Stressful Life Events and Risk of Homelessness After Active Duty: An Assessment of Risk and Resilience Among Servicemembers

Abstract

Objectives:

The transition from military to civilian life may present increased exposure to various stressful life events (SLEs) that can increase the risk of homelessness (eg, loss of employment, dissolution of romantic relationships). We assessed the extent to which exposure to SLEs occurring proximal to US Army soldier transitions out of active duty was associated with risk of homelessness.

Methods:

A total of 16 589 respondents who were no longer on active duty but participated while on active duty during 2011-2014 baseline surveys completed follow-up surveys during 2016-2018 and 2018-2019. The follow-up surveys assessed SLEs and homelessness occurring in the past 12 months. We used modified Poisson regression models to evaluate how much differential SLE exposure and effects explained the aggregate association of a risk index with homelessness among a sample of 6837 respondents, weighted to represent the full sample.

Results:

More than half (n = 3510, 52.8%) of respondents reported experiencing any SLEs in the past 12 months. Most (60.5%) of the difference in prevalence of homelessness among respondents defined as being at high risk of homelessness (vs lower risk) was explained by differential exposure to, and/or effects of, these SLEs. Personal betrayal by a loved one and economic problems played the largest roles in adjusted risk differences (0.045 and 0.074, respectively).

Conclusions:

Homelessness might be reduced by gearing interventions toward soldiers at high risk of homelessness who are transitioning out of active duty to reduce exposure to and effects of modifiable SLEs on experiencing homelessness.

Keywords

stressful life events homelessness servicemembers

Ending homelessness among military veterans is a top policy priority of the US government, backed by billions of dollars.¹ From 2009 to 2021, homelessness among veterans declined by approximately 50%.² However, veterans remain overrepresented in the homeless population: approximately 40 000 veterans experienced homelessness on any given night in 2020.³ One way to address this problem is to identify servicemembers at risk of homelessness as they transition to veteran status and develop preventive interventions.

Research has identified risk factors for veteran homelessness occurring prior to (eg, adverse childhood experiences), during (eg, military sexual trauma), and after leaving (eg, financial problems) military service.⁴ A recent report based on longitudinal surveys carried out in the Study to Assess Risk and Resilience in Servicemembers (STARRS-LS) showed that a risk model—developed while soldiers were still on active duty—had good accuracy in predicting self-reported homelessness after soldiers left active duty (area under the receiver operating characteristic curve = 0.78)⁵: the 20% of recently separated (ie, no longer affiliated with the US Army) or deactivated (ie, no longer on active duty but still in the Army National Guard or US Army Reserve) soldiers with the highest risk based on that model accounted for 61% of all homelessness in the STARRS-LS sample.

However, the STARRS-LS model did not include information about experiences that occurred after leaving service that may predict homelessness. It is important to know how these experiences relate to the preseparation/deactivation predictors in the model to guide the focus of preventive interventions. The transition to civilian life is a time of increased risk of exposure to a variety of stressful life events (SLEs) that may increase the risk of homelessness (eg, loss of employment, dissolution of romantic relationships).^6-9 The present study builds on the earlier STARRS-LS report by investigating the extent to which differential exposure to and effects of these SLEs account for the association between the high-risk indicator previously developed and subsequent homelessness.

Methods

Sample and Procedures

Baseline surveys

STARRS-LS is an epidemiological–neurobiological study designed to evaluate the risk and protective factors for suicidal behaviors among US Army soldiers.¹⁰ Field procedures are detailed elsewhere.^11-13 There were 3 baseline STARRS surveys: (1) the New Soldier Study (2011-2012; N = 38 733 soldiers), (2) the All Army Study (2011-2013; N = 25 088 soldiers), and (3) the Pre–Post Deployment Study (2012-2014; N = 8566 soldiers). All participants provided written informed consent. The human subjects committees of the University of Michigan and the Uniformed Services University of the Health Sciences approved recruitment, informed consent, and data collection protocols. In addition, we obtained approval from the Army Medical Research and Materiel Command for the component of the All Army Study survey that was carried out among soldiers who were deployed to Afghanistan and surveyed in Kuwait. As detailed elsewhere,¹³ calibration weights adjusted the baseline samples for discrepancies with administrative variables available for all soldiers.

Longitudinal surveys

Two subsequent STARRS Longitudinal Surveys (LSs) were conducted during September 2016–April 2018 (LS1) and April 2018–July 2019 (LS2) using a probability subsample of respondents from the baseline surveys who agreed to have their survey data linked to their US Army administrative data. We oversampled baseline respondents with a history of suicidality or mental disorder, women, members of the US Army Reserve or National Guard, and Special Operations soldiers. A total of 14 508 soldiers completed LS1, resulting in a weighted response rate of 35.6%. LS2 was subsequently completed by 12 156 of the LS1 respondents. We focus here on the subset of LS respondents who were either separated or deactivated at the time of their LS survey. We excluded LS2 respondents who reported being homeless in the 12 months before LS1 to avoid double counting any single respondent in the pooled LS1/LS2 analysis. The total sample included 16 589 observations, composed of 8797 respondents from LS1 and 7792 respondents from LS2.

