Suicide is the 10th leading cause of death in the United States, and veterans are twice as likely as nonveterans in the general population to die by suicide (
1,
2). Accordingly, suicide prevention is the top clinical priority of the U.S. Department of Veterans Affairs (VA), underscoring the need to identify factors that may increase veterans’ susceptibility to suicide morbidity and mortality (
1). Factors that have been linked to increased suicide rates in the general population that may be particularly relevant among veterans include male sex, younger and older ages, functional limitations, and availability of and experience with firearms (
3–
5). Psychiatric conditions also serve as significant risk factors for suicidal ideation and suicide attempts among veterans (
6). For example, veterans meeting criteria for mental health or substance use conditions such as depression, psychosis, posttraumatic stress disorder (PTSD), and substance use disorders are particularly likely to experience suicidal ideation and mortality, with psychiatric comorbidity placing veterans at an even greater risk (
7–
11).
The significant associations among mental health and substance use conditions and suicidal ideation and mortality have prompted a variety of suicide prevention efforts within the Veterans Health Administration (VHA) (
1,
12,
13). Recognizing the importance of early screening and risk assessment (
14), the VHA Office of Mental Health and Suicide Prevention, for instance, has recently announced plans to implement a standardized process of screening for suicide risk, which includes adding well-validated suicide ideation screeners (including item 9 from the Patient Health Questionnaire–9 [PHQ-9]) to existing clinical reminders to screen for depression and PTSD (
15–
17). Key efforts also have been implemented specifically within primary care, because primary care serves as a particularly high-leverage setting for suicide prevention. For example, research has shown that approximately three-quarters of individuals who die by suicide have had a primary care appointment in the year prior to their death, and 45% have seen a primary care provider within 1 month of suicide (
18). Moreover, primary care providers are often tasked with helping patients manage many of the conditions and factors that are associated with elevated suicide risk, including mental health and substance use conditions (
19,
20). Accordingly, under the Primary Care–Mental Health Integration (PCMHI) initiative, the VHA requires all medical centers to adopt a PCMHI collaborative care management program (
21). Central to PCMHI programs is ensuring timely, routine screening for mental health and substance use conditions (e.g., depression, PTSD, and substance use disorders) and suicidal ideation and providing algorithm-driven, measurement-based decision support for the management of these conditions in primary care (
22).
Although PCMHI programs screen for and target multiple factors that heighten the risk of suicide-related outcomes among veterans with mental health or substance use issues, these factors tend to focus on biological and psychiatric symptoms and conditions (
6). There are, however, additional factors that are not traditionally assessed in primary care and other clinical settings that may be associated with suicide rates in this vulnerable group. For example, the nature and quality of an individual’s connections with social network members represent factors that are associated not only with general medical and psychiatric health issues but also with suicide-related outcomes among primary care patients experiencing symptoms of mental health or substance use conditions (
23,
24). Positive social relationships and exchanges (e.g., interactions that are perceived as emotionally rewarding and result in a feeling of being supported, valued, and cared for) are essential in promoting physical and mental health (
25). Moreover, although negative exchanges (e.g., interactions that entail criticism and conflict or fail to offer support in times of need) occur less frequently than positive exchanges, they appear to have a particularly negative impact on mental health when they do occur (
24,
26), highlighting the value of examining both positive and negative aspects of social ties in clinical settings.
The association between social network factors and suicide outcomes, specifically, is also well documented, with literature on the impact of social connectedness on suicidal behaviors dating back to the 1960s (
27). For example, social support and greater social integration (i.e., the degree to which a person engages in various types of social relationships and interactions at both the individual and community level) are associated with lower suicidal ideation and completion rates in both veteran and nonveteran samples (
3,
10,
28–
33). Prior studies have typically relied on a single indicator to capture the nature of a person’s social ties (e.g., social integration or social support); however, to have the greatest clinical utility and more fully inform prevention efforts, it is important to tease apart which indicators have the greatest impact on suicide-related outcomes (e.g., suicidal ideation and attempts).
