Mediators of the Impact of a Home-based Intervention (Beat the Blues) on Depressive Symptoms among Older African Americans

Abstract

Older African Americans (N=208) with depressive symptoms were randomly assigned to a home-based nonpharmacologic intervention (Beat the Blues, BTB) or wait-list control group. BTB was delivered by licensed social workers and involved up to 10 home visits focused on care management, referral and linkage, depression knowledge and efficacy in symptom recognition, instruction in stress reduction techniques, and behavioral activation through identification of personal goals and action plans for achieving them. Structured interviews by assessors masked to study assignment were used to assess changes in depressive symptoms (main trial endpoint), behavioral activation, depression knowledge, formal care service utilization and anxiety (mediators) at baseline and 4-months. At 4-months, the intervention had a positive effect on depressive symptoms and all mediators except formal care service utilization. Structural equation models indicated that increased activation, enhanced depression knowledge, and decreased anxiety each independently mediated a significant proportion of the intervention’s impact on depressive symptoms as assessed with two different measures (PHQ-9 and CES-D). These three factors also jointly explained over 60% of the intervention’s total effect on both indicators of depressive symptoms. Our findings suggest that most of the impact of BTB on depressive symptoms is driven by enhancing activation or becoming active, reducing anxiety, and improving depression knowledge/efficacy. The intervention components appear to work in concert and may be mutually necessary for maximal benefits from treatment to occur. Implications for designing tailored interventions to address depressive symptoms among older African Americans are discussed.

Introduction

Depression in older adults is well recognized as a debilitating condition that heightens the risk for functional decline, comorbidity, poor quality of life, dementia and mortality (Ciechanowski et al., 2004; Cuijpers, Beekman, & Reynolds, 2012; Glaser, Robles, Sheridan, Malarkey, & Kiecolt-Glaser, 2003; Lenze et al., 2001). Even mild to moderate symptoms, if not successfully treated, may lead to poor health outcomes and increased healthcare utilization and costs (Areán, 2006; Glaser et al., 2003; Grabovich, Lu, Tang, Tu, & Lyness, 2010; Lee et al., 2012; Lyness, Chapman, McGriff, Drayer, & Duberstein, 2009).

Older African Americans, one of the fastest growing minority segments of the aging population (Institute of Medicine [IOM], 2012), are at risk for depressive symptoms due to their high rates of chronic illnesses (e.g., heart disease, high blood pressure, diabetes) and exposure to other jeopardies and chronic adversities including poor access to needed resources, low income, poor housing and unsafe neighborhoods, as well as a history of discrimination, all factors that increase burdens and risk for depression (Areán et al., 2010; Pickett, Bazelais, & Bruce, 2013).

Recent estimates of prevalence rates for this group are higher than the 7% to 15% previously reported (National Institute of Mental Health, 2010; Woodward, Taylor, Abelson, & Matsuko, 2013; Zivin, Pirraglia, McCammom, Langa, & Vijan, 2013). In the African American Health Study of 998 community-dwelling African Americans, 21.1% had clinically relevant depressive symptoms (Miller et al., 2004). A survey of 150 older poor African Americans attending outpatient rehabilitation found that 30% scored positively for depression (Kurlowicz, Outlaw, Ratcliffe, & Evans, 2005). Similarly, in a survey of 153 urban African American senior center members, 24.2% reported mild to moderate depressive symptoms (Gitlin, Chernett, Dennis, & Hauck, 2012). Of 440 temporarily homebound older African Americans evaluated for depressive symptoms, 31% (n=137) screened positively for mild to severe symptoms (Gitlin et al., 2012).

Nevertheless, older African Americans with depressive symptoms continue to be underdiagnosed and undertreated in primary care and mental health clinics (Areán & Unützer, 2003; Tai-Seale, McGuire, Colenda, Rosen, & Cook, 2007). Contributing to their underrepresentation in depression care are system (lack of access to treatment and trained health providers) and person (stigma, lack of knowledge about symptoms, lack of trust of providers) level factors (Conner et al., 2010; Ell, 2006; Gum et al., 2009). Studies also show that older African Americans may prefer nondrug approaches and have poorer compliance to drug therapies than their white counterparts (Kales et al., 2012). Moreover, drug therapies may not be as effective for those with mild to moderate symptoms and of short duration (Nelson, Delucchi, & Schneider, 2013). Although a growing evidence base supports the use of nonpharmacologic approaches including collaborative and integrated care models (Unützer et al., 2002) and psychotherapies (Scogin, 2005; Peng, Huang, Chen, & Lu, 2009; Lee et al., 2012; Wilson, Mottram, & Vassilas, 2008), their efficacy for minority populations has not been extensively examined (Fuentes & Aranda, 2012). Thus, a limited evidence base and persistent mental health disparities have prompted a national call for new models of care that address access and treatment preferences for nondrug approaches and which evaluate mechanisms by which approaches are effective for minority populations (Alexopoulos & Bruce, 2009; Bartels & Naslund, 2013; Callahan & Hendrie, 2010).