We selected a 70% randomly selected training sample from the full sample to build the prediction model described in the introduction of this article. Given the rarity of homelessness, we used a case-control subsampling scheme in that training sample, with all 310 respondents in the training sample who reported homelessness at any time in the 12 months before the LS survey compared with a probability sample of 1550 other respondents in the training sample (ie, 5 times the number who experienced homelessness). We gave the cases a weight of 5 to create equal weighted numbers of cases and controls to train the prediction model. We then used the remaining 30% of LS respondents in the holdout sample (133 who reported experiencing homelessness and 4844 who did not report experiencing homelessness) to test the model. We included all 6837 of these respondents in the current analysis with appropriate weighting to represent the full set of 16 589 LS1 and LS2 observations. Additional information on recruitment and sampling is available in supplemental materials (eFigure 1 and eFigure 2 in Supplementary Material).

Measures

Homelessness risk indicator

We developed the composite dichotomous homelessness risk indicator, used as the independent variable, by using information from STARSS baseline surveys, US Army/US Department of Defense administrative data, and small-area geospatial variables about the respondent’s county of residence after separation/deactivation.⁵ In this analysis, we defined the high-risk group as the 20% of respondents with the highest composite predicted risk of homelessness and the remaining 80% of the sample as the lower-risk group. The high-risk group accounted for 61% of all homelessness in the 12 months before the LS survey in the holdout sample.

SLEs

Respondents in each LS survey answered questions about exposure to 13 SLEs in the prior 12 months: serious illness/injury, economic events (job loss, major financial crisis), victimizations (burglary, armed robbery, physical or sexual assault), legal events (trouble with police, trouble with the law), interpersonal events that directly affected the respondent (separation/divorce/other breakup, personal betrayal by a loved one), and SLEs that occurred within one’s social network (death, serious illness/injury, life crisis of a loved one).

Homelessness

The LS surveys included a modified question from the Veterans Health Administration’s Homelessness Screening Clinical Reminder^14,15 on how much time during the past 12 months (or since most recently leaving or being released from active duty, if less than 12 months ago) respondents were “living in stable housing that you own, rent, or stayed in as part of a household.” If the response was anything other than “all of the time,” a follow-up question asked how many months in the past 12 months (or since most recently leaving or being released from active duty) respondents were homeless. A response of ≥1 month was coded as homeless in the past 12 months.

Analysis

We conducted the analysis during August–December 2021 and used conventional demographic rate standardization¹⁶ to evaluate the role of SLEs in accounting for the association between the high-risk indicator and subsequent homelessness (ie, a separate prediction model estimated the joint associations of a series of intervening variables [SLEs] in predicting an outcome [homelessness] within subsamples defined by a dichotomous primary predictor variable [homelessness risk index]). The observed prevalence of homelessness among high-risk respondents (P_h) can then be defined using the following equation:

P_{h} = p_{0 h} + V_{1 h} E_{1 h} + V_{2 h} E_{2 h} + \dots + V_{ih} E_{ih}

(1)

where V_ih is the difference in the probability of homelessness between respondents with versus without SLE_i, referred to as the vulnerability coefficient in the rate standardization literature, E_ih is the prevalence of SLE exposure among high-risk respondents, and p_0h is the predicted prevalence of homelessness under the model among respondents exposed to none of the SLEs. A similar expression can be made for the predictors of becoming homeless among lower-risk respondents (P_o), that is,

P_{o} = p_{0 o} + V_{1 o} E_{1 o} + V_{2 o} E_{2 o} + \dots + V_{io} E_{io}

(2)

The importance of differences in exposure and vulnerability to the SLEs in accounting for the observed difference in prevalence of homelessness between high-risk and lower-risk respondents (ie, P_h – P_o) can then be seen by subtracting the elements in equation 2 from those in equation 1 and rearranging to obtain

\begin{array}{l} P_{h} - P_{o} = {S_{i = 1}}^{n} (E_{ih} - E_{io}) V_{io} + {S_{i = 1}}^{n} (V_{ih} - V_{io}) \\ E_{io} + {S_{i = 1}}^{n} (V_{ih} - V_{io}) (E_{ih} - E_{io}) + (p_{oh -} p_{0 o}) \end{array}

(3)

summed over the n SLEs under consideration.

The first expression in equation 3, the differential exposure component, describes the extent to which the observed higher prevalence of homelessness among high-risk respondents could be accounted for by their higher exposure to SLEs even if vulnerability was the same as among lower-risk respondents. The second expression, the differential vulnerability component, describes the extent to which the observed higher prevalence of homelessness among high-risk respondents could be accounted for by the stronger association between SLEs and homelessness among high-risk respondents than among lower-risk respondents. The third expression, the interaction component, describes the extent to which the observed higher prevalence of homelessness could be accounted for by the joint occurrence of higher exposure and higher vulnerability to SLEs among high-risk respondents. The final residual component describes the elevation in prevalence of homelessness among high-risk respondents compared with lower-risk respondents in the absence of exposure to any of the SLEs.