To that end, several studies have examined the association between multiple dimensions of the nature and quality of a person’s social ties and suicidal ideation and behavior. For example, in an examination of patients recruited from residential substance use treatment programs, You and colleagues (
34) found that interpersonal conflict and lack of belongingness (i.e., having the innate need to belong in meaningful relationships met) were significantly associated with suicidal ideation and that a lack of perceived social support and belongingness and living alone were related to having made a suicide attempt. Among veterans with probable major depression who were receiving primary care at a VA hospital, loneliness (i.e., a state arising from the lack of perceived social connectedness) remained the sole predictor of suicidal ideation after the analysis controlled for other indices of social ties (e.g., social support, number of confidants, and interpersonal conflict) (
35). Finally, in one of the first examinations of the relative impact of positive and negative social exchanges on suicidal ideation among VA primary care patients referred for a mental health and substance use evaluation, our team found that after adjustment for sociodemographic factors, physical functioning, and comorbid mental health and substance use conditions, negative but not positive exchanges significantly predicted suicidal ideation (
23). Of note, subsequent analyses indicated that one category of negative social exchanges, social interactions that elicited feelings of rejection or neglect, were the most related to suicidal ideation. These results further support the notion that “thwarted belongingness,” or having unmet needs for meaningful and supportive social ties, may be particularly likely to elicit suicidal thoughts and behaviors (
34,
36,
37).
Despite the link between social network factors and suicidal ideation, few PCMHI programs routinely assess multiple indices of social ties in a standardized fashion. Information regarding veterans’ perceived social support and connectedness represents an important, unaddressed target that can be used to help guide more tailored, patient-centered, and measurement-based care. Veterans engaged in primary care who screen positive for a mental health or substance use condition are at high risk of suicidal ideation not only because of their physical or psychological health but also because they are likely to have fewer positive social interactions, resources, and supports (
38–
41).
Guided by empirical findings from prior studies highlighting the clinical value of evaluating multiple indices of social ties (
34) and by our work with veterans enrolled in a PCMHI program (
23), in 2010 the Behavioral Health Laboratory (BHL) (the PCMHI program at the Corporal Michael J. Crescenz Department of Veterans Affairs (VA) Medical Center) incorporated three items into its routine baseline clinical assessment that capture veterans’ perceived social support, the frequency of negative exchanges with social network members, and the frequency of contact with friends and relatives (a proxy for social integration). These three dimensions of social ties were chosen to reflect domains that have been found to have independent, unique associations with suicidal ideation (
23,
34) and with psychological health more broadly (
24,
25,
42). The study reported here sought to explore the relative association between each of these indices and suicidal ideation among veterans referred to the BHL by their primary care providers for a mental health and substance use assessment. To examine the added clinical value of including these items, our multivariate models adjusted for sociodemographic characteristics, perceived general medical health, and symptoms of mental health and substance use conditions. We hypothesized that all three items capturing social ties would be independently associated with increased odds of suicidal ideation, after the analysis adjusted for other known correlates of suicide outcomes (
3–
6).
Methods
Study Sample and Procedures
All study procedures were approved by the Corporal Michael J. Crescenz VA Medical Center Institutional Review Board. The study employed a retrospective analysis of clinical patient records. Data for 15,277 veterans who completed a BHL baseline assessment between September 2010 and March 2016 were extracted from the medical record. The BHL is an evidence-based, PCMHI program that focuses on the identification, assessment, triage, and clinical care management of primary care patients who may need care for mental health or substance use issues, such as depression, anxiety, substance use disorder, and PTSD. The veterans included in the analysis had all received primary care at the Corporal Michael J. Crescenz VA Medical Center or affiliated community-based outpatient clinics (
43,
44). Veterans were referred to or identified by the BHL through three mechanisms: referral by a primary care provider after a positive screen for PTSD, alcohol misuse, or depression; direct referral by a primary care provider on the basis of clinical judgment, independent of screening; or pharmacy-based casefinding of patients newly prescribed an antidepressant by their primary care provider.
On referral, patients were contacted to complete an initial, 20- to 30-minute mental health and substance use assessment. The assessments were conducted by BHL health technicians, who are extensively trained interviewers with a bachelor’s degree or higher. The clinical interview, conducted either over the phone or in person, included questions regarding sociodemographic characteristics, general health and mental health functioning, substance use, and assessments of axis I disorders (e.g., depression, PTSD, and psychosis).
Assessments and Measures
Patient characteristics.
The following sociodemographic variables were exported for analysis: age, sex (0, female; 1, male), race-ethnicity (0, other; 1, white), marital status (0, not married or partnered; 1, married or partnered), and inadequate financial status (0, have at least enough to get by; 1, can’t make ends meet). A single item (“In general, would you say your health is excellent, very good, good, fair, or poor?”) from the Veterans RAND 12-Item Health Survey assessed the patient’s perceived general health (
45).