Over the past decade, novel promising nonpharmacologic community and home-based care models have been tested. A systematic review of 23 effective models revealed that most were multi-component and included care management and a behavioral approach such as problem solving, cognitive behavioral or behavioral activation therapies, addressed unmet needs and provided participants specific skills to cognitively reframe and manage situational stressors such as poor health, housing or social needs (Gitlin, in press). These models are unified in their attempts to identify and tackle the multiple contextual factors that may contribute to or reinforce poor mood and which drug treatments cannot address (Frederick et al., 2007; Fuentes & Aranda, 2012; Gitlin, in press).

These models are also unified in their adoption of a social ecological framework for understanding the etiology of depression. This framework favors a behavioral explanation that links mood to daily life stressors and seeks to addresses contextual triggers of depressive symptoms such as unmet health and care management needs as well as helping participants become behaviorally activated and also reengage in pleasant activities or those meaningful to them (Kanter, Manos, Bowe, Baruch, Busch, & Rusch, 2010; Meeks, Looney, van Haitsma, & Teri, 2008; Scogin, Morthland, Kaufman, Chaplin, & Kong, 2011).

To address depressive symptoms in urban, low income older African Americans, we developed and tested the Beat the Blues (BTB) program. This multi-component home-based intervention was informed by previously effective interventions and findings from focus group and in-depth interviews conducted for the purposes of this trial to identify preferred treatment approaches. Treatment components included: care management (identifying unmet needs and deriving a plan of action), referral and linkage (making referrals to formal care and social service resources based on the care management assessment), education about depression and symptom recognition (instruction in early detection of symptoms, discussing feelings with physicians and those of a different race, therapeutic modalities including medication use and psychotherapy), stress reduction techniques (instruction in different approaches including deep breathing or counting, to reduce situational-induced anxiety), and behavioral activation (identifying daily routines, a behavioral goal, an action plan, introducing pleasant events in routines). As keeping busy and being engaged in activity was identified in focus groups and in-depth interviews as preferred coping mechanisms, we included behavioral activation as a treatment component versus more cognitive-based therapies such as structured problem solving (Agarwal, Hamilton, Crandell, & Moore, 2010). Each component was designed to address situational conditions that may be reinforcing depressed mood for this low resource population. Addressing unmet health management and daily living needs through care management and a behavioral activation plan may be particularly salient for this group (Areán et al., 2010). For example, this might include setting a goal for disease self-management and an action plan for how to achieve that goal which specifies the necessary behavioral steps.

Tested in a randomized trial involving 208 older African Americans, we found that study participants receiving the BTB program reported reduced severity in depressive symptoms with 43.8% in BTB vs 26.9% in the control group (p=0.020) in remission by 4-months. Following treatment, control group participants demonstrated benefits (4 to 8-months) similar in magnitude to the adjusted treatment effects for BTB participants in the first 4-months; and the initial BTB group maintained 4-month benefits at 8-months (Gitlin et al., 2013).

Although mounting evidence supports multi-component, nondrug, behavioral treatment approaches, the active ingredient(s) by which these approaches have their desired effects is unclear. Also unclear is whether all treatment components are necessary for benefits to be realized. Mediation analysis provides an analytic framework for evaluating underlying mechanisms or how interventions may work, whether intervention components are necessary, which components may need to be strengthened, and how to enhance intervention efficiency and design more effective and replicable treatments (Gitlin et al., 2000; Kraemer, Kiernan, Essex, & Kupfer, 2008; Roth & MacKinnon, 2012).

With few exceptions, mediational processes have not been examined in depression trials. Only a few caregiver studies addressing depression have examined mediators. These studies suggest different mediational processes may be operative depending upon treatment properties. A problem solving therapy for stroke caregivers found that perceived health, threat appraisals and use of rational problem solving significantly mediated the intervention effect on caregiver depressive symptoms (King et al., 2012). A counseling and support intervention for spouse caregivers of persons with dementia found that positive change in satisfaction with one’s social support network mediated a significant proportion of the impact of the intervention on caregiver depression; and this was further mediated by changes in caregiver stress appraisals (Roth, Mittelman, Clay, Madan, & Haley, 2005).

Only one trial to our knowledge that tested a home-delivered cognitive-behavioral therapy with 134 participants, most of whom were African American, has examined mediational processes. Scogin and colleagues (2007) found that while their intervention improved quality of life, cognitive and behavioral variables were not impacted and hence they did not mediate the positive benefits of the intervention.

Thus, it remains unclear as to the underlying mechanisms responsible for the effectiveness of home-based nonpharmacologic treatment approaches to reduce depressive symptoms in older adults, and specifically for African Americans. In this study, we applied a mediational framework to understand the active ingredient(s) of BTB. We examined four possible mediators of BTB’s impact on depressive symptoms: a measure of the change in levels of behavioral activation, or the extent to which participants were actively engaged; a measure of the change in depression knowledge and confidence in recognizing and managing symptoms; a measure of the change in anxiety levels; and a measure of formal care service utilization. These measures served as indicators of or broadly corresponded to each of the five intervention components; behavioral activation, depression knowledge/efficacy, stress reduction, and care management/ referral/ linkage respectively. As we did not have a direct measure of engagement in stress reduction techniques, we used state anxiety as an indicator. Also, as we did not have a direct measure of care management and referral/linkage components, we examined formal care service utilization as one indicator of these intervention processes.