Analysis began by examining the differential SLE exposure between high-risk respondents and lower-risk respondents. We then estimated adjusted risk differences (ARDs) in equations 1 and 2 from modified Poisson regression equations to define the vulnerability coefficients using the postestimation margins command in STATA MP version 17.0 (StataCorp), a postregression command that estimates the mean predicted probabilities within subsamples in Poisson models and then compares these differences across subsamples while adjusting for complex survey design effects.¹⁷ We computed the 95% CIs of ARD estimates using the Taylor series linearization method.¹⁸

We started with univariable models and then estimated multivariable models to evaluate the possibility of multicollinearity among the SLEs. We collapsed subsequent multivariate models across SLEs within conceptual domains to deal with some SLEs being very highly correlated with each other or being very uncommon. We also evaluated the possibility that joint associations of multiple SLEs with the probability of becoming homeless were nonadditive. We then used equation 3 to decompose the mediating and moderating effects of SLEs. We estimated the 95% CIs of the component proportions aggregated across all SLEs using the jackknife repeated replications simulation method.¹⁹

Results

SLE Prevalence

The prevalence (95% CI) of any SLE in the 12 months before one of their LS surveys was 52.8% (49.4%-53.1%) in the full sample (Table 1). Fewer than half this number experienced exactly 1 SLE (23.7%; 95% CI, 22.8%-26.2%), with the remaining divided relatively equally between those who experienced 2 SLEs (13.2%; 95% CI, 12.1%-14.5%) and ≥3 SLEs (13.5%; 95% CI, 12.2%-14.8%). The most prevalent individual SLEs were death of a loved one (19.5%; 95% CI, 18.0%-21.0%), major financial crisis (17.6%; 95% CI, 16.1%-19.1%), and personal betrayal by a loved one (14.0%; 95% CI, 12.6%-15.3%). The relative risk (RR) of SLE exposure among high-risk versus lower-risk respondents was 1.6 (95% CI, 1.4-1.8) for 1 SLE, 2.3 (95% CI, 1.9-2.8) for 2 SLEs, and 3.0 (95% CI, 2.6-3.5) for ≥3 SLEs. High-risk respondents were significantly more likely than lower-risk respondents to be exposed to each of the individual SLEs and groups considered, with RRs ranging from 1.6 (95% CI, 1.4-1.8) for any social network event to 7.4 (95% CI, 4.2-13.0) for trouble with the law.

Table 1.

Prevalence of stressful life events (SLEs) and relative risks (RRs) of homelessness based on SLEs that occurred after leaving active Army status among US Army respondents in STARRS surveys, United States, 2016-2019^a

Stressful life event	No.	Postseparation/deactivation SLE prevalence, % (95% CI)	High risk of homelessness^b predicting postseparation deactivation SLE exposure, RR (95% CI)^c
Physical health
Serious illness/injury	436	6.4 (5.5-7.3)	2.4 (1.8-3.3)
Interpersonal
Separation/divorce/other breakup	976	13.0 (11.8-14.2)	2.1 (1.8-2.6)
Personal betrayal by a loved one	1029	14.0 (12.6-15.3)	3.0 (2.5-3.6)
Economic
Job loss	882	11.2 (10.0-12.4)	2.5 (2.1-3.1)
Major financial crisis	1269	17.6 (16.1-19.1)	2.4 (2.0-2.8)
Any economic event^d	1719	23.4 (21.7-25.1)	2.2 (1.9-2.5)
Victimization
Burglary^e	281	3.8 (3.0-4.6)	2.5 (1.7-3.8)
Armed robbery^f	44	0.6 (0.3-1.0)	4.2 (1.5-11.4)
Physical or sexual assault	137	1.8 (1.3-2.3)	2.6 (1.5-4.5)
Any victimization^d	414	5.6 (4.7-6.5)	2.5 (1.8-3.4)
Legal
Trouble with police^g	195	2.2 (1.7-2.8)	3.0 (1.9-4.8)
Trouble with the law^h	137	1.4 (1.0-1.7)	7.4 (4.2-13.0)
Any legal event^d	286	3.4 (2.6-4.1)	4.0 (2.8-5.9)
Social network
Death of a loved one	1316	19.5 (18.0-21.0)	1.8 (1.5-2.1)
Serious illness/injury of a loved one	763	11.9 (10.8-13.1)	1.7 (1.4-2.1)
Life crisis of loved one	719	10.3 (9.3-11.2)	1.8 (1.4-2.3)
Any social network event^d	1683	29.2 (27.6-30.8)	1.6 (1.4-1.8)
Total
Any SLEⁱ	3510	51.3 (49.4-53.1)	1.6 (1.5-1.7)
1 SLE	1593	24.7 (22.8-26.2)	1.6 (1.4-1.8)
2 SLEs	902	13.3 (12.1-14.5)	2.3 (1.9-2.8)
≥3 SLEs	1015	13.5 (12.2-14.8)	3.0 (2.6-3.5)

Abbreviation: STARRS, Study to Assess Risk and Resilience in Servicemembers.