Clinical BHL mental health and substance use assessments.
The PHQ-9 was used to assess severity of depressive symptoms (
15). The 7-day Timeline Follow-Back Interview and a modified version of the Alcohol, Smoking and Substance Involvement Screening Test were used to assess whether patients met criteria for (coded as 1) or did not meet criteria for (coded as 0) at-risk drinking and illicit drug use, respectively (
46,
47). Given changes in the BHL assessment battery over the course of the data extraction time window, whether patients met criteria for (coded as 1) or did not meet criteria for (coded as 0) generalized anxiety disorder was determined by scores on either the Mini-International Neuropsychiatric Interview (
48) or the Generalized Anxiety Disorder–7 scale (GAD-7) (
49). The Patient Checklist–Civilian (PCL-C) for PTSD or the PCL-5 (assessments were modified during the data extraction time frame because of changes in the criteria set forth in the
DSM) assessed whether patients met criteria for (coded as 1) or did not meet criteria for (coded as 0) PTSD (
50).
DSM criteria were also used to assess whether patients met criteria for (coded as 1) or did not meet criteria for (coded as 0) mania and psychosis (
48). Diagnostic cutoffs for the GAD-7, PCL and PCL-5, mania, and psychosis were based on
DSM-IV or
DSM-5 criteria and symptom clusters, depending on the version and timing of the assessment administered.
Suicidal ideation.
Suicidal ideation was assessed with two separate measures. First, the five-item Paykel Suicide Scale was used to assess for the severity of suicidal ideation in the past year (
51,
52). Patients were asked to respond (yes or no) to a series of questions regarding whether they had felt as if life was not worth living (coded as 1); had wished they were dead (coded as 2); or had thought about taking (coded as 3), seriously considered taking (coded as 4), or made an attempt to take (coded as 5) their own life over the past year. Patients’ scores corresponded to the greatest severity of suicidal ideation or behavior endorsed (
53). On BHL assessment, patients who respond yes to items 3, 4, or 5 are contacted by a clinician for follow-up. Accordingly, we used Paykel scores to categorize patients in one of three groups reflecting the severity of suicidal ideation: no suicidal ideation (Paykel score of 0), low-severity suicidal ideation (response of yes to either item 1 or 2), or high-severity suicidal ideation (response of yes to item 3, 4, or 5).
Second, item 9 from the PHQ-9—“Over the past two weeks, how often have you been bothered by thoughts that you would be better off dead or of hurting yourself in some way?”—was dichotomized and treated as the outcome in separate analyses. Veterans who reported “several days,” “more than half the days,” or “nearly every day” in response to the item were coded as having a positive suicide ideation screen.
Social exchanges, support, and integration.
Three indices of social ties were assessed with individual items from validated measures. One item (“In the past month, how often did people you know act inconsiderate to you or unsympathetic about your personal concerns?”), from the Positive and Negative Social Exchanges Scale, captured the frequency of negative social exchanges (0, never, to 4, very often) (
24). A second item (“On the whole, how much do your friends and relatives make you feel loved and cared for?”), from the Dyadic Adjustment Scale, captured perceived social support (0, not at all, to 4, a great deal) (
54). The final item (“How often are you in contact with any of your friends or relatives that do not live with you (including visits, phone calls, letters, or electronic mail messages)?”) was modified from the Midlife in the United States study and dichotomized to capture the degree of social integration (0, once a month or less; 1, about once a week or more) (
55). Bivariate analyses of the three social tie indices yielded the following correlations: social support and negative interactions (r=–.33, p<.001), social support and social integration (r=.29, p<.001), and negative interactions and social integration (r=–.17, p<.001).
Analytic Strategy
To compare sample characteristics across veterans who did and did not endorse suicidal ideation, measured by both the Paykel Suicide Scale and PHQ-9 item 9, we ran a series of bivariate tests of significance, including t tests for equality of means and chi-square tests for continuous and dichotomous outcomes. We also ran a series of unadjusted binomial (with PHQ-9 item 9 as the outcome) and multinomial (with suicidal ideation severity from the Paykel Suicide Scale as the outcome) logistic regression models for each of the patient, clinical, and social tie variables summarized above.
