A Pilot Randomized Control Trial of a Dyadic Safety Planning Intervention: Safe Actions for Families to Encourage Recovery (SAFER)

Abstract

A recent systematic review on family and suicide prevention efforts identified a lack of family-based safety planning interventions for adults. To address this gap, The Safe Actions for Families to Encourage Recovery (SAFER) intervention was created. SAFER is a novel, manualized, 4- session, family-based treatment intervention that provides the tools and structure to support family involvement in Safety Planning Intervention (SPI) for Veterans at moderate risk for suicide. The SAFER intervention includes the use of psychoeducation, communication skills training, and development of a Veteran, and a complementary supporting partner, SPI. This Stage II (2aii) randomized clinical trial (RCT) evaluated the preliminary efficacy of this innovative and much-needed approach. Thirty-nine Veterans and an associated supporting partner were randomized to receive either SAFER or currently mandated (i.e., standard) individual Safety Planning Intervention (I-SPI).

Veterans in the SAFER condition as compared to I-SPI exhibited significant monthly decrements in suicide ideation as measured by the Columbia Suicide Severity Rating Scale (B=−0.37; p=.032). Moreover, a treatment-by-time interaction emerged when predicting improvements in Veteran suicide-related coping (B=0.08; p=.028) and supporting partner support of Veteran’s coping efforts (B=0.17; p=.032). However, the treatment effect for Veteran coping was not significant in dyadic analyses (B=0.07; p=.151) after controlling for the partner’s support (B=0.16; p=.009). Self-reported appraisals of relational factors and self-efficacy were not impacted by condition for either Veterans or supporting partners.

This initial efficacy pilot trial suggests that a brief dyad-based SPI has the potential to improve Veteran suicide symptoms and help family members support the Veteran’s coping efforts. However more intensive family work may be required for changes in self-perceptions of burdensomeness, belongingness, and caregiver perceptions of the Veteran as a burden. Nonetheless, SAFER’s discussion and disclosure about suicide symptoms facilitated more robust development of SPI for the Veteran and their accompanying supporting partner.

Suicide is the 10^th leading cause of death in the United States (US), with a global rate of 10.5 per 100,000 (WHO, 2019). Compared to the general US population, Veterans are at especially high risk for suicide (Kang, Bullman, Smolenski, Skopp, Gahm, & Reger, 2015). According to the most recent National Veteran Suicide Prevention Annual Report, since 2007 at least 16 Veterans per day died by suicide (VA, 2019). The suicide rates more than doubled for Veterans aged 18–34 between 2006 to 2016 (e.g., 22 deaths per 100,000 in 2006 to 45 per 100,000 in 2016; VA, 2018; VA, 2019). In March of 2019, the ‘President’s Roadmap to Empower Veterans and End the National Tragedy of Suicide’ (PREVENTS) task force responded to these statistics by establishing an Executive Order (EO 13861). This legislation called for the development of empirically validated interventions for Veterans at risk for suicide (Kemp & Bossarte, 2012).

One way that the Veterans Health Administration (VHA) has tried to combat suicide is through the implementation of the Safety Planning Intervention (SPI). The SPI is a mandated “best practice” that incorporates several steps an individual can take when resolving a suicide crisis (e.g., people or social setting that can provide distraction; Stanley & Brown, 2012; Stanley, Brown, Karlin, Kemp, & VonBergen, 2008; Stanley et al., 2018).

While the SPI has been implemented in multiple different contexts (e.g., individual and group settings; Goodman et al., 2020), there are no present guidelines specifying the involvement of family members in this intervention. Research suggests that Service Members with strong familial support have better mental health outcomes (Wilson, Gettings, Hall, & Pastor, 2015). Along with the support that these family members can provide, they also encourage treatment adherence and shed light on suicidal symptoms. Therefore, since family members are a part of the context in which suicide occurs, their involvement in SPI implementation could be especially beneficial (Frey & Cerel, 2015).

Despite family members’ critical role in protecting against suicide (Frey & Cerel, 2015), they lack education on how their behavior can help avert, or unwittingly aggravate suicidal thoughts and behavior. Even though they worry about their supportive partners they receive few resources on how they can help them (Grant, Ballard, & Olson-Madden, 2015). In addition to the lack of implementation guidelines, there is no real nuanced or detailed information on how to involve family in safety planning.

A substantial amount of literature has documented the influence of familial factors on clinical outcomes (Jobe-Shields, Flanagan, Killeen, & Back, 2015; Knobloch, Knobloch-Fedders, & Yorgason, 2019; Moriarty et al., 2015; Moriarty, Winter, Short, & True, 2018; Sayers, 2011; Solomon, Mikulincer, Fried, & Wosner, 1987; Tarrier et al., 1999). However, researchers have yet to fully understand the impact of familial relationships on suicide risk. For example, the Interpersonal Theory of Suicide highlights the critical role of thwarted belongingness and perceived burdensomeness in the development of suicidal thoughts and behaviors (Brenner et al., 2008; Monteith, Menefee, Pettit, Leopoulos, & Vincent, 2013). This theory highlights social relationships as both a protective factor for suicide (e.g., familial support providing a sense of belonging), and a risk factor for suicide (e.g., adverse feelings often found in a family context leading to more distress) in both civilians and Veterans (Joiner et al., 2005; Joiner et al., 2009).

There exists a robust literature detailing the benefits of couple and family-based treatments for mental conditions (Meis et al., 2013), consistently showing effectiveness similar or greater than corresponding individual treatments. In VHA, two models of family treatment exist with empirical backing. The first model is the SAFE program (Sherman, 2006), an 18-session family education program led by professionals, focusing mostly on Post Traumatic Stress Disorder. In contrast, the second model is led by trained family members from the National Alliance on Mental Illness (NAMI) in the Family-to-Family Education Program, teaching spouses, family members, and friends of Service Members about mental health conditions (Dixon et al., 2004). This was also adapted into a 6-session protocol by NAMI Homefront (Haselden et al., 2019; 2020). The success of these family-based interventions further strengthens the rationale for a dyadic approach to suicide safety planning.