The treatment components of BTB were conceptualized to complement and work in concert with each other. Thus, we expected each indicator of mediation to independently and significantly contribute to reducing depressive symptoms. We further expected that their joint effect would be even greater than their individual contributions.

Method

Participants

As described in depth elsewhere (Gitlin et al, 2012; Gitlin et al., 2013), a total of 208 African Americans were enrolled in the trial. Participants were >55 years of age, English speaking, cognitively intact (MMSE >24), and scored > 5 on the Patient Health Questionnaire (PHQ-9), a measure of depressive symptoms, on two sequential testing occasions. Other enrollment criteria to enhance study retention included having a home telephone and planning to live in the area for 8 months. Individuals were not eligible with a history of serious mental illness, life-limiting illnesses, involvement in another clinical depression trial, or who lived in assisted living or nursing home facilities. Anti-depressant or other medication use did not exclude participation.

The trial reflected a partnership between a senior center and an academic research center who shared recruitment, interviewing and intervention responsibilities. Recruitment targeted individuals enrolled in a short-term in-home support program for medically compromised individuals through on-going systematic screening, and the community at-large through media announcements and presentations at local events and social agencies.

Analyses for this present study were based on 179 participants (86.1%) who were retained through the 4-month assessment and provided complete data for all variables of interest in the mediation analyses. Of the 29 with incomplete data, 26 did not complete the 4-month follow-up (missing PHQ-9, CES-D and all mediators), 1 participant had missing data on the depression knowledge scale, 1 participant had missing data for both the behavioral activation and anxiety scales, and 1 had missing data for the CES-D and mediators. A comparison of the 179 participants included in the analyses to the 29 participants who were not, revealed no large or statistically significant differences at baseline on basic characteristics, indicators of mediation, and the two measures of depressive symptoms which were the outcome variables (all ps>.05).

Participants recruited from the two venues (in-home support group and community at-large) differed at baseline along certain characteristics as anticipated. The in-home group (n=51) participants were older, had more pain and more health conditions (ps<.05) compared to the community at-large group (n=128). For baseline depression measures, the two groups had similar PHQ-9 scores but did differ on CES-D scores (p=.04). The in-home group reported on average less symptomatology (Mean=13.06, SD=6.08) than the community at-large group (Mean=15.09, SD=5.73) on the CES-D, although both groups scored in a high range of symptomatology indicative of clinical depression. The groups did not statistically differ on the 4 measures of mediation. As we used a stratified randomization scheme based on recruitment source (in home support group vs. community at-large), all groups were comparable at baseline; that is, no large or statistically significant differences were found at baseline between the in-home participants assigned to experimental and control groups nor between the community at-large participants assigned to experimental and control groups.

Procedure

Enrollment procedures included two sequential depression screenings of individuals by trained senior center staff over two weeks using the Patient Health Questionnaire (PHQ-9). Those eligible (PHQ-9>5) and willing to participate in the trial provided written consent using an approved Institutional Review Board (IRB) form, completed a baseline home interview, and were then randomized to receive BTB immediately or 4 months later (wait-list control). All participants were reassessed at 4 and 8-months in their homes using the same interview battery conducted by assessors masked to participant group allocation. Participants did not incur any study-related expenses and were provided $15 for completion of each interview (baseline, 4 and 8 months) to recognize their time and participation in the study.

This paper used data from the baseline and 4-month assessments.

Intervention

Beat the Blues

The Beat the Blues intervention involved up to 10, one-hour sessions that occurred at home over 4 months and was conducted by licensed social workers who were trained in the protocol. The first few sessions focused on building rapport and assessing for unmet care management needs from which to derive a plan of action including referral and linkage to formal care, community and social services. Interventionists also educated participants about depressive symptoms and specifically, the link between behavior and mood, how to identify the onset of depressive symptoms and use specific strategies to manage symptoms early on including having discussions about symptoms with physicians and when physicians are of a different race. Participants also learned about the effects of stress and how to use a basic deep breathing technique before or during stressful points in the day. In the next few sessions, interventionists helped participants resolve identified unmet care management needs (medical, housing repairs, relocation needs, social, financial, benefits/entitlements), and worked on care management plans involving coordination, and referral and linkages to services if necessary. Additional stress reduction techniques (e.g., counting, music) were also introduced to provide other easy-to-use stress-reduction tools to address situational anxiety. In sessions 4 and 5, interventionists continued addressing care management needs and began behavioral activation exercises. First, interventionists reviewed daily routines and helped participants select a behavioral goal and a specific activity to add pleasure and personal satisfaction to daily or weekly routines. Active problem solving and motivational interviewing techniques helped participants achieve identified activity goals. Potential barriers to carrying out selected activities were identified and solutions derived that might also have required care management (arranging transportation) to engage in desired activities. Sessions 6-8 involved reinforcement of activity engagement, identification of new activity goals and specific steps to achieve them. In sessions 9-10, interventionists reviewed and reinforced all techniques, continued to help participants identify activity goals, break them down to achievable steps, and learn this process for future independent use.