To test our main hypotheses regarding the associations among the items capturing social ties and suicidal ideation, we ran two adjusted multivariate logistic regression models. In the first model, suicidal ideation (none, low severity, and high severity) over the past year, as measured by the Paykel Suicide Scale, was specified as a categorical outcome. Patients reporting no suicidal ideation served as the reference group. In the second model, item 9 from the PHQ-9 was specified as a dichotomous outcome (0 [reference group], negative suicide ideation screen; 1, positive suicide ideation screen). Both adjusted models controlled for age; sex; perceived general health; race-ethnicity; marital status; financial status; depressive symptom severity; illicit drug use; and meeting criteria for PTSD, mania, psychosis, anxiety, or at-risk drinking. The selected covariates not only have been shown in prior work to covary with suicidal ideation but also were significantly associated with suicidal ideation in our initial, bivariate analyses. Analyses were conducted with IBM SPSS Statistics 20.
Results
Table 1 summarizes sociodemographic and clinical characteristics of the full sample and the same characteristics stratified by suicidal ideation severity as measured by the Paykel Suicide Scale. The mean age of the patients was 51.3. Most were men, and most met criteria for at least one mental health or substance use condition on being assessed by the BHL. Roughly half were white, and 44% were married or partnered. Thirty-nine percent of the sample reported either low- or high-severity suicidal ideation, as measured by the Paykel Suicide Scale. Results from the bivariate analyses showed that across both measures used to assess for suicidal ideation, those who reported suicidal ideation differed from those who did not on all the variables examined, with one exception; sex was unrelated to suicidal ideation severity, as measured by the Paykel Stress Scale.
Results from the adjusted, multinomial logistic regression model suggested that when the analysis controlled for age, sex, race-ethnicity, marital and financial status, perceived overall health, depressive symptoms, and mental health or substance use comorbidity, all three social tie indices were independently associated with severity of suicidal ideation over the previous year (
Table 2). Specifically, patients reporting more social support or more frequent social contact with friends or relatives were significantly more likely to report no suicidal ideation as opposed to low- or high-severity suicidal ideation on the Paykel Suicide Scale. When the analysis adjusted for covariates, a one-unit increase in social support was associated with a reduction in the odds of reporting low-severity and high-severity suicidal ideation (adjusted odds ratio [AOR]=0.86 and 0.82, respectively), compared with no ideation. Similarly, for those who had a high degree of contact with friends or relatives, the odds of reporting suicidal ideation were lower than for those with less contact (low-severity ideation, AOR=0.88; high-severity ideation, AOR=0.83).
On the other hand, patients who reported a greater frequency of negative exchanges with network members were significantly more likely to report low- or high-severity suicidal ideation in the past year than to report no suicidal ideation. For every one-unit increase in negative exchanges, the odds of reporting low- or high-severity suicidal ideation, compared with no ideation, increased (low-severity ideation, AOR=1.08; high-severity ideation, AOR=1.07). These effects are comparable in size to those found in other multivariable analyses of the association between sociodemographic and psychosocial variables and suicidal ideation (
34,
56–
58). Results from analyses of the primary suicidal ideation screener from the PHQ-9 paralleled these findings; when the analysis adjusted for covariates, greater social support and more frequent social contact were each independently associated with lower odds of a positive suicidal ideation screen (
Table 3). Greater frequency of negative exchanges, however, was related to higher odds of suicidal ideation in the past 2 weeks.
Discussion
The aim of this study was to examine the value of evaluating multiple indices of social ties when assessing suicidal ideation among veterans referred by primary care to a PCMHI program for a mental health and substance use assessment. The results support our hypothesis; even after the analysis adjusted for other known, robust correlates of suicide outcomes in primary care and psychiatric patient samples (
3–
6), all three social tie items were independently associated with suicidal ideation. Of note, the association between the social tie items and suicidal ideation was significant both in terms of suicidal ideation severity over the past year (as assessed by the Paykel Suicide Scale) and in terms of the more proximal measure of suicidal ideation in the past 2 weeks (as assessed by PHQ-9 item 9). The convergence of findings across the two scales highlights the clinical utility of addressing the quality and frequency of social ties at multiple points over the course of assessment and treatment, particularly given the significant positive association between PHQ-9 item 9 and increased odds of suicide mortality in both VA and community samples (
16,
17).