With this increasing appreciation of family influence on mental health outcomes, and the beneficial impacts of family-based treatment, the VHA has begun to integrate family members in the care of Veterans (Sayers, Glynn & McCutcheon, 2014; Makin-Byrd, Gifford, & McCutcheon, 2011). Yet, despite the provision of family services by the VHA, high-quality programs are still not readily accessible (Fischer et al., 2015; Rutherford & Allegria, 2010). The Veterans Affairs Behavioral Health Autopsy Program (VA BHAP) Annual Report was able to emphasize this lack in family services and highlight the need to: 1) educate families about suicide warning signs; 2) improve communication between the Veterans and family members; 3) involve the family in the Veterans’ treatment to enhance support and trust; and 4) provide families with coaching on how to assist their loved one to seek help (VA, 2017). Even with the dissemination of these guidelines, to our knowledge, the current literature still does not provide family-based treatments for Veterans at risk for suicide.

Although there has been family treatment research addressing suicide risk with adolescents (Asarnow et al., 2017; Diamond et al., 2010; Diamond et al., 2019; Esposito-Smythers et al., 2019; Harrington et al., 1998; Spirito et al., 2015; Weinstein et al., 2018; Wharff et al., 2012; Wharff et al., 2019), this is not the case for adults. A recent systematic review on family and suicide prevention efforts identified a lack of family-based safety planning interventions for adults (Sullivan, Spears, Mitchell, Walsh, Love, & Goodman, 2021). While the VHA has provided national trainings that address the involvement of family members in suicide treatment (Glynn, personal communication), to date we are only aware of one intervention study that evaluates a relative-inclusive intervention for adult suicide risk (Anastasia, Humphries-Wadsworth, Pepper, & Pearson, 2015). This trial has promising results but lacks generalizability as the majority of participants were female inpatients. Thus, generalizability to a primarily male outpatient Veteran population is still unknown.

The SAFER Intervention

The Safe Action for Families to Encourage Recovery (SAFER) intervention is a family-based SPI treatment that was developed for Veterans at risk for suicide and their family members. This intervention aimed to fill the aforementioned gap in research and clinical care. Generally, the SAFER intervention was developed as a dyad-based treatment to help Veterans cope more effectively with their mental health symptoms, improve communication, decrease feelings of burdensomeness, and increase social support (e.g., a sense of belonging). This was all in the service of lessening Veteran suicide risk. Simultaneously, SAFER targeted family members by providing them with tools to help Veterans cope, improve their communication with one another, and decrease a family member’s own feelings of burden. For the purpose of this protocol, the family member involved was referred to as the “supporting partner.” Additionally, the Veteran, and their corresponding supporting partner in the SAFER intervention, were referred to as “Veteran-supporting partner dyads.”

Before conducting this RCT, six Veterans and their supporting partners participated in the open-label pilot testing of this intervention. During this time, the research team was able to finalize the manual and session content. In order to test the initial efficacy of the SAFER intervention, a small randomized clinical trial (RCT) was then designed to compare the SAFER intervention to a control condition (Individual-SPI; I-SPI). In this comparison condition, Veterans created a safety plan individually, which is standard care and VHA practice. In the case that it was recently created (e.g., in the last month), study clinicians reviewed the full SPI with Veteran and discussed revisions, as well as barriers in usage.

The SAFER intervention was developed as a manualized, 4-session, 90-minute, dyad-based, approach that includes psychoeducation about suicide risk and focuses on building a Veteran-supporting partner dyad safety plan (see Table 1). The first three sessions were conducted in consecutive weeks with the last session serving as a booster (three weeks after the third session). The entire treatment was completed within a six-week period.

Table 1.

SAFER Safety Planning Intervention for Veteran and Supporting Partner Dyad

Veteran	Supporting Partner
Step 1: Warning Signs	Step 1: Recognizing Warning Signs/Raising with Veteran
Step 2: Internal Coping Strategies	Step 2: Coaching Veteran on Use of Coping Strategies
Step 3: People and Social Settings that Provide Distraction	Step 3: Facilitating Veteran’s Use of Supportive Social Contacts
Step 4: People Whom I Can Ask for Help	Step 4: Providing Direct Support
Step 5: Professionals and Agencies to Contact for Help	Step 5: Facilitating Contact with Professionals/Agencies
Step 6: Making the Environment Safe	Step 6: Making the Environment Safe

Study Hypotheses

Specific hypotheses of this RCT included:

Hypothesis 1.

Veterans participating in SAFER (treatment condition) will experience reduced severity of suicidal ideation in comparison to Veterans completing I-SPI (control condition).

Hypothesis 2.

Veteran-supporting partner dyads participating in SAFER (treatment condition) will show improved suicide-related coping in comparison to I-SPI (control condition). Hypothesis 2A) Veterans will experience improvement in suicide-specific coping. Hypothesis 2B) The supporting partner will demonstrate greater resource knowledge and support of Veteran’s coping efforts.

Hypothesis 3.

Veteran-supporting partner dyads participating in SAFER (intervention condition) will show reductions compared to I-SPI (control condition) on interpersonal appraisals that are related to suicide risk. Hypothesis 3A) Veterans will experience reduced perceived burdensomeness and thwarted belongingness. Hypothesis 3B) Supporting partners will experience reduced caregiver burden.

Methods

Overview

The study was designed as an RCT comparing SAFER to I-SPI in a sample of 39 Veteran-supporting partner dyads struggling with recent suicidal ideation or past suicide behavior. Variables were evaluated at three time points: baseline, post-treatment, and extended follow-up. Due to the difficulty of recruiting this population, participants were re-contacted for a given assessment until it was finally completed. All procedures and instruments were approved by our institutional review board.