All participants received all five treatment components. The content of each component reflected or was tailored to the participants’ specific care management and referral/linkage needs, level of depression knowledge and ability to recognize symptoms, preferred stress reduction techniques, and self-identified goals and behavioral activation plans. For example, although all participants were exposed to stress reduction techniques, if a participant expressed a preference, that particular technique was reinforced. Similarly, all participants were exposed to the same behavioral activation steps but the specific goals and action steps identified were individualized. Further, the treatment components were integrated such that a care management need could be the basis of a behavioral activation goal (e.g., preparing and eating the right foods to manage diabetes), and depression knowledge/efficacy was linked to self-management and becoming actively engaged in one’s own health care. Interventionists presented each of their assigned cases in weekly one-on-one supervisor meetings and bi-weekly group debriefing sessions to assure protocol adherence and to review care management, referral/linkage and behavioral activation plans.

Approximately 7 sessions involved care management and behavioral activation, 5 sessions involved teaching and practicing a stress reduction technique and about 4 sessions involved referral and linkage and depression knowledge/efficacy (Gitlin et al., 2013).

Wait-list Control Group

The wait-list control group did not receive any study-related intervention or contact following the baseline interview. Participants in this group, as those in the BTB group, were free to engage in any non-study services, depression treatments or programs they chose. After completing the 4-month assessment, participants received BTB in its entirety and as delivered to the initial group. This study examined mediational processes at 4 months only given that by 8 months, all participants received the intervention.

Measures

Measures included the following background characteristics: marital status (not married/married, living as married), living arrangement (alone/with others), sex, education (<high school, high school or >high school level of education), age, financial difficulty (0=not very difficult to 3=very difficult paying for basics), self-reported health conditions and use of depression, anxiety or pain medications (yes/no).

Outcome Variable

The primary dependent variable was depressive symptoms (main trial endpoint). We used two different scales with psychometric adequacy for African Americans in order to examine the robustness of our effects across two measures that tapped into slightly different types of depressive symptoms (Nguyen, Kitner-Triolo, Evans, & Zonderman, 2004; Huang, Chung, Kroenke, Delucchi, & Spitzer, 2006). The PHQ-9 provided a brief, psychometrically sound 9-item self-report severity measure (Kroenke & Spitzer, 2002). As the scale was administered on two subsequent occasions two weeks apart to determine eligibility, the second screen was used as the participants’ baseline value for the present analyses. A total severity score was calculated by summing responses across the nine items which were rated as occurring not at all (0), several days (1), more than half the days (2), nearly every day (3). Possible scores ranged from 0-27 (Cronbach’s α=.78 for this sample).

We also used the 10-item Center for Epidemiology Studies Depression Scale (CES-D; Santor & Coyne, 1997). Participants indicated symptomatology in the past week (0=rarely/never; 3=most or all of time). Scores were summed across items (range=0-30) with a score > 8 indicative of clinical symptoms. Higher scores indicated greater symptomatology (α=.77 for this sample).

There is some overlap of items for the CES-D and PHQ-9 (e.g., sleep quality, felt sad). However, CES-D items tend to emphasize affective components (e.g., felt lonely, enjoyed life), whereas PHQ-9 items tend to emphasize somatic aspects (e.g., poor appetite, trouble concentrating, moving or speaking slowly) of depressed mood.

Indicators of Mediation

The measures used as mediating variables included depression knowledge/efficacy, state anxiety, behavioral activation and formal care service utilization.

We assessed depression knowledge/efficacy using 10-items reflecting symptom awareness (“can identify depression symptoms”) and efficacy (“know how to explain feelings”) rated from 1=not at all confident to 4=absolutely confident. A total mean score was derived by summing across items and dividing by the number of items (range=1-4). Higher scores represented greater knowledge/efficacy (α=.72 for this sample; Cretin, Shortell, & Keeler, 2004).

To measure anxiety, we used the 10-item State Anxiety Scale. Participants rated feelings (“I felt calm,” “tense”) from 1=very much to 4=not at all. A total anxiety score was computed as the mean across all items (α=.85 for this sample). Higher scores indicated greater anxiety (Speilberger, Gorsuch, Lushene, Vagg, & Jacobs, 1983).

Activation was assessed using a modification of the Behavioral Activation Scale which included eliminating 14 items (“My work/schoolwork suffered...”) that were not relevant to this sample and rewording 3 items to heighten their relevance. Participants rated the resulting 17-items from 0=not at all to 6=completely. Items reflect broad content including positive engagement (“accomplished goal,” “engaged in activities,” “did things even though hard because fit with goals”), avoidance of difficult situations (“there were certain things that I needed to do that I didn’t do”), or dwelling on negative feelings (“...spent time thinking about my past, people who have hurt me, mistakes I’ve made..”). A total activation score was computed as the mean across items (α=.83 for this sample). Higher scores indicated greater activation (Kanter, Mulick, Busch, & Berlin, 2007).

The use/nonuse of formal care services in the past month was measured by 14 items reflecting 4 service types: psychosocial (e.g., talking to clergy, social worker, counselor, psychologist, psychiatrist, attending support groups), home health (e.g., assistance with daily chores, self-care), social (e.g., transportation), and health care (e.g., physician visits, outpatient clinics) services. A total utilization index was computed by summing the number of services used.