The sample consists of STARRS survey respondents who were on active duty in the US Army at the time of their baseline survey in 2011-2013 and who participated after leaving active duty service during 1 or both of the STARRS longitudinal follow-up surveys (STARRS-LS) conducted in 2016-2018 (LS1) and 2018-2019 (LS2). The total sample used for analysis (N = 6837) included 3 subsamples from the original full stacked sample of 16 589 observations (ie, counting each LS survey completed by each respondent as a separate observation and censoring LS2 surveys of respondents who reported experiencing homelessness in the LS1 survey). The first 2 subsamples were a case-control training subsample of 1860 respondents, including all 310 respondents who experienced homelessness and a random 5 times as many respondents who did not experience homelessness (N = 1550) from the 70% of the sample in which the model was trained. The third subsample was the full 30% test subsample of 4977 respondents (133 cases and 4844 controls) in which the model was evaluated. Because the SuperLearner model (a machine learning algorithm) was trained in the case-control sample rather than the full 70% training sample, predicted probabilities from the best SuperLearner model were available only in the matched case-control training subsample. The training control subsample was weighted to represent all respondents in the 70% training sample who did not experience homelessness, yielding a total weighted sample of 16 589, which equals the size of the original sample used in developing and evaluating the model.

Respondents in the top 20% of predicted homelessness risk distribution based on a previously developed machine learning model using information available during active duty were classified as “high risk,” and the remaining 80% of respondents were classified as “lower risk.”

All values were significant at the .05 level using a 2-sided test.

Any of the individual SLEs within this conceptual domain.

Break-in or burglary of home, car, or workplace.

Being a victim of mugging or armed robbery.

For example, getting arrested.

For example, a tax audit or lawsuit.

Any individual SLE across all conceptual domains.

Difference in Predicted Risk of Homelessness

Univariable models documented that each SLE was associated with an elevated probability of becoming homeless among high-risk and lower-risk respondents; 5 SLEs had significantly elevated ARDs among high-risk respondents (eTable 1 in Supplementary Material): separation/divorce/other breakup (8.5%; 95% CI, 0.5%-16.6%), personal betrayal by a loved one (9.8%; 95% CI, 4.5%-15.1%), job loss (10.5%; 95% CI, 5.8%-15.1%), major financial crisis (11.0%; 95% CI, 5.5%-16.6%), and trouble with police (15.6%; 95% CI, 5.2%-25.9%). Four of these 5 univariate ARDs were also significant among lower-risk respondents; only major financial crisis was not (eTable 2 in Supplementary Material). However, the ARDs were consistently lower among lower-risk respondents than among high-risk respondents; the ARD was significant for job loss: 4.5% (95% CI, 2.3%-6.7%), t = 2.4, P = .02. Three other SLEs (victimization by burglary, physical or sexual assault, and trouble with the law), all with low prevalence (range: 0.6% to 2.9%), also had significantly elevated ARDs among lower-risk respondents but not among high-risk respondents, but none of the 3 ARD differences between subsamples was significant (t = 0.1-1.1; P = .27 to .92).

We then estimated sequential multivariable models to examine the overlap across SLEs and to collapse across rare SLEs within conceptual domains because of rarity or similarity of ARDs (eTable 1 and eTable 2 in Supplementary Material). In the final model, 1 individual SLE and 1 group of SLEs had significantly elevated ARDs among high-risk respondents (Table 2): personal betrayal by a loved one (ARD = 0.045; 95% CI, 0.010-0.079) and any economic event (ARD = 0.074; 95% CI, 0.034-0.114). The same 2 SLEs were significant among lower-risk respondents (personal betrayal by a loved one: ARD = 0.007 [95% CI, 0.001-0.012]; economic problems: ARD = 0.022 [95% CI, 0.013-0.031]) but in both cases significantly lower than among high-risk respondents (t = 2.1-2.6; P = .01 to .04). These differences represent the vulnerability components in the rate decomposition presented hereinafter. In addition, 1 individual SLE, separation/divorce/other breakup (ARD = 0.010; 95% CI, 0.003-0.016) and 2 other groups of SLEs (any victimization: ARD = 0.016 [95% CI, 0.004-0.028] and any legal event: ARD = 0.017 [95% CI, 0.001-0.034]) had significantly elevated ARDs among lower-risk respondents, but none was significantly higher than among high-risk respondents (t = 0.4-0.8; P = .45 to .70). Nonadditivity tests for joint effects of exposure to various multiple SLE combinations were not significant among both high-risk (eTable 1 in Supplementary Material) and lower-risk (eTable 2 in Supplementary Material) respondents.

Table 2.