Procedure

Approval was obtained from the James J. Peters VA Medical Center Institutional Review Board (IRB) and the Research and Development Office.

Participants

Eligible Veterans must have been receiving care at the VHA and present at moderate risk for suicide, defined as: having endorsed recent passive or active suicidal ideation (within the past 3 months) or a lifetime history of a suicide attempt. Given the novel nature of the SAFER intervention, the sample intentionally was recruited for having moderate risk. Additionally, Veterans needed to bring in an available, consenting, and qualifying supporting partner to participate with them. Supporting partners needed to meet at least three (two for nonrelatives) of five inclusion criteria established by (Pollak & Perlick, 1991). The supporting partner had to 1) be a spouse, co-habiting significant other or parent; 2) have more frequent contact than others in the support network; 3) help to support the patient financially; 4) be the one contacted by treatment staff for emergencies; 5) have been involved in the patient’s treatment. These criteria were selected to operationalize the definition of supporting partner, given that many of our Veterans did not have blood relatives available to participate or were estranged from their family of origin. Others developed close relationships with friends or romantic partners who helped with aspects of caregiving, including financial, emotional, and logistical support.

Veterans and supporting partners were excluded if: 1) they had a history of suicide attempt in the past 3-months; 2) they presented with untreated or un-medicated active psychosis; 3) they presented with alcohol or drug abuse or dependence within the past 1-month based on the Patient Health Questionnaire (PHQ; Spitzer, Kroenke, & Williams, 1999); 4) they presented with medical condition or life event (e.g., an upcoming move to another state) that would compromise participation; 5) they participated in another family-based psychosocial intervention trial with the past 6-months; 6) they had limited English proficiency; or 7) if they were participating with a romantic partner and endorsed recent, “severe” intimate-partner violence based on the revised 20-item Conflict Tactics Scale Short Form (CTS2S; Straus & Douglas, 2004). Immediately after consent participants were screened for inclusion/exclusion with measures described below.

The study aimed to recruit, consent, and randomize 40 Veterans at risk for suicide and 40 corresponding supporting partners over a 30-month period from a large Veterans Affairs Medical Center (VAMC) in a major metropolitan area in the Northeast. Recruitment sources included VAMC suicide prevention coordinators, a Veteran’s primary clinician from the VAMC’s psychiatric inpatient unit or outpatient care center, and community outreach (e.g., Vet Centers and higher education institutions). Please see Figure 1 for the consort diagram of study participants from recruitment to study completion.

Figure 1.

CONSORT Diagram Describing Recruitment Process

Note. All waves available from all participants included in final follow-up analyses.

Procedure

Assessments

After screening and consent, Veterans and supporting partners completed separate in-person baseline assessments which included demographic data, Veteran and supporting partner screening assessments, and full study measures. Each Veteran-supporting partner dyad was then randomly assigned to participate in either I-SPI or SAFER via a one-to-one ratio using a block randomization scheme. Regardless of randomization, Veterans were allowed to continue with all their other VHA care. Outcomes were re-assessed immediately post-intervention and then again through a follow-up assessment at least 3-months post-intervention. To minimize attrition in this difficult to obtain and track sample, participants were initially contacted at the end of their intervention and then re-contacted until at least one member of the dyad provided the posttreatment assessment (Range=0.30–7.47 months from baseline; M=2.84 months). The follow-up assessment was timed between 3-months from that point up until one year from enrollment (Range=3.10–11.43 months from participants’ baseline; M=6.32 months).

Attrition.

Of the 39 Veteran and supporting partner dyads enrolled in the study, 30 Veterans (77%) and 27 supporting partners (69%) provided at least one wave of follow-up data. Examination of demographic predictors of attrition found that participants who provided follow-up did not significantly differ from participants who did not with respect to assigned condition, years of education, gender, ethnicity, race, employment status, nature of caregiving relationship, Veteran status of supporting partner, or any other baseline measures. However, age did significantly predict attrition (t(76)= −2.36; p<.02), with those providing follow-up being generally older (M=51.37 years) than those who did not (M=42.76 years).

Individual Suicide-Safety Planning (Control)

As noted above, all participants were able to continue in standard VHA care based on their own determined needs (psychiatry; psychotherapy; and case management by the suicide prevention coordinator). Veterans in the I-SPI control group either developed a full safety plan with a study clinician, or if one already existed, updated and reviewed this safety plan including a discussion of barriers in usage.

SAFER Intervention (Treatment Condition)

The SAFER intervention is a novel, manualized, 90-minute, dyad-based intervention, that typically includes a joining session, two treatment sessions, and a booster. Session content includes the use of psychoeducation, communication skills training, and revision and development of both the Veteran and a complementary supporting partner SPI. SAFER follows standard skills training session format and include: 1) brief check-in with a brief assessment of mood, the current level of suicidality and use of safety plan; 2) homework review; 3) teaching of new material and skill (e.g., validation, making a positive request, active listening); 4) in-class practice of the skill; and 5) assignment of homework/outside practice of skill/ development of safety plan. The SAFER treatment goal is to provide the tools and structure to support social support involvement in SPI for Veterans at moderate risk for suicide (Table 1).

Interventionists, Competence, and Treatment Integrity

The three study interventionists included a psychiatrist, psychologist, and mental health counselor, who each had extensive training in either cognitive behavior or dialectical behavior therapies, and suicide risk assessment. All interventionists participated in a one-day training to review treatment manual components, instruction in dyad therapy principles, and suicide risk management. The interventionists followed the manual and received supervision after sessions by the treatment developer.

