Predicting Engagement With Mental Health Peer Specialist Services

We used data from a two-site trial that randomly assigned 166 veterans to VA housing services or to housing services plus PS services (3). Participating veterans were invited, but were not required, to engage in 40 weekly sessions of PS services. After being connected with PS support, engagement varied widely, as measured by PS visits recorded in the electronic medical record. Study participants had a history of co-occurring mental and substance use disorders and were enrolled on a case management team called Housing and Urban Development–VA Supportive Housing (HUD-VASH), which is present at most VA medical centers. PSs were well connected with HUD-VASH staff, and they provided services at the VHA facility, in the veteran’s home, and in various community settings.

This analysis included only veterans assigned to PS services who had no missing data and who completed study interviews before and after the approximate one-year treatment period (N=71). Veterans in the analytic sample did not differ from those with missing data (N=5) or from noncompleters (N=9) on the basis of age, race, lifetime history of homelessness, or PS engagement (the dependent variable; all ps>0.05). In addition, the analytic sample did not differ from noncompleters (those with missing data) on other baseline covariates included in Table 1 except for level of social support (completers reported more social support), t=−2.62, df=83, p=0.01. Less than 5% of data were missing, and Little’s missing completely at random test indicated that missingness was completely at random (χ²=429.47, df=1,022, p=1.00). Study outcomes were assessed via in-person study interviews before and after participation in the PS intervention, but only the baseline assessment was used in this analysis.

In addition to collecting demographic data (age, race, gender), veterans were assessed at baseline with measures targeting predisposing, need, and enabling factors. Race was included as a predisposing factor (dichotomized as white or nonwhite). Because of lack of variability, age (80% age 50–66) and gender (94% of sample was male) were excluded. Other predisposing factors were lifetime homelessness, assessed with the Residential Stability and Homelessness Interview (coded dichotomously ≤1 year or >1 year on the basis of data distributions [11]); the Herth Hope Index (HHI [12]); and the Temple University Community Participation scale, an assessment of the total number of community activities participated in over the past 30 days (13). Measures that targeted need factors were the Behavior and Symptom Identification Scale (BASIS [14]) and the Addiction Severity Index (ASI [15]). The 24-item BASIS is a comprehensive, psychometrically sound measure of psychiatric symptoms. The analysis used the overall summary score that includes six domains: depression, interpersonal relationships, self-harm, emotional lability, psychotic symptoms, and substance abuse. From the ASI, a widely used, reliable, and valid measure of substance use problems, the analysis included lifetime alcohol intoxication and drug use (both in years).

One enabling factor was measured, the number of social supports, and was assessed with the following question: “Which of the following are sources of social support (i.e., help you get along) for you? (check all that apply).” Response options were summed among eight choices (such as romantic partner, friends, veterans’ organizations), and no support was coded 0. In addition to these factors, we extracted the following data for each veteran from the VA common data warehouse (CDW) as a measure of previous health service usage: six months prior utilization of VA homelessness services, substance use treatment, and mental health services. These three variables were combined to create a “total services used in the past six months” variable. The dependent variable, total number of PS services received during the study, was also extracted from the CDW. Analyses were performed with IBM SPSS Statistics (version 24) and SAS (version 9.4).

We calculated means, standard deviations, and frequencies along with bivariate correlations between all variables of interest. To predict engagement in PS services, we tested a negative binomial regression model (a generalized linear model using maximum likelihood estimation). We fit this model with a logged time offset to allow for overdispersion in the PS count outcome and to account for variable times to study completion. A test for overdispersion was confirmed by p>0.05 (95% CI=0.52–1.10; p=0.829). To improve model fit and enhance the detection of differences, outliers on the service use (N=2) and drug use (N=3) variables were adjusted by assigning a value for those variables to be one unit larger than the next most extreme score in the distribution, as recommended by Tabachnick and Fidell (16).

Distributions and model results are presented in Table 1. Engagement in PS services ranged from zero to 42 visits. Participants were older (mean±SD=54.0±8.04), and 55% (N=39) were nonwhite; almost two-thirds (N=45, 63%) had a history of homelessness for more than one year. Bivariate correlations revealed that PS services engagement was significantly associated with two variables, time homeless (r=−0.27, p=0.02) and past six-month service utilization (r=0.25, p=0.03), indicating those with greater engagement in PS services had spent less time homeless and used more services.