Statistical Analysis

The general analytic approach to examine the proposed mediators individually is illustrated in Figure 1. The primary aim of the analyses was to determine whether improvements in four mediating variables (depression knowledge, anxiety, behavioral activation, formal care use) independently and jointly, mediated the beneficial impact of the intervention on depressive symptoms at 4-months. There are multiple ways to statistically test for mediation effects even with relatively straightforward, two-wave (pre-post) data from a randomized trial. In particular, the b path illustrated in Figure 1 can fluctuate substantially depending on whether covariated-adjusted change scores or 4-month scores are analyzed as the effects (Roth & MacKinnon, 2012).

Figure 1.

Two-wave mediation model used to examine mediators individually. BL = Baseline observation. Δ = 4-month score minus baseline score.

In our analyses, simple change scores (4 month minus baseline) were calculated for the mediators (behavioral activation, depression knowledge, anxiety, formal care service use) and the depressive symptoms measures (PHQ-9, CES-D). These change scores were then analyzed as a function of their baseline values and the intervention effect as illustrated in Figure 1. The mediated or indirect effect represents the joint impact of the a and b paths on the outcome, and is estimated as a*b. The c’ path represents unmediated or direct effect, and the sum of a*b and c’ comprises the total (baseline-adjusted) effect of BTB on depressive symptoms. These estimates were tested for statistical significance and used to estimate the proportion of the total effect that could be attributed to that mediator ((a*b)/((a*b)+c’)). We also conducted stratified analyses for each recruitment source group (in-home support group and community at-large) separately. As consistent results across mediators and outcomes were found, we only report analyses for the entire sample.

After the mediators were examined individually, those that were identified as statistically significant mediators were examined simultaneously to test for independent versus overlapping mediation effects. The c’ from those models represented any intervention effect on depressive symptoms that was independent of the other mediating mechanisms and was used to calculate a jointly mediated proportion. All statistical analyses were conducted using Mplus Version 7 (Muthén & Muthén, 2012).

Results

Table 1 provides descriptive data on the background characteristics and the baseline measures of behavioral activation, depressive symptoms, anxiety and formal care service utilization for the 179 participants included in this study for both the total sample and by treatment assignment. For this sample, participants had a mean age of 69.44 (SD=8.57). Most were female (N=142, 79.33%), not employed (N=164, 91.62%), not married (N=157, 87.71%), had >high school education (N=86, 48.04%), lived alone (N=106, 59.22%), and reported financial difficulties (N=121, 67.60%). Participants reported an average of 6.56 (SD=3.05) health conditions such as high blood pressure (n=141, 78.77%), high cholesterol (n=106, 59.22%), arthritis (n=136, 75.98%), and diabetes (n=79, 44.13%). Most were not taking medications for mood: 38 (21.35%) reported anti-depressant and 29 (16.29%) anti-anxiety medication use. However, most were on a pain management drug (N=92, 51.69%) suggesting this sample had significant health concerns.

Table 1.

Baseline characteristics of analytical sample (N=179)

Characteristic	Total sample (N=179)	Treatment group (N=86)	Control group (N=93)	p-value
Age	69.44 (8.57)	69.00 (8.70)	69.84 (8.48)	.51
Sex, n (%)				.77
Male	37 (20.67)	17 (19.77)	20 (21.51)
Female	142 (79.33)	69 (80.23)	73 (78.49)
Education, n (%)				.35
< HS	40 (22.35)	18 (20.93)	22 (23.66)
HS/GED	53 (29.61)	22 (25.58)	31 (33.33)
> HS	86 (48.04)	46 (53.49)	40 (43.01)
Employment status, n (%)				.04
Employed	15 (8.38)	11 (12.79)	4 (4.30)
Unemployed	164 (91.62)	75 (87.21)	89 (95.70)
Paying for basics, n (%)				.18
Not difficult at all	36 (20.11)	12 (13.95)	24 (25.81)
Not very difficult	22 (12.29)	13 (15.12)	9 (9.68)
Somewhat difficult	69 (38.55)	33 (38.37)	36 (38.71)
Very difficult	52 (29.05)	28 (32.56)	24 (25.81)
Marital status, n (%)				.12
Not married	157 (87.71)	72 (83.72)	85 (91.40)
Married	22 (12.29)	14 (16.28)	8 (8.60)
Number of health conditions	6.56 (3.05)	6.73 (2.75)	6.40 (3.31)	.47
Antidepressant medication, n (%)	38 (21.35)	22 (25.58)	16 (17.39)	.18
Anxiety medication, n (%)	29 (16.29)	12 (13.95)	17 (18.48)	.41
Pain medication, n (%)	92 (51.69)	44 (51.16)	48 (52.17)	.89
PHQ-9 Score (second screen)	12.87 (4.96)	12.99 (5.27)	12.75 (4.67)	.75
PHQ-9 Score (second screen), n (%)				.39
Minimal/no depression (0-4)	0 (0.00)	0 (0.00)	0 (0.00)
Mild depression (5-9)	54 (30.17)	27 (31.40)	27 (29.03)
Moderate depression (10-14)	63 (35.20)	30 (34.88)	33 (35.48)
Moderate/severe depression (15-19)	41 (22.91)	16 (18.60)	25 (26.88)
Severe depression (≥ 20)	21 (11.73)	13 (15.12)	8 (8.60)
CES-D score	14.51 (5.89)	14.53 (5.88)	14.48 (5.93)	.95
Behavioral Activation	2.89 (1.00)	2.94 (1.01)	2.85 (1.00)	.57
Depression knowledge	3.11 (0.43)	3.09 (0.46)	3.13 (0.40)	.51
Anxiety	2.52 (0.65)	2.51 (0.66)	2.53 (0.64)	.83
Formal care service use	4.04 (4.90)	4.36 (6.19)	3.75 (3.31)	.41