A fidelity scale was developed to assess core features of SAFER’s structure, contents, and treatment principles, as well as general clinical competence (e.g., building rapport, crisis management, etc.) The session-specific items (n=1–4) and general competence items (n=13) were rated on a 4-point Likert scale (0 = unacceptable to 3 = excellent) by one trained rater. This training included reviewing components of the treatment manual, instruction in group therapy principles and group didactics, and suicide risk assessment. All intervention sessions were recorded, and a randomization scheme was generated to have seven sessions randomly selected for review. Clinicians were required to maintain a total score of 80% or above on each session to demonstrate acceptable adherence to the intervention. Clinicians whose ratings fell below this criterion were given additional supervision and their adherence was monitored until satisfactory adherence was regained. The average percentages on the adherence scale ranged from 82 to 100 with the mean score being 90.7. No interventionists ever dropped below the 80% adherent threshold.

Measures

Primary Outcome Measure (Hypothesis 1)

Veteran suicidal ideation.

Moderate suicidality was measured to determine eligibility using the Columbia Suicide Severity Rating Scale (C-SSRS). Posner and colleagues (2011) found the C-SSRS to have divergent validity, predictive validity, sensitivity, specificity, sensitivity to change, and internal consistency in a multisite study (Posner, et al., 2011). The psychometric properties of the C-SSRS across multiple studies provide evidence of moderate to strong validity and internal consistency of the ideation intensity subscale (Cronbach’s alpha 0.73–0.946; Posner et al., 2011). The baseline C-SSRS measured severity of suicidal ideation in the past three months, as well as the most severe in lifetime. The baseline C-SSRS measured suicidal behavior in the lifetime, specifically noting those within the past year. The C-SSRS was also used across time points to record level of ideation, lifetime suicide attempts, and recent suicide attempts. The follow-up version of C-SSRS measured suicidal ideation and behavior that had occurred since the last assessment. Research staff was trained by Dr. Barbara Stanley and Dr. Ainsley Burke, two of the developers of the assessment. Additionally, staff had weekly meetings in which the C-SSRS was discussed to ascertain consensus when scenarios warranted further discussion.

Suicide-Specific Coping (Hypothesis 2)

Veteran suicide-related coping (Hypothesis 2A).

Suicide-related coping was evaluated by the Suicide-related Coping Scale (SRCS), a 17-item self-report measure assessing one’s ability to cope with suicidal ideation and urges (Stanley, Green, Ghahramanlou-Holloway, Brenner, & Brown, 2017). The scale demonstrated good internal consistency (α=.85) at baseline; scores were summed so that higher scores represent greater confidence and breadth of approaches to coping with suicidal thoughts and feelings.

Partner’s support of suicide-related coping (Hypothesis 2B).

Supporting partners also completed five items adapted from the SRCS that assessed their confidence in their ability to support the Veteran through their suicidal urges. The five items were: 1)“I know the nearest hospital or urgent care facility where my loved one can go if I cannot handle the crisis on my own,” 2) “When my loved one is feeling suicidal or showing signs of suicidal thinking/behavior, there are places we can go to help take his/her mind off the problems,” 3) “I have several things I can do to help my loved one get through a suicidal crisis,” 4) “I am able to put aside my own fears and focus on taking appropriate actions when my loved one is feeling suicidal or showing signs of suicidal thinking/behavior,” and 5) “Seeking help from health care professionals is a good way to keep my loved one safe when he/she is feeling suicidal or showing signs of suicidal thinking/behavior.” Items were rated on a 0 (Strongly Disagree) to 4 (Strongly Agree) scale and summed for a total score. Higher scores indicate greater self-efficacy when supporting the Veteran through suicidal crises. In addition to face validity, the scale showed acceptable internal consistency in the sample (α=.78).

Suicide-promoting appraisals (Hypothesis 3)

Veteran perceptions of burdensomeness and thwarted belongingness (Hypothesis 3A).

Perceived burdensomeness and thwarted belongingness were evaluated by the Interpersonal Needs Questionnaire (INQ-15; Van Orden, Cukrowicz, Witte, & Joiner, 2012). Scores on each subscale are summed so that higher scores represent a greater degree of their respective constructs. In this sample, Perceived Burdensomeness demonstrated excellent internal consistency (α= .95) and Thwarted Belongingness showed acceptable internal consistency (α =.79).

Supporting partner’s experience of caregiver burden (Hypothesis 3B).

Caregiver burden was evaluated using the Caregiver Burden Inventory (CBI; Novak & Guest, 1989), a 24-item scale assessing caregiver burden in five areas: time, physical, social, developmental, and emotional burden. Items were summed to create an overall measure of caregiver burden and showed excellent internal consistency in this study (α =.94).

Analytic Strategy

We examined predicted trajectories on the outcomes of interest using slope-intercept trajectories based on Hierarchical Linear Modeling (HLM; Raudenbush & Bryk, 2002). Specifically, we created a two-level model in which repeated measurements over time were modeled at Level 1 (within-person change) and differences between participants/dyads were modeled at Level 2 (between-person change, including difference in condition) using the following equation:

$$\text{Outcome}={\pi }_0+{\pi }_1\ast \text{(months}\text{since}\text{baseline)}+E$$ Level 1 (Individual)

$$\begin{gathered} {\pi }_0={\beta }_{00}+{\beta }_{01}\ast (\text{Condition})+{\beta }_{02}\ast \text{(Veteran}\text{History}\text{of}\text{Attempt})+{\text{B}}_{03}\ast \text{(Relationship}\text{with}\text{Supporting}\text{Partner)}+{\beta }_{04}\ast \text{(Age)}+r_0 \\ {\pi }_1={\beta }_{10}+{\beta }_{11}\ast (\text{Condition})+{\beta }_{12}\ast (\text{Veteran}\text{History}\text{of}\text{Attempt})+{\text{B}}_{13}\ast (\text{Relationship}\text{with}\text{Supporting}\text{Partner})+{\beta }_{14}\ast (\text{Age}) \end{gathered}$$ Level 2

The Level 1 individual equation represents a slope-intercept model predicting each participants’ estimated score on an outcome baseline (π₀) and then calculates the linear change in that outcome by each month across the study (π₁). Slope-intercept trajectories are ideal for the present data given the variable assessment windows across all participants and the ability of HLM to use all available information from each participant to estimate these slopes using restricted maximum likelihood estimation, retaining the maximum number of participants and datapoints in the analysis even if some waves are missing. The Level 2 equation models respondents’ trajectories are based on assigned condition (SAFER=1; I-SPI=0), reflecting an intent-to-treat approach. The treatment effect of SAFER is modeled as the effect of condition on change in functioning over time (β₁₁) controlling for differences between conditions at baseline (β₀₁) and demographic variables(all other βs).