The finding that greater social support and social integration and less frequent negative exchanges were significantly associated with reduced odds of reporting suicidal ideation is consistent with prior work (
23,
34,
35). The fact that each of the three social tie indices was independently associated with suicidal ideation, even when the analysis controlled for the other two indices and marital status, lends further support to the value of examining multiple domains of social ties when assessing suicide-related outcomes among individuals experiencing psychiatric symptoms. The benefits of social support are well documented; however, even supportive, close relationship partners can be a source of conflict and disappointments that are particularly detrimental to psychological well-being (
59). Indeed, positive and negative interactions with network members can best be characterized as two unique dimensions of interpersonal experience, rather than as two ends of a single continuum (
25), further underscoring the importance of evaluating both domains.
Similarly, our findings support the notion that frequency of contact with social network members, a marker for social integration or connectedness, also may confer a unique benefit for psychological well-being. Prior work has shown that social and community integration (e.g., frequency of social contact, availability of non–family members within driving distance, social network size, and marital status) and religious involvement are associated with suicide mortality even after adjustment for well-established covariates such as age; comorbid health conditions, including mental health conditions; and employment status (
31,
60). Moreover, other correlates of low social connectedness or integration, including feelings of rejection or neglect and loneliness, have all been shown to predict suicidal ideation among veterans with behavioral health conditions (
23,
34,
35). Our finding that frequency of contact with friends and relatives was uniquely associated with suicidal ideation is also consistent with interpersonal theories of suicide (
36,
37). It has been suggested, for example, that feeling socially excluded from others may deny individuals satisfaction of their essential human need for belonging (i.e., thwarted belongingness). Feeling as if one does not belong can further lead to a loss of self-worth, sense of meaning in life, and self-regulatory abilities, leading to an increase in self-defeating behavior and suicide risk (
61,
62).
Several caveats should be considered when interpreting the results. First, this study specifically focused on the clinical value of assessing various indices of social ties when examining suicidal ideation risk among veterans referred to a PCMHI program. Thus, although our sample afforded us the opportunity to examine a group at high risk of suicide morbidity and mortality, the focus on a primary care sample of veterans may limit the generalizability of our findings in terms of other high-risk or nonveteran populations. Generalizability of our results also may be limited because of the disproportionately high number of men in the sample. Second, the clinical data evaluated in this study were cross-sectional, thus precluding inferences about the directionality of the association between the social relationship indices and suicidal ideation. The association between interpersonal functioning and suicidal ideation is most likely reciprocal in nature, because acute and chronic symptoms of mental illness or substance use disorders and suicidal thoughts can also affect the quality and structure of one’s relationships with others (
40,
63). Finally, our analyses relied primarily on single items to capture the three indices of social ties. The inclusion of valid, reliable multi-item scales in future work is warranted. In a similar vein, future work would benefit from inclusion of measures that take into account other dimensions of social relationships (e.g., perceived burdensomeness to network members, perceived loneliness, instrumental and tangible support, and criticism and rejection by others) and the relationship type (e.g., spouse, child, friend, or peer), because these factors may be especially important when evaluating suicidal ideation among individuals experiencing symptoms of mental health or substance use conditions (
36,
38,
64).
Conclusions
Notwithstanding these limitations, the results from this study provide preliminary support for the clinical value of evaluating the nature and quality of social ties among veterans who screen positive for behavioral health issues in primary care and who are referred to a PCMHI program. Because this patient population is at high risk of suicidal thoughts, behavior, and mortality, the finding that each of the three indices of the veterans’ social ties was uniquely associated with suicidal ideation, even after control for known clinical and sociodemographic correlates, has multiple implications for clinical research and practice. Future investigations that use more comprehensive, multidimensional measures of social ties and examine the longitudinal mechanisms by which social ties, symptoms of mental health and substance use conditions, and suicidal ideation are linked would help inform the timing and content of suicide prevention efforts. Systematically including scales or items that capture multiple dimensions of veterans’ social ties in routine intake and follow-up PCMHI assessments may help identify patients at high risk of developing suicidal thoughts and inform care decisions. Moreover, care management or brief interventions that aim to improve the quality of preexisting social relationships, encourage engagement in social and community activities, and aid in the development of skills to reduce negative and maximize positive exchanges with network members may have a significant impact on reducing suicidal ideation among those at high risk.