Note: PHQ-9 = Patient Health Questionnaire; CES-D = Center for Epidemiologic Studies – Depression scale

Participants reported a moderate level of depressive symptoms on the PHQ-9 (Mean= 12.87, SD=4.96) with close to one third expressing mild symptoms, more than one third expressing moderate symptoms, close to one quarter indicating moderately severe symptoms and about 11% having severe symptoms. Similarly, for the CES-D, participants scored above the clinical cutoff for depressive symptoms and in the moderate symptom range (Mean=14.51, SD=5.89).

As to mediators, participants reported some to moderate anxiety (Mean=2.52, SD=0.65), some confidence recognizing symptoms (Mean=3.11, SD=0.43), low activation (Mean=2.89, SD=1.00), and using an average of four formal services (Mean=4.04, SD=4.90). (Table 1)

The correlations among the variables analyzed in the mediation models are displayed in Table 2. As expected, there were significant correlations across the different mediating variables, both at baseline and for the 4-month change scores. The two measures of depression were moderately correlated, both at baseline (r =.49) and in terms of changes over time (r = .50). This indicates some overlap, but also differences in the way depressive symptoms were measured in the two instruments. (Table 2)

Table 2. Summary of the Relationship of Mediators at Baseline and 4 Months.

		Baseline						Change at 4 month
		PHQ-9	CES-D	Behavioral Activation	Depression Knowledge	Anxiety	Formal care	PHQ-9	CES-D	Behavioral Depression	Activation Knowledge	Anxiety	Formal care
	Mean (SD)	12.87 (4.96)	14.51 (5.89)	2.89 (1.00)	3.11 (0.43)	2.52 (0.65)	4.04 (4.90)	−5.21 (6.01)	−4.53 (6.15)	0.63 (1.18)	0.25 (0.49)	−0.34 (0.69)	0.08 (2.67)
Baseline	CES-D	0 49***
	Behavioral Activation	−0 27***	−0 51***
	Depression Knowledge	−0.22**	−0.37***	0.43***
	Anxiety	0.35***	0.62***	−0.50***	−0.29***
	Formal care	−0.16*	−0.03	0.11	−0.04	−0.07
Change at 4 month	PHQ-9	−0.50***	−0.06	−0.00	−0.04	0.00	0.20**
	CES-D	−0.15*	−0.48***	0.16*	0.07	−0.23**	−0.01	0.50***
	Behavioral Activation	0.05	0.13	−0 55***	−0.12	0.17*	−0.05	−0.30***	−0.45***
	Depression Knowledge	0.08	0.16*	−0.17*	−0.52***	0.08	0.02	−0.34***	−0.43***	0.51***
	Anxiety Formal	−0.04	−0.20*	0.12	0.05	−0.50***	0.07	0.36***	0.56***	−0.46***	−0.38***
	care	0.09	0.16*	−0.09	−0.13	0.08	−0.18*	−0.07	−0.08	0.10 0.09	−0.06

^*p-value < 0.05;

^**p-value < 0.01;

^***p-value < 0.001

Note: PHQ-9 = Patient Health Questionnaire; CES-D= Center for Epidemiologic Studies – Depression scale

Effects of BTB on Depressive Symptoms and Mediators

Using standard analyses of covariance, with the baseline score as the covariate, the BTB intervention was found to have significant effects on both measures of depressive symptoms and on three (all ps < .0001) of four proposed mediating variables. In each case, the findings indicated that improvements were observed in the BTB group that exceeded any changes observed in the wait-list control group for behavioral activation, depression knowledge/efficacy and anxiety. There was no intervention effect for formal care service utilization. The significant intervention effects were previously published for the entire sample (Gitlin et al., 2013). The estimates for these effects on the mediators comprise the “a” paths in the mediation models and are provided in that column of Table 3. (Table 3)

Table 3. Summary of single mediator models on depression measures.

Outcome	Mediator	a	b	Ab	c’	ab/(ab+c’)
PHQ9	Behavioral Activation	0 71***	−1 21***	−0.86**	−1.74*	0.33
	Depression Knowledge	0.31***	−3 20***	−0.99**	−1.46	0.41
	Anxiety	−0.33***	2.69***	−0.89**	−1.67*	0.35
	Formal care	0.10	−0.07	−0.01	−2.53**	0.00
CESD	Behavioral Activation	0 71***	−1.82***	−1 29***	−1.93**	0.40
	Depression Knowledge	0.31***	−3.83***	−1 19***	−1.84*	0.39
	Anxiety	−0.33***	3 92***	−1.30**	−1.87**	0.41
	Formal care	0.10	0.00	0.00	−3 12***	0.00

^*p-value < 0.05;

^**p-value < 0.01;

^***p-value < 0.001

Note: PHQ-9 = Patient Health Questionnaire; CES-D= Center for Epidemiologic Studies – Depression scale; Refer to Figure 1 to understand letters (a, b, ab, c’, ab/(ab+c’) at top of columns.