Due to limited power, we were sparing with our controls and selected two theoretical control variables based on theory and practice within suicide and caregiving treatment. Namely, we controlled for lifetime history of suicide attempt (0= No attempts; 1= One or more suicide attempts). This is a common control/stratification variable in suicide treatment research (e.g., McCall et al., 2019; Diamond et al., 2019) that accounts for unique risk factors associated with suicide attempt (Klonsky & May, 2015). This variable could account for possible non-response due to higher risk. We also controlled for the nature of the relationship between partners (1=Romantic; 0=All others) as a common control/stratification variable in the caregiving treatment literature (e.g., Gitlin & Rose, 2016; von Heymann-Horan et al., 2018). This could account for different natures of caregiver burden between romantic partners and other types of family members (Conde-Sala et al., 2010). This variable may also explain possible non-response due to breakups/separations (Mannion et al., 1994). Finally, we also included participant age as a third control based on our empirical data because this variable predicted dropout. Together, these three controls reduced within condition variability and reduced the impact of dropout on the model. Significant treatment effects were followed by recalculating simple slopes within the SAFER and I-SPI conditions.

Additionally, we replicated all significant treatment effects using a dyadic model to account for potential dependence between outcomes. To conduct these analyses, we extended our two-level model so that the lowest level includes separate rows for each person and wave that are all nested in dyads following the approach to dyadic HLM models set in Kenny and colleagues (2006).

$$\text{Outcome}={\pi }_1\ast (\text{Veteran})+{\pi }_2\ast (\text{Veteran})\ast (\text{months}\text{since}\text{baseline})+{\pi }_3\ast (\text{Partner})+{\pi }_4\ast (\text{Partner})\ast (\text{months}\text{since}\text{baseline})+E$$ Level 1 (Dyadic)

The Level 1 equation of the dyadic model uses two dummy terms to represent Veteran partners (1=Veteran, 0=Supporting partner) and supporting partners (1=Supporting partner, 0=Veteran) to model parallel trajectories for both Veterans and supporting partners. The level two equations were identical to the level two equations in the individual model. Modeling both the Veteran and supporting partner intercept as random effects accounts for potential dependence between Veterans and supporting partners on each outcome.

Results

Descriptives

Participant characteristics and descriptives at each wave can be found in Table 3. This sample was primarily middle-aged, more likely to be Black, somewhat more likely to be Hispanic/Latino and somewhat more likely to be unemployed than the general population. These demographic characteristics are consistent with the Veteran population in the recruitment region. The final sample of 39 Veteran-supporting partner dyads included a range of supports including 14 romantic partners, 13 other family members of Veterans, and 12 close friends. Although supporting partners were not required to be Veterans, a third of supporting partner (n=13) identified as Veterans themselves. Nineteen Veterans reported a history of one or more suicide attempts at baseline. As can be seen in Table 3, participants were generally similar across both conditions with respect to demographics characteristics as well as all outcomes. This highlights the effectiveness of random assignment in our small sample and suggests the groups are roughly comparable for treatment analyses.

Table 3.

Demographics and Outcome Measures by Condition

	SAFER (n=19 dyads)				I-SPI (n=20 dyads)
	Veteran		Supporting Partner		Veteran		Supporting Partner
Individual Demographics	n/M	(%)/(SD)	n/M	(%)/(SD)	n/M	(%)/(SD)	n/M	(%)/(SD)
Age	53.26	(12.37)	48.68	(16.24)	51.90	(15.62)	42.55	(12.8)
Years of Education	13.79	(1.13)	12.71	(2.34)	13.63	(2.55)	13.58	(2.17)
Male	17	(89)	6	(32)	16	(80)	8	(40)
Race
Black/African American	10	(53)	8	(42)	10	(50)	10	(50)
White	5	(26)	5	(26)	2	(10)	2	(10)
Native American	0	(0)	1	(5)	1	(5)	0	(0)
Multiracial	2	(11)	1	(5)	4	(20)	5	(25)
Other	2	(11)	4	(21)	3	(15)	3	(15)
Hispanic/Latino	5	(26)	6	(32)	8	(40)	8	(40)
Employment Status
Employed / Student	3	(16)	5	(26)	5	(25)	10	(50)
Unemployed/Unlisted	10	(53)	9	(47)	6	(30)	8	(40)
Retired/Disabled	6	(32)	5	(26)	9	(45)	2	(10)
			SAFER		I-SPI
Measures	Range		M	(SD)	M	(SD)	t (df)
Baseline Measures
Veteran Suicidal Ideation	0 – 5		3.26	(169)	3.75	(1.33)	−1.00	.32
Veteran Suicide-Related Coping	0–4		3.03	(0.71)	2.91	(0.65)	0.59	.56
Partner Support of Suicide-Related Coping			3.23	(0.81)	3.37	(0.87)	−0.52	.61
Veteran Thwarted Belongingness	1–6		3.23	(102)	3.62	(0.9)	−1.25	.22
Veteran Perceived Burdensomeness	1 −6		2.84	(1.56)	3.03	(1.61)	−0.38	.71
Supporting Partner Caregiver Burden	0–4		1.39	(0.92)	1.06	(0.96)	1.08	.29
Post-Treatment Assessment
Veteran Suicidal Ideation	0 – 5		1.50	(1.93)	3.50	(1.43)	−2.53	.02
Veteran Suicide-Related Coping	0–4		3.15	(0.71)	2.69	(1.02)	1.43	.16
Partner Support of Suicide-Related Coping	0–4		3.65	(0.46)	3.36	(0.71)	1.17	.25
Veteran Thwarted Belongingness	1–6		2.85	(1.27)	3.57	(1.22)	−1.58	.12
Veteran Perceived Burdensomeness	1 −6		2.23	(1.39)	2.60	(1.68)	−0.65	.52
Supporting Partner Caregiver Burden	0–4		1.23	(0.85)	1.00	(0.86)	0.67	.51
Extended Follow-up (3+ Months)
Veteran Suicidal Ideation	0–5		1.00	(1.31)	3.38	(2.13)	−2.68	.02
Veteran Suicide-Related Coping	0–4		3.12	(0.59)	2.70	(0.86)	1.38	.18
Partner Support of Suicide-Related Coping	0–4		3.70	(0.57)	2.89	(1.4)	1.54	.14
Veteran Thwarted Belongingness	1–6		3.30	(1.46)	3.45	(0.95)	−0.30	.76
Veteran Perceived Burdensomeness	1 −6		2.10	(1.63)	2.86	(1.42)	−1.19	.25
Supporting Partner Caregiver Burden	0–4		0.90	(0.59)	0.72	(0.67)	0.61	.55

