes for the homeless who are not justice-involved. Finally, for the other poor subgroup, CEP showed suggestive evidence of shifts in utilization toward community services, such as an increased number of visits in informal sectors in which the client received help for depression or emotional problems at 12 months (P = .050; PFDR = 0.281; ES = 0.21) (Table 6). Detailed results for Model 1 and Model 2 with P values adjusted for false discovery rates for all outcomes are reported in the Online Appendix.
Health and Social Outcomes by Intervention Assignment Among Mutually Exclusive Client Subgroups Below the Federal Poverty Line (Model 2)
Service Utilization by Intervention Assignment among Mutually Exclusive Client Subgroups Below the Federal Poverty Line (Model 2)
Discussion
This article examines whether CPIC findings regarding the effects of community engagement and planning, over and above resources for services for the whole sample, also apply among impoverished clients of interest in under-resourced communities, with the nonpoor as a reference group. It also explores effects for low-income subgroups to inform future research. Overall, the pattern of findings suggesting some improved health under CEP compared with RS apply to the poor in terms of mental wellness but not for the primary outcomes of MHQOL and PHQ-9 score; comparable patterns of mental wellness for the nonpoor were not significant given the modest sample size. This is similar to the earlier findings of the Health Insurance Experiment, where variation in the amount of insurance coverage among low-income groups with poor mental health affected psychological well-being rather than distress.46 Low-income groups may be more comfortable reporting wellness than distress; or the network-based CEP may have specifically enhanced resiliency over the technical assistance–based RS. Evidence suggesting increased physical activity among the poor under CEP compared with RS was also consistent with overall findings within the sample. In addition, as a new suggestive secondary finding, CEP reduced worry about costs of mental health services, perhaps because of more depression training for cheaper, nonlicensed providers in CEP compared with RS.25
In exploring intervention effects for specific poor subgroups (eg, justice-involved, homeless but not justice-involved, other poor), we found no significant effects on preplanned primary outcomes, but we did observe suggestive evidence of ≥1 improved secondary outcome under CEP over RS for each subgroup: reduced homelessness risk factors and life difficulties for the justice-involved, improved physical health for the homeless but not justice-involved, and greater mental wellness for the other poor. Of note, under the community engagement arm relative to RS, fewer justice-involved clients reported having no place to stay, arrests, or being on probation. For the justice-involved, many of the intervention effect sizes were large even when not statistically significant (eg, ES = 0.41 for mental wellness compared with ES = 0.33 among the other poor and ES = 0.14 among the homeless but not justice-involved). We also found evidence suggesting fewer MHS visits for medication and counseling in the justice-involved group and of increased use of community-based depression services among the other poor, both of which are similar to the overall sample. These exploratory findings suggest that it may be important to examine the effects of CEP compared with RS to improve outcomes for depressed clients in larger samples of poor subgroups, especially those who are justice-involved, who tend to have many social and health risks: they are more likely to be homeless47; to die of a drug overdose; to commit homicide or suicide or be in an accident after release48,49; and to have high levels of chronic medical, mental health, and substance use problems while in prison.35 In our sample >65% of the justice-involved group are either currently homeless or had multiple risk factors for homelessness at baseline. Since engaging homeless and justice-involved populations was important for CPIC community and policy partners, several enrolled programs provided homeless and/or prisoner reentry services. The development of a community coalition across health care and community-based service programs—coupled with training in approaches to screen, refer, support coping, and deliver services to depressed clients—may have helped program staff alleviate stressors (eg, lack of food, shelter, clothing) or avoid arrest for those in a (depressed) population that is often difficult to engage in services. This potential to redirect therapeutic skills to enhance social services provision was mentioned by homeless agency staff during training and noted by community leaders while reviewing study findings. Such “task enhancement” may be an important area for future research on low-income populations.