Single Mediator Models

The unstandardized estimates that correspond to the paths illustrated in Figure 1 are reported in Table 3 for each mediator-depressive symptom measure combination. These findings indicate highly significant mediation effects for behavioral activation, depression knowledge, and anxiety (all ps < .01) for both depression measures. The mediation effects were similar in magnitude, with the proportion mediated effect ranging from .35 to .42.

Multiple Mediator Models

The single mediator models indicated that changes in three mediators (behavioral activation, depression knowledge, and anxiety) explained a portion of the intervention-induced changes on depressive symptoms. Figure 2 displays the standardized estimates of the effects from the model that examined intervention-induced changes on three of these mediators simultaneously and for both depressive symptom measures. Because the “a” paths did not change across the two models for different depressive symptom measures, only one estimate is provided for each “a” path in Figure 2. In both models, the unmediated or direct effect of the intervention was no longer statistically significant (−0.08 for PHQ-9 and −0.08 for CES-D) whereas the joint mediated effects explained more than 60% of the interventions’ total impact on depressive symptoms. A comparison of the standardized b paths indicated that changes in anxiety had somewhat stronger independent mediation effects than the other two mediators examined, although differences were not statistically significant.

Figure 2.

Multiple-mediator model of intervention effect on change in depressive symptoms. BL = Baseline observation. Δ = 4-month score minus baseline score. PHQ-9 = Patient Health Questionnaire; CES-D = Center for Epidemiologic Studies – Depression scale. Numbers on top refer to PHQ-9 whereas numbers in bracket refer to CES-D.

Discussion

To our knowledge, this is the first study to identify mediators of a multi-component, home-based intervention that reduces depressive symptoms in a resource strapped population, urban older African Americans. This group had significant health, pain and financial concerns and moderate to moderately severe levels of depressive symptoms. We examined whether the effects of four factors, depression knowledge/efficacy in symptom recognition, anxiety, behavioral activation, and formal service use mediated the relationship between intervention participation and the impact on depressive symptoms as measured by two indicators. Similar to our previous report involving the entire sample (Gitlin et al., 2013), results of these analyses confirmed that BTB had significant positive effects on depressive symptoms, and on three of four mediators. BTB reduced anxiety, and enhanced depression knowledge/symptom recognition and behavioral activation, all important outcomes for this group of older adults. The intervention did not impact formal care service use; that is, service use did not significantly increase or decrease as a consequence of intervention. We initially anticipated that service use would increase due specifically to the care management and referral/linkage activities of the BTB program. However, it appears that the intervention provided supportive care planning and referral and linkages without increasing utilization suggesting that inclusion of these two components may not result in increased service use and costs.

The mediation analyses provide an understanding of the pathways by which the intervention may have had its positive effect on reducing depressive symptom. A substantial proportion of the intervention impact on depressive symptoms was mediated by improvements in three of the four mediators examined. When the effects of the three mediators were examined jointly, the direct effects of the intervention on depressive symptoms were minimized, suggesting that the intervention had its effects by improving the ability of participants to understand depression and recognize symptoms, by reducing state anxiety and by enhancing activation. It would thus appear that multiple processes explain treatment benefit as each indicator independently and also conjointly contributed to improvements in mood.

In the psychotherapy literature, component analysis studies of cognitive-behavioral therapy identify behavioral activation as the more active ingredient with follow-up studies demonstrating it as an effective stand-alone approach (Cuijpers, Van Straten, & Warmerdam, 2007; Kanter et al., 2010; Hopko, Lejuez, Ruggiero, & Eifert, 2003; Jacobson et al., 1996; Jacobson, Martell, & Dimidjian, 2001; Lejuez, Hopko, & Hopko, 2001). In this study, for PHQ-9, behavioral activation was not a significant mediator in the joint model whereas anxiety reduction was the strongest mediating factor, although differences were not statistically significant. For CES-D, all three mediators jointly contributed to symptom reduction. As depression and anxiety may co-occur, strategies for managing state anxiety associated with stressful situations and life events may contribute to the reduction of depressive symptoms (Donker, Griffiths, Cuijpers, & Christensen, 2009; Hek, 2013). Further, as others have argued, for behavioral activation to be effective, a supportive and positive environment is necessary to help participants activate (Tursi, Baes, Camacho, Tofoli, & Juruena, 2013). Thus, our findings show that behavioral activation alone does not suffice. Enhancing education and symptom recognition and reducing anxiety appear to be necessary conditions to impact mood for this population.