In the SAFER intervention arm, participants on average attended 4.73 out of a possible 6 sessions (2 individual joining sessions; 4 joint sessions). Overall, 68% of participants completed the whole program. In the I-SPI (control arm), participants on average attended 0.9 out of one possible session and 90% completed the control arm intervention.

The bottom half of Table 3 includes the average scores for the scales at each wave divided by condition. The SAFER and I-SPI conditions significantly differed in suicidal ideation at both the post-treatment assessment and the 3-month follow-up assessment. However, further analysis of trajectories is needed to identify whether these distinctions between groups reflect genuine differences in trajectories of suicide ideation or simply changes in who responds at each wave. Among the 30 Veterans providing follow-up data, two Veterans reported a suicide attempt during their follow-up C-SSRS interview. Both of these Veterans were participating in the I-SPI condition, due to the low base-rate of suicide attempt, no further analyses were conducted.

Impact of Treatment Condition on Change in Outcomes over Time

The outputs of the full individual HLM models can be found in Table 4. Significant treatment effects were observed for three outcomes in the study. Specifically, Veterans in the active SAFER condition were predicted to see sharper declines in suicide ideation than those in I-SPI (B= −0.37; p=.032). To clarify the impact of treatment on trajectory over time, we then calculated the slope-intercept trajectory within each condition. As can be seen in Figure 2A, Veterans in the SAFER condition reported significant declines in severity of suicidal ideation (Figure 2A), but there was minimal change over time in the control condition.

Table 4.

Hierarchical Linear Model Outputs Predicting Change in Target Outcomes

Effect	Veteran Suicide Ideation Severity (0–5)			Veteran Suicide-Related Coping (0–4)			Partner Support of Suicide-Related Coping (0–4)
Predictor	B	SE	p	B	SE	p	B	SE	p
Predicted Baseline Score for Participants in I-SPI	3.71	0.32	<.01	2.86	0.16	<.01	3.43	0.18	<.01
Difference in Initial Score Due to:
Assignment to SAFER	−0.61	0.47	.20	0.14	0.24	.56	−0.12	0.27	.66
Romantic Relationship to Supporting Partner	0.01	0.56	.99	0.16	0.28	.56	−0.02	0.28	.94
History of 1+ Suicide Attempts	0.23	0.48	.63	0.39	0.24	.12	0.23	0.27	.40
Participanťs Age (in years)	0.00	0.02	.90	0.00	0.01	.98	0.00	0.01	.98
Predicted Monthly Change for Participants in I-SPI	0.10	0.12	.43	−0.03	0.02	.18	−0.08	0.04	.07
Difference in Monthly Change Due to:
Assignment to SAFER	−0.37	0.17	.03	0.08	0.04	.03	0.17	0.08	.03
Romantic Relationship to Supporting Partner	0.49	0.21	.02	−0.09	0.04	.04	−0.01	0.07	.93
History of 1+ Suicide Attempts	−0.08	0.17	.64	−0.02	0.04	.61	0.00	0.07	.95
Participanťs Age (in years)	0.01	0.01	.36	0.00	0.00	.14	0.00	0.00	.41
	Veteran’s Thwarted Belongingness (1 – 6)			Veteran’s Perceived Burdensomeness (1–6)			Supporting Partner’s Caregiver Burden (0–4)
	B	SE	p	B	SE	p	B	SE	p
Predicted Score at Baseline for Participants in I-SPI	3.67	0.23	<.01	3.01	0.34	<.01	1.09	0.19	<.01
Difference in Initial Score Due to:
Assignment to SAFER	−0.49	0.34	.16	−0.21	0.49	.66	0.23	0.27	.41
Romantic Relationship to Supporting Partner	0.01	0.40	.97	0.50	0.58	.40	0.74	0.28	.01
History of 1+ Suicide Attempts	−0.19	0.35	.59	0.40	0.50	.43	0.30	0.27	.28
Participanťs Age (in years)	0.00	0.01	.83	−0.02	0.02	.41	0.01	0.01	.60
Predicted Monthly Change for Participants in I-SPI	−0.03	0.04	.42	−0.02	0.05	.64	−0.04	0.03	.16
Difference in Monthly Change Due to:
Assignment to SAFER	0.01	0.07	.87	−0.12	0.08	.16	−0.02	0.05	.69
Romantic Relationship to Supporting Partner	0.09	0.08	.26	0.09	0.09	.32	−0.03	0.04	.55
History of 1+ Suicide Attempts	0.06	0.07	.39	−0.10	0.08	.20	−0.10	0.04	.03
Participanťs Age (in years)	0.00	0.00	.96	0.00	0.00	.57	0.00	0.00	.52

Notes. Each outcome represents a separate model. Romantic relationships and history of suicide attempts are dichotomized (so that 0=Non-Romantic relationships/No suicide attempts). Demographic controls are centered so that intercepts represent a typical participant in I-SPI. Values significant at p<.05 bolded. SAFER=Safe Actions for Families to Encourage Recovery; I-SPI=Individual Safety Planning Intervention

Figure 2.