Our findings have limitations. Only 2 communities in 1 large urban area are included and may not generalize to other regions. Interventions were randomized at the program level within the same communities, with potential for contamination (clients receiving services from both CEP and RS agencies). This would reflect a conservative bias (underestimate intervention effects). We are limited by sample size, especially for subgroups. We had multiple secondary outcomes but attempted to mitigate this challenge by reporting ESs and PFDR along with traditional statistical significance (ie, unadjusted P values). The strongest finding yielded a PFDR of .027 for mental wellness in subjects living below the FPL; many findings considered as suggestive evidence depended on relaxing the PFDR threshold to .10 to .20. We did not have data on important factors such as the chronicity of homelessness, the type of criminal offenses, enrollment in a medical home, and the severity of medical illnesses. Data are based on self-report. Future research is needed using larger sample sizes, more diverse geographic populations, the randomization of whole communities to minimize contamination, and the context of long-term outcomes for systems and clients. The results concern the added value of a community-engaged approach over technical assistance to support agencies in depression QI, not the value of treatment or services delivery per se. In this context one may expect a small ES, and it may be surprising that we observed some sizeable differences.
Conclusion
The increase in insurance coverage for lower-income populations through the Affordable Care Act provides an opportunity to incentivize community health homes, such as integrating primary care in behavioral health settings,50 with the potential to reduce disparities in access to and quality of care. Our findings suggest potential areas for future research on coalition approaches to improve outcomes for low-income clients with depression, particularly those who are also justice-involved. The burden of depression in impoverished and vulnerable persons and its collateral effects on communities underscore the need to understand how best to mobilize limited public health and social resources. Compared with more standard technical assistance, CEP offers an approach to better implement depression QI while addressing known disparities in care and outcomes for low-income, depressed clients through health care–community partnerships. The approach and findings of the study presented here recently helped provide an evidence-based foundation for a services model in LA County's Health Neighborhood initiative, which promotes services coordination, quality, and early intervention in mental health.51 Given the multiple health and social needs of vulnerable, low-income clients in under-resourced communities, even modest gains across health and/or social indicators can signal important improvements from coalition efforts.
Acknowledgments
The authors thank Dr. Mitchell H. Katz, Dr. David A. Ganz, and Susan Hirsch for their careful reading of and editorial comments about this manuscript.
Appendix 1. Detailed Results for Model 1 and Model 2 with P Values Adjusted for False Discovery Rates for All Outcomes
Health and Social Outcomes by Poverty Status and Intervention Assignment (Model 1) with P Values and the P Values Adjusted for False Discovery Rates
Service Utilization By Poverty Status and Intervention Assignment (Model 1) with P Values and the P Values Adjusted for False Discovery Rates
Health and Social Outcomes by Intervention Assignment Among Mutually Exclusive Client Subgroups Below the Federal Poverty Line (Model 2) with P Values and the P Values Adjusted for False Discovery Rates
Service Utilization by Intervention Assignment Among Mutually Exclusive Client Subgroups Below the Federal Poverty Line (Model 2) with P Values and the P Values Adjusted for False Discovery Rates
Notes
This article was externally peer reviewed.
Funding: These analyses of Community Partners in Care data were supported by the National Institute on Minority Health and Health Disparities (grant no. R01MH078853; P.I.s JM, KBW) and the Patient-Centered Outcomes Research Institute (award no. 1845; P.I. KBW). The parent study was funded by the National Institute of Mental Health (award no. R01MH078853, P30MH082760, and P30MH068639); the Robert Wood Johnson Foundation (award no. 64244), the California Community Foundation (award no. CMCH-12-97088), the National Library of Medicine (award no. G08LM011058), and the National Institutes of Health/National Center For Advancing Translational Science UCLA CTSI (award no. UL1TR000124). CAL is supported by the VA Quality Scholars program.
Conflict of interest: none declared.
Disclaimer: The contents of this article are solely the responsibility of the authors and do not necessarily represent the views of the funders.
- Received for publication September 15, 2015.
- Revision received January 19, 2016.
- Accepted for publication January 27, 2016.