As others have argued, addressing unmet care management needs followed by referral and linkage are strategies that can enhance the environment for activation (Areán et al., 2010). We believe care management and referral and linkage are critical components of the BTB program; however we were unable to demonstrate a mediational role for these components. This may be due in large part to the lack of an appropriate measure reflective of these components. We were able to examine one indicator, formal care service use which reflects only one aspect of care management. Future research should consider whether fulfilling unmet needs mediates intervention effects; unmet needs may be a better indicator of these two treatment components (care management and referral/linkage). Furthermore, as the care management and referral/linkage components identified unmet needs, provided resource ideas and areas for developing behavioral activation goals, it may be difficult to disentangle the specific effects of one intervention component from the other. For example, it may be that by addressing unmet needs, we also reduced anxiety. In this way, the indicators we used of mediational processes do not necessarily singularly align with each treatment component.

Taken as a whole, our results lend strong support for multi-component, nondrug approaches that help low income older African Americans address the contextual factors impinging on their mood. It appears that all treatment components are important and should be considered when replicating these findings with other minority populations or in different geographic regions. The results are consistent with behavioral theories suggesting that reducing situational stressors, educating about symptom recognition, and providing actionable plans leading to engagement in pleasant and/or meaningful activities can address mood disturbances and should be integrated in depression treatments for this population (Areán et al., 2010).

Noteworthy is that similar mediational results were obtained for each indicator of depressive symptoms, PHQ-9 and CES-D, lending support for the robustness of the findings. The two measures were moderately correlated suggesting overlap in symptoms but also differences in the aspects of depressed mood that were measured. Regardless, similar mediation relationships and effects were achieved suggesting that BTB impacts the breadth of depressive symptoms expressed by these measures. However, as noted above, although behavioral activation was significant in the independent model it was not a significant mediator in the joint model for PHQ-9. This may signify that behavioral activation impacts affective depressive symptoms more as emphasized by the CES-D items versus the more somatic aspects reflected by the PHQ-9 items.

Given the dearth of mediation analyses for home-based treatments, it is unclear how these findings compare to others. Scogin and colleagues’ (2007) cognitive behavioral intervention for rural older adults, most of whom were African American, could not account for treatment effects by cognitive or behavioral mediational processes. In contrast, for a dementia caregiver intervention involving individual and family counseling, change in satisfaction with social support mediated intervention effects for depressive symptoms (Roth et al., 2005). Thus, interventions may impact depressive symptoms through different processes depending upon the target population and properties of the intervention. No studies to our knowledge have attempted to identify indicators of components of complex interventions as we do here. Mediation analyses are an important analytic approach for understanding the relative contributions of treatment components and should be pursued in future depression trials.

Several study limitations should be noted. Our sample size was relatively small compared to large primary-care physician-based depression clinical trials. Nevertheless, this study compares favorably to intervention studies conducted in single community sites with sample sizes <600. A consequence of the small sample size, is the underrepresentation of men making it impossible to consider participant gender in this study. Previous research suggests that mediation processes may differ by gender (Park, Jang, Lee, Haley, & Chiriboga, 2013). As men and women experience depression differentially, examining gender effects could help to refine treatment components and tailor messaging (Gitlin et al., 2012).

Another related limitation is that participants were from one minority group and region. Our findings may not be generalizabale to other minority groups, geographic locations or rural populations. Additional research is warranted to determine if similar findings would be found for other older adults. Another limitation may be that we were unable to examine the long-term mediational effects as the wait-list control group received BTB after the 4-month followup. This limits our ability to determine whether our findings are sustained over time and whether the three factors continue to serve as the active ingredients impacting mood.

Although licensed social workers served as interventionists for this trial, BTB could be delivered by other mental health specialists. However, it is unclear whether peer educators or community health workers could provide BTB as knowledge and comfort with discussing mental health concerns, using motivational interviewing and behavioral activation techniques may require more training. An evaluation of a depression treatment with similar components as BTB implemented by care managers suggested that this group may be reluctant to use behavioral activation due to their lack of mental health training (Casado et al., 2008).

As a multi-component approach may be more impactful than a singular approach, future depression interventions should at a minimum include the treatment components reflected by the indicators tested here. One concern of a multi-component approach is that it may be more costly than drug therapy or a singular approach (Bosmans et al., 2013). A multi-component approach may require more training of interventionists and more treatment sessions to facilitate delivery of all treatment components. Another driver of cost is delivery in the home versus clinic. However, the cost of BTB, $146 per participant per month ($585 per person over 4-months) compares favorably to other drug and nondrug programs and has demonstrated cost-effectiveness (Pizzi et al., 2013). This is consistent with recent research showing that multi-component and home-based treatments can be cost-effective (Klug et al., 2010). Continued research is in order to evaluate the relative costs of its different treatment components.

Given the crisis in mental health care, novel care models that involve new partners, health providers and care sites and which complement, extend and serve as alternatives to primary care, are urgently needed to assure depression care is accessible and culturally appropriate to an increasingly diverse older adult population (Alexopoulous & Bruce, 2009; Callahan & Hendrie, 2010; IOM, 2012). BTB appears to confer an important benefit to an undertreated and resource limited population, older urban African Americans, and to consist of treatment components that jointly and effectively address depressive symptoms by tackling situational contingencies on mood. Based on the strong mediational relationships shown here, anxiety reduction, depression knowledge/efficacy, and behavioral activation appear to be important drivers of the effect of this intervention on depressive symptoms.