Predicted slope-intercept trajectories for significant condition by time interactions

Notes. Simple slopes are given in predicted unit change/month. Trajectories calculated after controlling for nature of the caregiving relationship, history of suicide attempt, and participant age. See Table 3 for original HLM models. I-SPI = Individual Safety Planning Intervention; SAFER = Safe Actions for Families to Encourage Recovery * p<.05

This pattern of findings was replicated when examining the impact of treatment participation on coping strategies (Hypothesis 2). An initial significant treatment-by-time interaction emerged when predicting Veterans’ suicide related coping (B=0.08; p=.028; Table 4); however, this was not replicated after accounting for the effect of condition on supporting partner’s support of their coping (B_Dyadic=0.07; p=.151; Supplemental Table 1). A similar treatment by time interaction emerged for partner support of suicide related coping (B=0.17; p=.032; Table 4) that was replicated in dyadic analyses (B_Dyadic=0.16; p=.009; Supplemental Table 1). Examination of simple slopes in the dyadic model found that supporting partners in the I-SPI condition experienced significant reductions in their confidence to support the Veteran after the initial wave (Figure 2B).

Contrary to expectations, self-reported appraisals of relational factors (Hypothesis 3) were not impacted by condition for Veterans Perceived Burdensomeness, Veteran Thwarted Belongingness, or Supporting Partner Caregiver Burden (see bottom half of Table 4).

Impact of Controls

Although not part of our primary hypotheses, we examined significant effects of controls on the outcomes of interest (Table 4). Notably, Veterans who joined the study with a romantic partner, as opposed to with a different support, showed corresponding increases in suicide ideation severity (B=0.49; p=.024) and decreases in suicide-related coping (B=−0.09; p=.037). Romantic partners of Veterans reported correspondingly higher overall levels of caregiver burden at baseline (B=0.74; p=.014) when compared to other types of supporting partners (e.g., parents, siblings, close friends). In contrast, Veteran history of suicide attempt was associated with greater declines in caregiver burden over the study (B=−0.09; p=.029). Taken together, this indicates that supporting partners of Veterans with a history of attempt showed general adaptation to the veteran’s elevated risk while romantic partners of Veterans experienced stable high levels of burden that were then followed by worsening functioning for the Veteran over time.

Discussion

Despite the extensive data supporting the need for family involvement (Chiang, Lu, Lin, Lin, & Sun, 2015; Prabhu, Molinari, Bowers, & Lomax, 2010; Manuel, Crowe, Inder, & Henaghan, 2018) and social support (Wilks et al., 2019) in suicide prevention, there is an alarming gap in the literature on exactly how best to involve families. There are also a lack of guidelines for families to follow on how best to assist their loved one with suicidal symptoms. The SAFER intervention is a brief 4-session treatment involving Veterans, and supporting partners in the development of a joint SPI. This study of the SAFER intervention is the first published pilot RCT of a manualized, dyad-based SPI. Even though one of the steps of an effective SPI (Stanley & Brown, 2012; Stanley, Brown, Karlin, Kemp, & VonBergen, 2008; Stanley et al., 2018) is to identify social supports to share the plan with, the current study is the first to incorporate family in SPI. This is particularly surprising given the widespread use of the SPI among Veterans at risk for suicide.

Additionally, the SAFER intervention facilitated discussions concerning suicide and coping. These topics may be difficult to broach as many suicidal individuals may be reluctant to disclose specific aspects of their suicidal impulses and acts (Ganzini et al., 2013, Calear & Batterham, 2019). While there is a paucity of literature in this area, initial data does exist on the rates, benefits, and challenges of suicide-related disclosure on suicidal individuals (Calear & Batterham, 2019; Choi et al., 2017; Frey, Hans & Cerel, 2016a). However, even less research has been conducted on suicide-related disclosures in military populations. One study by May and colleagues (2019) included suicidal individuals and their significant others in a brief suicide treatment. Among these participants 77% were aware of their partners’ suicidal ideation, but only 23% were aware of their partners suicidal behaviors. Although research around suicide-related disclosure is still evolving, developing strategies to facilitate safe communication around suicide symptoms, as demonstrated in our SAFER intervention, will be a critical component of dyadic and family treatments for suicide.

Our RCT results underscore that a brief suicide safety planning intervention can reduce the severity of Veteran ideation as compared to the control condition. As both our control condition (I-SPI) and SAFER both help the Veteran build a safety plan, family involvement was the primary difference between the conditions. Supporting partners in the control condition reported significant declines in suicide-related coping at each wave. Conversely, SAFER helped supporting partners retain their efficacy to support Veterans across these waves. While the number of subjects was too small to test a mediation effect, future larger studies should address this important question.

Despite our encouraging results, and contrary to our hypotheses, SAFER did not impact any of the interpersonal appraisal variables for Veterans (e.g., burdensomeness, belongingness) or supporting partners (e.g., burden). This pattern of findings is consistent with other skill-based treatments including Dialectical Behavioral Therapy, which in the original studies significantly changed suicidal behavior and symptoms, but not depression, hopelessness, or reasons for living (Linehan, Armstrong, Suarez, Allmon, & Heard, 1991). It is also possible that longer duration treatment in a family or dyadic context is required to positively impact views of the self on others. However, targeting the burdensomeness construct may be most helpful in reducing Veteran suicide risk (Bell et al., 2018, Monteith et al., 2013). This suggests that future refinements to the intervention should further expand on this topic. Nonetheless, our results indicate that changes in suicide risk are possible when supporting partners are equipped with support and tools to help their loved ones and do not require changes in self-appraisal.