Comparison of adjusted differences in predicted risk of homelessness associated with stressful life events (SLEs) within and between people classified as being at high risk and lower risk of homelessness after leaving active Army status among US Army respondents in STARRS surveys, United States, 2016-2019^a,b,c

Stressful life event	High-risk respondents		Lower-risk respondents
Stressful life event	SLE prevalence, % (95% CI)	Multivariate model, ARD^d,e (95% CI)	SLE prevalence, % (95% CI)	Multivariate model, ARD^d (95% CI)
Serious illness/injury	12.1 (9.2-15.0)	0.021 (−0.038 to 0.080)	5.0 (4.1-5.9)	0 (−0.005 to 0.005)
Interpersonal
Separation/divorce/other breakup	22.7 (19.6-25.9)	0.020 (−0.018 to 0.057)	10.6 (9.3-11.9)	0.010^g (0.003 to 0.016)
Personal betrayal by a loved one	29.9 (26.2-33.5)	0.045^f,g (0.010 to 0.079)	10.0 (8.6-11.3)	0.007^f,g (0.001 to 0.012)
Any economic event^h	41.3 (37.3-45.3)	0.074^f,g (0.034 to 0.114)	19.0 (17.2-20.7)	0.022^f,g (0.013 to 0.031)
Any victimization^h	10.8 (7.6-13.9)	0.008 (−0.031 to 0.046)	4.3 (3.5-5.1)	0.016^g (0.004 to 0.028)
Any legal event^h	8.3 (6.1-10.6)	0.039 (−0.016 to 0.094)	2.1 (1.5-2.7)	0.017^g (0.001 to 0.034)
Any social network event^h	41.5 (37.6-45.4)	−0.012 (−0.034 to 0.011)	26.2 (24.3-28.0)	−0.002 (−0.005 to 0.001)

Abbreviations: ARD, adjusted risk difference; STARRS, Study to Assess Risk and Resilience in Servicemembers.

The sample consists of STARRS survey respondents who were on active duty in the US Army at the time of their baseline survey in 2011-2013 and who participated after leaving active duty service during 1 or both of the STARRS longitudinal follow-up surveys conducted in 2016-2018 (LS1) and 2018-2019 (LS2). The total sample used for analysis (N = 6837) included 3 subsamples from the original full stacked sample of 16 589 observations (ie, counting each LS survey completed by each respondent as a separate observation and censoring LS2 surveys of respondents who reported experiencing homelessness in the LS1 survey). The first 2 subsamples were a case-control training subsample of 1860 respondents, including all 310 respondents who experienced homelessness and a random 5 times as many respondents who did not experience homelessness (n = 1550) from the 70% of the sample in which the model was trained. The third subsample was the full 30% test subsample of 4977 respondents (133 cases and 4844 controls) in which the model was evaluated. Because the SuperLearner model (a machine learning model) was trained in the case-control sample rather than the full 70% training sample, predicted probabilities from the best SuperLearner model were available only in the matched case-control training subsample. The training control subsample was weighted to represent all respondents in the 70% training sample who did not experience homelessness, yielding a total weighted sample of 16 589, which equals the size of the original sample used in developing and evaluating the model.

Comparison of risk differences between respondents classified as being at high risk and lower risk of homelessness were calculated from a single fully interacted model in the total sample.

ARDs and comparisons of ARDs across subsamples were computed using the margins postestimation command in Stata MP version 17.0 (StataCorp).

See eTable 1 in Supplementary Material for a nonadditivity test via a dummy predictor for ≥2 SLEs that was not significant among high-risk respondents (t = −1.4; P = .16) or between high-risk and lower-risk respondents (t = −1.3; P = .20).

Comparison of ARD of homelessness for a predictor among high-risk respondents versus lower-risk respondents, with significance at the .05 level using a 2-sided test.

Comparison of the mean predicted probability of homelessness when the predictor equals 1 versus when the predictor equals 0 among high-risk respondents, with significance at the .05 level using a 2-sided test.

Any of the individual SLEs within this conceptual domain.

Relative Importance of Differential Exposure and Differential Vulnerability

Decomposition showed that 18.9% (95% CI, 14.9%-22.9%) of the observed difference in the prevalence of homelessness between high-risk and lower-risk respondents was because of increased vulnerability to the effects of SLEs, 12.5% (95% CI, 9.8%-15.3%) to increased SLE exposure, and 29.0% (95% CI, 22.0%-36.1%) to the interaction of vulnerability with exposure. The residual was 39.5% (95% CI, 25.7%-53.3%), which means that SLE exposure and/or vulnerability accounted for 60.5% of the observed difference in prevalence of homelessness between the high-risk and lower-risk groups.

Discussion

More than half of respondents reported experiencing ≥1 SLE assessed during the 12 months before their LS survey. Compared with respondents in the lower-risk group, those in the high-risk group were more likely to be exposed to SLEs and had a significantly higher ARD of homelessness associated with some SLEs. Personal betrayal by a loved one and economic problems had the highest differences in ARDs between high-risk and lower-risk veteran respondents. Roughly 60% of the observed difference in the prevalence of homelessness between high-risk groups and lower-risk groups was associated with these SLEs.