Model results indicated that time spent homeless (incidence rate ratio [IRR]=0.63, p<0.05), HHI scores (IRR=1.08, p=0.04), BASIS scores (IRR=2.22, p=0.02), and service utilization (IRR=1.26, p<0.01) were all statistically significant predictors of PS service use (Table 1). The BASIS baseline assessment had the most marked association with engagement, with a one-unit increase resulting in two times greater use in PS services, holding all other variables constant (higher BASIS scores indicate more psychiatric difficulty). In addition, veterans who had a shorter history of homelessness, who reported feeling more hopeful at baseline, and who used more services in the six months before baseline were more likely than other veterans to engage in PS services.

Predicting engagement with PS services is a key step in better understanding these services and how to best deploy them. Using Andersen’s model of health service utilization, this study elucidated who engaged with PS services. In partial support of our hypothesis, the demographic variable (race) representing predisposing factors did not predict engagement in PS services; however, other aspects of predisposing factors, such as increased hope and history of homelessness, did. Having hope that one’s life can improve is widely viewed as an essential component of recovery. In a seminal article defining recovery, Jacobson and Greenley indicated that hope can lay a groundwork for healing and can sustain continued effort when obstacles are encountered (17). This positive outlook and fortitude could explain why participants who rated themselves as more hopeful were also more likely to engage with PS services in addition to their regular VHA services. In contrast, a longer period of homelessness was a barrier to PS engagement, as hypothesized, within the group of predisposing factors. Although this study was not able to examine the barriers experienced by those with a longer homelessness history, reviews of studies have stated that lack of trust, differing priorities, and complex needs and service systems may play a role (9). It is possible that the study design did not provide enough time for engagement with this especially challenging group.

In regard to the need factors, greater perceived symptoms (evidenced by the BASIS score) predicted greater engagement in PS services. As the Andersen model of health service utilization would predict, veterans who perceived themselves to have more problems were more likely to engage with PSs. Service utilization significantly predicted greater engagement in PS services, consistent with our hypothesis. This result suggests that those who engage in behavioral health services are also more likely to engage in PS services. Finally, the enabling factor of social support did not appear to influence the receipt of PS services, contrary to our hypothesis.

Clinical staff often lack guidance about how to best use PSs who are available (1). Findings from these analyses suggest that where PS services are limited, it may be more effective for PSs to focus on patients who have also demonstrated some willingness to use services and who are more hopeful about their future. This approach could increase the effectiveness of PSs by reducing multiple efforts to engage those who have little likelihood of engagement. However, suggestions for how to best deploy PSs should be viewed in the larger context of the needs of the clinical team in which PSs work. For example, if a PS is working on a team dedicated to engaging individuals who are homeless, engagement would take precedence over a focus on those with greater hope and service utilization. At the same time, these findings suggest that further investigation is needed to help identify alternative ways to engage patients who tend to be less responsive to PSs.

The study had certain limitations. First, it used a small sample of veterans from one type of clinical service (HUD-VASH) and from only two VA medical centers. Future research is needed with larger samples across a wider array of clinical settings to learn more about the factors that predict engagement in PS services. Second, the relationship between the PS and veteran is most certainly bidirectional, influenced by characteristics (age or gender) of both the veteran and the PS. This study was unable to examine this relationship dynamically; thus, future research that attempts to understand how PS characteristics influence engagement is needed. Although choice of predictors was driven by the Andersen model of health service utilization, other variables that could also affect engagement—for example, length of time in the HUD-VASH program—were not included and would be important to incorporate in future studies. Finally, one cannot infer causality from these analyses. Although veterans with less history of homelessness who were hopeful, symptomatic, and service users were more likely to engage with PS services, it is possible that PSs made more efforts with those veterans. Despite these limitations, the findings are a starting point to help identify characteristics of individuals who are most likely to engage with PSs, consistent with the PS research agenda (2).