This study did not distinguish family constellations and environment, which may be an important variable in thinking about including loved ones in care to reduce suicide risk. For example, the presence of an intimate partner has been associated with reducing suicide risk in a military sample (Skopp, Luxton, Bush, & Sirotin, 2011), but interpersonal conflict can presage increased suicide risk (Bryan, Clemans, Leeson, & Rudd, 2015). The nuances of perceived support and conflict in pre-existing family relationships are likely an important factor, which was not measured here. Clearer delineation of family relations and suicide triggers is warranted, both in terms of screening and in future treatment development for this population.

Clinical Implications

The results of our pilot trial for SAFER underscore the importance of more effectively involving family and loved ones in the treatment of suicidal Veterans. Our findings highlight that changes in Veteran suicide ideation severity are possible by equipping supporting partners with confidence, tools and coping strategies, which is achievable in a brief, 4-session dyadic format.

Limitations

The proposed evaluation of a dyad-based SPI has a few limitations. First, our strategy to target only moderate risk suicidal Veterans seemed prudent in this first trial of a conjoint intervention for suicide risk in Veterans. However, this exclusion criterion limits the applicability of the findings to high-risk Veterans. The SAFER intervention is not intended for high-risk suicidal Veterans nor couples with high levels of conflict. Second, our relatively small sample size of 39 suicidal Veterans and their supporting partners limits generalizability. The small sample also limits our ability to examine treatment mediators and potentially important moderators, such as Veteran suicide status (ideator vs. previous attempter) and gender. Third, conducting the study in a high density, urban setting may result in different findings than if conducted in a rural setting, where, for example, the type and level of stressors and accessibility to services may differ. However, a high density, urban environment, with multiple stressors, may promote increased uptake of our intervention. Fourth, the study arms were not matched for treatment dosage (SAFER: 4 sessions versus I-SPI: single session) and therefore our positive results may have been confounded by the additional clinician contact time in the SAFER arm. Specifically, SAFER participants may have benefitted from engaging in multiple sessions working on one’s safety plan or simply receiving the continued support of a mental health provider. Given this is a pilot RCT, next steps will be to disentangle these variables.

A challenge for this project was the recruitment and attrition of Veterans at risk for suicide and their supporting partners. While over 350 Veterans were approached during recruitment, only 39 Veteran participants were randomized. Reasons for refusal varied but for most Veterans, hesitancy to involve family members was paramount. When probed about the reluctance of participating in a family treatment, Veterans listed four main concerns: 1) Veteran could not identify a supporting partner to participate with them; 2) Veteran did not want to further burden social supports; 3) Veterans had not yet disclosed suicidal symptoms to their supportive partner, highlighting an important obstacle to be addressed in future refinements; 4) Veteran felt their current suicidal symptoms did not warrant participation (e.g., suicide attempt many years ago). As our research team conducts multiple intervention projects and subjects are offered a choice, as a result, many Veterans preferred to participate in a Veteran peer intervention that did not involve supportive partners (Goodman et al., 2020; Goodman et al., 2020; Marin, Sullivan, Spears, & Goodman, 2019). Additionally, family members were reluctant to participate for fears of feeling furthered burdened and adding additional caretaking responsibilities to already busy schedules. To address this, our manual evolved to promote selfcare rituals for supporting partners. Engagement during the treatment phase was excellent.. However, attrition during the follow-up assessments was problematic, particularly for supportive partners, and surprisingly, more pronounced in the SAFER arm for the third assessment. The reasons for this remain unclear, but scheduling conflicts, and other caretaking responsibilities was the most cited explanation for missed assessment sessions perhaps underscoring added burden associated with taking part of the study, however this explanation was not supported by our outcome data. Future refinements of this approach may consider telehealth delivery to facilitate compliance and attendance.

This first randomized control pilot trial for SAFER, a manualized, 4-session, novel dyad intervention, combining SPI and skills, targeting at-risk suicidal Veterans, demonstrated promising preliminary outcome results and underscores the need to optimize family involvement in the management and care of suicidal Veterans. Future development of SAFER and additional family approaches will need to better delineate when families and loved ones are triggers for the Veteran’s suicide experience versus sources of support and foster additional coping and self-care for families.

Table 2:

SAFER Intervention Sessions

Joining Sessions	- Held separately with the Veteran and supporting partner - Serves to gather background information, clarify treatment goals, assess motivation and identify any individual or couple issues that may impact the treatment.
Session #1	-Focuses on an educational overview on Veteran suicide and SPI - Didactic material is provided on Veteran suicide, risk assessment and importance of managing suicidal urges - Participants guided to monitor the Veteran’s mood and other identified risk factors unique to each Veteran, which is reviewed at the beginning of all subsequent sessions.
Session #2	-Focuses on SPI from the Veteran’s perspective - Explores potential obstacles to engaging the supporting partner in SPI. - Veteran describes each step of their SPI to their supporting partner. - Homework is assigned which entails Veteran compiling a “reasons for living” list and supporting partner compiles a list of “reasons for living” for the Veteran and a self-care list for themselves.
Session #3	-Focuses on maximizing social support for the Veteran - Creation of a parallel supporting partner’s SPI constructed collaboratively with the Veteran - Troubleshooting of potential problems in implementation are identified and addressed. - Discussion of how and when to practice using the Safety Planning Intervention - Session ends with members of the dyad sharing their homework - Homework is assigned for session 4 is to practice using communication skills to facilitate use of the SPI and to record the results
Session #4	-Serves as a booster to allow more time for practice of the SPI and skills - Focuses on review of communication skills both in general and particularly in the context of Veteran’s and supporting partner’s SPI use - Identifying and problem-solving obstacles to usage of SPI by either member of the dyad - Communication skills (e.g., validation and active listening) are reviewed and then role plays to practice skills - Emphasizes the value of a weekly, scheduled SPI check-in meetings