People experiencing homelessness are exposed to many SLEs, both before and during the transition to homelessness.²⁰ Prior literature has mostly focused on subgroup analyses (eg, men vs women, chronic homelessness with poor mental health vs without poor mental health) and has not quantitatively assessed the associations between SLEs and risk of homelessness. However, qualitative analyses found that economic problems, breakdown of social ties, and mental illness were perceived by people experiencing homelessness as being determinants of homelessness. The finding that personal betrayal by a loved one and economic problems had the highest differential ARDs among high-risk versus lower-risk respondents adds empirical evidence consistent with this previous qualitative literature.

Prior studies have not examined the prevalence and associations of SLEs with homelessness among veterans, but qualitative literature connects experiences of violence, abuse, social isolation, and housing instability (particularly in the context of intimate partner violence),²¹ which can precipitate social disruption leading to homelessness. A study that qualitatively explored “downward spirals” into homelessness among female veterans identified postmilitary violence, abuse, and relationship termination as key roots of homelessness.²² The associations of physical/sexual assault with homelessness did not hold in the present study, in part because the sample was primarily male; however, these results speak to a near-universal theme of a breakdown in relationships as a powerful risk factor for homelessness. Assessing veterans’ needs related to relationship issues—and providing services to address them—is indicated, perhaps building on the US Department of Veterans Affairs’ work implementing standard screening for intimate partner violence and offering related assistance.^23,24

The presence of recent economic problems was significantly associated with a risk of homelessness. Unstably housed individuals are known to be at elevated risk of homelessness when faced with an SLE and living in high-cost locations²⁵; veterans who have experienced housing instability have described that uncertainty about finances was a meaningful threat to housing.²⁶ Interventions to assist transitioning servicemembers to accrue and sustain their own safety nets, potentially through longer-term shallow subsidies (ie, provision of rental assistance to unstably housed households to supplement rental payments and enable generation of savings), may reduce the risk of a sudden financial crisis. Efforts are currently underway to evaluate the Supportive Services for Veteran Families Shallow Subsidy initiative, which may assist veterans in building a personal safety net.²⁷ Other strategies such as interventions to improve veterans’ money management skills may also be protective against homelessness.²⁸ Access to mental health supports and resilience training may be beneficial for preventing vulnerability to homelessness. Among high-risk servicemembers, it might be helpful to assess modifiable risk factors for SLEs during the transition out of service as well as coping and other resources to buffer the effects of these stressors on homelessness, although such efforts may be impeded by the transitions of these individuals from one system to another (ie, from the US Department of Defense to the US Department of Veterans Affairs) and may depend on eligibility for services.

Our study is novel in its attempt to differentiate the importance of SLE exposure and differential associations of SLEs with homelessness (ie, differential vulnerability) among high-risk compared with lower-risk respondents in accounting for the association of a previously developed index and subsequent homelessness. Our finding that both SLE exposure and vulnerability mediate this association has practical implications in suggesting that intervention efforts should both attempt to prevent modifiable SLEs from occurring to people at high risk of homelessness and increase the ability of these people to cope with these SLEs if they occur.

Limitations

This study had several limitations. First, the temporal associations between SLEs and homelessness are uncertain given that both were assessed retrospectively during the same 12-month recall period. Second, SLEs were not randomly assigned, making it hazardous to assume that predictive associations between temporally primary SLEs and subsequent homelessness were causal. Third, the focus on the past 12 months meant that we missed instances of homelessness that resolved more than 12 months before the LS surveys. Fourth, the measure of homelessness used was somewhat subjective and may not align with the federal definition of homelessness. Lastly, the focus on a limited set of SLEs likely underestimated the overall importance of stressful experiences by omitting consideration of chronic stressors and other SLEs. Results may be biased by nonrandom survey nonresponse.

Conclusions

Preventing and ending homelessness among veterans is an important policy priority, backed by a commitment from a variety of federal partners including the US Department of Defense, US Department of Veterans Affairs, and the US Interagency Council on Homelessness.²⁹ Results provide insights into potential intervention points among transitioning servicemembers related to their experience of SLEs, particularly those who are at high risk of housing instability.

Supplemental Material

sj-docx-1-phr-10.1177_00333549221149092 – Supplemental material for Stressful Life Events and Risk of Homelessness After Active Duty: An Assessment of Risk and Resilience Among Servicemembers

Supplemental material, sj-docx-1-phr-10.1177_00333549221149092 for Stressful Life Events and Risk of Homelessness After Active Duty: An Assessment of Risk and Resilience Among Servicemembers by Ann Elizabeth Montgomery, Katherine A. Koh, Andrew J. King, Robert O’Brien, Nancy A. Sampson, Aldis Petriceks, Murray B. Stein, Robert J. Ursano and Ronald C. Kessler in Public Health Reports

Footnotes

Acknowledgements

The Army STARRS Team consists of the following: coprincipal investigators: Robert J. Ursano, MD (Uniformed Services University of the Health Sciences [USUHS]) and Murray B. Stein, MD, MPH (University of California San Diego and VA San Diego Healthcare System); site principal investigators: Steven Heeringa, PhD (University of Michigan), James Wagner, PhD (University of Michigan), and Ronald C. Kessler, PhD (Harvard Medical School); Army scientific consultant/liaison: Kenneth Cox, MD, MPH (Office of the Deputy Under Secretary of the Army); other team members: Pablo A. Aliaga, MA (USUHS); COL David M. Benedek, MD (USUHS); Laura Campbell-Sills, PhD (University of California San Diego); Carol S. Fullerton, PhD (USUHS); Nancy Gebler, MA (University of Michigan); Robert K. Gifford, PhD (USUHS); Meredith House, BA (University of Michigan); Paul E. Hurwitz, MPH (USUHS); Sonia Jain, PhD (University of California San Diego); Tzu-Cheg Kao, PhD (USUHS); Lisa Lewandowski-Romps, PhD (University of Michigan); Holly Herberman Mash, PhD (USUHS); James A. Naifeh, PhD (USUHS); Tsz Hin Hinz Ng, MPH (USUHS); Matthew K. Nock, PhD (Harvard University); Nancy A. Sampson, BA (Harvard Medical School); COL Gary H. Wynn, MD (USUHS); and Alan M. Zaslavsky, PhD (Harvard Medical School).

Declaration of Conflicting Interests

The authors declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: In the past 3 years, Dr. Stein has received consulting income from Actelion, Acadia Pharmaceuticals, Aptinyx, atai Life Sciences, Boehringer Ingelheim, Bionomics, BioXcel Therapeutics, Clexio, EmpowerPharm, Engrail Therapeutics, GW Pharmaceuticals, Janssen, Jazz Pharmaceuticals, and Roche/Genentech. Dr. Stein has stock options in Oxeia Biopharmaceuticals and Epivario. He is paid for his editorial work on Depression and Anxiety (editor in chief), Biological Psychiatry (deputy editor), and UpToDate (co–editor in chief for psychiatry). He has also received research support from the National Institutes of Health, the US Department of Veterans Affairs, and the US Department of Defense. He is on the scientific advisory board for the Brain and Behavior Research Foundation and the Anxiety and Depression Association of America. In the past 3 years, Dr. Kessler was a consultant for Cambridge Health Alliance, Canandaigua VA Medical Center, Holmusk, Partners Healthcare, Inc, RallyPoint Networks, Inc, and Sage Therapeutics. He has stock options in Cerebral Inc, Mirah, PYM, and Roga Sciences.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Army STARRS was sponsored by the Department of the Army and funded under cooperative agreement #U01MH087981 (2009-2015) with the US Department of Health and Human Services (HHS), National Institutes of Health (NIH), and National Institute of Mental Health (NIMH). Subsequently, STARRS-LS was sponsored and funded by the US Department of Defense (USUHS grant #HU0001-15-2-0004). The contents are solely the responsibility of the authors and do not necessarily represent the views of HHS, NIMH, the Department of the Army, or the US Department of Defense. Dr Stanley was supported in part by a grant from NIMH (T32MH019836).

ORCID iD

Ann Elizabeth Montgomery, PhD

Supplemental Material

Supplemental material for this article is available online. The authors have provided these supplemental materials to give readers additional information about their work. These materials have not been edited or formatted by Public Health Reports’s scientific editors and, thus, may not conform to the guidelines of the AMA Manual of Style, 11th Edition.

References

House Committee on the Budget. President Biden’s budget fulfills our sacred obligation to care for our veterans and their families. April 25, 2022. Accessed November 9, 2022. https://budget.house.gov/publications/report/president-biden-budget-fulfills-sacred-obligation-care-our-veterans#:~:text=Invests%20%242.7%20Billion%20to%20End,budget%20for%20base%20homeless%20programs

Tsai

Pietrzak

Szymkowiak

The problem of veteran homelessness: an update for the new decade. Am J Prev Med. 2021;60(6):774-780. doi:10.1016/j.amepre.2020.12.012

Henry

de Sousa

Tano

, et al. The 2021 Annual Homeless Assessment Report (AHAR) to Congress. Part 1: Point-in-Time Estimates of Homelessness. US Department of Housing and Urban Development; 2022. Accessed January 28, 2022. https://www.huduser.gov/portal/sites/default/files/pdf/2021-AHAR-Part-1.pdf

Tsai

Rosenheck

RA.

Risk factors for homelessness among US veterans. Epidemiol Rev. 2015;37:177-195. doi:10.1093/epirev/mxu004

Koh

Montgomery

O’Brien

, et al. Predicting homelessness among U.S. Army soldiers no longer on active duty. Am J Prev Med. 2022;63(1):13-23. doi:10.1016/j.amepre.2021.12.028

Larson

Norman

. Prospective Prediction of Functional Difficulties Among Recently Separated Veterans. Naval Health Research Center; 2014. Report no. 12-51. Accessed January 24, 2022. https://apps.dtic.mil/sti/pdfs/ADA618146.pdf

Mobbs

Bonanno

GA.

Beyond war and PTSD: the crucial role of transition stress in the lives of military veterans. Clin Psychol Rev. 2018;59:137-144. doi:10.1016/j.cpr.2017.11.007

Morin

. The difficult transition from military to civilian life. December 8, 2011. Accessed January 24, 2022. https://www.pewresearch.org/social-trends/2011/12/08/the-difficult-transition-from-military-to-civilian-life

Pease

Billera

Gerard

Military culture and the transition to civilian life: suicide risk and other considerations. Soc Work. 2016;61(1):83-86. doi:10.1093/sw/swv050

10.

Ursano

Colpe

Heeringa

Kessler

Schoenbaum

Stein

, Army STARRS collaborators. The Army Study to Assess Risk and Resilence in Servicemembers (Army STARRS). Psychiatry. 2014;77(2):107-119. doi:10.1521/psyc2014.77.2.107

11.

Heeringa

Gebler

Colpe

, et al. Field procedures in the Army Study to Assess Risk and Resilience in Servicemembers (ARMY STARRS). Int J Methods Psychiatr Res. 2013;22(4):276-287. doi:10.1002/mpr.1400

12.

Kessler

Colpe

Fullerton

, et al. Design of the Army Study to Assess Risk and Resilience in Servicemembers (ARMY STARRS). Int J Methods Psychiatr Res. 2013;22(4):267-275. doi:10.1002/mpr.1401

13.

Kessler

Heeringa

Colpe

, et al. Response bias, weighting adjustments, and design effects in the Army Study to Assess Risk and Resilience in Servicemembers (ARMY STARRS). Int J Methods Psychiatr Res. 2013;22(4):288-302. doi:10.1002/mpr.1399

14.

Montgomery

AE.

Understanding the dynamics of homelessness among veterans receiving outpatient care: lessons learned from universal screening. Ann Am Acad Pol Soc Sci. 2021;693(1):230-243. doi:10.1177/0002716221995161

15.

Montgomery

Fargo

Byrne

Kane

Culhane

DP.

Universal screening for homelessness and risk for homelessness in the Veterans Health Administration. Am J Public Health. 2013;103(suppl 2):S210-S211. doi:10.2105/ajph.2013.301398

16.

Chevan

Sutherland

Revisiting Das Gupta: refinement and extension of standardization and decomposition. Demography. 2009;46(3):429-449. doi:10.1353/dem.0.0060

17.

Long

Mustillo

SA.

Using predictions and marginal effects to compare groups in regression models for binary outcomes. Sociol Methods Res. 2021;50(3):1284-1320. doi:10.1177/0049124118799374

18.

Dowd

Greene

Norton

EC.

Computation of standard errors. Health Serv Res. 2014;49(2):731-750. doi:10.1111/1475-6773.12122

19.

Ash

. Jackknife and other replication methods with a reduced number of replicates. 2018. Accessed February 10, 2022. http://www.asasrms.org/Proceedings/y2018/files/867226.pdf

20.

Muñoz

Vázquez

Bermejo

Vázquez

JJ.

Stressful life events among homeless people: quantity, types, timing, and perceived causality. J Community Psychol. 1999;27(1):73-87. doi:10.1002/(SICI)1520-6629(199901)27:1<73::AID-JCOP5>3.0.CO;2-#

21.

Montgomery

True

, et al. The intersection of interpersonal violence and housing instability: perspectives from women veterans. Am J Orthopsychiatry. 2020;90(1):63-69. doi:10.1037/ort0000379

22.

Hamilton

Poza

Washington

DL.

“Homelessness and trauma go hand-in-hand”: pathways to homelessness among women veterans. Womens Health Issues. 2011;21(4 suppl):S203-S209. doi:10.1016/j.whi.2011.04.005

23.

Dichter

Haywood

Butler

Bellamy

Iverson

KM.

Intimate partner violence screening in the Veterans Health Administration: demographic and military service characteristics. Am J Prev Med. 2017;52(6):761-768. doi:10.1016/j.amepre.2017.01.003

24.

Dichter

Wagner

Goldberg

Iverson

KM.

Intimate partner violence detection and care in the Veterans Health Administration: patient and provider perspectives. Womens Health Issues. 2015;25(5):555-560. doi:10.1016/j.whi.2015.06.006

25.

Curtis

Corman

Noonan

Reichman

NE.

Life shocks and homelessness. Demography. 2013;50(6):2227-2253. doi:10.1007/s13524-013-0230-4

26.

Montgomery

Byrne

Cusack

, et al. Patients’ perspectives on elements of stable housing and threats to housing stability. J Soc Social Work Res. 2020;11(4):545-567. doi:10.1086/712238

27.

US Department of Veterans Affairs. Supportive Services for Veteran Families (SSVF) program guide. 2021. Accessed February 12, 2022. https://www.va.gov/HOMELESS/ssvf/docs/SSVF_Program_Guide.pdf

28.

Elbogen

Hamer

Swanson

Swartz

MS.

A randomized clinical trial of a money management intervention for veterans with psychiatric disabilities. Psychiatr Serv. 2016;67(10):1142-1145. doi:10.1176/appi.ps.201500203

29.

United States Interagency Council on Homelessness. Ending veteran homelessness. Updated November 30, 2018. Accessed December 5, 2021. https://www.usich.gov/goals/veterans

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.22 MB